(Oxford Scholarship Online) Anderton, Charles H. - Brauer, Jurgen - Economic Aspects of Genocides, Other Mass Atrocities, and Their Prevention-Oxford University Press (2016)

Economic Aspects of Genocides,
Other Mass Atrocities,

and Their Prevention
Economic Aspects
of Genocides, Other
Mass Atrocities, and
Their Prevention
EDIT ED BY CH A R L ES H. A N DERTON
and
J U RG E N BR AU ER
1
1
Oxford University Press is a department of the University of Oxford. It furthers
the University’s objective of excellence in research, scholarship, and education
by publishing worldwide. Oxford is a registered trademark of Oxford University
Press in the UK and certain other countries.
Published in the United States of America by Oxford University Press

198 Madison Avenue, New York, NY 10016, United States of America.
© Oxford University Press 2016
All rights reserved. No part of this publication may be reproduced, stored in

a retrieval system, or transmitted, in any form or by any means, without the
prior permission in writing of Oxford University Press, or as expressly permitted
by law, by license, or under terms agreed with the appropriate reproduction
rights organization. Inquiries concerning reproduction outside the scope of the
above should be sent to the Rights Department, Oxford University Press, at the
address above.
You must not circulate this work in any other form

and you must impose this same condition on any acquirer.
Library of Congress Cataloging-i n-P ublication Data

Names: Anderton, Charles H., editor, author. | Brauer, Jurgen, editor, author.
Title: Economic aspects of genocides, other mass atrocities, and their prevention /
edited by Charles H. Anderton and Jurgen Brauer.
Description: New York : Oxford University Press, 2016. | Includes bibliographical references and index.
Identifiers: LCCN 2016007556 (print) | LCCN 2016008304 (ebook) | ISBN 978–0–19–937829–6 (alk.
paper) | ISBN 978–0–19–937830–2 () | ISBN 978–0–19–060699–2 ()
Subjects: LCSH: Genocide—Economic aspects. | Genocide—Case studies. | Genocide—Prevention. |
Atrocities—Economic aspects. | Atrocities—Prevention.
Classification: LCC HV6322.7 .E26 2016 (print) | LCC HV6322.7 (ebook) | DDC
364.151—dc23
LC record available at http://lccn.loc.gov/2016007556
135798642
Printed by Sheridan, USA
Dedicated to my children, Mark and James, and to
my wife, Roxane.
—CHA
Dedicated to my children, Mahlobo, Lwandle, and Nokuthula,

and to my wife, Jennifer.
—JB
CONTENTS
Acknowledgments xi
List of Contributors xiii
PART ONE ECONOM IC S A N D M A SS AT ROCI T I E S: OV ERV I E W
1. On the Economics of Genocides, Other Mass Atrocities, and Their

Prevention 3
C h a r l e s H . A n d e r t o n a n d J u r g e n B r au e r
2. “A Crime without a Name”: Defining Genocide and Mass Atrocity 28

J a m e s E . Wa l l e r
3. Datasets and Trends of Genocides, Mass Killings, and Other Civilian

Atrocities 52
Ch a r l e s H. A n derton
4. The Demography of Genocide 102

Ta deusz Kugl er
5. The Macroeconomic Toll of Genocide and the Sources of Economic

Development 125
Di m i t r i o s S ou di s , R o b e r t I n k l a a r , a n d R o b b e r t M a s e l a n d
vii
viii Contents
PART T WO T H EOR ET IC A L A PPROACH E S A N D R E V I E WS OF

E M PI R IC A L LI T ER AT U R E
6. Genocide and Mass Killing Risk and Prevention: Perspectives from

Constrained Optimization Models 143
C h a r l e s H . A n d e r t o n a n d J u r g e n B r au e r
7. Incentives and Constraints for Mass Killings: A Game-Theoretic

Approach 172
Joa n E st eba n, M a ssi mo Mor el li, a n d Dom i n ic Roh n er
8. Genocide: From Social Structure to Political Conduct 190

N é s t o r Du c h - B r o w n a n d A n t o n i o F o n f r í a
9. The Microeconomic Causes and Consequences of Genocides and Mass

Atrocities 211
Pa t r i c i a J u s t i n o
10. Development and the Risk of Mass Atrocities: An Assessment of the

Empirical Literature 230
A n k e Hoeffl er
11. Who Stays and Who Leaves during Mass Atrocities? 251
A na M a r í a I bá ñ ez a n d A n dr é s Moya
12. Media Persuasion, Ethnic Hatred, and Mass Violence: A Brief

Overview of Recent Research Advances 274
M a r i a P e t r o va a n d Dav i d Y a n a g i z awa- D r o t t
PART THREE C A SE ST U DI E S I
13. “For Being Aboriginal”—Economic Perspectives on Pre-Holocaust

Genocides 289
J u r g e n B r au e r a n d R au l C a r u s o
14. Identity and Incentives: An Economic Interpretation of the

Holocaust 318
R au l C a r u s o
15. The Economics of Genocide in Rwanda 339

W i ll a Fr i edm a n
Contents ix
16. Peace and the Killing: Compatible Logics in the Democratic Republic
of the Congo 356
Zoë M a r r i age
17. Gender and the Genocidal Economy 378

E l i s a vo n J o e d e n - F o r g e y
PART FOUR C A SE ST U DI E S I I
18. On the Logistics of Violence: Evidence from Stalin’s Great Terror,

Nazi-Occupied Belarus, and Modern African Civil Wars 399
Y u r i M . Zh u kov
19. Strategic Atrocities: Civilians under Crossfire—Theory and Evidence

from Colombia 425
J u a n F. Va r g a s
20. From Pax Narcótica to Guerra Pública: Explaining Civilian Violence in

Mexico’s Illicit Drug Wars 452
N e i l T. N . F e r g u s o n , M a r e n M . M i c h a e l s e n , a n d T o p h e r L . M c D ou g a l
21. Long-Term Economic Development in the Presence of an Episode of

Mass Killing: The Case of Indonesia, 1965–1966 481
S . M a n s o o b M u r s h e d a n d M o h a m m a d Z u l fa n T a dj o e d d i n
22. Economic Foundations of Religious Killings and Genocide with

Special Reference to Pakistan, 1978–2012 510
Pa r t h a G a n g o pa d h y a y
23. Understanding Civil War Violence through Military

Intelligence: Mining Suspects’ Records from the Vietnam War 536
R e x W. D ou g l a s s
PART FIVE TOWA R D PR EDICT ION A N D PR E V EN T ION
24. Economic Risk Factors and Predictive Modeling of Genocides and

Other Mass Atrocities 569
Ch a r l e s R . Bu tch er a n d Ben ja m i n E. Gol dsm it h
x Contents
25. Business in Genocide: Understanding and Avoiding Complicity 591

N o r a M . S t e l a n d W i m N au d é
26. Valuing Lives You Might Save: Understanding Psychic Numbing in

the Face of Genocide 613
Pau l S l o v i c , Da n i e l Vä s t f j ä l l , R o b i n G r e g o r y, a n d K i m b e r ly G . O l s o n
27. Genocides and Other Mass Atrocities: A Law and Economics

Approach 639
J u r g e n B r au e r , C h a r l e s H . A n d e r t o n , a n d Dav i d S c h a p
28. Local and National Democracy in Political Reconstruction 663

Roger B. M y er son
Name Index 675

Subject Index 689
ACK NOW L EDG M EN TS
In compiling this book, we accumulated great intellectual and personal debt to

family, friends, and colleagues worldwide. To use an analogy, while we composed
and orchestrated this work, without our publisher and without our musicians and
support staff, the score would not have been published and performed for the audi-
ence to enjoy. Under the supportive guidance of Scott Parris, our editor at Oxford
University Press, the book proposal underwent critical, anonymous review.
We thank those reviewers for their thoughts and comments, which assisted us
greatly in the final conceptualization of the work. In all, forty-one scholars then
contributed to the research and writing of the twenty-eight chapters that follow.
Except for our summative, introductory chapter, each chapter draft was internally
reviewed by ourselves and by at least one other contributor. Each chapter draft,
including our own chapters, also benefited from anonymous external review—by
more than fifty experts in all. As is professional practice, they will remain anony-
mous but we publicly thank them here for the critically constructive yet courte-
ous and thoroughly professional reviews they wrote of the various chapters, often
extending over many pages of engaged comments and arguments. Virtually every
one of our authors or author teams responded to the respective reviews gratefully
and thankfully. The final product—t his book—t hus is the outcome of a large, col-
laborative enterprise, and it has been our distinct privilege and pleasure to work
with such dedicated and knowledgeable professionals, colleagues, and friends—
dispassionate scholars who are passionate about helping to create a better world
for us all.
We also thank our spouses—Roxane Anderton and Jennifer A. Salamanca-
Brauer—for their forbearance during the course of this project and, in the final
stages of the manuscript preparation, for their practical assistance as well. In addi-
tion, Charles Anderton would like to thank the College of the Holy Cross for its
research support through the W. Arthur Garrity Sr. Professorship and his col-
leagues and students, especially those students who took a course on genocide
that he taught in 2011 and then again in 2014.
xi
LIST OF CONTR I BU TOR S
Charles H. Anderton is W. Arthur Garrity Sr. Professor in Human Nature,

Ethics and Society and Professor of Economics, Department of Economics
and Accounting, College of the Holy Cross, Worcester, Massachusetts. He was
formerly coeditor of Defence and Peace Economics.
Jurgen Brauer is Professor of Economics, Hull College of Business, Augusta
University, Augusta, Georgia, and Visiting Professor of Economics, Faculty of
Economics, Chulalongkorn University, Bangkok, Thailand. He is coeditor of the
Economics of Peace and Security Journal.
Charles R. Butcher is Associate Professor, Department of Sociology and Political
Science, Norwegian University of Science and Technology, Trondheim, Norway.
Raul Caruso is Professor of Economic Policy, Institute of Economic Policy,
Catholic University of the Sacred Heart, Milan, Italy. He is editor-in-chief of
Peace Economics, Peace Science and Public Policy.
Rex W. Douglass is a Postdoctoral Employee, Department of Mathematics,
University of California–San Diego, San Diego, California.
Néstor Duch-Brown is Researcher, Barcelona Institute of Economics, Lecturer
at the Department of Public Economics, and Director of the Chair of Markets and
Industrial Policy, University of Barcelona, Barcelona, Spain.
Joan Esteban is Senior Researcher at the Instituto d’Anàlisis Econòmica and the
Barcelona Graduate School of Economics, Barcelona, Spain.
Neil T. N. Ferguson is Research Affiliate, International Security and Development
Center, Berlin, Germany, and Researcher, Brandenburgisches Institut für
Gesellschaft und Sicherheit, Potsdam University, Potsdam, Germany.
Antonio Fonfría is at the Department of Applied Economics, Complutense
University of Madrid, Madrid, Spain.
xiii
xiv List of Contributors
Willa Friedman is Assistant Professor, Department of Economics, University of

Houston, Houston, Texas.
Partha Gangopadhyay is Associate Professor of Economics, School of Business,
Western Sydney University, Sydney, Australia. He is coeditor of the International
Journal of Development and Conflict.
Benjamin E. Goldsmith is Associate Professor, Department of Government and
International Relations, University of Sydney, Sydney, Australia.
Robin Gregory is Senior Researcher, Decision Science Research Institute,
Director of Value Scope Research, and Adjunct Professor, Institute for Resources,
Environment, and Sustainability, University of British Columbia, Vancouver,
Canada.
Anke Hoeffler is Research Officer, Centre for the Study of African Economies,
Department of Economics, St. Anthony’s College, University of Oxford, Oxford,
United Kingdom.
Ana María Ibáñez is Professor of Economics, Department of Economics,
Universidad de los Andes, Bogotá, Colombia.
Robert Inklaar is Associate Professor, Faculty of Economics and Business,
Department of Global Economics and Management, University of Groningen,
Groningen, The Netherlands.
Elisa von Joeden-Forgey is Assistant Professor of Holocaust and Genocide
Studies, Stockton University, Galloway, New Jersey, and Director of the
university’s Master’s of Holocaust and Genocide Studies Program.
Patricia Justino is Fellow and Honorary Professor, Institute of Development
Studies, University of Sussex, Brighton, United Kingdom.
Tadeusz Kugler is Associate Professor, Department of Politics and International
Relations, Roger Williams University, Bristol, Rhode Island.
Zoë Marriage is Reader in Development Studies, Department of Development
Studies, School of Oriental and African Studies, University of London, London,
United Kingdom.
Robbert Maseland is Associate Professor, Faculty of Economics and Business,
Department of Global Economics and Management, University of Groningen,
Groningen, The Netherlands.
Topher L. McDougal is Associate Professor of Economic Development and
Peacebuilding, Kroc School of Peace Studies, University of San Diego, San
Diego, California, and Research Affiliate, Centre on Conflict, Development, and
List of Contributors xv
Peacebuilding, Graduate Institute for International and Development Studies,

Geneva, Switzerland.
Maren M. Michaelsen is Assistant Professor of Microeconomics, Faculty of

Management and Economics, Ruhr University Bochum, Bochum, Germany.
Massimo Morelli is Professor of Economics, Department of Policy Analysis

and Public Management and the Innocenzo Gasparini Institute for Economic
Research, both at Bocconi University, Milan, Italy.
Andrés Moya is Assistant Professor of Economics, Department of Economics,

Universidad de los Andes, Bogotá, Colombia.
S. Mansoob Murshed is Professor, International Institute of Social Studies,

Erasmus University, Rotterdam, The Netherlands, and Professor of Economics,
Department of Economics, Finance, and Accounting, Coventry University,
Coventry, United Kingdom.
Roger B. Myerson is Glen A. Lloyd Distinguished Service Professor of

Economics, Department of Economics, University of Chicago, Chicago, Illinois,
and recipient of the 2007 Nobel Memorial Prize in Economic Sciences.
Wim Naudé is Dean- Director of the Maastricht School of Management,

Maastricht, The Netherlands, and Chair in Business and Entrepreneurship in
Emerging Markets, School of Economics and Business, Maastricht University,
Maastricht, The Netherlands.
Kimberly G. Olson is an independent researcher and policy analyst, Oregon

State Treasury, Eugene, Oregon.
Maria Petrova is Research Professor, Department of Economics and Business,

Universitat Pompeu Fabra, Barcelona, Spain.
Dominic Rohner is Assistant Professor, Department of Econometrics and

Political Economy, University of Lausanne, Lausanne, Switzerland.
David Schap is Professor of Economics, Department of Economics and

Accounting, College of the Holy Cross, Worcester, Massachusetts.
Paul Slovic is President, Decision Science Research Institute, and Professor of

Psychology, University of Oregon, Eugene, Oregon.
Dimitrios Soudis is Lecturer, Faculty of Economics and Business, Department of

Global Economics and Management, University of Groningen, Groningen, The
Netherlands.
xvi List of Contributors
Nora M. Stel is Research Fellow, Maastricht School of Management, Maastricht,

The Netherlands, and PhD candidate, Center for Conflict Studies, Utrecht
University, Utrecht, The Netherlands.
Mohmmad Zulfan Tadjoeddin is Senior Lecturer in Development Studies at the
School of Social Sciences and Psychology, Western Sydney University, Sydney,
Australia.
Juan F. Vargas is Professor of Economics, Department of Economics, Universidad
del Rosario, Bogotá, Colombia and Research Economist of CAF-Development
Bank of Latin America.
Daniel Västfjäll is Research Scientist, Decision Science Research Institute, and
Professor of Cognitive Psychology, Linköping University, Linköping, Sweden.
James E. Waller is Cohen Professor of Holocaust and Genocide Studies at
Keene State College, Keene, New Hampshire, and Academic Programs Director,
Auschwitz Institute for Peace and Reconciliation, New York City, New York.
David Yanagizawa-Drott is Associate Professor of Public Policy, Kennedy School
of Government, Harvard University, Cambridge, Massachusetts.
Yuri M. Zhukov is Assistant Professor, Department of Political Science,
University of Michigan, Ann Arbor, Michigan.
Economic Aspects of Genocides, Other
Mass Atrocities, and Their Prevention
PA R T O N E
ECONOMICS AND MASS

ATROCITIES: OVERVIEW
1
On the Economics of Genocides, Other

Mass Atrocities, and Their Prevention
C h a r l e s H . A n de rton a n d J u rge n Br au e r
1.1. Introduction
Genocide is a crime. It is not a crime of passion. Rather, it is deliberate, purpose-
ful, and focused—one might as well say “rational”—in its strategic conception
and in its practical execution. In the rationality of genocide, there lies hope. That
which comes into existence through calculation of expected benefits and costs
can also cease to exist, or never come into being in the first place, when such ben-
efits and costs change, or when underlying attitudes, beliefs, or support structures
change. The same can be said of mass atrocities generally, of which genocides are
an example.
Article 2 of the 1948 United Nations (UN) Convention on the Prevention and
Punishment of the Crime of Genocide defines genocide as “any of the following
acts committed with intent to destroy, in whole or in part, a national, ethnical,
racial or religious group, as such: (a) Killing members of the group; (b) Causing
serious bodily or mental harm to members of the group; (c) Deliberately inflicting
on the group conditions of life calculated to bring about its physical destruction
in whole or in part; (d) Imposing measures intended to prevent births within the
group; (e) Forcibly transferring children of the group to another group” (United
Nations 1951). Contrary to popular belief, genocide thus need not involve any kill-
ing at all, although it usually does. Nor do all mass killings constitute genocide, as
genocide requires intent to destroy a group of people as such (Waller 2007, 14).
Crimes distinct from but often associated with genocide and mass killing include
war crimes and crimes against humanity. Collectively, legal scholars refer to these
crimes as atrocity crimes. Including ethnic cleansing, we categorize all such crimes
under the more general heading of mass atrocities.1
Occurring on every continent except Antarctica, the sheer number, scale, and
geographic scope of mass atrocities are vast, as are their frequency and duration.
3
4 Economics and Mass Atrocities: Overview
We have compiled a record of some 201 distinct cases of mass atrocities since
1900, in each of which governments deliberately killed at least one thousand
noncombatant civilians over a period of sustained violence. Not even counting
atrocities committed by nonstate actors, total estimated fatalities for such cases
alone range from about eighty million to more than two hundred million people.
The frequency of such atrocities rivals the number of civil wars (237) and far sur-
passes the number of interstate wars (66) since 1900 as reported by the Correlates
of War (COW) project and the Uppsala Conflict Data Program/Peace Research
Institute Oslo (UCDP/PRIO). The severity of state-perpetrated mass atrocities
relative to other forms of violence is striking. For example, estimated fatalities
totaled over only three genocides (Cambodia 1975–1979, Pakistan/Bangladesh
1971, and Sudan 1983–2002) surpass the total estimated military fatalities for the
237 intrastate wars since 1900. Furthermore, data through 2015 provided by the
National Consortium for the Study of Terrorism and Responses to Terrorism at
the University of Maryland show that almost as many people were killed in just
one month at the height of the 1994 Rwandan genocide as were killed in all inter-
national and domestic terrorist incidents worldwide since 1970 (see c hapter 3 for
these and other details on mass atrocities data).
The systematic study of genocides and other mass atrocities (GMAs) grew
out of Raphael Lemkin’s pioneering 1944 book Axis Rule in Occupied Europe as
well as the efforts of other scholars to understand the Holocaust. In Axis Rule,
Lemkin characterized genocide as a “synchronized attack” against the political,
social, cultural, economic, biological, physical, religious, and moral foundations
of a targeted group’s existence with the purpose of destroying the national pattern
of the victim group and imposing in its place the national pattern of the oppressor
(Lemkin 1944, 79–90). The field of genocide studies, which emerged after World
War II and began to grow significantly in the 1970s, incorporates the multifac-
eted and multidisciplinary perspective of its founder and includes valuable schol-
arship from a variety of disciplines including political science (e.g., Harff 2003;
Valentino 2004), social psychology (e.g., Staub 1989; Waller 2007), sociology
(Kuper 1981; Fein 1993), history (Browning 1992; Kiernan 2007), law (Lemkin
1944; Schabas 2010), religion (Bartov and Mack 2001; Glick 2009), and public
policy (Albright and Cohen 2008; United Nations 2015).2
Although scholars from many disciplines have made important contribu-
tions to the field of genocide studies, economists specializing in the study of
defense and peace have devoted little attention to mass atrocities relative to
other forms of violence such as war and terrorism. That the much graver problem
of genocides, and of mass atrocities in general, has received so little attention by
economists is therefore rather remarkable and may be said to constitute a geno-
cide gap in the field of defense and peace economics. At a minimum, one would
expect the economic consequences of GMAs alone to be huge. Meanwhile,
noneconomists have pointed to important economic issues in historical and
Economics of Genocides , Mass Atrocities , and T heir Prevention 5
contemporary cases of mass atrocities, but such economic facets have not gen-
erally been considered in the context of the rich array of concepts, theories, and
mathematical and statistical tools available in the economics discipline. Thus,
one might say that there also exists an economics gap in the field of genocide
studies (Anderton 2014).
This book helps to fill both gaps. It features theoretical, empirical, case-study,
and policy-oriented research on the nature and importance of economic aspects
of mass atrocities and their prevention. With chapters written by accomplished
scholars drawn mostly, but not exclusively, from economics, the book addresses
some of the large, important gaps in the field of defense and peace economics and
in the field of genocide studies. Before the book moves into these contributions,
this introductory chapter categorizes and outlines economic aspects of GMAs
(section 1.2), highlights and contextualizes some of the important policy-and
research-relevant findings emerging in the chapters that follow (section 1.3), and
identifies important remaining gaps for future research (section 1.4).
1.2. What Has Economics Got To Do With It?

1.2.1. Economic Dimensions of Genocides
and Other Mass Atrocities
Figure 1.1 displays six dimensions in which the field of economics interacts with
genocides and other mass atrocities. The numbered set of boxes in the middle row
of the figure label the dimensions while the unnumbered, subsidiary boxes in the
last row provide additional information.
Box 1 of Figure 1.1 states that GMAs do not just happen: they are choices. The
subsidiary box indicates that these choices can stem from rational and nonra-
tional considerations. In economics, rational—t hat is, deliberate or purposeful—
choice results when a person weighs the expected costs and benefits of various
alternative, feasible options. Nonrational factors influence the choice process.
They include the sociopsychological and neurobiological context and elements
that can cause choices to deviate in systematic and often predictable ways from the
predictions of standard economic theory. The second box indicates that economic
conditions can affect the risk, seriousness, recurrence, and other such diagnostics
of GMAs. Conditions that have been hypothesized (but not necessarily empiri-
cally supported) to affect GMA risk include poverty, inequality, low trade open-
ness, and primary commodity export dependence. In box 3, GMAs affect the
economy through “the five Ds,” as we call them: disruption of economic activi-
ties such as trade; diversion of resources away from ordinary civilian goods to
attack, defense, or flight; displacement of people in the form of refugees and
internally displaced people (IDPs) and of capital investment (e.g., capital flight);
destruction of people, property, and the natural environment; and the difficulty
Economic Aspects of
Genocides and Other
Mass Atrocities (GMAs)
4. GMA as a 6. GMA Distorts

2. Economic 3. GMA
1. GMA is a Mode of 5. GMA as Business Social
Conditions Affects
Choice Wealth Organization Underpinnings
Affect GMA the Economy
Appropriation of Economy
Choice Types Diagnostics 5 Ds Lootable Assets Facets 1. Health

1. Rational 1. Risk 1. Disruption 1. People 1. Recruitment 2. Education
2. Nonrational 2. Seriousness 2. Diversion 2. Economic 2. Logistics 3. Security
3. Recurrence 3. Displacement 3. Cultural 3. Production of 4. Law
4. Destruction suffering and 5. Trust
5. Development Looting Agents death
1. Architects 4. R&D and
2. Collaborators innovation
3. Opportunists 5. Propaganda
6. Management
Figure 1.1 Economic aspects of genocides and other mass atrocities.
of post-GMA political, economic, and cultural reconstitution and reconstruc-

tion of society. The fourth box shows that GMAs can be modes of wealth appro-
priation including the enslavement and sexual exploitation of people from an
out-g roup; seizure of victims’ material assets (e.g., homes, businesses, and other
wealth); the theft and destruction of the out-g roup’s cultural goods and sym-
bols (e.g., artworks, religious sites, historical narratives); and even the taking
away of individuals’ very self-u nderstanding, self-i mage, self-identity, and sense
of belonging to a distinct group of human beings—the annihilation of self.
Looting agents include the GMA architects, their collaborators (e.g., the actual
perpetrators), and sundry opportunists. The fifth box says that, as a practical
matter, all GMAs involve business-related aspects such as the building of the
train tracks that led to Auschwitz and the use of coerced labor within otherwise
standard businesses. The draw on resources required to build, organize, and
manage GMAs thus lies within the domain of economics (and business school
disciplines). The subsidiary box highlights selected aspects of such business-
like activities. Box 6, finally, refers to the societal underpinnings of economies,
such as education, health, security, trust, and the rule of law. Without education
and health, both eminently economic topics, economies do not function well.
Without security in one’s person, property, family, and communities, economic
efficiency is compromised. Without trust, ordinary economic transactions grind
to a halt; and without the rule of customary or formal law, societies lose the lode-
star by which they organize themselves into coherent wholes. The architects and
perpetrators of GMAs routinely disrupt and undermine all of these to the detri-
ment of the afflicted populations and, ironically, usually to their own economic
disadvantage as well.
Although the categories in Figure 1.1 are placed in separate boxes, they over-
lap. For example, the looting of victims’ assets is a mode of wealth appropriation
(box 4), which undermines social trust (box 6), diminishes an economy (box 3),
and affects attitudes toward risk (box 2). Appropriation can be used to promote
the “business” of genocide (box 5) through the recruitment of opportunists seek-
ing to enrich themselves. Importantly, such choices can purposefully be set in
motion by genocide architects, often by employing sociopsychological tools to
sway participation decisions by individual genocide perpetrators (box 1). The
figure is not an explanatory scheme; it just emphasizes the many ways in which
GMAs can bear on the field of economics and in which economics is pertinent to
the field of genocide studies.
1.2.2. Economic Concepts, Theories, and Tools Applicable

to Genocides and Other Mass Atrocities
Economic aspects of GMAs affect the prevention, onset, and spread of GMAs,
but there is more to the subject matter than its economic dimensions alone. In
addition, economic concepts, theories, and tools can be brought to bear in research-
ing GMA choices, consequences, and policies aimed at prevention. Without
intention of completeness, we discuss here some relevant economic concepts.
1.2.2.1. Scarcity
Economics textbooks begin with the fundamental principle of scarcity, namely,
that resources such as the quantities and qualities of labor, capital, natural
resources, entrepreneurial spirit, and time are limited at any given point in his-
tory; necessarily, so are the goods and services produced with these resources. In
the context of this book, scarcity of resources is a critical constraint on the abil-
ity of victims of mass atrocity to escape victimization, the likelihood and nature
of third-party intervention to protect vulnerable populations, and the behavior
of would-be architects, perpetrators, bystanders, and resisters of genocide. Even
Stalin, it turns out, was subject to the law of scarcity (see c hapter 18).
1.2.2.2. Constrained Optimal Choice and Interdependent Choices

Another concept central to economics is constrained optimal choice, which can be
applied to the study of mass atrocities. A choice refers to an action about to be
undertaken in pursuance of a desired objective. Constituting a choice problem, this
action is to be selected from among a set of feasible options, each one of which
is constrained by the scarcity of available resources to defray the cost of any such
action. An optimal choice is one that best achieves the desired outcome, given the
constraints, and thus solves the choice problem. Importantly, the concept does
not question how the desired outcome itself came to be selected only that, once
selected, a person (an “agent” in the language of economics) is expected to go

about the action in an efficient, benefit-maximizing or cost-m inimizing, deliber-
ate, and purposeful manner. In this sense, and only in this sense, does the agent act
as if he or she were making a rational choice: What is rational is not the desired out-
come but the way in which one goes about achieving it, given the constraints. Constraints
(resources and costs) can be monetary and nonmonetary (e.g., income and time),
and desired outcomes (called preferences) are expected (or assumed) to be con-
sistent with various choice axioms economic theory has developed. 3
The decision maker’s choice problem gains in complexity when his or her
choices are affected by the choice options of others, and vice versa. In this case,
agents’ choices become interdependent, an entanglement that makes each agent’s
optimal choice more difficult to identify. Game theory is used to study the optimal
choices of interdependent agents. The formal development of constrained optimi-
zation and game theory models is pervasive in economics, and in many respects
the two approaches are the workhorses of formal economic theory today. Both
appear in this volume. For example, chapter 6 uses constrained optimization the-
ory to explain the ineffectiveness of civilian protection efforts in the Democratic
Republic of the Congo (in the context of the Kiwanja massacre of 2008), while
chapter 7 uses game-t heoretic analysis to show that, for GMAs in general, third-
party intervention to protect civilians can lead to greater victimization, owing to
unintended consequences that can arise in strategic contexts.
1.2.2.3. Rationality
Rational choice still is often wrongly construed to mean that a decision maker
only and selfishly cares about his or her own material objectives and rewards and
that he or she is extremely adept at calculating the costs and benefits of various
potential actions in complex decision-making environments. To be sure, this
view of rationality once served as a helpful, initial theoretical reference point for
the study of human choices, but it has long since been broadened, both in eco-
nomics and in the social sciences more generally. From the 1950s onward, social
scientists began conducting laboratory experiments on human behavior, quickly
discovering that choices often deviated in systematic ways from the predictions
of narrowly conceived rational choice theory. For example, subjects often were
motivated by concerns for fairness, altruism, or revenge; were subject to various
cognitive biases and collective choice failures; and were affected by background
conditions, or reference frames, within which they made choices.
Nowadays, people are held to be boundedly rational, or impurely rational,
whereby they make the best choice they can while subject to more or less severe
limitations on their abilities to undertake complex calculations in the presence
of their personal history, peer effects, time pressures, incomplete lists of relevant
variables—of uncertain magnitudes—and often operating under faulty logic and
ill-considered expectations. Such phenomena are nonrational (but not necessar-

ily irrational) aspects of human decision-making and have spawned the interdis-
ciplinary field of behavioral economics, in which economists, psychologists, and
other social scientists have been modifying rational choice and game theory mod-
els to account for instance for broad motivations, cognitive phenomena, framing
effects, and varieties of strategic behavior. The roles of both rational and nonra-
tional motives in GMA onset and prevention are featured in this book (see, e.g.,
chapters 6 and 26).
1.2.2.4. Opportunity Cost

Another foundational concept in economics is that of opportunity cost, the subjec-
tive or objective value of each alternative choice option available to a decision
maker. In selecting the highest-valued option, the decision maker sacrifices or
forgoes one or more other valuable options that also could have been chosen.
Mistakes notwithstanding, rational decision-makers are generally expected to
select the option with the highest value, so long as it is feasible given one’s resource
constraints.
In the GMA context, the key insight is that architects of atrocity crimes are
expected to manipulate the values of opportunity costs such that, for perpetrators
and victims alike, some options become feasible while others become infeasible.
For example, it is not so much the case, an economist surmises, that the people of
Nazi Germany had lost their moral compass per se. Rather, the price rose precipi-
tously when it was made socially so much more expensive to follow this compass,
especially when the price of atrocity participation was simultaneously lowered
by offering participants the looted assets of victims, social acceptance of hatred,
and career advancement for toeing the party line. The converse applies to the vic-
tims: their cost of escape was to be raised beyond feasibility, conversely lower-
ing the cost of self-entrapment. The notion of opportunity cost lies at the heart
of virtually all choices made in a mass atrocity context, including the choices of
architects, perpetrators, bystanders, resisters, and even third parties seeking to
prevent atrocity. Note that the concept of opportunity cost comes into full force
only when combined with the concepts of scarcity, rationality, and constrained
optimization. It is part of a class of concepts in economics. One lesson here is that
an extremely parsimonious set of concepts can help organize extremely complex
real-world phenomena.
1.2.2.5. Production Function

Another important, and related, class of economic concepts is available in the
realm of production theory, which is centered on the production function. A pro-
duction function specifies how a producer can best combine the inputs or fac-
tors of production (e.g., labor, capital, and raw materials) under a given state of
technology or knowledge to create the output of a good or service. Production

functions in economics assume productive efficiency—at the frontier of feasibility—
in which the inputs utilized generate the maximum possible output for a given
state of technology, or in which a predetermined level of output is created with
a minimum of inputs. (Note that constrained optimization again plays a role. In
effect, we are dealing with a single superclass of concepts.) Numerous affiliated
concepts can be embodied in a production function including economics of scale,
in which a larger scale of production can lower the producer’s average cost per
unit produced; economies of scope, in which the producer can achieve lower aver-
age costs by producing two or more related goods; economies of density or agglom-
eration, in which proximity can lower average communication, transaction, and
transport costs; and economies of learning, in which average costs fall over time
by learning how to refine production methods. For example, distasteful as it is to
think about it, there is no historical question about the scale effects of extermina-
tion camps. In a bizarre way, certain fantastic notions of Nazi leaders required
economies of scale, scope, density, and learning for their objective to be fulfilled.
Fortunately, the ultimate realization of their “final solution” was stopped just in
time, not least because they also had a war to “produce,” and on two fronts, again
pointing to the theme of constrained choice.
Furthermore, the concept of a production function can be applied to the whole
economy (i.e., an aggregate production function), which conceptualizes sources
of economic growth such as economy-w ide inputs of quantities and qualities of
labor, capital, natural resources, and entrepreneurship and their transformation
into economy-w ide outputs, measured for example as the sum total of the value of
goods and services produced. Thus GMAs can have dramatic effects on a society’s
total economic output and its year-over-year growth, which can be, and in this
book is, analyzed with the aggregate production function (chapter 5).
1.2.2.6. Comparative Advantage and the Exchange Economy

The aforementioned concepts also play a prominent role in trade or exchange
theory in regard to the principle of specialized production and trade according to
comparative advantage. Although nonhuman species sometimes display symbiotic
relationships, which are elementary forms of trade, humans have taken special-
ized production and trade to extraordinary levels. Few people in the world are
self-sufficient; most specialize in the production of one or just a few goods—
new economics knowledge, for instance—which they then trade, usually via the
medium of money, for hundreds of other goods that other people have specialized
in producing. The focused slotting of people into highly specialized production
activities unleashes great gains in productivity through repetition, learning, and
innovation. Productive energies liberated across hundreds of millions of people,
who then trade their specialized outputs with one another, can lead to quantities
and varieties of goods and services that would have astounded even recent ances-
tors. Unhappily, specialization and its attendant efficiencies apply to cases of
genocides and other mass atrocities as well, as they would to GMA intervention
where specialization is, as yet, mostly absent.
In addition to the study of the exchange economy, economics also features the
study of the grants economy and the appropriation economy. The latter two both refer
to one-sided actions, either conferring unilateral benefits or imposing unilateral
costs on another party. In the GMA context, appropriation is a sadly central con-
cept. And GMA intervention is—as we shall see—far more driven by exchange
than by grants consideration.
1.2.3. The Economics of Brutality

We have provided a sample of economic concepts relevant to studying GMAs
and their prevention that are applied in this book. Others include concepts from
public economics (e.g., public goods, club goods, and free-rider incentives), indus-
trial organization (e.g., market structures, supply chain economics), economics
of information (e.g., asymmetric information, risk, and uncertainty), political
economy (e.g., feminist and evolutionary economics perspectives), and law and
economics (e.g., law to prevent or punish GMAs), most of which are employed in
this book as well. Nevertheless, when these concepts are featured in standard text-
book treatments of economics (and most of them are), it is usually assumed that
economic phenomena generally promote human flourishing and that they can be
extraordinarily powerful in generating large quantities and varieties of goods and
services under certain societal characteristics such as respect for property rights,
relatively healthy legal and governance institutions, and sound monetary and fiscal
policies. And so it often is. For example, in their principles of economics textbook,
Krugman and Wells (2006, 1) highlight beneficial aspects of economic activities:
It’s Sunday afternoon in the summer of 2003, and Route 1 in central New
Jersey is a busy place. Thousands of people crowd the shopping malls that
line the road for 20 miles. . . . Most of the shoppers are cheerful—and why
not? The stores in those malls offer an extraordinary range of choice. . . .
There are probably 100,000 distinct items available along that stretch of
road. And most of these items are not luxury goods that only the rich can
afford; they are products that millions of Americans can and do purchase
every day.
The scene along Route 1 that summer day was, of course, perfectly
ordinary—very much like the scene along hundreds of other stretches
of road, all across America, that same afternoon. But the discipline of
economics is mainly concerned with ordinary things. As the great
nineteenth-century economist Alfred Marshall put it, economics is “a

study of mankind in the ordinary business of life.”
And so for well over a hundred years, since 1890 in fact if one regards Alfred
Marshall’s Principles of Economics as the profession’s first textbook, the stan-
dard texts that dominate the teaching and research in economics have blithely
assumed that people and property are secure, that peaceful interactions reign
in the economy, and that none of the great forces for economic good unleashed
in New Jersey and thousands of other places across America and the world are
turned toward nefarious ends. But imagine what can happen when these power-
ful forces for creating goods and services are turned to the creation of “bads”
and “disservices,” such as the destruction of specific groups of people. Imagine
human destruction as the desired output; genocide architects who have strong
and even rational motives to destroy a group; the structuring of opportunity
costs to favor perpetration and penalize resistance; production efficiencies of
scale, scope, density, and learning taken to the extreme for destructive purposes;
the extraordinary devastation caused by a destruction industry taking maxi-
mum advantage of specialized production according to comparative advantage;
and the potential for genocide architects to recruit a large number of perpetra-
tors by deploying best practices implied by collective action theory and labor
market economics. Imagine also how difficult it can be to turn off this engine
of destruction, once unleashed, and why it is so critically important to prevent
genocides and other mass atrocities before they get started. Such is the realm of
brutal economics and its particular manifestation in this book: the economics of
genocides and other mass atrocities.
1.3. What This Book Offers

1.3.1. Organization and Chapter Synopses
The book is divided into five parts. Despite this division, the book features numer-
ous instances in which economic facets of GMAs are treated integrally, with many
of the links noted within the chapters. In this subsection, we provide a synopsis
of each chapter.
Part I outlines foundational concepts, data trends, and macrolevel findings
related to economic aspects of GMAs. It begins with this chapter’s overview of
fundamental economic aspects of GMAs, telling summaries of all chapters, and
samples of policy lessons and research possibilities. Chapter 2 explores the his-
tory of the term “genocide,” critical aspects of the life of the founder of genocide
studies, Raphael Lemkin, and definitions and controversies associated with the
terminology of genocide and other forms of mass atrocity. Chapter 3 surveys
fifteen large-sample datasets on genocides, mass killings, and “lower-level”
intentional violence against civilians spanning the twentieth and twenty-fi rst
centuries; presents recent data trends for such atrocities; and provides an
extensive list of 201 genocides and mass killings since 1900. Chapter 4 shows
how genocides affect societies’ demographic characteristics (birth, death, and
migration rates), overall population levels and population regrowth, and cer-
tain gender and age cohorts. The chapter also discusses several research and
policy implications. Macroeconomic consequences of genocides are the focus
of c hapter 5. Statistical evidence is presented to show that overall economic
activity not only declines sharply when genocides commence but that econo-
mies subsequently do not recover this lost ground, at least not within ten years’
time postgenocide. Importantly, this finding is based on the Penn World Table
data (version 8.0), to our knowledge the first time that this data has been tapped
for this kind of work.
Part II covers theoretical and empirical economic perspectives on GMAs.
Taking cues from Raphael Lemkin and other genocide scholars, c hapter 6 ana-
lyzes the rationality and optimality of GMA-related behavior using constrained
optimization theory. The authors derive demand functions for GMA in general
and for specific GMA techniques in particular (e.g., killing versus forced deporta-
tion), and they examine the demand components for possible levers of interven-
tion. Chapter 7 provides a game-t heoretic model in which strategic considerations
in GMAs are analyzed. Among many interesting results, its authors show how
unintended consequences from intervention efforts can arise in strategic con-
texts, leading to more rather than less killing. Chapter 8 conceptualizes GMAs
as extremes of political exclusion, which is analyzed using entry deterrence, stra-
tegic groups, and other concepts from the field of industrial organization (I/O).
In addition to demonstrating that concepts from I/O can well be applied to the
study of GMAs, the authors show how and why political incumbents, to preserve
their monopoly power, will devote resources to destroying the sources of power
of opposing social groups.
Empirical perspectives in Part II include substantive reviews of empirical lit-
eratures relevant to GMA risks and consequences. Thus, c hapter 9 surveys micro-
economic theory and the microeconometric evidence of risks and consequences
of armed conflict at the individual, household, and community levels. To pull the
literature together and point to future research, the author provides an interpreta-
tive framework that focuses on distributional causes and consequences and on
distribution-related strategic aspects of GMAs. Chapter 10 surveys large-sample
(or, in the jargon, large-n) empirical studies focusing, in part, on the relative sta-
tistical importance of economic variables in understanding GMA risk. A provoc-
ative thesis advanced in this chapter is that what may appear, and be measured,
as genocide during or after the fact may not have been conceptualized as such
before the fact. Thus, policy may be misguided if it attempts to deal with violence
as a GMA problem when the root cause or causes may lie elsewhere. Chapter 11
reviews the literature on GMA-related population displacement. It explains how

strategies adopted by armed actors and by victims influence household decisions
to migrate, shows why some households decide to stay in conflict regions, and
explores what are the socioeconomic and behavioral consequences of violent con-
flict for those who migrate and those who stay. This is an important chapter as
the circumstances leading to the decision to stay rarely are examined in the GMA
literature. Chapter 12 takes up the theme of the economics of information (i.e.,
propaganda and the implied asymmetry of information) by reviewing the small,
evidence-based literature on media effects in GMAs. The primary finding—t hat
mass media can have direct and indirect powers (e.g., norm shifting, spillovers to
neighboring villages) to draw people to become perpetrators—is not unexpected,
of course. What is novel here, instead, is the application of sophisticated statisti-
cal techniques to unique datasets to empirically demonstrate the point. A specific
finding is that indirect effects of media persuasion in GMA contexts can be stron-
ger that the direct effects.
Without neglecting political, historical, and other contexts, case study evi-
dence on economic risk factors, consequences, and other aspects of GMAs are pre-
sented in Parts III and IV. The case studies, eleven in all, include some of the most
extreme and well known genocides (e.g., the Holocaust, Rwanda), cases that are
not generally conceived of as instances of genocide or mass killing (e.g., Mexico’s
drug wars and the targeting of suspects during the Vietnam War), and several
less well-k nown cases (e.g., the Yana people in nineteenth-century California,
the Aborigines peoples of Australia). Furthermore, the cases are wide-ranging
across time (spanning the nineteenth through the twenty-fi rst centuries) and
geographic space (all continents except Antarctica). They also vary methodologi-
cally, with those in Part III (chapters 13–17) mostly being qualitative, those in
Part IV (chapters 18–23) featuring quantitative material (e.g., mathematical
models, econometric tests), and one (chapter 17) treating gender as a “people
case” as distinct from the conventional country case.
Several important themes connect the case study chapters. One, of course,
is that each case is fruitfully examined through an economic lens, which serves
to add depth and breadth to previous case study work. Another theme concerns
the role of identity in understanding GMAs. The economics of identity shows
how entrenched and self-reinforcing the self-image of perpetrators can become
and how difficult it can be to change behaviors associated with such self-image
(chapters 12–14, 17, 21, and 22). Furthermore, several of the case study chapters
conceptualize GMAs not along ethnic, religious, or political lines, but along eco-
nomic class lines. For example, chapter 15 on Rwanda includes the usual treatment
of the genocide along ethnic lines, but also by economic class (in which Hutu
and Tutsi are classified, not ethnically, but economically). Similarly, Stalin’s
pogrom against non-Russian rural populations (chapters 13, 18), the marginal-
ization of local peoples in aid programs in the Democratic Republic of the Congo
(chapter 16), and Mao’s starvation policy (not covered in this book), can be seen
as genocide-by-economic-class cases.
This book also offers quantitative case studies in which substantial background
information on GMA cases is presented in combination with case-specific theo-
retical models, datasets, and/or empirical analyses. Thus, c hapter 18 provides
empirical tests of logistics-related hypotheses using new and extraordinarily
detailed case-specific datasets for Stalin’s Great Terror and civilian killings by
Nazi Germany in occupied Belarus during World War II. (Hypotheses are also
tested in the context of modern African civil wars.) The findings are a wonderful
empirical confirmation of the constrained optimization idea: even the most ruth-
less of dictators face constraints and must—and do—adapt their strategies of
causing harm accordingly. Chapter 19 presents a game-t heoretic model in which
two armed groups fighting over territory use atrocities strategically to secure the
compliance of unarmed civilians. Nontrivially, the model predicts that greater
power of the armed groups leads to more, not less, killing of civilians, a prediction
the empirical testing confirms with detailed data drawn from Colombia’s long-
running violence. Chapter 20 analyzes violence against civilians in the Mexican
drug wars of the mid-2000s to mid-2010s. Few will regard the case as one of
genocide, but collectively, tens of thousands of civilians have been intentionally
killed in these wars. Certainly, this is a case of mass atrocity, if not an atrocity
crime under international law. The chapter, too, presents a game-t heoretic model
of strategic interactions in Mexico among drug cartels, the national government,
and subnational state governments. It argues that among the parties an implicit
agreement prevailed prior to Felipe Calderón’s ascension to the presidency in
2006: You scratch our back by not interfering overmuch with the drug trade; we
scratch yours by not killing overmuch. This balance was upset, the authors argue,
when Calderón came to power, and the gangs responded with demonstrative
mass murder. The chapter shows how strategic mistakes may have amplified the
civilian death toll.
Chapter 21 places an episode of politicide—t he Indonesian mass murder of
suspected communists in 1965–1966—w ithin a long-term economic develop-
ment perspective. The authors also make a theoretical contribution in outlining
how an individual person’s identity or self-image is formed, and how it is influ-
enced, even inflamed, to the point of participating in GMAs. An unsettling result
of the case study is that an episode of GMA can bring about seeming political
stability in an otherwise politically fractured society. While this in turn can allow
an economy to grow at a greater rate than it might have otherwise, the case should
not be taken to suggest, however, that GMAs can pay: as is well known, Indonesia
collapsed in yet another set of GMA-related upheavals in the late 1990s (Aceh
and East-Timor). True, repression can seemingly cement society, but putting a
lid on a cauldron is not the same as reducing the heat. Chapter 22 also focuses
on identity and GMA, here in the context of seemingly religiously motivated
violence against minority Shiite Muslims perpetrated by majority Sunni Muslims

in Pakistan. The theoretical contribution of this chapter lies in its development of
a game-t heoretic framework that permits two possible equilibria, one violent and
one peaceful. The dynamics of the tipping point toward either violence or peace
depend, in part, on economic factors, and the empirical evidence in the chapter
shows possible routes of intervention to prevent or at least mitigate the ongoing
Sunni-on-Shiite violence. As in the Mexican case, the Vietnam War is not thought
by most scholars to be a case of genocide or even of mass killing, but c hapter 23
considers aspects of repression and mass violence against people who were, or had
some characteristics of, civilians. The chapter is important methodologically in
that it applies sophisticated data-m ining and machine-based learning techniques
to a unique data record on well over 73,000 people—over 15,000 of whom were
killed—whom the government of South Vietnam and its US advisers suspected
of collaborating with the Viet Cong and North Vietnamese forces. In the process,
the chapter makes valuable points regarding the interpretation and handling of
highly detailed microlevel datasets.
Part V focuses on the immense challenges associated with mass atrocity pre-
vention and postatrocity reconstitution and reconstruction. Chapter 24 pro-
vides an overview of critical distinctions between risk factor studies as opposed
to forecasting studies of genocide, what must be done to better predict geno-
cides, and an analysis of empirical forecasting of genocides. While a fair bit of
material exists on the fact and the how of corporate complicity, the authors of
chapter 25 find that why businesses end up as accomplices to genocide is hardly
ever explored from the perspective of these firms and their representatives them-
selves. Presumably, firms are profit-maximizing entities, but surmising is not
explaining. In advancing some categorical concepts, the chapter breaks open the
topic of businesses in genocide from economic and management perspectives.
Chapter 26 draws upon insights from behavioral economics to show that large-
scale killings run into human limits on affective and cognitive abilities for appre-
ciating the true scale of such atrocities. Based on discoveries of phenomena such
as psychophysical numbing and compassion fade, the chapter details how geno-
cide prevention efforts can fail owing to such stumbling blocks. It concludes with
a long list of constructive, feasible advice on how to improve atrocity reporting
and atrocity-crime-related law based on insights from behavioral economics.
Turning to the economic analysis of law, chapter 27 examines just why inter-
national GMA-related law appears to be so weak in design and force. Relatively
few GMAs are effectively punished, and the law does not appear to function well
in its role of preventing atrocity crimes. In addressing the costs and benefits of
GMA-prevention-contracting states, the chapter helps to break open a new front
of potential scholarship. Finally, chapter 28 focuses on what happens after con-
flict or breakdown of the state and what must occur for democratic transitions
to take root and succeed. The chapter might be regarded as an example of an

economic analysis of prescriptive constitutional law.
1.3.2. Preview of Selected Findings

We now turn to some policy and research lessons. Literally dozens of insights arise
throughout the book, so our summary is selective. Some of the lessons may seem
obvious, others surprising, and yet others beyond feasibility. But the strength of
these lessons is that, based on theoretical and evidence-based research, they are not mere
conjectures. These lessons do not guarantee GMA prevention, of course, but ignor-
ing them will not help the case either.
1.3.2.1. The Rational and Strategic Nature of Mass Atrocities:

Varied, Pervasive, and Consequential
A critically important theme already noted concerns the purposeful and stra-
tegic nature of mass atrocities. For instance, c hapter 6 presents a bleakness theo-
rem according to which perpetrators can overcome piecemeal GMA prevention
efforts owing to extensive substitution possibilities across means of genocide.
Similarly, c hapter 7’s game-t heoretic model also warns of substitution effects.
For example, if exploitation of vulnerable groups is made harder by threats
of intervention, perpetrators might as well resort to elimination. Likewise,
chapter 19’s case study of civilian atrocities in Colombia and chapter 20’s case
study of violence against civilians in Mexico’s drug wars demonstrate that third-
party efforts to protect oppressed groups or to help one group relative to another
can lead to greater civilian killing owing to unintended consequences that lurk
within strategic contexts. Furthermore, chapter 13 draws heavily upon concepts
from constrained optimization theory to understand agents’ motives and choices
in pre-Holocaust genocide cases and offers a rich typology for future applica-
tions of rational-choice thinking to the behaviors of perpetrators, victims, and
third parties. A telling example of constrained decision-making in strategic
environments is chapter 19’s analysis of how it may be virtually impossible for
civilians to remain neutral in armed conflict. One of the more disturbing illus-
trations related in this book is chapter 17’s evidence that extremely brutal means
of atrocity (e.g., torture, rape, forced killing of family members) are not gratu-
itous, but can advance the aims of the perpetrators, for example, by severing chil-
dren’s ties with family, which enhances recruitment of child soldiers, creating
new social orders in which the perpetrators thrive, and increasing the generative
power of the perpetrator group. Despite these dark sides of rationality, there is
also a hopefulness theorem implied by the concept. To the extent that incentives
cause the onset and spread of GMAs and may lead third parties to do nothing,
greater understanding of such incentives also can lead to better policy designs
for thwarting mass atrocities.4
1.3.2.2. Microeconomic Incentives and the Macroeconomic

Environment Matter
Microeconomic incentives of individuals in households, communities, busi-
nesses, and polities to participate in, stand by, or resist GMAs, as well as society-
wide macroeconomic variables that affect GMA risk and prevention, reverberate
throughout this book. Chapter 13, in particular, demonstrates just how far the
microeconomic triad of resource constraints, costs of various possible actions,
and preferences of individuals can go in understanding how people are incentiv-
ized to become perpetrators, bystanders, resisters—and victims—of mass atroci-
ties. Such incentives often operate in insidious ways in which the social acceptance
of atrocity becomes locked in or, worse, magnified owing to economic, psycho-
logical, and other variables that reinforce one another. True, of Hitler’s Willing
Executioners, as one book title has it, there were many. Our book adds examples
but also shows the presence of “unwilling executioners” (child soldiers being
a well-k nown but far from the only example) who simply didn’t have it within
themselves to escape their constraints—and who then acquiesced to murder.
At the microeconomic level—the level of individuals in their various social
settings—t he threat or actual outbreak of violence thus presents enormous chal-
lenges in regard to the care and survival of oneself and of one’s family and imme-
diate circle of friends and triggers decisions whether to fight, stay, or flee. Violence
compels choices whether to join or support an armed group (perhaps impressed
by force)—and, if so, whether to join openly or in secret—or whether to look the
other way or even to resist or refuse the committing of atrocities. Not surprisingly,
over the course of an armed conflict, circumstances can change constantly, and
any person’s behavior can then be fluid as he or she moves in and out of the various
categories identified here.
Economic conditions affect survival and other prospects in insecure environ-
ments. An individual’s incentive to participate in an armed group depends on two
self-reinforcing conditions: vulnerability to poverty and vulnerability to violence
(e.g., chapters 9, 11, 15, 19, 21, and 22). Furthermore, an individual’s decision to
stay or to flee depends in part on the individual’s resources (the wealthy usually
have options, the poor less so), which in turn depend on employment, access to
markets, the ability to avoid expropriation by violence-producing organizations,
and the logistical (im)possibilities of fleeing (chapters 11, 18). In an emerging or
actual violent setting, all these aspects are intertwined with information flows
that can lead to new norms and even new identities for individuals. For example,
there is evidence that in-g roup members tend to conform to atrocity-supporting
media campaigns, especially if they are predisposed to hatred already and
additionally suffer from poverty, low education, or live in close proximity to other
in-g roup members (chapters 12, 21). The theme of mass atrocity–supporting
forms of norm establishment, norm shifting, and new identity formation is one
of the more frightening aspects of the economics of identity highlighted in this
book. Social interactions among in-g roup members can dramatically escalate the
risk of mass killing through reinforcement mechanisms (chapters 12–14, 17, 21),
and relatively small changes in seemingly innocuous initial economic or other
conditions can tip a society away from peace toward mass atrocity (chapter 22).
Economic variables and conditions at the macroeconomic level can also play
important roles in affecting the risk of GMAs and how they might be prevented.
Among the strongest correlates of GMA risk reduction is a healthy economy in
the form of real GDP growth, improved formal or informal employment pros-
pects, and a reasonably fair resource distribution. When economies decline
and formal and informal employment prospects are dire, GMA risk tends to
become elevated. Such results have theoretical, empirical, and case study support
(chapters 9, 10, 15, 16, 20–22). Furthermore, there is evidence that the substan-
tial presence of natural resources and adverse shocks to key economic sectors
(e.g., agriculture, exports) can raise GMA risk, especially in developing countries
(chapters 7, 15, 16, 20–22). There is even the disturbing aspect that a repressive
regime can operate for an extended period of time, so long as the economy func-
tions reasonably well (chapter 21). Once favorable economic conditions falter,
however—as they did in Indonesia in the late 1990s—t he prospect of GMA can
quickly emerge again.
1.3.2.3. Micro and Macro Consequences: Reinforcement

of Economic Dysfunction
At the microeconomic level (individuals, households, communities), the negative
effects of GMA can be devastating and persist into the post-GMA environment.
Unsurprisingly, atrocity survivors tend to experience harsh economic and other
outcomes including diminished employment, education, access to social capi-
tal, and increased risk aversion and vulnerability to poverty (chapters 9, 11, 16,
and 22). For instance, in the Democratic Republic of the Congo in the 1990s and
2000s, persistent exclusion of vulnerable groups from economic and political life
(“othering”), coupled with continuous and pervasive violence against civilians
and insensitive aid efforts, all restricted individuals’ access to food, markets, and
trade opportunities; led to excess taxes and asset seizures; and forced people into
producing low-value goods (e.g., hemp, charcoal) (chapter 16). Such individuals
become ensnared in behavioral poverty traps, situations in which violence leads
to increased risk aversion, diminished hope, and other behavioral changes that
increase vulnerability to and intergenerational transmission of persistent pov-
erty (chapters 11, 16). One of the most traumatic aspects of GMAs are “life force
atrocities” (e.g., rape, genital mutilation, forced sexual intercourse among fam-
ily members), which not only create severe psychological harm but also seriously
disrupt community social networks and the ability of individuals and families to
remain productive in the locations of their victimization (chapter 17), all with
attendant adverse economic consequences (chapters 11, 16).
The overall demographic effects of GMAs can vary significantly from one case
to another depending on the country’s pre-GMA population growth and level of
development. Over the long run, countries with a low level of development and
high population growth can absorb negative population effects of GMAs (e.g.,
Afghanistan). Mature economies with relatively slow population growth, how-
ever, tend to find it more difficult to absorb the demographic consequences of
GMA (e.g., Bosnia). Hence, as societies develop and mature demographically,
future GMAs will tend to experience more severe demographic consequences
(chapter 4). The specific population structure, such as in regard to gender and
age cohorts, can also vary significantly from one case to another. For example,
the Cambodian genocide (1975–1979) led to a male-to-female ratio in the 45–
49 years of age cohort of 0.59 (chapter 4)—just six males for every ten females—
implying huge cultural and socioeconomic consequences.
At the macroeconomic level, GMAs adversely shock the growth path of afflicted
countries. Even as the economy-w ide growth rate eventually returns to its pre-
GMA number, it does so from a significantly reduced base. Thus, there is a per-
manent negative effect on the level of economic activity (chapter 5). The empirical
evidence leads to a reasonable conjecture to explain this distinct effect. It is that,
relative to interstate and civil wars, GMAs are on average more prone to destroy
human and social capital rather than to destroy physical capital. The apparent
asymmetry in the types of economy-w ide inputs destroyed may explain why post-
GMA economic consequences can be more persistent, and recoveries slower, rel-
ative to post-interstate and post-civil war contexts. It is easier to rebuild a factory
destroyed in war than to rebuild the social bonds of a community destroyed by
genocide.
1.3.2.4. Policy Lessons for GMA Prevention and Post-G MA

Reconstitution and Reconstruction
The book is replete with policy (and research) insights regarding GMA preven-
tion and post-GMA societal reconstitution and economic reconstruction. Here is
a sample of five themes.
First, healthy economic opportunities in employment, education, key economic
sectors, and overall avoidance of economic shocks can reduce GMA risk. Of
course, economists always favor economic development, but the profession’s bias,
if there is one, does not make the facts disappear. As a rule of thumb, healthier
economies do see fewer instances of GMAs, everything else the same. Conversely,
weaker economies are inherently more vulnerable to suffer GMA outbreaks.

(Recall, for instance, that the rise of Nazism and Stalinism followed in the wake
of the Great Depression.) Of course, not all weak economies succumb to GMAs,
nor do stronger economies always avoid GMAs, but among other danger signs, a
weak—and perhaps especially a weakening—economy is always a cause for con-
cern. Healthy economic opportunities may be an important long-r un (structural)
variable of GMA risk reduction. Special consideration should be given toward
promoting reasonably fair distributions of income-earning opportunities and
productive asset accumulation (e.g., education, access to land). It is important to
distinguish macroeconomic and microeconomic indicators of economic health
because a healthy macroeconomy can hide unhealthy micro outcomes (e.g., in the
treatment of women, lack of economic opportunities for particular groups, lack of
access to education, and decline of key economic sectors).
Second, GMA prevention and post-GMA reconstruction efforts appear skewed
toward political elites and easily identified groups of victims (e.g., refugees).
In contrast, too little attention is paid to local conditions, local and regional
leaders, and to unrepresented people, for example, those who stay behind in
conflict zones (c hapter 11). For instance, in Iraq, reconstitution and recon-
struction efforts focused on national political and economic elites, yet efforts
must achieve a balance of support for top-down national leaders and bottom-up
local and regional leaders (c hapter 28). In Mexico, when the national govern-
ment gained strength relative to state government, civilians were put at greater
risk. This was because external aid directed at the national level, along with a
reduction of internal transfers, left subnational levels of government vulner-
able to co-option by drug cartels (c hapter 20). In the Democratic Republic
of the Congo, collaboration of aid groups and political leaders at the elite
level left large segments of the population shut out of the peace process. This
served to reinforce incentives for violent conflict (c hapter 16). Furthermore,
when policymakers, practitioners, and scholars consider business complicity,
they almost always focus on large-scale multinational corporations—easy tar-
gets. Ironically, just as firms may be co-opted into GMA complicity, they may
perhaps also be co-opted into ongoing-G MA or post-G MA policy schemes
that are not necessarily beneficial to afflicted populations. The same applies
to foreign nongovernmental organizations (NGOs): they, too, can be swept
into complicity. Alternatively, is pulling foreign staff out and leaving the local
staff behind really the best option? In fact, we know virtually nothing about
just why companies, or NGOs, big or small, get wittingly or unwittingly drawn
into GMAs and their aftermath, which is a huge research and knowledge gap
(c hapters 16, 25).
Third, well-intended policy efforts can backfire and the wrong post-GMA design
can entrench vested interests. In attempting to put Humpty Dumpty together
again, policymakers, practitioners, and scholars working to protect vulnerable
populations should be cautioned by the nonobvious and surprising outcomes

that can arise in strategic GMA contexts. This book provides substantive anal-
yses of how and when such backfiring can unfold. As mentioned, third-party
efforts to protect oppressed groups, support rebel groups, or support one group
relative to another, as well as resource discoveries and democratizations, can
increase civilian killing (chapters 6, 7, 16, 19, 20, 28). Aid and peace efforts in
post-G MA contexts can reinforce the “othering” of out-g roups, elevating rather
than diminishing the risk of future atrocities (chapter 16). The design of any
future policy efforts to prevent GMAs should take the full suite of caveats
offered in this book into account.
Fourth, GMA prevention will require new laws and institutions based on reformu-
lated incentives that consider insights from economics generally and from law and
economics, behavioral economics, and collective action economics in particular
(chapters 25–28). More thought than before has to go into considering the mul-
tiplicity of agents, their interests and incentives, and the design of domestic and
international law, and accompanying institutions, to make credible prevention a
primary objective. We also must think more about firms and overseas customers
who through their purchases wittingly or unwittingly aid and abet GMA regimes.
In the meantime, as in the financial markets or in US bankruptcy law, automatic
intervention triggers could be considered by states and intergovernmental orga-
nizations for cases of GMA. Unilateral intervention could be preauthorized and
prefinanced, and interveners placed under administrative law to hold interven-
ers potentially liable for abuse or neglect (chapters 26, 27). 5 Also important is
the need to fully appreciate and work around the affective and cognitive limita-
tions that lead to psychic numbing and other behavioral phenomena, which can
severely hamper third-party prevention efforts. Chapter 26 offers a rich menu of
policy options for doing just that.
1.4. What’s Missing?

The book covers much ground in regard to economic aspects of GMAs and their
prevention. Nevertheless, the editors were unsuccessful in recruiting authors to
write chapters on some important topics such as specific country datasets; eco-
nomics of culture; looting of cultural artifacts; nonkilling aspects of GMAs; treat-
ment of indigenous populations; pre-t wentieth-century cases including cases in
ancient history; the industrial organization—that is, the actual conduct—of
genocide; and others. Furthermore, even though this book helps to fill the geno-
cide gap in defense and peace economics and the economics gap in genocide stud-
ies, virtually all chapters point to new research that should be undertaken in the
years ahead if the economics discipline is to play a role in understanding and pre-
venting GMAs. We now turn to several such lines of needed research.
Alongside standard economic perspectives (e.g., rationality, strategic choice,

opportunity cost) and models (constrained optimization, game theory), there are
numerous tools and subfields of economics that have the potential to generate
new and rich insights into GMAs but they are only beginning to be applied, or
have not been applied at all, to the topic. First, in the field of behavioral economics,
chapters 6 and 27 note, respectively, that the negative and positive quadrant of the
famous S-shaped function in prospect theory implies that potential perpetrators
will overvalue existential threat more than rational choice theory predicts, and
potential third-party helpers will undervalue helping victims more than rational
choice theory predicts. Conceived of as magnification theorems, owing to limita-
tions in affective and cognitive abilities, the perceived threat to leaders in power is
magnified as is the failure to assist victims in need. The S-shaped function could
be a prominent tool in future theorizing, empirical testing, and case studies of
GMAs. Furthermore, there are numerous other behavioral economics discover-
ies relevant to GMAs including psychic numbing (chapter 26), overweighting of
small probabilities, confirmatory bias and the polarization of attitudes, how peo-
ple habituate to new contexts, interdependent preferences, and collective action
failures (Anderton 2014). Except for the S-shaped curve and psychic numbing,
most insights from behavioral economics into human actions remain relatively
undeveloped and unapplied in GMA contexts.
Second, another important area for future GMA research is the economics of
identity. A person’s identity can be multifaceted and subject to manipulation by
elites, leading to potentially significant changes in preferences (chapters 12–14,
17, 21, 22). Modeling and empirically testing hypotheses related to preference for-
mation is a weak aspect of economic analysis, owing in part to Stigler and Becker
(1977) and others who have claimed (for good reasons) that theoretical explana-
tion based on preference change is hollow because anything can be explained by
claiming that somebody has a (new) preference for it, which is really no explana-
tion at all. And yet, preferences can and do change, and sometimes quickly and
radically. What is important given the Stigler and Becker critique is not that pref-
erences should be held constant but that, if they do change (see c hapter 13 for
examples), we have theoretical reasons for why they do and how we can model
such changes. In GMA contexts, people adopt new behavior (and rationalize it
ex post) in order to survive and/or prosper. In addition to economics of identity
models, related methods for analyzing how norms (or preferences) for or against
GMA can shift are evolutionary game theory (chapter 22), preference falsifica-
tion (Frank 1987; Kuran 1997), reference-dependent utility functions and fram-
ing effects (chapters 6 and 26), and positional goods (Frank 2005).
A third relatively untapped area for future research is how economic variables
and conditions affect GMA risk and prevention. With a few exceptions, knowl-
edge of such variables and conditions is provisional and additional research
is needed to clarify their nature and effects on GMA risk and prevention,
including how such variables interact with political and cultural conditions.
For instance, some economic variables hypothesized to affect GMA risk and
prevention do not as yet have robust empirical support (e.g., higher trade
reduces risk, greater inequality increases risk; c hapters 9, 10, 14, 22); and some
have barely been studied empirically (economic discrimination and rapid
economic decline increase risk; c hapters 9, 10) or have not empirically been
studied at all (e.g., various possible interactions between economic and non-
economic variables; c hapter 10). Moreover, the empirical GMA literature has
not yet matured to move beyond identifying correlates of GMA risk to ascertain
causal properties of economic and other variables, which is probably critical for
GMA prevention (c hapters 7, 10, 12). In addition, c hapter 23 offers a caution-
ary tale for future empirical work using events datasets. Specifically, newspa-
per or other forms of relatively shallow reporting may truncate data to killings
only, or omit critical information about those killed and the killers, and miss
important information in the process of aggregating information into a final
dataset. We also note that almost all empirical research on GMAs is ex post
explanatory work and statistical testing of theoretically derived hypotheses for
GMAs that have already occurred. In contrast, c hapter 24 offers a forward-
looking forecasting analysis of GMAs with potentially important implications
for preventing future atrocities.
Fourth, another major line of future scholarly inquiry concerns short-and long-
term, micro-and macroeconomic consequences of mass atrocities (e.g., chapters 4, 5,
9, 13, 21). Mass atrocities involve psychological costs that can far outweigh eco-
nomic considerations. That said, the loss of people, capital, networks, and institu-
tions diminish a society’s productive base, and such consequences can persist for
years after an atrocity has ended. Not only are family and community economic
relationships harmed by GMAs, but future economic growth is jeopardized as
long-standing social capital undergirding economic discourse and investments in
new capital, infrastructure, and educational opportunities for the next generation
are torn asunder.
Fifth, the field of genocide studies and the emerging subfield of genocide eco-
nomics both often treat victims as passive, as helpless, and as part of a homogenous,
undifferentiated group. Yet some victims have more resources at their disposal
than do others (including networks of escape) and face very different constraints
related to minimizing victimization (e.g., geographic location, ability to blend in,
age, gender). Chapter 13 is especially instructive in this regard because it finds,
by and large, that victims’ options and actual behaviors tend to be understudied,
perhaps a grave scholarly and policy oversight, and it demonstrates how an eco-
nomic lens can be applied to GMA victims as well as to perpetrators and third
parties. In this way, the chapter opens an important door for future research on
economic aspects of victimization, including prevention and protection.
Space limitations preclude a more detailed look at the many other areas of
economics that are only beginning to be applied to GMAs in this book and in
the literature, but further research is obviously needed on the industrial orga-
nization of mass atrocities (e.g., supply chains for industrial-scale mass killing;
c hapters 8, 14, 18, 25), public economics (e.g., public goods, club goods, free-
rider incentives in intervention, and collective action; c hapter 27), econom-
ics of information (e.g., media and propaganda, asymmetric information, risk
and uncertainty; c hapters 12, 21), political economy and gender perspectives
(c hapters 16, 17), labor economics (e.g., the recruitment of perpetrators), oper-
ations research (supply chains, data-m ining techniques; c hapters 18, 23), and
law and economics (e.g., laws and institutions to punish and prevent GMAs,
constitutional design; c hapters 25–2 8). Inasmuch as very many of our cases
deal with developing economies, the entire field of development economics
should continue to consider questions of peace and security, and the condi-
tions that underpin any culture’s foundational social contract, as cornerstone
issues.
Future research on economic aspects of GMAs will require collaboration
with scholars from multiple fields including, to name just a few, political science,
psychology, sociology, history, and law. We trust that the chapters in this book
demonstrate that economics has much to offer to the field of genocide studies,
such as in understanding genocide choices; why genocide prevention efforts
often are “too little, too late”; why well-i ntentioned policy can backfire; and how
future mass atrocities might be prevented. We also believe this book is impor-
tant for the field of defense and peace economics. It certainly demonstrates that
we who work in the field have done “too little, too late” ourselves in researching
mass atrocities and offering economic perspectives on, and policy insights into,
how GMAs can be prevented. So many research avenues lie wide open for future
scholarship.
Notes
1. Crimes against humanity are systematic attacks against civilians involving inhumane means
such as extermination, forcible population transfer, torture, rape, and disappearances.
War crimes are grave breaches of the Geneva Conventions including willful killing, will-
fully causing great suffering or serious injury, extensive destruction and appropriation of
property, and torture. Ethnic cleansing is the removal of a particular group of people from a
state or region using such means as forced migration and/or mass killing (Pégorier 2013).
Ethnic cleansing is not, however, defined as an atrocity crime under the Rome Statute of
the International Criminal Court.
2. For coverage of historical and contemporary perspectives in genocide studies, see
Meierhenrich (2014) and Bloxham and Moses (2010).
3. For such axioms see, e.g., Varian (2014).
4. We use the term “theorem” in several places in this chapter (and in chapter 6) not in the
sense of a formally derived outcome from a mathematical model, but as an intuitive moni-
ker to aid remembrance and teaching of what we believe to be an important general (the-
matic) idea related to GMA prevention.
5. See, e.g., the inaction of Dutch peacekeepers in Srebrenica or of alleged rapes by peace-
keeping troops. Of course moral hazard, unintended consequence, and related issues need
to be considered. If an automatic intervention trigger were set at, say, 500 people killed,
any armed group could try to kill “just” 499 at a time. Automatic triggers could be acted
upon by rebel or other groups seeking to strategically forestall or create intervention. As
in monetary and other policy arenas, a menu of criteria would have to be considered, pos-
sibly by an outside, independent panel of experts who may make a recommendation to the
UNSC to be decided on in an up-or-down vote.
References
Albright, M. K., and W. S. Cohen. 2008. Preventing Genocide: A Blueprint for U.S. Policymakers.
New York: United States Holocaust Memorial Museum.
Anderton, C. H. 2014. “A Research Agenda for the Economic Study of Genocide: Signposts from
the Field of Conflict Economics.” Journal of Genocide Research 16, no. 1: 113–38.
Bartov, O., and P. Mack, eds. 2001. In God’s Name: Genocide and Religion in the Twentieth Century.
New York: Berghahn.
Bloxham, D., and A. D. Moses, eds. 2010. The Oxford Handbook of Genocide Studies.
New York: Oxford University Press.
Browning, C. R. 1992. Ordinary Men: Reserve Police Battalion 101 and the Final Solution in Poland.
New York: HaperCollins.
Fein, H. 1993. Genocide: A Sociological Perspective. London: Sage.
Frank, R. F. 1987. “If Homo Economicus Could Choose His Own Utility Function, Would He
Want One with a Conscience?” American Economic Review 77, no. 4: 593–6 04.
Frank, R. F. 2005. “Positional Externalities Cause Large and Preventable Welfare Losses.”
American Economic Review 95, no. 2: 137–41.
Glick, L. B. 2009. “Religion and Genocide.” In S. L. Jacobs, ed., Confronting Genocide: Judaism,
Christianity, Islam. Lanham, MD: Lexington, 95–118.
Harff, B. 2003. “No Lessons Learned from the Holocaust? Assessing Risks of Genocide and
Political Mass Murder since 1955.” American Political Science Review 97, no. 1: 57–73.
Kiernan, B. 2007. Blood and Soil: A World History of Genocide and Extermination from Sparta to
Darfur. New Haven, CT: Yale University Press.
Krugman, P., and R. Wells. 2006. Macroeconomics. New York: Worth.
Kuper, L. 1981. Genocide: Its Political Use in the Twentieth Century. New Haven, CT: Yale
University Press.
Kuran, T. 1997. Private Truths, Public Lies: The Social Consequences of Preference Falsification.
Cambridge, MA: Harvard University Press.
Lemkin, R. 1944. Axis Rule in Occupied Europe: Laws of Occupation, Analysis of Government,
Proposals for Redress. Washington, DC: Carnegie Endowment for International Peace.
Meierhenrich, J., ed. 2014. Genocide: A Reader. New York: Oxford University Press.
Pégorier, C. 2013. Ethnic Cleansing: A Legal Qualification. New York: Routledge.
Schabas, W. A. 2010. “The Law and Genocide.” In D. Bloxham and A. D. Moses, eds., The Oxford
Handbook of Genocide Studies. New York: Oxford University Press, 123–41.
Staub, E. 1989. The Roots of Evil: The Origins of Genocide and Other Group Violence.
New York: Cambridge University Press.
Stigler, G. J., and G. S. Becker. 1977. “De Gustibus Non Est Disputandum.” American Economic
Review 67, no. 2: 76–9 0.
United Nations. 1951. “Convention on the Prevention and Punishment of the Crime of
Genocide.” https://t reaties.un.org/doc/P ublication/U NTS/ Volume%2078/volume-78-I-
1021-English.pdf [accessed June 26, 2015].
Valentino, B. A. 2004. Final Solutions: Mass Killing and Genocide in the Twentieth Century. Ithaca,
NY: Cornell University Press.
Varian, H. 2014. Intermediate Microeconomics. 9th ed. New York: W. W. Norton.
Waller, J. 2007. Becoming Evil: How Ordinary People Commit Genocide and Mass Killing. 2nd ed.
2
“A Crime Without a Name”

Defining Genocide and Mass Atrocity
Ja m e s E . Wa l l e r
2.1. Introduction
This chapter reviews the historical development of the word “genocide,” both
through the personal story of Raphael Lemkin, who created the term, as well as
through the drafting history of the United Nations (UN) Genocide Convention.
Central to this is an understanding of three conceptual areas that reappear as
recurring threads in the development of international law against the destruction
of groups. First, which groups would be protected? Second, what acts would be
defined as criminal? And third, who would have the jurisdictional responsibility
to prosecute individuals accused of those criminal acts? The chapter concludes
by placing genocide within the larger definitional nexus of mass atrocity crimes,
including war crimes, crimes against humanity, and ethnic cleansing. While
I hope that the chapter includes novel perspectives even for genocide scholars, its
primary purpose is to introduce those who are unfamiliar with the topic, includ-
ing perhaps the bulk of professional economists and assorted other readers, to the
“story” of how the term and the concept of genocide came to be and how genocide
relates to other mass atrocity crimes.
2.2. Raphael Lemkin and the Defining of Genocide

2.2.1. Nazi Atrocities: A Crime without a Name
On August 24, 1941, after a three-day Atlantic sea meeting with US President
Franklin D. Roosevelt, British Prime Minister Winston Churchill returned to
England and made a live radio address on the BBC. It was only two months after
the German invasion of the Soviet Union, code-named Operation Barbarossa.
More than three million German soldiers, reinforced by half-a-m illion auxiliaries
28
“A C r i m e w i t h o u t a N a m e ” 29
from Germany’s allies, had attacked the Soviet Union across a broad front, from
the Baltic Sea in the north to the Black Sea in the south. Special action squads fol-
lowed the German forces as they advanced east. These squads, the Einsatzgruppen,
included four battalion-sized operational groups, with a total strength of about
three thousand men. They were mobile killing units charged with the murder of
anyone in the newly occupied Soviet territories whom the Nazis deemed racially
or politically unacceptable. These included Soviet political commissars and other
state functionaries, partisans, prisoners of war, Roma (Gypsies), and Communist
Party leaders. Specifically targeted for annihilation were Jews.
The first sweep of killing began on Sunday morning, June 22, 1941. Less than
a month later, British cryptographers, having broken the latest Enigma code,
were decoding regular reports from the Einsatzgruppen (see Breitman 1998).
These reports, meant for Berlin, gave detailed accounts and specific numbers of
those killed in mass executions. While Churchill could not reveal the extent of
the detailed knowledge about these killings, lest he undermine the British intelli-
gence objectives, he felt compelled to describe the barbarity being inflicted by the
German forces. In his address, he claimed, “[W]‌hole districts are being extermi-
nated. Scores of thousands—l iterally scores of thousands—of executions in cold
blood are being perpetrated by the German police troops [that is, Einsatzgruppen]
upon the Russian patriots who defend their native soil. Since the Mongol inva-
sions of Europe in the sixteenth century, there has never been methodical, merci-
less butchery on such a scale, or approaching such a scale.” Churchill referred to
the ongoing Nazi atrocities as “a crime without a name” (Churchill 2003, 297–
300). Churchill’s evocative phrase struck a chord with a young European scholar
who already had become immersed in the study of “race murder.” That scholar’s
name was Raphael Lemkin, and he would soon become obsessed with the pursuit
of giving this crime a name.
2.2.2. Raphael Lemkin and Development

of the Term “Genocide”
The biographical facts of Raphael Lemkin’s life have been well chronicled,
buttressed by various versions of his incomplete and unpublished autobiography.1
These facts, though, must be contextualized in the larger frame surrounding the
first half of the twentieth century in which Lemkin was born and lived. His was an
unparalleled time of sweeping historical, political, social, religious, technological,
and cultural revolutions. Lemkin’s obsessive pursuit to name, and criminalize,
Churchill’s “crime without a name” is inseparably twinned with many of the chal-
lenges raised by these transitions. It is a pursuit that touches on, and is touched by,
global issues of modernity, nationalism, and the rise of the nation-state; related
developments in human rights norms and legislation; and deliberations on race,
politics, and the cold war. It is a pursuit whose footsteps still echo today as it
informs several enduring debates that remain as relevant, and as pressing, for us
as they were in Lemkin’s time.
A geography underlies Raphael Lemkin’s biography, a “power of place,” that is
essential to understanding the sense of precariousness that marked much of his
life. Raphael Lemkin was born in 1900 on a farm fourteen miles from the city of
Wolkowysk. Now in Belarus, the Wolkowysk of Lemkin’s youth lay, in Lemkin’s
words, “between ethnographic Poland to the west, East Prussia to the north,
Ukraine to the south, and Great Russia to the east” (Frieze 2013, 3). The second
of three children, he was a precocious boy, mastering nine languages by the age of
fourteen. As soon as he could read, he would “devour books on the persecution of
religious, racial, or other minority groups” (Frieze 2013, 1). A bibliophile, Lemkin
seemed particularly affected by Henryk Sienkiewicz’s Quo Vadis, a novel of Nero’s
persecution of Christians in 64 CE. As part of a traditional Jewish family, home-
schooled by his mother, Lemkin’s reading of far-away suffering was all too often
translated into real-l ife experiences of exclusion, extortion, persecution, and even
nearby pogroms. Forced to temporarily flee his home during World War I, his
family was driven into the forest where his younger brother, Samuel, died of pneu-
monia and malnourishment. Never secure in the Poland of his birth, the young
Lemkin internalized these experiences—both those read about and lived—so
deeply that he “sometimes … felt physically the tension of blood in [his] veins”
(Frieze 2013, 19).
In 1920, Lemkin enrolled at the University of Lvov in Poland (present-day
Ukraine) to study philology. While there, Lemkin came across the story of
Soghomon Tehlirian. Tehlirian was a survivor of the Armenian massacres in
which, from 1915 to 1923, up to a million-and-a-half Armenians perished at the
hands of Ottoman and Turkish military and paramilitary forces and through
atrocities intentionally designed to eliminate the Armenian demographic pres-
ence in Turkey. In 1915, on a death march, Tehlirian had witnessed the rape of his
sisters, the beheading of his brother, and the murder of his parents during the mas-
sacres and had escaped only by being mistakenly left for dead in a pile of corpses
(Balakian 2003, 345). Tehlirian, as part of the radical wing of the Dashnak Party,
and to avenge his family, assassinated Talaat Pasha, one of the Ottoman leaders
who were the architects of the Armenian massacres, on March 15, 1921 in the
Charlottenburg district of Berlin. “This is for my mother,” he told Pasha as he shot
him (Frieze 2013, 20). Tehlirian, in what was a sensational trial for its time, was
eventually acquitted on the grounds of “psychological compulsion,” or what today
would be called temporary insanity, rooted in the soul-w renching trauma he had
endured and continued to suffer.
The then twenty-one-year-old Lemkin, in conversation with his professor
at the University of Lvov, asked a deceptively simple question: “It is a crime for
Tehlirian to kill a man, but it is not a crime for his oppressor to kill more than a
million men? This is most inconsistent” (Power 2002, 17). His professor cited the
banner of state sovereignty—t he right of every state to conduct its internal affairs
independently. That is, states and statesmen could do as they pleased within their
own borders. His professor continued: “There was no law under which he [Talaat]
could be arrested. … Consider the case of a farmer who owns a flock of chickens.
He kills them, and this is his business. If you interfere, you are trespassing” (Strom
2007, 3). Lemkin’s response, that “sovereignty cannot be conceived as the right to
kill millions of innocent people,” was a moral-t hreshold moment that anticipated
his subsequent transfer to the Lvov law school, where he began to search for legal
codes that would punish as well as prevent the mass murder of civilians (Frieze
2013, 20).
Following graduation, working as a public prosecutor in Warsaw, Lemkin’s
next step in what would become a lifelong crusade toward making such law
came when he developed a proposal that would commit the Polish government
and others to stopping the targeted destruction of ethnic, national, and religious
groups. He was scheduled to present the proposal, arguing for the establishment
of an international law, at a League of Nations conference for the unification of
criminal law in Madrid, Spain in October 1933. At the last minute, the Polish
minister of justice denied Lemkin the travel visa necessary to attend the meeting.
The denial was explained on the basis that Lemkin’s proposal was “anti-German
propaganda” and there was concern that Lemkin might give the wrong impres-
sion to other governments about Polish foreign policy. An influential anti-Semitic
Polish newspaper also denounced Lemkin for being solely concerned to protect
his own race.
Undeterred, Lemkin found a delegate who agreed to present his proposal (Strom
2007, 10). The proposal—titled “Acts Constituting a General (Transnational)
Danger Considered as Offences Against the Law of Nations”—called for a new
type of international law to legislate against “general (transnational) danger [that]
threatens the interests of several States and their inhabitants” (Lemkin 1933).
Lemkin’s paper was presented, in his absence, and tabled. Delegates were not
given the opportunity to accept or reject the proposal. Samantha Power, now US
Ambassador to the UN, points out that some delegates believed that these crimes
happened too seldom to legislate and most were skeptical about the “apocalyptic
references to Hitler,” appearing even as early as October 1933 (Power 2002, 22).
Moreover, nearly all seemed to be in agreement that state sovereignty trumped
mass atrocities against a state’s own citizens. Sovereignty holds that states should
enjoy political independence and autonomy without outside interference. States
have the right to govern and control without external interference, and they also
have the right to nonintervention from external actors in internal state affairs.
International law, such as that proposed by Lemkin, should never usurp the sanc-
tity of domestic laws. As a dogged Lemkin noted, however, the lawyers at the
conference “would not say yes, but they could not say no” (Frieze 2013, 24). In
Power’s words: “They were not prepared to agree to intervene, even diplomati-
cally, across borders. But neither were they prepared to admit that they would
stand by and allow innocent people to die” (Power 2002, 22).
Not to be dissuaded by the cool reception his proposal received in Madrid,
Lemkin continued to push his agenda over the next several years at law confer-
ences in Budapest, Copenhagen, Paris, Amsterdam, and Cairo. After being dis-
missed by the Polish government for refusing to curb his criticisms of Hitler, he
opened a private law practice in Warsaw in 1934, focusing on the international
ramifications of tax law. Lemkin’s practice thrived, and he refocused his scholarly
efforts on the drafting of a treatise on exchange control regulations.
With the Nazi invasion of Poland on September 1, 1939, however, Lemkin
soon became an internally displaced refugee. Fleeing the bombing of a train on
which he was a passenger, Lemkin was forced to retreat to the woods outside of
Warsaw. There, he witnessed further bombing attacks and the death of many from
starvation, disease, and exhaustion. After six months of this nomadic existence,
Lemkin decided to flee and, failing to persuade his family to join him, he escaped
to then-neutral Lithuania before receiving a visa to Sweden, where he taught at the
University of Stockholm. Cleared for immigration to the United States in 1941,
he made an arduous 10,000-m ile journey across the Baltic Sea, Siberia, Japan, the
Pacific Ocean, Canada, and the continental United States. In April 1941, Lemkin
finally arrived in Durham, North Carolina to teach at Duke University. There, and
later at Yale University, Lemkin continued to sharpen his 1933 Madrid proposal.
After the United States entered the war in late 1941, Lemkin joined the ranks of
public service, first as a consultant to the Board of Economic Warfare (later to
become the Foreign Economic Administration) and later as a special adviser on
foreign affairs and international law to the War Department.2
In 1942, Lemkin sent a carefully worded memo to President Roosevelt in which
he suggested the adoption of a treaty to make the mass destruction of civilians
an international crime. He urged “speed” and that “it was still possible to save at
least a part of the people.” Several weeks later, Roosevelt responded with his own
urging—“patience.” In his autobiography, a frustrated Lemkin writes: “ ‘Patience’
is a good word for when one expects an appointment, a budgetary allocation, or
the building of a road. But when the rope is already around the neck of the victim
and strangulation is imminent, isn’t the word ‘patience’ an insult to reason and
nature?” (Frieze 2013, 115).
Realizing that he “was following the wrong path” and that, indeed, “statesmen
were messing up the world,” Lemkin turned his attention to publishing his collec-
tion of documents on Nazi laws and decrees of occupation. He saw in this work a
“picture of the destruction of peoples” that, he held, would give people “no choice
but to believe.”3 In November 1944, the Carnegie Endowment for International
Peace published Lemkin’s manuscript as Axis Rule in Occupied Europe (Lemkin
1944). The major part of the 721-page book dealt with detailed commentaries on
laws and decrees of the Axis powers, and of their puppet regimes, for the govern-
ment of occupied areas.
One chapter, however, was devoted specifically to the subject of genocide.
Lemkin restated his 1933 Madrid proposal to outlaw the targeted destruction of
groups and urged the creation of an international treaty that could be used as a
basis for trying and punishing perpetrators. Most importantly, however, it was in
this chapter that Lemkin proposed the term “genocide,” which he had coined the
year before and briefly introduced in the preface, from the ancient Greek word
geno (race, tribe) and the Latin cide (to kill). As he defined genocide, it meant
a coordinated plan of different actions aiming at the destruction of

essential foundations of the life of national groups, with the aim of anni-
hilating the groups themselves. The objectives of such a plan would be
disintegration of the political and social institutions, of culture, language,
national feelings, religion, and economic existence of national groups,
and the destruction of the personal security, liberty, health, dignity, and
even the lives of the individuals belonging to such groups. . . . Genocide is
directed against the national group as an entity, and the actions involved
are directed against individuals, not in their individual capacity, but as
members of the national group. (Lemkin 1944, 79)
As historian Thomas Butcher argues, Lemkin’s conception of genocide was

multifaceted; a “synchronized attack” that was part of a complete, integrated
policy (Butcher 2013, 253–56). For Lemkin, the concept of genocide would
protect the right to life of national groups, just as the concept of homicide pro-
tects the right to life of individuals. Lemkin broke from the nationalist ideology
that national groups were defined by language, blood, and territory; for Lemkin,
national groups were “families of mind” who shared common beliefs and sen-
timents and whose identities were mutable and constantly changing (Irvin-
Erickson 2013). While one scholar critiques Lemkin’s preliminary conception of
genocide, as articulated in Axis Rule, as “extremely vague, confused, and, when
illustrated by empirical referents, invalid” (Stein 2005, 180), Lemkin had, at least,
finally given the crime a name.
2.2.3. Lemkin’s Drive to Make Genocide a Crime

in International Law
Now, Lemkin turned his attention toward making international law against the
crime of genocide. In 1945–1946, he left his position with the War Department
to become an unofficial adviser to Robert Jackson, US Supreme Court Justice and
US Chief of Counsel at the International Military Tribunal (IMT) in Nuremberg.
As legal scholar John Q. Barrett has demonstrated, while only nominally affiliated
with Jackson’s staff, “Lemkin’s intellectual work was known to and influenced
Jackson and his staff” (Barrett 2010, 39). While the word “genocide” does not
appear in the Tribunal’s Charter, it does appear in the drafting history of the
Charter, as well as in count three (War Crimes) of the IMT indictment, and was
spoken for the first time in a courtroom litigation proceeding when the Nuremberg
trial began on November 20, 1945. On June 25, 1946, a British deputy prosecutor,
Sir David Maxwell-Fyfe, read the definition of genocide directly from Lemkin’s
book during a cross-examination of defendant Konstantin von Neurath. On July
27, Sir Hartley Shawcross, the British chief prosecutor, spoke in his closing argu-
ment of the defendant’s crimes of genocide, as did French deputy prosecutor
Charles Dubost in his closing argument.
While progress was made in the use of the word “genocide” during the trial, the
word does not appear in the final judgment on October 1, 1946. A disappointed
Lemkin wrote: “The Allies decided their case against a past Hitler but refused
to envisage future Hitlers. They did not want to, or could not, establish a rule of
international law that would prevent and punish future crimes of the same type”
(Frieze 2013, 118). More disconcertingly for Lemkin, the IMT maintained that
states and individuals who did not cross an international border were still free
under international law to commit genocide. In other words, the Allies did not
question Germany’s absolute authority over its internal affairs before the war.
As legal scholar William Schabas points out, “although there was frequent refer-
ence [during the trials] to the preparation for the war and for the Nazi atrocities
committed in the early years of the Third Reich, no conviction was registered for
any act committed prior to September 1, 1939” (Schabas 2006, 95). In essence,
had the Nazis killed only German Jews, they would not have been liable for any
international crime. Later, Lemkin was to call the Nuremberg judgment “the
blackest day of my life” (Korey 2001, 25)—a rather high bar, given Lemkin’s life
to that point. (As a minor victory, however, it should be noted that several of the
twelve subsequent Nuremberg trials, held from 1946 to 1949, that followed the IMT
did include genocide as a separate charge. In addition, the Polish Supreme National
Tribunal adopted Lemkin’s framework and convicted Amon Goeth, Rudolf Hoess,
and Artur Greiser of genocide under Polish law, becoming the first state to use the
word “genocide” in its domestic criminal proceedings; see Nersessian 2002.)
By 1946, Lemkin’s work took on a new, distinctly personal, urgency. He had
lost forty-n ine relatives, including his parents—likely gassed at Treblinka—to
the Holocaust. The only European members of his family to survive the Holocaust
were his brother, Elias, and Elias’s wife and two sons. In a draft preface to his auto-
biography, Lemkin wrote of his reaction to this enormous personal loss:
When I have conceived the idea of outlawing genocide, I hardly could

imagine that it will affect me personally. During the war 49 members
of my family perished from Genocide, including my parents. Suddenly
I felt that the earth is receeding [sic] from under my feet and the sense
of living is disappearing. But soon I have transformed personal disaster

into a moral striking force. Was I not under a moral duty to repay my
mother for having stimulated in me the interest in Genocide? Was it not
the best form of gratitude to make a ‘Genocide pact’ as an epitaph on her
symbolic grave and as a common recognition that she and many millions
did not die in vain? I redoubled my efforts and found temporal relief from
my grief in this work.4
Affected by these personal losses, and stung by the disappointment of Nuremberg,

Lemkin would be transformed from a “crusader” into a “zealot” for making inter-
national law against the crime of genocide (Korey 2001, 26).
Outside legislative circles, Lemkin’s new term caught on quickly. A December
3, 1944 Washington Post editorial (written at Lemkin’s urging) claimed that “geno-
cide” was the only word that properly described the murder of Jews at Auschwitz
(Washington Post 1944). On January 21, 1945, a full-page review of Axis Rule in
Occupied Europe appeared on the lead page of the New York Times Book Review. The
term “genocide” resonated in ways that “race murder,” “mass murder,” “denation-
alization,” “barbarity,” “vandalism,” “terrorism,” and other descriptions had not.
In his unpublished autobiography, Lemkin recounts a 1946 conversation in which
Judge Abdul Monim Bey Riad of Saudi Arabia said: “It [a convention against
genocide] is a beautiful concept. It is something worth living for. The word ‘geno-
cide’ has so much appeal, so much force” (Frieze 2013, 129). By 1950, the word
had made its first appearance in the Merriam-Webster English dictionary. In the
United States, the word quickly became so commonplace that, in 1962, Governor
Ross Barnett of Mississippi, in defending his reactionary decision to prevent the
enrollment of James Meredith at the University of Mississippi, could state pub-
licly, “we will not drink from the cup of genocide”—w ith near certainty that his
fellow white supremacists would grasp the meaning of the word (Mitchell 2012).
Still searching for international legislative weight to back the new term, Lemkin
devoted himself tirelessly, and even more obsessively, to a single-handed cam-
paign to make “genocide” an international crime. He looked to the newly estab-
lished United Nations organization (founded October 24, 1945) to construct an
international law that did not link the destruction of groups to internationally
recognized cross-border aggression. First written by Lemkin “on a soft sofa in the
Delegates’ Lounge” of the UN offices in Lake Success, New York—and supported
with sponsorships from Panama, Cuba, and India—a draft resolution was pre-
pared for presentation at the first session of the General Assembly in late 1946
(Frieze 2013, 122). In requesting the inclusion of the resolution on the agenda, the
Cuban delegate, Ernesto Dihigo, reminded the Committee that “at the Nurnberg
trials, it had not been possible to punish certain cases of genocide because they
had been committed before the beginning of the war. Fearing that such crimes
might remain unpunished … [Dihigo] asked that genocide be declared an inter-
national crime. This was the purpose of the resolution” (United Nations 1946).
Clearly, the initial Nuremberg trial’s failure to recognize the criminality of atroci-
ties committed in peacetime left a lingering dissatisfaction within the interna-
tional community. Vulnerable emerging states of the underdeveloped world were
particularly invested in developing an instrument to protect them from repressive
acts and atrocities that could be committed against them in peacetime (Schabas
2006, 96). There was a legal gap to fill and codifying the crime of genocide was
seen by many as a legal, and moral, necessity.
The resolution forwarded to the UN General Assembly, Resolution 96 (I), con-
demned genocide as “a denial of the right of existence of entire human groups. …
Many instances of such crimes have occurred, when racial, religious, political and
other groups have been destroyed, entirely or in part” (United Nations General
Assembly 1946, 188–89). The brief resolution analogized genocide to the crime
of homicide. As legal scholar Howard Shneider has stated: “Just as the key ele-
ment of homicide is the taking of another human being’s life, regardless of who
that human being is, the Resolution argued that the key element of genocide is
the taking of a human group’s life, regardless of the characteristics that bind the
group” (Shneider 2010, 318).
Perhaps most significantly, Resolution 96 (I) went beyond a mere symbolic
declaration and also tasked the Economic and Social Council, one of the six prin-
cipal organs of the UN, with drafting a convention on the crime of genocide. On
December 11, 1946, the General Assembly of the UN unanimously passed the
resolution without debate.
2.2.4. Drafting of the UN Genocide Convention

The subsequent drafting process of what would become the Genocide Convention
would go through three stages, and end up taking two years, with Lemkin’s direct
involvement varying throughout. Behind the scenes, Lemkin became a “one-man,
one-globe, multilingual, single-issue lobbying machine” to ensure passage of the
convention (Power 2002, 61). In July 1948, Lemkin dropped everything to fly
to Geneva to find “new friends” at the Economic and Social Council to support
the convention. He strategically surmised “that the delegates, away from home,
were more lonely here than in New York and might have more time to listen to me”
(Frieze 2013, 135). He then moved on to Paris, “learning the composition of the
delegations and lining up strategic forces,” in preparation for the General Assembly
discussion of the convention (Frieze 2013, 151). While there, he cajoled “a number
of friendly delegates” into streamlining the subcommittee procedure necessary for
presentation of the Convention to the General Assembly (Frieze 2013, 151).
Historian Jay Winter describes Lemkin as “a man of great conviction, but not
great charm, he was an uncomfortable presence, someone not easily deflected.
Speaking to him a second, third, or 33rd time could be tiresome or worse”
(Winter 2013, 6). Benjamin Ferencz, chief prosecutor of the SS-Einsatzgruppen
case at Nuremberg, recalled thinking that Lemkin was odd, maybe even “crazy”
(Earl 2013, 323). Not above guilt-inducing hyperbole, Lemkin once wrote to the
head of the Methodist women’s council: “I know it is very hot in July and August
for work … but let us not forget that the heat of this month is less unbearable to
us than the heat in the ovens of Auschwitz and Dachau” (Earl 2013, 318). Various
accounts refer to him as a pest, crank, or nudnik. Never one for small talk, this
fits with A. M. Rosenthal’s (a columnist and the executive editor at the New York
Times) personal recollections of Lemkin walking “the corridors of the UN. He
stopped journalists, took junior delegates by the arm and hung on until they lis-
tened, at least a moment. To see an ambassador, he would plan and plot for weeks
and sit for days in reception rooms.” Rosenthal continued: “He [Lemkin] had no
money, no office, no assistants. He had no UN status or papers, but the guards
always let him pass… . He would bluff a little sometimes about pulling political
levers, but he had none” (Rosenthal 1988, A31).
Because of his persistence, and despite the fact that he could certainly reach
the bounds of pesky annoyance, Lemkin was able to rally a worldwide network
of support for his cause. Whether viewed by his contemporaries as a “dreamer”
or “fanatic,” most would agree, as archivist Tanya Elder argues, “as a lobbyer,
Lemkin was brilliant” (Elder 2005, 481). He used correspondence, petitions, arti-
cles, radio addresses, interviews, public lectures, and a broad range of coalitions
to build support for the passage of the Convention.
Due, in large part, to his tireless campaigning, the United Nations Convention
on the Prevention and Punishment of the Crime of Genocide (commonly known
as the Genocide Convention) was finally adopted at the Palais de Chaillot in Paris
on December 9, 1948, one day before the end of the assembly. Fifty-five delegates
voted yes to the pact; none voted no. The Genocide Convention—U N Resolution
260 (III)—became the first human rights treaty adopted by the General Assembly
of the United Nations.
An ill and destitute Lemkin managed to make the Paris Conference and was
present when the treaty was adopted. Hours after the vote, Rosenthal recounts
finding an exhausted Lemkin in a darkened Assembly Hall, “weeping as if his
heart would break” and asking “please to be left in solitude” (Elder 2005, 481).
Days later, Lemkin fell gravely ill and was hospitalized. For nearly three weeks, the
doctors struggled with a diagnosis. Lemkin finally offered one himself: “genocidi-
tis; exhaustion from the work on the Genocide Convention” (Frieze 2013, 179).
2.2.5. The Articles of the UN Genocide Convention

The Genocide Convention includes nineteen concise articles. 5 Throughout, we
can see fingerprints of Lemkin’s influence, some traceable to his original 1933
Madrid proposal. We also see, however, some significant deviations from how
Lemkin conceived of the crime of genocide and its punishment. Some were
political compromises, necessary to ensure passage of the Convention. Others

were rooted in intransigent notions of state sovereignty, race, and jurisdictional
responsibilities that would prove divisive in committee deliberations.
The first two articles give the substance of the Convention. Article I noted that
genocide can be “committed in time of peace or in time of war” (Human Rights
Web 2011). This was a significant deviation from the International Military
Tribunal’s initial confining of crimes against humanity to only those acts perpe-
trated after the outbreak of war. While later international jurisprudence would
clarify that crimes against humanity may also occur in peacetime, the 1948
Convention was groundbreaking in its decision that mass destruction of peoples
need not be limited to armed conflict; war does not have to be present for geno-
cide to occur. Peacetime atrocity was no longer beyond the reach of law.
Article II, however, is the central defining article of the Convention: “In the
present Convention, genocide means any of the following acts committed with
intent to destroy, in whole or in part, a national, ethnical, racial or religious group,
as such: (a) killing members of the group, (b) causing serious bodily or mental
harm to members of the group, (c) deliberately inflicting on the group condi-
tions of life calculated to bring about its physical destruction in whole or in part,
(d) imposing measures intended to prevent births within the group, or (e) forcibly
transferring children of the group to another group” (United Nations 1951).
Within Article II, we see three conceptual areas at play. First, Article II defines
the protected groups as “national, ethnical, racial, or religious.” Although the
inclusion of political groups had survived for several sessions of the deliberations,
it had been a matter of contention throughout. Supporters of its inclusion pointed
out future genocides would likely be committed mainly on political grounds and
that the inclusion of political groups was consistent with the protection of reli-
gious groups—both were defined by beliefs in an idea. Many even argued that
genocide, in its most serious form, was nearly always a political crime. If political
groups were not included, some worried that perpetrators might hide behind the
pretext of the political opinion of a racial or religious group to destroy it, with-
out being liable to any international sanctions. It was commonly understood that
political groups constituted an especially vulnerable population, and there was no
legally defensible principle that could justify their exclusion.
The proponents of protecting political groups under the Convention may
have been equipped to win the battle with moral and logical force, but they were
unable to win the political war that ensued. The core of the opposition came from
Eastern bloc countries led by the Soviet Union, who were joined by a number
of Latin American states (Brazil, Peru, Venezuela, the Dominican Republic,
and Uruguay) as well as Lebanon, Sweden, the Philippines, Egypt, Belgium,
and Iran. These diverse voices argued for the exclusion of political groups on the
grounds that such groups—mutable, imprecise, transient, unstable, and lacking
in distinguishing characteristics—could not be objectively designated (under a
mistaken assumption that “national,” “ethnical,” “racial,” or “religious” could be

objectively designated). They maintained that membership in political groups
was voluntary—unlike membership in national, ethnical, racial, or religious
groups—and that they lacked permanency and consistency. While people can
certainly change nationality and religion, such changes did not happen as fre-
quently as people changing their political beliefs. In short, opponents held that
political groups did not conform to the definition of genocide and would weaken
the entire Convention.
While not voiced publicly, legal scholar Beth van Schaack also reminds us
of the implicit reality that the Convention could not implicate member nations
on the drafting committee. For instance, if political groups were included in the
Convention, it could “inculpate Stalin’s politically motivated purges of the kulaks
(the petty bourgeois) during the forced collectivizations of agriculture in the late
1920s and early 1930s” (van Schaack 1996, 2268). Clearly, many governments
could have found themselves in a similar position—threatened with charges
of genocide if political identity were to be included as a protected group in the
definition.
The most significant objection to the inclusion of political groups centered,
however, on the issue of state sovereignty. Opponents feared the inclusion of
political groups in the Convention would expose nations to external interven-
tion in their domestic concerns, and that political conflict within a country could
become an international issue. In other words, including political groups in the
Convention would enter into the controversial issue of civil war and inhibit states
that were attempting to suppress internal armed revolt. Countries like Argentina,
Brazil, the Dominican Republic, Iran, and South Africa joined the Soviet dele-
gation and its allies in expressing concern they could be accused of genocide if
they fought against domestic political insurgencies (Naimark 2010, 22). Lemkin
recounts the argument of Ambassador Gilberto Arnado of Brazil: “We in Latin
America make revolutions from time to time, which involves the destruction of
political opponents. Then we reconcile and live in peace. Later the group in power
is thrown out in another revolution. Why should this be classified as the crime of
genocide?” (Frieze 2013, 161).
Ultimately, however, the exclusion of political groups from the listing of pro-
tected groups would come down, ironically, to political compromise. The fear was
simply too great that the passage of the Convention itself would be jeopardized
by the inclusion of political groups. The United States reversed its previous insis-
tence on the inclusion of political groups and its representative, Ernest Gross, gave
voice to the reality behind this decision: “The United States delegation continued
to think that its point of view was correct but, in a conciliatory spirit and in order
to avoid the possibility that the application of the convention to political groups
might prevent certain countries from acceding to it, he would support the pro-
posal to delete from article II the provisions relating to political groups” (United
Nations General Assembly 1948, 662). Most agreed with the need to produce an
instrument that would be acceptable to a large number of member states, and the
inclusion of political groups was simply too contentious to risk. Even Lemkin,
who had procedural doubts about the inclusion of political groups since the initial
deliberations, had reconciled himself with the reality that its inclusion would be
an obstacle to the Convention. In his words: “I thought the destruction of politi-
cal opponents should be treated as the crime of political homicide, not as geno-
cide” (Frieze 2013, 161).
Second, Article II enumerates the five acts of genocide: (a) killing members of
the group, (b) causing serious bodily or mental harm to members of the group,
(c) deliberately inflicting on the group conditions of life calculated to bring about
its physical destruction in whole or in part, (d) imposing measures intended to
prevent births within the group, or (e) forcibly transferring children of the group
to another group. As Article II clearly explains, genocide means “any” of the listed
acts; not “most” and certainly not “all.” An allegation of genocide is supported
when any one of the five acts have been committed with intent to destroy, in whole
or in part, a national, ethnical, racial, or religious group, as such. To say that geno-
cide consists of any one act is to run the risk that other acts are omitted; or to lose
the spirit of Lemkin’s original conception of genocide as a “synchronized attack.”
Eventually, however, the principle of enumeration carried the day.
For a third conceptual issue, we turn to Article VI. Whereas Lemkin’s 1933
proposal had advocated for a principle of universal “repression” or jurisdiction in
which the crime of genocide could be prosecuted by any state, even in the absence
of a territorial or personal link, the drafters of the Genocide Convention explicitly
rejected this principle and recognized only territorial jurisdiction. As Article VI
states: “Persons charged with genocide … shall be tried by a competent tribu-
nal of the State in the territory of which the act was committed.” As van Schaack
(1996, 2266) points out, choosing territorial over universal jurisdiction virtu-
ally guarantees impunity for perpetrators of genocide, because states will rarely
prosecute their own. Article VI went on to suggest that the crime of genocide
may be tried by an “international penal tribunal as may have jurisdiction with
respect to those Contracting Parties which shall have accepted its jurisdiction”
(United Nations 1951). This clause would eventually lead, nearly fifty years later,
to the creation of ad hoc international tribunals to deal with genocides in the for-
mer Yugoslavia and Rwanda as well as the adoption of the Rome Statute of the
International Criminal Court.
2.2.6. Passage and Ratification

Although passed in 1948, the Genocide Convention would become operative in
law only after a sufficient number of domestic ratifications (twenty) by member
states. Lemkin’s work was far from done as he continued to lobby countries for
ratification, often borrowing money to invite representatives to lunch. Meeting

resistance from several major powers, he refocused his attention on small nations
because they “need the protection of international law more than big nations”
(Frieze 2013, 187). Ethiopia was the first country to ratify the Convention on July
1, 1949. Australia followed on July 8, Norway on July 22, and Iceland on August
29. Lemkin also enlisted the support of religious leaders, foreign offices, and
widely read newspapers and magazines in a broader campaign to mobilize popu-
lar support for ratification of the Convention.
Finally, on October 16, 1950, the number of domestic ratifications surpassed
what was needed for the Convention to come into effect. Article XIII of the
Convention stipulated that the Convention “shall come into force on the nine-
tieth day following the date of deposit of the twentieth instrument of ratifica-
tion” (United Nations 1951). At that point—January 12, 1951—the Genocide
Convention became codified in international law, binding on those nations that
signed it. Just months later, the International Court of Justice issued an Advisory
Opinion in which it asserted that the principles underlying the Convention
are binding on all states, including those that had not ratified the Genocide
Convention (Meron 2011). Samantha Power reminds us that the passage of the
Genocide Convention enshrined, at least in principle, a new political reality—
“states would no longer have the legal right to be left alone” (Power 2002, 58).
Political scientist Kathryn Sikkink goes further in viewing the Convention as the
foundation for the development of a new norm—what she calls the “justice cas-
cade”—t hat gave legitimacy and strength to the emerging notion that state offi-
cials, including heads of state, should be held criminally accountable for human
rights violations (Sikkink 2011).
All told, the Genocide Convention stands as one of the most widely accepted
treaties in international relations. As of this writing, 142 states have ratified or
acceded to the Genocide Convention. (Once closed for signature, on January 12,
1951, nonsignatory states now can only accede to the treaty.) The most recent
accession came from Cape Verde on October 10, 2011. Around fifty states are
not yet parties to the Genocide Convention; the Dominican Republic signed the
treaty on December 11, 1948, but has yet to ratify it. On April 16, 2014, twenty
years after the genocide in Rwanda, a unanimously adopted UN Security Council
resolution called for universal ratification of the Genocide Convention by chal-
lenging “States that have not yet ratified or acceded to the Convention … to con-
sider doing so as a matter of high priority” (United Nations 2014).
2.2.7. Lemkin’s Legacy

Throughout the 1950s, Lemkin was repeatedly nominated for the Nobel Peace
Prize, but it was never awarded to him. He did receive the Grand Cross of
Cespedes from Cuba in 1950, the Stephen Wise Award of the American Jewish
Congress in 1951, and the Cross of Merit of the Federal Republic of Germany
in 1955. Despite this minor celebrity status, however, Lemkin faded from
public view after the ratification of the Genocide Convention in 1951. He was
plagued by poor health, particularly high blood pressure, and lived an indi-
gent life marked by piles of unsorted papers, poverty, hunger, and a few moth-
eaten clothes. In the words of Michael Ignatieff, Lemkin “appears to have been
one of Kafka’s hunger artists, those moving, self-punishing creatures who cut
themselves off from the world, preyed upon by a guilt they cannot name, mak-
ing their misery into their life’s work. … His work on genocide finally became
a trap from which he could not—a nd in the end did not wish to—escape”
(Ignatieff 2013, 3).
On August 28, 1959, after a heart attack at a bus stop on Forty-second Street in
New York City, Lemkin was taken to the nearest police station, where he died. He
never lived to see a conviction for the crime that he had given a name. Lemkin, who
once described loneliness as “an essential condition of my life,” passed away as he
had lived for much of his life—a lone (Frieze 2013, 163). The closing sentence in
his New York Times obituary read succinctly: “He was a bachelor” (New York Times,
1959). Having lost most of his family in the Holocaust and alienated many of his
friends over the years with his “self-lacerating obsession,” Lemkin’s funeral would
draw only seven people (Ignatieff 2013, 9). Lemkin is buried in Mount Hebron
Cemetery in Queens, New York, with a headstone that reads, “The Father of the
Genocide Convention.”
Today, however, Lemkin is rightly recognized as one of the heroes in human
rights history. In 2001, on the fiftieth anniversary of the Convention entering
into force, Lemkin was honored by then-U N Secretary-General Kofi Annan as
“an inspiring example of moral engagement” (United Nations 2001). Poland’s
Ministry of Foreign Affairs, under the leadership of Adam Rotfeld, named a con-
ference room after him in 2005 and then mounted a commemorative plaque on the
house in which Lemkin had lived in Warsaw (AJC 2013). Each year, T’ruah: The
Rabbinic Call for Human Rights grants a Raphael Lemkin Human Rights Award
to preeminent leaders in the human rights field. Every other year, the Institute
for the Study of Genocide grants the Lemkin Book Award for the best scholarly
book in genocide studies. Four times a year, I develop the curriculum for, and
teach in, the Raphael Lemkin Seminars for Genocide Prevention. Hosted by the
Auschwitz Institute for Peace and Reconciliation, and held on the grounds of the
Auschwitz-Birkenau State Museum in Poland, these seminars embody Lemkin’s
determination to build a worldwide network of government policymakers com-
mitted to the prevention of genocide.
Lemkin’s story is one of an idea and a word. Due, in large part, to his single-
minded activism, Churchill’s “crime without a name” now has a name, and that
name is “genocide.” Lemkin’s legacy reminds us that words matter; names matter;
labels matter.
2.3. Genocide and Other Mass Atrocity Crimes

The word Lemkin coined—and the act to which he devoted his adult life to defin-
ing it as an international crime—quickly acquired considerable weight. It was a
weight that international political leaders were unwilling to pick up for fear of
being compelled to act. For activists and politicians intent on responding to mass
murder, it was a weight frequently swung as a cudgel of moral judgment or politi-
cal one-upmanship. As Ignatieff said: “Those who should use the word ‘genocide’
never let it slip their mouths, and those who do use the word ‘genocide’ banal-
ize it into a validation of every kind of victimhood” (Ignatieff 2001, 7). “What
remains is not a moral universal which binds us all together, but a loose slogan
which drives us apart” (Ignatieff 1998, 2).
The name itself became what Dirk Moses described as a “Janus-faced keyword,”
having multiple forms for multiple audiences (Moses 2013, 26). Indeed, however
much weight it had assumed, “genocide” took on an ever greater amount of defi-
nitional controversy over the years—as a political, legal, empirical, moral, and
analytical concept. “Genocide” joined concepts such as democracy, justice, rule
of law, citizenship, war, art, morality, nature, and science as what social theorist
W. B. Gallie defined in 1956 as “essentially contested concepts.” For Gallie, these
are “concepts the proper use of which inevitably involves endless disputes about
their proper uses on the part of their users” (Gallie 1956, 169).6 Put differently,
essentially contested concepts are concepts that are understood, and used, incon-
sistently. Such “conceptual blockages” (Moses 2002, 7) are not simply a result of
confusion; rather, they are functional disagreements, often the cause of conten-
tious contestation, that threaten the coherence of research in a field and its appli-
cations. Gallie describes well the nature of these disputes: “Each party continues
to maintain that the special functions which the term ‘work of art’ or ‘democracy’
or ‘Christian doctrine’ fulfils on its behalf or on its interpretation, is the correct
or proper or primary, or the only important, function which the term in question
can plainly be said to fulfil. Moreover, each party continues to defend its case with
what it claims to be convincing arguments, evidence, and other forms of justifica-
tion … although not resolvable by argument of any kind, [such apparently end-
less disputes] are nevertheless sustained by perfectly respectable arguments and
evidence” (Gallie 1956, 168–69).
Fortunately, over nearly seven decades of advances in international and
domestic humanitarian law—as well as the fruit of academic, policy, and civil
society discussions—have helped clarify many of the questions raised by the
original conception of “genocide.” As a result, academics, lawyers and jurists,
policymakers, and civil society increasingly are coming to see value in under-
standing genocide in the broader context of an umbrella conceptual frame-
work of mass atrocities that also includes war crimes, crimes against humanity,
and—in subsumed form—ethnic cleansing. These categories of mass atrocity

are not mutually exclusive. They certainly contain some overlap, as evidenced by
the fact that defendants at international tribunals are often indicted on the same
criminal acts for two, or even all three, of the legal categories of genocide, war
crimes, and/or crimes against humanity. These categories are also, though, indi-
vidually distinguishable in some important conceptual, and practical, features.
2.3.1. War Crimes

War crimes, according to legal scholar Steven Ratner, have been a concept in inter-
national law for many centuries. Tracing limitations on the conduct of armed con-
flict back to the Chinese warrior Sun Tzu (sixth century bce), through the Hindu
code of Manu (200 bce), and on through Roman and European law, Ratner writes
that the “first true trial for war crimes is generally considered to be that of Peter
von Hagenbach, who was tried in 1474 in Austria and sentenced to death for war-
time atrocities” (Ratner 2007, 420). The Hague Conventions of 1899 and 1907
officially regulated categories of combatants and noncombatants as well as the
types of weapons deemed legitimate for warfare. Article 6(b) of the 1945 London
Charter of the International Military Tribunal at Nuremberg defined war crimes
as “violations of the laws or customs of war,” including, but not limited to, “mur-
der, ill-treatment or deportation to slave labour or for any other purpose of civil-
ian population of or in occupied territory, murder or ill-treatment of prisoners of
war or persons on the seas, killing of hostages, plunder of public or private prop-
erty, wanton destruction of cities, towns or villages, or devastation not justified by
military necessity” (Nuremberg Trial Proceedings 1945). Subsequently, each of
the four 1949 Geneva Conventions (on wounded and sick on land, wounded and
sick at sea, prisoners of war, and civilians) included its own list of “grave breaches”
of rules of war, and the 1977 Additional Protocols to the Geneva Conventions
clarified and expanded that list. Building on this corpus of work, the most recent
catalogue of war crimes appeared in the 1998 Rome Statute of the International
Criminal Court (ICC) (International Criminal Court 1998a).
War crimes are serious violations of a body of law known as international
humanitarian law or the law of armed conflict. Some war crimes are prohibited
by treaty, others by customary law; some are prohibited in international conflicts
alone, some in internal conflicts alone, and some in all conflicts (Fenrick 2004, 2).
In terms of individual criminal responsibility, war crimes can be committed by
military personnel against enemy military personnel or civilians, but also by
civilians against enemy military personnel or other civilians. While there is no
comprehensive and complete list of war crimes, Joseph Stefanelli (2014) of the
International Criminal Tribunal for Rwanda (ICTR) suggests that war crimes
include crimes committed against (a) protected persons, including noncom-
batants or those no longer taking part in the conflict (for example, the injured
or prisoners of war); (b) enemy combatants or civilians using prohibited meth-

ods (for example, directly targeting civilians or militarizing a civilian location);
(c) enemy combatants or civilians using prohibited means (for example, chemical
or biological agents); and (d) specially protected persons and objects (including
medical personnel, hospitals and medical transports, humanitarian relief workers
or agencies, and so on). In short, war crimes are a range of prohibited behaviors
that occur in the context of armed conflict.
The distinguishing definitional context is that war crimes only cover atroci-
ties committed during armed conflict. So, war crimes must be perpetrated
either between two states (an international armed conflict) or between two
forces in a civil war (or other internal armed conflict). It is not necessary that
a state of war per se be recognized by either side. Armed conflict is the con-
textual key for war crimes. If there is no armed conflict, there cannot be a war
crime. In a sense, “war crimes” are really more aptly understood as “armed con-
flict crimes.” Riots or isolated, sporadic acts of violence, while they may lead to
criminal behavior, will not evidence—by definition—w ar crimes.
War crimes differ from genocide in three important respects. First, war crimes
need not evidence any intent to destroy a group in whole or in part. Second, the
protective reach of war crimes is expansive and not limited to national, ethnical,
racial, or religious groups. Third, while genocide is a crime that can be commit-
ted in time of peace or in time of war, war crimes are contextually limited only to
times of armed conflict.
2.3.2. Crimes against Humanity

Crimes against humanity is a more recent, twentieth-century concept, originat-
ing in the 1907 Hague Convention preamble. When the phrase was used by the
Allies in 1915 to describe what we know of today as the Armenian genocide, it
had no recognized legal definition. In 1945, however, the London Charter of the
International Military Tribunal at Nuremberg codified crimes against humanity in
Article 6(c) as: “Murder, extermination, enslavement, deportation, and other
inhumane acts committed against any civilian population, before or during the
war; or persecutions on political, racial or religious grounds in execution of or in
connection with any crime within the jurisdiction of the Tribunal, whether or not
in violation of the domestic law of the country where perpetrated” (Nuremberg
Trial Proceedings 1945). As defined at Nuremberg, the concept of crimes against
humanity was created to prosecute atrocities committed within the borders of
Germany against German civilians.
The Allies were worried, however, about the extent to which this new concept
could be applied. In the woefully understated words of Justice Robert Jackson,
“We [the United States] have some regrettable circumstances at times in our
own country in which minorities are unfairly treated” (Schabas 2007, 33). The
British, French, and Russians shared similar concerns. So, the Tribunal decided
to limit the scope of crimes against humanity to wartime. Registering no con-
viction for any criminal act committed prior to Germany’s invasion of Poland
on September 1, 1939, the court confirmed that crimes against humanity could
not—by definition—be committed in peacetime (see Schabas 2006, 94–95).
The limited scope of crimes against humanity established at the initial
Nuremberg trial was frustrating for many involved in the drafting of the Genocide
Convention. As Schabas points out, vulnerable emerging states were particularly
desirous of international criminalization of atrocities in peacetime for their own
protection (Schabas 2006, 96). Over time, this desire would be met. In addition to
gaining some measure of protective satisfaction in the passage of the Convention,
other advances in humanitarian law would also broaden the scope of crimes
against humanity—including in peacetime. While, to date, there has been no
specialized international convention on crimes against humanity, there was, as
Schabas notes, a “dramatic enlargement of the ambit of crimes against human-
ity during the 1990s” (Schabas 2008, 3). Most notably, crimes against humanity
would be included in the statutes of the ICTR and the International Criminal
Tribunal for the former Yugoslavia (ICTY), and case law from both tribunals
would clarify and enlarge the scope of the crime under international human
rights law.
The most recent step in this evolving process of giving legal shape to the crimes
was the codification—and further expansion—of crimes against humanity in the
Rome Statute of the ICC as acts “committed as part of a widespread or systematic
attack directed against any civilian population” (International Criminal Court
1998b, Article 7). The criminal acts listed in Article 7 of the statute include mur-
der, extermination, enslavement, deportation or forcible transfer, imprisonment,
torture, rape and other forms of sexual violence, persecution, enforced disappear-
ance, apartheid, and a catch-a ll category of other inhumane acts that cause great
suffering or serious injury. These acts must be part of a widespread (a matter of
scale) or a systematic (a matter of organization) attack—d isjunctive rather than
conjunctive; either rather than both. There is no statutory limitation for crimes
against humanity. Article 7 contains no reference to armed conflict as a contex-
tual requirement of crimes against humanity.
Crimes against humanity are particularly odious because they constitute an
egregious attack on human dignity. The act injures not just the victim, but tears
at the fabric of what it means to be human. Hannah Arendt captured this sense
of an offense against all humanity in her description of the Holocaust as “a crime
against humanity perpetrated upon the body of the Jewish people” (Wald 2007,
624). Crimes against humanity can be perpetrated, in time of war or time of peace,
by individuals acting in a state capacity (for example, military commanders, sol-
diers, police officers) or by private individuals, not in isolation but with knowledge
that their acts are part of a widespread or systematic attack. The primary victims
protected by international laws against these crimes include all civilian popula-
tions (a government’s own citizens or those of another state), although increasing
protections for military personnel are also being included under the continual
legal evolution of crimes against humanity in customary law (Stefanelli 2014).
Wald notes that “crimes against humanity is a big tent set up on ground that
overlaps both war crimes and genocide” (Wald 2007, 625). Crimes against
humanity differ from the crime of genocide because they do not require intent to
destroy a group in whole or in part. Rather, crimes against humanity must only
target a given group (much broader in reach than the limited protected groups of
the Convention) and be either widespread or systematic. In addition, the crimi-
nal acts included under crimes against humanity are also much broader than the
five acts of destruction criminalized for genocide. In relation to war crimes,
crimes against humanity are distinct because they are not limited to times of
armed conflict; rather, crimes against humanity, like genocide, are now under-
stood to occur in times of peace or in times of war. While crimes against human-
ity carry some additional proof burdens over war crimes, they are not particularly
onerous (Wald 2007, 629).
2.3.3. Ethnic Cleansing

While having no formal legal definition, ethnic cleansing often is included as a
distinct, autonomous category of mass atrocity. Journalists, politicians, and non-
governmental organizations quickly seized on the term as intuitively descriptive—
even the defining characteristic—of the violence that was occurring throughout
the former Yugoslavia. By 1993, William Safire, the noted New York Times colum-
nist, coronated ethnic cleansing as “this generation’s entry in the mass-murder
category.” He went on to assert that ethnic cleansing “has become a major coin-
age, now used without quotation marks or handled with the tongs of so-called”
(Safire 1993, 1). Indeed, by 1994, the UN General Assembly no longer bracketed
the term in quotation marks (Schabas 2009). In academic circles it soon became
in vogue to use the term in reference to a wide variety of geographical expulsions
in the nineteenth and twentieth centuries.
For all of its popular appeal, however, the term offers more heat than light. For
most academics, ethnic cleansing is a distinction without a difference. No concep-
tual singularity is attached to the term that usefully distinguishes it from genocide,
war crimes, or crimes against humanity. For lawyers and jurists, ethnic cleans-
ing remains a nebulous blanket term that can cover a host of criminal offenses
related to forced removal or displacement of civilian populations—all of which are
already codified under other legally recognized mass atrocity crimes. For instance,
Article 7 of the Rome Statute outlining crimes against humanity includes as one of
its listed acts “(d) deportation or forcible transfer of population” and understands
this to mean “forced displacement of the persons concerned by expulsion or other
coercive acts from the area in which they are lawfully present” (International
Criminal Court 1998b). As an example from case law, an ICTY Trial Chamber
found that the forcible displacement of women, children, and elderly people from
Srebrenica amounted to the crime of genocide because it reached a “requisite level
of causing serious mental harm”—even in the absence of intent for physical exter-
mination (International Criminal Tribunal for the former Yugoslavia 2005, para-
graph 650). Finally, in a tragic reversal of intention, we should be wary that ethnic
cleansing not be reduced, for policymakers and civil society, to simply a catchy
emotive phrase—w ithout legal standing—that can end up excusing the inter-
national community from complying with duties laid down by international law
for the recognized crimes of genocide, war crimes, and crimes against humanity
(Petrovic 1994).
2.4. Conclusions
Just as one must carefully parse the word “genocide,” we must carefully dis-
tinguish it from, and relate it to, other forms of mass violence. I see genocide
as one form of mass violence directed at civilian populations. I do not see it as
the “apex of the pyramid,” the “crime of crimes,” or even the general framework
under which all other forms of mass atrocity should be placed. Rather, I see geno-
cide as one of three criminal categories of mass atrocity—joining war crimes
and crimes against humanity under that conceptual framework. Each of these
categories—d istinguishable from each other in important respects and overlap-
ping in others—is a serious violation of international humanitarian law and each
is subject to universal jurisdiction.
Vigilance must be taken with each of these terms because of the different pur-
poses they serve, and the different implications they have, for different audiences.
For academics, a clear understanding of what constitutes mass atrocity is impor-
tant because it allows us to compare and contrast the complex social phenomena of
similar historical events—our universe of cases—that fall within the boundaries
of those definitional classifications. For lawyers and jurists, mass atrocity crimes are
legal categories, focused on the elements of legal culpability, seen as foundational
for international human rights law. For policymakers, a clear understanding of mass
atrocity crimes is important because it animates decision making in response to
crisis. Finally, for civil society, mass atrocity is an activist and mobilization term
with implications for prevention and humanitarian intervention strategies.
We must remember, however, that the terminology employed to describe large-
scale destruction is only one facet of its prevention. Ultimately, naming crimes is
not the same as eliminating them. As legal scholar Martha Minow has argued,
“the problems … will not be cured by new words … renaming legal categories
will do little to address underlying problems of leadership and will … [and] new
names will not undo the reluctance of individuals, nationals, and international
organizations to respond to mass violence” (Minow 2007, 37–38). It is to those
problems and that reluctance to which the work of this volume is dedicated.
Notes
1. For those desiring a more detailed biographical account, I strongly recommend Cooper
(2008) and Korey (2001). Lemkin’s major archival papers are spread across three institu-
tions in the United States—t he Jacob Marcus Rader Center of the American Jewish Archives
in Cincinnati, Ohio (donated 1965); the American Jewish Historical Society at the Center
for Jewish History in New York City (donated 1975); and the Forty-second Street Branch
of the New York Public Library in New York City (donated 1982). There are at least three
versions, one typed and two handwritten, of Lemkin’s incomplete and unpublished autobi-
ography, Totally Unofficial, begun about 1951. The most complete copy was donated to the
New York Public Library in August 1982. There, in the Manuscripts and Archives Division,
Totally Unofficial can be found in Box 1, Folders 35–43 (Accession #83 M 39). While I have
worked often with those materials, I have chosen to reference selections from Lemkin’s auto-
biography from Frieze (2013). For the general reader, this is a much more accessible, and
readable, source than the original papers and is the most current transcription of Lemkin’s
unpublished autobiography.
2. Much of this timeline is drawn from Cooper (2008).
3. Quoted material taken from Frieze (2013, 115–16).
4. New York Public Library, “Raphael Lemkin Papers, 1947–1959,” Reel 2, Box 1, Folder 36,
“Writings—Autobiography,” “Chapters 1–4,” 3.
5. See https://t reaties.un.org/Pages/showDetails.aspx?objid=0800000280027fac (accessed
May 19, 2015).
6. W. B. Gallie (1956) first discussed “essentially contested concepts” at the March 21, 1956
Meeting of the Aristotelian Society at Bedford Square in London. Powell (2007) offers an
insightful analysis of Gallie’s work.
References
[AJC] American Jewish Committee. 2013. http://w ww.ajc.org/site/apps/n lnet/content2.aspx?
c=7oJILSPwFfJSG&b=8479733&ct=1248210 [accessed November 14, 2013].
Balakian, P. 2003. The Burning Tigris: The Armenian Genocide and America’s Response.
New York: Perennial.
Barrett, J. 2010. “Raphael Lemkin and ‘Genocide’ at Nuremberg, 1945–1946.” In C. Safferling
and E. Conze, eds., The Genocide Convention Sixty Years after Its Adoption. The Hague,
Netherlands: Asser Press, 35–5 4.
Breitman, R. 1998. Official Secrets. New York: Hill and Wang.
Butcher, T. 2013. “A ‘Synchronized Attack’: On Raphael Lemkin’s Holistic Conception of
Genocide.” Journal of Genocide Research 15, no. 3: 253–71.
Churchill, W. 2003. Never Give In! The Best of Winston Churchill’s Speeches. New York: Hyperion.
Cooper, J. 2008. Raphael Lemkin and the Struggle for the Genocide Convention. New York: Palgrave
Macmillan.
Earl, H. 2013. “Prosecuting Genocide before the Genocide Convention: Raphael Lemkin and
the Nuremberg Trials, 1945–1949.” Journal of Genocide Research 15, no. 3: 317–37.
Elder, T. 2005. “What You See before Your Eyes: Documenting Raphael Lemkin’s Life by Exploring
His Archival Papers, 1900–1959.” Journal of Genocide Research 7, no. 4: 469–99.
Fenrick, W. 2004. “Crimes in Combat: The Relationship between Crimes against Humanity
and War Crimes.” Guest Lecture Series of the Office of the Prosecutor, The Hague (March
5, 2004).
Frieze, D. 2013. Totally Unofficial: The Autobiography of Raphael Lemkin. New Haven, CT: Yale
University Press.
Gallie, W. 1956. “Essentially Contested Concepts.” Proceedings of the Aristotelian Society (n.s.) 56
(1955–1956): 167–98.
Human Rights Web. 2011. http://w ww.hrweb.org/legal/genocide.html [accessed January
23, 2011].
Ignatieff, M. 1998. “The Treacherous Sands of Time.” Independent, September 5, 1998. http://
www.independent.co.uk/a rts- entertainment/t he-t reacherous- s ands- of-t ime-1196282.
html [accessed December 29, 2015].
Ignatieff, M. 2001. “Lemkin’s Word.” The New Republic, February 26, 2001. http://w ww.newre-
public.com//a rticle/politics/lemkins-word [accessed November 14, 2013].
Ignatieff, M. 2013. “The Unsung Hero Who Coined the Term ‘Genocide.’” The New Republic,
September 21, 2013. http://w ww.newrepublic.com//a rticle/114424/raphael-lemkin-
unsung-hero-who-coined-genocide [accessed November 14, 2013].
International Criminal Court. 1998a. “Rome Statute of the International Criminal Court,
Article 8.” November 10, 1998. http://legal.un.org/icc/statute/romefra.htm [accessed July
28, 2014].
International Criminal Court. 1998b. “Rome Statute of the International Criminal Court,
Article 7.” November 10, 1998. http://legal.un.org/icc/statute/romefra.htm. [accessed
July 28, 2014].
International Criminal Tribunal for the Former Yugoslavia. 2005. Prosecutor v. Vidoje Blagojevic
and Dragan Jockic, Case No. IT-02-6 0-T (January 17, 2005).
Irvin-Erickson, D. 2013. “Genocide, the ‘Family of Mind’ and the Romantic Signature of
Raphael Lemkin.” Journal of Genocide Research 15, no. 3: 273–96.
Korey, W. 2001. An Epitaph for Raphael Lemkin. New York: Jacob Blaustein Institute for the
Advancement of Human Rights.
Lemkin, R. 1933. “Acts Constituting a General (Transnational) Danger Considered as Offences
Against the Law of Nations.” Special Report presented to the Fifth Conference for the
Unification of Penal Law in Madrid.
Proposals for Redress. Washington, DC: Carnegie Endowment for International Peace,
Division of International Law.
Meron, T. 2011. Remarks titled “Enforcing the Genocide Convention.” From a keynote speech
given by the Hon. Judge Theodor Meron, Berlin, Germany (May 12, 2011). Available at
https://w ww.youtube.com/watch?v=IlXlHrfDQXg.
Minow, M. 2007. “Naming Horror: Legal and Political Words for Mass Atrocities.” Genocide
Studies and Prevention 2, no. 1: 37–41.
Mitchell, F. 2012. Ghosts of Ole Miss. ESPN Films. http://espn.go.com/30for30/fi lm?page=ghosts-
of-ole-m iss [accessed May 19, 2015].
Moses, A. D. 2002. “Conceptual Blockages and Definitional Dilemmas in the ‘Racial Century’:
Genocides of Indigenous Peoples and the Holocaust.” Patterns of Prejudice 36, no. 4: 7–36.
Moses, A. D. 2013. “Genocide.” Australian Humanities Review 55: 23–4 4.
Naimark, N. 2010. Stalin’s Genocides. Princeton, NJ: Princeton University Press.
Nersessian, D. 2002. “The Contours of Genocidal Intent: Troubling Jurisprudence from the
International Criminal Tribunals.” Texas International Law Journal 37 no. 2: 231–76.
New York Public Library. n.d. “Raphael Lemkin Papers, 1947–1959.” Reel 2, Box 1, Folder 36.
“Writings—Autobiography.” “Chapters 1–4,” 3.
New York Times. 1959. “Raphael Lemkin, Genocide Foe, Dies.” New York Times, August 30, 1959.
Nuremberg Trial Proceedings. 1945. Vol. 1, Charter of the International Military Tribunal,
August 8, 1945. http://avalon.law.yale.edu/i mt/i mtconst.asp [accessed July 29, 2014].
Petrovic, D. 1994. “Ethnic Cleansing—A n Attempt at Methodology.” European Journal of

International Law 5, no. 1: 342–59.
Powell, C. 2007. “What Do Genocides Kill? A Relational Conception of Genocide.” Journal of
Genocide Research 9, no. 4: 527–47.
Power, S. 2002. “A Problem from Hell”—America and the Age of Genocide. New York: Perennial.
Ratner, S. 2007. “Categories of War Crimes.” In R. Gutman, D. Rieff, and A. Dworkin, eds.,
Crimes of War: What the Public Should Know. New York: W. W. Norton, 420–22.
Rosenthal, A. 1988. “On My Mind: A Man Called Lemkin.” New York Times, October 18,
1988, A31.
Safire, W. 1993. “On Language: Ethnic Cleansing.” New York Times, March 14, 1993. http://
www.nytimes.com/1993/03/14/magazine/on-language-ethnic-cleansing.html [accessed
December 29, 2015].
Schabas, W. 2006. “The ‘Odious Scourge’: Evolving Interpretations of the Crime of Genocide.”
Genocide Studies and Prevention 1, no. 2: 93–106.
Schabas, W. 2007. “Semantics or Substance?” Genocide Studies and Prevention 2, no. 1: 31–36.
Schabas, W. 2008. “Convention for the Prevention and Punishment of the Crime of Genocide.”
http://legal.un.org/avl/pdf/ha/c ppcg/c ppcg_e.pdf [accessed August 1, 2014].
Schabas, W. 2009. Genocide in International Law. 2nd ed. Cambridge, UK: Cambridge University
Press.
Shneider, H. 2010. “Political Genocide in Latin America: The Need for Reconsidering the
Current Internationally Accepted Definition of Genocide in Light of Spanish and Latin
American Jurisprudence.” American University International Law Review 25, no. 2: 313–55.
Sikkink, K. 2011. The Justice Cascade: How Human Rights Prosecutions Are Changing World Politics.
New York: W. W. Norton.
Stefanelli, J. 2014. “War Crimes: Legal Frameworks and Lessons for Prevention.” Paper pre-
sented at a conference sponsored by the Auschwitz Institute for Peace and Reconciliation.
Arusha, Tanzania.
Stein, S. 2005. “Conceptions and Terms: Templates for the Analysis of Holocaust and
Genocides.” Journal of Genocide Research 7, no. 2: 171–2 03.
Strom, A., ed. 2007. Totally Unofficial: Raphael Lemkin and the Genocide Convention. Brookline,
MA: Facing History and Ourselves.
United Nations. 1946. Document A/C .6/SR.22. November 22, 1946. New York: United
Nations.
United Nations. 2001. United Nations Press Release SG/ SM/7842. June 13, 2001.
New York: United Nations.
United Nations. 2014. UN S/R ES/2150. “Threats to International Peace and Security.” Adopted
April 16, 2014. New York: United Nations.
United Nations General Assembly. 1946. Fifty-Fifth Plenary Meeting. UN Document A/R ES/
96 (I). December 11, 1946. New York: United Nations.
United Nations General Assembly. 1948. Sixth Committee Summary Records of Meetings, UN
Document A/C .6/SR.128. November 29, 1948. New York: United Nations.
van Schaack, B. 1996. “The Crime of Political Genocide: Repairing the Genocide Convention’s
Blind Spot.” Yale Law Journal 106, no. 7: 2259–91.
Wald, P. 2007. “Genocide and Crimes against Humanity.” Washington University Global Studies
Law Review 6, no. 3: 621–33.
Washington Post. 1944. “Genocide.” Washington Post, December 3, 1944, B4.
Winter, J. 2013. “Prophet without Honors.” Chronicle of Higher Education, June 3, 2013. Available
at http://chronicle.com/a rticle/a rticle-content/139515.
3
Datasets and Trends of Genocides, Mass

Killings, and Other Civilian Atrocities
C h a r l e s H . A n de rton
3.1. Introduction
This chapter surveys fifteen datasets on genocides, mass killings, and “lower-level”
intentional violence against civilians. Based on selected datasets, I compile a list
of mass atrocities (i.e., genocides and mass killings) perpetrated by states since
1900 and by independent nonstate groups since 1989. The lists reveal that mass
atrocities have been almost as frequent as intrastate wars, and far more numer-
ous than interstate wars, over comparable time periods. I then explore trends in
the number of intentional attacks against civilians in recent decades and compare
and contrast the deadliness of mass atrocity relative to war and terrorism. Also
considered are historical perspectives on a “two-track” development of conflict
datasets and empirical peace research. Along the first track is the pioneering data
work on wars, subwar conflicts, and terrorism developed in the 1960s and 1970s
by the Correlates of War (COW) Project, International Crisis Behavior (ICB)
Project, International Terrorism: Attributes of Terrorist Events (ITERATE),
World Events Interaction Survey (WEIS), and Conflict and Peace Databank
(COPDAB), along with the large number of empirical studies built upon such
datasets. Along the second track is the later development of civilian atrocity
datasets in the 1990s and 2000s by the Political Instability Task Force (PITF),
Uppsala Conflict Data Program (UCDP), events data organizations, and various
scholars, as well as correspondingly recent empirical research on civilian atroci-
ties. Although there have been substantial benefits from the advance of conflict
datasets over the past half-century, the two-track development has had several
negative consequences for empirical research on both wars and civilian atrocities,
which I explore in light of conceptual issues related to war and genocide. The con-
clusion maintains that greater data awareness among scholars can improve future
theoretical and empirical research on war, terrorism, and civilian atrocities.
52
D a t a s e t s a n d Tr e n d s o f G e n o c i d e s 53
3.2. Datasets on Genocides, Mass Killings,

and Lower-Level Civilian Atrocities
3.2.1. Conceptual Issues
I survey datasets on collective political violence against civilians (as distinct from
homicide and mass murder, such as school massacres or mall shootings) that
track intentional civilian killings as distinct from civilian deaths from collat-
eral damage in war or government (or other agent) incompetence. I consider
datasets that track genocides (including politicides), mass killings, and lower-level
civilian atrocities. Genocide is intentional destruction, in whole or in part, of
a specific group of people (see c hapters 1 and 2 for details). For mass killing,
perpetrators do not intend to eliminate the whole group or those targeted for
elimination cannot be clearly identified as part of a specific group (Waller
2007, 14).1 I use the term mass atrocity to encompass both genocide and mass
killing. Lower-level civilian atrocities are attacks against civilians in which
the number killed is relatively small (i.e., less than 1,000 per case or per year).
I include datasets that specialize in intentional civilian attacks and also those
that focus on other conflict types with substantial information on civilian vio-
lence. Due to space limitations I do not review country-specific civilian atroc-
ity datasets, which feature surveys of perpetrators, victims, and/or locations of
attacks. 2 Finally, rather than staking out my own civilian atrocity conceptual-
izations, I present the definitions of the data organizations themselves, which
allow readers to make their own judgments about strengths and weaknesses of
the data. 3
3.2.2. State/Nonstate Group Civilian Atrocity Datasets

Table 3.1 summarizes eight large-sample civilian atrocity datasets in which per-
petrators are states and, in some cases, nonstate groups. The top row shows char-
acteristics such as civilian atrocity type and unit of analysis. First is the Political
Instability Task Force’s geno-politicides (PITF-G); it is the only dataset that
specifically tracks genocides.4 Next are mass killing datasets from Ulfelder and
Valentino (2008) (UV) and Easterly, Gatti, and Kurlat (2006) (EGK). The fourth
dataset is from Rummel (1998), which tracks democides. Next, the Uppsala
Conflict Data Program One-Sided Violence Dataset (UCDP-V) tracks primarily
cases of lower-level violence against civilians (i.e., cases with 25–999 fatalities),
but also includes a few with relatively high fatalities (i.e., 1,000 or more). Sixth is
the Political Terror Scale (PTS), which categorizes states by the level of “politi-
cal terror” they direct within their borders to groups. High-end political terror
includes death squad killings of political foes and intentional killing of civil-
ians. Seventh, the Cingranelli-R ichards (CIRI) Human Rights Dataset includes
Table 3.1 Selected State/Nonstate Group Civilian Atrocity Datasets
Dataset Civilian Atrocity Unit of Sample Fatality Fatality Time Period Data
Type(s) Analysis Scope Criterion Estimates Updates
Available? Likely?
1. Political Instability Task Genocide and States, World None Yes (ordinal 1955–2013 Yes
Force Geno-Politicide politicide Nonstate specified scale)
Dataset (PITF-G) Groups
2. Ulfelder and Valentino Mass killing States World 1,000 or Yes 1945–2006 No
(UV) more per
sustained
case
3. Easterly, Gatti, and Kurlat Mass killing States World None Yes, for 1820–1998 No
Data (EGK) specified most cases
4. Rummel Democide, States, World None Yes 20th and pre- No
one-sided lower- Nonstate specified 20th centuries
level atrocity Groups
5. Uppsala Conflict Data Mass atrocity States, World 25 or more Yes 1989–2013 Yes
Program One-Sided and one-sided Nonstate per year
Violence Dataset lower-level Groups
(UCDP-V) atrocity
6. Political Terror Scale Includes States World None No 1976–2013 Yes
(PTS) high-level specified
terror against
population
7. Cingranelli-Richards Includes one- States World 1–49 for Yes (ordinal 1981–2011 Yes
(CIRI) Human Rights sided lower-level moderate- scale)
Dataset atrocity level killings,
50+ for high-
level killings
(per year)
8. Minorities at Risk Forced Minority World None No 1945–2006 Yes
(MAR) resettlement, Groups specified
systematic
killings, ethnic
cleansing
information on extrajudicial killings and other human rights abuses by govern-

ments within their borders. Finally, the Minorities at Risk (MAR) Dataset tracks
information on minority groups across the world, including those that experience
various forms of severe persecution.
The datasets in Table 3.1 are diverse in regard to civilian atrocity types,
whether they focus only on state or also include nonstate perpetrators, treat-
ment of fatalities, and time period. For datasets that employ a fatality criterion
(i.e., UV, UCDP-V, and CIRI), there are differences in whether fatalities are
counted per case or per year. A common feature of the datasets is their world-
wide scope. Note also that the first five datasets specialize in civilian atrocities
while the last three include civilian atrocities as a subset of broader human rights
information that they track. Table 3.2 presents additional information on the
data sources in Table 3.1 such as websites and/or citations in which additional
details are available including, in some cases, downloadable data, coding manu-
als, and dataset history.
Table 3.2 Definitions and Sources for State/Nonstate Group Civilian

Atrocity Datasets
1.
Political Instability Task Force Geno-Politicide Dataset (PITF-G)
(systemicpeace.org/i nscr/i nscr.htm)
“Genocide and politicide events involve the promotion, execution, and/or implied
consent of sustained policies by governing elites or their agents—or in the case
of civil war, either of the contending authorities—t hat result in the deaths of a
substantial portion of a communal group or politicized non-communal group. In
genocides the victimized groups are defined primarily in terms of their communal
(ethnolinguistic, religious) characteristics. In politicides, by contrast, groups are
defined primarily in terms of their political opposition to the regime and dominant
groups” (Marshall, Gurr, and Harff 2014, 14–15).
2. Ulfelder and Valentino Data (UV) Mass Killings (Ulfelder and
Valentino 2008)
Mass killing is “any event in which the actions of state agents result in the
intentional death of at least 1,000 noncombatants from a discrete group in a period
of sustained violence…. Mass killing events were considered to have begun in
the first year in which at least 100 intentional noncombatant fatalities occurred.
If fewer than 100 total fatalities are recorded annually for any three consecutive
years during the event, the event was considered to have ended during the first
year within that three-year period in which fatalities dropped below 100 per year”
(Ulfelder and Valentino 2008, 2,7; emphasis removed).
(continued)
Table 3.2 (Continued)
3. Easterly, Gatti, and Kurlat Data (EGK) Mass Killings (Easterly, Gatti, and
Kurlat 2006)
In defining mass killing, EGK use the broad definition of genocide from Charny
(1999, 7): “Genocide … is the mass killing of substantial numbers of human
beings, when not in the course of military action against the military forces of an
avowed enemy, under conditions of the essential defenselessness and helplessness
of the victims” (Easterly, Gatti, and Kurlat 2006, 132).
4. Rummel Democides (Rummel 1998 and hawaii.edu/powerkills/M URDER.
HTM)
Democide is the “murder of any person or people by a government, including
genocide, politicide, and mass murder” in which the government is “intentionally
or knowingly reckless and [displays a] depraved disregard for life” (www.hawaii.
edu/powerkills/DBG.CHAP2.HTM; definition downloaded April 29, 2014).
5. Uppsala Conflict Data Program One-Sided Violence Dataset (UCDP-V)
(pcr.uu.se/research/ucdp/datasets/ucdp_one-sided_v iolence_dataset/)
“One-sided violence is the use of armed force by the government of a state or by
a formally organized group against civilians which results in at least 25 deaths.
Extrajudicial killings in custody are excluded” (Pettersson 2012, 2; see also Eck
and Hultman 2007). The dataset presents aggregated information on mass and
lower-level atrocities based on automated events data collection across various
news sources using Virtual Research Associates technology.
6. Political Terror Scale (PTS) (politicalterrorscale.org)
“The PTS measures … violations of physical integrity rights carried out by a state
(or its agents)…. The PTS focuses on state-sponsored killings that take place
outside of the normal judicial setting. The ‘extrajudicial’ executions or killings
include death squad killings of political enemies, unlawful use of lethal force by
police forces (e.g., shooting unarmed suspects), intentional killing of civilians
by security forces during combat, and other arbitrary deprivation of life by state
actors” (Wood and Gibney 2010, 369, 370–71).
7. Cingranelli-Richards (CIRI) Human Rights Dataset
(www.humanrightsdata.org)
The CIRI dataset “contains standards-based quantitative information on
government respect for 15 internationally recognized human rights for 202
countries, annually from 1981–2011” (www.humanrightsdata.org; quote
downloaded December 10, 2013). CIRI reports close to 2,500 country years in
which at least one person was extrajudicially killed and almost 900 country years
in which 50 or more were so killed.
(continued)

8. Minorities at Risk (MAR) Dataset (http://w ww.cidcm.umd.edu/mar/)
The MAR dataset tracks information on discrimination for 284 politically active
ethnic groups around the world. Data related to repression and violence against
civilians include political and economic exclusion, property confiscation, exile/
deportation, forced resettlement, torture, reprisal killings, systematic killings, and
ethnic cleansing.
3.2.3. Events Datasets

Table 3.3 shows selected events datasets that track information on intentional
violence against civilians. What distinguishes the datasets in Table 3.3 from
those in Table 3.1 is that the unit analysis of the former is civilian atrocity events,
whereas the latter focus on aggregate yearly atrocity behavior of states and, in
some cases, nonstate groups. Events data provide a detailed “micro” picture of day-
to-day civilian atrocities as distinct from yearly cumulative “macro” atrocity out-
comes of perpetrators tracked in the datasets in Table 3.1. Moreover, events data
can provide early warning of rising instability within states, including emerging
threats against civilians. As just one example, events data provided to the author
by Virtual Research Associates (VRA) showed that Egypt’s internal stability
began to decline a full month before eruption of the Arab Spring in that country
in late January 2011.
Table 3.3 shows substantial variation in events datasets for sample scope, treat-
ment of fatalities, and time period. Political Instability Task Force Worldwide
Atrocities Dataset (PITF-A) and Konstanz One-Sided Violence Event Dataset
(KOSVED) specialize in civilian atrocities whereas the others include civilian
atrocities as a subset of other conflict/cooperation events between and within
states. Also, PITF-A and VRA are worldwide in scope, and VRA is updated daily
and Armed Conflict Location and Event Data (ACLED) monthly. Table 3.4 pres-
ents further details about events datasets including websites, data sources, coding
methods, and civilian atrocity definitions.
3.2.4. Terrorism Datasets

To date, most genocide and terrorism scholars conducting empirical work have
not viewed terrorism from the perspective of genocide studies. 5 Nevertheless,
many terrorist attacks are intentional violence against civilians. For example,
of the 124,191 incidents reported by the Global Terrorism Database (GTD)
for the period 1970–2 013, many were against primarily civilian targets: pri-
vate citizens and property (30,078 attacks), businesses (16,812), educational
institutions (3,379), journalists and other media personnel (2,271), airports
Table 3.3 Selected Conflict Events Datasets with Civilian Atrocities
Dataset Civilian Unit of Sample Scope Fatality Fatality Time Period Data
Atrocity Type Analysis Criterion Estimates Updates
Available? Expected?
1. Political Instability One-sided Events World 5 or more per Yes 1995–2014 Yes
Task Force Worldwide lower-level incident
Atrocity Dataset atrocity
(PITF-A)
2. Konstanz One-Sided One-sided Events 20 civil wars None specified No Mostly mid- No
Violence Event Dataset lower-level 1990s to early
(KOSVED) atrocity 2000s
3. Armed Conflict Location Includes one- Events Africa, some None specified Yes 1997–2014, Yes
and Event Dataset sided lower- Asian states monthly
(ACLED) level atrocity updates for
select African
nations
4. Event Data on Armed Includes one- Events Sub-Saharan Events with Yes 1990–2009 No
Conflict and Security sided lower- Africa (7 at least one
(EDACS) level atrocity states) fatality
(continued)
Dataset Civilian Unit of Sample Scope Fatality Fatality Time Period Data
Atrocity Type Analysis Criterion Estimates Updates
Available? Expected?
5. Social Conflict in Includes one- Events Africa At least 1 for Yes 1990–2013 Yes
Africa Database sided lower- high-level
(SCAD) level atrocity repression
6. Uppsala Conflict Data Includes one- Events Africa, 1 or more per Yes 1989–2014 Yes
Program sided lower- Middle East, incident
Geo-referenced Event level atrocity Asia
Data (UCDP-GED)
7. Virtual Research Includes mass Events World None specified Yes Mid-1990s to Yes
Associates (VRA) killing and present and
one-sided real-t ime basis
lower-level
atrocity
Table 3.4 Additional Information for Events Datasets
1.
Political Instability Task Force Worldwide Atrocities Dataset (PITF-A)
(eventdata.parusanalytics.com/data.dir/atrocities.html)
Data sources: Agence France Presse, Associated Press, New York Times, and Reuters
Coding method: Manual
Number of event records: About 10,000
Definition of Violence against Civilians: “We define an atrocity as implicitly or
explicitly political, direct, and deliberate violent action resulting in the death of
noncombatant civilians… . We have only coded incidents involving five or more
non-combatant deaths” (PITF-A 2009, 3, 5; emphasis removed).
2. Konstanz One-Sided Violence Event Dataset (KOSVED) (polver.uni-konstanz.
de/en/gschneider/research/kosved)
Data sources: Numerous news sources from around the world
Number of event records: 2,850
Definition of Violence against Civilians: “Violent acts perpetrated by an organized
group, which can be either a rebel organization or government troops, directed
against a group of unarmed non-combatants during, shortly before, or after a
conflict. These acts result in the immediate physical harming or death of more than
one non-combatant” (Bussmann and Schneider 2012, 3; see also Schneider and
Bussmann 2013).
3. Armed Conflict Location and Event Dataset (ACLED) (acleddata.com)
Data sources: Numerous news sources and other datasets listed in the data bank file
at the ACLED website
Number of event records: About 23,000 for the violence against civilians dataset and
more than 80,000 for all conflict events
Definition of Violence against Civilians: “Violence against civilians is defined as
deliberate violent acts perpetrated by an organized political group such as a rebel,
militia or government force against unarmed non-combatants” (Raleigh and Dowd
2015, 13).
4. Event Data on Armed Conflict and Security (EDACS) (conflict-data.org/
edacs)
Data source: Numerous news sources and other datasets from around the world
Coding method: LexisNexis searches of keywords from news sources and manual
coding of data
Number of event records: About 7,800
Definition of Violence against Civilians: “We define one-sided attacks as direct
unilateral violence by organized groups aimed at civilian or military targets”
(EDACS 2012, 4).
(continued)
5. Social Conflict in Africa Database (SCAD) (du.edu/korbel/sie/research/
hendrix_scad_database.html)
Data sources: Associated Press and Agence France Presse
Coding method: LexisNexis searches of keywords from news sources and manual
coding of data
Number of event records: More than 10,300
Definition of Violence against Civilians: “Distinct violent event waged primarily by
government authorities, or by groups acting in explicit support of government
authority, targeting individual, or ‘collective individual,’ members of an alleged
opposition group or movement” (Salehyan and Hendrix 2012, 3).
6. Uppsala Conflict Data Program Geo-referenced Event Data (UCDP-GED)
(ucdp.uu.se/ged)
Data sources: UCDP’s armed conflict, nonstate conflict, and one-sided violence
datasets
Coding method: Manual and automated scripts to check the data
Number of event records: 103,665
Definition of Violence against Civilians: “An incident where armed force was used by
an organized actor … against civilians resulting in at least 1 direct death” (Croicu
and Sundberg 2015, 9, emphases removed).
7. Virtual Research Associates (VRA) (vranet.com)
Data sources: Agence France Presse and Reuters
Coding method: Computer-coded using specialized VRA software
Number of event records: Extensive historical and real-t ime data across about
200 countries
Definition of Violence against Civilians: VRA tracks information on states’
repressiveness (SR) against civilians, civil contentiousness (CC), and
civil violence (CV). A country’s stability index (CS) is calculated as
æ CC 2 × SR2 ö÷
CS = 1 − çç ÷÷ × CV .
çè 2 ø
and aircraft (1,242), and tourists (427). Furthermore, the September 11, 2001
attack by al Qaeda against the United States fits Ulfelder and Valentino’s (UV)
definition for mass killing noted in Table 3.2 and also appears in the UCDP-V
dataset. Anderton and Carter (2011, 26, 28–29) provide detailed information
on five well-k nown terrorism datasets, including the GTD that is used in data
analyses below.
3.3. State-and Nonstate-Perpetrated Mass

Atrocities
Appendices A1 and A2 respectively list mass atrocities perpetrated by states
and independent, nonstate groups. The lists are created from the PITF-G ,
UV, EGK, UCDP-V, and PITF-A datasets summarized in Tables 3.1–3 .4.
Appendix A1 shows state-perpetrated mass atrocities from 1900 to 2013.
Appendix A2 shows mass atrocities perpetrated by independent, nonstate
groups from 1989–2 013. The appendices also show genocides identified by
the PITF-G dataset for the period 1956–2 013. Many state-perpetrated mass
atrocities in Appendix A1 are reported by multiple data sources. Hence,
one cannot add up the number of rows in Appendix A1 to obtain a count of
state-perpetrated mass atrocities. Furthermore, some state-perpetrated cases
also involve independent, nonstate group mass atrocities; hence, some cases
appear in both appendices. Further details about the lists are available at the
end of the appendices.
While many implications can be drawn from the appendices, I emphasize
five. First, from 1900 to 2013, the number of distinct mass atrocity cases per-
petrated by states (201 can be identified in Appendix A1) is close to the num-
ber of intrastate wars (237) and far exceeds the number of interstate wars (66)
reported by the Correlates of War (COW) Project and the Uppsala Conflict Data
Program (UCDP) over the same 1900–2 013 period. Hence, mass atrocities are
not extremely rare events. Second, more than 93 percent of state-perpetrated
mass atrocities in Appendix A1 occurred during periods of intrastate war.
Furthermore, 86 percent of COW intrastate wars since 1900 overlapped during
one or more calendar years with an associated state-perpetrated mass atrocity.
Hence, war and mass atrocity are often coterminous (see section 3.5 for further
discussion). Third, for a few states, mass atrocity is the norm rather than an
exception. For example, since becoming a state in 1956 (according to COW’s
state member dataset), Sudan experienced mass atrocity in 46 of its subsequent
57 years. Similarly, since 1948 North Korea experienced mass atrocity in essen-
tially all of its subsequent years. Fourth, estimated fatalities for some cases are
truly staggering. For example, about a dozen distinct cases show estimated
fatalities of one million people or more and about another three dozen report
estimated fatalities of 100,000 people or more. Finally, many nonstate groups
acting independently of explicit or implicit state endorsement have perpetrated
mass atrocities. Hence, to the extent that genocide prevention efforts focus on
states (e.g., the Responsibility to Protect), important perpetrators of mass atroc-
ity may be overlooked. 6
3.4. Civilian Atrocity Trends and Comparative

Measures of Seriousness
This section focuses on state-perpetrated mass atrocities from Appendix A1
(based on the PITF-G, UV, EGK, UCDP-V, and PITF-A datasets) and lower-level
civilian atrocities (from the PITF-A and civilian attacks within GTD datasets) to
highlight trends in civilian atrocities and the seriousness of mass atrocity relative
to war and terrorism. To avoid double-counting mass atrocity cases and to arrive
at (rough) estimates for fatalities per case, I carefully treat overlaps across datasets
as one case and I take the average of fatality estimates reported by the datasets.
For example, PITF-G, UV, and EGK report mass atrocity for Algeria in 1962 with
estimated fatalities of 48,000, 90,000, and 81,000, respectively. I treat this as one
mass atrocity in 1962 with estimated fatalities of 73,000 (the average of the three
fatality numbers).
3.4.1. Civilian Atrocity Data Trends

Figure 3.1 shows the number of states perpetrating genocides and mass atroci-
ties (genocides and mass killings) per year over the period 1956–2 013 and
1900–2013, respectively. Shown in Figure 3.2 is the number of civilian atroc-
ity incidents (mostly “lower-level”) reported by PITF-A from 1995 to 2013 and
the number of GTD terrorist attacks worldwide from 1970 to 2013 against six
types of civilian targets.7 In the more than fifty years covered by the genocide
data there were one or more genocides in the world each year except 2012. For
the more than 100 years spanned by the mass atrocity data there were two or
more mass atrocities in the world each year. The average number of genocides,
mass atrocities, lower-level atrocities, and terrorist attacks against civilians
per year for the time periods shown are 4.8, 15.7, 475, and 1,243, respectively.
While genocides and mass atrocities have trended downward in recent years,
lower-level atrocities and terrorist attacks against civilians appear to be trend-
ing upward.
3.4.2. Seriousness of Mass Atrocities Relative to Wars

and Terrorism
The seriousness of mass atrocities relative to other forms of conflict is shown
in Table 3.5. Since 1900, state-perpetrated mass atrocities have been about 2.7
times more severe than interstate wars in total estimated fatalities. Estimated
fatalities per case are about the same for mass atrocities and interstate wars.
Note, however, that World Wars I and II skew estimated fatalities for interstate
No. of Genocides
and Mass Atrocities
40
35
Mass Atrocities
(Genocides and
30 Mass Killings)
25
20
15
10
Genocides
5
10
70
75
80
85
90
95
00
05
00
05
10
15
20
25
30
35
40
45
50
55
60
65
15
19
19
19
19
19
20
20
20
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
20
No. of Genocides No. of Mass Atrocities (Genocides and Mass Killings)
Figure 3.1 Number of genocides and mass atrocities by year, 1956–2013 and
1900–2013.
No. of Atrocity Incidents and

Terrorist Attacks against Civilians
GTD Terrorist
5000
Attacks against
Civilians
4500
4000
3500
3000
2500
2000
PITF-A
Atrocity
1500 Incidents
1000
500
0
70
72
74
76
78
80
82
84
86
88
90
92
94
96
98
00
02
04
06
08
10
12
14
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
20
20
20
20
20
20
20
20
No. of PITF-A Atrocity Incidents No. of GTD Terrorist Attacks against Civilians
Figure 3.2 Number of PITF-A atrocity incidents and GTD civilian terrorist attacks by
year, 1995–2013 and 1970–2013. Note: GTD data for 1993 are missing.
Table 3.5 Comparative Measures of Seriousness for State-Sponsored Mass Atrocities (Genocides and Mass Killings), Intrastate and
Interstate Wars, and Terrorism
Conflict Type Number of Distinct Time Period Total Estimated Estimated Fatalities
Cases Fatalities for the Cases Per Case
Mass Atrocities 201 1900–2013 83,659,568 469,998a
Interstate Wars 66 1900–2013 30,698,060 465,122
Interstate Wars Excluding WW I and WW II 64 1900–2013 5,485,122 85,705
Intrastate Wars 237 1900–2013 5,576,677 27,201b
Terrorism (Domestic and International) 124,191 1970–2013c 263,643 2
Sources: For Mass Atrocities: Political Instability Task Force geno-politicide dataset (PITF-G) (Marshall, Gurr, and Harff 2014); Ulfelder and Valentino (UV) (2008);
Easterly, Gatti, and Kurlat (EGK) (2006); Uppsala Conflict Data Program One-Sided Violence Dataset (UCDP-V) (Eck and Hultman 2007); Political Instability Task
Force Worldwide Atrocities Dataset (PITF-A 2009). For UCDP-V and PITF-A data, mass atrocities were based on the UV definition of mass killing. For Wars: Sarkees
and Wayman (2010) and COW (www.correlatesofwar.org/[.]) for 1900–2 007; Uppsala Conflict Data Program Armed Conflict Dataset (Gleditsch et al. 2002; Themnér
and Wallensteen 2014); and UCDP Battle-R elated Deaths Dataset, Version 5.0 (http://w ww.pcr.uu.se/[.]) for 2008–2 013. For Terrorism: National Consortium for the
Study of Terrorism and Responses to Terrorism (2015).
a
23 of the mass atrocity cases have no reported fatalities. Hence, estimated fatalities per case is calculated as total estimated fatalities divided by 178.
b
30 intrastate wars have no reported fatalities and two report 550 and 9 fatalities. Estimated fatalities per case is calculated as (total estimated fatalities –550 –
9) divided by (237 –30 –2).
c
Data for 1993 are missing.
wars; removal of the two wars in row 3 shows mass atrocities to be more than
five times more severe than interstate wars on a per-case basis. For intrastate
wars, mass atrocities are about 15 times more severe in total estimated fatalities
and about 17 times more serious on a per-case basis. Furthermore, estimated
fatalities totaled over only three genocides, based on PITF-G data (Cambodia
1975–1979, Pakistan/Bangladesh 1971, and Sudan 1983–2 002), surpasses
total estimated military fatalities for the 237 intrastate wars fought from 1900
to 2013. Based on GTD data (see row 5 of Table 3.5), international and domes-
tic terrorism incidents worldwide over the period 1970–2 013 led to estimated
fatalities of 263,643. Hence, during the 1994 Rwandan genocide, almost as
many people were killed in one month (estimated to be 241,750) as died in all
worldwide international and domestic terrorist incidents over a recent forty-
three-year period. 8
3.5. Two-Track Development of Conflict Datasets

In the history of quantitative peace research, datasets on war and terrorism
emerged in the 1960s and 1970s along one track while datasets for civilian atroci-
ties developed much later (in the 1990s and 2000s) along a second track. There
were good reasons for conflict datasets to develop along the two tracks, but there
were also some potentially negative consequences regarding how quantitative
researchers tended to conceptualize and study war and civilian atrocities.
3.5.1. Examples of the Two-Track Development

of Conflict Datasets
An important impetus for quantitative peace research was the Correlates of
War (COW) Project’s creation of large-sample datasets on wars between and
within states (interstate and intrastate wars), between states and external non-
state groups such as colonies (extrastate wars), and, later, between nonstate
groups (nonstate wars). The COW also created data for interstate subwar con-
flicts known as militarized interstate disputes. The Project’s work began in
the 1960s and fostered an explosive growth of empirical research on war and
subwar conflict that continues to this day. This work has been truly valuable
in promoting greater research into violent conflicts and how they can be pre-
vented. Nevertheless, COW tended to set aside intentional killing of civilians
from its conception of war and, correspondingly, from its datasets (Anderton
2014). For example, Sarkees and Wayman (2010) describe the treatment of
civilian fatalities in COW’s coding protocol for war as follows: “Yet, not all tak-
ing of human life is war. A murder of an individual or a slaughter or massacre
of civilians would not be considered as fatalities from sustained combat” (40).
Excision of civilian fatalities is most obvious in COW’s coding of intrastate

wars, which follows an “effective resistance criterion,” whereby the weaker
party “is able to inflict upon the stronger opponents at least five percent of
the number of fatalities it sustains” (Small and Singer 1976, 67). According to
Sarkees and Wayman (2010, 66), “the effective resistance criterion was specifi-
cally utilized to differentiate civil wars from massacres.”
The Uppsala Conflict Data Program/ Peace Research Institute Oslo
(UCDP/PRIO) armed conflict dataset, which emerged in the early 2000s,
tracks interstate, intrastate, and extrastate wars and subwar conflicts. Unlike
COW, UCDP/PRIO includes in its coding protocols civilian fatalities that are
the direct result of war or subwar conflict involving a state. But like COW,
UCDP/PRIO excludes intentional large-scale killings of civilians. For exam-
ple, in describing the UCDP/PRIO armed conflict dataset, Gleditsch et al.
(2002, 626) state:
“The massacres carried out by Hutu militias and Hutu civilians in 1994
are often estimated as having resulted in 500,000–800,000 deaths. The
deaths are not classified as battle-related and are not included in this
study” [Wallensteen and Sollenberg 1997, 353n2]. Thus, Rwanda in
1994 is listed as having only an intermediate conflict (and in the three
subsequent years, no conflict at all).
In the post-Holocaust world in which quantitative research on peace and war

emerged, the Holocaust and other mass atrocities appeared as something unlike
war, even as phenomena that contradict the notion of war. To the extent that war
is conceptualized and coded only as violence between combatants who are able
to put into the fighting a degree of military resistance, mass atrocities tend to be
treated as something other than war.
3.5.2. The Two-Track Development and the Genocide Gap

in Defense and Peace Economics
The separation of datasets for war from those for mass atrocities in the historical
development of conflict datasets is not necessarily a problem because distinctions
in conflict types can be important, and scholars can always explore connections
between them. But what actually happened was that mass atrocities were “set
aside” in the early decades of quantitative peace research. Hence, mass atrocity
received a scant amount of attention and delayed development of quantitative
research relative to war and terrorism.9 This is especially true, as noted in the
introductory chapter, regarding the “genocide gap” in defense and peace econom-
ics.10 Economic analysis of conflict took off in the 1960s and 1970s alongside the
emerging conflict datasets on wars and terrorism. As Anderton (2014, 119) notes,
to test their theoretical models, defense and peace economists “went with the
available datasets, which excluded genocides and mass killings. The separation
of wars from mass atrocities in the most influential datasets in quantitative peace
research meant that genocide and mass killing tended to be ignored within …
[defense and peace] economics from its inception.”
3.5.3. The War/Mass Atrocity Nexus

Potential problems associated with the two-track development of conflict datasets
run deeper than the relative lack and late development of quantitative research on
mass atrocities. Defense and peace economists and quantitative peace research-
ers more generally tend to view wars and mass atrocities as separate phenomena
rather than as interrelated aspects of violent human behavior. But most mass
atrocities occur in the context of wars and are interdependent with wars. For
example, 147 distinct state-perpetrated mass atrocity onsets can be identified in
Appendix A1 over the period 1946–2007. Lining up these mass atrocities with
COW intrastate war data (that run until 2007), 117 have a time overlap with a
corresponding intrastate war. Of the 30 that do not overlap, 10 have a time overlap
with a UCDP/PRIO intrastate war and an additional three have a similar overlap
with a PITF ethnic or revolutionary war.11 Hence, 88 percent (130 of 147) of state-
perpetrated mass atrocities in Appendix A1 over the period 1946–2007 occurred
in the context of intrastate war as coded by COW, UCDP/PRIO, and PITF. But
what is even more intriguing than these time overlaps is that on a calendar-year
basis a large proportion of mass atrocities begin simultaneously with or even
before the start of war. Specifically, of 117 cases of mass atrocity that line up with a
COW intrastate war, 24 (21 percent) begin after the onset of war, but 55 (47 per-
cent) begin simultaneously with war and 38 (32 percent) begin before the onset
of war. Mass atrocity can be an outgrowth of war, but it seems more common for
mass atrocity to occur alongside or even before war. Hence, wars and mass atroci-
ties may be interconnected in ways that quantitative research on war and peace
has tended to overlook.12
The data support the observation of some genocide scholars that war and geno-
cide cannot be easily separated. For example, Lemkin (1944, 80) distinguished
war and genocide. War involved attacks against states and armed forces, whereas
genocide represented destruction of populations. But Lemkin also character-
ized Nazi genocide as part of its program to wage a total war in which occupied
nations would be weakened and destroyed and, ultimately, superseded by the
German nation (Lemkin 1944, 81). As Lemkin noted, many ancient wars were
wars of extermination in which mass atrocities were features of war. Hence, the
Nazi program of total war can be viewed as a throwback to ancient war, but with
one important difference: it was ancient war modernized, that is, total war with
modern military machines, revolutionary tactics and strategies, and innovations

in techniques and technologies of mass atrocity.
Some genocide scholars distinguish war and genocide to the point of catego-
rizing them as separate phenomena, but others, like Lemkin, emphasize sym-
bioses between them. For example, Levene (2005, 51) notes: “The whole thrust
of Lemkin’s conceptualization … [of genocide] suggests a phenomenon which
does not simply take place within a war context but is itself a form of warfare.”
Furthermore, Shaw (2007, 111–12) proposes the concept of degenerate war in
which armed forces target civilians as means of weakening armed opposition
(see also Shaw 2003). According to Shaw, city bombing during World War II and
intrastate attacks against civilians to undermine support for state or rebel mili-
tary forces are forms of degenerate war. While Shaw notes that genocide seems to
go beyond even degenerate war, he is not willing to completely separate war and
genocide; instead his main objective is to understand interconnections between
them. He states:
The most striking fact about the process that produced the [UN
Genocide] Convention was its separation of genocide from war. In one
sense, this was entirely valid: clearly genocide was not ordinary warfare
and it was conceivable that it could occur outside pre-existing contexts
of warfare. However, the major commonly recognized instances of geno-
cide . . . have been clearly connected with war contexts, and this is an
overwhelming empirical trend. Thus the legal separation of genocide
from war left unresolved the more general conceptual questions: what
are the connections of war and genocide in terms of their meanings, and
in terms of causation? (Shaw 2007, 28, his emphases)
From the perspective of quantitative civil war research, Sambanis (2004,

815–16) also questions the separation of war from genocide and other forms
of conflict:
Although a core set of “ideal” cases of civil war may exist, too many cases
are sufficiently ambiguous to make coding the start and end of the war
problematic and to question the strict categorization of an event of politi-
cal violence as a civil war as opposed to an act of terrorism, a coup, geno-
cide, organized crime, or international war. In the end, it may be difficult
to study civil war without considering how groups in conflict shift from
one form of violence to another, or it may be profitable to analyze politi-
cal violence in the aggregate, rather than cut across that complex social
phenomenon with arbitrary definitions.
3.5.4. Symbioses between War and Mass

Atrocities: Implications for Empirical Research
The two-track development of conflict datasets and conceptual issues about war
and genocide raised by Lemkin, Levene, Sambanis, Shaw, and others are likely
to have important implications for large-sample empirical research on both mass
atrocities and wars. Of the 27 published large-sample (cross-country) empirical
studies of civilian atrocity risk and seriousness cited in Anderton’s (2014) supple-
mentary web appendix, 24 include a measure of conflict, or begin with a sample
of conflicts, to assess correlates of civilian atrocity risk or seriousness. Hence,
war and/or subwar conflict is recognized as a fundamental context in empirical
research on mass atrocities. But the converse does not hold; empirical studies of
civil war rarely consider the possibility that civilian atrocity could be a risk factor
for war. For example, of the 31 large-sample empirical studies of civil war risk
cited in Hoeffler’s (2012) survey article, none control for or incorporate measures
of mass atrocity. I suspect that the asymmetry is rooted, at least in part, in the
historical two-track development of conflict datasets. Recently, Esteban, Morelli,
and Rohner (2015) have taken important steps in theoretically and empirically
analyzing symbioses between war and mass atrocity. Specifically, their formal
game-t heoretic model and empirical tests treat war initiation, war intensity, and
mass killing as jointly determined activities.
3.6. Conclusions
Numerous large-sample and events datasets on intentional violence against
civilians have appeared within the last two decades, which has fostered new
empirical research on civilian atrocity risk and seriousness and allowed schol-
ars to compare and contrast the frequency and seriousness of mass atrocities
relative to war and terrorism. Nevertheless, dataset development and empiri-
cal research for wars and terrorism on the one hand, and for civilian atrocities
on the other, have occurred essentially along two distinct and mostly sepa-
rate tracks, which continue the tradition of quantitative peace research that
wars and intentional civilian atrocities are separate phenomena. This two-
track development of conflict datasets and corresponding quantitative peace
research was not arrived at through theoretical and empirical inquiry. It seems,
rather, to have been assumed rather than scrutinized.
As civilian atrocity datasets become well established, quantitative research
into the risks, seriousness, and prevention of civilian atrocities can be expected
to flourish. Moreover, greater data awareness among scholars of peace and war
and the development of combined datasets of conflict types (e.g., war and mass
atrocity) should promote new quantitative research on interconnections and
distinctions between and among war, terrorism, and civilian atrocities. Such
awareness would help close the “genocide gap” in defense and peace economics
and promote fruitful interactions and integrative work among scholars of war, ter-
rorism, mass killing, and genocide.
Acknowledgments
I am grateful to Jurgen Brauer, Kristian S. Gleditsch, Belen Gonzalez, and
Mansoob Murshed for helpful comments on earlier drafts and to Joe Bond for
early warning data from VRA on instability in Egypt. I alone am responsible for
any errors and omissions.
Appendix A1
Selected Mass Atrocities (Genocides and Mass Killings)
Perpetrated by States, 1900–2013
State Time Estimated Source Source Notes (if any,

Period Fatalities annotated) [My own
notes are in square
brackets]
Afghanistan 1880–1901 2,000 EGK Killings during
unification by Abdur
Rahman.
1978–1992 1,176,000 PITF-G Political purges,
systematic terror,
destruction of villages,
and execution of
prisoners.
1978–1992 1,400,000 UV
1978–1992 1,750,000 EGK Mass murder of
supporters of old regime
and rural supporters of
rebels.
1992–1996 6,250 UV Rabbani/Massoud vs.
Taliban et al.
1996– 7,500 UV Taliban vs. United
2001 Front.
(continued)
Appendix A.1 Continued

notes are in square
brackets]
1998–2001 6,513 UCDP-V
1998– 3,338 PITF-A
2002
Albania 1944–1985 7,500 UV Political repression.
Algeria 1962 48,000 PITF-G Algerian militants attack
Europeans and Muslim
civilians who collaborated
with the French.
1962 90,000 UV Postindependence
retribution.
1962 81,000 EGK Mass murder of Harkis
(French-Muslim troops)
and OAS supporters.
1991–2005 37,500 UV Islamists.
1992–1998 70,000 EGK Killing of civilians by
Islamic fundamentalists.
Angola 1975–1994 546,000 PITF-G UNITA rebels and
MPLA-led government
forces perpetrate
atrocities against
civilians.
1980–1990 EGK Massacre of indigenous
group: San.
1998– 120,000 PITF-G Contending forces target
2002 civilian populations in
their attempts to gain
tactical advantages.
1975– 217,500 UV Civil war.
2002
Argentina 1976–1980 12,000 PITF-G Military stages
coup. Death squads
target subversives for
kidnappings, torture,
and murder.
(continued)

notes are in square
brackets]
1976–1980 19,500 EGK Mass murder of leftists.
1976–1983 19,500 UV
Australia 1824–1908 10,000 EGK Removal/k illings of the
Aborigines.
Azerbaijan 1991–1994 1,500 UV Nagorno-K arabakh.
Bangladesh 1979–1999 EGK Massacres of indigenous
groups: Tribals.
1980–1997 2,250 UV Chittagong Hill
insurgency.
Brazil 2003– 1,306 PITF-A
2007
Bulgaria 1944–1956 55,138 UV Political repression.
Burundi 1965–1973 129,750i PITF-G 1965: Hutu leaders
killed after Tutsi
massacre. 1972: Hutu
massacre Tutsi, Tutsi
regime kills Hutus.
1965–1973 150,000i UV
1988 12,000 PITF-G Tutsi-dominated army
massacres Hutus.
1988– 175,000 UV
2005
1993 60,000i EGK Massacres during Hutu-
Tutsi conflicts.
1993 48,000i PITF-G Three waves: Tutsi
soldiers against Hutu
civilians, Hutus against
Tutsis, Tutsi against
Hutus.
1995– 14,581 PITF-A
2004
1995– 4,159 UCDP-V
2004
(continued)

notes are in square
brackets]
Cambodia 1967–1975 142,500 UV Civil war.
1975–1979 2,700,000a PITF-G Khmer Rouge policies
lead to mass deaths by
deprivation, executions,
and massacres.
1975–1979 1,500,000 EGK Massacre of ethnic
Vietnamese,
intellectuals, middle
class.
group: Cham.
1975–1979 1,500,000 UV Khmer Rouge.
Canada 1500s– EGK Massacre of indigenous
1900 peoples.
Central 2013 1,500i PITF-G Muslim community
African Rep. targeted in fighting
among Seleka rebel
forces and anti-Balaka
militia groups.
2013 1,188 UCDP-V
Chad 1982–1990 26,000 UV Political repression/civil
war, Habre regime.
1991–2003 1,500 UV Political repression/civil
war, Deby regime.
Chile 1973–1976 18,000 PITF-G Supporters of former
regime and other leftists
are disappeared, exiled,
and executed.
1973–1976 16,000 EGK Mass murder of
leftists.
1973–1978 3,200 UV
1986 EGK Massacre of indigenous
group: Mapuche.
(continued)

notes are in square
brackets]
China 1920s– 10,000,000 EGK Killing by nationalists
1949 before final defeat on
mainland.
1947 25,000 EGK Mass murder of
Taiwanese nationalists
[by Kuomintang-led
Chinese government].
1947 22,500 UV
1949–1956 4,500,000 EGK Murder of landlords and
rich peasants during
land reform; murder of
Kuomintang cadre.
1949–1977 35,500,000 UV Communist.
1954–1977 375,000 UV Tibet.
1959 96,000 PITF-G Suppression of
groups: Tibetan
Buddhists, landowners,
supporters of former
regime.
1959–1979 1,200,000 EGK Occupation of Tibet.
1966–1975 625,000 EGK Cultural Revolution
victims.
1966–1975 480,000 PITF-G Wide spectrum of
society targeted for
arrest, harassment,
reeducation, torture,
and execution.
1989 2,600 UCDP-V
Colombia 1948–1958 100,000 UV La Violencia.
1948–1958 180,000 EGK Massacres by
liberal/conservative
governments.
1965– 50,000i UV FARC, ELN, etc.h
2006 b
(continued)

notes are in square
brackets]
1999– 1,000i EGK Killings by right-w ing
2000 paramilitary groups of
alleged sympathizers of
left-w ing guerrillas.
Congo 1960–1963 5,000 UV Kasai.
1964–1965 3,500 UV CNL/Simbas.
1992–1997 5,000 UV Congo Brazzaville/
Republic of Congo,
Lissouba regime.
1997–1999 1,128 UCDP-V
1997–2003 4,000 UV Congo Brazzaville/
Republic of Congo,
Sassou regime.
Côte d’Ivoire 2000– 1,588 PITF-A
2005
Croatia 1993–1995 EGK Killings of Muslim and
Serbian civilians during
Bosnia war (see also
Yugoslavia below).
Cuba 1959–1970 6,668 UV Castro—political
repression.
Czechoslovakia 1945–1946 24,500 UV Expulsion of Germans.
1948–1963 18,030 UV Political repression.
Democr. Rep. 1964–1965 6,150 PITF-G In state consolidation,
of Congo [former] rebels
massacre Congolese,
missionaries, and other
Europeans.
1977–1979 14,150 PITF-G Episodic rebellions
countered by killings
of political opponents,
dissidents, and
prisoners.
(continued)

notes are in square
brackets]
1977–????c 3,500 EGK Mass murder of tribal
and political opponents
of Mobutu.
1993–1997 9,000 UV Kabila/Tutsi vs.
Mobutu.
1996–2012 9,386 UCDP-V
1997– 3,993 PITF-A
2004
1998– 100,000 UV
2006 b
Dominican 1937 17,500 EGK Massacre of
Republic Haitians and black
Spanish-speakers.
1965–1978 4,500 UV Civil war.
El Salvador 1932 30,000 EGK Government repression
of Indians (Pipil) and
peasants.
1977–1992 55,000 UV
1980–1989 49,450 PITF-G Military, security units,
and death squads kill and
harass suspected leftists.
1980–1992 45,000 EGK Mass murder of leftists.
Equatorial 1969–1979 31,300 PITF-G An unsuccessful coup
Guinea attempt triggers a
violent crackdown on all
political opposition.
1969–1979 25,500 EGK Mass murder of Bubi
tribe and political
opponents of Macias.
1969–1979 45,000 UV
Ethiopia 1961–1991 190,000 UV Eritrea, civil war.
1974–1979 30,000 EGK Mass murder of political
opposition.
(continued)

notes are in square
brackets]
1974–1991 250,000 UV Political repression by
Dergue, Tigre civil war.
1977–1985 50,000 UV Ogaden.
1976–1979 13,500 PITF-G Army, security units,
and civilian defense
squads massacre those
opposed to the regime.
1989–1990 1,058 UCDP-V
2003 1,137 PITF-A
France 1945 25,750 EGK Reprisals from colonial
authorities after attack
on European settlers in
Algeria.
Malagasy nationalists.
1955 12,000 EGK Reprisals from colonial
authorities after attack
on European settlers in
Algeria.
Georgia 2008 1,692 PITF-A
Germany 1900–1918 132,000 EGK Colonial massacres
(including 65,000
Herero between 1904
and 1907).
1933–1945 9,939,850 EGK Jews, Poles, Roma,
disabled people,
homosexuals,
communists, Jehovah’s
Witnesses.
Guatemala 1954–1996 7,500 UV Civil war and
repression.
indigenous group
(Maya) and leftists.
(continued)

notes are in square
brackets]
1978–1990 71,400 PITF-G Attacks against
suspected leftists and
indigenous Mayans.
1989–1994 2,097 UCDP-V
Guinea 1960–1980 28,000 UV Sekou Toure, political
repression.
Haiti 1958–1986 35,000 UV Duvalier, political
repression.
1991–1994 3,500 UV Political repression,
Cedras/FRAPH
government.
Hungary 1919 5,000 EGK White terror.
1941 3,300 d
EGK Massacre, mostly Serbs.
India 1946–1947 500,000 EGK Massacre and postwar
flight of Muslims.
Naxalites.
1984–1994 3,000 UV Punjab.
1984 7,500 EGK Killing of Sikhs during
anti-Sikh violence.
1990– 30,000 UV Kashmir.
2006d
1990–1991 1,000 UV Assam.
2003– 1,772 PITF-A
2005
Indonesia 1949–1962 2,750 UV Darul Islam.
1965–1966 750,000a PITF-G After attempted coup,
suspected communists
and ethnic Chinese are
massacred.
1965–1966 375,000 UV Anti-communist
massacres.
(continued)

notes are in square
brackets]
1965–1967 750,000 EGK Anti-communist, anti-
Chinese massacre.
1969– 40,000 UV West Papua—OPM
2006 b [Organisasi Papua
Merdeka].
1975–1992 198,750 PITF-G Following invasion
of East Timor, resisting
Timorese are killed in
massacres and famine.
1975–1999 115,000 UV East Timor.
1976–1987 149,000 EGK East Timor;
fatality estimate
is for those killed
1976–1979.
1989 EGK Killings of indigenous
group: Auyu.
1989– 9,000 UV Aceh.
2005
1998– 3,864 PITF-A
2000
2003– 1,772 PITF-A
2004
Iran 1978–1979 2,500 UV Political repression,
Reza Shah Pahlevi.
1979– 15,000 UV Postrevolution
2006 b repression, Kurds.
1981–????c 15,000 EGK Mass murder of
Kurds, Baha’is, and
Mujahadeen.
1981–1992 13,700 PITF-G Government suppresses
dissident Muslims
and rebel Kurds and
selectively executes
Baha’is.
(continued)

notes are in square
brackets]
Iraq 1959 1,000 UV Mosul uprising.
1959–1975 EGK Mass murder of Kurdish
nationalists.
1961–1991 160,000 UV Kurds.
1963–1975 42,500 PITF-G Large-scale massacres
to suppress repeated
rebellions for
independent Kurdistan.
1963– 85,000 UV Saddam, political
2003 repression.
1988 5,000 EGK Kurds.
1988–1991 336,000 PITF-G Al-A nfal campaign
in Iraqi Kurdistan
involving gassing,
massacres, village
demolitions.
1990–1996 1,695 UCDP-V
1991 EGK Killing of Kurds during
Persian Gulf War.
1991–2003 45,000 UV Shiites.
Israel 1987–1993 1,095 EGK Palestinians killed by
Israeli security forces
during Intifada.
2000– 2,308 PITF-A
2010
Italy 1937 30,000 EGK Mass executions of
Ethiopians after failed
assassination.
Japan 1935–1939 5,150,000 EGK Of Chinese (Rape of
Nanking, 200,000).
1935–1939 20,000 EGK Indonesian civilians.
Jordan 1920–1921 EGK Massacre of Jewish
refugees.
(continued)

notes are in square
brackets]
1929 EGK Massacre of Jewish
refugees.
1946 EGK Massacre of Jewish
refugees.
1970–1971 1,000 UV PLO.
Kenya 1991–1994 1,500 i
EGK Violence in tribal clashes.
Korea North 1948– 950,000 UV Political repression.
2006 b
Korea South 1948–1950 21,363 UV Civil violence in South.
[According to COW,
S. Korea becomes a state
in 1949].
Laos 1960–1973 60,000 UV Communists, civil war.
1963–????c 19,000 EGK Mass murder of Meo
tribesmen.
1975– 20,500 UV Communist repression,
2006 b Hmong civil war.
1979–1986 100,000e EGK Mass killings
of indigenous
group: Hmong.
Lebanon 1975–1990 200,000i EGK Deaths during civil
war caused by religious
hatred.
Liberia 1989–1990 11,250 UV Civil war.
1990 1,250 UCDP-V
1995– 1,088 PITF-A
2004
2000– 1,000 UV Civil war, LURD
2003 [Liberians United
for Reconciliation
and Democracy] and
MODEL [Movement for
Democracy in Liberia].
(continued)

notes are in square
brackets]
2001– 2,844 UCDP-V
2003
Libya 2011 7,188 PITF-A
Malawi 1964–1994 6,000 UV Political repression.
Mali 1988–1990 EKG Massacre of indigenous
group: Tuareg.
Mongolia 1929–1932 EGK Killings among
monastic class, nobility,
political opposition, and
by communists.
Mozambique 1975–1992 31,500 UV RENAMO.
Myanmar 1978 6,000 PITF-G Military units and
(Burma) militant Buddhist
elements attack
Arakanese Muslims in
Western Burma.
1978 EGK Mass murder of
Muslims in border
region.
1948– 145,000 UV Ethnic separatists.
2006 b
1988 3,000 UV Political repression.
1988 3,000 EGK Army opening fire on
peaceful pro-democracy
demonstrations.
1948–1990 100,000 UV Communist insurgency,
political repression.
Nepal 1995– 7,000 UV Civil war vs. Maoists.
2006 b
Netherlands 1873–1913 65,000 EGK Massacre during
occupation of Sumatra.
Nicaragua 1974–1979 7,000 UV Somoza.
(continued)

notes are in square
brackets]
group: Miskito.
1981–1990 1,500 UV Contras.
Niger 1988–1990 EGK Massacre of indigenous
group: Tuareg.
Nigeria 1966–1970 1,000,000 EGK Massacre, expulsion,
and starvation of Igbos.
1967–1970 1,225,000f PITF-G Government blockade
of relief assistance
leads to deaths from
starvation and disease.
1967–1970 1,300,000 UV Biafra.
1980 4,100 UV Kano/a nti-Muslim.
1990– 4,500 UV Niger Delta/Ogoni.
2006 b
1998– 1,963 PITF-A
2005
Pakistan 1946–1947 EGK Massacre and flight of
indigenous peoples.
1958–1974 EGK Mass murder of Baluchi
tribesmen.
1971 2,000,000a PITF-G Military deploys against
Bengali nationalists and
indiscriminately attacks
civilians.
1971 2,125,000 EGK Mass murder of Bengali
nationalists.
1971 650,000 UV Bangladesh.
1973–1977 1,750 UV Baluchistan.
1973–1977 7,300 PITF-G Baluchi rebellion
suppressed by military
using indiscriminate
violence against civilians.
(continued)

notes are in square
brackets]
Papua New 1988 EGK Massacre of indigenous
Guinea group: Dani.
1988–1998 11,250 UV Bougainville.
group: Nasioi.
Paraguay 1990–1991 EGK Massacre of indigenous
group: Pai Tavytere.
Peru 1980–1992 25,000 UV Shining Path.
Philippines 1946–1954 2,000 UV Huks.
1968–1985 55,000 EGK Mass murder of Moro
(Muslim) nationalists.
1969– 6,500 UV New People’s Army.
2006 b
1972–1986 45,000 UV Moro civil war.
1972–1976 48,000 PITF-G Military and
paramilitary terror
tactics: Moros are
killed in massacres and
napalm bombings.
1987 EGK Massacre of indigenous
group: Atta.
Poland 1945–1948 1,421,700 UV Expulsion of Germans.
1945–1948 1,583,000 EGK Removal of Germans.
1945–1947 5,000 UV Ukrainian nationalists.
Portugal 1961–1962 40,000 EGK Mass murder of
Bakongo during
suppression of
nationalist uprising in
Angola.
Romania 1907 10,000 EGK Government
suppression of peasant
revolt.
(continued)

notes are in square
brackets]
1919, EGK Romania-Hungary
1936, conflict in Transylvania
1940, (under Romanian
1949, control).
1956,
1959, 1990
1989 1,000 UV
Russia/USSR 1903–1906 3,188 EGK Jewish people killed
during wave of
pogroms.
1919 700,000 EGK Mass killing of Cossacks
during suppression of
the Don Cossack revolt.
1922–1941 EGK Killings of the
bourgeoisie, aristocracy,
kulaks.
1930–1937 6,500,000 EGK Killings of kulaks.
1932–1933 6,000,000 EGK Artificially induced
famine of peasants,
mostly Ukrainians.
1937–1938 1,000,000 EGK Execution of
Communist Party
members during Great
Terror.
1940–1953 EGK Antinationalist
deportation of the Balts
in Estonia.
repatriated Soviet
nationals.
Chechens, Ingushi,
Karachai, and Balkars.
(continued)

notes are in square
brackets]
1944– 116,000 EGK Mass murder of
1968 Meskhetians and
Crimean Tatars.
1945–1953 EGK Killings of Jewish
people, intelligentsia,
etc.
1947–????c EGK Mass murder of
Ukrainian nationalists.
1949 55,000 EGK Deportation of
Estonians.
1992–1993 2,000i UV [Killings of Georgians
by Abkhazia separatists
with alleged support
from Russian forces.]
1994– 65,278 PITF-A
2003
1994– 62,500 UV Civil war in Chechnya.
2006 b
1999– 2,301 UCDP-V
2002
Rwanda 1962–1964 11,000 EGK Mass murder of Tutsi
ruling class.
1963–1964 12,150 PITF-G Vengeance attacks by
Hutu officials following
cross-border incursions
by Tutsi rebels.
1963–1967 16,000 UV
1990–1994 650,000 UV Hutu vs. Tutsi.
1990–1997 511,302 UCDP-V
1994 750,000 a
PITF-G Hutu-dominated
military and gangs
systematically slaughter
Tutsis and Hutu
moderates.
(continued)

notes are in square
brackets]
1994 855,000 EGK Massacres during Hutu-
Tutsi conflicts.
1994–1999 13,000 UV Tutsi vs. Hutu.
1995–1999 9,522 PITF-A
1996–1997 75,000 EGK Killing of Hutu refugees
from Rwanda and
Burundi in Congo
by Tutsi army from
Rwanda.
Sierra Leone 1991–2002 4,500 UV
Somalia 1982–1990 55,000 UV Barre vs. SNM
[Somali National
Movement]—Isaaqs
and others.
1988–1991 29,150 PITF-G Indiscriminate
antiinsurgency
operations cause
civilian deaths (esp.
among Issaq clan).
1989–1991 1,138 UCDP-V
2007–2012 1,273 PITF-A
South Africa 1976–1994 2,700 UV
group in occupied
Namibia: San.
1990–1994 14,000 EGK Political violence during
transition before first
all-race election.
Spain 1936–1939 430,000 EGK Killing of soldiers
and civilians during
and after Spanish Civil
War.
(continued)

notes are in square
brackets]
Sri Lanka 1958, 40,000 EGK Tamil-Sinhalese
1971, conflict.
1977,
1981,
1983–
1986,
1995
1971 5,000 UV JVP [Janathā Vimukthi
Peramuṇa] I.
1983– 16,000 UV Tamil.
2002
1989–1990 25,500 PITF-G Revolutionary
campaign prompts
government to unleash
military and police
death squads.
1989–1992 42,500 UV JVP [Janathā Vimukthi
Peramuṇa] II.
2006– 24,942 PITF-A
2010
2008– 48,650 PITF-G Indiscriminate
2009 decimation of LTTE
[Liberation Tigers of
Tamil Eelam] follows
government withdrawal
from cease-fi re.
Sudan 1952–1972 500,000 EGK Mass murder of
southern nationalists.
1955–1972 500,000 UV Civil war.
1956–1972 441,000 PITF-G Government uses
indiscriminate violence
against mostly non-
Muslim Africans in the
south.
(continued)

notes are in square
brackets]
1983– 1,750,000 UV
2005
1983– 2,012,000g PITF-G Government targets
2002 secessionist non-
Muslim southerners and
Nuba for destruction.
1989–2011 10,969 UCDP-V Excludes Janjaweed
killings (2001–2008) of
3,501.
1991–1992 EGK Mass killing
of indigenous
group: Nuba.
1992–1993 EGK Mass killing
of indigenous
group: Dinka, Nuer.
2002– 53,618 PITF-A
2012
2003–2011 400,500 PITF-G Government-backed
Janjaweed militias
attack non-A rab peoples
in Darfur.
2003– 250,000 UV Darfur.
2006 b
Syria 1979–1985 17,500 UV Muslim Brotherhood.
1981–1982 36,000 PITF-G Government military
and security forces
attack Muslim
Brotherhood.
1981–1982 15,000 EGK Mass murder of Muslim
Brotherhood.
2011–2013 3,862 UCDP-V
2011–2012 5,343 PITF-A
Tajikistan 1992–1997 17,500 UV United Tajik
Opposition.
(continued)

notes are in square
brackets]
Turkey/ 1822–1923 68,000 EGK Greeks killed.
Ottoman Emp.
1909 27,500 EGK Massacre of Armenians
in Adana.
1914–1918 EGK Massacres/e viction of
Nestorian and Jacobite
Christians and the
Maronites of Lebanon.
1915–1918 1,562,000 EGK Massacres of Armenians
in Armenia, Baku, and
Shusha.
1922, EGK Massacre of Christians
1974 (mostly Greeks); 30,000
fatalities.
1924–1927 30,000 EGK Anti-Kurdish
campaigns.
1937–1938 EGK Military campaigns
against Dersim tribes of
Kurds.
1984–1999 17,500 UV Kurds.
1992–1996 1,300 PITF-A
Uganda 1971–1979 216,000 PITF-G Amin regime systemati
cally exterminates
political opponents and
personal enemies.
1971–1979 165,000 UV Amin.
Karamojong, Acholi,
Lango, Catholic clergy,
political opponents of
Amin.
Karamojong, Nilotic tribes,
Bagadans, and supporters
of Amin regime.
(continued)

notes are in square
brackets]
1980–1986 240,000 PITF-G Political and tribal rivals
of Obote are targeted by
army and armed bands.
1981–1986 250,000 UV Civil war.
1986– 1,500 UV NRA [National
2006 b Resistance Army] vs.
Holy Spirit–UPDA
[Uganda People’s
Democratic Army] LRA
[Lord’s Resistance Army].
United 1830s– EGK Removal/k illings of
Kingdom 1950s Aborigines in Tasmania.
USA 1500s– EGK Massacre of indigenous
1900 peoples.
Vietnam 1953–1954 15,000 EGK Mass murder of
Catholic landlords and
rich and middle-class
peasants.
1965–1972 475,000 EGK Mass murder of civilians
in NLF areas.
1975–1987 250,000 EGK Mass murder of “boat
people.”
1975–????c 82,500 UV Postwar political
repression by
communist
government.
Vietnam 1954–1957 41,000 UV Political repression.
North
Vietnam 1954–1975 225,000 UV Civil war.
South
1965–1975 528,000 PITF-G Military and
paramilitary forces
engage in killings
against villagers
supporting Viet Cong.
(continued)

notes are in square
brackets]
Yemen 1962–1970 35,000 UV
1986 1,500 UV
Yugoslavia 1941–1945 577,500 EGK Massacre of Serbs,
(including Jewish people, and
former) Roma (in Croatia
by Nazi-supported
regime).
1945–1948 57,178 UV Expulsion of Germans.
1945–1948 48,000 EGK Crushing of Albanian
resistance to Serbian
rule in Kosovo.
1991–1992 3,500 UV Croatian civil war.
1991–1995 200,000 EGK Killings of Muslims and
Croats in Bosnia during
Bosnia war.
1992–1995 228,000 PITF-G Bosnian Muslims are
subject to massacres by
Serb and Croat forces.
1992–1995 145,000 UV
1992–1995 12,454 UCDP-V Serbian Republic
of Bosnia and
Herzegovina.
1995 14,505 PITF-A [1995 is the first year in
the dataset.]
1998–1999 10,000 UV Kosovo.
1998–1999 13,500 PITF-G Serb militias backed
by Yugoslavian armed
forces target ethnic
Albanians.
1998–1999 3,478 PITF-A
(continued)

notes are in square
brackets]
1999 55,000 EGK Killings of Albanian
Muslims by Serbians
during Kosovo war.
Zanzibar 1964 4,500 UV Political repression.
Zimbabwe 1972–1979 7,000 UV Civil war.
1982–1983 EGK Killings of indigenous
group: Tyua.
1982–1987 20,000 UV
Appendix A2
Selected Mass Atrocities (Genocides and Mass Killings)
Perpetrated by Nonstate Groups, 1989–2013
Nonstate Group Location Time Estimated Data

Period Fatalities Source
Abkhazia Abkhazia 1992–1993 2,000i UV
Separatists
Alliance of Congo-Kinshasa 1996–1997 35,126 UCDP-V
Democratic Forces
for the Liberation
of Congo-Kinshasa
Al Qaida USA 2001 2,753 UCDP-V
Army for the Rwanda, DR 1996–1998 3,059 UCDP-V
Liberation of Congo, Burundi,
Rwanda Uganda
Communist Party India 2006–2013 1,503 UCDP-V
of India—Maoist
Congolese Rally for DR Congo 1998–2003 5,314 UCDP-V
Democracy
(continued)

Congolese Rally DR Congo 2002 1,200 UCDP-V
for Democracy—
Kisangani—
Liberation
Movement and
Patriotic Force of
Resistance in Ituri
(Coalition)
Democratic Rwanda, Uganda, 2009–2012 1,845 UCDP-V
Liberation Forces Tanzania
of Rwanda
FARC, ELN, etc.h Colombia 1965–2006 50,000i UV
Paramilitary Groups Colombia 1999–2000 1,000i EGK
Front for National DR Congo 2002–2004 2,280 UCDP-V
Integration
Hutu Rebels Burundi 1965–1973 129,750i PITF-G
Burundi 1993 60,000 i
EGK
Burundi 1993 48,000 i
PITF-G
Burundi 1995–2000 2,136 UCDP-V
ISIS [Islamic State] Iraq, Jordan, Syria 2005–2013 8,198 UCDP-V
Jama’atu Ahlis Nigeria, Cameroon 2012–2013 1,298 UCDP-V
Sunna Lidda’awati
wal-Jihad
Kashmir Insurgents India 1996–2006 2,783 UCDP-V
Liberation Tigers of Sri Lanka 1989–1996 1,712 UCDP-V
Tamil Eelam
Liberia Peace Liberia 1994–1995 2,168 UCDP-V
Council
Lord’s Resistance Uganda, Sudan 1995–1998 1,418 UCDP-V
Army Uganda, DR Congo, 2002–2011 5,722 UCDP-V
Sudan, Cen. Afr. Rep.
Mozambican Mozambique 1989–1992 1,400 UCDP-V
National Resistance
(continued)

National Patriotic Liberia, Ivory 1990–1995 6,042 UCDP-V
Front of Liberia Coast, Sierra Leone
National Union Angola, Namibia, 1999–2001 1,671 UCDP-V
for the Total Zambia
Independence of
Angola
Patani Insurgents Thailand 2004–2012 1,744 UCDP-V
Religious Groups Lebanon 1975–1990 200,000 i
EGK
Rwandan Patriotic Rwanda, Burundi 1993–1994 1,222 UCDP-V
Front
Seleka Rebels, Anti- Central African 2013 1,500i PITF-G
Balaka Militias Rep.
Sikh Insurgents India 1989–1992 1,314 UCDP-V
Taleban Movement Pakistan 2008–2013 1,891 UCDP-V
of Pakistan
Tribal Clashes Kenya 1991–1994 1,500i EGK
Union of Congolese DR Congo 2002–2003 2,362 UCDP-V
Patriots
United Liberation Liberia, Guinea 1994–1995 1,359 UCDP-V
Movement for
Democracy in
Liberia—Alhaji
Kromah Faction
United Liberation Liberia 1996 1,025 UCDP-V
Movement for
Democracy in
Liberia—Roosevelt
Johnson Faction
United Liberation Liberia, Sierra 1993 1,272 UCDP-V
Movement for Leone
Democracy in
Liberia
Notes for Appendices A1 and A2:

1. PITF-G = Political Instability Task Force Geno/politicide Dataset.
UV = Ulfelder and Valentino (2008) Mass Killing Dataset.
(continued)

EGK = Easterly, Gatti, and Kurlat (2006) data.
UCDP-V = Uppsala Conflict Data Program One-Sided Violence Dataset.
PITF-A = Political Instability Task Force Worldwide Atrocities Dataset.
2. UCDP-V and PITF-A data were used to identify cases of mass killing as follows: I adopted
Ulfelder and Valentino’s (2008) description of a mass killing as the intentional death of at least
1,000 noncombatants in a period of sustained violence. A mass killing begins in the first year in
which there are at least 100 intentional fatalities. If fewer than 100 fatalities are recorded each year
for any three consecutive years, the event ended during the first year of the three-year period in
which fatalities dropped below 100.
3. Cases from EGK with start dates prior to 1900 that continued to at least 1900 are included
in Appendix A1.
4. Two cases of nonstate mass killings from the UCDP-V dataset were excluded from this
appendix because they were in alliance with a state: Janjaweed (2003–2 007) and South Sudan
Defence Force (1991–1992).
5. Footnotes for appendices:
a
Based on Harff (2003).
b
UV data indicate that mass killing is ongoing as of 2006.
c
End date unknown.
d
Easterly, Gatti, and Kurlat (2006) do not provide fatality estimates, but Rummel (1998,
235) estimates 3,300 Serb and Jewish fatalities for this case, which is known as the Novi Sad raid
or Újvidék massacre.
e
Easterly, Gatti, and Kurlat (2006) do not provide fatality estimates, but Rummel (1998,
313) cites 100,000 fatalities.
f
Based on Valentino (2004, 88).
g
Based on Harff (2003) estimate for 1983–2 001 plus PITF-G data point for 2002.
h
The UV comments note FARC and ELN, which are nonstate groups. The case is included
in both appendices because the comment indicates nonstate groups and the UV dataset focuses on
cases involving state actions of mass killing (see Ulfelder and Valentino 2008, 2, 5–6).
i
Case recorded in both tables owing to evidence of substantial state and independent nonstate
mass atrocity behavior.
Notes
1. Some scholars treat genocides and mass killings as essentially the same phenomenon (see,
e.g., Shaw 2007; Ulfelder and Valentino 2008).
2. Space precludes an extensive list of specific-country datasets, but examples for the Rwandan
genocide include McDoom (2013) on locations of participants, nonparticipants, and vic-
tims from one community; Verpoorten (2012) on intensity of Tutsi deaths across more
than 1,000 administrative units; and Verwimp (2003) on the fate of 352 peasant house-
holds during the transition from civil war and genocide to a postconflict environment.
3. Chapter 2 of this volume provides a thorough treatment of conceptual issues pertinent to
intentional atrocities against civilians.
4. My use of the term “genocide” includes politicides.
5. There are exceptions. For example, Armstrong and Davenport’s (2008) empirical analy-
sis of genocide fatalities includes a control variable for the number of terrorist incidents.
Moreover, the Uppsala Conflict Data Program (UCDP) notes in the FAQ section of
its webpage that its “category of ‘one-sided violence’ often overlaps with definitions of
terrorism with a lethal outcome,” although “it does not make use of the term ‘terrorism’ to
classify any type of violence” (quotes downloaded from pcr.uu.se/research/ucdp/faq/ on
January 2, 2014). Furthermore, some scholars emphasize interconnections between war,
terrorism, and civilian atrocities (e.g., Kalyvas 2003, 2004, 2005, 2006; Sambanis 2004;
Valentino 2004).
6. Several empirical studies assess risks of nonstate groups perpetrating civilian atrocities
(e.g., Wood 2010, 2014; Hicks et al. 2011; Hultman 2012; Wood et al. 2012; Fjelde and
Hultman 2014).
7. Counts for genocides and mass atrocities in Figure 3.1 are the number of states perpe-
trating genocides and mass atrocities, not the number of specific-g roup cases being perpe-
trated by states. For example, the genocide count for Germany in 1943 is one (not three,
four, etc.). Data for genocides is for 1956–2 013 and comes solely from the PITF-G data-
set. Mass atrocity data is for 1900–2 013 and comes from the datasets used to construct
Appendix A1, with the proviso that double counting is avoided. Counts for terrorist attacks
against civilians are determined from the six types of civilian targets from GTD: private
citizens and property, businesses, educational institutions, journalists and other media
personnel, airports and aircraft, and tourists.
8. This paragraph updates similar summaries in Anderton (2014). Estimated monthly fatali-
ties of 241,750 for the Rwandan genocide is the average of the fatalities reported by the
PITF-G and UV datasets in Appendix A1 divided by 100 days (the frequently reported
length of the genocide) multiplied by 30 to scale on a monthly basis. The period 1970–2 013
is forty-four years, but 1993 GTD data are missing.
9. Geller and Singer (1998) summarize risk factors for interstate conflict based on over 500
large-sample empirical studies. Given the many review articles on empirical studies of
civil war risk (e.g., Collier and Sambanis 2005; Collier and Hoeffler 2007; Dixon 2009;
Blattman and Miguel 2010; and Hoeffler 2012) and Enders and Sandler’s (2011) survey
of empirical literature on terrorism risk, it seems reasonable to believe that there are at
least 200 published large-sample empirical studies of civil war risk and 100 or more on
terrorism risk. Meanwhile, the number of published large-sample empirical studies of
civilian atrocity risk or seriousness appears to be less than three dozen at the time of this
writing.
10. A “genocide gap” in defense and peace economics appears to exist in theoretical and empir-
ical models in which conflict is the dependent variable. I am grateful to Mansoob Murshed
for pointing out to me that such a gap may also exist in studies that include war or military
expenditure as an independent variable in studies of economic outcomes (e.g., growth).
11. I define a “war-level conflict” for UCDP/PRIO as that dataset’s designation that the case
reached a cumulative intensity of war.
12. For the 43 specifically identified genocides since 1956 (i.e., those in the PITF-G data-
set), 39 (91 percent) have a time overlap with a COW intrastate war. Of these 39 cases, 13
(33 percent) begin after the onset of war, 19 (51 percent) begin simultaneously with war,
and only 7 (18 percent) begin before war onset. Hence, genocides appear to be much less
likely to occur before the onset of war relative to mass killings.
References
Anderton, C. H., and J. R. Carter. 2011. “Conflict Datasets: A Primer for Academics,
Policymakers, and Practitioners.” Defence and Peace Economics 22, no. 1: 21–42.
Armstrong, D., and C. Davenport. 2008. “Six Feet Over: Internal War, Battle Deaths and
the Influence of the Living on the Dead.” In S. M. Saideman and M. J. Zahar, eds.,
Intra-State Conflict, Governments and Security: Dilemmas of Deterrence and Assurance.
New York: Routledge, 33–53.
Blattman, C., and E. Miguel. 2010. “Civil War.” Journal of Economic Literature 48, no. 1: 3–57.
Bussmann, M., and G. Schneider (with A. Bräunig, C. Ruhe, A. Scharpf, and R. van der Haer).
2012. “Appendix A: Konstanz One-Sided Violence Event Dataset (KOSVED) Codebook
Version 1.2 –January 27, 2012.” Retrieved from www.polver.uni-konstanz.de/gschneider/
forschung/kosved/codebooks.
Charny, I., ed. 1999. Encyclopedia of Genocide. Vols. 1 and 2. Santa Barbara, CA: ABC-CLIO.
Collier, P., and A. Hoeffler. 2007. “Civil War.” In T. Sandler and K. Hartley, eds., Handbook of
Defense Economics. Vol. 2. New York: Elsevier, 711–39.
Collier, P., and N. Sambanis, eds. 2005. Understanding Civil War: Evidence and Analysis. Vol. 1:
Africa. Vol. 2: Europe, Central Asia. Washington, DC: World Bank Publications.
Croicu, M., and R. Sundberg. 2015. “UCDP Georeferenced Event Dataset Codebook Version
3.0.” Department of Peace and Conflict Research, Uppsala University. Retrieved from
http://www.ucdp.uu.se/ged/data.php.
Dixon, J. 2009. “What Causes Civil Wars? Integrating Quantitative Research Findings.”
International Studies Review 11, no. 4: 707–35.
Easterly, W., R. Gatti, and S. Kurlat. 2006. “Development, Democracy, and Mass Killing.” Journal
of Economic Growth 11, no. 2: 129–56.
Eck, K., and L. Hultman. 2007. “One-Sided Violence against Civilians in War: Insights from
New Fatality Data.” Journal of Peace Research 44, no. 2: 233–4 6.
EDACS. 2012. “Event Data on Armed Conflict and Security (EDACS) Codebook.” Available
at http://w ww.conflict-data.org/edacs/downloads/EDACS_Codebook_Version3_7_
20121128.pdf#.
Enders, W., and T. Sandler. 2011. The Political Economy of Terrorism. 2nd ed. New York: Cambridge
University Press.
Esteban, J., M. Morelli, and D. Rohner. 2015. “Strategic Mass Killings.” Journal of Political
Economy 123, no. 5: 1087–1132.
Fjelde, H., and L. Hultman. 2014. “Weakening the Enemy: A Disaggregated Study of Violence
against Civilians in Africa.” Journal of Conflict Resolution 58, no. 7: 1230–57.
Geller, D. S., and J. D. Singer. 1998. Nations at War: A Scientific Study of International Conflict.
Gleditsch, N. P., P. Wallensteen, M. Eriksson, M. Sollenberg, and H. Strand. 2002. “Armed
Conflict 1946–2 001: A New Dataset.” Journal of Peace Research 39, no. 5: 615–37.
Hicks, H., U. R. Lee, R. Sundberg, and M. Spagat. 2011. “Global Comparison of Warring Groups
in 2002–2 007: Fatalities from Targeting Civilians vs. Fighting Battles.” PLoS ONE 6, no.
9: 1–14.
Hoeffler, A. 2012. “On the Causes of Civil War.” In M. R. Garfinkel and S. Skaperdas, eds., The
Oxford Handbook of the Economics of Peace and Conflict. New York: Oxford University Press,
179–2 04.
Hultman, L. 2012. “Attacks on Civilians in Civil War: Targeting the Achilles Heel of Democratic
Governments.” International Interactions 38, no. 2: 164–81.
Kalyvas, S. 2003. “The Ontology of ‘Political Violence’: Action and Identity in Civil Wars.”
Perspectives on Politics 1, no. 3: 475–94.
Kalyvas, S. 2004. “The Paradox of Terrorism in Civil War.” Journal of Ethics 8, no. 1: 97–138.
Kalyvas, S. 2005. “Warfare in Civil Wars.” In I. Duyvesteyn and J. Angstrom, eds., Rethinking the
Nature of War. Abingdon: Frank Cass, 88–108.
Kalyvas, S. 2006. The Logic of Violence in Civil War. New York: Cambridge University Press.
Levene, M. 2005. Genocide in the Age of the Nation-State. Vol. 1, The Meaning of Genocide.
London: I. B. Tauris.
Marshall, M. G., T. R. Gurr, and B. Harff. 2014. “PITF—State Failure Problem Set: Internal
Wars and Failures of Governance, 1955–2 013.” Vienna, VA: Societal-Systems Research.
McDoom, O. S. 2013. “Who Killed in Rwanda’s Genocide? Micro-Space, Social Influence
and Individual Participation in Intergroup Violence.” Journal of Peace Research 50, no.
4: 453– 67.
National Consortium for the Study of Terrorism and Responses to Terrorism (START). 2015.
Global Terrorism Database. Online Data. Retrieved from www.start.umd.edu/g td [data
downloaded January 2015].
Pettersson, T. 2012. “UCDP One-Sided Violence Codebook Version 1.4.” Department of Peace
and Conflict Research, Uppsala University. Retrieved from www.pcr.uu.se/research/
ucdp/datasets/ucdp_one-sided_v iolence_dataset.
PITF-A . 2009. “Political Instability Task Force Worldwide Atrocities Event Data Collection
Codebook Version 1.0B2.” Political Instability Task Force. Retrieved from http://event-
data.parusanalytics.com/data.dir/atrocities.html.
Raleigh, R., and C. Dowd. 2015. “Armed Conflict Location and Event Data (ACLED) Project
Codebook 2015.” Armed Conflict Location and Event Data Project (www.acleddata.com).
Rummel, R. J. 1998. Statistics of Genocide: Genocide and Mass Murder since 1900. Piscataway,
NJ: Transactions.
Salehyan, I., and C. Hendrix. 2012. “Social Conflict in Africa Database Version 3.0.” http://w ww.
du.edu/korbel/sie/research/hendrix_ scad_database.html [downloaded January 2014].
Sambanis, N. 2004. “What Is Civil War? Conceptual and Empirical Complexities of an
Operational Definition.” Journal of Conflict Resolution 48, no. 6: 814–58.
Sarkees, M. R., and F. W. Wayman. 2010. Resort to War: 1816–2007. Washington, DC: CQ Press.
Schneider, G., and M. Bussmann. 2013. “Accounting for the Dynamics of One- Sided
Violence: Introducing KOSVED.” Journal of Peace Research 50, no. 5: 635–4 4.
Shaw, M. 2003. War and Genocide. Cambridge, UK: Polity Press.
Shaw, M. 2007. What Is Genocide? Cambridge, UK: Polity Press.
Small, M., and J. D. Singer. 1976. “The War-Proneness of Democratic Regimes, 1816–1995.”
Jerusalem Journal of International Relations 1, no. 4: 50–69.
Themnér, L., and P. Wallensteen. 2014. “Armed Conflict, 1946–2 013.” Journal of Peace Research
51, no. 4: 541–5 4.
Ulfelder, J., and B. Valentino. 2008. “Assessing Risks of State-Sponsored Mass Killing.” Working
Paper. http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1703426.
Verpoorten, M. 2012. “Leave None to Claim the Land: A Malthusian Catastrophe in Rwanda?”
Journal of Peace Research 49, no. 4: 547–63.
Verwimp, P. 2003. “Testing the Double-Genocide Thesis for Central and Southern Rwanda.”
Journal of Conflict Resolution 47, no. 4: 423–42.
Wallensteen, P., and M. Sollenberg. 1997. “Armed Conflicts, Conflict Termination and Peace
Agreements, 1989–96.” Journal of Peace Research 34, no. 3: 339–58.
Waller, J. 2007. Becoming Evil: How Ordinary People Commit Genocide and Mass Killing. 2nd ed.
Wood, R. M. 2010. “Rebel Activity and Strategic Violence against Civilians.” Journal of Peace
Research 47, no. 5: 601–14.
Wood, R. M. 2014. “Opportunities to Kill or Incentives for Restraint? Rebel Capabilities, the
Origins of Support, and Civilian Victimization in Civil War.” Conflict Management and Peace
Science 31, no. 5: 461–80.
Wood, R. M., and M. Gibney. 2010. “The Political Terror Scale (PTS): A Re-I ntroduction and a
Comparison to CIRI.” Human Rights Quarterly 32, no. 2: 367–4 00.
Wood, R. M., J. D. Kathman, and S. E. Grant. 2012. “Armed Intervention and Civilian
Victimization in Intrastate Conflict.” Journal of Peace Research 49, no. 5: 647–6 0.
4
The Demography of Genocide

Ta deusz K ugl e r
4.1. Introduction
This chapter focuses on the demography of genocide. The goals are to investigate
the measurable effects of genocide and to evaluate the lasting influences of its
resulting casualties and forced migrations on the structural foundations of a pop-
ulation. Genocide’s shock to society can fundamentally reshape a population’s
underlying demographic components—t he inertial forces of ordinary birth and
death rates. The shock, however, although it may be large in absolute numbers,
can be “small enough” to be obfuscated by ordinary yearly population variance
and population measurement and estimation errors. The specific characteristics
of genocide matter, as do the likely political goals of the perpetrators. Establishing
causality in regard to which portion of a population is the victim of genocide to
begin with, and which portion of postgenocide population recovery is due to
postgenocide population “catch-up”—rather than ordinary birth rates—a re very
complex matters. The populations targeted (e.g., young, old, educated or not) and
the choices made after the event (e.g., encouraging return migration or contin-
ued persecution) each influence the interaction between population recovery and
prospects for postgenocide economic development. This chapter offers a primar-
ily descriptive evaluation of the demographic fundamentals of genocide, their
interactions with economic development, and their effects on refugees; addition-
ally, there is a call for future research on policy requirements needed for lasting
population recovery and peace.
Casualties resulting from violent conflict, including those involving mass
atrocities against civilians, are not normally distributed. Males, the very young,
and the elderly often take the brunt of the loss of life, while women often face orga-
nized sexual violence. Expected age distributions can thus be disrupted, resulting
in disproportionately limited numbers of the elderly, the young, or working-
age males in the immediate postviolence phase. Gender ratios—in personal,
political, labor market, or other terms—a re commonly reshaped in genocide,
102
T h e D e m o g ra p h y o f G e n o c i d e 103
creating extraordinarily high gender-i mbalance ratios, in fact among the highest
ever observed. The political realities of postgenocide society affect population
growth rates as well, determined for instance by a polity’s willingness to welcome
return migrants back home (and those migrants’ own willingness and/or abil-
ity to return). Lastly, the trauma of genocide itself may temporarily cause a rapid
increase in birth rates. This effect stems from personal choices made to increase
family (and/or group) size within surviving households. While the increase accel-
erates population recovery, this very population bonus can exceed the recovery
rate of the economy. Thus, rather than aiding overall societal recovery, this popu-
lation bonus may produce negative effects that affect the long-term destiny of a
nation. Individual decisions leading to birth are a critical aspect of recovery.
Section 4.2 is a descriptive overview and discussion of concepts, theories, and
measurement issues in the field of demography as they apply to cases of mass atroc-
ities such as genocide. Among other things, we find that non-genocide-related
population shifts, that is, birth-, death-, and migration-rate changes unrelated to
genocide, can camouflage the population effects of genocide in population statis-
tics, particularly in states that are in the early stages of economic development.
Similarly, it is important to differentiate population recovery to its pregenocide
population forecast from mere population growth. The latter may return a popu-
lation to its pregenocide level of population without returning it to its pregenocide
population trend. Each factor may imply different postgenocide policies to deal
with the population effect of genocide. Section 4.3 moves to a more specific, case-
based appraisal of population effects in different genocides, with an emphasis on
disruptions in expected age distributions and gender-ratio imbalances. Section 4.4
discusses the potential for population recovery and the long-term population
implications of genocide. Some policy and research-related issues and options are
discussed in section 4.5.
4.2. Demography and Evaluation of Violence-

Related Population Losses
4.2.1. Demography
One difficulty in analyzing the effect of genocide on the size and structure of a
population lies in differentiating between normal demographic fluctuations and
how they affect population projections and the specific influence of genocide-
related deaths and migrations. Establishing causality, if any, between the choices
made by individuals and institutions during and after the violence is a profoundly
complex matter.1 For example, an event may be classified as genocide, but if
the scale is not large relative to the total population it will not have much influ-
ence on national-level statistics, nor seem to necessitate large-scale policies of
reconstruction. Another difficulty lies in appreciating, and then disentangling,

the difference between recovery and mere growth. Postgenocide population recov-
ery moves the size of a population to the level of the pregenocide population fore-
cast or trend, whereas mere growth is sustained, postgenocide population growth
but does not have sufficient momentum to overcome the population-d iminishing
effect of genocide. Simply returning to the pregenocide level of population is not
the same as recovery to the pregenocide population forecast. In current research,
conflation of these two distinct aspects of rebuilding a population is a common
problem and a difficult distinction to make.
One reason to make the distinction between recovery and mere growth is that
it directly affects the very measurement of genocide-related losses. In most geno-
cides, infant and child mortality rates increase dramatically, with some level of
fluctuation also seen in the birth rate (see, e.g., Holck and Cates 1982; Hill 2004).
The issue here is to decide what should be counted as a loss. Should disrupted
births (increases in miscarriages and stillbirths) and losses not directly caused
by a genocidal episode be amalgamated into total casualty estimates (Seybolt,
Aronson, and Fischhoff 2013)? When dealing with long-term population recov-
ery, the issues are exacerbated when changing fertility rates may hide the true
nature of genocide’s consequences. Moreover, failed public health plans, eco-
nomic crises, and local environmental factors all can create small (and sometimes
large) movements in population growth rates (Montgomery and Cohen 1998;
Feng, Kugler, and Zak 2000).
Population losses occur directly both in terms of casualties and as a result of
other changes that adversely affect the population. With few exceptions, most of
the countries with long periods of instability and ongoing or repeated episodes
of genocide also have very high mortality (primarily infant) as well as high birth
rates, which can signal that the country in question is in the first or second phase
of a demographic transition (explained shortly). These two phases characterize
countries that are at the lowest level of development, fit the general demographic
profile of traditional societies, and have the potential to experience rapidly chang-
ing population growth rates. The overlapping coincidence of all these factors can
create considerable uncertainty as to the total losses suffered due to genocide per
se as opposed to other population-damaging phenomena.
Demographic transition theory is the basic framework for our analysis (Preston
1975; Restrepo and Rozental 1994; Eastwood and Lipton 2011). As Gould (2009)
explains, unlike much of economics or political science, demographic transition
theory is predominantly based on empirical evidence as opposed to conceptual
logic.2 Lee (2011, 569) notes:
Lacking practical guidance from grand dynamic theories, forecast-

ers rely on a largely descriptive framework known as the demographic
transition, which summarizes historical patterns initially observed in
Europe but which have been found appropriate for less developed coun-
tries (LDCs) as well. Over the course of the demographic transition,
populations move from an initial state of high mortality and high fertil-
ity to a state of low mortality and low fertility. Typically mortality begins
to decline first, continuing at a gradual and steady pace, with a later and
faster decline in fertility that may move from a high to a relatively low
level in a span of two or three decades.
These changes in vital rates cause dramatic changes in the popula-
tion size, the rate of population growth, and the age distribution. During
the period in which mortality has begun to decline but fertility remains
high, the population growth rate rises and the proportion of youth in
the population rises as well. Once fertility begins to decline the propor-
tion of population in the working ages rises, and continues to rise for
five or six decades, until well after fertility decline ceases. Eventually the
growth of the working age population slows while that of the older popu-
lation accelerates.
Demographic transition theory identifies four population phases, which

later researchers have linked to issues of economic and political development
(Teitelbaum1975; Caldwell 1976; Kirk 1996; Dyson 2010). 3 The limitation is cau-
sality: the precise why of population change is still much under discussion but the
how of change is at least roughly measurable.4 Relying on Gould (2009, 83–84)
and Omran (1982), the four phases may be stylized as follows:
• First phase: Least developed societies with high levels of both birth and death
rates (40 to 50 per thousand); low life expectancies (between 30 and 35 years
of age) caused by high levels of both famine and pestilence and resulting in
a pyramid-shaped age distribution (that is, large base, small apex); low total
population; and low overall population growth. Maddison (2009) shows that
this phase holds for most human societies until the early nineteenth century.
• Second phase: Early stages of economic development with population growth of
roughly 2 percent a year or higher (sometimes dramatically higher), caused by
the first sustained fall in mortality but not yet by large declines in birth rates;
increased life expectancy to over fifty-five years of age; rapid increases in total
population; and frequently a youth surge. For the cases discussed in this chap-
ter, this phase is the most commonly observed one.
• Third phase: Continued population expansion, with life expectancies gradu-
ally moving to the high sixties and a continuous decline in mortality rates
and now-rapid decline in birth rates and fertility. These factors are gener-
ally seen in middle-income countries. Much of the sustained population
growth is due to momentum and disproportionate age-d istributions (such
as baby booms), which cause the total population to grow even as fertility
rates decline. The reason is that the disproportionately large segment of the
young population enters adulthood; thus, even with declines in fertility,
growth continues. This phase has been linked to population recovery after
war, with temporary, postviolence increases in birth rates having a very sim-
ilar effect.
• Fourth phase: As seen in the economically advanced world, there are sharply
lower birth and mortality rates (single digits to fifteen per thousand), and
rapid increases in average life expectancy (eighty-plus years) but also the yet-
unknown effect of an increasingly aging population. This end phase of demo-
graphic transitions is the outcome of a number of successes, such as increased
life expectancy, declines in mortality of all types, and decreased fertility rates.
These common characteristics of the developed world are likely the future of
most of the developing world as well.
The first and second phases, commonly but not exclusively seen in states that
have experienced genocide, are the most volatile. Correspondingly, states with
these characteristics have the potential to see the most dramatic population
changes given small increases in either economic or political success (Acemoglu
and Johnson 2007). 5
Births and deaths are the primary influence on the expected total population,
with the latter the most easily abated by public health, investment, and infrastruc-
ture policies such as simple sanitation.6 Controlling death rates can cause dra-
matic increases in population growth rates. When combined with unmitigated
birth rates, a demographic youth bulge results. The youngest generation being the
largest, proportionally, is a normal characteristic of phases one and two. The dif-
ference lies in the scale of that proportionality. At the extremes, commonly seen in
phase two, the likelihood of political instability and violent conflict can increase.
So can the possibility of remarkable economic growth, as youth growth leads to a
high level of labor potential and the resulting comparative advantage that might
be harnessed for growth (Bloom, Canning, and Sevilla 2003; Urdal 2006, 2012;
Cincotta and Doces 2012). Importantly, the population increases observed in
phases one or two can be sufficiently large to statistically camouflage any effects
of genocide, as states that undergo substantial levels of trauma may simultane-
ously undergo rapid demographic changes unrelated to genocide (declines in fer-
tility or mortality as well as population growth).
Violent conflict appears to be a characteristic of demographic change,
although not necessarily a key one. The possibility of reworked public finances,
an increasingly stable political system, and economic success all are effects that
have the potential to mitigate the risk and/or expected damage resulting from
violence. Importantly, the confluence of all these factors and considerations leads
to an increasing complexity in the estimation of population size and its variations
over time.
4.2.2. Total Population and Demographic Growth Rates

The intertwined complexity of genocide-related population losses and demo-
graphic growth rates may be seen in Table 4.1. For a number of states, it lists war
years and estimates of total mass-atrocity-related deaths. The death estimates vary
widely, and sometimes there is no clear “start” year for a mass atrocity episode.7
Total pre-and postwar populations are taken from the United Nations (UN)
Population Division database and supplemented by Maddison’s data (2009) for
pre-1950 time periods. Note the apparent disconnect between the estimates of
deaths and the listed population size at the end of the episode (or episodes) of
violence. In twenty of the listed cases, the total population was larger after the vio-
lence than before. In seven cases there is a population reduction, and in only two
cases—Rwanda and Bosnia—does one see double-d igit percentage-point reduc-
tions in total population. In the majority of the cases, total population grew sig-
nificantly, generally following or recovering to pregenocide population forecasts.
The birth, migration, and natural growth rates shown in Table 4.1 illuminate
the likely causes of population changes. Note that the computation of the natural
population growth rate—which is, simply, the birth rate minus the death rate—
assumes a migration rate of zero. Natural growth may thus be thought of as endoge-
nous population growth, all other things (e.g., migration) being unchanged.8 Data
quality is a constant worry because as Hill (2004, 8) explains: “Demographers are
obsessive about two things: age and exposure time. We obsess about age because
demographic processes are highly age-dependent, with the result that quite small
differences can have a substantial effect on summary measures, such as the crude
birth rate.” Data quality thus should be such as to make “small differences” really
small for fear of otherwise throwing off any summary measures.
The most commonly used source of modern demographic data is the UN
Population Division (UN 1955a; UN 2011).9 Like many other cross-national
sources, it needs constant updates and reappraisal (Lee 2011; Andreev, Kantorová,
and Bongaarts 2013). Currently, a combination of amalgamating national sur-
veys and indirect methods are used to arrive at population estimates. Tabeau
and Zwierzchowski (2013) illustrate the dangers of making demography-related
assumptions based on limited supporting data, which then flow into population
estimates. For example, early death estimates in the Cambodian genocide of the
1970s were based on rough, and unverified, estimates of population totals. Tabeau
and Zwierzchowski explain (2013, 227–28):
In the absence of reliable data on population size and structure, migra-

tion, and fertility and mortality trends, such balancing equations have
no value. They express no more than individual authors’ subjective
views on one of the many possible variants of population development
during a given time frame (here, the Khmer Rouge period). The lack of
Table 4.1 Population Losses and Demographical Rates
Rates per Thousand
Belligerents War Years Civilian and Battle Prewar Postwar Loss (%) Birth Migration Natural
Population Losses Population* Population* Growth
Rwanda 1990–1994 650,000–855,000 6,804 5,461 –2 4.6% 40.4 –41.2 4.4
Bosnia 1992–1995 unknown 4,449 3,608 –23.3% 12.5 –2 6.7 3.2
Turkey 1914–1923 1,562,000 15,000 13,877 –8.1%
Cambodia 1970–1979 142,500–2 ,500,000 7,329 6,888 – 6.4% 41.3 –16.6 7.4
Liberia 1989–1996 1,088–11,250 2,340 2,201 –6.3% 44.2 35.0 23.8
Germany 1939–1945 9,939,850 68,558 64,678 –6.0%
Afghanistan 1979–1989 1,175,000–1,750,000 15,269 14,669 –4.1% 52.4 2.1 30.0
Algeria 1961–1962 48,000–90,000 10,909 11,273 3.2% 50.4 –12.5 30.9
China 1939–1945 10,000,000 513,336 535,418 4.1%
South Korea 1950–1953 21,363 20,208 21,259 4.9% 38.0 1.6 21.6
Lebanon 1975–1990 200,000 2,988 3,196 6.5% 25.7 1.7 18.6
Japan 1939–1945 5,150,000 71,879 77,199 6.9%
Burundi 1993–1997 48,000–60,000 5,809 6,271 7.4% 40.5 –15.6 22.3
Nigeria 1967–1970 1,225,000–1,300,00 49,491 57,234 13.5% 46.0 – 0.5 24.2
China 1949–1956 4,500,000 541,085 637,408 15.1% 37.4 – 0.2 16.5
Yemen 1962–1969 35,000 5,994 7,098 15.6% 50.3 –9.5 26.0
Somalia 1982–1997 1,138–55,000 5,829 6,965 16.3% 45.9 –1.9 27.5
Uganda 1980–1986 75,000–250,000 12,138 15,520 21.8% 49.3 2.4 32.9
Mozambique 1976–1992 31,500 10,433 13,691 23.8% 43.3 10.6 23.3
Nicaragua 1982–1990 1,500 2,901 3,873 25.1% 36.8 –6.6 29.3
Ethiopia 1982–1991 50,000–250,000 40,359 54,761 26.3% 47.9 3.8 30.0
Colombia 1949–1958 100,000–180,000 11,332 15,447 26.6% 45.2 –2 .8 32.1
Iraq 1980–1988 85,000–336,000 12,768 17,568 27.3% 38.7 –7.9 30.7
Iran 1980–1988 13,700–15,000 37,839 54,479 30.5% 37.7 5.8 27.1
Vietnam 1965–1978 528,000 36,099 52,668 31.5% 34.7 –5.4 20.4
Sudan 1983–1997 2,000,000 20,367 32,511 37.4% 38.8 –1.1 26.7
Angola 1961–1994 546,000 4,797 9,419 49.1% 51.8 2.1 29.2
*In thousands.
Source: Maddison (2009), United Nations, Department of Economic and Social Affairs, Population Division (2011).
more elaborate justifications in terms of methods and sources makes

these methods hard to accept and subjects their results to considerable
distrust.
The numeric size of the demographic growth rates are an important charac-
teristic of states. In Table 4.1, for instance, Angola has a birth rate of 51.8 per
thousand, which feeds into a natural population growth rate of 29.2, the differ-
ence coming from the death rate.10 For traditional, or phase one, societies, the
large size of these rates are not unexpected: states with high death rates create
the “need” for high birth rates to compensate for losses that occurred during peri-
ods of violence. But for Angola, the ordinary death rate is so large that it effec-
tively obscures deaths that occurred during its repeated periods of violence. For
example, with the violence estimated at averaging over 16,000 deaths, in 1994, or
1.76 per thousand, it is difficult to extract the effect of the violence from fluctua-
tions in the ordinary, “routine” size of Angola’s death rate. The ordinary death rate
itself, of course, contains an estimation error. For developing countries, this error
is thought to lie between 3 and 8 percent and thus it can totally absorb or mask
genocide-related deaths.11
In Table 4.1, Bosnia and Rwanda show the highest levels of total population
losses at minus 24.6 percent and minus 23.3 percent, respectively. Emigration
stood at the extraordinarily high rate of 41.2 percent for Rwanda and a less dra-
matic but still very high 26.7 percent for Bosnia. Yet the demographic profile of
the two countries differs fundamentally.12 The Bosnian birth rate of 12.5 is much
lower than Rwanda’s 40.4, even as their natural rates of growth are quite simi-
lar: 4.4 and 3.2, respectively. Both countries’ populations are growing at roughly
the same rate, but Rwanda has vastly higher birth and death rates. This suggests
that the radical population decline in Rwanda likely stems from genocide-related
deaths and migration, whereas for Bosnia the likely cause is overwhelmingly
migration, not deaths. Correspondingly, the long-term effect of genocide-related
deaths should be more pronounced in Bosnia than in Rwanda. With relatively
low birth and death rates, the Bosnian population fundamentals do not have the
flexibility to mask the population outcome of the war. Small fluctuations in either
birth or death rates would easily be seen in national datasets and in the resulting
total population.
Bosnia, an example of a developed country that has already entered the fourth
phase of the demographic transition (rapid declines in birth and death rates),
would be more affected by violent conflict such as genocide than countries with
the phase one profile of a traditional society (high birth and death rates). Without
large-scale societal changes, including policy-induced changes, in average fam-
ily size, Bosnia thus has less ability than Rwanda to recover its pregenocide
population.
This discussion also helps to illustrate the scale of the Rwandan genocide in
relation to other states in its region—states with similar demographic structures.
Of all cases in Table 4.1, Rwanda is the only case that had a genocide-related death
(and migration) rate so large that it undercut its overall population growth; it is
likely that half its death rate in the event years was directly due to the genocide.
Thus, in countries that fit the phase one or phase two demographic profiles, only
violence events that resulted in extraordinarily high death rates will likely be
“seen” in national statistics. In the majority of our cases, however, the levels of
likely genocide-related deaths are simply not large enough to radically reorder
their population statistics or destinies.
4.2.3. Migration
Migration, a fundamental of population statistics, must be accounted for whenever
population losses or recovery are evaluated.13 From a national-level perspective,
total population losses are not simply those due to deaths or disruptions of births
but also to the effects of refugees who do not subsequently return. Yet in demo-
graphic studies, “migration” is often considered to be an error term with charac-
teristically rapid, large, and unexpected variance. Migration is difficult to measure
and, of the three major components of demography (birth, death, and migration),
it tends to have the highest level of inaccuracy. This measurement error can be
exacerbated in cases of genocide. For example, some death counts, arrived at from
a state’s underlying demographic structure, overestimate conflict mortality, with a
percentage of that population likely having migrated. Further, the policy of geno-
cide is often not simply slaughter but forced displacement beyond the boundaries
of the home country.14 The ability to return is, in part, a policy decision as well. It is
not simply a question of whether or not refugees would want to return home; it is a
question of if they can, if they should, and if it is safe to return. Postviolence politics
and policies of peace thus can exacerbate population losses.
In Table 4.1, most of the cases show a negative migration rate (i.e., emigra-
tion) that exceeds not only many of the natural rates of growth but even the
birth rates. Emigration in these cases is a rough indicator of the number of
refugees. In cases such as Bosnia, emigration likely far outweighed the conflict
deaths, driven in part by a program of ethnic cleansing, a process by which
communities were targeted in part to mobilize movement of their popula-
tions. To illustrate migration flows, the cases in Figure 4.1 are states that expe-
rienced at least five years of conflict followed by at least five years of peace.
Cases were chosen across a range of economic development and conflict types.
In the most prominent case, Rwanda, the emigration levels are very high, at
almost 75 per thousand, or roughly half a million people fleeing by the third
year into the conflict. That is a remarkable number in a country of just over
Migration rates per 1000 starting within 5 years of violence followed by 5 years of recovery
Migration rate
During violence Post-violence
75
65
Rwanda
55
45
Liberia Afghanistan
35
25
Lebanon Bosnia
15
Cambodia
5
–5 Angola
Burundi
–15
–25
–35
–45
–55
–65
–75
1 2 3 4 5 6 7 8 9 10
Years
Figure 4.1 Net migration rates (per 1,000 population). Source: Constructed from data in
UN (2011).
6.8 million people. For the whole of the conflict period, close to two million
people emigrated, likely an underestimation. Equally remarkable is the scale
of the return, as immigration rates are nearly as high as emigration rates. Each
of the eight cases in Figure 4.1 shows significant numbers of emigration fol-
lowed by immigration. For example, Afghanistan saw return swings similar to
Rwanda’s, as did Liberia. In Bosnia, however, the return immigration rates are
not equal to the emigration rate, which highlights the overall population loss
in that country. About five years after the end of hostilities, immigration rates
tend to fall off, likely meaning that those who wished to return did so within
the five-year time frame.15
4.3. The Profile of Population

4.3.1. Age Distribution
Population losses in genocide are not equally distributed by age. The younger
generations tend to bear the brunt of the violence, and with a higher degree of
mortality seen among males, but there is still some degree of variance of course,
and the state’s place in the population transition is itself an important character-
istic in this regard. In Figure 4.2, Cambodia, Rwanda, and Afghanistan exemplify
traditional societies. Their most populous generation is the youngest one, shown
Rwanda Cambodia
80+ 80+
70–74 70–74
Five-year age groups

60–64 60–64
50–54 50–54
40–44 40–44
30–34 30–34
20–24 20–24
10–14 10–14
0–4 0–4
750 500 250 0 250 500 750 750 500 250 0 250 500 750
In Thousands In Thousands
1995 Male 1989 Male 1995 Female 1989 Female 1980 Male 1970 Male 1980 Female 1970 Female
Afghanistan Bosnia
80+ 80+
70–74 70–74
60–64 60–64

50–54 50–54
40–44 40–44
30–34 30–34
20–24 20–24
10–14 10–14
0–4 0–4
1500 1000 500 0 500 1000 1500 175 125 75 25 25 75 125 175
In Thousands In Thousands
1989 Male 1979 Male 1989 Female 1979 Female 1995 Male 1992 Male 1995 Female 1992 Female
Figure 4.2 Pre-and postgenocide (or mass atrocities) population age-d istributions. Source: Constructed from data in UN (2011).
visually by steep pyramid-shaped age distributions (oldest at the top and young-
est at the bottom). In contrast, Bosnia’s age distribution is nearly cylinder-shaped,
as it had entered a demographic transition. Figure 4.2 also shows the population
effects of the genocidal period in each of the four states. The black-pyramid out-
line is the age distribution prior to genocide and the grey-colored outline is the
age distribution postgenocide. Thus, the difference is the total population loss,
stemming from deaths, birth disruptions, and migration.
In each of the four cases, the effects of genocide are starkly visible; the pre/
post age distributions, however, help one to distinguish among the cases. Both
Rwanda and Cambodia experienced large declines in their youngest generations.
These generations would have been expected to be the largest in the absence of
genocide. The losses are due to direct deaths and indirect deaths (as well as unre-
alized births). Indirect deaths, seen in increased infant mortality rates, stem from
the many aspects of a disrupted society such as limited food and lack of inocu-
lations and medical aid. In Afghanistan, in contrast, the youngest generation is
still the largest. Because losses occurred at each level of the age distribution, the
overall demographic profile remained unchanged. This is an example of a country
at the start of the demographic transition, likely just entering phase two (rapid
population increases).
4.3.2. Gender Ratio

Gender plays a fundamental role in society. The rise of women’s rights has the poten-
tial to accelerate economic growth and assist recovery efforts after wars. Gender
parity is not, however, generally seen in war or genocide. In war, it is generally
males who take most losses; in genocide, one might expect more parity, losses per-
haps even weighted toward women. The data, however, do not clearly illustrate this
point. The societies and the individual genocidal episodes themselves make this
choice, with women being underrepresented in mortality data even in the case of
major genocides. (On gender and genocide, see c hapter 17 in this volume.)
In a natural state, the human gender ratio is about 101 males to 100 females
(UN 1955b). With aging, the ratio becomes increasingly weighted toward females
due to their generally longer life expectancy. This affects the age distributions.
Depending on the phase of demographic development, ratios weighted toward
women start roughly at the fifty-to fifty-four-year-old category. The imbalance is
exacerbated in states that are in the latter stages of the demographic transition due
to rapid, but unequal, increases in life expectancy that favor females. Although
noticeable, the imbalance is not vast, with commonly seen male-to-female ratios
of 97 to 100 or even 95 to 100 or an increase in the birth to elder ratio of nearly six
additional women within the cohort.
In Cambodia, prior to its genocide, the age distribution at each five-year
cohort from zero to four years to forty to forty-four years saw the expected ratio
of about 101:100, very close to the expected natural state in a country with lim-
ited emigration and, equally importantly, limited sex selection. The postgeno-
cide demographic data then suggest that the perpetrators of the Cambodian
genocide made the decision to concentrate overwhelmingly on males, some-
thing not seen at a similar scale in our other cases. In Bosnia, we have a simi-
lar pregenocide condition with a normal gender ratio that becomes weighted
toward women by ages thirty to thirty-four. In both Afghanistan and Rwanda,
we see early skewing prior to the violence. Afghanistan had a gender ratio of
over 110:100 for people in their mid-t hirties, a likely sign of sex selection prac-
tices at birth (Guilmoto 2009). Rwanda’s ratio is skewed toward women from
between five to six additional women at each age distribution, a somewhat
unexpected observation.
Figure 4.3 illustrates on which sex and at what age the population losses are
concentrated. Cambodia’s case is the most dramatic, particularly for males in
their forties—a population that would have been well into its labor-force par-
ticipation years and likely the productive cornerstone of society. Population
losses in this age group are in the 50 to 70 percent range. A generally higher
degree of losses for males than for females is also seen in the cases of Bosnia
and Rwanda. In addition, both of these countries show losses of roughly 10 per-
cent of the young-adult population (people in their early to mid-twenties),
who most likely would have been active participants in the violence as well
as the most likely victims of selected killings. Only in Afghanistan are the
population losses nearly equally split between males and females, with females
0-4 5–9 10–14 15–19 20–24 25–29 30–34 35–39 40–44 45–49 50–54 55–59 60–64 65–69 70–74 75–79 80+
30%
20%
More
males Cambodia
10%
0%
Afghanistan
–10%
Bosnia
–20%
Fewer
males
–30%
–40%
–50%
–60% Rwanda
–70%
Figure 4.3 Percentage change in gender ratio by age distribution from genocide
inception to genocide end (five-year age groups). Source: Constructed from data in UN
(2011).
experiencing a slightly higher level of losses than in the other countries stud-
ied, perhaps due to continuous sex-selective birth practices even during the
years of violence. The wide range seen in the loss of elderly populations in both
Rwanda and Cambodia are probably due to the limited numbers of people over
eighty years of age, where even small absolute changes can cause large percent-
age changes.
Table 4.2 displays the cases numerically. In particular, in Cambodia the male-
to-female gender ratio of the population between 35 and 59 years of age ranges
from 75:100 to a sadly remarkable 59:100. The absolute scale and the proportional
imbalance of the losses are vast, comparable in current research with sex-selective
imbalances of 110:100 to 140:100 in some Chinese and Indian provinces as well
as in the Caucasus region (Das Gupta et al. 2003; Zhu, Li, and Hesketh 2009). It
is remarkably difficult to imbalance a population on this scale: it takes a sustained
societal choice of birth selection (or genocide), affecting the vast majority of the
population over several generational cohorts.16
Table 4.2 Postconflict Male-to-Female Gender Ratio
Age Rwanda Cambodia Afghanistan Bosnia

0–4 100 100 106 107
5–9 99 100 107 103
10–14 98 94 108 104
15–19 95 88 109 102
20–2 4 88 85 109 94
25–29 88 84 110 92
30–3 4 93 80 111 96
35–39 94 75 112 99
40–4 4 92 66 111 99
45–49 96 59 110 95
50–54 95 60 108 90
55–59 93 60 105 90
60–6 4 90 92 101 89
65– 69 81 84 97 78
70–74 65 80 93 54
75–79 50 99 89 51
80+ 39 79 84 57
Source: United Nations, Department of Economic and Social Affairs, Population Division (2011).
4.4. Population Recovery

A postgenocide, female-weighted population structure becomes an issue when
considering population recovery. While an overly biology-focused view would
ignore the critical roles of marriage structure (e.g., polygamy versus monogamy),
cultural expectations, and birth control, a population ratio overweighed toward
women improves the potential for population recovery more than a ratio over-
weighed toward men. Recovery remains key and is an underresearched aspect
of conflict studies. The potential for rapid, radical change to societies is likely the
outcome of the significant levels of destruction highlighted earlier in this chapter.
Kugler et al. (2013) show that population recovery after war lasts far longer than
war itself, and that the more developed countries have the potential to enter a
period of higher economic development than might have been expected prior to
war. One likely cause lies in the (even if unexpected) increase in postwar birth
rates. As Heuveline and Poch (2007, 409) note: “The extant record suggests that
such crises [e.g., war, famine, or epidemics] might induce a birth dearth, but that
afterward, fertility often rebounds to pre-crisis levels, sometimes temporarily
rising above those levels.” A youth surge, combined with investment in human
capital, eventually allows for a larger than expected and more productive labor
force—a demographic and economic dividend (Frumkin 1951; Urlanis 1971;
Bloom, Canning, and Sevilla 2003, 2004). In contrast, Kugler et al. (2013) also
show that while the outcome is more mixed in less-and least-developed nations,
a common characteristic is the poverty trap: a period of demographic recovery
without corresponding economic gains causes each generation to become poorer
than the prior one.
Regardless of economic development, nearly all societies that have suffered
significant losses do recover demographically. However, they do not all recover
in the same way or for the same reasons. The differences in demographic recovery
are important to understand. Due to the complexity of the proper measurement of
recovery, I use estimations of the expected total population to evaluate not simply
what became of the country postviolence but where it might have been without
the violence (see Kugler et al. 2013).17 The method allows a greater degree of con-
sideration of recovery as a separate issue, not simply recovery to the earlier status
quo, but recovery to where one would have been in the absence of genocide. This
method also combines losses from emigration, disrupted birth rates, and deaths.
Recovery is defined as reaching the expected prewar population forecast. Table
4.3 illustrates the descriptive relation between demographic transition phases and
either the pure recovery from war or the camouflaging effects of war. Initial post-
war growth, in column 2, is the average population growth rate, as a percentage,
for the recovery period. A key supposition is that if the country did not sustain
losses large enough to be measurable outside expected, normal population fluc-
tuations, then the recovery period is “completed” prior to the first postviolence
Table 4.3 Demographic Transition and Recovery after War

Percent per Years Needed for Recovery
Phase Year Growth Belligerents 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
3 0.656 Germany
4 0.899 Bosnia
3 1.185 Japan
2 1.519 Yemen
2 1.668 Vietnam
3 1.809 Lebanon
2 1.869 Burundi
2 2.062 China
2 2.169 Algeria
2 2.398 Nicaragua
2 2.439 Somalia
2 2.495 Iraq
2 2.512 Sudan
2 2.869 Iran
2 2.934 Angola
2 2.958 Colombia
2 3.123 South Korea
2 3.199 Cambodia
2 3.386 Ethiopia
2 3.631 Uganda
2 3.672 Mozambique
2 4.135 Liberia
2 4.336 Rwanda
2 7.736 Afghanistan
Key: Recovery in Process
Recovered
Percent per Year Growth: Average population growth during recovery years as percent.
Phase: Demographic Transition Phase (1–4). Estimated at beginning of recovery period.
Source: Author constructed numbers, Kugler at al (2012), Maddison (2009), United Nations, Department of
Economic and Social Affairs, Population Division (2011)
year (the empty lines in the table). In effect, the population consequence of the
violence was not large enough to directly affect the population destiny of the state
or could not otherwise be differentiated from its basic population foundations
(births, deaths, migrations). The table is capped at twenty years postviolence, or
one full generation. Even in the most severe cases, this is normally an adequate
time for population levels to recover. The table shows two distinct categories of
countries, one for states with recovered populations, the other for those still in
the recovery process.
Population recovery consists of gains due to increased domestic births as well
as from return migration.18 The majority of the cases are countries in the second
phase of the demographic transition, with very high birth rates and the possibility
of extremely high population growth. Note the different time-spans of popula-
tion recoveries. The postviolence population growth rates in Bosnia, Germany,
Japan, and Lebanon, all in phases three or four, range from 0.656 to 1.809 per-
cent per year. In contrast, except for Burundi, phase two countries experience
growth well above 2 percent. These differences illustrate the connections among
scale, development, and the possibilities of population recovery. Recovery itself is
a remarkably malleable thing. In ten of the cases—Yemen, Vietnam, Nicaragua,

Iraq, Sudan, Iran, Angola, Colombia, Ethiopia, and Uganda—the countries
had either such extreme levels of population growth or, possibly, lower scales of
losses that the expected future population was reached within the first postvio-
lence year, meaning that there was in effect no national-level demographic recov-
ery period at all. The population effects of the mass atrocities are “swept up” in
the overall population statistics. In Colombia, in particular, population effects
are hidden due to massive numbers of internal migrants. To see the effects there,
one would have to conduct demographic studies at the subnational, that is,
departmental or provincial, levels.
In the following five country examples, recovery to previolence population
forecasts has not occurred. In Bosnia, post–World War II Germany (despite its
postwar baby boom), and Lebanon, this is likely due to small population increases.
Japan also had a baby boom, large enough to more than compensate for its high
levels of losses and thus to allow for its recovery to be completed. Interestingly,
this recovery was then followed by one of the fastest recorded fertility declines.
For the case of Somalia, it is likely that its recovery period has not yet started. In a
word, some degree of recovery is common but not constant, and society’s stage of
development appears to play an important role in its potential recovery.
Figure 4.4 uses a subset of cases. In all states but Lebanon, population recovery
can be seen in a combination of return migration, elevated birth rates (i.e., baby
booms), and likely progression through phase two of the demographic transition.
This is not, however, a quick process. Nonetheless, most countries come reason-
ably close to full recovery within fifteen years postviolence. Both Cambodia and
Rwanda, the two largest in terms of losses, recovered fairly quickly (Cambodia
with a lag). Both countries had high birth rates and fairly high rates of return
migration. Similar effects can be seen in Afghanistan, Burundi, Japan, and Liberia.
Bosnia is a special case. Recall the earlier statement that it should be among the
most affected countries. First, the scale of its loss was quite large. Second, it had
progressed well into the demographic transition with limited population growth
and a near-stable population. A shock, such as the violence it experienced, was
irreversible without dramatic changes to the fertility rate or compensatory immi-
gration. As the world develops economically and demographically, the potential
for this special case to become the norm also increases.
4.5. Conclusion
Most countries do recover to the level of their previolence population forecasts.
A caveat is that this is primarily true for cases where the countries are in the early
phases of demographic transition (phases one or two), phases that are becoming
rare as the majority of states have either entered or soon will enter phase four
Percentage difference of expected population starting with 5 years of violence followed by 15 years of recovery
Percent During violence Post-violence
20
Afghanistan
15
Burundi
10
5
Japan Liberia Rwanda
0
Cambodia
–5
–10
Bosnia
–15
Lebanon
–20
–25
–30
–35
–40
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Years
Figure 4.4 The influence of mass atrocity on population and population recovery
(percentage difference of expected population relative to t = 0). Source: Author-constructed
from Kugler et al. (2013) and Maddison (2009).
(aging). Only in a handful of Central and South Asian states as well as in sub-
Saharan African countries are birth rates high enough to compensate for losses
that might be experienced in large-scale violence. Therefore, violent conflict with
mass atrocity increasingly is not an event that can be camouflaged by phase one or
phase two population growth. In a demographically developed (i.e., phase four)
world, future violence involving mass atrocity will result in a greater degree of
lasting population damage than before.
Notes
1. For a guide to the interaction between fertility and conflict see, e.g., Hill (2004) who
headed a round table jointly organized by the US National Research Council and the
Program on Forced Migration and Health at the Mailman School of Public Health at
Columbia University, New York. The outcome was a comprehensive review of the then-
extant literature and an evaluation of the complex relation between and among the various
factors that influence population and a call for a universally agreed-upon data collection
practice.
2. Galor (2011) is an exceptional example of an attempt to link all aspects of the human con-
dition under a single theoretical framework.
3. See Kugler (2016) for an overview of interdisciplinary demography. Also see Kugler and
Kugler (2013). Teitelbaum (1975) is of particular interest as he discusses the problematic
aspects of instituting policies, which may be based on demographic transition empirics,
but without having a clear understanding of the causality (reason) for observed population
outcomes.
4. See Kirk (1996) for an extensive overview of causality in demographic transition theory
and the attempts (economic, historical, and sociological) to explain complex interrela-
tionships. This corpus of work is still very much under development as the pattern of the
observed transitions are only insufficiently explained by theory.
5. The interactions and mutual influences between demography and development are a vast
topic well beyond the scope of this chapter. Starting in 1662 with John Graunt’s (1665)
project (primarily focused on methods to decrease mortality), then moving to Malthus
(expectations of population growth outpacing both technological innovation and eco-
nomic growth), demography has been a key consideration of development research. See
Goldstone, Kaufmann, and Toft (2012) and Bashford (2014) for additional explanations
of policy and of current understandings of the interaction among demography, economics,
and politics.
6. Tilly (1978, 19) quotes Carr-Saunders (1925, 40): “There is no great mystery about the
fall in the death-rate…. It was due to improved sanitary conditions and to advances in the
study of medicine.” This practical statement relatively early in the twentieth century has
continued to have a great deal of support. The largest declines in mortality are at the earli-
est levels (due to scale), and the technological necessities to achieve successes are reason-
ably small with policy and organization being the key. Sanitation, for example, is primarily
an issue of structure, not of cost.
7. For data sources and details, see chapter 3 of this volume, as well as Cunningham and
Lemke (2013).
8. Using the cases of Bosnia and Cambodia, Tabeau and Zwierzchowski (2013) help illus-
trate the complexity of the issue, showing the need to move beyond demography to a
mixed-methods approach. In the case of Cambodia, for example, the basic census data was
highly suspect. In contrast, in Bosnia a modern statistical apparatus was in place and thus
there existed a greater degree of data reliability. The problem with both was the issue of
migration as well as of scale.
9. See Lee (2011) for an overview of population projections and discussion on data
accuracy.
10. At the end of 1995 the world’s average birth rate per thousand was 24.0. At 51.8 per thou-
sand, Angola’s is over double that. In central Africa the average was 47, with Angola’s rate
being the highest in the region. Angola’s rate is certainly high in comparison to the world,
but not inordinately different from central Africa’s rate.
11. The United Nations (2002) “Methods for Testing Adult Mortality” (1) highlights the com-
plexity of mortality measurements: “In developed countries, adult mortality can be mea-
sured using data from civil registration systems and population estimates derived from
censuses or population registers. In most developing countries, however, the estimation
of adult mortality is seriously constrained by the absence of reliable, continuous, and com-
plete data registration systems. Most of our cases are well within the developing countries
category and some assumption of error is needed in our evaluations.”
12. Estimates of Bosnian casualties are still not agreed upon in the literature. Tabeau and
Zwierzchowski (2013, 216) report twelve different estimates, ranging from losses of
42,500 to 329,000 people. Their own, preferred estimate is a loss of 104,732 people.
13. Due to this chapter’s generalization across a wide range of countries and years, I use both
voluntary and forced migration (i.e., refugees) as the fundamental measure. Internally dis-
placed people (IDPs) are separate from refugees in domestic and international statistics
but better dealt with in works with a greater degree of country specificity. For a literature
review on who stays and who leaves in genocides and mass atrocities, see chapter 11 in this
volume.
14. In this regard, Greenhill (2010, 14) writes: “Dispossessed engineered migrations are those in
which the principal objective is the appropriation of the territory or property of another
group or groups, or the elimination of said group(s) as a threat to the ethnopolitical or
economic dominance of those engineering the (out-) migration; this includes what is
commonly known as ethnic cleansing. Explorative engineered migrations are those migra-
tions engineered either to fortify a domestic political position—or to discomfit or destabi-
lize foreign governments(s); militarized engineered migrations are those conducted, usually
during armed conflict, to gain military advantage against an adversary—v ia the disruption
or destruction of an opponent’s command and control.”
15. Angola, again, is an interesting case. Its high birth and death rates led to a high potential
for rapid population growth. Combined, the number of victims of violence there was large,
but the yearly rate was, on average, relatively low and lies within the error rate of the death
rate, a prime example of how underlying population fundamentals can hide crisis effects.
16. In this regard, research suggests that increased instability and security concerns can arise
from an overweighed male population (Hudson and den Boer 2002, 2004).
17. From Kugler et al. (2013, 6): “To forecast demographic trajectory, we use a linear extrapo-
lation. The basic reason is fit. Our forecast excludes possibilities of acceleration or unan-
ticipated declines that may take place during the expansion process and hide distortions
resulting from a fast demographic transition. More complex estimates would require a full
modeling of demographic structures—a nd we reserve that task for the future.”
18. The precise reasons for increased births, post-v iolence, are a topic for future research.
Likely reasons include a complex combination of trauma and psychological need, com-
bined with policy. For example, Fargues (2000) and Kaufmann (2010) refer to direct
attempts by ultra-Orthodox Jews to rebuild populations after the Holocaust as both policy
decision and religious need, particularly in the state of Israel. A similar policy impetus, car-
ried out via various measures such as monetary subsidies paid per child born, can be seen
in much of the world, primarily but not only for economic reasons, in countries such as
Italy, Japan, and South Korea. The effectiveness of national policies seems to be somewhat
limited, with more personal, and religiously motivated reasons in particular, showing a
higher degree of effectiveness on increased birth rates. See Bashford (2014) for a guide to
population policy globally.
References
Acemoglu, D., and S. Johnson. 2007. “Disease and Development: The Effect of Life Expectancy
on Economic Growth.” Journal of Political Economy 115, no. 6: 925–85.
Andreev, K., V. Kantorová, and J. Bongaarts. 2013. Demographic Components of Future Population
Growth. United Nations Department of Economic and Social Affairs. Population Division
Technical Paper No. 2013/3. New York: United Nations Publication.
Bashford, A. 2014. Global Population: History, Geopolitics, and Life on Earth. New York: Columbia
University Press.
Bloom, D. E., D. Canning, and J. Sevilla. 2003. “The Demographic Dividend: A New Perspective
on the Economic Consequences of Population Change.” Population Matters Series. Santa
Monica, CA: Rand.
Bloom, D. E., D. Canning, and J. Sevilla. 2004. “The Effect of Health on Economic
Growth: A Production Function Approach.” World Development 32, no. 1: 1–13.
Caldwell, J. C. 1976. “Toward a Restatement of Demographic Transition Theory.” Population and
Development Review 2, nos. 3/4: 321–6 6.
Carr-Saunders, A. M. 1925. Population. Oxford: Oxford University Press.
Cincotta, R. P., and J. Doces. 2012. “The Age-Structural Maturity Thesis: The Impact of the Youth
Bulge on Advent and Stability of Liberal Democracy.” In J. A. Goldstone, E. P. Kaufmann,
and M. Duffy Toft, eds., Political Demography: How Population Changes Are Reshaping
International Security and National Politics. Oxford: Oxford University Press, 98–116.
Cunningham, D. E., and D. Lemke. 2013. “Combining Civil and Interstate Wars.” International
Organization 67, no. 3: 609–27.
Das Gupta, M., J. Zhenghua, L. Bohua, X. Zhenming, W. Chung, and B. Hwa-Ok. 2003. “Why
Is Son Preference So Persistent in East and South Asia? A Cross-Country Study of China,
India, and the Republic of Korea.” Journal of Development Studies 40, no. 2: 153–87.
Dyson, T. 2010. Population and Development: The Demographic Transition. London: Zed Books.
Eastwood, R., and M. Lipton. 2011. “Demographic Transition in Sub-Saharan Africa: How Big
Will the Economic Dividend Be?” Population Studies 65, no. 1: 9–35.
Fargues, P. 2000. “Protracted National Conflict and Fertility Change: Palestinians and Israelis
in the Twentieth Century.” Population and Development 26, no. 3: 441–82.
Feng, Y., J. Kugler, and P. Zak. 2000.”The Politics of Fertility and Economic Development.”
International Studies Quarterly 42, no. 2: 667–94.
Frumkin, G. 1951. Population Changes in Europe since 1939. New York: Augustus M. Kelley.
Galor, O. 2011. Unified Growth Theory. Princeton, NJ: Princeton University Press.
Goldstone, J. A., E. P. Kaufmann, and M. D. Toft, eds. 2012. Political Demography: How
Population Changes Are Reshaping International Security and National Politics. Oxford: Oxford
University Press.
Gould, W. T. S. 2009. Population and Development. New York: Routledge.
Graunt, J. 1665. Natural and Political Observations Mentioned in a Following Index, and Made upon the
Bills of Mortality ... The Fourth Impression. Oxford: Printed by William Hall for John Martyn
and James Allestry, Printers to the Royal Society. Early English Books Online. http://eebo.
chadwyck.com/home [accessed July 7, 2013].
Greenhill, K. M. 2010. Weapons of Mass Migration. Ithaca, NY: Cornell University Press.
Guilmoto, C. Z. 2009. “The Sex Ratio Transition in Asia.” Population and Development Review 35,
no. 3: 519–49.
Heuveline, P., and B. Poch. 2007. “The Phoenix Population: Demographic Crisis and Rebound
in Cambodia.” Demography 44, no. 2: 405–2 6.
Hill, K. 2004. War, Humanitarian Crises, Population Displacement, and Fertility: A Review of the
Evidence. Washington, DC: National Academies Press.
Holck, S. E., and W. Cates Jr. 1982. “Fertility and Population Dynamics in Two Kampuchean
Refugee Camps.” Studies in Family Planning 13, no. 4: 118–2 4.
Hudson, V., and A. den Boer. 2002. “A Surplus of Men, A Deficit of Peace: Security and Sex
Ratios in Asia’s Largest States.” International Security 26, no. 4: 5–38.
Hudson, V., and A. den Boer. 2004. Bare Branches: The Security Implications of Asia’s Surplus Male
Population. Cambridge, MA: MIT Press.
Kaufmann, E. P. 2010. Shall the Religious Inherit the Earth? London: Profile Books.
Kirk, D. 1996. “Demographic Transition Theory.” Population Studies 50, no. 3: 361–87.
Kugler, T. 2016. “Demography and International Relations: Politics, Economics, Sociology and
Public Health.” In P. James and S. Yetiv, eds., Interdisciplinary Approaches to International
Studies. London, UK: Palgrave Macmillan.
Kugler, T., K. Kang, J. Kugler, M. Arbetman, and J. Thomas. 2013. “Demographic and Economic
Consequences of Conflict.” International Studies Quarterly 57, no. 1: 1–12.
Kugler, T., and J. Kugler. 2013. “Revised Political Demography.” In R. A. Denemark, ed., The
International Studies Compendium Project. Oxford: Wiley-Blackwell. Available online at
http://w ww.isacompendium.com/subscriber/u id=2857/?authstatuscode=202.
Lee, R. 2011. “The Outlook for Population Growth.” Science 333, no. 6042: 569–73.
Maddison, A. 2009. Historical Statistics of the World Economy: 1–2008 AD. http://w ww.ggdc.net/
maddison/[accessed May 24, 2010].
Montgomery, M. R., and B. Cohen, eds. 1998. From Death to Birth: Mortality Decline and
Reproductive Change. Washington, DC: National Academy Press.
Omran, A. R. 1982. “The Epidemiologic Transition.” In J. A. Ross, ed., International Encyclopedia
of Population. New York: Free Press, 172–83.
Preston, S. H. 1975. “The Changing Relation between Mortality and Level of Economic
Development.” Population Studies 29, no. 2: 231–4 8.
Restrepo, H. E., and M. Rozental. 1994. “The Social Impact of Aging Populations: Some Major
Issues.” Social Science and Medicine 39, no. 9: 1323–38.
Seybolt, T. B., J. D. Aronson, and B. Fischhoff. 2013. Counting Civilian Casualties: An
Introduction to Recording and Estimating Non-Military Deaths in Conflict. Oxford: Oxford
University Press.
Tabeau, E., and J. Zwierzchowski. 2013. “A Review of Estimation Methods for Victims of
the Bosnian War and the Khmer Rouge Regime.” In T. B. Seybolt, J. D. Aronson, and
B. Fischhoff, eds., Counting Civilian Casualties: An Introduction to Recording and Estimating
Nonmilitary Deaths in Conflict. Oxford: Oxford University Press, 213–43.
Teitelbaum, M. S. 1975. “Relevance of Demographic Transition Theory for Developing
Countries.” Science 188, no. 4187: 420–25.
Tilly, C., ed. 1978. Historical Studies of Changing Fertility. Princeton, NJ: Princeton
University Press.
[UN] United Nations. 1955a. “Methods of Appraisal of Quality of Basic Data for Population
Estimates.” Population Studies 23. Department of Economic and Social Affairs.
New York: United Nations Publication.
[UN] United Nations. 1955b. Manual II: Methods of Appraisal of Quality of Basic Data for Population
Estimates. New York: United Nations Publication.
[UN] United Nations. 2002. Methods for Estimating Adult Mortality. Working Paper No.
ESA/ P/
W P.175. Department of Economic and Social Affairs. Population Division.
New York: United Nations Publication.
[UN] United Nations, Department of Economic and Social Affairs, Population Division. 2011.
World Population Prospects: The 2010 Revision. CD-ROM Edition—E xtended Dataset in
Excel and ASCII formats. New York: United Nations Publication.
Urdal, H. 2006. “A Clash of Generations? Youth Bulges and Political Violence.” International
Studies Quarterly 50, no. 3: 607–29.
Urdal, H. 2012. “Youth Bulges and Violence.” In J. A. Goldstone, E. P. Kaufmann, and M. Duffy
Toft, eds., Political Demography: How Population Changes Are Reshaping International Security
and National Politics. Oxford: Oxford University Press, 117–32.
Urlanis, B. 1971. Wars and Population. Moscow: Progress Publishers.
Zhu, W. X., L. Li, and T. Hesketh. 2009. “China’s Excess Males, Sex Selective Abortion and One
Child Policy: Analysis of Data from 2005 National Inter-Census Survey.” British Medical
Journal 338, no. 7700: 1136–41.
5
The Macroeconomic Toll of Genocide and

the Sources of Economic Development
Di m i t r ios Sou dis, Robe rt I n k l a a r , a n d Robbe rt M a se l a n d
5.1. Introduction
The costs of conflict and mass killings are not limited to the immediate victims
of the violence. Episodes of violent conflict also have severe consequences for the
survivors, part of which is in lost economic output. In the short term, civil con-
flict is bad for the economic growth of afflicted countries (Collier 1999; Koubi
2005), and these negative effects can spill over to contingent states (De Groot
2010). After peace, some of the immediate negative effects of conflict disappear
and countries may find themselves in a recovery process, with higher than normal
growth rates (Organski and Kugler 1977; Abadie and Gardeazabal 2003; Koubi
2005). Others, though, have questioned whether economic activity recovers from
its initial fall (Mueller 2012). Conflict may cause permanent damage to a coun-
try’s social or physical infrastructure, so that it continues to experience adverse
consequences long after the termination of the conflict itself (Ghobarah, Huth,
and Russett 2003; Hoeffler and Reynal-Querol 2003).
Genocides have received much less attention than other types of conflict.
They are generally treated as separate events in the literature (Sambanis 2004) or
even as mere consequences of civil wars (Krain 1997; Stewart 2011). Yet, there is
ample reason to expect the economic consequences of genocides to be worthy of
independent analysis. First, genocides are characterized by a higher intensity of
violence that is more explicitly directed at the civilian population and its liveli-
hood. This results in a relatively higher loss of lives and thus of human capital, but
a possibly lower degree of destruction of physical capital. Since a loss of physi-
cal capital is easier to restore than losses in human capital, genocides are likely
to have more persistent negative effects and may generate less swift recoveries
than nongenocidal conflicts. Second, genocides may have different economic
effects through their impact on trust and the formal and informal institutions
125
underlying cooperation. Genocides are likely to be more divisive and tear up

the social fabric of society more completely than other forms of violent conflict.
Ethnic and communal violence may increase the degree of trust within the in-
group, at the expense of trust in out-g roups. The resulting damage to social infra-
structure increases transaction costs, reducing the scope for impersonal exchange
and worsening allocative efficiency (Wallis 2011).
Given the differences between genocides and other forms of conflict, there is
good reason to investigate genocides more systematically. This chapter does so by
providing a broad, cross-country empirical investigation. Thus, we aim to answer
the question of how genocides and mass political killings (henceforth both events
will be referred to as “genocides”) affect economic outcomes. Extant genocide lit-
erature mainly uses economic variables to predict or explain the occurrence of
genocide, although empirical support for this relationship is weak at best (Stewart
2011). There is not much work on the opposite relation: Do genocides affect eco-
nomic outcomes and, if so, how? What is more, the few studies that have addressed
this question focus on the Rwandan case, which might not be representative of
genocides in general. At the macro level, Lopez and Wodon (2005) argue that
Rwanda’s per capita gross domestic product (GDP) levels would have been 25 to
30 percent higher had the genocide of 1994 not happened. On the individual level,
Serneels and Verpoorten (2013) argue that, six years after the Rwandan genocide,
households that experienced more violence are lagging behind in terms of con-
sumption. However, Rogall and Yanagizawa-Drott (2013) find the exact opposite
result. The conflicting and country-specific evidence suggests there is a clear need
for more systematic analysis.
We provide such an analysis by combining data from the newly released version
8.0 of the Penn World Table (PWT; see Feenstra, Inklaar, and Timmer 2015) and
the Political Instability Task Force (Marshall, Gurr, and Harff 2014) data on
genocides and politicides, creating a dataset covering thirty-five episodes of geno-
cide in twenty-t hree countries. Our two goals are, first, to distinguish between
the short-term and long-term effects of genocides on economic outcomes; and,
second, to determine whether any negative effects on GDP can be traced to
decreases in capital or decreases in productivity.
Given the broader literature on the effects of conflicts on economies, it could
be that (1) genocides have only a transitory negative effect on economic activ-
ity, as a short-term decline is compensated by a longer-term recovery; (2) geno-
cides have a permanent adverse effect on the level of economic activity, and while
growth subsequently resumes on its old path, it thereby does not catch up to its
pregenocide trend level; or (3) genocide leads to a permanently lower growth
path. To distinguish among these three scenarios, we follow the methodology
of Cerra and Saxena (2008) and estimate an autoregressive (AR) model for eco-
nomic outcomes and a variable indicating when genocide starts. Using impulse
response functions (IRFs), we show that economic activity follows scenario (2): It
T h e M a c r o e c o n o m i c To l l o f G e n o c i d e 127
declines sharply at the start of a genocide and does not subsequently recover this
lost ground—a result in line with the findings for civil wars in Mueller (2012).
Furthermore, we find that the reduction in GDP per worker can be traced to a
decline in productivity, not to a decline in physical capital per worker. This leads
us to conclude that the start of genocide results in a significant disruption of the
economic process in the afflicted countries—leading to a permanent drop in the
level of productivity compared to a scenario without genocide—but not a long-
run destruction of growth potential. We note, though, that this is not a statement
of causality as we cannot rule out that genocide has its roots (partly) in economic
circumstances.
5.2. How Genocides May Effect

Economic Outcomes
Apart from their costs in terms of human suffering, genocides are likely to bring
about serious economic costs. A broad literature has documented the economic
effects of civil wars and conflicts, showing that civil conflict does considerable
contemporary harm to the economy (Collier 1999; Abadie and Gardeazabal
2003; Hoeffler and Reynal-Querol 2003; Koubi 2005; Lopez and Wodon 2005;
De Groot 2010; Serneels and Verpoorten 2013). Collier (1999) identifies three
main channels through which civil conflict may affect economic growth. First,
civil war may directly destroy capital, cause capital flight, and reduce savings,
eroding a country’s capital stock. Second, militarization and increased security
expenses imply that resources are diverted from productive activities toward
nonproductive activities, resulting in a lower factor of productivity. Third, higher
transaction and transport costs due to disruption of social and physical infra-
structure and the undermining of the state may result in a less efficient allocation
of productive factors (Collier 1999).
To formalize how these channels may be distinguished (in part), it is helpful
to consider a growth-accounting decomposition. This decomposition is origi-
nally attributed to Solow (1957) (see Hulten [2010] for a comprehensive and
modern survey). Assume that the output of an economy, Y, is produced using
capital, K, labor, L, and with productivity level, A, in a Cobb-Douglas produc-
tion function:
Y = AK α L1 − α.(1)
The output of an economy is its level of GDP, the capital stock consists of cumu-
lated past investments in buildings and machinery, labor is the number of work-
ers in the economy, α is the output elasticity of capital, and productivity is the
efficiency with which capital and labor are combined into output. This concept
of productivity is usually referred to as total factor productivity (TFP). Equation

(1) can be expressed in terms of growth rates of output per worker y ≡ Y ∕ L as:
Δ log y = α Δ log k + Δ log A , (2)
where k ≡ K ∕ L. This implies that the growth of GDP per worker, Δ log y, on the
left-hand side of equation (2) can be expressed as the contribution from changes
in the capital stock per worker plus TFP growth on the right-hand side. In the
original work by Solow (1957), TFP growth was given the interpretation of tech-
nological change, but this interpretation relies on strict assumptions such as
perfect competition. More generally, TFP growth is computed as a residual; and
Hulten (2010) discusses that TFP growth can reflect much more, including the
efficiency of resource allocation (see also Fernald and Neiman [2011] more spe-
cifically on this point). Indeed, the fact that TFP is measured as a residual means
that we can empirically distinguish only the effect of genocide on the capital/
labor ratio (k) from the effect on all other factors influencing economic activity.
Seen through this lens, all three channels identified by Collier (1999) imply
that economic activity is negatively affected by conflict; but in the case of direct
destruction (the first channel), this is because the input of capital is reduced. In
the case of the other two channels—m ilitarization and security spending, and
transaction and transport costs—t he effect would be seen in the country’s pro-
ductivity. The shift toward military activities and/or higher transaction costs and
social, institutional, and physical disruption would lead to a less efficient use of a
given set of inputs, and so result in a reduction in TFP, the second component on
the right-hand side of equation (2).
Leaving aside the precise channels, it seems safe to conclude that there is a
net negative contemporaneous effect of civil conflict on the economy. However,
the longer-term effects are more hotly debated (Cerra and Saxena 2008; Chen,
Loayza, and Reynal-Querol 2008; Voors et al. 2012). Collier (1999) maintains
that the main harm done through capital flight and dissaving is reversible once
peace is restored. In his argument, agents respond to the fall in productivity from
disruption during a conflict by moving capital out of the country. When the con-
flict’s immediate negative effect on productivity is lifted, the resulting lack of
capital in postconflict economies makes returns on investment relatively high.
A recovery phase may set in, with temporarily higher growth rates. The occur-
rence of such a recovery phase depends on the duration of the conflict, however. If
a conflict constitutes a short, negative shock to the economy, the economy has not
yet completely adjusted to the conflict-time equilibrium when peace is restored. If
the war is long, there is a higher chance of seeing a recovery, as all negative adjust-
ments have been made and the economy can begin anew. Studying the effects of
the long-r unning conflict in the Basque region of Spain, Abadie and Gardeazabal
(2003) indeed find empirical support for a recovery after a truce is called.
Others have argued that conflicts continue to do harm long after the fighting
has stopped, in terms of both casualties and economic costs (Ghobarah, Huth,
and Russett 2003; Hoeffler and Reynal-Querol 2003). Part of the reason is that
while the destruction and diversion of capital associated with violent conflict may
have ended, the damage done to a country’s social infrastructure and to the legiti-
macy and effectiveness of the state may be permanent. The ethnic nature of vio-
lence in genocides is especially likely to do lasting harm to social ties, trust, and
institutions (Rohner, Thoenig, and Zilibotti 2013; Bauer et al. 2014). This matters
because the decline in institutions that facilitate interactions between individuals
and groups in the economy erodes the potential for cooperation, exchange, and
specialization, with adverse economic consequences (North 1990; Putnam 1993;
Easterly 2001; Wallis 2011). Indeed, a recurrent result in the empirical growth
literature is that past shocks to institutional quality often have enduring effects on
future economic performance (Guiso, Sapienza, and Zingales 2008; Nunn 2009;
Spolaore and Wacziarg 2013). There is also increasing evidence that violent con-
flict may change the preferences of individuals, inducing them to become more
risk-seeking and developing higher discount rates, thus pushing down savings
rates permanently (Voors et al. 2012; Bauer et al. 2014; Callen et al. 2014). For
these reasons, the economic damage done by conflict may not be easily restored.
If the erosion of institutions and social capital affects the potential for adaptation
and innovation, it may even set economies on a permanently lowered growth path.
Genocidal conflicts may be particularly prone to have such long-term effects.
Aimed less at harming an opponent’s fighting capacity and more at eliminating
entire segments of the civilian population, the effects of genocides on physical
capital may be relatively limited compared to regular violent conflict, while the
costs in terms of human capital are relatively high (Serneels and Verpoorten
2013). As a consequence, genocides may have more persistent income effects. The
ethnic or communal component of most genocides is also likely to do particular
harm to societal institutions and social capital. For these reasons, we expect geno-
cides to have stronger long-term effects on productivity and output.
On the basis of the arguments discussed here, we identify three possible sce-
narios for growth after genocide (see Figure 5.1). The first is a recovery scenario
(left-hand panel), wherein the decline in productivity and output occurring dur-
ing genocide is only temporary, and societies bounce back to the previous growth
trajectory after peace is restored. In this scenario, genocides have the same effects
as mild civil conflicts, causing contemporary diversion and destruction during
the conflict itself but doing no permanent damage to the economy.
The second scenario (middle panel) is that of a onetime permanent drop in
productivity levels, with no changes in subsequent growth rates but without ever
catching up to the prior trend line of economic growth. This may be due to last-
ing damage to institutions and social capital, causing a reduction in allocative
efficiency.
Y Y Y
Time Time Time

Scenario 1 Scenario 2 Scenario 3
Figure 5.1 Growth after genocide—possible scenarios.
Scenario three (right-hand panel) is the one in which genocides set societies on
a permanently lower growth path. Adverse effects of genocides on savings rates,
innovation, and adaptive efficiency of a society’s institutions may cause such
long-term effects. The aim of our empirical analysis is to determine which of these
scenarios best fits the economic experience of countries that have experienced
genocide.
5.3. Data and Methods

To quantify the effect of genocide on economic performance, we combine three
datasets. Our economic variables come from the PWT, version 8.0, covering
167 countries between 1950 and 2011 (Feenstra, Inklaar, and Timmer 2015).
We calculate the annual growth rate of GDP per worker for each country and
decompose it into the growth of physical capital per worker, plus the growth
in TFP, using the growth accounting decomposition on the right-hand side of
equation (2). The PWT also provides its own growth accounting decomposition
(see Feenstra, Inklaar, and Timmer 2015), but coverage of countries is incomplete
due to missing data on human capital and capital income shares. We therefore use
the average capital share as our estimate of α, used to weight the growth of (physi-
cal) capital per worker and construct TFP growth as a residual. We use all three
indicators as dependent variables in order to identify the channel through which
genocides affect growth.
While these three indicators provide a more detailed view of the economic
consequences of genocide, they are not without limitations. In general, data
collected in a turbulent period may be subject to greater measurement error,
although without a bias in a specific direction this will only make it harder to draw
any type of statistically valid conclusion. It could be that demographic informa-
tion, and thus data on the number of workers that we use to scale GDP and capital,
is particularly less reliable around genocide (on the demography of genocide, see
chapter 4 in this volume). We will thus also estimate a model with GDP growth as
the dependent variable.
A final remark on the growth data concerns the data on capital. The capital data
from the PWT are based on accumulated and depreciated past investment, which
means that direct destruction of capital is not measured. The impact of such
destruction will be apparent in TFP growth, since that is measured as a residual.
However, destruction of capital would, ceteris paribus, raise the marginal pro-
ductivity of new investments and stimulate faster growth in capital in the future.
This would show up as a delayed positive effect of genocide on capital per worker
growth and would, in principle, be captured by our analysis.
Our genocide indicator comes from the Genocide and Politicide Problem
Set produced by the Political Instability Task Force (PITF). The PITF provides
annual information on genocides and politicides for all countries with a total pop-
ulation of 500,000 or greater, covering the period from 1955 to 2011. (On geno-
cide and mass atrocity data generally, see c hapter 3 in this volume.) Genocides in
our study “involve the promotion, execution, and/or implied consent of sustained
policies by governing elites or their agents or in the case of civil war, either of the
contending authorities that result in the deaths of a substantial portion of a com-
munal group or politicized non-communal group” (Marshall, Gurr, and Harff
2014, 14). The victimized groups are defined both in terms of their communal
(ethnolinguistic, religious) characteristics (in the case of genocides) as well as
their political opposition to the regime and dominant groups (for politicides). The
PITF dataset covers 35 episodes of genocide across 23 countries with a median
length of 5.5 years.
Our estimation is based on the approach used by Cerra and Saxena (2008).
We use an autoregressive distributed lag model to estimate the effects of genocide
on our three dependent variables and subsequently construct IRFs to represent
these effects graphically. In particular we estimate:
3 3
zit = βi + ∑ βjzi , t − j + ∑ δG + εit , (3)
j s=0 s i , t −s
where zit is one of the three measures of economic outcomes for country i at time
t (i.e., Δ log yit, Δ log kit, or Δ log Ait) and G is an indicator coding whether a geno-
cide started in year t in country i. As argued by Mueller (2012), this start-year coding
makes it possible to correctly assess the economic impact of a genocide (or other con-
flict), in contrast with coding all years in an episode as in Cerra and Saxena (2008).
Our choice of lag-length is based on the statistical significance of the lagged coef-
ficients of the dependent variables; after the third lag, none of the coefficients reach
conventional levels of statistical significance. The IRFs are graphed together with a
95 percent confidence interval based on the results of 1,000 bootstrapped samples.
This empirical strategy is well suited to identify not only the initial effect of
genocide but also to trace its dynamic impact on the economy. However, it does
not say anything about causality. We effectively have to assume that genocide
starts at a random point in time and then, having started, we can determine how
the economy evolves. It could, of course, be the case that genocide has (in part)
roots in previous economic developments—a lthough Stewart (2011) finds no
clear evidence of this—and in that case our estimates will be biased. Our analysis
can thus best be compared with event studies in finance: Given the start of geno-
cide, how does the economy evolve in subsequent years?
5.4. Results
Table 5.1 shows summary statistics for the three growth indicators for our main
sample of 23 countries that experienced genocide and, for reference, the statistics
for all 167 countries in the PWT. Moreover, the statistics for countries that expe-
rienced genocide are split between periods—when there was genocide, and when
there was not (see Table 5.3 for the list of genocide countries and periods). In
nongenocidal years, genocide countries experience, on average, somewhat higher
growth in GDP per worker than the average PWT country—namely, 2.7 percent
average annual growth versus 1.9 percent—and correspondingly faster growth
in capital per worker (2.6 versus 2.2 percent) and faster TFP growth (1.5 versus
0.9 percent). However, this changes drastically when focusing on the genocide
years: growth in GDP per worker is –0.3 percent on average, growth in capital per
worker is only 1.9 percent, and TFP growth is –1.2 percent. Note, though, that
variation around the average growth rates tends be higher in genocide countries
Table 5.1 Summary Statistics for Growth
Genocide Countries All Countries
Growth in: Nongenocidal Genocidal Period

Period
GDP per worker 0.027 (0.082) –0.003 (0.088) 0.019 (0.063)
Capital per worker 0.026 (0.062) 0.019 (0.045) 0.022 (0.043)
Total factor 0.015 (0.073) –0.012 (0.084) 0.009 (0.058)
productivity
No. of observations 861 204 7,026
Notes: The table shows average growth, with the standard deviation in parentheses. Statistics
for “genocide countries” are based on data from PWT (version 8.0) for the 23 countries that
experienced genocide since 1955 (see Table 5.3), with genocidal period(s) as shown in Table 5.3.
Statistics for “all countries” cover all 167 countries in the PWT.
than in nongenocide countries, and observations for all countries and years are
simply lumped together. So while these summary statistics are surely suggestive
of sharply lower growth during genocides, a formal analysis is needed to more
firmly establish any effect.
We present the results from the estimation of equation (3) in the form of cumu-
lative impulse responses. Figure 5.2 depicts the effect of the start of genocide on
the growth of GDP per worker and its constitutive parts. The left panel shows
that in the first three years after the start of genocide, GDP per worker declines,
on average, by 9.7 percent. This decline is statistically significant, as indicated by
the 95 percent confidence interval, although the broad-w idth confidence inter-
val could indicate that there is considerable variation in the impact of genocides
across countries. After the initial decline, there is very little change in the cumula-
tive response function, indicating that the loss in the level of GDP per worker is
not recovered in subsequent years. This outcome is consistent with Scenario 2 in
Figure 5.1, as Scenario 1 would imply a total recovery from the negative shock of
genocide; while for Scenario 3, the cumulative effect should continue to decline.
This impulse response function (IRF) is estimated based on changes in GDP per
worker, to match the concept in the growth accounting framework, but we have
also estimated the same model on changes in the level of GDP per se. The result-
ing IRF matches the IRF for GDP per worker quite closely, as shown in Figure 5.3.
Next, we look at the effects of genocide on the growth of capital per worker and
on TFP. Starting with the latter, the IRF for TFP is essentially identical to that for
GDP/worker growth TFP growth Capital/worker growth
0 0 0
Cumulative effect on growth
–5% –5% –5%
–10% –10% –10%
–15% –15% –15%
0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10
Years since start genocide Years since start genocide Years since start genocide
Figure 5.2 The effect of the start of genocide on GDP per worker and its constituent
parts. Notes: Impulse response functions (solid lines) represent GDP/worker growth and its
constituent parts. The vertical axis shows the cumulative effect on growth, while the horizontal axis
indicates the number of years since the start of genocide. Dashed lines show 95 percent confidence
intervals.
GDP growth
0
–5%
Cumulative effect on growth
–10%
–15%
–20%
0 2 4 6 8 10
Years since start genocide
Figure 5.3 The effect of the start of genocide on GDP. Notes: Impulse response functions
represent growth in GDP (solid lines). The vertical axis shows the cumulative effect on growth, while the
horizontal axis indicates the number of years since the start of genocide. Dashed lines show 95 percent
confidence intervals.
GDP per worker, with a negative cumulative effect after three years of 9.5 percent.
In contrast, the confidence interval in the IRF for physical capital is on both sides
of zero, indicating no statistically significant effect for the start of genocide on
capital per worker. As discussed before, our measure of capital is accumulated
and depreciated past investment and does not take into account any physical
destruction. However, physical destruction would, under normal circumstances,
increase the marginal product of capital and thus stimulate investment. As a
result, capital would increase with some lag following any physical destruction
unless a perfectly countervailing decline in TFP would reduce the marginal prod-
uct of capital to remove the incentive to invest. So our finding—t hat there is no
effect on capital—suggests that physical destruction alone cannot account for the
decline in GDP. If there were significant physical destruction, it would have to be
in fine balance with the loss of TFP, since there is no significantly higher or lower
investment with a lag.
To check on the robustness of these results and to understand whether vari-
ables such as the duration of genocide or its intensity—as measured by the num-
ber of deaths as a percentage of the population—have a moderating effect on the
severity of the economic impact, we repeat the previous estimation exercises
using subsamples of the data. In particular, we calculate the median duration of
genocide (in years) and the median magnitude of deaths and estimate IRFs for the
group of countries that fall above or below each median. Table 5.2 summarizes
Table 5.2 The Effect of Genocide on Growth of GDP per Worker in the Short
Run and Long Run by Genocide Characteristics
Short Run (Year 1) Long Run (Year 10) Leveling off

All Genocides – 0 .043 (–0.076, – 0.096 (–0.193, Yes
–0.006) –0.018)
High –0.047 (–0.126, –0.134 (–0.293, Yes
Casualties –0.033) –0.038)
Low –0.044 (–0.067, –0.067 (–0.159, Yes
Casualties –0.021) –0.030)
Short –0.045 (–0.102, –0.057 (–0.209, Yes
Duration –0.005) –0.003)
Long –0.045 (–0.062, –0.160 (–0.285, Yes
Duration –0.000) –0.044)
Notes: The table shows the effects of genocide on the growth of GDP per worker at Years 1 and
10 (as well as the 95 percent confidence intervals in parentheses), using the IRF estimates obtained
by equation (3) estimated on subsamples, split by the median death toll and by the median dura-
tion of a genocide.
the main results for GDP per worker, with the short-r un effect in the first year
after the start of genocide and the long-r un effect ten years after the start of geno-
cide. As the table shows, the short-r un effects are all very similar, with a decline in
GDP per worker of approximately 4.5 percent. There is more variation in the long-
run effects, with the point estimates showing a decline of 13.4 percent for geno-
cides with above-median deaths and a decline of 6.7 percent in genocides with
below-median deaths. Likewise, in genocides with duration above the median of
5.5 years, the long-r un effect is larger, with a 16 percent decline, than is the nega-
tive effect in briefer genocides of 5.7 percent. However, the smaller sample size of
genocides in these subsamples means that it is harder to draw statistically signifi-
cant contrasts between these effects.
Our results should be interpreted with caution since it is empirically not fea-
sible to separate the effects of genocide from civil war. Table 5.3 presents the
episodes of genocide in our dataset where, in the last column, we show the num-
ber of genocide years that overlapped with a civil war. In our sample there are
only two countries (Ethiopia and Cambodia) where the two crises do not over-
lap. On average, about 60 percent of country-year episodes of genocide overlap
with a civil war. Those that do not most often take place before, after, or between
recurring episodes of civil war. Controlling for civil war and estimating condi-
tional IRFs does not change the results presented above, but we do not want to
claim that this empirical strategy fully disentangles the effects of the two types
of conflict.
Table 5.3 Genocidal Episodes and Overlap with Civil Wars
Country Genocidal Period Duration Civil War

Overlap (no. of
years)
Angola 1975–1994, 1998–2002 25 7
Argentina 1976–1980 5 2
Bosnia and 1992–1995 4 3
Herzegovina
Burundi 1965–1973, 1988, 1993 11 1
Cambodia 1975–1979 5 No
Chile 1973–1976 4 1
China 1959, 1966–1975 11 1
Congo (Dem. 1963, 1965, 1977–1979 5 No data
Republic)
El Salvador 1980–1989 10 10
Equatorial Guinea 1969–1979 11 1
Ethiopia 1976–1978 4 No
Guatemala 1978–1990 13 13
Indonesia 1965–1966, 1975–1992 20 19
Iran 1981–1992 12 11
Iraq 1964–1975, 1988–1991 17 15
Nigeria 1967–1970 4 4
Pakistan 1971, 1973–1977 6 5
Philippines 1972–1976 5 5
Rwanda 1963–1964, 1994 3 1
Sri Lanka 1989–1990, 2008–2009 15 2
Sudan 1956–1972, 1983–2011 46 No data
Syria 1981–1982 2 2
Uganda 1971–1986 16 9
Source: Data from Political Instability Task Force (Marshall, Gurr, and Harff 2014).
Similarly, it is very difficult to establish through what channel(s) TFP growth

is reduced. One prominent potential channel is a decrease in the degree of trust
within society or even a broader loss of social capital. However, systematic time-
varying data is not available to test this. There is more information about political
institutions that might serve as a proxy for their capacity to exchange and work
together with other parts of society. In particular, data on the competitiveness

of the political environment—t hat is, whether executives are openly recruited as
opposed to whether there is an elite that rules over the state—may provide rel-
evant information (data from Marshall, Gurr, and Jaggers 2014). However, in our
sample of twenty-t hree countries, only five experienced a reduction in political
competitiveness at or around the time of genocide. This small sample size pre-
vents us from testing the proposed relationship.
Summing up our results, genocides act as a negative shock to the growth path
of afflicted countries through the effects they have on productivity. This has a
permanent negative effect on the level of economic activity and productivity; the
growth rates do return to their previous path, but at a lower level of GDP than
before.
5.5. Concluding Remarks

Genocides are not just atrocious events that exact a great toll of human suffer-
ing, but they also have a negative effect on economic activity. We find that in the
first three years after the start of genocide, GDP per worker decreases by about
10 percent. In contrast to civil wars, where somewhat smaller effects disappear
over the long run, this negative effect of genocide on the economy persists for ten
or more years. Some of the initial economic damage of violent conflicts is subse-
quently restored, but for genocides this apparently is an incomplete process. We
present evidence that the income reduction is primarily caused by a reduction in
TFP, that is, the efficiency with which production factors are used in the econ-
omy. Production factors per se are much less affected, economically speaking. All
these results are in line with the interpretation that genocide typically tears social
bonds in society and harms social capital in a more pervasive fashion than does
civil war, thereby reducing allocative efficiency.
Some caveats are in order. First, although our results fit an erosion of social
capital narrative, direct proof that the reduction in TFP growth is attributable to
a reduction in social capital cannot be provided because of lack of data. Reliable
historical information on trust and social capital is not available. Second, while
our analysis focuses on a systematic assessment of the economic effects of geno-
cide in general, we acknowledge that each episode of genocide is a historically
unique event. Economic impacts are likely to vary accordingly. The fact that our
figure for the economic costs of genocides differs substantially from the reduc-
tion of 25 to 30 percent found by Lopez and Wodon (2005) for the Rwandan
genocide attests to this. Genocides differ on dimensions such as the number of
people affected, geographic concentration, duration and intensity, and political
outcome. However, while genocides may not be rare (see chapter 3), there are still
too few of them to allow for a systematic analysis of the effect of these differences
on economic outcomes. Third, as indicated, our analysis of the economic damage

associated with genocide should not be read as a statement on causality. Although
evidence points to the contrary (Stewart 2011), we cannot rule out that genocide
has its roots (partly) in economic circumstances. (For literature reviews on eco-
nomic correlates, causes, and consequences of genocide, see chapters 9, 10, 21, 22,
and 24 in this volume.)
These caveats notwithstanding, the results presented in this chapter strongly
contribute to the view that genocides carry serious economic consequences that
go beyond those of “regular” civil wars. The finding of lasting negative effects
makes the avoidance of genocide even more important and suggests that poli-
cies after genocide should be firmly focused on restoring some of the broken ties
between communities.
References
Abadie, A., and J. Gardeazabal. 2003. “The Economic Costs of Conflict: A Case Study of the
Basque Country.” American Economic Review 93, no. 1: 113–32.
Bauer, M., A. Cassar, J. Chytilová, and J. Henrich. 2014. “War’s Enduring Effects on the
Development of Egalitarian Motivations and In-Group Biases.” Psychological Science 25, no.
1: 47–57.
Callen, M., M. Isaqzadeh, J. D. Long, and C. Sprenger. 2014. “Violence and Risk
Preference: Experimental Evidence from Afghanistan.” American Economic Review 104, no.
1: 123–4 8.
Cerra, V., and S. C. Saxena. 2008. “Growth Dynamics: The Myth of Economic Recovery.”
Chen, S., N. V. Loayza, and M. Reynal-Querol. 2008. “Aftermath of Civil War.” World Bank
Economic Review 22, no. 1: 63–85.
Collier, P. 1999. “On the Economic Consequences of Civil War.” Oxford Economic Papers 51, no.
1:168–83.
De Groot, O. J. 2010. “The Spillover Effects of Conflict on Economic Growth in Neighbouring
Countries in Africa.” Defence and Peace Economics 21, no. 2: 149–6 4.
Easterly, W. 2001. “Can Institutions Resolve Ethnic Conflict?” Economic Development and
Cultural Change 49, no. 4: 687–706.
Feenstra, R. C., R. Inklaar, and M. Timmer. 2015. “The Next Generation of the Penn World
Table.” American Economic Review 105, no. 10: 3150–82.
Fernald, J., and B. Neiman. 2011. “Growth Accounting with Misallocation: Or, Doing Less with
More in Singapore.” American Economic Journal: Macroeconomics 3, no. 1: 29–74.
Ghobarah, H. A., P. Huth, and B. Russett. 2003. “Civil Wars Kill and Maim People—L ong after
the Shooting Stops.” American Political Science Review 97, no. 2: 189–2 02.
Guiso, L., P. Sapienza, and L. Zingales. 2008. Long Term Persistence. National Bureau of Economic
Research (NBER). Working Paper No. 14278. Cambridge, MA: NBER.
Hoeffler, A., and M. Reynal-Querol. 2003. “Measuring the Costs of Conflict.” Oxford: Centre
for the Studies of African Economies. University of Oxford. Oxford: UK. http://conflic-
trecovery.org/ bin/2 003_ Hoeff ler_ R eynal-M easuring _ t he_C osts_of_C onf lict.pdf
[accessed December 22, 2015].
Hulten, C. M. 2010. “Growth Accounting.” In B. H. Hall and N. Rosenberg, eds., Handbook of the
Economics of Innovation. Vol. 2. Amsterdam: Elsevier, 987–1031.
Inklaar, R., and M. P. Timmer. 2013. Capital, Labor, and TFP in PWT8.0. Mimeo. Downloadable
from www.ggdc.net/pwt.
Koubi, V. 2005. “War and Economic Performance.” Journal of Peace Research 42, no. 1: 67–82.
Krain, M. 1997. “State-Sponsored Mass Murder: The Onset and Severity of Genocides and
Politicides.” Journal of Conflict Resolution 41, no. 3: 331–6 0.
Lopez, H., and Q. Wodon. 2005. “The Economic Impact of Armed Conflict in Rwanda.” Journal
of African Economies 14, no. 4: 586–6 02.
Marshall, M. G., T. R. Gurr, and B. Harff. 2014. PITF–State Failure Problem Set: Internal Wars
and Failures of Governance, 1955–2013. Vienna, VA: Center for Systemic Peace, http://w ww.
systemicpeace.org/i nscr/PITFProbSetCodebook2013.pdf.
Marshall, M. G., T. R. Gurr, and K. Jaggers. 2014. Polity IV Project: Political Regime Characteristics
and Transitions, 1800–2013. Vienna, VA: Center for Systemic Peace, www.systemicpeace.
org/polity/polity4.htm.
Mueller, H. 2012. “Growth Dynamics: The Myth of Economic Recovery: Comment.” American
North, D. C. 1990. Institutions, Institutional Change and Economic Performance. New York: Cambridge
University Press.
Nunn, N. 2009. “The Importance of History for Economic Development.” Annual Review of
Economics 1, no. 1: 62–92.
Organski, A. F. K., and J. Kugler. 1977. “The Costs of Major Wars: The Phoenix Factor.” American
Political Science Review 71, no. 4: 1347–6 6.
Putnam, R. D. 1993. Making Democracy Work: Civic Traditions in Modern Italy. Princeton,
NJ: Princeton University Press.
Rogall, T., and D. Yanagizawa-Drott. 2013. “The Legacy of Political Mass Killings: Evidence
from the Rwandan Genocide.” http://w ww.hks.harvard.edu/fs/dyanagi/R esearch/
Legacy_of_Rwandan_Genocide.pdf.
Rohner, D., M. Thoenig, and F. Zilibotti. 2013. “Seeds of Distrust: Conflict in Uganda.” Journal
Sambanis, N. 2004. “Using Case Studies to Expand Economic Models of Civil War.” Perspectives
on Politics 2, no. 2: 259–79.
Serneels, P., and M. Verpoorten. 2013. “The Impact of Armed Conflict on Economic
Performance: Evidence from Rwanda.” Journal of Conflict Resolution. Published
online: http://jcr.sagepub.com/content/early/2 013/12/13/0 022002713515409.abstract.
Solow, R. M. 1957. “Technical Change and the Aggregate Production Function.” Review of
Economics and Statistics 39, no. 3: 312–2 0.
Spolaore, E., and R. Wacziarg. 2013. “How Deep Are the Roots of Economic Development?”
Journal of Economic Literature 51, no. 2: 325–69.
Stewart, F. 2011. “Economic and Political Causes of Genocidal Violence: A Comparison with
Findings on the Causes of Civil War.” MICROCON Research Working Paper 46. Brighton,
UK: MICROCON.
Voors, M. J., E. Nillesen, P. Verwimp, E. H. Bulte, R. Lensink, and D. P. Van Soest. 2012. “Violent
Conflict and Behavior: A Field Experiment in Burundi.” American Economic Review 102, no.
2: 941–6 4.
Wallis, J. 2011. “Institutions, Organizations, Impersonality and Interests: The Dynamics of
Institutions.” Journal of Economic Behavior and Organization 79, nos. 1–2: 48–6 4.
PA R T T W O
THEORETICAL APPROACHES
AND REVIEWS OF EMPIRICAL
LITERATURE
6
Genocide and Mass Killing Risk and

Prevention
Perspectives from Constrained Optimization Models
C h a r l e s H . A n de rton a n d J u rge n Br au e r
6.1. Introduction
To highlight an economic way of thinking about mass killing, Ferrero (2013,
333) posed an important question: “Are rulers who kill more people more evil
than those who kill fewer, or is their choice driven by opportunity costs?” Put
differently, are those rulers who kill few people less evil than those who kill many,
or do they simply face different constraints that direct their ultimate choice of
action? Is mass killing a ruler’s direct objective, an end, or is it an outcome that
emerges when alternative means of inflicting harm are considered “too expensive”
relative to the cost of mass killing? The question is relevant beyond mass killing
and pertains to all types of mass atrocities. In all cases, perpetrators make choices
as to the means used to inflict harm, and Ferrero simply asks whether the means
chosen might not reflect the cost of choosing and applying them.
For example, suppose that a dictator presides over an “in-group” and per-
ceives—however that perception comes about—that eliminating another group
of people (the “out-group”) is necessary to defend, maintain, or enhance the
in-group’s hold on power. Under what conditions would destruction of the out-
group be seen as necessary and, if feasible, what would be the most cost-effective
way to destroy the out-group? Posed this way, mass atrocities do not “just hap-
pen.” Instead, they result from and are carried out according to the in-group’s
assessment of the benefits and costs of such choices relative to alternative means
of achieving in-group objectives. Analyzing mass atrocity as arising from an in-
group’s assessment of benefits and costs can offer scholars and policymakers new
insights regarding such choices and how mass atrocities, including genocides and
mass killings (GMKs), might be mitigated or altogether prevented.
143
144 Theoretical Approaches Empirical Liter ature
In this chapter, then, our objective is to promote a wider and deeper applica-
tion and appreciation of constrained optimization theory in the study of GMK
than has occurred in the literature to date. In addition to applying constrained
optimization models to the choices of GMK architects and perpetrators, we con-
nect such models to key insights of genocide studies scholars, particularly Raphael
Lemkin (1944), regarding techniques of genocide. To enhance understanding of
the models, we provide intuitive narratives and graphical depictions of key con-
cepts and offer case examples of GMK risk and prevention. Our chapter addition-
ally highlights a critical GMK prevention conundrum suggested by constrained
optimization theory, namely, the ineffectiveness (and sometimes the complete
ineffectiveness) of piecemeal policies meant to protect civilians.
It is important to understand fully and precisely what constrained optimiza-
tion means. For example, during the course of a research project a research team
necessarily has to allocate its limited budget of time and other resources, such
as research grants and research assistants. Given the resource constraints, the
objective is to maximize an outcome variable that yields satisfaction (or “utility,”
as economists call it), such as the number of high-quality, high-impact publica-
tions produced under the researchers’ names. When combined in the most effi-
cient (productive) way, research inputs—time, grants, and assistants—yield the
highest possible, that is, maximum, utility. The process of efficiently combining
limited and costly inputs (the constraints) to maximize output given the con-
straints is called constrained optimization. Sometimes, the objective is to minimize
rather than to maximize, for instance, to minimize the number of accidents in a
research laboratory. Again, limited and costly resources to prevent accidents can
be combined efficiently or inefficiently. “Optimization” refers to the most efficient
combination, given the objective of accident minimization. Constrained optimiza-
tion models specify the objective (the desired output) and the constraints (limited
and costly inputs) for a particular research problem at hand and then examine
various aspects of the optimization process. In the GMK context, general con-
strained optimization theory is well suited to model, hypothesize, and test claims
about GMK-related choices made by an in-group regarding the onset of an atroc-
ity and the predicted use of various atrocity techniques (e.g., killing, deportation,
enslavement).
We begin with a brief overview of the promise and limits of constrained opti-
mization theory in the study of mass atrocity (section 6.2). The substance of the
chapter is then laid out in the next three sections. First, we present a baseline
model analyzing an in-group’s incentives and constraints to either fight rebels
or to intentionally kill civilians (jointly, the out-group) or to choose a mixture
of both when facing a perceived threat (section 6.3). We refer to this as the pre-
GMK model, for it allows us to highlight risk factors as well as potential prevention
policies before or just as a GMK event may occur. But assuming that an in-group
decision has in fact been made to conduct an actual GMK event, we then turn,
Genocide and Mass Killing R isk and Prevention 145
second, to optimal GMK techniques, including those identified by Raphael

Lemkin in his classic book, Axis Rule in Occupied Europe (section 6.4). To the
(neoclassical) economic theorist, the techniques identified by Lemkin and other
genocide scholars are inputs into a victim group’s destruction. In effect, we trans-
late Lemkin’s narrative into the language of a constrained optimization model
and show how an in-group may achieve its objectives by optimally substituting
between and among various means of out-group elimination, such as immediate
killing, deportation, starvation, work-to-death camps, and the locations and time
periods of attacks. Importantly, however, while perpetrators’ resources may be
ample, none are unlimited. They are scarce, and in this constraint there lies one
hope for GMK prevention. Third, we add perspectives from behavioral econom-
ics to constrained optimization models and the study of GMK, in particular the
importance of perpetrators’ reference points and degree of loss aversion (section
6.5).1 To highlight the applicability of the economic concepts, case examples of
GMK are used throughout the chapter. In the conclusion (section 6.6), we pres-
ent future research directions for constrained optimization modeling in GMK
studies and summarize what such models may reveal to scholars and policymak-
ers about GMK risks and prevention.
6.2. Promise and Limits of Constrained Optimization

Theory in the Study of Mass Atrocities
Choice models in economic theory generally fall into two categories: (1) con-
strained optimization and (2) game theory. Constrained optimization models
focus on a decision maker’s optimal choices in isolation from the strategic behav-
ior of others who may affect and be affected by such choices. Game theory goes
further by also considering how the actions of decision makers influence one
another. In this chapter, we focus on constrained optimization models. Other
chapters in this book (i.e., chapters 7, 14, and 19–22) address game-theoretic
perspectives.
A natural question follows: Why focus on one decision maker’s choices in
isolation from the interdependent and strategic choices of other decision mak-
ers relevant to the situation? First, models for studying such “isolated choices”
are pervasive in economics and rich in conceptual content, having been built
up over the course of more than a century. It is certainly time for much wider
application of such tools to mass atrocity. Second, various choice models in eco-
nomics (and other social sciences) have comparative strengths and weaknesses.
Game theory models are relatively strong in studying strategically interdepen-
dent choices among decision makers (the collection of choices), whereas con-
strained optimization models are comparatively good at analyzing finer details
related to one decision maker’s choices (individual choice). Finally, many game
theory models are built upon the theoretical foundations of constrained opti-
mization; as such, each type of model can be used to better understand and
apply the other.
Constrained optimization models are also called rational choice models,
where the word “rational” is a technical term and does not mean “reasonable” as
in the commonsensical meaning of the word. Obviously, mass atrocities are not
reasonable, and they are an affront to common sense. Instead, the word means no
more than that given one’s objective, however arrived at and however altruistic,
selfish, or malevolent it may be, one chooses actions to best achieve that objec-
tive in light of the constraints one faces, such as limits of time and budget (see
Anderton 2014a; also see section 6.5 of this chapter). Mass atrocity outcomes are
also affected by the costs and constraints that impinge upon the out-group as it
seeks to avoid the destruction unleashed by a genocidal regime and by resisters
as they seek to help the targeted group. Furthermore, the choices made by third
parties affect mass atrocity outcomes; specifically, their choices can block, deter,
ignore, accommodate, or support genocide (see Brauer and Anderton 2014; also
see chapter 13 in this volume). The potential for applying constrained optimiza-
tion models to the various decision makers in mass atrocity contexts is wide open
and ripe for much new research. In the sections ahead, we focus on the applica-
tion of constrained optimization models to the choices of potential or actual mass
atrocity perpetrators.
6.3. A Baseline Model: When is GMK an Optimal Choice?

To set up our pre-GMK model, assume, solely by way of example, that an in-
group (the “authority” group) encompasses key leaders of a state. Further assume
that this group has a well-defined objective such as defending, maintaining, or
expanding political or territorial power (hereafter “control”) and that this objec-
tive is perceived to be under severe threat from an armed out-group of a given
size, which we generically call “rebels.”2 In addition, there exist unarmed people
(“civilians”) who may be or could become supportive of the armed opposition or
may be seen as a threat in their own right. It is important to note the exact point
of the analysis upon which we are about to embark. We are not theorizing about
why an in-group has control as its objective, nor why an armed out-group is chal-
lenging it, nor whether this out-group might change the size of its forces given the
behavior of the in-group. For now, such elements all are taken as given and immutable.
It is in this context, however, that constrained optimization models are ideal for
investigating why the in-group might, perhaps surprisingly, choose to intention-
ally destroy civilians and, if such a choice is made, how (by what means) such
destruction might occur. In modeling possible atrocities, we position ourselves as
if we were assessing the incentives for and implementing the means of out-group
elimination. This is unpleasant to think about. But it will help us to discover risk
factors for GMK, understand how atrocity architects can work around policy
efforts to protect civilians, and provide pointers on how such atrocities may be
prevented.
6.3.1. Contesting an Armed Opposition and Killing Civilians:

A Baseline Constrained Optimization Model
Imagine that we are the leaders of an in-group regime that perceives a serious
threat to our control of the state. If we lose control we might be incarcerated and
executed. In short, we perceive an “existential threat” (Mayersen 2010). Assume
our objective, and that of our inner circle of colleagues, is to maximize control,
which we designate, for short, with the letter Q (think of Q as quantity of con-
trol). Collectively, we may have ample yet ultimately limited resources at our
disposal with which to achieve control. These we label with the letter I (think
of I as earned income, credit, grants, loot, loyalty, fanaticism, energy, knowl-
edge, time, and other monetary and nonmonetary resources). We could use the
resource pool to negotiate with and perhaps buy off the rebels such that violence
is avoided and a peaceful settlement is achieved, but virtually all post–World War
II GMKs occurred in a context of civil war or other serious state breakdown (see
chapter 3 in this volume; Harff 2003). Hence, we assume that any prospects for
peaceful settlement have fallen through and the context is indeed one of exis-
tential threat. An extensive literature in the social sciences seeks to explain why
peace can fail to hold even though fighting is costly. Such explanations include
asymmetric information, commitment problems (e.g., preemptive war technolo-
gies, shifts in power, indivisibilities), and political bias (Garfinkel and Skaperdas
2007; Anderton and Carter 2011). We do not model the emergence of an envi-
ronment of threatened or actual fighting; rather we take it as given and as the
starting point of our analysis.
In this threatening environment, our options for achieving control have been
narrowed to fighting or threatening to fight rebels, R, or attacking civilians, C. 3
Note that R and C are not ends but means of achieving control (Q). Following the
work of scholars on the economics of terrorism (e.g., Enders and Sandler 2011),
we assume that there is an average price to be paid from one’s resources per unit
of rebels contested—one unit is one rebel or a battalion of rebels—and this is
labeled as Pr. Similarly, an average price is expended per unit of civilians to be
killed (Pc). Taking the view of the in-group, this can be expressed in symbolic
shorthand as a constrained optimization model, namely:
maxR , C Q = f (R , C)
, (1)
subject to I = PrR + PcC
where f (R , C ) stands for the production function in which the elimination of

rebels or civilians, or both, are needed to produce maximum control, Q. This is
subject to the resource constraint of limited income. The resource bill for elimi-
nating rebels is PrR (the price per rebel-unit times the number of rebels elimi-
nated), and the bill for eliminating civilians is PcC. These bills can be paid until
the resource pool, I, is used up. If all resources are used to contest rebels, none
are left to kill civilians, and vice versa. Note also that by including just one output
and two inputs, we will be able to present several key results in two-dimensional
graphs while also retaining essential aspects of the choice problem facing the
in-group.
Before jumping to the model’s solution and its predictions about GMK,
we emphasize the importance of the production function in (1), Q = f (R , C ),
which reflects the in-group’s beliefs (or experiences) about how best to combine
the inputs of “contesting rebels” and “killing civilians” to achieve control, Q.
Achieving control through such inputs in specific contexts will depend on many
factors including the geographic terrain, weapons technologies, training and tac-
tical sophistication of the combatants, and so forth. Such factors will play a role
in whether the in-group treats fighting rebels and attacking civilians as inputs
that should go hand-in-hand (complements) or whether they can be traded off
against each other (substitutes). The production function also governs whether
increases in one of the inputs (with the other held constant) causes control to
increase at an increasing, constant, diminishing, or even negative rate (this con-
cerns the nature of marginal productivity and marginal returns).4 The produc-
tion function also captures whether a percentage increase in both inputs by the
in-group would cause control to rise by a greater, same, or lesser percentage (this
concerns increasing, constant, or decreasing returns to scale in the production
of control).
As mentioned, the production function assumes that the in-group has only
one output and two inputs. Instead, the in-group could have multiple outputs,
for example, when it is attempting to achieve one type of control in one part of
the country and another type of control elsewhere. It could also have (and mostly
certainly does have) more than two inputs, with other inputs related to control
including administration, diplomatic relations, and so on. Extensions of produc-
tion theory to more than one output and to many inputs would allow one to study
whether there may be productivity-enhancing opportunities for the in-group to
apply some of its inputs across both domains in its efforts to achieve control in
both areas (so-called economies of scope). Finally, the model in (1) is static, but
production theory has been extended into dynamic contexts, which allows one
to study productivity-enhancing learning curves. It is beyond the scope of this
chapter to apply many of these production theory concepts to GMK; we are sim-
ply opening the door to such inquiry. But we do wish to emphasize just how much
production theory can be developed and applied to better understand GMK.
6.3.2. Solution of the Baseline Model

Panels (a) and (b) of Figure 6.1 depict the symbolic solution to the model in equa-
tion (1). The two panels show the in-group’s optimal choice between the two
inputs, either contesting rebels (R), measured on the vertical axis, or killing civil-
ians (C), measured on the horizontal axis. Shown in each panel is the in-group’s
straight-line resource constraint, the height of which is determined by the amount
of the in-group’s resources (I) and the prices paid to contest rebels and kill civilians
(Pr and Pc). Also shown in each panel is a curve (called an isoquant), which shows
all combinations of R and C deemed to achieve the same level of control, Q. The in-
group’s optimal choice between R and C is determined by the highest isoquant, the
maximum level of control, which can be reached given the resource limits. In panel
(a), such an optimum occurs at a point where the in-group chooses to allocate all
resources to only contest rebels and none to kill civilians (called a corner solution).
The choice for R is R*>0 and for C it is C* = 0, where the asterisk (*) indicates “opti-
mal,” the maximum achievable in light of the available resources. Panel (b) depicts
a decidedly different outcome in which the in-group’s optimal choice is an inte-
rior solution such that it allocates some of its resources to contesting rebels (R*>0)
and some to killing civilians (C*>0). When C* is very much larger than zero, a
mass killing occurs. Put differently, when in the pursuit of the control objective it is
cheaper to contest rebels than it is to kill civilians, then the reasonable (“rational”)
choice is to pursue rebels alone. But if it becomes sufficiently cheap to kill civilians,
given the objective of maximizing control, the in-group will be predicted to pursue
them in addition to the out-group’s rebel forces.
6.3.3. Risk Factors for Killing Civilians

Why would an in-group choose to kill civilians rather than not? What deter-
mines this choice? The model depicted in Figure 6.1 provides an abstract answer,
namely, when the isoquant and the resource constraint line tilt sufficiently toward
the C axis, the in-group “pulls away” from the corner solution in panel (a) and
arrives at an interior solution like panel (b) in which some level of killing of civil-
ians is chosen.
The position and shape of the isoquant are based upon the production func-
tion, f (R , C ), in equation (1), which again reflects the in-group’s beliefs (or expe-
riences) about how best to combine “contesting rebels” and “killing civilians” to
achieve equivalent levels of control, Q. If the rebel group is relatively strong such
that contesting it is relatively unproductive, then the in-group might find it advan-
tageous to attack civilians who support (or could support) the rebels and/or to
terrorize civilians into supporting the in-group. In this case, the isoquant will be
positioned (tilted) away from R (toward C) and the in-group will choose to kill
civilians as shown in panel (b) of Figure 6.1, assuming everything else is the same.
Panel (a)
Optimal choice, R*>0 and

R*>0 ● C* = 0 (corner solution)
Isoquant
Contesting
Rebels, R
Income line
0 Killing Civilians, C
C* = 0
Panel (b)
Optimal choice, R*>0 and

C*>0 (interior solution)
R*>0 ●
Contesting
Rebels, R
Isoquant (steeper than in panel (a))
Income line (flatter than in

panel (a))
0 C*>0
Killing Civilians, C
Figure 6.1 Authority group’s optimal allocation of resources to R and C.
Separately or additionally, if civilians are a strong source of support for the rebels,
the in-group will find that killing civilians is relatively productive to achieving
control, which will tilt the isoquant toward C as well.
The position of the in-group’s resource line is based upon the resource equa-
tion in (1). If the unit price of killing civilians, Pc, is relatively low in terms of
overall resource expenditure, the resource line will rotate toward C, leading to an
increase in the amount of civilian killing. This inverse relationship between the
price of killing civilians and the number of civilians killed is an example of econo-
mists’ law of demand applied to GMK. Specifically, the cheaper it is to kill civil-
ians, the greater the amount of killing demanded, everything else the same. If the
price of contesting rebels, Pr, is relatively high, the resource line will rotate away
from R, which will lead to more civilian killing if the two inputs are sufficiently close
substitutes for achieving control by the authority group. 5 Finally, if the in-group
has more resources at its disposal, the resource line will shift outward (away from
the origin) in a parallel fashion. This will cause the in-group to kill more civilians,
assuming killing is what economists call a normal input.6
Figure 6.2 summarizes those elements within the model in equation (1) that
would increase an in-group’s propensity to kill civilians. Also included are our
connections of such economic principles to real phenomena that put civilians at
risk in civil wars and state breakdowns. Our analyses in Figures 6.1 and 6.2 rep-
resent a way of looking at risk factors for civilian killing through the lens of con-
strained optimization theory.
6.3.4. Tools of Genocide and Mass Killing Prevention

Just as the baseline model can be used to identify risk factors for GMK, so it
can be used to highlight tools of GMK prevention. Following scholars who
have applied such models to terrorism we highlight three classes of GMK
prevention policy: (1) resource policies, (2) productivity (or isoquant) poli-
cies, and (3) price policies.7 An example of a resource policy is shown in panel
(a) of Figure 6.3. Given a decline in the in-group’s resources, it will decrease its
demand for civilian killing from C* to C*’ assuming C is a normal input. Policies
that might reduce the in-group’s resources include asset freezes and embargoes.
The second category is productivity (or isoquant) policy. As noted, when
isoquants are relatively flat (more horizontal), the model tends toward a corner
solution in which the in-group only contests rebels and there is no civilian killing
(C* = 0). When the isoquants become steeper (more vertical), they begin to “roll
down” the resource line leading to an interior solution in which civilian killing
now occurs (C*>0). Thus, in panel (b) of Figure 6.3, as isoquants become steeper,
civilian killing increases from C* = 0 to C*’ to C*’’. Steeper isoquants occur when
there is an increase in the productivity of killing civilians relative to the pro-
ductivity of contesting rebels. Hence, policies that make attacking civilians less
productive (e.g., safe havens) and contesting rebels more productive will tilt the
isoquants toward the corner solution in which killing civilians is not chosen.
The third policy category is price policy, which can be broken into two catego-
ries: (1) increases in Pc and (2) changes in Pr. Examples of policies that increase Pc
are third-party on-the-ground (defensive) protection of vulnerable civilians; third-
party coercive (offensive) intervention against civilian-harming forces; threatened
maxR,C Q = f(R,C)
subject to I = PrR +PcC
Production function, Q = f(R,C) Resource constraint, I = PrR + PcC
(assuming a given resource constraint) (assuming a given production function)
Low productivity of High productivity of High Pr Low Pr Low Pc High Pc High I Low I
contesting rebels killing civilians
More C and More C and More C if R More C if R More C More C if C More C if More C if
less R less R and C are gross and C are based on the is Giffen C is normal C is
substitutes* gross law of demand input** inferior***
complements*
No increasing Civilians provide Relatively Relatively Geographically Foreign aid
marginal returns key support to easy to difficult to concentrated for in-group
or bandwagon rebels substitute substitute civilians Control of
effects in Killing civilians between C between C Geographically territory
contesting rebels improves and R to and R to unprotected rich in
State effectiveness of in- control control civilians resources
institutionally group forces (e.g., territory territory Little 3rd party
weak (anocracy, control over key interest in victim
new state) terrain) group
Little disruption to
Notes: trade and other
*Substitutes and complements are covered in sections 6.3.3 and 6.4. political/economic
**Giffen inputs are impossible under standard profit maximization assumptions in activities when
economic theory. Anderton (2014b) explains how such inputs might be possible attacking civilians
in a GMK context.
***See Anderton (2014b) for analysis of inferior input characteristics and GMK.
Figure 6.2 Risk factors for civilian killing in baseline constrained optimization model (conditional on the level of perceived existential threat).
(a) Panel (a): Resource reduction (normal input case)
Initial optimal choice
R* New optimal choice

R*' Initial isoquant
Contesting
Rebels, R
Highest attainable isoquant at new income level
Initial income line
New income line after decline in

income (I)
0 C*' C* Killing Civilians, C

Civilian killing declines
Panel (b): Productivity policy (“rolling back” the isoquants)
Optimal choice for relatively flat isoquant (corner solution)
Optimal choice for steeper isoquant

Contesting
Rebels, R
Optimal choice for even steeper isoquant
Income line
0 C*' C*''
C* = 0 Killing Civilians, C
Figure 6.3 GMK prevention policies to reduce civilian killing.
litigation against perpetrators through the International Criminal Court; and other
political, economic, and cultural actions against an atrocity-perpetrating regime.
Panel (c) of Figure 6.3 depicts the effect of a higher price of killing civilians on the
optimal choice of the in-group. The increase in Pc causes the resource line to rotate
inward along the C axis (the same resources expended at higher cost reduce the kill-
ing, C* falls, all else—especially killing productivity—being equal).
Panel (c): Higher price of killing civilians (increase in Pc)
R* New optimal choice

R*'
Contesting
Rebels, R
Initial isoquant
Highest attainable isoquant given higher Pc
Initial income line
New income line after increase in Pc
0 C*' C*
Killing Civilians, C
(law of demand)
Figure 6.3 (Continued)
The second price policy noted above would be alterations in the price of con-
testing rebels, Pr. Our focus in this chapter is on the problem of intentional attacks
against civilians. Nevertheless, the price of contesting rebels can have an impor-
tant cross-effect on the level of civilian killing chosen by the in-group as shown
in Figure 6.4. In panel (a), the price to the in-group of contesting rebels has risen,
which causes the resource line to rotate downward along the vertical axis. The
in-group’s optimal response is to choose fewer attacks against rebels and against
civilians. Both R* and C* decline in panel (a). This cross-price effect is negative,
that is, a higher price of contesting rebels leads to a lower demand for civilian
killings. Panel (a) depicts the case in which the in-group views contesting rebels
and killing civilians as complementary inputs in the production of control. When
inputs are complementary, policies designed to raise the price of contesting reb-
els also serve to protect civilians. Such policies might include third-party troops
operating as a buffer between rebel and in-group forces; political, economic, or
cultural penalties against the in-group should it renew hostilities with the rebels;
or financial or military aid to the rebels. Such policies would make it more diffi-
cult (pricey) for the in-group to contest a given number of rebels, leading to fewer
attacks against rebels and civilians.
The model, however, contains a big warning for such rebel-helping price poli-
cies. In Figure 6.4, the same price-increasing policies that helped civilians in panel
(a) backfire in panel (b). Specifically, when contesting rebels and killing civil-
ians are substitutes rather than complements, as perhaps in the Guatemalan mass
Panel (a) (R and C are complementary inputs)
Contesting

Rebels,R
R*
New optimal choice
R*' Initial isoquant
Highest attainable isoquant given higher Pr

Income line after increase in Pr
Initial income line
0 C*' C* Killing Civilians, C

Panel (b) (R and C are substitute inputs)
Initial income line
Income line after increase in Pr

Contesting
Rebels, R
R* Initial optimal choice
Initial isoquant
R*' New optimal choice
Highest attainable isoquant

given higher Pr
0 C* C*' Killing Civilians, C

Civilian killing increases
Figure 6.4 Input complements and substitutes and “cross price effects” on civilian killing.
killings in the early 1980s, an increase in the price of contesting rebels will cause
the in-group to reduce actions against rebels and increase attacks against civilians.
In panel (b), the resource line rotates downward along the R axis just as in panel
(a). Now, however, C* rises rather than declines. Hence, price policies designed to
help rebels may help vulnerable civilians under some circumstances, as in panel
(a), but harm them under other circumstances, as in panel (b).
Constrained optimization theory thus helps to reveal circumstances under

which rebel-helping price policies will backfire, leading to more civilian harm.
Specifically, when the two inputs—contesting rebels (R) and killing civilians
(C)—are relatively easy for the in-group to substitute when it seeks to achieve
control (Q), the in-group will be relatively sensitive to an increase in the price of
contesting rebels. When this is so, a higher Pr will lead to a cross-effect such that
the optimal amount of civilian killing, C*, will rise.8
6.3.5. The NATO Bombing of Yugoslavia and the Backfire Condition

Is this backfire outcome a depiction of what happened when NATO supported
the Kosovo Liberation Army (KLA) in its contest with the Yugoslavian govern-
ment during the 1998–1999 Kosovo war? Obviously the war was more complex
than depicted in Figure 6.4 (b) but, then, the very point of a model is to simplify
complexity so as to focus on essential elements the analyst finds important. Thus,
Kuperman (2008) maintains that third-party support of the KLA, motivated at
least in part by the desire to protect civilians, had the unintended consequence of
exacerbating violence, including against civilians. He focuses on the potential for
moral hazard in humanitarian interventions in which “the provision of protection
against risk … unintentionally promotes irresponsible or fraudulent risk-taking,
and thereby perversely increases the likelihood of the undesired outcome” (50).
Kuperman maintains that Western support for the KLA and NATO’s threat to
bomb Yugoslavia emboldened the KLA to push for independence and thus served
to escalate violence. Furthermore, following the initiation of NATO’s bombing
campaign, Kuperman argues that Yugoslavia “escalated from counter-insurgency
[against the KLA] to ethnic cleansing, demonstrating that under coercive pres-
sure to surrender sovereignty, a state may instead opt to perpetrate genocidal
violence in hopes of retaining sovereignty” (66). Hence, by making Yugoslavia’s
contestation of rebels more difficult (i.e., raising the price of R), and assuming
that Yugoslavia viewed contesting rebels and killing civilians as substitutable
inputs to achieve control, the constrained optimization model would predict the
backfire outcome of Figure 6.4 (b).
6.4. A GMK Model: Substitution Possibilities in GMK

The constrained optimization model developed above depicts an existentially
threatened in-group allocating resources to contesting rebels and/or killing civil-
ians to increase its control. Such models, however, are flexible regarding the objec-
tives of the in-group. Thus, in this section, we assume that the in-group already
has decided to eliminate an out-group. We (i.e., the in-group) no longer distin-
guish between rebels and civilians. All are viewed as members of an out-group
to be destroyed. We now consider the inputs that an in-group might choose to

optimally achieve such destruction, and its ability to substitute across such inputs
when facing third-party attempts to protect civilians. Before presenting the model
appropriate to this context, we turn to the founder of the field of genocide stud-
ies to learn more about the means of group elimination available to a determined
in-group.
6.4.1. A Synchronized Attack against an Out-Group’s Existence:

Raphael Lemkin’s Perspectives on Techniques of Genocide
In his path-breaking 1944 book on the Holocaust, Axis Rule in Occupied Europe,
Raphael Lemkin characterized genocide as a “synchronized attack” designed to
destroy an oppressed group and replace it with the national pattern of the oppres-
sor (Lemkin 1944, xi and 79). Lemkin (1944, xi–xii) identified eight dimensions
in which the plan of group elimination was carried out:
[I]‌n the political field (by destroying institutions of self-government and

imposing a German pattern of administration, and through colonization
by Germans); in the social field (by disrupting the social cohesion of the
nation involved and killing or removing elements such as the intelligen-
tsia . . .); in the cultural field (by prohibiting or destroying cultural insti-
tutions and cultural activities; by substituting vocational education for
education in the liberal arts . . .); in the economic field (by shifting wealth
to Germans and by prohibiting the exercise of trades and occupations by
people who do not promote Germanism “without reservations”); in the
biological field (by a policy of depopulation and by promoting procre-
ation by Germans in occupied countries); in the field of physical exis-
tence (by introducing a starvation rationing system for non-Germans
and by mass killings . . .); in the religious field (by interfering with the
activities of the Church . . .); in the field of morality (by attempts to create
an atmosphere of moral debasement . . .).
Each field can be viewed as an input into a genocidal regime’s production of group
elimination. Moreover, each field may have sub-inputs. For example, Lemkin
mentions starvation rations and direct mass killing as alternative means to
destroy the physical existence of an out-group. If the eight fields each have, say,
three sub-inputs, an in-group could draw upon a total of 24 inputs to produce
out-group destruction. The “24” here is just an example, but to the economist, the
presence of many input possibilities into a genocidal program suggests the poten-
tial for many input substitution possibilities. This is not good news for genocide
prevention policy, which we will now demonstrate using constrained optimiza-
tion theory.
6.4.2. High Genocidal Input Substitution Possibilities in a Constrained

Optimization Model
Assume an in-group’s objective is to maximize an out-group’s destruction, desig-
nated by Q (here, the quantity or number of people destroyed). It has allocated
a limited amount of resources with which to achieve such destruction. These are
designated by I (as in section 6.3). It also has many inputs with which to achieve
out-group destruction such as direct killing (K), cutting off the out-group’s means
of physical existence, such as starvation (S), work-to-death enslavement (E), and
coercive relocation (R). To keep the exposition simple, assume the in-group uses
only two inputs: direct killing (K) and starvation (S). There is an average price per
unit of victims directly killed, Pk (e.g., a soldier’s time, a firearm, and a bullet), and a
price per unit of civilians starved, Ps (e.g., the cost of restricting a victim’s access to
food). This leads to the following constrained optimization model for the in-group:
maxK , S Q = g (K , S)
, (2)
subject to I = PkK + PS s
where g (K , S ) is the production function for destroying people from the out-
group, Q, and I = P kK + PsS stands for the perpetrators’ resource constraint.
Assuming a high degree of input substitutability, Figure 6.5 shows the in-
group’s optimal choice of destructive inputs. At the initial optimum the in-group
chooses K* and S* in direct killing and starvation. Consider now a piecemeal pol-
icy intervention, say by the United Nations (UN), to bring food relief to the out-
group to undermine the in-group’s starvation program. Assume this intervention
raises the in-group’s price of implementing its starvation program, Ps. This causes
the resource line to rotate downward along the vertical axis, resulting in a smaller
absolute level of possible starvation. But if K and S are highly substitutable inputs,
the isoquants in Figure 6.5 will have relatively little curvature and the result of
the intervention policy designed to undermine the in-group’s starvation program
is that more direct killing of civilians will be chosen. Notice on the new resource line
in Figure 6.5 that S* falls to S*’, but K* rises to K*’ as the in-group redirects its
genocide program to the input made relatively cheaper by the well-intentioned
but misdirected (backfiring) intervention.
6.4.3. A Bleakness Theorem: Multidimensional Input

Substitution Possibilities and the Ineffectiveness of Piecemeal
Civilian Protection Policies
Figure 6.5 points to a policy dilemma, namely, efforts to restrict what a genocidal
regime can do along one or several dimensions will not stop a determined in-group
from working around such efforts. The challenge of substitution possibilities is
Initial income line
Income line after increase in Ps

Program, S
Starvation
S* Initial optimal choice
Initial isoquant
S*' New optimal choice
Highest attainable isoquant given

higher Ps
0 K* K*' Direct Killing, K

Direct killing increases
Figure 6.5 High substitutability among destruction inputs, S and K.
demonstrated in Figure 6.5 even when there are only two inputs. What if there are
dozens? In that case, the in-group’s ability to substitute across inputs to achieve its
genocidal objective would be rendered easier, and the challenge of civilian protec-
tion would be rendered more difficult.
Our concern is that we do in fact see dozens of input substitution possibilities
available to a genocidal in-group. In addition to direct killing, starvation, work-
to-death enslavement, and coercive relocation, there are many additional inputs
within Lemkin’s eight genocide fields including decapitation attacks against a tar-
geted group’s leaders; destruction of cultural assets; coerced reproductive, edu-
cational, and familial assimilation; wealth appropriation; destruction of trades
and occupations; and so on.9 Moreover, such inputs can be further differentiated
according to the times, locations, and intensities of such attacks. For example, killing
victims at time t is a different input than killing victims at time t+1 and killing vic-
tims at location L1 is different than killing victims at location L2. There really are
dozens of input possibilities for a genocidal regime to consider and, correspond-
ingly, a disturbingly high degree of substitution possibilities. One could model
the many substitution possibilities by expanding the input menu in equation
(2) from two to five, or any other number, but one would no longer be able to use
two-dimensional graphs to illustrate key concepts. There exists, however, a con-
strained optimization model that is ideally suited to analyzing such a many-input
world, namely, Lancaster’s (1966) household production model (also known as
the product or input attributes model).
For illustration, assume that the in-group has the four input possibilities for
group destruction noted earlier: direct killing (K), starvation (S), work-to-death
enslavement (E), and coercive relocation (R). Each generates attributes the in-
group deems desirable. Moreover, we assume the in-group has in fact preferences
for two attributes or objectives (rather than just one): destruction of the out-
group, designated as Av, and imposition of the national pattern of the in-group,
designated as Aa.10 Each input generates a certain amount of each attribute. For
example, one unit of direct killing (K) might generate one unit of out-group
destruction (Av) and one-half unit of imposition of the national pattern of the
in-group (Aa). There are, of course, many numerical possibilities for connecting
inputs to attributes. In general, the attributes will be generated from the inputs
according to the following linear equations:
A v = akvK + asvS + aevE + a rvR (3)
Aa = akaK + asaS + aeaE + a raR, (4)
where a kv stands for the number of victims destroyed per unit of killing K, a vs for
the number of victims destroyed per unit of starvation S, and so on. The in-group’s
budget constraint will now be:
I = PkK + PS
s + PeE + PrR, (5)
where I is the resource budget and Pi (i = k,s,e,r) is the unit price of input i. Finally,
assume the in-group achieves satisfaction over the attributes according to the fol-
lowing utility function:
U = U(A v , Aa ). (6)
The in-group’s optimization problem is to choose K, S, E, and R to maximize

(6) subject to the constraints in equations (3) to (5).
Suppose, by way of a numerical example, that the price of each input
is 1 (Pi = 1 for all i) and the in-group has I = 100 resource units to allocate to
its genocide program. Suppose the four inputs contribute to the in-group’s
desired attributes, Av and Aa, according to the following attribute coefficient
values: akv = 1, aka = 0.5, asv = 0.9, asa = 0.6, aev = 0.8, aea = 0.8, arv = 0.5, ara = 1.
For example, if the in-group allocated all of its income to direct killing, then K
would be equal to 100. This in turn would generate 100 units of the out-group
extermination attribute (Av) because a kv = 1 and 50 units of imposition of the
national pattern of the in-group attribute (Aa) because a ka = 0.5. Similar calcula-
tions can be made for the other three inputs.
Figure 6.6 shows four rays emanating from the origin, each showing the
amounts of the two desirable attributes the in-group can attain by allocating all
of its resources to the respective input. For example, the ray 0K shows the point
K*, which represents the amounts of the two attributes that can be obtained
when the in-group allocates all 100 resource units to direct killing. The other
three rays are the same for the starvation (S), work enslavement (E), and coerced
relocation (R) inputs. Of course, the in-group could obtain attributes through
combinations of inputs. Thus, the line joining K* and S* shows combinations of
the attributes available through combinations of direct killing and starvation.
Similarly, the line joining K* and E* shows attributes available through combi-
nations of direct killing and work enslavement, and so on. The preferences of
the in-group for attributes, based upon the utility function in (6), can be used
to generate combinations of the attributes that deliver a certain level of utility.
Such combinations are called an indifference curve. (The in-group is indiffer-
ent to the exact combination chosen so long as a certain level of satisfaction is
achieved.) The in-group would like to achieve the highest indifference curve it
can, subject to its budget constraint and attribute generation functions. Assume,
by way of example, that the best feasible indifference curve is the one labeled U*
in Figure 6.6, which is just tangent to the line segment K*E*. The figure shows
Aa(Imposition of national
pattern of authority group)
R*
100
E
E*
80
S
O*
60 S*
U*
K
50
K*
0 50 80 90 100
Av (Victim extermination)
Figure 6.6 The input attributes model of “optimal” genocidal programs.

that the in-group’s optimal allocation of resources to the inputs to maximize its
utility over attributes is a combination of direct killing and work enslavement as
shown by the optimal choice point O*.
The input attributes model suggests a number of critical results. First, given
the linearity of the input attribute equations (3) and (4), generally at most only
two inputs will be chosen. In Figure 6.6, direct killing and work enslavement are
chosen at point O*. Second, from the observed choices of the in-group, one may
be able to deduce certain properties of the underlying choice model. For example,
an in-group choosing combinations of S and K or of E and K rather than of R and
E or of R and S would have preferences geared more toward victim extermination
than in-group territorial control. Such knowledge could be useful to third par-
ties in designing policies to prevent victim extermination. Third, notice that some
inputs are not chosen in Figure 6.6, specifically, starvation and coerced relocation.
Hence, policy efforts designed to diminish an in-group’s efforts to deploy certain
inputs will be completely ineffective. For example, third-party intervention to
raise the price of the starvation input to the in-group, Ps, will cause point S* to
move toward the origin along ray 0S, drawing line segments E*S* and S*K* down
accordingly (not shown). But this will have no effect on the optimal choice O*
and, thus, on the amount of out-group direct killing. Fourth, suppose we return
the S* point to its initial location and assume that third-party support for the out-
group is able to raise the cost of direct killing to the authority group, Pk. This will
cause the K* point to move toward the origin along ray 0K. If K* declines enough,
then an input combination involving only E and S along line segment E*S*, or
possibly input S alone, will be chosen. This appears to be good news in regard to
avoiding out-group direct killing, but Figure 6.6 shows that such a policy could
have no effect on the overall number of out-group members killed. Specifically,
suppose line segment E*K* is no longer optimal because K* has fallen substan-
tially along ray 0K. Suppose the new best-achievable indifference curve is one that
just touches point S*. As drawn, the same number of out-group members is killed
as before (90), just that they are killed with starvation alone rather than with a
combination of direct killing and work enslavement.
We characterize the possible ineffectiveness of GMK intervention of the previ-
ous paragraph in a qualitative manner as a “bleakness theorem.” What we mean is
that when an in-group has numerous possible inputs (perhaps dozens) to achieve
what it deems to be desirable attributes from GMK, piecemeal protection poli-
cies along just one or a few dimensions will have relatively little overall effect,
and sometimes no effect, in protecting the out-group. Figure 6.6 shows only four
inputs, but even under such a limited input menu we can see the ineffectiveness
of piecemeal protection policy. Following the economics of terrorism literature,
only multithwarting protection policies would protect the out-group even after a
determined in-group’s best-possible responses to such efforts. In the context of
Figure 6.6, multithwarting protection would require a reduction in the resources
of the perpetrating regime or increases in the prices of all inputs. Such policies
would pull down K*, S*, E*, and R* along their respective rays and pull down the
attribute possibilities line segments as well. Only then can we be sure that out-
group killings will decline.
6.4.4. Ineffectiveness of Civilian Protection in the DR Congo

The second Congo war began in the Democratic Republic of the Congo (DRC)
in 1998 and allegedly ended when the Transitional Government of the DRC
took power in 2003. Since then, violence has continued between various out-
groups and the DRC government, especially in the eastern provinces of North
and South Kivu and Oriental. Moreover, severe violence against civilians has
persisted in the DRC including rapes and mass executions. The United Nations
Mission in the DRC (MONUC, which became the UN Stabilization Mission
in the DRC or MONUSCO in 2010), has made civilian protection the highest
priority among its several dozen peacekeeping tasks. But MONUC, as well as
the DRC government, have been heavily criticized for their failures to protect
civilians. It is beyond the scope of this section to provide any detailed analysis
of MONUC, MONUSCO, and the DRC government regarding the persistence
of violence and civilian atrocities in the DRC (but see Reynaert 2011). Instead
we offer several observations about a particular failure of civilian protection in
the DRC—the Goma crisis and the Kiwanja massacre of 2008—in light of the
bleakness theorem described above.
In January 2008, the Goma Conference led to a peace agreement between
various rebel groups and the DRC government to end the violence in North and
South Kivu. Nevertheless, in October and November 2008, violence escalated
between the main rebel group CNDP (National Congress for Defense of the
People) and DRC government forces in which the provincial capital of North
Kivu, Goma, was under threat of conquest by the rebels. During this crisis, most
of the MONUC forces were stationed around Goma owing to limited resources.
As Reynaert (2011, 17) notes: “MONUC was vastly overstretched … [and could]
only protect the people in the major towns and along key roads. Elsewhere,
MONUC could only protect itself.” In an input attributes model such as Figure
6.6, suppose the attributes sought by the CNDP are territorial control and bar-
gaining leverage vis-à-vis the DRC government. The activities that could achieve
such attributes would be to control various locations (towns), including those in
North Kivu. To the extent that MONUC protects Goma, it raises the price to the
CNDP of attacking that location, but it is relatively easy to substitute and attack
other locations because MONUC is “vastly overstretched” and could only protect
civilians in a piecemeal fashion. As such, the mission failed to protect civilians in a
smaller town in North Kivu, Kiwanja, in which at least sixty-seven civilians were
massacred (Reynaert 2011, 17). Human Rights Watch investigated the Kiwanja
massacre and made the following observations (among others), as summarized

by Reynaert (2011, 18):
MONUC was confronted with competing priorities, as the mission was

facing the dilemma to choose whether to prioritize the defence of a small
community residing in Kiwanja or to protect the larger area around
Goma. Moreover, with only a limited number of peacekeepers present
around Kiwanja, MONUC was largely tied up in securing a few humani-
tarian workers, a journalist and a group of MILOBs [military observers].
An economist would tend to surmise, as the bleakness theorem suggests, that

raising the price to a rebel group of attacking civilians in one or a few locations
(e.g., Goma) would lead that group to substitute into other locations in which to
attack civilians (e.g., Kiwanja).
6.5. Behavioral Economics: Perspectives

on GMK Risk and Prevention
Based on several decades of experimental research on human decision-making,
it is now generally accepted that peoples’ choices often are motivated by more
than narrow self-interest (Tversky and Kahneman 1991; Kahneman 2011).
Experimentalists find that laboratory subjects are motivated by their own returns,
but also by such concerns as altruism, fairness, reciprocity, envy, and revenge.
Moreover, human beings seem to be affected in significant ways by background
conditions (also known as frames or reference points) and various cognitive biases
such that choices often deviate from the predictions of standard constrained
optimization models as generated in sections 6.3 and 6.4. The field in which such
research is occurring is known as behavioral economics, wherein “insights from
laboratory experiments, psychology, and other social sciences [are applied] in eco-
nomics [in conjunction] with the standard economic model to get a better under-
standing of economic behavior” (Cartwright 2011, 4). In this chapter we do not
provide in-depth coverage of constrained optimization theory and behavioral
economics (for that, see c hapter 26 in this volume). Instead, we offer within con-
strained optimization theory a particular focus on the importance of reference
point dependence and loss aversion for understanding GMK risk and prevention.
6.5.1. Reference Point Dependence and Loss Aversion

Imagine a small group of in-group leaders of a state who have a preference for
consumption goods (e.g., luxury food, fine clothing, and exquisite homes)
and control. Assume that repression of the citizens of the state, the out-group,
is the least-cost method of achieving control. Moreover, the in-group must
allocate scarce resources (designated I) to acquire consumption goods (des-
ignated C) and control through repression (designated R). To fix the idea by
hypothetical illustration, the cadre of the current North Korean leadership
might be assumed to have such preferences and resource constraints. Panel
(a) of Figure 6.7 shows the optimal allocation of resources to C and R, labeled
O*, based on the in-group’s initial amount of the two goods at time t and its
indifference curve, Ut, defined relative to that initial point. In the figure, the
in-group chooses a relatively large amount of repression (R*) that, over time,
leads to repeated episodes of atrocities committed against the out-group. Note
also in panel (a) that the in-group would have been equally satisfied to apply
less repression and acquire more consumption goods at point B relative to O*,
but such an outcome falls outside the resource constraint and thus is not eco-
nomically feasible.
Panel (b) of Figure 6.7 turns to two critical ideas from behavioral economics—
reference point dependence and loss aversion—to show why the establishment
of high repression as a reference point makes it even more difficult (i.e., pricey)
to dislodge than standard constrained optimization theory would already pre-
dict. A key discovery of behavioral economics is loss aversion in which “losses
(outcomes below the reference state) loom larger than corresponding gains (out-
comes above the reference state)” for decision makers (Tversky and Kahneman
1991, 1047). Recall in panel (a) that initially the in-group is equally happy with
O* (with high repression) and B (with lower repression), relative to reference
point t. It chooses O* because it is the best the in-group can do given its resource
constraint. Over time, O* became the reference point.11 Panel (b) shows the impli-
cations of such reference point lock-in. Whereas the in-group used to be indif-
ferent between O* and B, eventually it will come to have a preference in favor of
O* over B. This is because now that point O* is the reference point, a change in
outcome from O* to B would imply a loss of control (measured by distance ab)
and a previously corresponding gain of consumption goods (measured by dis-
tance cd). But loss aversion implies that the loss will loom larger for the in-group
than the corresponding gain. Hence, the reference-dependent indifference curve
for the in-group going through point B in panel (b) will be steeper than the one
going through point B in panel (a), because O* became the reference point when
in-group preferences are governed by loss aversion. The now steeper indifference
curve in panel (b), labeled UO*, reflects the in-group’s higher marginal valuation
of repression because of the gradual establishment of a high repression reference
point, O*. Tragically, previous repression can serve to lock in and even expand
repression over time owing to human tendencies toward reference dependence
and loss aversion.12
Panel (a): Optimal choice relative to reference point t
B
Economic Goods, C
O*
C* Indifference curve based on
reference point t
t
Ut
0 R* Political Control Through

Repression, R
Panel (b): Increase in marginal valuation of repression owing to reference point O*
d B
Economic Goods, C
O*
c
Indifference curve based on
reference point O*
UO*
0 a b Political Control Through

Repression, R
Figure 6.7 Reference point, loss aversion, and the “lock in” of repression.
6.5.2. Policy Implications for Civilian Atrocity Prevention

Figure 6.7 is abstract and may not be helpful to scholars from all disciplines. But
the policy implications of reference point dependence and loss aversion for mass
atrocity prevention are important for scholars and policymakers across all fields.
Although an overstatement, a key message of behavioral economics for genocide
studies is “once repressive, then ever more repressive.” This might be character-
ized as a “repression inertia theorem.” Once repression gets established in an
in-group’s decisions, it becomes difficult to dislodge in future decisions and may

even worsen over time. Policies that attempt to take away an in-group’s ability
to repress its own citizens will be framed as a loss, assuming everything else is
unchanged, and loss aversion will cause the in-group to magnify the cost of that
loss beyond what the standard constrained optimization model would predict.
Given reference dependence and loss aversion, policies designed to move an in-
group away from repression, whether coercive or peaceful, are likely to face more
resistance than even standard theory would predict.
But there is good news as well: To the extent that nonrepression can become
the reference point, it too can be difficult to dislodge owing to reference point
dependence and loss aversion. Hence, investments in crafting institutions and
norms that keep nonrepressive states nonrepressive can pay potentially huge divi-
dends by keeping in-groups tethered to accommodation rather than repression.
Moreover, moving a repressive state to nonrepression is not a one-off policy but
one that can generate a stream of peace dividends year-by-year as accommodation
becomes a new and self-reinforcing norm. Research in the economics of culture,
multi-equilibrium modeling (with low- and high-repression equilibria), and evo-
lutionary game theory may be quite relevant here (see, e.g., chapters 7 and 22 in
this volume).
6.6. Conclusions
This chapter presented a number of standard constrained optimization models
from economic theory and adapted them to the analysis of GMK choices. The
models were designed to focus on two critical aspects of GMKs. The first was the
conditions under which leaders of an in-group would choose GMK. The second
was the frighteningly wide menu of atrocity mechanisms or inputs over which a
regime could choose once it had decided to undertake GMK. The models pointed
to a number of key ideas and results and suggested many possible future research
possibilities, some of which we now summarize.
First, constrained optimization models (COMs) can be useful in the study of
GMK choices and prevention by focusing attention on the motives (reflected in
the preferences) and constraints of leaders contemplating mass atrocity. These
COMs are ideally suited to the study of conditions that give rise to a “demand
for GMK.” Moreover, by focusing analysis on the incentives for GMK and, ulti-
mately, the opportunity cost or “price” of GMK choices, COMs can be used to
better understand prevention policies that might reduce or even eliminate lead-
ers’ demands for GMK.
Second, COMs are models. As models should be, COMs are simple (but not
simplistic) and flexible to accommodate a variety of different real-world events
of GMK. The purpose of models is to focus attention on certain features so as to
better understand, explain, and predict factors that drive real-world behavior. For
example, COMs reveal a “bleakness theorem” in which anti-GMK policies along
just one or a few dimensions can have relatively little impact, and under some
conditions no impact at all, in restraining GMK outcomes. As a second example,
addition of behavioral economics considerations to GMK, COMs shows that ref-
erence dependence and loss aversion help to explain why, once started, a regime’s
past repression can become locked in as a regime norm, our “repression inertia
theorem.”
Third, the notion of constrained optimal choice can fruitfully be applied to
the study of GMKs. While controversial and, we believe, often misunderstood in
the social and behavioral sciences, its perspectives and models can integrate and
be integrated into theoretical perspectives from multiple disciplines. Hence, the
choice for scholars is not constrained optimal choice theory versus but along with
(interconnected with) other theoretical perspectives (Anderton 2014a, 126–32).
Moreover, it is because GMK choices are driven in part by optimal choice consid-
erations that policies can be designed to alter regimes’ incentives such that they
choose to abstain from GMK.
Finally, we note several of many possible future research avenues on COMs
and GMKs. The standard constrained optimization model of economic theory
has been formalized and refined for well over a century. Lying within this model
is an extraordinarily rich array of concepts that we have only begun to tap in this
chapter. Particularly important for future research is a fuller exploration of the
precise nature of substitution possibilities available to decision makers as rep-
resented by the Slutsky equation of economic choice theory. In this chapter we
applied COMs to the choices of GMK leaders. Also relevant is the potential to
apply such models to the constrained choices of other agent types such as perpe-
trators (as distinct from GMK architects), victims, and third parties (see Brauer
and Anderton 2014 and chapter 13 in this volume for nonformal treatments of
such ideas). We note also that COMs can be productive for generating empiri-
cally testable propositions in GMK research (similar to what COMs have done
in empirical terrorism research). It is also important that future research consider
two weaknesses of standard COMs. The first is that behavioral economists have
discovered many instances in which human behavior does not strictly follow the
predictions of narrowly conceived COMs (e.g., reference dependence and loss
aversion to name just two). As such, social scientists from a variety of disciplines
(e.g., economics, psychology, political science) are developing more general and
refined choice models, and these will almost certainly have the potential to shed
new insights into GMK choices and behaviors. A second weakness of standard
COMs is that preferences are typically taken as given. The application of COMs
to GMKs will require that one take account of preference formation, for example
by analyzing how preferences are rooted in history and culture, how preferences
can be manipulated through propagandistic advertising by a regime intent on
GMK, and how the long-term creation of benevolent norms results in desirable
preferences such as multigroup tolerance or the development of superidentities
(beyond “us” versus “them”). Finally, it is likely that significant teaching and
learning potential exists inasmuch as COMs convey understanding and increase
awareness about GMKs and their prevention to any interested party, including
formal academic study at the undergraduate and graduate levels.
Notes
1. Chapter 26 in this volume addresses behavioral economics perspectives on GMKs in more
detail. See also Midlarsky (2005), who applies loss aversion to GMKs.
2. “Rebels” is a generic term for armed groups that are or could oppose the in-group’s control.
Hence, rebels could be insurgent groups in the context of intrastate violence or instability,
resistance groups in the context of interstate war and territorial conquest, or anticolonial
resistance groups in the context of extrastate tensions.
3. Many genocide scholars have studied conditions in which an authority group believes that
there are tactical and strategic benefits from attacking civilians (e.g., Kalyvas 1999 and
2006; Valentino 2004; Humphreys and Weinstein 2006; Shaw 2007; and chapters 7, 19–21,
and 23 in this volume).
4. Although an unusual form of economic behavior, negative (and not just diminishing) mar-
ginal returns to greater input use can be incorporated in a production function. Such a
possibility would arise in a GMK context when greater attacks against civilians reduce the
control of the attackers, everything else the same. As one possible example, Lyall, Blair,
and Imai (2013) report survey evidence across 204 villages in Afghanistan that harm
inflicted by the International Security Assistance Force (ISAF) led to reduced support
from the villages for the ISAF and increased support for the Taliban.
5. If, instead, the two inputs are complements rather than substitutes in production, an
increase in the price of contesting rebels will lead to a decline in civilian killing. See later
sections for in-depth analyses of substitutes and complements in a GMK context.
6. A normal input is one whose use increases when the scale of production expands, which it
would in Figure 6.1 when the in-group’s resources expand. In contrast, an inferior input is
used less when resources and the scale of production expand (see Anderton 2014b).
7. A fourth class, preference policies, is addressed in section 6.4. For coverage of various
classes of counterviolence policies see also Frey and Luechinger (2003), Frey (2004),
Anderton and Carter (2005), and Enders and Sandler (2011).
8. Following Anderton and Carter (2005, 276–78), the backfire condition arises if and only
if the price elasticity of demand for contesting rebels (R) is price elastic. The proof is based
on the resource constraint, I = PrR+PcC, and the total expenditure test of elasticity. The
total expenditure test says that if a good is price elastic, an increase in the price of that good
will cause spending on the good to decline. The proof of the backfire conditions is as fol-
lows: When Pr goes up, R* will decline via the law of demand. If the demand for R is price
elastic, the regime will spend fewer resources on R. Since all resources are spent and total
resources (I) are fixed, the regime will spend more resources on C. Since Pc is unchanged,
attacks against civilians will rise (C* increases). This establishes that an elastic demand
for R will lead to more C when Pr goes up. A reversal of the argument establishes that if an
increase in Pr leads to more C, the demand for R must be price elastic.
9. See Brauer and Anderton (2014) and chapter 13 in this volume.
10. For Lemkin (1944, 79), the dual purpose of genocide is to destroy the victim group and to
replace it with the national pattern of the oppressor.
11. For scholars in other disciplines, this may sound like the concept of path-dependence.
12. Of course, the new reference point, O*, itself becomes “old” as time progresses from t+1 to
t+2. We do not analyze this problem further here.
References
Anderton, C. H. 2014a. “A Research Agenda for the Economic Study of Genocide: Signposts
from the Field of Conflict Economics.” Journal of Genocide Research 16, no. 1: 113–38.
Anderton, C. H. 2014b. “Killing Civilians as an Inferior Input in a Rational Choice Model
of Genocide and Mass Killing.” Peace Economics, Peace Science and Public Policy 20, no.
2: 327–46.
Anderton, C. H., and J. R. Carter. 2005. “On Rational Choice Theory and the Study of Terrorism.”
Defence and Peace Economics 16, no. 4: 275–82.
Anderton, C. H., and J. R. Carter. 2011. “A Bargaining Theory Perspective on War.” In D.
Braddon and K. Hartley, eds., Handbook on the Economics of Conflict. Cheltenham, UK:
Edward Elgar, 29–51.
Brauer, J., and C. H. Anderton. 2014. “Economics and Genocide: Choices and Consequences.”
Seton Hall Journal of Diplomacy and International Relations 15, no. 2: 65–78.
Cartwright, E. 2011. Behavioral Economics. New York: Routledge.
Enders, W., and T. Sandler. 2011. The Political Economy of Terrorism. 2nd ed. New York: Cambridge
University Press.
Ferrero, M. 2013. “You Shall Not Overkill: Substitution between Means of Group Removal.”
Peace Economics, Peace Science and Public Policy 19, no. 3: 333–42.
Frey, B. S. 2004. Dealing with Terrorism—Stick or Carrot. Cheltenham, UK: Edward Elgar.
Frey, B. S., and S. Luechinger. 2003. “How to Fight Terrorism: Alternatives to Deterrence.”
Garfinkel, M. R., and S. Skaperdas. 2007. “Economics of Conflict: An Overview.” In T.
Sandler and K. Hartley, eds., Handbook of Defense Economics. Vol. 2. New York: Elsevier,
649–709.
Humphreys, M., and J. M. Weinstein. 2006. “Handling and Manhandling Civilians in Civil
War.” American Political Science Review 100, no. 3: 429–47.
Kahneman, D. 2011. Thinking, Fast and Slow. New York: Macmillan.
Kalyvas, S. N. 1999. “Wanton and Senseless? The Logic of Massacres in Algeria.” Rationality and
Society 11, no. 3: 243–85.
Kalyvas, S. N. 2006. The Logic of Violence in Civil War. New York: Cambridge University Press.
Kuperman, A. J. 2008. “The Moral Hazard of Humanitarian Intervention: Lessons from the
Balkans.” International Studies Quarterly 52, no. 1: 49–80.
Lancaster, K. J. 1966. “A New Approach to Consumer Theory.” Journal of Political Economy 74,
no. 2: 132–57.
Lyall, J., G. Blair, and K. Imai. 2013. “Explaining Support for Combatants during
Wartime: A Survey Experiment in Afghanistan.” American Political Science Review 107, no.
4: 1–27.
Mayersen, D. 2010. “On the Timing of Genocide.” Genocide Studies and Prevention 5, no. 1: 20–38.
Midlarsky, M. I. 2005. The Killing Trap: Genocide in the Twentieth Century. New York: Cambridge
University Press.
Reynaert, J. 2011. MONUC/MONUSCO and Civilian Protection in the Kivus. Antwerp,

Belgium: International Peace Information Service.
Tversky, A., and D. Kahneman. 1991. “Loss Aversion in Riskless Choice: A Reference-Dependent
Model.” Quarterly Journal of Economics 106, no. 4: 1039–61.
7
Incentives and Constraints

for Mass Killings
A Game-Theoretic Approach
Joa n E st e ba n, M a s si mo Mor e l l i, a n d Dom i n ic Roh n e r
7.1. Introduction
The question of what motivates mass atrocities has puzzled scholars for the longest
time, even going back to Thucydides’ masterpiece The War of the Peloponnesians
and the Athenians. The goal of this chapter is to present a rational analysis of the
incentives that propel groups who control power to engage in mass killings. In
contrast to terror organizations, where the decision makers are typically leaders
of rebel groups, the rational incentives to perpetrate genocide have to do with
extending or keeping control, rather than challenging control. Mass killings, espe-
cially genocide, are typically perpetrated by governments and targeted at groups
that a group in power wishes to weaken or dispossess.1
Understanding the potential for rational, or purposeful, recourse to mass atrocities
by governments is important not only from the point of view of positive analysis but
also for potential improvements in risk assessment and the design of early warning
systems, which are key necessary conditions for prevention and/or effective interven-
tion.2 Thus we will also devote some attention to the role of intervention and sanctions.
The chapter proceeds as follows. Section 7.2 outlines six “stylized facts” about
genocide; and section 7.3 discusses third-party intervention, or rather the diffi-
culty or lack thereof. Section 7.4 presents a model, and section 7.5 concludes.
7.2. Stylized Facts

Fact 1: Genocides Are Planned
The first stylized fact is that genocides are planned. As the US Genocide Prevention
Task Force reports: “Genocide is not the inevitable result of ‘ancient hatreds’ or
172
I n c e n t i v e s a n d C o n s t ra i n t s f o r M a s s K i l l i n g s 173
irrational leaders. It requires planning and is carried out systematically” (Albright

and Cohen 2008, xv). Further, “the task force finds that mass atrocities are gener-
ally perpetrated when underlying risk factors such as ethnic or sectarian discrimi-
nation, nationalist myths, armed insurgency or political and economic exclusion
are exploited by opportunistic elites seeking to amass power and to eliminate
competitors” (Albright and Cohen 2008, 36). In other words, nationalism, eth-
nic sectarianism, insurgencies, and political exclusions are (l) endogenous and
(2) cannot by themselves explain recourse to mass killings. Understanding mass
killings requires a direct focus on the interests of the groups in power. Formal
analysis is necessary to explain under what economic or institutional conditions
an insurgency takes place such that the group in power decides to use mass kill-
ings as part of its strategy.
Even though hatred, distrust, and uncontrolled passion can certainly play a big
role, 3 “to understand ethnic cleansing we need a sociology of power more than
a special psychology of perpetrators as disturbed or psychotic people—though
some may be… . All cases of cleansing involve material interests. Usually, mem-
bers of an ethnic group come to believe they have a collective economic inter-
est against an out-group” (Mann 2005, 9, 31). Also Chirot and McCauley (2006,
5) argue “that most political massacres are quite deliberate, are directed by or at
least approved by the authorities, and that they have a goal.” These authors “take
the position that mass killing is neither irrational nor in any sense ‘crazy’ ” (Chirot
and McCauley 2006, 7). As for explanations of war, explanations of episodes of
mass killings require reference to history, ideological clashes, religious cleavages,
and the like, but the presence of such cleavage-related motivations alone cannot
explain why in their presence there are cases in which government mass killings
take place and other cases in which they do not. A rationalist explanation of gov-
ernmental mass-killing decisions can be crucial for this type of positive analysis
even when material incentives are not the sole motivations. (By “rationalist” we
mean that groups select actions that maximize their payoffs, given their prefer-
ences. This of course does not mean that their preferences should be endorsed.)
Fact 2: Power Kills

To carry out genocide, a group needs to handle power and control the military.
Hence genocides and most forms of large-scale mass killings are perpetrated by
governments (Harff 2003; Valentino, Huth, and Balch-Lindsay 2004). According
to Rummel (1995), “political regimes—governments—have probably murdered
nearly 170,000,000 of their own citizens and foreigners in this century—about
four times the number killed in all international and domestic wars and revolu-
tions” (Rummel 1995, 3). Only rebel groups that are militarily very strong are
able to commit some forms of mass killings (Hultman 2009), and this typically
after having won a military battle (Schneider, Bussmann, and Ruhe 2012). Krain
(2000, 43) concludes that “military victories by definition enable the winner to
set the terms of the post–internal war period. This may include the decision to
punish the losing side by eradicating them, thereby eliminating the problem of
having to live side by side with the enemy in the post–internal war state. This
was the solution chosen by the Congolese rebels who took control of what would
become Zaire in the mid-1960s.” Or as put by Chirot and McCauley (2006, 2),
“conflict can become genocidal when powerful groups think that the most effi-
cient means to get what they want is to eliminate those in the way.”
Fact 3: Genocides Happen at the End or after Civil War

Genocide events, as evident by looking at the Political Instability Task Force (PITF)
dataset, take place predominantly at the end or after civil wars (see also Krain 2000;
Valentino, Huth, and Balch-Lindsay 2004). In the words of Krain (2000, 46), “inter-
nal wars are lethal twice over—in the actual bloody conflict, and in the enhanced
potential for state-sponsored mass murder subsequently.” Once one party has been
defeated on the battlefield and is powerless, the likelihood of massacres by the vic-
torious group is highest. For this reason, the model we adopt in section 7.4 assumes
that the proper and most consistent timing involves first a decision by the powerless
group to rebel or not; only afterward is there a decision stage, by whoever conquers
(or keeps) power, whether to enact mass killings or not.4
Fact 4: Civil Wars Are the Least Civil

An interesting new stylized fact is presented by Esteban, Morelli, and Rohner
(2015, 5): “Not all forms of war are equally likely to be accompanied by mass
killings. A substantial fraction of civil wars entail deliberate mass killings of civil
non-combatants on a large scale perpetrated by the dominant group, while there
is almost no record of mass killings of this sort in post-WWII interstate wars.
Between 1960 and 2000 roughly a third of all civil wars (50 out of 152) featured
mass killings, while in none of the interstate wars (23) were there mass killings.”
Fact 5: The Dark Side of Democratization

The existing literature finds contradictory results about the effect of democracy
and democratization on mass killing incentives, depending on whether they focus
on the level of democracy or the process of democratization. Most existing papers
focus on the level of democracy and find that nondemocratic regimes are more
likely to commit mass killings than democracies, especially when the autocrats
are powerful (Rummel 1994, 1995; Harff 2003; Valentino, Huth, and Balch-
Lindsay 2004; Easterly, Gatti, and Kurlat 2006; Colaresi and Carey 2008). In
contrast, there is evidence that the process of democratization heightens the risk
of mass killings. Numerous case studies make this point, as shown in the books
of Snyder (2000), Mann (2005), and Mansfield and Snyder (2005). According
to Mann (2005, 4), the process of democratization is one of the main causes of
ethnic cleansing: “Stably institutionalized democracies are less likely than either
democratizing or authoritarian regimes to commit murderous cleansing…. But
their past was not so virtuous. Most of them committed sufficient ethnic cleans-
ing to produce an essentially mono-ethnic citizen body in the present. In their
past, cleansing and democratization proceeded hand in hand.” Or, in the words
of Mansfield and Snyder (2005, 5) on the cases of Burundi and Rwanda, “[t]‌he
1993 elections in Burundi—even though internationally mandated, free, and
fair—intensified ethnic polarization between the Hutu and Tutsi ethnic groups,
resulting in some 200,000 deaths.” Moreover, “[p]ower sharing and pluralism
[were] precursors to the Rwandan genocide. In Rwanda, as in Burundi, the pres-
sures to democratize applied by the international donors that were the source of
60 percent of the Rwandan government’s revenue played a central role in trigger-
ing ethnic slaughter” (Mansfield and Snyder 2005, 255). An article by Esteban,
Morelli, and Rohner (2015) has presented the first systematic large-scale statisti-
cal evidence that recent democratizations indeed have led to a sharp rise in the
risk of genocides. 5
Fact 6: The Curse of Polarized, Poor, and

Resource-Abundant Countries
Among other papers in the empirical literature studying mass killings, Krain
(1997), Heger and Salehyan (2007), Bae and Ott (2008), and Querido (2009)
find that large levels of ethnic fractionalization reduce the risk of mass killings,
while Montalvo and Reynal-Querol (2008) and Esteban, Morelli, and Rohner
(2015) show that ethnic polarization increases the risk of mass killings. Some
studies also find that richer countries tend to display fewer mass killings (Scully
1997; Bae and Ott 2008; Esteban, Morelli, and Rohner 2015). 6 Inequality
(especially human capital inequality) tends to increase the risk of mass killings
(Besançon 2005). Trade openness reduces the risk of mass killings (Harff 2003;
Esteban, Morelli, and Rohner 2015). Countries abundant in natural resources
are more likely to experience mass killings (Querido 2009; Esteban, Morelli, and
Rohner 2015).
7.3. Third-Party Intervention

In 2011, the United Nations (UN) recognized for the first time the importance
of limiting a leader’s power to kill: Resolution 1973 (March 17, 2011) on Libya
marked the first time the UN Security Council (UNSC) authorized the use of
force for human protection against the wishes of a functioning state. According to
Bellamy and Williams (2011, 825): “The closest it had come to crossing this line
previously was in Resolutions 794 (1992) and 929 (1994). In Resolution 794, the
Council authorized the Unified Task Force to enter Somalia to ease the humani-
tarian crisis there, but this was in the absence of a central government rather than
against one.” That recent resolutions like this show recognition that perpetrators
of mass killings are often (if not always) state-controlling groups or states them-
selves is a step in the right direction, although some powerful states still oppose
action by the UN against governments. However, even if this opposition were
lifted, the theory that we have so far, which we develop in section 7.4, suggests
that the optimal level of intervention depends on the economic and institutional
characteristics of the country where civilians are at risk. That different economic
and institutional characteristics induce different normative considerations on
what intervention is optimal, even when the objectives of the third parties are
unchanged, calls for a more integrated study of intervention categories.
There is a very limited empirical literature on the desirability of intervention
in order to constrain the exercise of power: Since the end of the cold war, the
question of whether to intervene to stop states from committing atrocities has
become central (Hoffman et al. 1996; Teson 1997; Wheeler 2002; Holzgrefe and
Keohane 2003; Weiss 2007; Rotberg 2010). There is some evidence of ambiguous
effects of such tightening of the power to kill: Hultman (2010), for example, finds
that UN interventions mandated to protect civilians do reduce civilian deaths,
but other UN interventions increase rebel targeting of civilians. Also, the theo-
retical literature is relatively slim. Two recent relevant papers that we discuss in
detail in section 7.4 are Kydd and Straus (2013) and Esteban, Morelli, and Rohner
(2015) (hereafter referred to as KS and EMR, respectively).
Both KS and EMR deal with the effects of external intervention in civil war.
A consideration that emerges from both perspectives is that international inter-
vention is bound to increase the expected payoff of the rebel group relative to the
government’s, and hence groups in opposition will be more inclined to trigger
conflict because their effective cost will be lower than without intervention (see
also Kuperman 2008). The costs of rebellion can be affected through different
channels, such as economic sanctions on the government repressing the rebel-
lion, the imposition of a “fairer” division of the surplus, or moderating or prohib-
iting the killing of noncombatant civilians. International imposition of limits on
the exploitation of the minority group not only has an effect on the likelihood of
conflict but also, importantly, may increase the incentives of committing mass
killings for or by the group in government. Put differently, once exploitation
becomes harder, the ruler may have incentives to substitute exploitation with
elimination. In either model, though, international military intervention scales
down, or stops, mass atrocities. The net result is to increase the likelihood of con-
flict onsets but with ambiguous effects on the level of mass killings taking place.
Committing to directly reduce atrocities has ambiguous effects and may end up
increasing extreme genocidal incentives.
In KS, the government announces a sharing of the surplus. If accepted, peace
follows and the game ends. If rejected, there is violent conflict and the govern-
ment decides on the extent of mass killings. The murders are carried out right
away in order to increase the government’s probability of victory on the battle-
field, the sole benefit of such atrocity. The international community experiences
a cost by viewing or experiencing such atrocities from the outside and decides
between either imposing economic sanctions on the government or becoming a
third military actor in support of minimum fairness toward the opponents they
deem best. External intervention reduces the probability of victory for either of
the other two parties and results in an effective scaling-down of the level of atroci-
ties committed. Economic sanctions are an exogenous parameter and have the
effect of increasing the cost of conflict to the government. The game ends with
the winner implementing the most desired policy: keeping the entire surplus in
the case of the government or the opposition or, in case of victory of the exter-
nal actor, distributing the surplus to the two parties in the proportion considered
adequate.
In EMR, genocides can be committed by whoever ends up being victorious
(consistent with the aforementioned stylized fact 3) and have the goal of weakening
the opponent in future iterations of the game. The probability of winning depends
on the relative size of each group and hence mass killings have the effect of reduc-
ing the threat of future conflict onsets. This lowers the likelihood of future rebel-
lion and permits the group in power a higher level of exploitation of the defeated
group in the future as well. The benefit of a smaller share of surplus allocated to
the defeated group has to be compared with the loss in surplus produced by the
shrinking of the working population.7 The international community intervenes in
two ways. One is with a cap on the level of deaths in mass killings. From a modeling
perspective, this is equivalent to assuming that mass killings have zero cost below
the threshold and an unbounded cost above. The second type of intervention is the
minimum level of fairness with which the group that conquers power can treat
the powerless. Beyond some threshold, unfairness is considered completely unac-
ceptable. But here, too, the intertemporal nature of the model resurfaces because
constraints on unfairness can arise not only from external intervention but also
from the threat of future rebellion. Therefore, even without external pressure for
fairness, the group in power might have an incentive to exterminate the opposition
in order to eliminate any future contestation of its power.
Considering KS and EMR together allows one to shed new light on the phe-
nomena under study. It is true that reducing the cost of rebellion may induce more
rebellions, as pointed out in KS; but it is also true that the defeated group will have
elevated incentives to rebel in the future. This will moderate the level of exploi-
tation that the winner can impose on the loser and hence reduce the incentive
to enter a conflict in order to control power, unless one considers the possibility
of the elimination of future claims. Consequently, the murdering of part of the
opponent’s population becomes a strategy that could be considered profitable by
both sides. External pressure for a fair division of surplus, as pointed out in EMR,
can increase the likelihood of mass killings further.
In section 7.4 we present a formal analysis for the last paragraph, using a two-
period simplification of the EMR (2015) model, which allows for an easier inte-
gration with KS’s perspective. The integrated model will also be consistent with
the six stylized facts of mass killings discussed in section 7.2 and with the pros and
cons of intervention discussed in this section.
7.4. An Integrated Model

We now present a modified and simplified version of the EMR model that
brings together the main questions raised in KS (2013) and EMR (2015),
informed as much as possible by the observed facts. While the two models
examine the role of international intervention in civil wars, and while both
include the option of rebellion and the possibility of mass killings, the ques-
tions addressed are not the same. KS are interested in the inf luence of foreign
intervention on conf lict onset and EMR on the economic and social condi-
tions leading to mass killing incentives, with or without foreign intervention.
The goal of what follows is to clarify the complementarity between the two
models and sometimes to contrast the two approaches within an integrated
framework.
Consider a country populated mainly by two groups, i and j, of population size
(Ni, Nj). Hence, the total population is N = Ni + Nj. We take the convention that
group j is currently in power. The group in government decides on the distribu-
tion of the surplus, G, between the two groups. This surplus is composed of eco-
nomic output and rents derived from natural resources. Output, Y, is produced
by labor. We assume a rigid labor supply, so that output is proportional to popula-
tion: Y = βN. We can think of β as determined by the existing stock of capital and/
or of land, as well as by technology. In addition, the country obtains income from
the exploitation and export of natural resources, E. Hence, the government’s reve-
nue is G = βN + E. We denote by σ the share of output-generated (rather than rent-
βN
generated) revenue, σ = . Let the share going to the group in opposition be
Ni G
λ j , where λ j stands for the fairness of the distribution chosen by group j, which
N
is the group in power. Thus, when λ j = 1, shares are distributed equally to both
groups, in proportion to their size within the overall population. If the opposition
accepts the share of the revenue, there is no international intervention and hence
no externally imposed fairness.
This situation may be disrupted by rebellion of the group that is out of

power and that seeks to seize political control. In order to capture the various
static and dynamic incentives in the simplest possible form, it is sufficient to
consider two periods, of which period 2 is simply a stylized representation of
“the future.” Hence, when no ambiguity is possible, time 1 variables will be left
without time superscript. In case of rebellion at time 1, output is reduced by d
(destruction) and the win probability for group i is equal to the group’s relative
size, Ni . 8 Whichever group h = i, j wins, the victorious group seizes (or keeps)
N
power, consumes all the current surplus, perpetrates Mh mass killings on the
opponent population, and sets new distribution shares over future government
revenues, that is, it sets λ h2.
As to future resource rents, they are discounted by a factor δ and their
distribution is conditioned by the threat of rebellion by the group now in
opposition and by external pressure for a minimum level of fairness. In KS,
these external pressures are modeled as a defined third agent with her own
preferences that can end up victorious if she decides to intervene in the con-
f lict in period l, and there is no period 2. In order to make the model com-
mon to both contributions, we shall assume all the pressure is external and
hence exclude the role of changes in the future threat. Hence, the group in
power chooses the maximum level of unfairness compatible with the exter-
nal pressure for a minimum degree of fairness, λ. Clearly, whichever group
is in power, it will always choose the maximum externally tolerated degree
of unfairness.
Once one of the groups conquers power, it decides on the number of oppo-
nents to murder.9 In both approaches, KS and EMR, the (partial) elimination of
the opponent has “positive” and negative effects and hence the optimal choice will
be a balance between the two. In EMR, mass killings have a double effect on the
distribution of the surplus. The opponents are fewer, and hence they are less of a
future threat (the “positive” effect);10 but the elimination of opponents reduces
the overall surplus to be created and distributed in period 2 (the negative effect).
The KS model also has a double effect of murders.11 Following EMR, we assume
{ }
here that there is a cap M ≤ min Nj , Ni beyond which the cost imposed on the
perpetrator is unbounded.
We also assume that group members act in a coordinated manner so that each
group maximizes the payoff of a representative group member. Group i decides to
rebel if the group’s expected payoff from conflict exceeds the payoff from remain-
ing peaceful.
The payoff to group i if remaining peaceful is
Ni δ N 1 N
Πip = λ j G+ λj i G = λj i G (1)
N 1−δ N 1−δ N
and, if rebelling,
Ni éê δ æç Nj − Mi ö÷ ù
Πri = ê G−d+ çç1 − λ ÷÷(G − βMi)úú
Nê 1 − δ èç N − Mi ÷÷ø úû
ë (2)
Nj δ Ni − Mj
+ λ
N 1 − δ N − Mj
(
G − βMj . )

The corresponding payoffs for the group initially in power are
1 é Nù
Π pj = ê1 − λ j i úG (3)
1 − δ êë N úû
and, in conflict,
Nj éê δ çç
æ Ni − Mj ö÷ ù
ú
Π rj=
Nê
êG − d − s + ç
1 − δ ççè
1 − λ
N − Mj ÷÷÷ø
(
÷÷ G − βM
j )
ú
ú
ë û (4)
N δ Nj − Mi
+ i λ (G − βMi).
N 1 − δ N − Mi

Following KS, we assume that the government can be subject to economic

sanction, s, by the international community. Note that in our temporal extension
of their model, the fairness of the distribution has to be increasing in s. The argu-
ment is that because of the higher sanction-related cost of conflict to government,
the opposition would rebel more easily, so that peace demands a higher level of
fairness.
We start by characterizing the “optimal” mass killing policy by whoever is the
winner in case of conflict. Player h in power will choose M h ≤ M , below or equal
to the killing “cap,” to maximize
é N − Mh ù é N − Mh ù
ê1 − λ k ú(G − βM ) = ê1 − λ k ú[E + β(N − M )]. (5)
ê ú
N − M h ûú h ê N − M h ûúú h
êë êë
Differentiating the payoff after victory with respect to Mh, we obtain:
∂Πvh λNhE
= − (1 − λ)β. (6)
∂M h (N − M h)2
Note that when all the surplus is produced by labor, that is, rents E = 0, we have
∂Π vh
< 0 , so that the highest payoff is obtained without mass killings, Mh=0.
∂M h
Remark 1: The existence of nonproduced rents is a necessary condition for mass

killings.
It can be checked that for E > 0 the payoff is convex in Mh.12 Let us denote by
( )
M h0 N h the threshold mass killing, such that Π ch(M h0) = Π hc(0). That is,
λ 1−σ
M h0(Nh) = N − N h. (7)
1−λ σ
Therefore, when in power, group h will murder M h* = M opponents if M ≥ M h0;
otherwise it will choose M h* = 0.13 It follows that the larger is the threshold, M h0,
the less likely it is that there will be mass killings.
From (7) we can deduce the following observation:
Remark 2: Given M , mass killings are more likely the smaller is the share of pro-
duced, rather than rent-derived, revenue, σ, and the higher is the required fairness, λ.
The relative importance of nonproduced rents, typically coming from natural
resources, is a fundamental determinant of conflict and mass killings.14
The pressure for fairness can have an indirect effect on mass killings. Whenever
λ ≤ σ , we have M 0 > N and hence group h would not have incentives to carry
h k
out mass killings even without any external intervention on the level of atroci-
ties committed. But, for any parameter configuration, there always is a threshold
level of fairness, λ h0, above which group h will try to exterminate the opponent.
Formally,
Remark 3: (1) With λ ≤ σ , M h* = 0.15 (2) With λ > σ, we have: (2i) if
Nh 1−λ σ , Nh 1−λ σ ,
≥ M h* = M ; and (2ii) if < M h* = M if M ≥ M h0, and
N λ 1−σ N λ 1−σ
M h* = 0 if M < M h0.
With this characterization result, one can compute the expected payoff from
conquering power in any situation of conflict. Given the expected payoffs, we can
now examine the decision to trigger conflict or not.
If λ < σ , no mass killings occur even in the case of open conflict. In this case,
one can obtain the level of fairness that makes i indifferent between peace and
rebellion. It is
d
λ j = 1 − (1 − δ) . (8)
G
Suppose the group in power starts with the level of fairness exactly fixed at this
point. Then we have the following result.
Remark 4: Let the group, j, in power apply the level of fairness that just makes the
opposition indifferent between peace and conflict. Then an increase in the government
revenue, G, or a decrease in the cost of conflict, d, will induce the opposition to initiate
conflict.
As in KS, a reduction of the cost of rebellion makes groups in opposition more
likely to trigger conflict. A peaceful status quo can also be broken by an unex-
pected change in λ or σ. In the first case, this could result from a decision by the
international community to tighten the fairness requirement. The second kind
of change may come about from a sudden increase in the relative size of nonpro-
duced rents within government revenue.
Let us now assume that the conditions for mass killings are met, so that
we can examine the core questions addressed by KS and EMR. In particu-
λ 1−σ
lar, we henceforth assume that λ > σ and hence that > 1. As is con-
1−λ σ
ventional in the literature, we take the group in power to be the largest group,
Nj / N > 1 / 2 > Ni / N .
7.4.1. Equilibrium with One-Sided Mass Killings Risk

We start by considering the case in which the group in power is the only one with
the incentive to carry out mass killings. This will be the case when
λ 1 − σ Nj M λ 1 − σ Ni
1− ≤ < 1− . (9)
1−λ σ N N 1−λ σ N
λ 1 − σ Nj
When ≥ 1, the condition is satisfied for all M > 0 and hence the
1−λ σ N
λ 1 − σ Nj
choice is to exterminate the maximum tolerated. However, when < 1,
1−λ σ N
the group j in power will carry out mass killings only if M is sufficiently large.
We have already obtained the result that the maximum payoff for group j that
d
is compatible with i remaining peaceful is λ j = 1 − (1 − δ ) . Now we have to
G
verify whether j prefers the payoff associated with peace at λ j, or triggers conflict,
and eliminates M opponents when victorious.
The peace (maximum) payoff for the group in power j is
Nj G N
Π pj = + i d. (10)
N 1−δ N
The conflict (with mass killings) payoff for j is
Nj é æ N − M ö÷ ù N δ Nj
Πrj = êG − d − s + δ çç1 − λ i ÷÷(G − βM )úú + i λ G. (11)
ê 1 − δ çè N − M ÷ø
Në û N 1−δ N
Comparing the two payoffs, the following result can be verified.
Remark 5: Let the group in power, j, be the only one with incentives for mass killings
in case of conflict. Then, the profitability of conflict with mass killings relative to peace
decreases with higher costs of conflict, d and s, and a lower cap on mass killings, M , but
increases with the level of required fairness, λ.
It comes as no surprise that the larger the costs of conflict, the more the group
in power will prefer peace to conflict. This also means that it will be more ready
to make concessions to the opposition in order to avoid conflict, as KS point out.
However, the roles of the pressure on fairness and the cap on mass killings are
more complex. More fairness leaves no better option for increasing the payoff
than decimating the opponent, and a stricter cap on mass killings will reduce the
likelihood of conflict (when associated with the elimination of the opponent).
7.4.2. Equilibrium with Two-Sided Mass Killings Risk

In view of Remark 3, besides the necessary condition that λ > σ, incentives for
mass killings depend on whether the size of the group is larger or smaller than
1−λ σ
λ 1−σ
. We focus first on the case in which the size of both groups is smaller
than this threshold. This is the case when λ is not much larger than σ. Then,
N 1−λ σ
the group h = j, i of size h < will commit mass atrocities whenever
N λ 1−σ
M M h0 λ 1 − σ Nh
≥ = 1−
N N 1−λ σ N .
Taking into account that Nj > Ni, a first observation is that a policy of being
more strict on the number of potential murders will deter the smaller group first,
the one in opposition. Therefore, moderate intervention may end up hurting the
smaller (rebel) group and not the government. It is useful to see this argument
from the other extreme. Suppose that the international community starts with a
very strict policy, a very low M , and no murders are being observed. If it is very
costly to keep the necessary high number of observers for a strict implementation
of a low M , efficiency considerations may push toward reducing such cost and
hence implicitly accepting a higher number of (unreported) murders. This reduc-
tion will eventually have a discontinuity and mass killings will jump from zero to
the full M . The first perpetrator will clearly be the group in power.
Remark 6: Suppose that λ > σ and that Nh < 1 − λ σ and
M h0 N λ 1−σ
M λ 1 − σ Nh
≥ = 1− for h = j, i. A reduction of M reduces mass killings com-
N N 1−λ σ N
mitted by the two contenders and, beyond a threshold, the rebels stop having incentives
to kill in case of victory; the government remains the sole perpetrator of atrocities below
that threshold.
We now also briefly discuss the case where both groups h = j, i are of size
Nh 1−λ σ
> . This corresponds to a situation where fairness constraints are
N λ 1−σ
powerful and nonproduced rents amount to a large share of the economy. In this
case both groups will always have incentives to perform mass killings after vic-
tory, independently of the size of M .
The effect of international intervention is more complex than in the one-sided
case. Since the rebels would also perpetrate atrocities, the concession that the
group in government will have to make to keep peace depends on M . Therefore,
in order to examine the effect of policies on the net incentive for conflict (followed
by mass killings), one needs to take into account the change in both the peace and
the conflict payoffs.
In particular, the payoff from conflict for group h is
éN
Πrh =
Nh δ ê h (Nk − Nh)M úù ,
(G − d − sh) + (G − βM ) ê −λ ú (12)
N 1−δ êN N(N − M ) ú
êë úû
where sh = s for h = j and sh = 0 for h = i.

Let us first discuss the incentives of group h = i, initially in opposition.
Remember, M < Ni < 1 / 2 < Nj. One can now see that a decrease in M increases
the conflict payoff to group i and hence decreases the peace payoff to the govern-
ment (as the rebels are more expensive to “buy off”). In contrast, for the govern-
ment (h = j) the impact of M on the conflict payoff is ambiguous, and thus the
total effect is ambiguous. Trying to directly reduce mass atrocities has ambiguous
effects and may end up increasing mass killings.
From equation (12), it follows that an increase in the required fairness, λ,
reduces the payoff of conflict for the rebels i and hence increases the peace payoff
for the group j in power, as a lower compensation is sufficient for peace. However,
the payoff to the government j from conflict with mass killings also goes up, and
once more we have an ambiguous net effect on the relative profitability of mass
murders from the point of view of the group in power.
To summarize, we obtain that international intervention has ambiguous
effects. When pressing for a fair treatment of the losers of conflict, the interna-
tional community may end up triggering further conflict and mass killings. Also,
when becoming stricter on the number of killings toward which to close one’s
eyes, the intervention may unwittingly help the group in power rather than an
insurgent minority.

In this chapter we highlighted six stylized facts about mass killings that have
occurred since World War II. We then proposed a model that incorporates (or
is consistent with) such features and that allows us to reach a number of theory-
driven predictions. The proposed simplified version of Esteban, Morelli, and
Rohner (2015) also allows us to draw some precise comparisons with the third-
party intervention theory by Kydd and Straus (2013).16 The overall picture that
we obtain is full of caveats. Not only does the positive analysis point out that
resource discoveries and democratizations can be a curse (by fueling mass killing
incentives) if materializing at the “wrong” time, but also that the normative impli-
cations—in terms of third-party intervention—have to be carefully considered.
Even if the international community were to agree that defending populations
from mass killings is a top priority (something not altogether clear prior to the
UN resolution on Libya), we point out that neither a threat of direct intervention
(effectively putting a cap on allowed mass killings) nor the imposition of mini-
mum standards to be used for the treatment of defeated minorities can be evalu-
ated in the absence of considering the economic structure and social divisions. In
some cases (e.g., in ethnically polarized societies and in low-labor-productivity
cases), such measures could even backfire.17 How to design a policy function that
maps various initial conditions to an optimal M and λ is beyond the scope of this
chapter, but a “flexible” direction has to be advocated.
Obviously, some of the episodes of mass killings in the last sixty years fit the
assumptions of the model better than others. The Darfur event, for example, is
one that arguably fits the model well, since there were indeed two well-identified
groups, a great prevalence of natural resources, relatively low labor productiv-
ity, and a looming democratization (see, e.g., Straus, 2005, 2006; de Waal 2007).
However, even if some cases do not conform exactly to the stylized assumptions of
the model, EMR (2015) show that the theoretical predictions of the model (more
or less reproduced by the aforementioned remarks) find significant support in the
data. Hence the mechanism operating in the model must play some relevant role
even when the exact situation on the ground does not match the model assump-
tions completely, for example involving more than two groups or not involving
valuable natural resources at all.
The sad recent history of Syria, Iraq, and Libya will certainly push scholars to
focus on multilateral conflict more than before, and we believe that this (expected)
new focus will bring new and important insights for the understanding and pre-
vention of future outbreaks of violence and repressions. However, given stylized
fact 2, the understanding of mass killings relates mostly to the incentives of gov-
ernments, and hence we do not believe that the main insights of the proposed
model will change much when considering situations with more than one group
not controlling power. Similarly, the imposition of power sharing among multi-
ple groups should have, qualitatively, very similar effects to increases in λ in the
model. Thus, we believe that the main tenets of the rationalist explanation that
we have provided will prove robust to most extensions involving multiple groups.
One distinctive feature of mass killings that clearly separates this deadly
option from other forms of weakening an opposition group (e.g., imprisonments,
internments, expropriations, and disenfranchisements) is that mass killings are
designed to reduce the size of the opponent group or groups, either directly or
by causing refugee outflows and displacements (i.e., a multiplier effect; see Krain
2000, 41). The possibility of a multiplier effect, caused for example by the vicin-
ity of a country expected to keep open borders, could constitute an incentive
amplification factor to be considered in future work about the dynamics of forced
migration. However, if a government tries to displace minority groups without
killings, the underlying logic is somewhat different, because killings are irrevers-
ible, while displaced populations are often looking for opportunities to return or
retaliate. We will certainly study in future research the important dynamics of
forced migration, as sometimes a complement to and sometimes a substitute for
direct eliminations.18
Notes
1. Recall the distinction generally endorsed throughout this volume: mass atrocities include,
but are not limited to, mass killings and genocides. The latter require the specific intent to
target and eliminate a group enumerated in the UN Genocide Convention (on definitions
and data, see c hapters 1, 2, and 3). When looking at the data (see Anderton’s c hapter 3 in
this volume), it is indeed the case that the genocide events in the PITF dataset (Political
Instability Task Force 2010) are perpetrated by government-controlling groups, whereas
in datasets that include broader categories of mass killings, the frequency of events with
mass killings on both sides in the same event is larger.
2. On early warning, see chapter 24 in this volume.
3. On trust and violent conflict, see Rohner, Thoenig, and Zilibotti (2013a, 2013b).
4. See Anderton (2014) for a decision-theoretic model in which the two forms of violence can
be chosen simultaneously.
5. They also point out that autocracy does not remain a statistically significant explanatory
variable for genocides (there called mass killings) when one reduces the omitted variable
bias and accounts for unobserved heterogeneity.
6. Easterly, Gatti, and Kurlat (2006) find that mass killings are most likely for countries with
intermediate income levels.
7. Notice that this dynamic argument explains why mass killings seem to be specific to civil
wars and not of international wars (stylized fact 4). In KS, instead, the atrocities would be
equally effective in an interstate war.
8. In a general game-theoretic model of conflict, Esteban and Ray (2011) show that the equi-
librium win probabilities are approximately equal to the population shares of the groups
involved. See also Morelli and Rohner (2015).
9. Note that in KS only government can commit mass killings if victorious. The subject group
simply consumes the entire surplus but does not kill the former enemies.
10. This implies that a minority group could remain peaceful even with a more unfair distribu-
tion. As mentioned, for the sake of simplicity we are excluding this effect.
11. In KS, the “positive” effect is the increase in the probability of victory, while for EMR it is
the increase in the expected share of the surplus; the cost in KS is the international sanc-
tion proportional to the number of murders, while in EMR the cost is (in proportion β) the
lost surplus.
12. The convexity of the payoff function with respect to Mh would be further exacerbated,
as argued in Casper and Tyson (2014), if we also consider explicitly the collective action
problem and the related military compliance. Their focus is on individual participation
decisions under incomplete information, whereas we abstract from collective action prob-
lems and information issues.
13. This corner solution results from the convexity of the benefit in the reduction of the size
of the opponent together with a linear cost. In KS, the effect of mass killings on the payoff
also has a linear cost but, with no explicit justification, the effect on the expected benefit is
assumed to be concave. Hence, KS obtain an interior “optimal” level of atrocities.
14. Of course if a country has a lot of resources (Norway being a top example) but (l) is ethni-
cally homogeneous and (2) has a high opportunity cost of conflict, resources may be a
blessing rather than a curse.
15. Notice that when E = 0, as in Remark l, we have σ = 1 ≥ λ and hence no mass killings.
16. See also Favretto (2009) and Grigoryan (2010).
17. Other moral hazard concerns regarding humanitarian interventions are discussed, e.g., in
Rowlands and Carment (1998), Kuperman (2008), and Rauchhaus (2009).
18. On these topics, also see chapter 6 (substitution; backfire conditions), chapter 11 (migra-
tion), and chapter 13 (open borders), all in this volume.
References
Albright, M. K., and W.S. Cohen. 2008. Preventing Genocide: A Blueprint for U.S. Policy Makers.
Genocide Prevention Task Force. Washington, DC: United States Holocaust Memorial
Museum.
Anderton, C. H. 2014. “Killing Civilians as an Inferior Input in a Rational Choice Model of
Genocide and Mass Killing.” Peace Economics, Peace Science and Public Policy 20, no. 1:
327–46.
Bae, S., and A. Ott. 2008. “Predatory Behavior of Governments: The Case of Mass Killings.”
Bellamy, A. J., and P. D. Williams. 2011. “The New Politics of Protection: Cote d’Ivoire, Libya
and the Responsibility to Protect.” International Affairs 87, no. 4: 825–50.
Besançon, M. 2005. “Relative Resources: Inequality in Ethnic Wars, Revolutions, and
Genocides.” Journal of Peace Research 42, no. 4: 393–415.
Casper, B. A., and S. A. Tyson. 2014. “Military Compliance and the Efficacy of Mass Killings at
Deterring Rebellion.” Working paper. New York University.
Chirot, D., and C. McCauley. 2006. Why Not Kill Them All? The Logic and Prevention of Mass
Political Murder. Princeton, NJ: Princeton University Press.
Colaresi, M., and S. Carey. 2008. “To Kill or to Protect: Security Forces, Domestic Institutions,
and Genocide.” Journal of Conflict Resolution 52, no. 1: 39–67.
de Waal, A. 2007. “Darfur and the Failure of the Responsibility to Protect.” International Affairs
86, no. 6: 1039–54.
Easterly, W., R. Gatti, and S. Kurlat. 2006. “Development, Democracy, and Mass Killings.”
Journal of Economic Growth 11, no. 2: 129–56.
Economy 123, no. 5: 1087–1132.
Esteban, J., and D. Ray. 2011. “Linking Conflict to Inequality and Polarization.” American
Favretto, K. 2009. “Should Peacemakers Take Sides? Major Power Mediation, Coercion, and
Bias.” American Political Science Review 103, no. 2: 248–63.
Grigoryan, A. 2010. “Third-Party Intervention and the Escalation of State-Minority Conflicts.”
Heger, L., and I. Salehyan. 2007. “Ruthless Rulers: Coalition Size and the Severity of Civil
Conflict.” International Studies Quarterly 51, no. 2: 385–403.
Hoffman, S., R. C. Johansen, J. P. Sterba, and R. Vayrynen. 1996. The Ethics and Politics of
Humanitarian Intervention. South Bend, IN: University of Notre Dame Press.
Holzgrefe, J. L., and R. Keohane, eds. 2003. Humanitarian Intervention: Ethical, Legal and Political
Dilemmas. Cambridge, UK: Cambridge University Press.
Hultman, L. 2009. “Uncivil Warfare in Civil War.” Mimeo. Swedish National Defence College
and Uppsala University.
Hultman, L. 2010. “Keeping Peace or Spurring Violence? Unintended Effects of Peace
Operations on Violence against Civilians.” Civil Wars 12, nos. 1–2: 29–46.
Politicides.” Journal of Conflict Resolution 41, no. 3: 331–60.
Krain, M. 2000. “Democracy, Internal War, and State-Sponsored Mass Murder.” Human Rights
Review 1, no. 3: 40–48.
Kuperman, A. J. 2008. “The Moral Hazard of Humanitarian Intervention: Lessons from the
Balkans.” International Studies Quarterly 52, no. 1: 49–80.
Kydd, A. H., and S. Straus. 2013. “The Road to Hell? Third-Party Intervention to Prevent
Atrocities.” American Journal of Political Science 57, no. 3: 673–84.
Mann, M. 2005. The Dark Side of Democracy: Explaining Ethnic Cleansing. Cambridge,
UK: Cambridge University Press.
Mansfield, E., and J. Snyder. 2005. Electing to Fight: Why Emerging Democracies Go to War.
Cambridge, MA: MIT Press.
Montalvo, J., and M. Reynol-Querol. 2008. “Discrete Polarisation with an Application to the
Determinants of Genocides.” Economic Journal 118, no. 533: 1835–65.
Morelli, M. and D. Rohner. 2015. “Resource Concentration and Civil Wars.” Journal of
Development Economics 117: 32–47.
Political Instability Task Force. 2010. “Genocides.” Dataset: http://globalpolicy.gmu.edu/pitf/.
Querido, C. 2009. “State-Sponsored Mass Killing in African Wars—Greed or Grievance?”
International Advances in Economic Research. Published online.
Rauchhaus, R. 2009. “Principle-Agent Problems in Humanitarian Interventions: Moral Hazard,
Adverse Selection, and the Commitment Dilemma.” International Studies Quarterly 53, no.
4: 871–84.
Rohner, D., M. Thoenig, and F. Zilibotti. 2013a. “War Signals: A Theory of Trade, Trust and
Conflict.” Review of Economic Studies 80, no. 3: 1114–47.
Rohner, D., M. Thoenig, and F. Zilibotti. 2013b. “Seeds of Distrust? Conflict in Uganda.” Journal
Rowlands, D., and D. Carment. 1998. “Moral Hazard and Conflict Intervention.” In M. Wolfson,
ed., The Political Economy of War and Peace. Boston: Kluwer Academic Publishers.
Rotberg, R., ed. 2010. Mass Atrocity Crimes: Preventing Future Outrages. Washington,
DC: Brookings Institution.
Rummel, R. 1994. “Power, Genocide and Mass Murder.” Journal of Peace Research 31, no. 1: 1–10.
Rummel, R. 1995. “Democracy, Power, Genocide, and Mass Murder.” Journal of Conflict Resolution
39, no. 1: 3–26.
Schneider, G., M. Bussmann, and C. Ruhe. 2012. “The Dynamics of Mass Killings: Testing
Time-Series Models of One-Sided Violence in the Bosnian Civil War.” International
Interactions 38, no. 4: 443–61.
Scully, G. 1997. “Democide and Genocide as Rent-seeking Activities.” Public Choice 93, nos.
1–2: 77–97.
Snyder, J. 2000. From Voting to Violence: Democratization and Nationalist Conflict. New York: W. W.
Norton.
Straus, S. 2005. “Darfur and the Genocide Debate.” Foreign Affairs 84, no. 1: 123–33.
Straus, S. 2006. “Rwanda and Darfur: A Comparative Analysis.” Genocide Studies and Prevention
1, no. 1: 41–56.
Teson, F. R. 1997. Humanitarian Intervention: An Inquiry into Law and Morality. 2nd ed. Irvington-
On-Hudson, NY: Transnational.
Valentino, B., P. Huth, and D. Balch-Lindsay. 2004. “‘Draining the Sea’: Mass Killing and
Guerrilla Warfare.” International Organization 58, no. 2: 375–407.
Weiss, T. G. 2007. Humanitarian Intervention: Ideas in Action. Cambridge, UK: Polity Press.
Wheeler, N. J. 2002. Saving Strangers: Humanitarian Intervention in International Society.
Oxford: Oxford University Press.
8
Genocide
From Social Structure to Political Conduct
N é stor Duc h-Brow n a n d A n ton io Fon f r í a
8.1. Introduction
Economists have become increasingly interested in studying conflict and the
different mechanisms that can be put toward its resolution. After some decades
of theoretical and empirical research on conflict, many questions remain unan-
swered and constitute valuable avenues for ongoing and future research. The
results that emanate from these efforts will, one hopes, help eradicate the humani-
tarian disasters that these confrontations generate.
Millions of people are systematically injured and killed through violent
conf licts in many parts of the world. In addition, these extreme episodes of
violence have several long-term effects that obstruct social, political, and
economic progress. Violence is one-sided in asymmetric conf licts, where
complex aggressive actions are undertaken by powerful actors—generally
political authorities—against largely defenseless civilian minorities. In this
case, how, why, and when do violent conf licts become viable strategies for
perpetrators? This is one of the most basic questions that researchers in the
conf lict literature have posited, yet suitable and satisfactory answers are still
to unfold. Our aim is to contribute to the literature by trying to understand
the conditions under which genocides and other violent conf licts are more
likely to appear.
Anderton and Carter (2009; also see chapter 1 in this volume), when refer-
ring to the broader field of conflict economics, argue that conflict and economics
combine in four distinct ways. For the purposes of this chapter, we find their
first three ways particularly relevant. First, they say, conflict is a choice, meaning
that there is no need to rely on irrationality arguments in order to understand
violent conflicts. Conflicts imply costs and benefits to the agents involved, and
when conflict arises it must be optimal for at least one of the rivals. Hence, we
190
Genocide 191
should study the conduct of the different sides in a conflict situation in order to
understand their decisions, their strategies, and their objectives. They continue,
second, by stating that conflict affects economic outcomes. According to these
authors, economic performance will be modified depending on the existence
and/or the severity of conflict. And third, conflict is affected by economic vari-
ables, meaning that the framework within which conflict occurs conditions the
actions of the different sides.
If we place choice at the center of the analysis, it is possible to devise a
conceptual scheme whereby a set of economic variables affects the conduct
(choice) of the participating agents and this, in turn, affects economic out-
comes. Here, the most relevant aspect of the analytical proposal is related to
conduct, given that the agents’ behavior can alter both the structural condi-
tions that shape the set of possible choices available and also the outcomes
of the interaction. Strategy is important for almost any human activity. If we
define social interactions—as in the sociology of markets—as “social struc-
tures characterized by extensive social relationships between markets, sup-
pliers, customers, and governments” (Fligstein and Dauter 2007, 105), then
the previous argumentation could easily be applied to many different human
actions. We believe that this is precisely the case for conf licts in general and
genocide in particular.
Our conceptual scheme consists of two main blocks that will be described
in the following sections in detail. In addition, these two blocks are comple-
mented by a set of preconditions that shape and constrain the set of choices the
agents in our framework can make. Taking all these elements into consider-
ation, it is possible to explain the outcome of the interplay of different concep-
tual constructs. The first block is formed by the analysis of the social structure
of societies that gives rise to intergroup asymmetries—perpetrators on one
side and victims on the other. In the second block, the conduct of the perpetra-
tors is studied. This part of the analysis is particularly important because it
influences enormously the frequency and intensity of violence. It rests deeply
on the decisions made by the authority group as well as on the nature of inter-
group rivalry. This treatment of genocide tries to shed some light on under-
standing the mechanisms that can be found behind the atrocities committed
by violent groups.
The chapter is structured as follows. In section 8.2 we briefly sketch the sug-
gested framework to tackle the study of genocide. Section 8.3 elaborates on
the concept of social structure and how it is mapped to the political structure.
Political conduct, as a consequence of the political structure, is analyzed in sec-
tion 8.4. Here, particular attention is placed on the nature of political rivalry and
the role barriers to mobility play in the explanation of the incumbent’s choice of
repression. Section 8.5 connects social structure and political conduct with geno-
cide, and section 8.6 concludes.
8.2. The Analytical Framework

Violent conflict can take several forms. We focus on genocides and mass killings,
although we believe that many of the ideas we develop can be applied to other
forms of violent conflict. Genocide and related terms and data are discussed in
chapters 1, 2, and 3 in this volume, and so are not discussed here. For our pur-
poses, we rely on a practical definition, viewing genocide as systematic killings
of a social group based on their ethnicity, race, religion, class, or political beliefs,
and perpetrated by a state or other hegemonic group (Stewart 2011). One impor-
tant feature is that in genocide, violence is one-sided, which requires some sort of
dominance or incumbency.
Our framework to understand social rivalry (conflict) and to analyze geno-
cide and mass killings relies on two basic concepts: social structure and political
conduct. According to some scholars (for instance, Stewart 2011 and references
therein), the majority of genocides involve government as the perpetrator,
although some other socially dominant groups can also participate.1 Dominant
groups gain power by political competition in ways similar to how dominant
social groups gain supremacy by social competition. Hence, one should analyze
links between social structure and political behavior. Social structure is defined
as the social configuration resulting from intra- and intergroup social interactions
in a given society. The social structure is mapped onto a corresponding politi-
cal structure according to a complex set of social interactions taking place within
any given society. In cases of high concentration, the political structure can be
suitable for exercising monopoly-like, dictatorial power. Dominant social and/or
political groups use a wide range of strategies either to protect or to extend their
power against actual and/or potential rivals.
The political market includes agents undertaking mutually advantageous
exchange under the rules determined by a social contract. These rules can be
modified by means of an electoral process in the case of democracies or by other
mechanisms in nondemocratic regimes (Tisdell and Hartley 2008). In addition,
technological change is believed to be one important source of power, both for
unstable structures, when potential innovators develop a radical innovation that
gives them a privileged position (at least for some time), and for increasing an
already existing dominance, when the incumbent monopoly has strong incentives
to innovate and keep its dominant position. In a political setting, the incumbent
ruler or party has strong incentives to keep office to sustain the extraction of politi-
cal rents, but potential entrants (opposition groups or parties) also have incentives
to become the new rulers or leader since the gains (rents) can be sizeable. This
political competition can be fiercer if, in addition, the degree of rivalry between
agents—the intensity of hatred (Glaeser 2005)—is high. There can be cases where
incumbents perceive threats that are actually nonexistent. However, what matters
in these cases is the perceived threat of rivalry and not necessarily any actual rivalry.
Genocide 193
As mentioned, our approach requires the interplay of social structure and

political conduct. Interactions between them occur in imperfectly competi-
tive environments characterized by notorious power imbalances. This frame-
work allows one to study how social groups can derive competitive advantages
through the adoption of strategies conditioned on a given social context. Here, a
distinction is needed between tactical and strategic decisions (Church and Ware
2000, 465). Short-term actions conditional on an existing situation are tactical
decisions. Strategic decisions (conduct) have long-run implications for social
structure, the competitive environment faced by agents. Strategic decisions are
relevant because, by influencing the environment (structure), they affect tacti-
cal decisions. The ability of strategic variables to affect tactical decisions arises
because of commitment. Strategic decisions commit agents to follow a given tac-
tic because it is in their own best interest; commitment, in turn, depends on the
irreversibility of strategic decisions previously adopted, where irreversibility is
one way of making a commitment credible.
Before explaining in detail, in sections 8.3 and 8.4, the two main building
blocks of our conceptual framework to explain violent conflict, we briefly describe
some of the preconditions for the appearance of violent conflict in general, and
genocide in particular, that are discussed in the literature. They can be grouped in
three broad categories: (1) political and institutional factors, (2) socioeconomic
factors, and (3) resource and environmental factors.
Political and institutional factors refer to weaknesses that can allow the con-
ditions for violent conflict to appear. These include, for example, institutional
structures that allow political exclusion, corruption, rotten state institutions, or
identity politics, among many others. Socioeconomic factors refer mostly to dif-
ferent sources of social inequalities, exclusion, and marginalization. Hence, cul-
tural, ethnic, or religious minorities are more exposed to violence than are other
groups. Poverty is also strongly linked to conflict. Many conflicts are motivated
by economic aspects such as access to strategic resources, or even resource scar-
city or environmental insecurity, which thus can cause violent conflict as well.
Each of these factors can decisively affect the likelihood of the emergence of vio-
lent conflict. In addition, once a conflict unfolds, any of the factors mentioned can
arise and sustain it over time.
8.3. Social Structure

The characteristic question in the social sciences is how social interactions affect
individual and collective behavior. Economists have been criticized for not tak-
ing social interactions properly into account, focusing exclusively on economic
transactions using markets. However, in the last few decades, economics research
related to other disciplines (mainly sociology and psychology) has generated
new insights under the umbrellas of social and/or behavioral economics. These
relatively new fields in economics have expanded our knowledge about individual
decision-making and collective conduct (Becker and Murphy 2003; Kahneman
2003).
Societies, defined as groups of people involved in interpersonal relation-
ships, are formed by communities, and these in turn are composed of persons.
Individuals differ in many dimensions but at the same time share some character-
istics with others, creating specific social groups or communities. These groups
are based on cultural, political, economic, religious, racial, or ethnic common
traits and shared interests. One individual can be a part of different communities
at the same time. The pattern of social arrangements that simultaneously emerge
from and determine the actions of individuals and groups is the social structure
of a given society. Social interactions, also called nonmarket interactions, are a
particular form of externalities defining situations in which the preferences of an
individual are affected by the actions of a reference group (Scheinkman 2008).
Social structure can be defined at the aggregate level, and refers to the system
of social and economic stratification and social institutions that define the inter-
actions between large social groups. The set of relationships between the different
communities or social groups are intergroup social interactions. Social structure
has also a microscale, composed of the set of norms and rules that outline the
behavior of individual actors in a social system. Within communities, individu-
als also interact with each other, developing intragroup social interactions. In the
middle, an intermediate level defining the structure of the social networks linking
individuals and/or organizations also plays an important role.
In the economics literature, social interactions are believed mainly to exhibit
strategic complementarities. These occur when the marginal utility to a given indi-
vidual of undertaking an action increases with the average amount of the action
taken by members of the reference group (Scheinkman 2008). However, also rel-
evant is the consideration of strategic substitutes. As a matter of fact, in the field
of the economics of industrial organization, the distinction between strategic
substitutes and complements structures the analytical approach to understand
strategic interactions. It is useful to understand firms’ strategic behavior, but it is
also relevant to the study of international relations, political science, and conflict
(Baliga and Sjöström 2011).
Social interactions among communities manifest in patterns of stratification
and dominance. The type and intensity of social interactions and their corre-
sponding patterns are context-specific and depend on many factors. For instance,
social composition, defined as the number of communities within a given society,
is a relevant characteristic. In societies where there are a large number of equally
sized and relatively homogeneous communities, the likelihood of the appear-
ance of a dominant group is rather low. Still, an arrangement would be needed
to organize the different social groups, that is, a political or institutional system.
Genocide 195
Similarly, when a society is formed by just one large and perfectly homogeneous
community, there is no room for rivalry either, and hence the pattern of intergroup
dominance is also absent. When there are a reduced number of communities,
however, the likelihood that intergroup interactions will take the form of rivalry
or conflict is higher. In this case, many different outcomes of intergroup social
interactions are possible. For instance, to govern the late stages of the Ottoman
Empire, the so-called Young Turks imposed a strictly hierarchical social system
that subordinated non-Muslims as second-class subjects deprived of basic rights
(Adalian 2013).
A scenario with a small number of communities is not a sufficient condition
for extreme rivalry or dominance. For simplicity, consider a case with only two
social groups. If the two groups are close in the sense that they share all char-
acteristics except one—for instance, they only differ in terms of political ideol-
ogy but share all other cultural and institutional characteristics—then there is
little room for dominance even though the political rivalry could be intense. If
we define a social niche as the set of resources that is capable of supporting a set of
communities, intergroup competition will depend on the degree of niche overlap
or attempts to obtain very similar resources. Hence, when niche overlap is large,
there is going to be intense intergroup competition; when niche overlap is small,
then intergroup competition will be weak. Obviously, if the niches do not overlap
at all, two communities will not be competing for the same resources and hence
there is going to be a vacuum of intergroup social interactions. For example, in
Guatemala the Mayan population represented more than half of the total popula-
tion. Economic conditions broke down the barriers that had kept them relatively
socially isolated and forced them to migrate from the highlands to the coast and
the capital in search of jobs. This brought them into increased contact with the
non-Mayan population. Rather than losing their identity, this experience rein-
forced their indigenous identity (language, customs, and religious practices) but
also increased their claims for land and their demands for other rights. While
Mayans were demanding more recognition and social participation, indigenous
organizations were defined by the government as subversive and excluded from
even minimal forms of political expression that were permitted for other popula-
tion segments (Jonas 2013).
Intergroup social interactions are also impacted by conditions that block the
participation of some groups in cultural or political spheres. These are defined as
structural, institutional, or behavioral conditions that allow dominant groups to
enjoy superior benefits for a significant length of time (Cabral 2008). Here, we
focus on the structural and institutional components, dealing with the behavioral
one in the next section. For instance, the control of essential resources by one
community could obstruct other social groups to participate in certain activities.
The existence of intragroup network effects—benefits to individuals of one com-
munity derived from activities undertaken by members of the same group—could
eventually block members of other communities and relegate them to marginal

positions or niches. In addition, propaganda in favor of (or against) certain groups
can also have discriminatory effects and could eventually deter participation of
other groups in certain social niches.
Intergroup interactions can generate cooperation, through specializa-
tion or division of labor, but also can spur competition for resources or status.
Competition between social groups tends to produce conflict between them. The
social and economic changes that intensify competition between social groups
indirectly raise the rate of conflict in a given society. Those social groups occupy-
ing certain critical positions within that structure will have more power. Power, as
used in political science, economics, and sociology, describes asymmetries in the
interactions between individuals and social groups (Simon 1953). This asymme-
try may be related to different factors. Some of them were identified already with
the structural barriers defined previously and hence are exogenous to the social
structure; others result from specific actions or behaviors of individuals or social
groups and hence are endogenous to the social structure. Whereas dominance indi-
cates an advantaged position that permits a privileged group to act independently
of rivals, and hence to benefit from leadership, the abuse of a dominant position is
a completely different issue.
Social structure can be used as a source of power. However, power as a concept
encompasses two components: dominance and leverage. Dominance is defined as
power based upon force or the threat of force. Two types of dominance can be dis-
tinguished: intrinsic dominance, based upon the ability to use force, and derived
dominance, based upon the ability to form coalitions or alliances with partners.
Yet power can also be based upon resources that cannot be taken by force or lever-
age (Hand 1986). Leverage is defined as power based upon inalienable resources,
and it can arise from any number of asymmetries. For instance, specific skills or
knowledge could be sources for leverage at the individual level, whereas social
capital could be a source of leverage at the collective level.
Although it seems clear that the more dominance (leverage) a social group
has the less additional power it can gain by adding more dominance (leverage),
the precise interaction between dominance and leverage is less clear. Apparently,
situations of extreme dominance may neutralize leverage. In extremely concen-
trated social structures, smaller groups may have very few options for leverage
since most resources can be taken by force by the larger group. The source of lever-
age for these groups relies on their social capital. This can be illustrated with the
case of Armenians in the late Ottoman Empire, who had an extensive and solid
network of elementary and secondary schools. Through education, they hoped to
preserve their culture and identity and to obtain participation in government and
other social dimensions (Adalian 2013).
Factors that amplify levels of intergroup competition also boost the likelihood
of collective action, constituting the basis for the manifestation of social conflict.
Genocide 197
Hence, interactions that generate intergroup inequalities tend to generate con-

frontations, and interactions that foster social integration and intergroup equality
of opportunities reduce grievances and diminish the odds of social conflict.
According to Sidanius and Pratto (1999), group-based inequalities can be cre-
ated and maintained through three primary intergroup behaviors: institutional
discrimination, aggregated individual discrimination, and behavioral asymme-
try. Intergroup interactions can hence generate and maintain culturally defined
group-based social hierarchies, which do not necessarily exist in all societies.
Arbitrarily set hierarchies can be based on ethnicity, religion, nationality, and so
on. Human social hierarchies consist of a hegemonic group at the top and nega-
tive reference groups at the bottom.
This approach suggests that widely shared cultural ideologies, also known
as legitimizing myths, provide the moral and intellectual justification for these
intergroup behaviors. Sidanius and Pratto (1999) differentiate between hier-
archy-enhancing (e.g., racism, chauvinism) and hierarchy-attenuating (e.g.,
feminism) legitimizing myths. These are the types of beliefs, ideologies, and
institutions that greatly facilitate violent threats and, eventually, genocidal epi-
sodes. As an example, consider the case of the Jewish people in Germany. Before
the Holocaust, Jews were concentrated in a few highly visible economic sectors
such as banking, publishing, and the metal and clothing trades. This proved con-
venient to blame them for economic depressions, bankruptcies, and unemploy-
ment (Niewyk 2013).
Social division is exacerbated when social domination is institutionalized.
Even an imperfect mapping from social structure onto the political dimension
can exclude minorities from political participation and representation. We have
argued already that a salient characteristic of genocide is the asymmetric nature
of the conflict. The literature on genocide, mass killing, and other mass atroci-
ties and violent conflicts has emphasized its unbalanced nature. In this case, an
authority group uses specific tactics and strategies—a continuous implementa-
tion of policies to damage rival groups. This authority group has power, however
defined (political, military, social), to inflict on a weaker group severe damage
and impede its social or political mobility. Hence, genocide is more likely to occur
when the structure of the political market shows some degree of concentration.
In such a setting, the degree of competition is low. This could be the result of the
existence of entry barriers, giving more power to the incumbent ruler. As an illus-
tration, consider the case of East Pakistan (now Bangladesh). Conflict originated
when in 1948 the ruling elites declared their intention to make Urdu, the lan-
guage of only 7 percent of the population, the sole and official state language, at a
time when Bengalis comprised 55 percent of the population. This is an example of
barriers erected by powerful elites that prevented Bengalis from participating in
the political process and resulted in their exclusion from state power. It gradually
drove Bengalis to demand autonomy and self-determination (Jahan 2013).
8.4. Political Conduct

Under some circumstances, such as a small number of social groups, diversified
ideologies, and blockaded social participation, the resulting social structure could
be heavily concentrated and characterized by the presence of a hegemonic group.
This group normally will be able to control many institutional aspects, such as the
political arena, leaving a fringe of smaller dominated groups with limited partici-
pation in relevant dimensions of public life. The dominant or hegemonic group
may have attained such a position through large investments, for instance by cre-
ating and sustaining legitimizing myths. Hence, the hegemonic group has the
characteristics of an incumbent, a leader that has already invested large amounts
of resources that can be considered sunk. (A sunk cost, in economics, is an outlay
of resources into an investment that cannot be redeployed to other uses.) On the
contrary, if the leader and the follower(s) had the same size and the same capabili-
ties, instead of one-sided violence we would observe an escalation of violence—in
the form of a civil war, say—since the follower would be in a position to retaliate
if attacked or would even be able to initiate a rebellion or insurrection. A peace-
ful scenario based on mutual deterrence could also develop, sustained by each
group’s fear of retaliation from the rival.
The analysis of the political market requires the consideration of the social
structure in which the different communities of a society operate. We have seen
that demographic, social, political, and economic factors shape the social struc-
ture of a society. In addition, a nation’s political system determines its voting and
collective choice rules. The institutional arrangement for collective governance
will be heavily influenced by the characteristics of the social structure. Political
participation will be dominated by the hegemonic social group, and social minor-
ities will be underrepresented.
Genocides and mass killings are purposeful actions carried out by authority
groups. We should pay attention then to conduct that allows agents to attain power
and, more importantly, the abuse of that power that extends the incumbents’
authority once achieved. The sources and consequences of authority can be var-
ied. Hegemony can be obtained and sustained in a number of ways. For instance,
it can be the result of relative efficiency, that is, citizens’ ability to participate in
social and political life. Similarly, it will be preserved fundamentally by remaining
more efficient than the rivals. There are other sources of power, and leaders can
maintain their advantage even if they are inefficient. In many cases, leaders are
early innovators that rise to leadership using economically efficient strategies or
by developing new alternatives. For instance, the equivalent in political markets
would be parties that win elections through political innovation offering to imple-
ment new or innovative policies or by offering an alternative to existing options.
In some cases, however, hegemony can be created by means of anticompetitive
tactics or inefficient strategies. Again, the corresponding example in political
Genocide 199
markets would be the acquisition of office by coups d’état or other antidemocratic

means as is customary with authoritarian regimes. Once they reach power, lead-
ers engage in a number of strategies, some efficient, others inefficient, to main-
tain their hegemonic positions (Rosenbaum 1998). However, at some point, the
leader eventually loses ascendancy. In some cases, the framework of interactions
develops too fast for any leader to maintain control. In other cases, the change
is guided by third-party intervention. In addition, bad administration and other
inefficiencies can also contribute to a dominance loss. Nonetheless, as a general
rule, even when a powerful leader becomes inefficient, the system works very
slowly to remove that leader. As an illustration of two opposing episodes with
genocidal consequences, consider the cases of the Armenians and the Holocaust.
In the former, the Young Turks, the dictatorial triumvirate that governed in the
late phase of the Ottoman Empire, started out as members of a clandestine orga-
nization that eventually staged a revolution in 1908, replaced the incumbent ruler
one year later, and seized power by a coup in 1913 (Adalian 2013). In the latter, in
the late 1920s and early 1930s, the Nazis assembled a sufficient amount of elec-
toral support to become the largest political party in Germany. Hitler had joined
the party in 1919 and, by 1933, managed to be appointed chancellor of a coalition
government, a position that allowed him to become a dictator (Niewyk 2013).
Government has long been viewed as a natural monopoly of force. Nevertheless,
from a dynamic perspective, government may only be a monopoly in a transitory
sense. A well-functioning democracy regularly has perfect (or at least some) com-
petition for the right to run the monopoly until the next election (Schumpeter
1942; Tullock 1965). Mulligan and Tsui (2008) suggest (and show) that this idea
can be generalized to all regimes, even in cases where regime turnover may be
motivated by revolutions or insurrections. Hence, all types of regimes (democratic
and nondemocratic) can be viewed as regulated public utilities. For instance, only
one firm at a time can deliver water or electricity, but the firm doing so competes
with others for the contract via multiple mechanisms. This interpretation allows
Mulligan and Tsui (2008) to define the degree of political competitiveness as the
size of entry barriers into the process, allocating the rights to temporarily run the
government, or the natural monopoly on force. The literature on political compe-
tition has dealt with the issue (Tullock 1987; Grossman and Noh 1994; Bueno de
Mesquita et al. 2003). In their paper, Mulligan and Tsui (2008) develop a model
of political competition that endogenizes entry barriers. They show that politi-
cal market structure, measured by the number of actual challengers to an incum-
bent regime, can be a poor indicator of political competitiveness. Rather, political
competitiveness is better measured by the size of political entry barriers.
Scholars in the genocide and conflict literatures are well aware of this. For
instance, Hirshleifer (1994) refers to the concept of the decisiveness of a conflict,
which he relates to a natural monopoly (in the conflict industry). In a model pro-
posed by Bae and Ott (2008), genocide and mass killing are equilibrium outcomes
only when there is an authoritarian regime, associated with a “pure monopoly”

equivalent in the political market literature. Anderton (2010) uses a bargaining
model to explain four rationalist conditions for genocide—preventive genocide,
indivisibility, elimination of a rival, and political bias—and concludes that all
of these motives have a common denominator: the authority group’s monopoly
power.
A social and political structure characterized by a hegemonic dominant group
and a fringe of small and marginalized groups, where sunk investments play a
relevant role in the determination of conduct, can be defined in terms of interac-
tions of strategic groups. Hence, strategic groups are one relevant dimension of
the social and political structure, and considering their role and evolution would
greatly improve our understanding of social conflict. In the conjectured mapping
of social structure onto political structure, an interesting question is whether
rivalry is greater between groups or within groups (between members of the same
group). The latter approach would call for a thorough analysis of the industrial
organization of genocide. Some recent papers have devoted some effort to delin-
eating theoretical explanations in particular conflict cases. Recent literature ana-
lyzing organized crime (Baccara and Bar-Isaac 2008; Leeson and Rogers 2012),
rebellions (Weinstein 2007), and street gangs (Skarbek and Sobel 2012) are
examples. Moreover, Stokes and Gabriel (2010) argue that genocide should
lie at the heart of organizational studies. According to these authors, “[g]‌enocide
requires the collaboration of numerous formal organizations, including armies,
suppliers, intelligence and other services, but also informal networks and groups.
Far from being peripheral to the world of organizations, we shall argue that geno-
cide is central to some of its core concerns” (462).
The former approach—rivalry between groups—is the object of this chapter
and deals with intergroup social rivalry that eventually turns into one-sided vio-
lence against a weaker group, defined in terms of relative power and not necessar-
ily in terms of relative size. The fundamental question here relates to how groups
act competitively. Even if previous research has indicated that competition and
cooperation are not necessarily mutually exclusive, and that thus a multiplicity of
permutations between competitive and cooperative actions can simultaneously
be present (Dranove, Peteraf, and Shanley 1998), in what follows we focus mainly
on the effects of intergroup competition.
Strategic (social or political) groups should be defined as sets of individuals
within a society that share common cultural qualities which make them different
from individuals outside the group on one or more dimensions of their strategy.
This means that strategic groups will make strategic decisions that will be dif-
ficult for other groups to imitate without incurring significant additional costs.
Indeed, strategic groups will devote resources to raise the cost of strategy rep-
lication to other groups. Known as mobility barriers, these moderate the abil-
ity to move from one group to the other, thus preventing imitation (Caves and
Genocide 201
Porter 1977; Porter 1979). The concept of mobility barriers implicitly assumes
that the degree of rivalry differs both within and between groups (Peteraf 1993).
Mobility barriers in the strategic groups approach are equivalent to entry barri-
ers and can be defined “either as absolute costs of movement from one strategic
position (i.e., group) to another (becoming vertically integrated, for example), or
as the operating cost penalty relative to the group incumbents that the imitators
must face” (McGee 2003, 264). It would then be decisive to determine whether
the rivalry between groups is determined systematically and predictably in terms
of asymmetry.
Mobility barriers provide strategic groups with a relative advantage over other
groups. Indeed, adopting a group structure would make no sense unless it pro-
tected against and imposed overheads on outsiders’ attempts to imitate group
members. Thus, differences in performance between group members and out-
siders tend to persist in the medium to long term (McGee and Thomas 1986).
However, mobility barriers provide insufficient theoretical support for the con-
nection between group characteristics and performance. Along these lines,
recent studies have integrated mobility barriers into a new theorization of the
strategic group, one based on members’ strategic interactions. Group-level effects
on performance derive from power, efficiency, or differentiation produced by the
members’ strategic interactions. While group members’ strategic interactions are
critical to group-level effects on performance, mobility barriers help to sustain
the group-level effects that these strategic interactions produce by deterring com-
petitors from gaining entry into the group and by enhancing the strategic interac-
tions among members.
Conduct proceeds on the assumption that decision makers are motivated
in the choice of their actions by the goal of long-run benefits (even if this is
a simplification) and that they have reasonable estimates of how their actions
affect the likelihood of their success and that of their rivals. In the quest for
benefits, agents can employ potentially predatory conduct on three interrelated
fronts: (1) engage in practices designed to deter potential entrants; (2) engage
in practices that disadvantage current rivals without necessarily causing their
exit, but that relax the competitive constraint exercised by them over the hege-
monic group; and (3) engage in actions that cause the exit of an existing rival
(or rivals).
But what is a social or political barrier to mobility? The literature offers a
wide and varied collection of definitions and interpretations (see, e.g., McAfee,
Mialon, and Williams 2004). We take the view that a barrier to political mobility
is a benefit that is derived from supremacy. It is the additional advantage that a
group can have as a sole consequence of being hegemonic. Additionally, we define
incumbency as the ability to move first. This “first mover advantage” is sometimes
gained by leadership, which is normally related to the fact that the first participant
can eventually gain authority that rivals simply cannot match.
Caves and Porter (1977) recognized that entrants can be entirely new partici-
pants or agents already there, but they eventually may find it profitable to enlarge
their operations and participate in a different social dimension. Hence, mobil-
ity barriers are erected not to impede or block entry by new contenders but to
avoid competing rivals in another sphere to increase or extend their participa-
tion at the expense of the hegemonic leader in a given niche. This type of pro-
cess occurs in political markets and in societies at large. Incumbents in political
markets, for instance, may not worry about the appearance of small nationalist
parties in remote regions. However, incumbents would otherwise try to impede
the creation of new nationwide competitors that could represent a real alternative
for citizens. Along the same line of argument, the incumbent government will try
to extend its dominance (time in office) at the expense of their rivals. This can be
achieved by not allowing rival parties to thrive.
Social groups differ in multiple dimensions that necessarily condition the set of
strategies they can effectively employ. A group with technological leadership can
direct its barrier-creating investments toward (social) innovation; and a group
with salient communication skills can focus on political or ideological advertise-
ment.2 These decisions create and sustain competitive advantages against rivals,
whether they contribute or not to blocking participation. However, “mediocre”
groups—those that achieved dominance by chance, by past efficiencies that are
no longer effective, or by force—can only resort to sabotage to impede rivals from
expanding their participation in shared niches.
We focus exclusively on intergroup rivalry, that is, the competitive response
of a group to tactical actions taken by another group, characterizing the rivalry
between defined strategic groups as asymmetric and of the leader-follower type.
In this framework, an incumbent group can adopt different types of optimal
behavioral strategies in response to the threat of a competitor (Fudenberg and
Tirole 1984). The nature of the strategy depends on the effect of the incumbent’s
strategic investment on its own benefits, the slope of the reaction curves—that
is, strategic complements or substitutes—and whether the incumbent chooses to
accommodate or deter new entrants.
If a strategic investment makes the incumbent look tough—that is, it reduces
the rival’s benefits—and the reaction curves have negative slopes (strategic sub-
stitutes), the incumbent should adopt the top-dog strategy that means overin-
vestment: being big or strong to look tough and aggressive to accommodate an
existing rival. However, if the investment level makes the incumbent look tough
and the reaction curves have positive slopes, the incumbent should underinvest to
minimize competition (a puppy-dog strategy). On the other hand, when invest-
ment makes the incumbent look soft, it will underinvest when competition is
based on strategic substitutes (a lean-and-hungry-look strategy), and it will over-
invest if the reaction curves have positive slopes (a fat-cat strategy). If instead the
incumbent wants to deter entry by potential rivals, then the type of competition
Genocide 203
is no longer relevant, only the effect of investment is (Tirole 1989). In this case, if
the investment makes the incumbent look tough, it should adopt a top-dog strat-
egy again. However, if the investment makes the incumbent look soft, then the
incumbent should adopt a lean-and-hungry-look strategy (underinvestment).
In the taxonomy of Fudenberg and Tirole (1984), aggressive behavior falls
into the top-dog category, whereby the incumbent overinvests in order to accom-
modate a rival or deter new competitors from entering the market. This is the
optimal-deterrence strategy in models such as Spence (1977) and Dixit (1981).
Investment makes the incumbent appear tough and, in response, the entrant cow-
ers and produces less (the reaction curve is downward sloping). Top-dog behavior
is optimal in this case, whether or not the entrant is actually prevented from enter-
ing the market. Even if entrance is allowed, as in the Dixit model, the incumbent
will play the top-dog role to increase its profits thereafter (Gilbert 1989).
Genocides can be interpreted as contests where one side aims to increase its
power by destroying the rival’s power and the incumbent is prepared to engage
in mass killing of the enemy in order to achieve these goals. In general, a contest
can be described as a situation in which individuals or groups devote resources
in order to win valuable prizes. The efforts of the participants in such contests
can be productive or destructive (Amegashie 2012). In this analytical setting,
productive effort means actions that directly affect the probability of success,
while destructive effort (also known as sabotage) refers to actions that reduce the
rival’s performance (Konrad 2000). From this perspective, individuals or groups
employ strategies that are intended to damage someone else’s success instead of
improving one’s own.
Sabotage is defined as “actions that put the rival at a competitive disad-
vantage” (Ordover and Saloner 1989, 565). The Wikipedia page for the entry
“sabotage” describes it as “a deliberate action aimed at weakening a polity or cor-
poration through subversion, obstruction, disruption, or destruction.”3 History
is full of cases and examples of unethical behavior (see Pench 1996 for business
stories of sabotage). From an industrial organization perspective, sabotage is
associated with raising rivals’ costs (alternatively, reducing rivals’ revenues), for
example via an exclusionary strategy (Salop and Scheffman 1983; Chowdhury
and Gürtler 2013).
We mentioned before that the nature of interactions among rivals is funda-
mental to understanding their behavior. From industrial organization theory
we distinguish between strategic complements and strategic substitutes. Using
these concepts, two types of conflicts are distinguished. First are conflicts char-
acterized by strategic complements, where aggression feeds on itself cyclically,
as the incentive to be aggressive for one player increases the probability that the
other player chooses violence as well. This triggers an escalating violent spiral as
in civil wars or interstate wars. Second, a scenario where strategic substitutes pre-
vail: here, the toughness of one player forces the opponent to back down. Hence,
the incentive of one player to choose violence decreases in the probability that the
opponent chooses violence. Hence, these types of models can be used to repre-
sent one-sided violence, as in genocides and mass killings.
Jews in Nazi Germany were not only deemed a racial enemy but also repre-
sented a perceived threat in terms of social privilege and political opposition.
They were defined as a powerful enemy, not based on any actual dominance but
on their leverage. The Nazis promised to restore German greatness by overpower-
ing those social groups that in their national-racist ideology they saw as a threat
to “pure” German people and values. Similarly, Stalin defined kulaks as enemies
of the state; in Rwanda, Tutsi were defined as adversaries even as the distinction
to Hutu was often fuzzy and based only on an identity card system (Totten and
Parsons 2013; chapter 15 in this volume). Almost all genocides have relied on
mechanisms of social classification, sustained by legitimizing myths built from
fantastical arguments, which in addition are rapidly mutated according to the
requirements of political skirmishes in which the perpetrators are involved. In
all these cases, and in many more, the incumbent ruler identifies a social group as
a military or otherwise threatening enemy (which justifies systematic violence).
This identification lies in the real or imputed power of the enemy group, including
its social, economic, political, cultural, and ideological power (Shaw 2015). As
we discussed, part of this power comes from dominance, part comes from lever-
age. Enemy social minorities only have leverage left, since they are dominated, but
they still have power. The unfolding of violence is geared to destroy the leverage-
based power of the enemy group and to prevent them from participation in social,
economic, and political spheres.
Power is the outcome of social interactions and structure. Two parties in an
interaction each have some power. The analysis of power lies in the assessment of
discovering and describing the parties’ relative strength. In an asymmetric inter-
action, as in the one we have been describing in this chapter, the balance of power
would be also asymmetric, but even the weak party will still have some power. We
defined power as being composed of dominance and leverage. In an asymmetric
interaction, even if the strong party is dominant, and removes any possibility of
dominance of the weak party by force, the weak party will still keep some leverage.
In contests with strategic substitutes and asymmetric players, in order to lower
the rival’s power, the incumbent has to find ways to increase its own power. If
the power of the rival group is based only on leverage—because dominance has
been eliminated by means of social competition or by the use of softly coercive
actions—the only way to further increase its own power is by annihilation of
the rival group’s sources of leverage. In social contexts, as we argued before, that
leverage is based on the community’s social capital.
It is our view that genocide can be interpreted within a framework of strategy
that raises a rival’s costs (reduces a rival’s revenue). An innovative incumbent pol-
ity can rely on novel and original policies to maintain its appeal to society. In so
Genocide 205
doing, it can not only perpetuate its power but also avoid opposing parties or social
groups augmenting their social relevance. Similarly, a government with great com-
munication skills would base its strategy on propagandistic or populist actions to
remain in office. However, a mediocre, greedy, fearful, and unethical incumbent
ruler can only use force to increase its relative dominance over rival groups when
the strength of rivals depends exclusively on their leverage (social capital).
8.5. Genocide
In the preceding sections, we defined social structure in terms of the number
and size distribution of the participating social groups. Social interactions are
shaped by a set of factors (historical, institutional, and economic) that determine
the relative power of each social group. Next, we discussed conduct within inter-
group political rivalry and focused particularly on situations of relevant asym-
metries between a small number of groups. To some extent, the social structure
and its mapping over the political structure can condition the range of strategies
the groups can use. For instance, in societies where there is only one homoge-
neous social group or the power of different social groups is uniformly distrib-
uted, the resulting political structure can be competitive and political rivalry
can be intense. In contrast, highly concentrated social structures will generally
be mapped onto very unequal political structures, giving rise to authoritarian
or totalitarian regimes. In this case, hegemonic groups have some discretion
to deliberately change the social and political structure. Using the appropriate
strategy, social and/or political participation (entry) could be deterred, thereby
increasing social concentration and political power.
Political markets can be imperfect (Tisdell and Hartley 2008). Some of the
most relevant imperfections in political markets come from the fact that competi-
tion in democratic political markets is infrequent and discontinuous and is of the
all-or-nothing type. In addition, political incumbents have discretion since voters
cannot hold politicians to an agreed, and clearly specified, set of policies, so there
is room for opportunistic behavior. Finally, in some democracies there are only
two or three effective parties so that political competition is imperfect (a high
degree of market concentration). All these characteristics would imply that the
outcome of political markets is far from ideal. Some consequences of these market
imperfections from a conflict perspective could be a disproportionate share of
military expenditure vis-à-vis social expenditure (e.g., education, health), more
repression, and disruption of trade leading to slower economic growth, among
many others.
Imperfections in political markets can also arise from concentration where
one social group is able to dominate the political system. This incumbency gives
authority groups a first-mover advantage that they can use to shape the nature
and intensity of rivalry. Depending on their expectations about rivals, and con-
sidering their own gains, authority groups can adopt soft or tough behavior. Still,
even a tough hegemonic group can choose either to employ productive effort to
increase its own performance or destructive effort to reduce rivals’ performance.
This would be the case when the dominated social groups have some power in
terms of leverage. Genocide and mass killing is the outcome of destructive effort
(sabotage) by an incumbent authority group whose power—a strategic substitute
with rivals’ group power—can only be maintained or increased by eliminating
the leverage of a dominated group or groups.
Social capital improves productivity by reducing the social costs of doing busi-
ness and sharing a common set of norms and institutions, that is, information and
transaction costs. Genocide reduces social capital by increasing these costs not
only for the victims, but also for the perpetrators by means of reduced labor sup-
ply—which will have an impact on its relative price—and through the elimina-
tion of weak ties in the social network of society (Granovetter 1985).
8.6. Conclusions
This chapter proposes a conceptual framework to analyze genocide and other
violent asymmetric conflicts. In particular, we suggest that political conduct is
shaped by an underlying social structure and by the nature of rivalry between dif-
ferent social groups. Even dominated minorities have power in the form of lever-
age. But the social and political power of different groups function as strategic
substitutes. From this perspective, the accent should be placed on strategic con-
duct within a context of strong asymmetries across the agents involved. Genocide
and mass killing can be interpreted as a political or social strategy of raising rivals’
costs. This strategy encompasses deliberate, and often costly and illegal, acts to
damage a rival’s likelihood of success—in our case, success in the form of social
and political participation—by an incumbent authority group that believes its
power can only be maintained or increased by such means.
The main contribution of this chapter lies in the development of an analytical
framework to explain asymmetric conflicts that have genocide and mass killings
as a motivation. In essence, we are grafting concepts from the field of the econom-
ics of industrial organization onto political markets, particularly those that can
lead to mass murder and genocide. However, our approach has limitations. The
most relevant limitation, shared with models based on strategic conflict theory
and the public-choice approach to the study of conflict, lies in the fact that it lacks
an appropriate behavioral explanation for conduct. The approach taken in this
chapter, mapping social structure onto political structure, however, is an attempt
to overcome one of the main criticisms of this type of approach, namely that these
models “ignore the institutional set-up of the societies, and the strategies of the
Genocide 207
agents are supposed to be rational but not history dependent” (Vahabi 2009, 852,
italics in the original). Future research will have to further tackle this issue, as
well as others. That said, two important research tasks in the near future should be
the formalization of the conceptual scheme developed here, perhaps in line with
recent contributions in the literature (Baliga and Sjöström 2004, 2011), and the
validation of the proposal by means of empirical analysis (Anderton 2014).
Acknowledgments
We thank the editors and three anonymous reviewers for extremely helpful com-
ments and suggestions on an early version of this chapter. Errors and omissions
are our sole responsibility. We also thank our home institutions for providing a
nice intellectual atmosphere and administrative support for our research. The
views and opinions expressed in this chapter are those of the authors and do not
necessarily reflect those of our home institutions.
Notes
1. Broadly speaking, government has the basic resources—that may or may not be sufficient
for its purposes—in terms of productive means, raw materials, and political, legal, mili-
tary, and institutional power to compel part of the armed forces and population against
the victims. For example, in Darfur, both the Janjaweed and the Sudanese government
have been involved in attacks on the black African population. See Totten (2013) for more
details.
2. The possibility that government carries out a marketing plan, subject to the constraint of
available resources, can trigger demand for messages related to the existence of opposite
groups based on religious or political ideas, etc. It is perhaps more likely that people with
low income and low skills are influenced by such marketing, and so they become the main
target population in order to be recruited and to serve the incumbent government’s strate-
gies. (On media persuasion and mass violence, see c hapter 12, and on messaging and iden-
tity politics, see chapter 21, both in this volume.)
3. https://en.wikipedia.org/wiki/Sabotage, accessed January 30, 2015.
References
Adalian, R. P. 2013. “The Armenian Genocide.” In S. Totten and W. S. Parsons, eds., Centuries of
Genocide: Essays and Eyewitness Accounts. 4th ed. Abingdon, UK: Routledge, 117–56.
Amegashie, J. A. 2012. “Productive versus Destructive Efforts in Contests.” European Journal of
Political Economy 28, no. 4: 461–68.
Anderton, C. H. 2010. “Choosing Genocide: Economic Perspectives on the Disturbing
Rationality of Race Murder.” Defence and Peace Economics 21, nos. 5–6: 459–86.
Anderton, C. H. 2014. “Killing Civilians as an Inferior Input in a Rational Choice Model of
Genocide and Mass Killing.” Peace Economics, Peace Science and Public Policy 20, no. 2:
327–46.
Anderton, C. H., and J. R. Carter. 2009. Principles of Conflict Economics: A Primer for Social
Scientists. Cambridge, MA: Cambridge University Press.
Baccara, M., and H. Bar-Isaac. 2008. “How to Organize Crime.” Review of Economic Studies 75,
no. 4: 1039–67.
Bae, S. H., and A. Ott. 2008. “Predatory Behavior of Governments: The Case of Mass Killing.”
Baliga, S., and T. Sjöström. 2004. “Arms Races and Negotiations.” Review of Economic Studies 71,
no. 2: 351–69.
Baliga, S., and T. Sjöström. 2011. “Conflict Games with Payoff Uncertainty.” Mimeo.
Northwestern University. http://econpapers.repec.org/paper/rutrutres/200905.htm.
Becker, G. S., and K. M. Murphy. 2003. Social Economics: Market Behavior in a Social Environment.
Cambridge, MA: Harvard University Press.
Bueno de Mesquita, B., A. Smith, R. M. Siverson, and J. D. Morrow. 2003. The Logic of Political
Survival. Cambridge, MA: MIT Press.
Cabral, L. 2008. “Barriers to Entry.” In S. N. Durlauf and L. E. Blume, eds., The New Palgrave
Dictionary of Economics Online. London: Palgrave Macmillan [accessed February 15,
2015].
Caves, R. E., and M. E. Porter. 1977. “From Entry Barriers to Mobility Barriers: Conjectural
Decisions and Contrived Deterrence to New Competition.” Quarterly Journal of Economics
91, no. 2: 241–62.
Chowdhury, S., and O. Gürtler. 2013. “Sabotage in Contests: A Survey.” Mimeo. University of
Cologne. https://ideas.repec.org/p/uea/aepppr/2012_51.html.
Church, J. R., and R. Ware. 2000. Industrial Organization: A Strategic Approach. Boston: McGraw
Hill.
Dixit, A. 1981. “The Role of Investment in Entry Deterrence.” Economic Journal 90, no.
357: 95–106.
Dranove, D., M. Peteraf, and M. Shanley. 1998. “Do Strategic Groups Exist? An Economic
Framework for Analysis.” Strategic Management Journal 19, no. 11: 1029–44.
Fligstein, N., and L. Dauter. 2007. “The Sociology of Markets.” Annual Review of Sociology
33: 105–28.
Fudenberg, D., and J. Tirole. 1984. “The Fat Cat Effect, the Puppy Dog Ploy, and the Lean and
Hungry Look.” American Economic Review 74, no. 2: 361–68.
Gilbert, R. J. 1989. “Mobility Barriers and the Value of Incumbency.” In R. Schmalensee
and R. D. Willig, eds., Handbook of Industrial Organization. Vol. 1. Amsterdam: Elsevier,
475–535.
Glaeser, E. L. 2005. “The Political Economy of Hatred.” Quarterly Journal of Economics 120, no.
1: 45–86.
Granovetter, M. 1985. “Economic Action and Social Structure: The Problem of Embeddedness.”
American Journal of Sociology 91, no. 3: 481–510.
Grossman, H. I., and S. J. Noh. 1994. “Proprietary Public Finance and Economic Welfare.”
Journal of Public Economics 53, no. 2: 187–204.
Hand, J. L. 1986. “Resolution of Social Conflicts: Dominance, Egalitarianism, Spheres of
Dominance, and Game Theory.” Quarterly Review of Biology 61, no. 2: 201–220.
Hirshleifer, J. 1994. “The Dark Side of the Force.” Economic Inquiry 32, no. 1: 1–10.
Jahan, R. 2013. “Genocide in Bangladesh.” In S. Totten and W. S. Parsons, eds., Centuries of
Jonas, S. 2013. “Guatemala: Acts of Genocide and Scorched-Earth Counterinsurgency War.” In
S. Totten and W. S. Parsons, eds., Centuries of Genocide: Essays and Eyewitness Accounts. 4th
ed. Abingdon, UK: Routledge, 355–94.
Kahneman, D. 2003. “A Perspective on Judgment and Choice: Mapping Bounded Rationality.”
American Psychologist 58, no. 9: 697–720.
Genocide 209
Konrad, K. 2000. “Sabotage in Rent-Seeking Contests.” Journal of Law, Economics, and

Leeson, P. T., and D. B. Rogers. 2012. “Organizing Crime.” Supreme Court Economic Review 20,
no. 1: 89–123.
McAfee, R. P., H. M. Mialon, and M. A. Williams. 2004. “What Is a Barrier to Entry?” American
McGee, J. 2003. “Strategic Groups: Theory and Practice.” In D. O. Faulkner and A. Campbell,
eds., The Oxford Handbook of Strategy. Oxford: Oxford University Press, 261–301.
McGee, J., and H. Thomas. 1986. “Strategic Groups: Theory, Research and Taxonomy.” Strategic
Management Journal 7, no. 2: 141–60.
Mulligan, C. B., and K. Tsui. 2008. “Political Entry, Public Policies, and the Economy.”
National Bureau of Economic Research (NBER). Working Paper No. 13830. Cambridge,
MA: NBER.
Niewyk, D. L. 2013. “The Holocaust: Jews, Gypsies, and the Handicapped.” In S. Totten and
W. S. Parsons, eds., Centuries of Genocide: Essays and Eyewitness Accounts. 4th ed. Abingdon,
UK: Routledge, 191–248.
Ordover, J. A., and G. Saloner. 1989. “Predation, Monopolization, and Antitrust.” In
R. Schmalensee and R. D. Willig, eds., Handbook of Industrial Organization. Vol. 1.
Amsterdam: Elsevier, 537–96.
Pench, M. 1996. Dirty Business: Explaining Corporate Misconduct. London: SAGE.
Peteraf, M. 1993. “Intra-Industry Structure and the Response Towards Rivals.” Managerial and
Decision Economics 14, no. 6: 519–89.
Porter, M. 1979. “The Structure within Industries and Companies’ Performance.” Review of
Economics and Statistics 61, no. 2: 214–27.
Rosenbaum, D. I. 1998. Market Dominance: How Firms Gain, Hold, or Lose It and the Impact on
Economic Performance. Westport, CT: Praeger.
Salop, S. C., and D. T. Scheffman. 1983. “Raising Rivals’ Costs.” American Economic Review 73,
no. 2: 267–71.
Shaw, M. 2015. War and Genocide: Organized Killing in Modern Society. New York: Wiley. Kindle
edition.
Scheinkman, J. A. 2008. “Social Interactions (Theory).” In S. N. Durlauf and L. E. Blume, eds.,
The New Palgrave Dictionary of Economics Online. 2nd ed. New York: Palgrave Macmillan
[accessed January 15, 2015].
Schumpeter, J. A. 1942. Capitalism, Socialism, and Democracy. New York: Harper.
Sidanius, J., and F. Pratto. 1999. Social Dominance: An Intergroup Theory of Social Hierarchy and
Oppression. Cambridge, UK: Cambridge University Press.
Simon, H. A. 1953. “Notes on the Observation and Measurement of Political Power.” Journal of
Politics 15, no. 4: 500–516.
Skarbek, D., and E. Sobel. 2012. “The Industrial Organization of Street Gangs.” Journal of Gang
Research 20, no. 1: 1–17.
Spence, A. M. 1977. “Entry, Capacity, Investment, and Oligopolistic Pricing.” Bell Journal of
Stewart, F. 2011. “Economic and Political Causes of Genocidal Violence: A Comparison with
Findings on the Causes of Civil War.” MICROCON Research Working Paper 46. Brighton,
UK: MICROCON.
Stokes, P., and Y. Gabriel. 2010. “Engaging with Genocide: The Challenge for Organization and
Management Studies.” Organization 17, no. 4: 461–80.
Tirole, J. 1989. The Theory of Industrial Organization. New York: Cambridge University Press.
Tisdell, C., and K. Hartley. 2008. Microeconomic Policy. A New Perspective. Cheltenham, UK: Elgar.
Totten, S. 2013. “Genocide in Darfur, Sudan.” In S. Totten and W. S. Parsons, eds., Centuries of
Totten, S., and W. S. Parsons, eds. 2013. Centuries of Genocide: Essays and Eyewitness Accounts. 4th
ed. Abingdon, UK: Routledge.
Tullock, G. 1965. “Entry Barriers in Politics.” American Economic Review 55, no. 1: 458–66.
Tullock, G. 1987. Autocracy. Boston: Kluwer.
Vahabi, M. 2009. “A Critical Review of Strategic Conflict Theory and Socio-Political Instability
Models.” Revue d’Economie Politique 119, no. 6: 817–58.
Weinstein, J. M. 2007. Inside Rebellion: The Politics of Insurgent Violence. New York: Cambridge
University Press.
9
The Microeconomic Causes and

Consequences of Genocides and Mass
Atrocities
Pat r ic i a J ust i no
9.1. Introduction
For a long time, researchers and policymakers were interested in the causes and
consequences of armed conflict from the perspective of the security and capacity of
states. But since the mid-2000s or so, the scholarly literature has seen an increased
amount of published research on armed conflict from a household-level perspective.
This microlevel research on conflict, violence, and peace building has offered com-
pelling evidence that “the security and capacity of states may be closely entwined
with the security and welfare of their people” (Justino, Brück, and Verwimp 2013,
302; italics in original). Furthermore, “inattention [of peace building interventions]
to local conflicts leads to unsustainable peacebuilding in the short term and poten-
tial war resumption in the long term” (Autesserre 2010, 39–40). Economists and
other quantitatively oriented scholars thus have begun to do what genocide scholars
have done for a long time, for example, collection and analysis of detailed eyewit-
ness accounts (e.g., in Totten and Parsons 2013). Raphael Lemkin—the founder
of the field of genocide studies—of course displayed in extraordinary detail micro-
level aspects of genocide in his classic book, Axis Rule in Occupied Europe (1944).
The duration and potential resolution of armed conflict is affected by impor-
tant microfoundations that operate across two dimensions (Justino 2013a). The
first dimension involves the complex interactions that emerge in areas of conflict
among the behavior, choices, aspirations, perceptions, and expectations of ordi-
nary people; the behavior and preferences of armed groups; and the structure and
organization of violence. The second dimension has to do with the ways in which
violence (in its various forms and across time) may transform institutions, social
organizations, and social norms at the local level.
211
This relatively new field of the microeconomics of violent conflict has resulted
in important insights about processes of conflict and violence. But so far very few
studies have distinguished between different types of conflict and the forms that
violence may assume within different conflicts, including violence that expresses
itself in mass atrocities such as genocide. Thus, most microlevel findings to date
are generated either from contexts of civil war or from studies wherein conflict
is analyzed as a homogenous category measured by discrete violent events that
mark the onset or the end of specific episodes of armed conflict. The specific
microeconomic conditions that may facilitate mass atrocities, and the economic
and welfare consequences of such atrocities for individuals, households, and local
communities, have not been analyzed in the microeconomics of the conflict field
in much detail.
One important constraint is the absence of fine-grained data on different forms
of conflict and types of violence that can be matched with economic data at indi-
vidual, household, and community levels. Despite this constraint, some studies
have adopted creative ways of matching different administrative, household sur-
vey, and event datasets to analyze some of the microeconomic effects of known
mass atrocities such as those experienced in Bosnia (Kondylis 2005, 2007),
Cambodia (de Walque 2006), the Holocaust (Akbulut-Yuksel 2014), and Rwanda
(Friedman 2013; Justino and Verwimp 2013; Verwimp 2003, 2005). What these
episodic studies imply, among other things, is that we lack a systematic analysis of
the microeconomic causes and consequences of mass atrocities as distinct forms
of violence, both within and separate from civil war events.
The objective of this chapter, then, is to suggest a conceptual framework to
better understand the microeconomic causes and consequences of mass atrocities
that will, I hope, encourage future theoretical and empirical research. In particu-
lar, the chapter focuses on the analysis of one important factor in the study of the
microeconomic causes and consequences of mass atrocities in relation to other
forms of armed violence: their distributional dimension.
A large body of literature has given rise to an ongoing debate across policy
arenas and the social sciences about what constitutes a mass atrocity such as
genocide. Genocide is defined by the United Nations (UN) Convention on the
Prevention and Punishment of the Crime of Genocide as “acts committed with
intent to destroy, in whole or in part, a national, ethnical, racial or religious group”
(United Nations 1951). This definition has been contested in the literature due
to its narrowness, and several scholars have adopted more encompassing defini-
tions. For instance, Harff (2003) defines genocide and politicide as “the promo-
tion, execution, and/or implied consent of sustained policies by governing elites
or their agents—or, in the case of civil war, either of the contending authori-
ties—that are intended to destroy, in whole or part, a communal, political, or
politicized ethnic group” (58). Other scholars have adopted broader notions that
include the targeting of groups defined along specific characteristics, but also the
T he Microeconomic Causes 213
targeting and killing of large numbers of civilians “not in the course of military
action against the military forces of an avowed enemy, [but] under conditions of
the essential defenselessness and helplessness of the victims” (Charny 1999, 7,
cited in Easterly, Gatti, and Kurlat 2006). One of the more widely used defini-
tions of mass atrocity is that of Valentino (2004) and Valentino, Huth, and Balch-
Lindsay (2004) as “the intentional killing of a massive number of noncombatants.
Victims of mass killings may be members of any kind of group (ethnic, political,
religious, and so on) as long as they are noncombatants and as long as their deaths
were caused intentionally” (Valentino, Huth, and Balch-Lindsay 2004, 377–78).
This chapter adopts this broader definition (and when general reference is made
to mass atrocities, it includes genocides). Further discussion about the range of
definitions, and associated data, is provided in chapters 1, 2, and 3 of this volume.
A common thread across the definitions that is of importance to the arguments
made in this chapter is that mass atrocities specifically target certain population
groups intentionally. Often this targeting is the result of either perceived or real
conflicts over the distribution of power and resources in society (see Valentino
2004; Verwimp 2005; Straus 2006). This targeting, in turn, may result in profound
changes in the way in which resources and power will be distributed across soci-
ety. This chapter argues that distributional concerns play an important role in our
understanding of the microeconomic causes and consequences of mass atrocities.
Theoretically, a process of mass violence that results in fairer distributions (even if
not justified morally or ethically) may result in more even distributions and lower
the potential for further intergroup conflict and violence. Frequently, these types
of violence redress imbalances in favor of those committing the atrocities. Whether
this will be an improvement in relation to the status quo is debatable, however.
The chapter offers a detailed discussion of the distributional dimension of
mass atrocities over the following three sections. Section 9.2 summarizes recent
literature on the microeconomic causes and consequences of armed conflict in
general, particularly by drawing on the civil war–related literature. Section 9.3
then grafts that discussion onto the distributional dimensions of mass atrocities
such as genocides. Section 9.4 concludes and considers potential new directions
for further research.
9.2. Economic Causes and Consequences of Violent

Conflict: A Microlevel Perspective
Economic factors are closely entwined with the outbreak and legacy of violent
conflict, whatever its type (e.g., war, civil war, rebellion, mass atrocity, genocide).
From a general microlevel perspective, then, economic factors have been shown
to shape why individuals, households, and groups participate in acts of organized
violence or engage in armed groups. Violent conflict, in turn, has considerable
consequences for the economic welfare of individuals, households, and local

communities—sometimes over very long periods of time—which may have an
effect on the likelihood of conflict recurrence. Justino (2009) developed this
endogenous relationship between violent conflict and welfare outcomes at the
microlevel theoretically. This section summarizes the main points of that frame-
work and links them to extant theoretical and, especially, empirical work. While
this section relates to mass atrocities as well, a more focused discussion of mass
atrocities is deferred to section 9.3.
9.2.1. Causes: Individual Participation in Organized Violence

The analysis of economic factors in the outbreak of violent conflict has a long
history in political science and international relations. Mancur Olson (1965)
famously proposed that collective action problems can be solved through the use
of selective incentives, and this idea was adopted by many conflict scholars who
viewed the puzzle of individual participation in political violence as a particularly
extreme example of the collective action problem (Keen 1998, 2005; Hirshleifer
2001; Grossman 2002; this literature is reviewed in Lichbach 1995).
So, why do individuals participate in group violence? Through in-depth sur-
veys of ex-combatants, recent literature has shown that monetary incentives may
play only a marginal role in explaining the participation of individuals or groups
in forms of political violence. Other factors have been shown to be equally, if not
more, prominent such as (1) the role of coercion and abduction (Blattman and
Annan 2010; Humphreys and Weinstein 2008); (2) fear of violence from oppos-
ing factions (Kalyvas 2006; Kalyvas and Kocher 2007); (3) peer pressure allied
to community norms and sanction (Petersen 2001; Verwimp 2005; Pinchotti and
Verwimp 2007); (4) ideological and cultural identification with an armed group
(Goodwin 2001; Petersen, 2001; Wood 2003); and (5) the opportunity to voice
group or class discontent and accumulated grievances (Gurr 1970; Paige 1975;
Scott 1976; Wickham-Crowley 1992; Richards 1996; Wood 2003).
These reasons may or may not overlap with economic motivations. However,
it is clear that in some cases—even if not in all cases or at all times—the threat
of poverty and destitution, unmet aspirations, or lack of productive employment
may increase the likelihood of some individuals joining armed violence for rea-
sons related to economic incentives (Collier and Hoeffler 1998; Grossman 2002;
Walter 2004; Friedman 2013) or in order to better manage economic and security
risks that they and their families may face during conflict (Kalyvas and Kocher
2007; Humphreys and Weinstein 2008; Justino 2009). In particular, Kalyvas and
Kocher (2007, 183) have argued that “individuals may participate in rebellion not
in spite of risk but in order to better manage it.”
Even though participation in armed groups may be associated with an
increased risk of death or injury, nonparticipation also incurs considerable risks such
as the danger of a particular individual, household, or group being identified

with opposite factions and thus as potential enemies (chapter 19 in this volume).
Under these circumstances, armed groups may offer protection from indiscrim-
inate violence from enemy groups, in addition to privileged access to informa-
tion (Kalyvas 2006; Kalyvas and Kocher 2007). This argument was extended to
include economic protection in Justino (2009, 7–8), who argues that “the (in)
ability of households to protect their economic status in conflict areas may well
increase substantially the risk of nonparticipation (i.e., the probability of poverty
and destitution).”
Building on these findings, Justino (2009) suggested that individual or house-
hold participation in armed groups is a function of two interdependent vari-
ables. The first is the set of initial individual or household characteristics, which
determine the extent of their vulnerability to poverty. These include variables like
income, employment status, levels of education, and other welfare markers typ-
ically used in the literature to justify the participation of individuals in armed
groups, whether due to monetary incentives or fear of deprivation. A second
variable is the (expected) extent of exposure of individuals, their households, or
immediate social groups to violence during conflict, that is, their vulnerability to
violence. Theoretically, the interaction between these two variables determines
the probability of a given household or group (or its members) participating in
and supporting armed groups due to their effects on the costs of participation in
relation to the costs of nonparticipation. To a large extent, these costs are shaped
by the (expected) direct and indirect consequences of violent conflict on welfare
outcomes at the individual, household, and group levels (Justino 2009).
9.2.2. Consequences: The Microlevel Effects of Violent Conflict

The effects of violent conflict may be anticipated or not. Direct effects include
(1) changes in household composition caused by killings, injuries, and recruit-
ment, reflected in changes in who heads the household (conflict-affected commu-
nities usually experience an increase in female-headed households, particularly
widows), fertility decisions, and dependency ratios; (2) changes in the economic
status of households due to the destruction of assets, livelihoods, markets, and
infrastructure; and (3) the breakdown of families, economic structures, and social
networks caused by forced displacement and migration (further discussed in
chapter 11 in this volume).
These direct effects are now well documented: Exposure to violence, or simply
to widespread insecurity, often results in significant reductions in the income and
consumption expenditure of affected individuals and households due to the loss of
livelihoods and reductions in productivity and household accumulation of assets
and human capital. A large and growing body of literature has focused in par-
ticular on how violent conflict results in largely negative and long-lasting human
capital effects,1 sometimes affecting individual and household labor outcomes

and earnings capacity over several decades of time and across generations (Ichino
and Winter-Ebmer 2004; Merrouche 2006; Akbulut-Yuksel 2014; Justino, Leone,
and Salardi 2014). These effects can be further aggravated by the breakdown of
households and their social and economic protection mechanisms when fighting
results in the displacement of people and communities to safer zones. In addition,
individuals, households, and sometimes whole communities may not be able to
cope with the effects of violence as traditional economic coping strategies such
as informal social networks or income transfers may be considerably restricted
during violent conflict (Justino 2012).
Indirect effects, although less studied, have to do with local changes regarding
the access of households to exchange, employment, credit, and insurance mar-
kets; in social relations and norms of trust and cooperation; and in local institu-
tions and organizations. Indirect effects can also be due to national-level changes
in economic growth and the distribution of income, power, and resources.
For example, armed conflict may affect exchange, employment, and credit
markets through changes in the market prices of goods sold and purchased by the
household, changes in savings, and changes in risk behavior and preferences. At
the same time, social relations, cooperation, trust, local forms of governance, and
wider economic conditions are affected dramatically as well, leading to signifi-
cant changes in how households and communities relate to each other and how
society is organized. While the specifics of these institutional changes on indi-
viduals, households, and communities are underresearched, it is however known
that they translate into profound changes in the behavior and characteristics of
local populations, including their participation (voluntary or otherwise) in forms
of armed violence.
Thus, an endogenous relationship emerges at the microlevel between and
among welfare outcomes, institutional change, and conflict processes. Defined
as the microlevel foundations of violent conflict, it has been shown to explain in
statistically significantly ways the evolution and duration of violent conflict (see
Justino, Brück, and Verwimp 2013).
How do these effects transpose into our understanding of mass atrocities? All
forms of violent conflict carry largely adverse effects on local populations and
communities. However, it is important to note that violent conflict arises for
(more or less) well-defined reasons and that some population groups may well
benefit from violence, for instance through looting (Keen 1998), redistribution
of assets in favor of certain population groups (Brockett 1990; Wood 2003), and
privileged access to markets and political institutions (Richards 1996). In the
next section, I discuss how these distributional effects of violent conflict are par-
ticularly pronounced in cases of mass atrocity and may, in many circumstances, lie
at the heart of explanations for why they occur in the first place, or may recur in
specific societies.
9.3. Microlevel Causes and Consequences of Mass Atrocities

The microlevel perspective on processes of violent conflict outlined in the pre-
vious section has two important insights to offer to the understanding of the
causes and consequences of mass atrocities. The first has to do with a better grasp
as to why ordinary citizens may be recruited into perpetrating acts of genocide
or other mass atrocities, or why they may be mobilized into supporting groups
that perpetrate such atrocities, while others become bystanders (neither partici-
pating nor condemning atrocities), and still others actively resist and try to pre-
vent the commitment of atrocities (Straus 2006; Anderton 2010). While some
of these issues have been addressed theoretically and empirically at more aggre-
gate levels of analysis (e.g., Collier and Hoeffler 2004), research that focuses on
individuals as the unit of analysis is arguably better equipped to answer questions
about individual motivations and preferences, which may or may not aggregate to
meso- or national-level social outcomes. The second perspective is related to the
specific consequences of mass atrocities on the economic characteristics, rela-
tions, and preferences of people directly affected by mass atrocities, of people that
commit or support them, and the relations between the two (or more) groups.
Again, while a large economics literature has provided useful insights into the
global, national, and regional costs of armed conflict, research at the microlevel
has offered more specific analyses of the mechanisms at play in different contexts
(Verwimp, Justino, and Brück 2009).
9.3.1. Causes: Why Do People Participate in or Support Mass Atrocities?

The microlevel literature on individual motivations for participation in orches-
trated mass violence offers two possible answers. On the one hand, individuals
(or their households and immediate social groups) may have specific motivations
involving opportunities, aspirations, feelings, or attitudes that may explain par-
ticipation and engagement with armed groups (Wood 2003; Humphreys and
Weinstein 2008). An alternative explanation has to do with the social context of
these individuals and how leaders are able to coerce or convince others to join
in collective violence (Petersen 2001; Straus 2006; Fujii 2009; McDoom 2014).
Given that mass atrocities generally are an exercise in violent collective action, it
is worth reflecting in detail on this latter explanation. In this, an important ques-
tion—stemming from the framework outlined in section 9.2—is whether and
how the organizers of mass atrocities may be able to strategically manipulate the
twofold vulnerabilities of ordinary citizens (to poverty and to violence) to help in
either committing or supporting the perpetration of mass atrocities.
The literature offers several explanations for why different individuals or groups
may agree (though many may be coerced) to being recruited or mobilized to support
mass atrocity (Fein 1979, 1993; Kuper 1981; Straus 2006; Waller 2007; Fujii 2009;
Smeulers and Hoex 2010; McDoom 2013, 2014). One dominant set postulates
that individual participation is shaped to a large extent by the promotion among
the architects of the violence of dislike or hatred toward “others”; the entrench-
ment of economic, social, cultural, and political cleavages between groups; and the
enforcement of exclusionary norms and ideologies. A second, related set of expla-
nations has to do with individual or group perception of being a target of violence
by the “hated other,” or the manipulation of that perception by those intending to
commit the atrocities. These factors are used to facilitate recruitment and/or sup-
port because they further the social, economic, and political distances between
groups (Kuper 1981; Harff 2003; Montalvo and Reynal-Querol 2008).
These perceptions and attitudes stem from imbalances in society or from a per-
ceived or fabricated sense of grievances, which become part of a strategy of build-
ing identities of “us” against “them” that characterize many instances of mass
atrocity (Fein 1993; Waller 2007). Typically, these identities are constructed
around forms of group membership such as race, ethnic, and religious affilia-
tion, geographic location, or economic status (McDoom 2014). Economic, social,
or political distances (real or perceived) between specific groups in turn shape
people’s motivations to participate in mass atrocities, either voluntarily or due to
pressures from within the group (Bhavnani 2006; chapter 21 in this volume).
Some literature has linked the occurrence of armed violence to how institu-
tional structures manage different forms of exclusion, inequality, and other social
differences that may shape distances between groups (Boix 2003; Acemoglu and
Robinson 2006). The literature on civil wars, in particular, has shown a strong
association between low levels of GDP, adverse economic shocks, and the out-
break of armed conflict (Fearon and Laitin 2003; Collier and Hoeffler 2004;
Miguel, Satyanath, and Sergenti 2004). Other studies have long identified rapid
uneven economic growth as a destabilizing force that may spur civil unrest and
political violence (Horowitz 1985; Tilly 1990).
One point to take into consideration is that while forms of exclusion and
inequality persist in many countries, no small number of them have experienced
(or are likely to experience) instances of mass atrocities (see c hapter 3 in this vol-
ume). Two factors may be central to understanding the relationship between (real
or perceived) inequality and violence (Justino 2013b). The first is the nature of
inequality processes; and the second is the type of structures and institutions in
place to manage social conflicts that may allow (or not) violence as a strategy to
access power.
9.3.1.1. The Nature of Inequality Processes

Studies have shown a close association between violent conflict and (1) income
and asset inequality (Muller and Seligson 1987; Schock 1996); (2) class divides
(Paige 1975; Scott 1976); (3) inequalities in access to power decisions (Richards
1996); (4) horizontal inequality between ethnic, religious, and other cultural
groups (Stewart 2000, 2002; Langer 2004; Murshed and Gates 2005; Østby
2006); (5) relative deprivation (Gurr 1970); (6) levels of polarization (Esteban
and Ray 1994; Esteban and Schneider 2008; Montalvo and Reynal-Querol
2008); and (7) ethnic fragmentation (Easterly and Levine 1997). While most of
this literature has been concerned with grievances that may lead groups to con-
test the state, a small literature has analyzed the potential role of grievances and
distributional concerns in explaining mass atrocities. Montalvo and Reynal-Querol
(2008), for example, developed a discrete index of ethnic polarization that mea-
sures perceived distances between “us” and “them” and shows that ethnic polar-
ization is a strong explanation for genocide. Esteban, Morelli, and Rohner (2015)
have highlighted the role of natural resource rents in mass killings. Kim (2010)
found that political grievances are associated with an increased probability of
governments committing mass killings in civil wars. Harff and Gurr (1998) and
Harff (2003) have discussed the role of economic differences and exclusionary
ideologies as risk factors for the onset of genocide. Vargas (2011) developed a
theoretical model in which mass killings may result from initial inequalities and
subsequent repression of social discontent or rebellion.
But not all forms of inequality will result in rebellion and/or atrocities, and
some studies have reported a lack of association between measures of inequality
and violent conflict (e.g., Collier and Hoeffler 2004). To date, the debate continues
and much more research is needed to better understand which types of inequali-
ties may (or may not) result in which types of violence. It is, however, clear that
even if not all forms of inequality may be direct causes of violence, inequalities
may be used by political leaders to solidify identities among particular population
groups and establish chasms between “us” and “them” that may help mobilization
to participate in mass atrocities (Straus 2006; Waller 2007; Fujii 2009; McDoom
2013, 2014).
9.3.1.2. Structures and Institutions That Support or Hinder Violence

Although some forms (or levels) of inequality may result in violence, in practice
the onset of mass violence is often triggered by specific mechanisms. The mecha-
nisms that lead to the “actualization [of discontent] in violent action against polit-
ical objects and actors” (Gurr 1970, 12–13) are, however, not well understood,
and it is possible that real or perceived inequalities may not be sufficient to trig-
ger violent conflict on their own. In many instances of mass atrocity, discourses
emphasize potential threats of violence (from the out-group) in terms of “if we do
not kill them, they will kill us” (Waller 2007). Furthermore, the kill or be killed
dynamic can also operate within a group. For instance, Bhavnani (2006) analyzes
the emergence of a violence-promoting norm within the Hutu population in the
context of the Rwandan genocide that shaped the behavior of its members toward
their perceived rivals and led even reluctant Hutu to participate in the genocide,
“because they were left with no other choice: ‘kill or be killed’ ” (652) (see also
Smeulers and Hoex 2010).
These discourses may or may not be based on real underlying divergences
in the distribution of power, income, and resources. They are, however, often
used by leaders of fighting units as a deliberate strategy to gain supporters and
win wars (Valentino, Huth, and Balch-Lindsay 2004; Kalyvas 2006). This may
also explain why instances of mass atrocity may take place in contexts of civil
wars. For instance, Harff (2003) shows that most genocides since World War
II occurred during, or in the immediate aftermath of, internal conflicts, while
Valentino, Huth, and Balch-Lindsay (2004) discuss evidence demonstrating that
mass atrocities are more likely to take place in guerrilla wars than during other
types of internal war. They show, in addition, that mass atrocities are associated
with increased levels of population support for the rebel group and are more likely
when the group poses a serious military threat to an incumbent regime. Eck and
Hultman (2007) and Wood (2010) show that violence against civilians also takes
place among rebel groups, particularly those that are weaker and unable to resort
to other, nonviolent strategies to gain and maintain population support (Wood
2010) or face difficult economic conditions (Verwimp 2003). Mass atrocities
have been observed as calculated military strategies in contexts of civil wars in
Algeria, Cambodia, El Salvador, Ethiopia, Uganda, and Rwanda, among others
(Kalyvas 1999, 2006; Harff 2003; Valentino 2004; Valentino, Huth, and Balch-
Lindsay 2004; Kim 2010).
Whether or not inequalities result in mass atrocities is also determined by
forms of institutional change that accompany these forms of violence (Justino
2013b). In some situations, it is possible that the institutional changes that
take place during conf lict may result in positive social, economic, and politi-
cal outcomes—such as the mechanisms set in place to resolve social conf lict
(electoral and judicial systems), property rights arrangements, systems for
resource distribution (taxation, budgeting, and revenue allocation), mar-
kets, collective organizations, or systems for the provision of public goods
and services (see Justino 2013a for a review). Although such outcomes are
observed in some civil wars (Arjona 2009; Mampilly 2011), it is unlikely that
mass atrocities will result in many positive institutional outcomes given that
the breakdown of the social and political status quo is at the heart of the vio-
lence in the first place. In these contexts, violence is likely to be used strategi-
cally by political actors to transform society in ways that benefit only certain
population groups. The resulting distributional changes will, in turn, have
profound effects on the survival and security of people that were not killed
outright (either victims or perpetrators) due to the legacies of the violence
itself, as well as the probability that in the future violence may target them
(see c hapter 7 in this volume).
The economic effects of fear, distrust, and the breakdown of social relations
between groups in contexts of genocide and mass atrocity remain, however,
largely underresearched. The most likely scenario is that such violence gives
rise to “extractive institutions,” in the words of Acemoglu and Robinson (2012),
which may eventually collapse under the effects of events that may trigger further
violence. This is particularly likely if processes of legitimacy and accountability
are weak and authoritarian regimes and systems of patronage prevail. In support
of this argument, a number of studies have found a nonlinear association between
democracy and mass atrocities (Rummel 1995, 1997; Harff 2003; Easterly, Gatti,
and Kurlat 2006). It is less clear, however, what type of subnational institutional
processes may support or hinder mass violence against civilians.
Amid the destruction, suffering, and trauma caused by mass atrocities, it is
important to remember that, at some level, violence takes place for a reason.
Sometimes actors have used violence—including genocidal violence—as a
means to try to improve their own position and to take advantage of oppor-
tunities offered by the conflict. In particular, mass killings have been used
throughout history to open opportunities for new groups to challenge incum-
bent political power and address existing disparities in the distribution of
power, resources, and wealth (Keen 1998; Reno 2002; Cramer 2006; McDoom
2013, 2014). These changes are likely to profoundly affect the lives of individu-
als and households, the organization of communities, and hence how societies
transition from violence to cohesion. Norms and organizations that favor cor-
rupt, rent-seeking, and predatory behavior will perpetuate dysfunctional eco-
nomic, social, and political structures. Norms and organizations that protect
rights, enforce acceptable and just norms of conduct, and impose sanctions for
unwanted behavior may create the conditions necessary for the establishment
of more inclusive societies (Justino 2013b). Disentangling these complex nor-
mative and institutional effects will be, one hopes, the focus of new microlevel
empirical research on the economic causes and consequences of genocides and
mass atrocities in the future, and of armed conflict more generally.
9.3.2. Consequences: What Are the Microeconomic

Effects of Mass Atrocities?
At the individual level, it is unlikely that mass atrocities will have welfare effects
that are substantially dissimilar to those caused by other forms of violence, other
than a larger-magnitude effect due to specific targeting of civilians during mass
atrocities. The main differences may be more visible at the level of groups and
communities. Section 9.2 highlighted some of the effects of armed violence in
general in terms of changes in household composition, population movements,
and asset and human capital accumulation. Some of the findings reviewed
in that section originate from contexts characterized by genocides and mass
atrocities such as Bosnia (Kondylis 2005, 2007), Cambodia (de Walque 2006),
the Holocaust (Akbulut-Yuksel 2014), and Rwanda (Friedman 2013; Justino and
Verwimp 2013; Verwimp 2005, 2003).
However, because mass atrocities so clearly pitch major groups in society
against each other, it is reasonable to postulate that the distributional conse-
quences of these forms of violence may be much more significant than those of
other violent processes. In particular, it is realistic to expect that mass atrocities
may result in the widespread exclusion of certain groups (if not massacred) from
social, economic, and political opportunities (Acemoglu, Hassan, and Robinson
2011). The large body of literature discussed previously has examined the effects of
inequalities and other economic factors on the onset of civil conflict. In contrast,
only a few studies have examined the effect of armed violence on distributional
arrangements in societies affected by armed violence.2 This is particularly impor-
tant because mass atrocities—including those that take place within contexts of
civil war—are targeted to specific groups due to their ethnic or religious affilia-
tion, their geographical location, their wealth, or other characteristics salient to
local or national conflict cleavages. The expectation of obtaining or accessing cer-
tain distributional advantages may therefore be the reason (or the main reason)
why these forms of violence happen in the first place.
It is accepted that violent conflict may result in new forms of social arrange-
ments and political structures that are bound to benefit some groups to the detri-
ment of others (Justino 2013a, 2013b; chapter 8 in this volume). For instance,
Justino and Verwimp (2013) find that the Rwandan genocide resulted in large dis-
tributional effects because better-off provinces and households were particularly
targeted during the violent events. These changes in the distribution of economic
resources, and the potential association with new forms of distribution of social
and political power in postviolence periods, may lead to recurring outbreaks of
violence. But I postulate that mass atrocities—and genocides in particular—are
likely to result in profound, direct distributional effects more generally than has
hitherto been realized, limited to specific case studies as the literature is. Thus,
in many contexts of violent conflict, including those where mass atrocities took
place, winners have been known to restrict access to education for the losers by
limiting enrollments in some levels of education and/or by segregating schools
along racial (South Africa), ethnic (pre-1994 Rwanda), or religious (Northern
Ireland) lines (Bush and Saltarelli 2000; Shemyakina 2011). In addition, low lev-
els of economic growth combined with weak sociopolitical institutions and spe-
cific political agendas may reinforce existing inequalities or produce new forms of
inequality. This may, in turn, fuel further resentment and generate tensions across
population groups, creating a cycle of impoverishment, violence, and instability
from which many countries cannot recover fully.
Another important distributional effect of mass atrocities and, again, espe-
cially for larger-scale atrocities such as genocides, has to do with changes to the
social fabric of affected communities, including social relations between family

members, neighbors, and friends; how differently affected communities relate to
each other; and on the functioning of local citizen organizations and their relation
with state-level institutions (Justino 2013b; McDoom 2013, 2014). This impact
is likely to substantially affect the distribution of power and wealth within local
communities because it will affect the ability of local people—depending on their
group affiliation—to rely on social relations and networks in times of difficulty,
to access employment or credit markets, and to integrate into local society and
norms (Justino 2012). As mentioned, to date these likely effects are underre-
searched and will constitute an important research agenda in the future.
9.4. Conclusion: A New Research Agenda

A new body of literature on the microlevel causes and consequences of violent
conflict has led to new insights about conflict processes, one that may complement
extant research by the community of genocide scholars. However, much is still
left to do. In particular, this literature has not yet provided a good handle on how
different forms of violent conflict—such as genocides and mass atrocities—may
be analyzed from the viewpoint of individuals, households, and local communi-
ties (i.e., the microlevel). Questions like “who participates in genocides and mass
atrocities, and why?” or “what are the consequences of genocides and mass atroci-
ties for individuals, households, and local communities?” have remained underre-
searched. This chapter provides a conceptual entry-point to think through some
of these questions and, I hope, motivates further empirical analysis in the future.
Four main lines of future enquiry arise from the chapter. The first is related
to the distributional foundations of genocides and mass atrocities. These particular
forms of violence are both the result of distributional arrangements within soci-
eties that may be perceived as unfair by some groups and also are likely to lead
to important changes in the distribution of economic resources and social and
political power when winning factions restrict access to other population groups.
A significant body of literature has analyzed the impact of inequalities on the onset
and duration of violent conflict, although the results for contexts affected by
genocides and mass atrocities are limited. This offers an important path to new
research, particularly in terms of understanding what inequalities may trigger
genocides and mass atrocities, and under what circumstances. But better knowl-
edge about the distributional consequences of genocides and mass atrocities—and
other forms of violence—may provide additional insights into root causes of these
atrocities and offer new entry-points to possible policy interventions to minimize
the occurrence of genocides and mass atrocities in societies across the world.
A second area for further research will be to think through the macrolevel conse-
quences of the distributional foundations of mass atrocities. This chapter has discussed
how processes of mass atrocities may be affected by the behavior, choices, and aspi-
rations of individuals, households, and groups. These local dynamics may, in turn,
have important consequences for how these types of violence aggregate to the coun-
try level. Understanding the microlevel distributional dimensions of mass atrocities
in particular may provide important clues as to what risk factors may explain why
some socially divided societies experience mass atrocities, while in others different
groups may live alongside each other without violence breaking out.
A third line of enquiry has to do with the institutional setup of societies at risk
of mass atrocities. One suggestion made in section 9.3 was that violence may be
mitigated—even in countries already afflicted by violent conflict—by norms and
organizations that may lay the seeds for stability, trust, and inclusion. We have,
however, only started to open up the black box of institutional transformation
in areas of conflict and violence (see chapter 28 in this volume). The processes
whereby institutional frameworks shape the likelihood of the use of violence—
including genocides and mass atrocities—or the types of institutional change
promoted by different processes of armed violence are yet to be understood in the
literature. This is an extremely important area of future research, and an impor-
tant challenge to us all as these factors lie at the center of how we understand
processes of transition from conflict-ridden societies to inclusive, legitimate, and
accountable states, and how we minimize the risk of horrific atrocities being com-
mitted against human beings living in the same societies.
Fourth, there is a need to further disaggregate violence-related data at the micro-
level in order to provide better understandings of types of violence (including
genocides and mass atrocities); how households and communities perceive and
are affected by interdependencies between different types of armed violence, the
actors involved, and their motivations; as well as allowing researchers to match
disaggregated data on mass violence events to detailed socioeconomic informa-
tion on how people live under conditions of violence. This is a challenging but not
impossible task given the ongoing progress in event-data collection3 and in ana-
lytical methods to conduct research on violent conflict at the microlevel.4 I expect
that the distribution-focused framework proposed in this chapter can provide a
point of reference for further empirical work on the microlevel causes and conse-
quences of genocides and mass atrocities.
Notes
1. See, for instance, Akresh, Verwimp, and Bundervoet (2011), Alderman, Hoddinott, and
Kinsey (2006), Akresh and de Walque (2008), Rodriguez and Sanchez (2012), Camacho
(2008), Chamarbagwala and Morán (2008), and Shemyakina (2011) for Tajikistan, and
Swee (2009) for Bosnia, among others. This literature is reviewed in Justino (2012).
2. Exceptions are McKay and Loveridge (2005) and Justino and Verwimp (2013) at the
microlevel.
3. The Armed Conflict Location and Event Data (ACLED) Project datasets have made impor-
tant strides in this direction as they contain information on the specific dates and locations
of political violence, the types of events that have taken place, the groups involved, the
number of fatalities, and changes in territorial control. The datasets record, in addition,
information on battles, killings, riots, and recruitment activities of rebels, governments,
militias, armed groups, protesters, and civilians. See http://www.acleddata.com/.
4. See, for instance, the working paper series of the Households in Conflict Network: www.
hicn.org. Also see chapter 23 in this volume on data mining and machine learning
techniques.
References
Acemoglu, D., and J. Robinson. 2006. Economic Origins of Dictatorship and Democracy.
Acemoglu, D., and J. Robinson. 2012. Why Nations Fail: The Origins of Power, Prosperity and
Poverty. London: Profile.
Acemoglu, D., T. Hassan, and J. Robinson. 2011. “Social Structure and Development: A Legacy
of the Holocaust in Russia.” Quarterly Journal of Economics 126, no. 2: 895–946.
Akresh, R., and D. de Walque. 2008. “Armed Conflict and Schooling: Evidence from the 1994
Rwandan Genocide.” HiCN Working Paper No. 47. The Households in Conflict Network.
http://www.hicn.org/wordpress/wp-content/uploads/2012/06/wp47.pdf.
Akresh, R., P. Verwimp, and T. Bundervoet. 2011. “Civil War, Crop Failure and Child Stunting
in Rwanda.” Economic Development and Cultural Change 59, no. 4: 777–810.
Akbulut-Yuksel, M. 2014. “Children of War: The Long-Run Effects of Large-Scale Physical
Destruction and Warfare on Children.” Journal of Human Resources 49, no. 3: 634–62.
Alderman, H., J. Hoddinott, and B. Kinsey. 2006. “Long Term Consequences of Early Childhood
Malnutrition.” Oxford Economic Papers 58, no. 3: 450–74.
Arjona, A. 2009. “Variants of Social Order in Civil War.” PhD dissertation. Department of
Political Science. New Haven, CT: Yale University Press.
Autesserre, S. 2010. The Trouble with the Congo: Local Violence and the Failure of International
Peacebuilding. Cambridge, MA: Cambridge University Press.
Bhavnani, R. 2006. “Ethnic Norms and Interethnic Violence: Accounting for Mass Participation
in the Rwandan Genocide.” Journal of Peace Research 43, no. 6: 651–69.
Blattman, C., and J. Annan. 2010. “The Consequences of Child Soldiering.” Review of Economics
and Statistics 92, no. 4: 882–98.
Boix, C. 2003. Democracy and Redistribution. New York: Cambridge University Press.
Brockett, C. D. 1990. Land, Power, and Poverty: Agrarian Transformation and Political Conflict in
Central America. Boston: Unwin Hyman.
Bush, K. D., and D. Saltarelli, eds. 2000. Two Faces of Education in the Ethnic Conflict: Towards a
Peacebuilding Education for Children. Florence, Italy: Innocenti Research Center, UNICEF.
Camacho, A. 2008. “Stress and Birth Outcomes: Evidence from Terrorist Attacks in Colombia.”
Chamarbagwala, R., and H. E. Morán. 2008. “The Human Capital Consequences of Civil
War: Evidence from Guatemala.” HiCN Working Paper No. 59. The Households in
Conflict Network.
Charny, I. 1999. Encyclopedia of Genocide. 2 vols. Santa Barbara, CA: ABC-CLIO.
Collier, P., and A. Hoeffler. 1998. “On Economic Causes of Civil War.” Oxford Economic Papers
50, no. 4: 563–73.
Collier, P., and A. Hoeffler. 2004. “Greed and Grievance in Civil War.” Oxford Economic Papers
56, no. 4: 563–95.
Cramer, C. 2006. Civil War Is Not a Stupid Thing: Accounting for Violence in Developing Countries.
London: Hurst.
de Walque, D. 2006. “The Socio-Demographic Legacy of the Khmer Rouge Period in Cambodia.”
Population Studies 60, no. 2: 223–31.
Easterly, W., and R. Levine. 1997. “Africa’s Growth Tragedy: Policies and Ethnic Divisions.”
Quarterly Journal of Economics 112, no. 4: 1202–50.
New Fatality Data.” Journal of Peace Research 44, no. 2: 233–46.
Esteban, J., and D. Ray. 1994. “On the Measurement of Polarization.” Econometrica 62, no.
4: 819–52.
Esteban, J., and G. Schneider. 2008. “Polarization and Conflict: Theoretical and Empirical
Issues.” Journal of Peace Research 45, no. 2: 131–41.
Economy 123, no. 5: 1087–1132.
Fearon, J., and D. Laitin. 2003. “Ethnicity, Insurgency and Civil War.” American Political Science
Review 97, no. 1: 75–90.
Fein, H. 1979. Accounting for Genocide: National Responses and Jewish Victimization during the
Holocaust. New York: Free Press.
Fein, H. 1993. Genocide: A Sociological Perspective. London: Sage.
Friedman, W. 2013. “Local Economic Conditions and Participation in the Rwandan Genocide.”
HiCN Working Paper No. 160. The Households in Conflict Network.
Fujii, L. A. 2009. Killing Neighbors: Webs of Violence in Rwanda. Ithaca, NY: Cornell University
Press.
Goodwin, J. 2001. No Way Out: States and Revolutionary Movements 1945–1991.
Grossman, H. I. 2002. “Make Us a King: Anarchy, Predation, and the State.” European Journal of
Gurr, T. R. 1970. Why Men Rebel. Princeton, NJ: Princeton University Press.
Harff, B., and T. R. Gurr. 1998. “Systematic Early Warning of Humanitarian Emergencies.”
Hirshleifer, J. 2001. The Dark Side of the Force: Economic Foundations of Conflict Theory.
Horowitz, D. 1985. Ethnic Groups in Conflict. Berkeley: University of California Press.
Humphreys, M., and J. M. Weinstein. 2008. “Who Fights? The Determinants of Participation in
Civil War.” American Journal of Political Science 52, no. 2: 436–55.
Ichino, A., and R. Winter-Ebmer. 2004. “The Long-Run Educational Cost of World War II.”
Journal of Labor Economics 22: 57–86.
Justino, P. 2009. “Poverty and Violent Conflict: A Micro-Level Perspective on the Causes and
Duration of Warfare.” Journal of Peace Research 46, no. 3: 315–33.
Justino, P. 2012. “War and Poverty.” In M. Garfinkel and S. Skaperdas, eds., Handbook of the
Economics of Peace and Conflict. New York: Oxford University Press, 676–705.
Justino, P. 2013a. “Research and Policy Implications from a Micro-Level Perspective on
the Dynamics of Conflict, Violence and Development.” In P. Justino, T. Brück, and P.
Verwimp, eds., A Micro-Level Perspective on the Dynamics of Conflict, Violence and Development.
Oxford: Oxford University Press, 290–306.
Justino, P. 2013b. “Shared Societies and Armed Conflict: Costs, Inequality and the Benefits of
Peace.” HiCN Working Paper No. 125. The Households in Conflict Network. http://www.
hicn.org/wordpress/wp-content/uploads/2012/06/HiCN-WP-125.pdf.
Justino, P., and P. Verwimp. 2013. “Poverty Dynamics, Violent Conflict and Convergence in
Rwanda.” Review of Income and Wealth 59, no. 1: 66–90.
Justino, P., T. Brück, and P. Verwimp, eds. 2013. A Micro-Level Perspective on the Dynamics of
Conflict, Violence and Development. Oxford: Oxford University Press.
Justino, P., M. Leone, and P. Salardi. 2014. “Short and Long-Term Impact of Violence on
Education: The Case of Timor Leste.” World Bank Economic Review 28, no. 2: 320–53.
Society 11, no. 3: 243–85.
Kalyvas, S. N. 2006. The Logic of Violence in Civil Wars. New York: Cambridge University Press.
Kalyvas, S. N., and M. Kocher. 2007. “How ‘Free’ is Free-Riding in Civil Wars?” World Politics
59, no. 2: 177–216.
Kim, D. 2010. “What Makes State Leaders Brutal? Examining Grievances and Mass Killing
During Civil War.” Civil Wars 12, no. 3: 237–60.
Keen, D. 1998. “The Economic Functions of Violence in Civil Wars.” Adelphi Paper 320.
London: International Institute of Strategic Studies.
Keen, D. 2005. Conflict and Collusion in Sierra Leone. London: James Currey.
Kondylis, F. 2005. “Agricultural Production and Conflict Refugee Status: Quasi-Experimental
Evidence from a Policy Intervention Programme in Rwanda.” Unpublished manuscript.
Economics Department. Royal Holloway, University of London.
Kondylis, F. 2007. “Agricultural Outputs and Conflict Displacement: Evidence from a Policy
Intervention in Rwanda.” HiCN Working Paper No. 28. The Households in Conflict
Network.
Kuper, L. 1981. Genocide: Its Political Use in the Twentieth Century. New Haven, CT: Yale University
Press.
Langer, A. 2004. “Horizontal Inequalities and Violent Conflict: The Case of Cote D’Ivoire.”
CRISE Working Paper No. 13. Centre for Research on Inequality, Human Security and
Ethnicity. Queen Elizabeth House, University of Oxford.
Lemkin, R. 1944 [2008]. Axis Rule in Occupied Europe: Laws of Occupation, Analysis of Government,
Proposals for Redress. Clark, NJ: The Lawbook Exchange.
Lichbach, M. 1995. The Rebel’s Dilemma. Ann Arbor: University of Michigan Press.
Mampilly, Z. 2011. Rebel Rulers: Insurgent Governance and Civilian Life during War. Ithaca,
McDoom, O. 2013. “Who Killed in Rwanda’s Genocide? Micro-Space, Social Influence and
Individual Participation in Intergroup Violence.” Journal of Peace Research 50, no. 4: 453–67.
McDoom, O. 2014. “Antisocial Capital: A Profile of Rwanda Genocide Perpetrators’ Social
Networks.” Journal of Conflict Resolution 58, no. 5: 866–94.
McKay, A., and S. Loveridge. 2005. “Exploring the Paradox of Rwandan Agricultural Household
Income and Nutritional Outcomes in 1990 and 2000.” Staff Paper 2005-06. Department of
Agricultural Economics. Lansing: Michigan State University. http://ageconsearch.umn.
edu/bitstream/11582/1/sp05-06.pdf.
Merrouche, O. 2006. “The Human Capital Cost of Landmine Contamination in Cambodia.”
Miguel, E., S. Satyanath, and E. Sergenti. 2004. “Economic Shocks and Civil Conflict: An
Instrumental Variables Approach.” Journal of Political Economy 112, no. 4: 725–53.
Montalvo, J., and M. Reynal-Querol. 2008. “Discrete Polarisation with an Application to the
Muller, E., and M. Seligson. 1987. “Inequality and Insurgency.” American Political Science Review
81, no. 2: 425–51.
Murshed, M., and S. Gates. 2005. “Spatial-Horizontal Inequality and the Maoist Insurgency in
Nepal.” Review of Development Economics 9, no. 1: 121–34.
Olson, M. 1965. The Logic of Collective Action: Public Goods and the Theory of Groups. Cambridge,
MA: Harvard University Press.
Østby, G. 2006. “Horizontal Inequalities, Political Environment and Civil Conflict: Evidence
from 55 Developing Countries.” CRISE Working Paper 28. University of Oxford. http://
www3.qeh.ox.ac.uk/pdf/crisewps/workingpaper28.pdf.
Paige, J. 1975. Agrarian Revolutions. New York: Free Press.
Petersen, R. 2001. Resistance and Rebellion: Lessons from Eastern Europe. New York: Cambridge
University Press.
Pinchotti, S., and P. Verwimp. 2007. “Social Capital and the Rwandan Genocide: A Micro-Level
Analysis.” HiCN Working Paper No. 30. The Households in Conflict Network.
Reno, W. 2002. “The Politics of Insurgency in Collapsing States.” Development and Change 33,
no. 5: 837–58.
Richards, P. 1996. Fighting for the Rainforest: War, Youth and Resources in Sierra Leone. London: James
Currey.
Rodriguez, C., and F. Sanchez. 2012. “Armed Conflict Exposure, Human Capital Investments
and Child Labor: Evidence from Colombia.” Defence and Peace Economics 23, no. 2: 161–84.
Rummel, R. J. 1995. “Democracy, Power, Genocide, and Mass Murder.” Journal of Conflict
Resolution 39, no. 1: 3–26.
Rummel, R. J. 1997. Power Kills: Democracy as a Method of Nonviolence. New Brunswick,
NJ: Transaction.
Schock, K. 1996. “A Conjuctural Model of Political Conflict: The Impact of Political
Opportunities on the Relationship between Economic Inequality and Violent Political
Conflict.” Journal of Conflict Resolution 40, no. 1: 98–133.
Scott, J. 1976. The Moral Economy of the Peasant: Rebellion and Subsistence in Southeast Asia. New
Haven, CT: Yale University Press.
Shemyakina, O. 2011. “The Effect of Armed Conflict on Accumulation of Schooling: Results
from Tajikistan.” Journal of Development Economics 95, no. 2: 186–200.
Smeulers, A., and L. Hoex. 2010. “Studying the Microdynamics of the Rwandan Genocide.”
British Journal of Criminology 50, no. 3: 435–54.
Stewart, F. 2000. “Crisis Prevention: Tackling Horizontal Inequalities.” Oxford Development
Studies 28, no. 3: 245–62.
Stewart, F. 2002. “Horizontal Inequalities: A Neglected Dimension of Development.” Queen
Elisabeth House. Working Paper Series No. 81. Oxford: University of Oxford.
Straus, S. 2006. The Order of the Genocide: Race, Power and War in Rwanda. Ithaca, NY: Cornell
University Press.
Swee, E. L. 2009. “On War and Schooling Attainment: The Case of Bosnia and Herzegovina.”
Tilly, C. 1990. Coercion, Capital, and European States, AD 990–1990. Oxford: Blackwell.
Totten, S., and W. S. Parsons, eds. 2013. Centuries of Genocide: Essays and Eyewitness Accounts. 4th
ed. London: Routledge.
Genocide.” https://treaties.un.org/doc/Publication/UNTS/Volume%2078/volume-78-I-
Valentino, B. 2004. Final Solutions: Mass Killing and Genocide in the 20th Century. Ithaca,
Valentino, B., P. Huth, and D. Balch-Lindsay. 2004. “‘Draining the Sea’: Mass Killings and
Guerrilla Warfare.” International Organization 58, no. 2: 375–407.
Vargas, J. F. 2011. “Rebellion, Repression and Welfare.” Defence and Peace Economics 22, no.
5: 563–79.
Verwimp, P. 2003. “The Political Economy of Coffee, Dictatorship, and Genocide.” European
Journal of Political Economy 19, no. 2: 161–81.
Verwimp, P. 2005. “An Economic Profile of Peasant Perpetrators of Genocide: Micro-Level
Evidence from Rwanda.” Journal of Development Economics 77, no. 2: 297–323.
Verwimp, P., P. Justino, and T. Brück. 2009. “The Analysis of Conflict: A Micro-Level
Perspective.” Journal of Peace Research 46, no. 3: 307–14.
Waller, J. 2007. Becoming Evil: How Ordinary People Commit Genocide and Mass Killings. 2nd ed.
Walter, B. 2004. “Does Conflict Beget Conflict? Explaining Recurring Civil War.” Journal of
Peace Research 41, no. 3: 371–88.
Wickham-Crowley, T. 1992. Guerrillas and Revolution in Latin America: A Comparative Study of
Insurgents and Regimes since 1956. Princeton, NJ: Princeton University Press.
Wood, E. J. 2003. Insurgent Collective Action and Civil War in El Salvador. New York: Cambridge
University Press.
Wood, R. 2010. “Rebel Capability and Strategic Violence against Civilians.” Journal of Peace
Research 47, no. 5: 601–14.
10
Development and the Risk of Mass

Atrocities
An Assessment of the Empirical Literature
A n k e Hoe f f l e r
10.1. Introduction
Mass atrocities have been studied from many different perspectives. Within
the disciplines of psychology, sociology, anthropology, and political science
a large number of case studies, comparative studies, and large-n statistical
analyses have been undertaken. In contrast, economists have conducted few
studies of mass atrocities. Many large-n studies of genocides and mass killings
consider economic explanations, but there is little robust evidence that the
level of economic development measured for the entire country is associated
with the risk of mass atrocities (see c hapter 24 in this volume). More recently
researchers have analyzed the connection between resources and strategies in
civil wars and the implication of this link for violence against civilians in wars
(see c hapter 19 in this volume). There is some evidence suggesting that armed
groups that have access to outside finance are less reliant on support by the
civilian population. In these cases the civilian population is more at risk of
mass atrocities.
This chapter is structured in the following way. Section 10.2 provides gen-
eral background that informs the statistical modeling of mass atrocities. This
is followed in section 10.3 by a very detailed, in-depth presentation and discus-
sion of the most common findings in large-n studies of mass atrocities. Civil
war is one of the most robust predictors of mass atrocities, followed by a num-
ber of democracy-related and institutional variables. Evidence for the influ-
ence of economic variables on the occurrence and conduct of mass atrocities,
such as income per capita, inequality, and trade, is more mixed. Section 10.4
concludes.
230
Development and the R isk of Mass Atrocities 231
10.2. Background
While there are numerous large-n studies on the correlates of civil war, there
is a much smaller body of literature on mass atrocities. Thus, unlike the study
of civil war, the study of mass atrocities has not attracted a lot of attention in
the discipline of economics. Very few economists have studied the phenom-
enon. Notable exceptions include Easterly, Gatti, and Kurlat (2006) and
Montalvo and Reynal-Querol (2008). The aim of this review is to provide an
assessment of the empirical literature on economic development and the risk
of mass atrocities, and for this purpose I define development broadly as a pro-
cess that improves individual and social welfare. The measurement of the level
of economic development is controversial; often economists focus solely on
income measures. One popular alternative measure is provided by the Human
Development Index (HDI), which is the average of three key dimensions of
development: health, education, and income. Extensions of this measure con-
sider inequality, poverty, human security, and empowerment.1 While mass
atrocities have a negative impact on economic development (c hapter 5 in
this volume), it is less clear how economic development affects the risk of mass
atrocities. Easterly, Gatti, and Kurlat (2006) suggest that the a priori relation-
ship between economic development and mass atrocities is ambiguous. On
the one hand, higher standards of living make killings more costly, and more
developed societies may have designed safeguarding measures to avoid the
cost of violence. In addition, a more educated society may be more tolerant
toward other groups. On the other hand, more development leads to advances
in technology and social organization that lower the cost of violence, thus
making mass atrocities potentially more likely.
Next, I point to the discussion and definition of mass atrocities in chapters 1
through 3 in this volume, where the term mass atrocity is taken as an umbrella
term that encompasses genocides, politicides, and other types of mass violence
against noncombatants. To recap, genocides and politicides are the intentional
destruction, in whole or in part, of a specific group of people. For other mass kill-
ings, perpetrators either do not intend to eliminate the group as such or victims
are not limited to one or more specific groups. The perpetrators of such atrocities
can be the state or nonstate actors.
Since there is no established literature on the economics of mass atrocities
per se, I review the existing large-n studies, mainly conducted by political scien-
tists. Many of these studies include proxies for income, wealth, and resources and
examine the relative importance and interaction between economic and strategic
reasons for mass atrocities. Before turning to the discussion of the results from
these large-n studies, this section provides some background as to why there has
been relatively little interest in the economic explanation of mass atrocities.
In contrast to economists, researchers from other disciplines, for example

psychologists, historians, sociologists, anthropologists, and political scientists,
have a long tradition of studying violence in its many different forms. One part of
this literature has a strong focus on genocide (e.g., Staub 1989; Goldhagen 1996;
Valentino 2004; Waller 2007; Baum 2008). Genocide appears to be a category
of violence that seems particularly difficult to understand and analyze. Pinker
(2011, 386) states that “of all the varieties of violence of which our sorry species
is capable, genocide stands apart, not only as the most heinous but as the hard-
est to comprehend … killing-by-category targets people for what they are rather
than what they do and thus seems to flout the usual motives of gain, fear, and
vengeance” (italics in original).
When genocide occurs independent of war, the phenomenon seems particu-
larly difficult to comprehend. In peacetime, it is difficult to think of other reasons
for collective actors to kill a large number of individuals other than to destroy
the group (Shaw 2007). This may be the reason why many of the psychological
studies concentrate on explaining genocide as a social or collective action. The
development of an extreme leadership and ideology, and recruitment of individu-
als to “the cause,” are key aspects in the understanding of genocide. However,
as c hapter 3 shows, the overwhelming majority of genocides occur during civil
war. During these wars combatants on both sides rely on civilian support. As
discussed in more detail later on, violence against civilians in civil wars can be
used for different purposes: to destroy support for the opposing forces, to coerce
civilian cooperation, to finance the war, to provide short-term benefits for com-
batants, and to settle old scores. Often the motivation does not appear to be the
destruction of a group of people, but ex post these atrocities can be interpreted as
such. In particular, because political violence is often organized along ethnic lines
and many locations are dominated by one group, mass killings as a strategic and
economic instrument are then interpreted as “genocide” even though there may
not have been any prior intent. Thus, I argue that in many cases civilians are not
killed because of who they are, but because of strategic and economic reasons. To
summarize, since it is difficult to distinguish among the categories of genocide,
politicide, and mass killings, it appears to be sensible to use a broad definition of
mass atrocities that includes all of these categories. Furthermore, if the killing
of civilians is motivated, in part, by strategic and economic reasons, this can in
principle be tested in statistical models.
This view—that one can formulate and test distinct hypotheses regarding
mass atrocities—is contradicted by an understanding of mass atrocities as a
Hobbesian scenario of “war of all against all,” for if people turn against and kill
each other just because of who they are, we should not be able to observe any
empirical patterns at all for this type of violence other than group membership.
However, there is a wealth of evidence suggesting that mass atrocities are not in
fact a “war of all”: violence is not perpetrated by the majority of ordinary people.
Studies of mass atrocities in Bosnia and Rwanda (Mueller 2007) find that in both
countries leaders found it difficult to use regular armed forces to kill civilians.
Instead they recruited thugs (for example, by releasing criminals from prisons)
who formed small, predatory bands that carried out “ethnic cleansing” and geno-
cide. The analysis by Rogall (2014) suggests that the majority of the atrocities
committed during the Rwandan genocide were due to a relatively small number
of killers: about 10 percent of the total number of perpetrators caused 83 percent
of the deaths. Other microlevel evidence also suggests that the killing of civil-
ians can be explained by factors other than group membership. Kalyvas (2006)
focuses on the local dynamics of violence against civilians. His theoretical work
and case study evidence on Greece suggests that violence can be driven by com-
batants’ desire to obtain information about the opposing side and by noncomba-
tants’ desire to gain security.
Weinstein (2007) focuses on violence against civilians perpetrated by rebel
movements. He distinguishes two types of rebellion. In activist rebellions,
rebels can commit to internal discipline by drawing on established norms
and networks. Power within the rebel army can be decentralized without vio-
lence against civilians. In opportunistic rebellions, the rebels cannot credibly
commit to nonabusive behavior and without local ties it is difficult to identify
defectors. Rebels work on short-term rewards that they receive in terms of loot
and violence. However, many activist rebellions turn, over time, into opportu-
nistic rebellions as more fighters with less strong convictions join. Weinstein
argues that violence against civilians in these cases can be the result of material
strength, not rebel weakness. In situations of material strength, due to foreign
sponsorship or income from natural resources, rebel forces have to rely less on
local support. Further microlevel evidence on why civilians are killed is pro-
vided by the case study of northwestern Rwanda by André and Platteau (1998).
They carried out a detailed survey before the genocide, and survivors were
traced afterward. The pre-1994 survey identified only one person as Tutsi. In
the genocide about 5 percent of the local population was killed, that is, not only
the one person identified as Tutsi but also a number of Hutu. The postgenocide
survey reveals that the victimization of the population was not random. Many
victims had previously been identified as “troublemakers,” people who “aroused
jealousy and hatred” and owned more land. Thus, the wave of violence offered
opportunities to settle old scores and to seize land. To summarize, there is case
study evidence to suggest that violence against civilians in civil war is not inex-
plicable and irrational. States and rebels use violence against civilians as a tac-
tic in civil wars. For example, terror against civilians drives people away, thus
restricting civilian support for the opposing side. In addition, fighters may be
incentivized with short-term rewards. Furthermore, the havoc created in situa-
tions of mass atrocity opens a window of opportunity to settle old scores and to
misappropriate property.
Recently collected datasets offer information on violence against civilians

without relying on any assumptions of intent, that is, they are not restricted
to genocides and politicides. These data are detailed and allow researchers
to analyze violence by state and nonstate groups and to specify the location
of the violence (see, e.g., Eck and Hultman 2005; Raleigh 2012). Since mass
atrocities strongly overlap with civil war, similar approaches to the study of
civil war are applied in the study of mass atrocities. However, potential eco-
nomic determinants are not as systematically studied and as the following
section shows, many of the variables that are robustly correlated with the risk
of civil war are not found to be correlated with the risk or the magnitude of
mass atrocities.
10.3. Large-n Studies of Mass Atrocities: An Overview

This section highlights some of the findings obtained in large-n studies of mass
atrocities. The main focus is on models using cross-country regressions. (The
burgeoning microlevel evidence is reviewed in c hapter 9, and the forecasting of
mass atrocities is discussed in chapter 24, respectively, in this volume. Microlevel
evidence specifically in regard to refugees is reviewed in chapter 11.) Table 10.1
summarizes the findings from some of the key studies.
In general, very few large-n studies consider problems of autocorrelation,
endogeneity, and simultaneity. For example, often there are cycles of violence—
previous mass atrocities that increase the risk of experiencing another mass atroc-
ity. Atrocities depress income, and the resulting economic stress could then, in
turn, increase the risk of another mass atrocity. Throughout this section I there-
fore do not refer to causes of mass atrocities, but to correlates of mass atrocities. In
contrast to the large-n literature, some of the case study literature has applied the
method of instrumental variables in order to try to establish causality in studies
of mass atrocities. One example is the study by Yanagizawa-Drott (2014), where
he uses data on radio reception to trace the effect of propaganda on killings in
the Rwandan genocide. Rwanda is a hilly country, and the evidence suggests that
hate speech against the minority Tutsi population increased militia violence in
the villages with radio reception. The resulting violence then spilled over into
neighboring villages. Another example is the study by Rogall (2014). The level of
violence at the village level during the genocide in Rwanda is estimated by using
distance from the main road as well as levels of rainfall. This is a plausible choice
of instrumental variables, because in this centrally planned mass atrocity, armed
groups were sent to the villages to kill civilians and to incite other civilians to
turn against the minority Tutsi population. Local roads are often impassable after
heavy rainfall, and the empirical evidence suggests that levels of violence were
lower in difficult-to-reach villages.
Table 10.1 Selected Large-n Studies of Mass Atrocities
Author(s) Civilian Atrocity Unit of Analysis Sample Scope and General Findings and Economic Correlates
Type Time Period
1. Rummel (1995) Democide, one- States, nonstate Focus on regimes Democide is mainly explained by the level of democracy,
sided lower-level groups with democides domestic rebellion, and war. GNP/capita, energy
atrocity, fatality 1900–1987 consumption, educational level not significant in regressions
criterion not of democide.
specified
2. Krain (1997) Genocide and States, nonstate 1949–1982 Main explanatory variables of the risk and severity of
politicide groups genocide/politicide are (civil) wars and extraconstitutional
changes; level of democracy is less robustly correlated.
The percentage of a country’s trade in world trade
(marginalization) is not significant.
3. Harff (2003) Genocide and States, nonstate States with state Prior genocide, regime type, ideology, ethnic minority
politicide groups failure 1955–1997 leaders, and political upheaval predict probability of a
genocide/politicide. Trade openness reduces the risk. Infant
mortality not significant.
4. Valentino, Huth, Mass killings in States 147 wars Guerilla tactics make mass killings more likely. The relative
and Balch- war, over 50,000 1945–2000 military capability of the guerrilla and the support for the
Lindsay (2004) fatalities guerilla make mass killings more likely. Democracy only
has a small effect on the probability of mass killings. No
economic variables considered.
(continued)
Author(s) Civilian Atrocity Unit of Analysis Sample Scope and General Findings and Economic Correlates
Type Time Period
5. Easterly, Gatti, Mass killing, States World 1820–1998 High levels of democracy and income make mass killings
and Kurlat fatality criterion less likely and reduce scale of killings. Wars make mass
(2006) not specified killings more likely and larger in scale.
6. Besançon Genocide (State States, nonstate World 1960–2001 Nonelite representation and higher GDP/capita lower the
(2005) Failures Data Set) groups risk of genocide. Economic and human capital inequality
more than 300 increase the risk and there is some evidence that natural
fatalities resource income increases the risk.
7. Eck and One-sided States, nonstate World 1989–2004 Lagged dependent variable and civil war dummy positive
Hultman violence groups and significant. Only some evidence that democracy reduces
(2007) risk of violence and number fatalities. Trade not significant,
no other economic variables reported.
8. Montalvo and Genocide (State States, nonstate World 1960–1999 Ethnic polarization increases the risk of genocide. Weaker
Reynal-Querol Failures Data Set) group evidence that higher levels of democracy and income
(2008) more than 300 decrease the risk. No evidence of significant relationship
fatalities between natural resource income and genocide risk.
9. Esteban, Mass Killings, States, nonstate World 1960–2007 Level of democracy not robustly correlated with risk of
Morelli, and PITF data group mass killings. Lagged dependent variable and civil war are
Rohner (2015) Minorities at Risk significant, as are a number of economic variables: GDP/
data for group- capita (-), oil and diamond production (+), and trade (-).
level analysis Group level: When groups are involved in a civil conflict
the risk of violence against civilians is high. Economic
geography measures also have a positive relationship with
this risk (soil quality, oil and diamond production).
Thus, the use of geographic and weather data enables Yanagizawa-Drott (2014)
and Rogall (2014) to establish why levels of violence varied across Rwanda.
However, while the application of the method of instrumental variables is useful
in the identification of causal relationships, it appears more difficult to find valid
instruments in the broader cross-country context.
10.3.1. War and Political Upheaval

As shown in c hapter 3, almost all mass atrocities committed during 1900–2013
occurred during periods of intrastate war (about 93 percent). It is therefore
unsurprising that conflict variables carry a positive and statistically significant
coefficient in models of mass atrocities. These include dummy variables for civil
war and international war (Krain 1997; Easterly, Gatti, and Kurlat 2006; Esteban,
Morelli, and Rohner 2015)2 or measures of the severity of the war, such as the
number of war deaths (Rummel 1995; Wood 2014). Interestingly, war duration
does not appear to be statistically significant (Rost 2013; Wood 2014).
In her study of genocides and politicides, Harff (2003) uses the concept of
“political upheaval,” which includes civil war, illegitimate regime transitions,
defeat in international wars, and new state formation. She terms these events
“state failure” and measures the magnitude of state failure on an ordinal scale,
summing up all values over the preceding fifteen years. This state failure variable
is statistically significant. When comparing a country in the twenty-fifth percen-
tile to a country in the seventy-fifth percentile, the risk of genocide occurring
is 1.7 times as high. Krain (1997) also finds evidence that major changes in the
political structure create windows of opportunity. International and civil wars,
extraconstitutional changes, and decolonization all are associated with a signifi-
cantly higher risk of genocides and politicides.
10.3.2. Past Atrocities

One of the most robust results in the civil war literature is that countries which
experienced a war in the past are more likely to experience another one. This has
been referred to as the “conflict trap” (Collier et al. 2003). In the mass atrocities
literature there is also considerable evidence that countries that experienced geno-
cides and politicides or other mass killings are more likely to experience further
atrocities against civilians. Harff (2003) accounts for genocides and politicides in
the previous forty-five years and finds this dummy variable statistically highly sig-
nificant. Specifically, countries with prior genocides and politicides are more than
three times as likely to experience another atrocity as are countries without such
a prior event. The study of mass killings by Esteban, Morelli, and Rohner (2015)
includes a lagged dependent variable, an indicator of whether mass killings took
place in the previous year. Eck and Hultman (2007) carry out a similar regression,
but in addition they explain the number of civilians killed by including a lagged
dependent variable. In both studies, the coefficients of past indicators of atrocities
are positive and statistically significant at the 1 percent level in all specifications.
10.3.3. Ethnicity
The importance of ethnicity has been evaluated in a number of studies. Genocides
are defined as the deliberate killing of members of a specific group, including eth-
nic groups, and a number of studies in the mass atrocities literature have included
variables capturing ethnic diversity, polarization, and discrimination. Ethnic
diversity is often measured by an ethnolinguistic fractionalization index (ELF),
which measures the probability that two randomly drawn citizens of one state
do not speak the same language (see Alesina et al. 2003). This diversity measure
has been included in regressions of mass atrocities, but the evidence is mixed.
Rummel (1995), Aydin and Gates (2008), and Kim (2010) find that there is no
statistically significant relationship between ELF and the risk of mass atrocities.
Wood (2014) documents a positive relationship, while Querido (2009) finds a
negative one in her analysis of violence against civilians by African governments.
Easterly, Gatti, and Kurlat (2006) find a nonmonotonic relationship in some
specifications, meaning that higher values of ELF are associated with a higher
risk, but very high values of ELF are associated with a lower risk. Montalvo and
Reynal-Querol (2008) investigate this relationship by using their measure of eth-
nic polarization and find that measure to be positive and statistically significant
in all of their regressions. Their results suggest that homogenous as well as rather
more heterogeneous societies face a lower risk of mass atrocities than societies
characterized by ethnic polarization. The countries with the highest risk are those
where large ethnic minorities face large ethnic majorities. Harff (2003) provides
support for this finding. She includes a dummy variable indicating whether the
political elite is based mainly or entirely on an ethnic minority. However, this is
in contrast to Valentino, Huth, and Balch-Lindsay (2004), who do not find any
evidence that identity-based conflicts are more likely to experience mass killings.
10.3.4. Ideology and Democracy

One of Rummel’s central arguments is that “Power kills; absolute Power kills
absolutely” (Rummel 1994, 1). He defines “Power,” with a capital P, as govern-
ment power and its holders (leaders), agencies (government departments and
bureaucracies), and instruments (e.g., propaganda, armies, concentration camps).
Rummel concentrates his analysis on democide, which he defines as the murder
of any person or people by a government, including genocide, politicide, and mass
murder. The so-defined democides overwhelmingly take place in totalitarian
regimes: 82 percent of all the deaths were caused by communist, fascist, militarist,
or Islamist regimes (Rummel 1997, 367). The most murderous regimes were the
USSR, China, and Nazi Germany, together committing about 75 percent of the
170 million deaths due to democides during the twentieth century (Rummel 1994,
10). Using data from a variety of sources, including the Polity dataset, Rummel
constructs a measure of Total Power that considers, among other political mea-
sures, democracy, autocracy, totalitarianism, and communist regime types. This
composite index, as well as some of the component parts, is highly correlated with
the occurrence of democides, and Rummel concludes that democracy prevents
democides (Rummel 1995). A number of other large-n studies also find evidence
that the level of democracy is negatively and statistically significantly related to the
risk of democide (e.g., Valentino, Huth, and Balch-Lindsay 2004; Easterly, Gatti,
and Kurlat 2006; Montalvo and Reynal-Querol 2008). Harff (2003) also con-
cludes that the probability of genocide is highest in autocratic regimes, when elites
advocate an exclusionary ideology or represent an ethnic minority. Using her data
on genocides, Aydin and Gates (2008) take a closer look at the institutions within
political regimes. They argue that the structural relationship between mass killing
and broad categorizations of political regimes is weak. Many autocratic regimes do
not perpetrate mass murder, and it is unclear why some partial democracies com-
mit atrocities while others do not. Furthermore, it is unclear which aspect of auto-
cratic rule makes regimes more murderous. It could be due to weak or nonexistent
constraints on the executive, the manner of executive recruitment, and the oppres-
sion of popular participation in the (s)election process. Aydin and Gates (2008)
argue that institutional limitations on executive power are central to understand-
ing when mass atrocities are likely to happen. Killings primarily occur in states
lacking constitutional checks on the political power of the sovereign.
A number of studies do not find a strong association between the level of democ-
racy and the risk of democide (e.g., Esteban, Morelli, and Rohner 2015). The study
by Wayman and Tago (2010) examines this relationship in more detail. Their
model includes a dummy variable for democracy or autocracy (based on the Polity
IV data) and a yes/no dummy variable for communist regime type. The dependent
variable, mass atrocities, is either (1) Rummel’s democide data or (2) Harff and
Gurr’s data (in Harff 2003). The indicators of democracy and communist regime
are statistically significant only in the regressions using Rummel’s democide
data. Wayman and Tago (2010) argue that Rummel’s concept of mass atrocities is
broader than the one developed by Harff and Gurr. The latter define genocides and
politicides as “sustained policies … that are intended to destroy, in whole or in part,
a communal, political, or politicized ethnic group” (Harff 2003, 58). In addition to
these killings, targeted at specific groups, Rummel’s definition also includes “mass
murder,” which he defines as “the indiscriminate killing of any person or people by
government” (Rummel 1994, 31). Thus, Wayman and Tago (2010) conclude that
autocratic regimes, especially communist ones, are more likely to engage in non-
targeted mass murder than in targeting specific groups (genocide and politicide).
10.3.5. Economy
There is sparse evidence that economic outcomes are associated with the risk and
magnitude of mass atrocities. This is in contrast to the civil war literature that has
identified a number of economic correlates of war, such as the level and growth
of income, as well as the structure of income (for overviews, see Blattman and
Miguel 2010; Hoeffler 2012).
Income could affect the risk of mass atrocities in a number of different ways.
Low incomes result in economic and emotional stress, which could trigger
aggression. Higher incomes of a minority group could act as an incentive to com-
mit genocide or politicide to appropriate these higher incomes. During civil wars
governments as well as the opposition rely on civilians for recruitment, supply
of finance, food, shelter, transport, and information. The strategic decision of
whether to coerce civilians to provide support may depend on alternative sources
of finance and support.
Richer countries are also more likely to have higher state capacity. Their tax
income is higher, and they are able to afford disciplined armies that do not have
to rely on loot and the associated violence against civilians. Their justice system
tends to be more efficient, thus avoiding the build-up of grievances and aggres-
sion. In states with well-functioning justice systems, individuals are less likely to
use private means (e.g., violence) to achieve justice. Higher-income countries also
have higher levels of democracy (as discussed) and more educated populations,
which may be less likely to commit mass atrocities. They are also likely to be part
of international organizations that provide additional scrutiny.
Thus, the level of income, its structure, and its distribution may potentially
explain the risk and magnitude of mass atrocities. In addition, international trade
can be interpreted as a proxy for international openness. More open countries
receive more scrutiny, and mass atrocities may be less likely to occur in such coun-
tries. The following discussion of economic variables considers the empirical evi-
dence for the relationship, if any, of income and trade with mass atrocities.
10.3.5.1. Level of Income

Skully (1997) presents a simple positive correlation between the number of peo-
ple killed in mass atrocities and the level of per-capita income, but multivariate
regressions show a statistically insignificant coefficient on income measures in
the models that attempt to statistically explain the risk of mass atrocity (Rummel
1994, 1995; Harff 2003). Easterly, Gatti, and Kurlat (2006) examine mass kill-
ings from 1820 to 1998 and include a measure of income per capita. They draw
on a number of databases to collect information on state-sponsored mass kill-
ings of civilians. They find that higher levels of income are associated with a lower
likelihood of episodes of mass killings, and higher income is also associated
with a lower number of civilians killed. Allowing for a nonlinear relationship by
adding a squared income term, the authors show that the risk of mass killings first
rises with income and then falls. Thus, intermediate levels of income are associ-
ated with a higher risk. When the sample excludes the nineteenth-century obser-
vations, the relationship between income and risk is linear: higher incomes are
associated with lower risks. Qualitatively similar results are obtained when the
magnitude, that is, the number of deaths, is examined.
Easterly, Gatti, and Kurlat (2006) examine the correlation of income and
democracy and find that, for their sample, the correlation coefficient for the two
variables is ρ = 0.55. When the model only accounts for democratic regimes,
the income variable is statistically significant, but democracy as a regime form
becomes statistically insignificant at conventional levels once income is included.
The relatively high correlation between income and democracy makes it difficult
to interpret the significance of income as a pure economic effect. The significance
of income could, for instance, also capture aspects of state capacity. Montalvo and
Reynal-Querol (2008) estimate a similar model for a shorter time period (1960–
1999) and include income per capita and levels of democracy. They allow only
for linear relationships and find that across a number of different specifications
income is negatively, and at the level of democracy, positively related to the risk
of genocides and politicides. However, the relationship of democracy seems to be
more robust. Democracy is statistically significant in most specifications, whereas
income is sometimes not significant at conventional levels. When Montalvo and
Reynal-Querol (2008) restrict the analysis to those genocides and politicides that
occur during civil wars, the partial correlation between income and risk is more
robust than is the relationship between democracy and risk. This mirrors the find-
ings in the civil war literature, where many studies find no significant relationship
between democracy and the risk of war when controlling for income (see, e.g., the
discussion in Hoeffler 2012).
A number of studies provide a subnational analysis to address the question of
whether higher local incomes are correlated with mass atrocities. There is insuf-
ficient subnational GDP data available to analyze atrocities in a large-n panel sam-
ple. Thus, various proxies have been used. Esteban, Morelli, and Rohner (2015)
examine what makes ethnic groups more likely to be victimized. One variable
that seems to be robustly correlated with that risk is local soil quality, which can
be interpreted as a determinant of income in rural areas. Valentino, Huth, and
Balch-Lindsay (2004) measure civilian support for the opposition in civil wars
as a dummy variable, taking a value of one if the active support by the civilian
population is high—that is, they provide supporters with food, shelter, informa-
tion, transport, and other logistical aid—and zero otherwise. This could be inter-
preted as income for rebels. However, the statistical analysis shows that civilian
support is positively correlated with the risk of mass killing. Valentino, Huth, and
Balch-Lindsay (2004) discuss the problem of possible endogeneity. Civilian sup-
port may be high because past experiences of violence may have instilled fear of
future mass killings in the civilian population. It is unclear in which cases state
repression diminishes civilian support and in which cases it increases the support
for the opposition. Keeping these endogeneity issues in mind, Valentino, Huth,
and Balch-Lindsay (2004) argue that guerilla wars are more prone to experienc-
ing mass atrocities and that because of their tactics guerillas depend on civilian
support, thus making civilians more likely to be a target for government atroci-
ties. (For further discussion, see c hapter 19 in this volume.) However, given these
findings, it is impossible to decide whether civilian support should primarily be
interpreted as a strategic or as an economic variable.
10.3.5.2. Structure of Income: Aid, Natural Resources, and Foreign Support

One variable that can be interpreted as income to the government is international
development aid. Most aid (e.g., budget support) cannot be appropriated by the
opposition, with food aid being the notable exception. Food aid can be appropri-
ated by the opposition and used to support their fighters (see, e.g., Anderson 1999;
Nunn and Qian 2014). However, over time the amount of food aid has declined
to about two percent of total aid. Thus, most aid goes directly to the government.
Azam and Hoeffler (2002) examine the relationship between aid and violence
against civilians in Africa. They approximate this violence by the number of refu-
gees and show that countries that receive more aid appear to experience higher
levels of violence thus measured. They interpret this as evidence that government
armies can terrorize civilians because they have to rely less on civilian support,
due to their aid income. However, this result is open to interpretation because the
authors do not address any endogeneity concerns. It may be the case that coun-
tries with mass killings are perceived as in need of aid and thus attract more donor
funding in the first place. Azam and Hoeffler (2002) also assume that violence
against civilians is perpetrated by the government. More recent evidence, from
Raleigh (2012), suggests that in Africa opposition groups kill more civilians than
do government forces and militias.
Income from natural resources has been analyzed extensively in the civil war
literature. Natural resources can motivate rebellion (the “honeypot” argument in
DeSoysa and Neumayer 2007) or provide a channel for rebel finance (Collier and
Hoeffler 2004; Lujala, Gleditsch, and Gilmore 2005). Diffuse resources (such as
alluvial diamonds) are considered to be more likely to finance rebels than point
resources (such as crude oil), which require technical know-how and typically
involve international companies for their exploitation. More generally, states that
derive a relatively high income from natural resources suffer from severe political
economy problems because these states rely less on tax income and thus are sub-
ject to less budgetary scrutiny from their populations. The revenue from natural
resources is often used to provide private goods to the elite and not for redistribu-
tion to the wider population or for public goods. The unwillingness, or incapacity,
to redistribute natural resource–derived wealth may even cause grievances that

motivate rebellion (see chapter 21 in this volume). Thus, Isham et al. (2005) argue
that crude oil producing countries tend to have weaker institutional capacity,
which may then leave them less able to deter rebellion effectively.
Esteban, Morelli, and Rohner (2015) examine the effects of natural resources
on the risk of experiencing episodes of mass killings. Using different resource
measures, they find that countries with high crude oil–related income, and dia-
mond producers as well, are prone to a higher risk of violence against civilians.
In their analysis of ethnic groups, they also find that these groups are more likely
to be victimized if the region in which the group lives is characterized by oil or
diamond production. They interpret these findings as evidence that mass killings
follow from material interests: groups are being targeted because of their wealth.
This appears plausible for violence perpetrated by government because crude oil
production in particular relies on international companies, which typically coop-
erate with governments and do not operate in rebel territories. However, Esteban,
Morelli, and Rohner (2015) analyze violence against civilians perpetrated by
state and nonstate actors. As much of the violence against civilians is committed
by the opposition (Eck and Hultman 2007; Raleigh 2012), it is thus difficult to
understand opposition-perpetrated violence on the basis of the material interest
hypothesis.
Querido (2009) concentrates on violence against civilians perpetrated by
African governments. She presents evidence that a higher risk of these atrocities
is positively correlated with alluvial diamond deposits and with onshore and off-
shore crude oil production. The number of civilians killed is also higher when
a country produces both diamonds and oil. Although these results are not dis-
cussed in much detail, they could be interpreted as providing evidence for the
hypothesis that if governments have other options of financing war, they do
not have to rely on civilian support. In the case of outside funding, atrocities
against the civilian population have little consequence in financing government
conflict with a rebel movement. Violence against civilians can, in these cases, also
become a strategy to undermine support for the rebels, a tactic that is not open to
the government when it has to rely on civilian support.
The interaction between strategy and resources in civil wars and their implica-
tion for violence against civilians perpetrated by opposition groups is examined
by Wood (2014). Based on a rebel-government/dyad-year sample, he shows that
even when they receive moderate levels of support, rebel groups kill civilians.
Mere popular sympathy is not sufficient for recruitment, and this result indicates
that rebel groups often coerce individuals to join up. However, when the relative
military capability of the group increases, violence against civilians declines in
situations of popular support. The interaction of military capability with exter-
nal finance is statistically positive, indicating that victimization increases when
outside sponsorship is provided. In these cases, financial support substitutes for
popular support. However, financing derived from natural resources, such as gem-
stones or illicit drugs, does not have an effect on violence against civilians: there is
neither a direct effect nor an interactive effect with military capability.
Using a principal-agent model, Salehyan, Siroky, and Wood (2014) concen-
trate their analysis on the impact of foreign-state funding for rebel organizations
on violence against civilians. Their empirical results suggest that foreign-state
funding for rebel organizations greatly reduces incentives to “win the hearts and
minds” of civilians because it diminishes the need to rely on the local popula-
tion. In addition to this result, they show that the extent of atrocities committed
against the civilian population depends on the regime type of the foreign sup-
porter and on the number of supporters. Democracies, for example, tend to house
strong human rights lobbies, and in this case the levels of violence are lower. If
there are several foreign supporters, the levels of civilian abuse tend to be higher
because no single state can effectively restrain the rebel organization.
10.3.5.3. Income Inequality

One of the most frequently cited causes of civil war is inequality (e.g., Sen 1973;
chapter 9 in this volume), but commonly used measures of inequality are not sta-
tistically significant in civil war onset regressions. One of the most commonly
used proxies for income inequality is the Gini coefficient, which captures inequal-
ity among individuals. What might matter more is the inequality between groups,
and Stewart (2002) makes the distinction between “vertical” and “horizontal”
inequality. Horizontal inequality is understood to be the outcome of discrimi-
nation against groups in an inequitable society. This is close to Regan’s concept
of “structural” poverty (Regan 2009). There is some evidence that horizontal
inequality increases the risk of civil war (Østby 2008; Cederman, Gleditsch, and
Buhaug 2013); correspondingly, there is no evidence that vertical inequality is
associated with a higher risk of civil war.
For genocide, it may be that inequality forges a sense of “relative deprivation”
that causes frustration and turns into aggression (Gurr 1970). Genocide could be
motivated by the inequality between groups; inequality could cause aggression in
a sufficiently large number of individuals to kill significant members of another
group. However, in the mass atrocity literature there is very little testing of any
statistical significance of inequality. Rost (2013) includes a qualitative measure
of economic discrimination, which appears to be associated with a higher risk
of genocide onset (significant at the 10 percent level), but the variable receives
no discussion in the study. A more detailed analysis of political and economic
discrimination is presented in Anderton and Carter (2015). They conclude that
although the two measures are highly correlated, the explanatory power of eco-
nomic discrimination appears to be greater. Besançon (2005) includes the Gini
index of income and finds that genocides are more likely in countries with higher
economic inequality. Similarly, she uses an index of human capital inequality,

showing that genocides are more likely in countries where educational oppor-
tunities are unequally distributed. Thus, there is some evidence that vertical
inequality is associated with a higher risk of genocide.
However, for the study of genocide it may be of interest to investigate the
impact of horizontal inequality. To my knowledge, none of the large-n studies
have considered measures of horizontal inequality. Further investigation of the
concepts of vertical and horizontal inequality and how they may relate to the risk
of genocide may be useful. If societal groups are relatively small, and horizon-
tal inequality is large, this should also manifest itself in large vertical inequality
measures. The potential role of within-group inequality in genocide would also be
interesting to explore. Based on the theoretical work of Esteban and Ray (2008),
group heterogeneity makes collective action more difficult, but this very hetero-
geneity may enable a division of labor such that the poorer members of the group
carry out the atrocities and the richer members provide the finances, thus making
it easier to carry out mass atrocities than in a more homogenous group, which may
have less opportunity to divide labor.
10.3.5.4. International Trade

The relationship between international trade and the risk of mass atrocities has
been examined by Harff (2003) and Krain (1997). Trade connects countries,
and these international linkages may make it less likely that countries engage in
mass atrocities because trade partners may act as an external enforcer of human
rights. Countries with high trade-to-GDP ratios may fear the response from their
trading partner, possibly in the form of sanctions. International trade may also be
closely related to memberships in other (political) international organizations.
These international memberships may make (nonviolent) conflict management
more likely, may provide scrutiny of one’s internal affairs, and thus may make
genocides and politicides less likely. Harff (2003) finds that openness to interna-
tional trade is statistically significant, and negatively correlated with genocides
and politicides. However, when she substitutes trade openness with memberships
in international organizations, this membership variable is not statistically sig-
nificant. Thus, economic openness and international memberships do not cap-
ture the same aspects of international interdependence. In Harff’s (2003) study,
economic interdependency appears to be more important in determining the risk
of violence against civilians than the fact that it captures a country’s international
political connections.
Krain (1997) examines a different aspect of international trade. He analyzes
whether economic marginalization makes countries more prone to experiencing
mass atrocities. Economic marginalization is calculated as a country’s percentage
of world trade. Similar to the reasoning advanced by Harff (2003), marginalized
countries are less well connected internationally and receive less attention in any
efforts to stem violence. Alternatively, following from psychological approaches,
one could argue that the sense of “relative deprivation” is higher in marginalized
countries, which could make more individuals frustrated and induce them to act
aggressively. In any case, there is no statistical evidence that global economic
marginalization has an impact on the risk of mass atrocities.
10.4. Conclusions
This chapter reviews the large-n literature on the correlates of mass atrocities.
Psychologists have concentrated on explaining genocides, which are one form of
mass atrocities. Genocides are the intentional destruction, in whole or in part, of a
specific group of people. The underlying assumption is that individuals are being
killed not for what they do but for what they are. In addition to these intentional
destructions of groups, there are many mass killings, defined as killings where the
perpetrators either do not intend to eliminate the group as such or victims do not
belong to any specific group. In many cases, it is difficult to detect intention to
eliminate or weaken a group; intention is usually inferred from actions and political
statements (Shaw 2007). However, the overwhelming majority of mass atrocities
(genocides, politicides, and mass killings) occur during a civil war. In this context,
violence against civilians can be interpreted as a strategic choice. The civilian pop-
ulation provides fighters, food, shelter, transport, and information. States as well
as rebel armed groups depend to a large extent on civilian support, and violence
against civilians has received a number of different explanations. Violence against
civilians can destroy this support for the opposing forces; it can also coerce a civil-
ian population to cooperate and hand over informants. Violence against civilians
also depends on how combatants are organized and funded. Many armed groups
have little military discipline and are not well resourced. To provide short-term
benefits, violence against civilians is encouraged, often helped by the recruitment
of thugs who are driven by their individual preferences for violence and not by an
overarching political aim. The instability and insecurity that civil war creates also
opens up opportunities to settle old scores and to seize property. Again, these are
private motivations, not aimed at destroying a specific group of people. However,
ex post, these atrocities against civilians can look like genocide (i.e., like the inten-
tional destruction of a group). Ethnic and other well-defined groups are often geo-
graphically concentrated, and if their location becomes contested territory during
the war, violence in this location can look like genocide. Thus, it does not seem to
be useful to categorize mass atrocities by prior intent.
A number of large-n studies of mass atrocities do not consider any economic
variables. There is only weak evidence that countries with higher incomes
are less prone to mass atrocities. There is stronger evidence that the level of
democratization is negatively correlated with mass atrocities. However, income

and democracy are highly correlated, and few studies consider the consequences
of this correlation for their statistical models and for the interpretation of the
results. Income to an armed group, rather than the average income of the country,
has been considered in some studies. Results suggest that if combatants are well
funded through external support or natural resource rents, there is more violence
against civilians, that is, support from civilians can be substituted by income from
other sources.
Income inequality could provide a motive for violence. It has been found to
be statistically significant in the literature on homicides (Fajnzylber, Lederman,
and Loayza 2002), but there is almost no evidence that inequality is a driver of
civil war. Surprisingly, there are very few large-n studies of mass atrocities that
consider income inequality. The study by Besançon (2005) is a notable exception;
she finds a positive relationship between the Gini coefficient and the risk of geno-
cide. Instead, research seems to concentrate on ethnicity as a potential explana-
tion of mass atrocities. The evidence for this is mixed, however. Using the same
proxy for ethnic diversity, some researchers find no relationship with the risk of
mass atrocities, while some find a negative relationship, yet others find a positive
one. A measure of ethnic polarization appears to be positively correlated with the
risk of mass atrocities, but is only supported by one study (Montalvo and Reynal-
Querol 2008). Studies considering other measures of economic and political dis-
crimination of groups do not provide cohesive evidence either.
Economic variables seem to be important in conjunction with strategic deci-
sions, and the further study of this interdependence appears promising. However,
since state and nonstate armed groups have different access to financial support,
it is of interest to examine the violence against civilians by perpetrators (state or
nonstate). Lastly, much of this literature does not consider the possible endoge-
neity of some of the explanatory variables used. It is important to establish the
direction of causality, in particular if these models are used for forecasting in
early warning systems. The models of mass atrocities produce many false posi-
tives (e.g., Harff 2003; Goldsmith et al. 2013; Rost 2013), possibly because they
show no more than that “bad things go together.” For example, civil war is one of
the most robust predicters of mass atrocities, but not all civil wars are character-
ized by mass atrocities.
Acknowledgments
The research leading to these results has received funding from the European
Community’s Seventh Framework Programme (FP7/2007–2013) under grant
agreement number 290752 (NOPOOR project).
Notes
1. Further details on the measurement of development can be obtained from the UN HDI
website: http://hdr.undp.org/en/content/human-development-index-hdi [accessed
March 30, 2015].
2. Krain (1997) and Easterly, Gatti, and Kurlat (2006) include a dummy variable for civil
war and a variable for other types of war, which includes international, extrasystemic, and
colonial war. Their dummies are based on the Correlates of War dataset. Esteban, Morelli,
and Rohner (2015) only include a civil war dummy.
References
Alesina, A., A. Devleeschauwer, W. Easterly, S. Kurlat, and R. Wacziarg. 2003. “Fractionalization.”
Anderson, M. B. 1999. Do No Harm: How Aid Can Support Peace—or War. Boulder, CO: Lynne
Rienner.
Anderton, C. H., and J. R. Carter. 2015. “A New Look at Weak State Conditions and Genocide
Risk.” Peace Economics, Peace Science and Public Policy 21, no. 1: 1–36.
André, C., and J. P. Platteau. 1998. “Land Relations under Unbearable Stress: Rwanda Caught in
the Malthusian Trap.” Journal of Economic Behavior and Organization 34, no. 1: 1–47.
Aydin, A., and S. Gates. 2008. “Rulers as Mass Murderers: Political Institutions and Human
Insecurity.” In S. M. Saideman and M.-J. J. Zahar, eds., Intra-State Conflict, Governments and
Security: Dilemmas of Deterrence and Assurance. New York: Routledge, 72–95.
Azam, J.-P., and A. Hoeffler. 2002. “Violence against Civilians in Civil Wars: Looting or Terror?”
Baum, S. 2008. The Psychology of Genocide: Perpetrators, Bystanders, and Rescuers.
Besançon, M. L. 2005. “Relative Resources: Inequality in Ethnic Wars, Revolutions, and
Cederman, L.-E., K. S. Gleditsch, and H. Buhaug. 2013. Inequality, Grievances, and Civil War.
Collier, P., and A. Hoeffler. 2004. “Greed and Grievance in Civil War.” Oxford Economic Papers
56, no. 4: 563–95.
Collier, P., V. L. Elliot, H. Hegre, A. Hoeffler, M. Reynal-Querol, and N. Sambanis. 2003. Breaking
the Conflict Trap: Civil War and Development Policy. Oxford: Oxford University Press.
DeSoysa, I., and E. Neumayer. 2007. “Resource Wealth and the Risk of Civil War Onset: Results
from a New Dataset of Natural Resource Rents, 1970–1999.” Conflict Management and Peace
Science 24, no. 3: 201–18.
New Fatality Data.” Journal of Peace Research 44, no. 2: 233–46.
Economy 123, no. 5: 1087–1132.
Esteban, J., and D. Ray. 2008. “On the Salience of Ethnic Conflict.” American Economic Review
98, no. 5: 2185–2202.
Fajnzylber, P., D. Lederman, and N. Loayza. 2002. “What Causes Violent Crime?” European
Goldhagen, D. J. 1996. Hitler’s Willing Executioners: Ordinary Germans and the Holocaust.
London: Little, Brown.
Goldsmith, B. E., C. R. Butcher, D. Semenovich, and A. Sowmya. 2013. “Forecasting the Onset
of Genocide and Politicide: Annual Out-of-Sample Forecasts on a Global Dataset, 1988–
2003.” Journal of Peace Research 50, no. 4: 437–52.
Gurr, T. R. 1970. Why Men Rebel. Princeton, NJ: Princeton University Press.
Political Mass Murder Since 1955.” American Political Science Review 97, no. 1: 57–73.
Hoeffler, A. 2012. “On the Causes of Civil War.” In M. Garfinkel and S. Skaperdas, eds., Oxford
Handbook of the Economics of Peace and Conflict. New York: Oxford University Press,
179–204.
Isham, J., M. Woolcock, L. Pritchett, and G. Busby. 2005. “The Varieties of Resource
Experience: Natural Resource Export Structures and the Political Economy of Economic
Growth.” World Bank Economic Review 19, no. 2: 141–74.
Kalyvas, S. 2006. The Logic of Violence in Civil War. Cambridge, UK: Cambridge University
Press.
Kim, D. 2010. “What Makes State Leaders Brutal? Examining Grievances and Mass Killing
During Civil War.” Civil Wars 12, no. 3: 237–60.
Politicides.” Journal of Conflict Resolution 41, no. 3: 331–60.
Lujala, P., N. P. Gleditsch, and E. Gilmore. 2005. “A Diamond Curse? Civil War and a Lootable
Resource.” Journal of Conflict Resolution 49, no. 4: 538–62.
Montalvo, J. G., and M. Reynal-Querol. 2008. “Discrete Polarisation with an Application to the
Mueller, J. E. 2007. The Remnants of War. Ithaca, NY: Cornell University Press.
Nunn, N., and N. Qian. 2014. “U.S. Food Aid and Civil Conflict.” American Economic Review 104,
no. 6: 1630–66.
Østby, G. 2008. “Polarization, Horizontal Inequalities, and Violent Civil Conflict.” Journal of
Pinker, S. 2011. The Better Angels of Our Nature: A History of Violence and Humanity. London: Penguin
Books.
Querido, C. M. 2009. “State-Sponsored Mass Killing in African Wars—Greed or Grievance?”
International Advances in Economic Research 15, no. 3: 351–61.
Raleigh, C. 2012. “Violence against Civilians: A Disaggregated Analysis.” International
Interactions 38, no. 4: 462–81.
Regan, P. 2009. Sixteen Million One: Understanding Civil War. Boulder, CO: Paradigm Press.
Rogall, T. 2014. “Mobilizing the Masses for Genocide.” Job Market Paper. Institute for
International Economic Studies. Stockholm University.
Rost, N. 2013. “Will It Happen Again? On the Possibility of Forecasting the Risk of Genocide.”
Journal of Genocide Research 15, no. 1: 41–67.
Rummel, R. J. 1994. Death by Government. New Brunswick and London: Transaction.
Resolution 39, no. 1: 3–26.
Salehyan, I., D. Siroky, and R. M. Wood. 2014. “External Rebel Sponsorship and Civilian
Abuse: A Principal-Agent Analysis of Wartime Atrocities.” International Organization 68,
no. 3: 633–61.
Sen, A. 1973. On Economic Inequality. Oxford: Clarendon Press.
Skully, G. W. 1997. “Democide and Genocide as Rent-Seeking Activities.” Public Choice 93, nos.
1–2: 77–97.
Staub, E. 1989. The Roots of Evil: The Origins of Genocide and Other Group Violence. Cambridge,
Stewart, F. 2002. “Horizontal Inequalities: A Neglected Dimension of Development.” Fifth

Annual Lecture. Helsinki: UNU-WIDER. http://www.wider.unu.edu/publications/
annual-lectures/en_GB/AL5/ [accessed April 7, 2015].
Guerrilla Warfare.” International Organization 58, no. 2:375–407.
Waller, J. E. 2007. Becoming Evil: How Ordinary People Commit Genocide and Mass Killing. 2nd ed.
Wayman, F. W., and A. Tago. 2010. “Explaining the Onset of Mass Killing, 1949–87.” Journal of
Weinstein, J. M. 2007. Inside Rebellion: The Politics of Insurgent Violence. Cambridge, UK: Cambridge
University Press.
Origins of Support, and Civilian Victimization in Civil War.” Conflict Management and Peace
Science 31, no. 5: 461–80.
Yanagizawa-Drott, D. 2014. “Propaganda and Conflict: Evidence from the Rwandan Genocide.”
11
Who Stays and Who Leaves During Mass

Atrocities?
A na M a r í a I bá ñ ez a n d A n dr é s Moya
11.1. Introduction
The number of people forcibly displaced from their homes or places of habitual
residence as a consequence of “persecution, conflict, generalized violence, or
human rights violations” is distressingly large: some 59.5 million people by the
end of 2014, according to the United Nations High Commissioner for Refugees
(UNHCR) (UNHCR 2015). Of those, 38.2 million people were classified as inter-
nally displaced persons (IDPs), a figure about equal to the population of Poland.
Syria is the country with the largest number of IDPs in absolute terms in 2014
(7.6 million people), followed by the number of IDPs in Colombia (6.0 million
people, or about 12.4 percent of Colombia’s total population), the only “hotspot”
in the Americas by the UNHCR’s reckoning. The IDPs seldom return to
their hometowns even when the conflict is over. Thus, UNHCR data show that in
2014 only about 1.8 million IDPs returned to their home residence (UNHCR 2015).1
Forced displacement occurs during episodes of genocides and mass killings
but also, and to a larger extent, in nongenocidal violent conflicts. To clarify our
terminology, we refer in this chapter to large-scale violent conflict of a collective
rather than of an interpersonal nature or due to organized crime—or simply to
“conflict,” “episodes of mass atrocities,” or similar shorthand expressions. It is
important to appreciate that, to generate its consequences, violence need not be
carried out in fact but only credibly threatened.
If armed actors are hegemonic in a territory, violence against groups of civil-
ians is infrequent and is targeted at particular individuals instead. Conversely, in
contested territories, attacks intensify and are targeted against a collective of peo-
ple, are targeted at a group of people, or are altogether indiscriminate (Kalyvas
2006; Steele 2009; Wood 2010). The escalation of aggression toward civilians is
not a by-product but rather a strategy of war. Attacks are a deliberate and effective
251
strategy to terrorize whole populations, weaken support provided to rival groups,

force collaboration, seize valuable assets, and control territories (Kalyvas 1999;
Azam and Hoeffler 2002; Kalyvas 2006; Wood 2010).
While published research exists on the causes and consequences of displace-
ment, much less is written on the decision to stay or on the consequences of that
decision. In violent environments, migration becomes an extreme coping mecha-
nism to minimize the risk of victimization (or a reaction after having become a vic-
tim) and often is the only remaining option available for civilian populations. As
discussed in detail later, the literature shows that exposure to violence, such as assas-
sinations, attacks, or threats, among others, is the main cause of forced displace-
ment. Nevertheless, economic and political dimensions also play a role in the decision
to leave even as violence continues to be the dominant push-factor (Morrison and
May 1994; Engel and Ibáñez 2007; Williams 2008; Czaika and Kis-Katos 2009;
Lindley 2010; Lozano-Gracia et al. 2010; Bohra-Mishra and Massey 2011; Balcells
and Steele 2012; Williams 2013). To be clear, when we use the phrase decision to
leave in this chapter, we mean it in the sense of both physically forced removal and
households evaluating their options and leaving because they view that decision as
their best remaining option under the circumstances. Since there is no established
terminology, “forced” versus “compelled” displacement will serve the purpose.
The decision to stay in a conflict region likewise reflects financial or economic
constraints. Some households stay because they lack sufficient financial resources
to fund migration or lack contacts in destination locations (Kalyvas 1999; Lindley
2010). Other households decide to stay because of the high opportunity cost of
migration (Engel and Ibáñez 2007; Ibáñez and Vélez 2008). Yet other house-
holds navigate through violent conflict in ways that make migration somewhat
less likely. For example, some civilians interact strategically with armed groups to
reduce their risk of victimization, such as joining combatants, collaborating with
the dominant group, or forming alliances with armed actors (Korf 2004; Kalyvas
and Kocher 2007; Steele 2009; Lindley 2010). Or, in some cases, they do so even
to obtain economic gains from conflict (Korf 2004; Kalyvas 2006; Steele 2009;
Lindley 2010; Zetter, Purdekova, and Ibáñez 2013). All such interactions play an
important role in the stay-or-leave decision.
In this chapter, then, we review the literature with three objectives in mind.
First, in section 11.2, we discuss how strategies adopted by armed actors to con-
trol civilian populations influence household migration decisions. Second, also
in section 11.2, we explore why some households decide to stay. Third, we distin-
guish between the impacts on those who migrate as opposed to those who stay
(section 11.3). Section 11.4 concludes our discussion. Note one important topic
that we do not discuss: Who returns and who remains (at the destination location),
and why, and with what consequences? The circumstances of the plight of remain-
ees and returnees, and associated topics such as statelessness, would require
another chapter.
W h o S t a y s a n d W h o L e a v e s d u r i n g M a s s A t r o c i t i e s? 253
11.2. Who Leaves and Who Stays?

During irregular wars, armed groups require the support of civilians. Civilians
provide food, shelter, protection, and information (Kalyvas 2006; Kalyvas and
Kocher 2007). Since frontlines are blurry, rebels may hide among the population,
making combatants and noncombatants difficult to distinguish (Kalyvas and
Kocher 2007). In marshaling civilian support, armed groups can provide benefits
and public goods, among them security, or else resort to violence against civil-
ians (Kalyvas 1999; Wood 2010). This violence can result in forced or compelled
migration. Although the immediate trigger is often expected or actual violence,
displacement is the result of the particular dynamics of violence and interactions
between armed groups and civilians.
In this section we first analyze how armed groups, seeking to control terri-
tory, adopt strategies to rule over civilians, either by coercion or persuasion.
The perceived need to control a population can result in deliberate and targeted
aggression, causing forced migration of particular groups of civilians. Second, we
examine the strategies civilians use to minimize the risk of aggression and to sur-
vive while staying in conflict regions. Third, apart from strategy, we discuss how
(the threat of) violence helps shape the migration decision. And, fourth, we focus
on economic factors that enter the migration decision.
11.2.1. Strategies of Armed Groups

The strategies of armed groups to control a population may cause an escalation of
aggression against civilians and ensuing displacement. For genocides and politi-
cides, displacement rates are higher than for other mass atrocities since they are
directed toward specific groups of the population and thus affect a higher number
of civilians (Uzonyi 2014). For other internal conflicts, aggression against civil-
ians occurs more often when two or more groups are contesting a region. In con-
tested regions, armed groups cannot as easily provide benefits to the population
and thus resort to terror tactics to spread fear (Kalyvas 1999; Steele 2009). In
addition, attacking civilians aims at alienating support for rival groups, redistrib-
uting political power, and seizing valuable assets from the population (Morrison
and May 1994; Azam and Hoeffler 2002; Balcells and Steele 2012). Violence is
also an effective strategy used by weaker armed groups to force civilian collab-
oration, collect valuable information, and reduce the benefits to civilians from
remaining neutral (Kalyvas 2006; Steele 2009; Wood 2010).
Country studies corroborate these hypotheses. Stanley (1987) shows that
armed groups in El Salvador, including the army, used violence to control civil-
ians. Morrison and May (1994) argue that armed groups attacked rural popu-
lations in Guatemala for economic gain. Williams (2008) and Czaika and
Kis-Katos (2009) discuss how armed groups in Nepal and Aceh (Indonesia) used
displacement strategically to reshape the political landscape and separate armed

groups from their civilian base. Wood (2010) describes different stages of interac-
tions between the National Resistance Army (NRA) in Uganda and civilians. At
first, the NRA targeted civilians suspected of collaborating with the government.
Once the NRA controlled the territory, it provided services and public goods.
When control collapsed, the NRA expelled some population groups.
Attacks against civilian populations are seldom random. Although this
requires rich information on the behavior of civilians and control over the terri-
tory, armed groups use selective violence to target particular population groups
(Kalyvas 1999). Econometric and qualitative evidence show that attacks are pur-
poseful. On the one hand, to change the distribution of political power in their
favor or to terrorize the population, armed groups direct violence against groups
with particular political affiliations or against individual community leaders. For
instance, Balcells (2012), Balcells and Steele (2012), and Adhikari (2013) find
that displacement in Spain, Colombia, and Nepal is correlated with the politi-
cal affiliation of migrants, while Engel and Ibáñez (2007) show that in Colombia
community leaders are the more likely targets of direct threats. On the other
hand, targeting wealthier households is a strategy to accumulate valuable assets
and augment the loot of combatants (Azam and Hoeffler 2002). In Colombia,
large-scale landowners and better-educated individuals reported a higher like-
lihood of receiving direct threats from armed groups (Engel and Ibáñez 2007;
Ibáñez and Vélez 2008). During the Rwandan genocide, large-scale landholders
and cattle owners were overrepresented among the victims (André and Platteau
1998; Verwimp 2005; Verpoorten 2009).
11.2.2. Strategies of Civilians

Despite facing violence and redistribution of power within conflict regions, indi-
viduals have agency and often some leverage in their interactions with armed groups
(Wood 2010). The main objective is to minimize the risk of becoming a victim
of violence, and thus the priority is to remain alive (Kalyvas 1999; Wood 2010).
In pursuit of this objective, people may decide to participate in armed groups as
supporters or combatants, form alliances with one of the armed groups or power-
holders, remain neutral, or migrate (Korf 2004; Kalyvas and Kocher 2007; Steele
2009; Justino 2011). These decisions are not static, and they change according to
the dynamics of war.
Kalyvas and Kocher (2007) develop a theoretical framework to understand
the decision of individuals to join rebel groups. Because armed groups may pro-
vide selective protection to combatants and their families as well as basic goods
(e.g., food), some individuals become combatants simply to improve their likeli-
hood of survival and not due to their political preferences. Using data for Greece
and El Salvador, they indeed find that noncombatants who decided to stay in their
towns and do not collaborate with armed groups in a period of escalating violence
experienced a higher likelihood of being murdered than combatants. Most civil-
ians killed in Greece had limited or no links with any of the armed groups other
than some forced acts of collaboration with the controlling armed group.
Participation in armed groups may also stem from private incentives in the
form of the accumulation of assets, wealth, and power (Kalyvas and Kocher
2007). Combatants may seize assets and accumulate wealth. André and Platteau
(1998) argue that increasing landlessness in Rwanda, and the breakdown of
customary laws to transfer land across generations, alienated young people.
The genocide thus became an opportunity to gain access to land (Verwimp
2005). Based on a household survey, Verwimp (2005) finds that perpetrators
of the Rwandan genocide belonged to households with abundant labor but lim-
ited access to land, implying less land per productive household member (see
chapters 15 and 22 in this volume). Also, perpetrators were more likely to have
higher nonfarm incomes and to participate more actively in land rental markets
due to land scarcity.
Instead of joining armed groups, civilians may stay in conflict regions and nego-
tiate their daily lives in other ways (Korf 2004; Lindley 2010; Zetter, Purdekova,
and Ibáñez 2013). Stayers adjust their behavior to the controlling group and the
new rules imposed (Kalyvas 1999). Some households may form alliances with the
controlling group and obtain economic and political gains from conflict, while
others may retreat to their private lives to minimize the risk of victimization (Korf
2004; Lindley 2010). Also, households modify their economic behavior by reduc-
ing visible investments, increasing the percentage of land left idle, and retreating
from markets (Deininger 2003; Grun 2008; Bozzoli and Brück 2009). Responses
of, and final outcomes for, individuals thus differ according to their economic
condition, social status, and political resources (Korf 2004; Lindley 2010).
11.2.3. Beyond Strategy: The Decision to Migrate

Migration is another alternative to minimize the risk of victimization (Verpoorten
2009; Zetter, Purdekova, and Ibáñez 2013). As mentioned, micro-level quantita-
tive studies conducted at the community or district level show that the incidence
of violence is the main trigger of displacement. Case studies find a strong cor-
relation between violence and displacement in El Salvador (Stanley 1987), Nepal
(Williams 2008; Bohra-Mishra and Massey 2011), China (Gottschang 1987),
Colombia (Lozano-Gracia et al. 2010), Guatemala (Morrison and May 1994),
Spain (Balcells 2012), and Indonesia (Czaika and Kis-Katos 2009). Adhikari
(2013), Engel and Ibáñez (2007), and Ibáñez and Vélez (2008) use measures of
direct victimization, victimization of family and friends, or perceived risk of vio-
lence for Nepal and Colombia. These studies confirm violence as the strongest
correlate of the migration decision.
The impact of violence on migration can be nonlinear. Violence deters migra-

tion up to a threshold beyond which the correlation between violence and migra-
tion becomes positive (Morrison and May 1994; Bohra-Mishra and Massey
2011). Nonlinearity between violence and migration may be picking up on the
heterogeneity of the impact of conflict in contested and controlled territories.
When a region is controlled by a dominant group, a low incidence of violence may
not produce migration. In contested territories, violence is high and unpredict-
able. People may respond by migrating to safer places.
A massive displacement in Mogadishu, Somalia, highlights some of the dynam-
ics of violence and the decisions that civilians have to make (Lindley 2010). Most
of the city’s population had lived for several decades in an extremely violent envi-
ronment, yet certainty about the likely incidence of violence in time and space
allowed individuals to navigate through conflict. Once violence became unpre-
dictable and the population failed to negotiate their daily lives with armed actors,
displacement erupted. Some individuals migrated soon after family members
were killed or their main productive assets were destroyed. For others, migration
became the last resort after seeking recourse to other coping strategies. The loss of
protection from armed actors, due to the reconfiguration of political power, also
prompted many to migrate.
Yet dynamics beyond (the threat of) violence also cause displacement.
Households may migrate in spite of not facing direct victimization. Some results
of the above mentioned studies point in this direction, although in differenti-
ated ways. Thus, Engel and Ibáñez (2007) for example find that the presence of
police forces lowers the likelihood of receiving direct threats, while military pres-
ence increases this likelihood. Police presence may signal government control of a
region, implying lower expected levels of aggression against civilians. In contrast,
military forces are present when they are contesting territories with rebel groups,
and violence may escalate. Czaika and Kis-Katos (2009) find a similar correlation
between low levels of violence and the presence of police forces in Aceh, Indonesia.
In addition, their study shows that institutional presence, a proxy of governance
and control of a dominant group, reduces the incidence of political violence.
11.2.4. Economic Factors

Besides the dynamics of war, economic factors help determine the migration
decision. Some individuals may perform cost-benefit analyses in which they
trade reductions in income for improved safety and a lower risk of victimiza-
tion (Morrison and May 1994; Engel and Ibáñez 2007; Ibáñez and Vélez 2008;
Lozana-Gracia et al. 2010; Williams 2013). However, distinguishing between the
impact of violence and economic variables on migration is difficult as violence
also disrupts economic activity and destroys household assets (Lindley 2010).
Thus, economic variables can be endogenous.
Five broad channels are related to population displacement through eco-

nomic variables. First, armed groups often attack the better-off individuals in
the community to seize assets and terrorize populations (Verwimp 2005; Engel
and Ibáñez 2007; Ibáñez and Vélez 2008; Adhikari 2013). Second, conflict may
severely disrupt economic activities, destroy or deplete physical assets, and reduce
the income-generating capacity of households. Individuals may migrate not after
being victims of attacks or to prevent victimization, but due to absolute destitu-
tion brought about by war (Lindley 2010). Third, economic variables determine
the ability of households to generate expected income in their hometowns and
destination locations. Ownership of valuable assets, such as land or work oppor-
tunities, increases the opportunity cost of leaving, reducing the likelihood of
migration for some individuals in spite of high levels of violence (Morrison and
May 1994; Engel and Ibáñez 2007; Ibáñez and Vélez 2008; Williams 2008;
Bohra-Mishra and Massey 2011; Adhikari 2013). Economic opportunities in des-
tination locations may prompt people to migrate not only to be safer but also to
gain economically from migration (Lozano-Gracia et al. 2010). Fourth, financial
and economic resources may facilitate migration. Wealthier households may sell
assets to fund migration and resettlement (Kalyvas 1999; Engel and Ibáñez 2007;
Lindley 2010; Bohra-Mishra and Massey 2011). Fifth, the presence of social
networks may reduce migration costs. Dense social networks in the hometown
may provide support to individuals facing difficult conditions and protect pri-
vate property (Adhikari 2013; Williams 2013), while social networks at destina-
tions may facilitate migration by providing initial shelter and work opportunities
(Bohra-Mishra and Massey 2011).
Although literature on displacement is not plentiful, research on why people
decide to stay in spite of high levels of violence is practically nonexistent. Stayers
may face a lower risk of victimization (Morrison and May 1994; Engel and Ibáñez
2007). As mentioned, alliances with armed actors may provide protection from
a rival group or may help avoid targeting by the dominant group. In some cases,
alliances with armed actors may provide economic benefits and political power.
However, other population groups may decide to stay simply because they lack
resources to migrate. Poorer people, with fewer assets, might not be able to fund
migration and resettlement. Also, less educated people might face more difficul-
ties to settle in a new location and find work there, effectively compelling them to
stay in their hometown in spite of violence. Stayers are subject to extreme vulner-
ability and may face unique difficulties even when conflict ends.
Population redistribution—migration—on account of violence depends on
the strategic behavior of armed groups and victims, political affiliations, and eco-
nomic resources (Williams 2008). In turn, population reallocation has important
socioeconomic implications for conflict regions and reception sites. If better-off
individuals migrate to avoid the hardships of war, towns with forced emigration
may lose economic resources, thus reducing the likelihood of full recovery once
the violence ends. Conversely, if poorer individuals migrate, destination locations

will receive inflows of populations in need of significant social service provision
and state support. Migration and strategic interactions may also redistribute
political and economic power in the expelling regions. Stayers with relatively
strong leverage over a dominant armed group may benefit in the conflict period
by accumulating assets, increasing income, expanding social networks, and
acquiring or strengthening political power. In contrast, people who stay because
of sheer necessity—willing but unable to migrate—may face extreme conditions
of vulnerability.
11.3. The Socioeconomic Consequences

of Staying or Leaving
The literature reviewed thus far highlights that to understand migration one
needs to go beyond the analysis of immediate economic and political determi-
nants and comprehend in detail the factors that influence the decision to migrate
or to stay behind. Research shows that violence affects socioeconomic, political,
and behavioral outcomes mostly in adverse ways. Unfortunately, as we discuss
in this section, previous research often fails to distinguish between the conse-
quences for those who migrated as opposed to those who stayed behind, and in
general ignores that people adopt strategies to minimize the expected negative
consequences of conflict. Distinguishing between stayers and leavers, then, and
understanding the different coping strategies they adopt, is important from aca-
demic and public policy perspectives. Improved knowledge and understanding
may allow us to design better policies for different population subsets, not only
according to the degree but also according to the circumstances by which they
have been affected by violence. In this section we review the evidence on the eco-
nomic consequences of large-scale violence, focusing on studies in which we can
distinguish between stayers and leavers.2
In general, it is widely assumed that violence has negative consequences on
poverty dynamics, economic growth, and social and economic inequality. After
all, countries and regions afflicted by violence suffer devastating consequences,
such as the loss of lives, displacement of civilians, destruction of physical capital
and infrastructure, and institutional decay. Perhaps surprisingly, some macro-
oriented research suggests that the economic consequences of warfare eventually
dissipate. Evidence from postwar Germany (Brakman, Garretsen, and Schramm
2004), Japan (Davis and Weinstein 2002), and Vietnam (Miguel and Roland
2011) finds few persistent effects of World War II and the Vietnam War, respec-
tively, on long-run economic performance, growth, or poverty rates. Similarly, the
economic performance of countries like Mozambique, Sierra Leone, and Uganda
suggests that war-torn countries can catch up with prewar levels (Bellows and
Miguel 2009; Justino and Verwimp 2013). That said, the finding for Sierra Leone
is highly dependent on just which gross domestic product (GDP) and other eco-
nomic data one studies; and Mozambique is “catching up” only about now, some
forty years after the start of its civil war in 1975. Cerra and Saxena (2008) study all
countries, finding evidence for each of three possibilities: some countries suffer
permanent decline in their baseline growth trend, some do eventually catch up to
the level of economic activity implied by their prewar baseline growth, and some
experience permanent trend growth that exceeds their prewar trend. However,
for countries that suffered through episodes of genocide, chapter 5 in this volume
provides evidence of permanent economic damage.
The global findings notwithstanding, the experience of specific conflict-torn
countries suggests that ongoing wars do not necessarily hinder current economic
performance. One such case is Colombia, a country afflicted by a long-lasting,
and ongoing, civil conflict. And yet, in recent years Colombia has achieved higher
growth rates than its relatively conflict-free neighbors. However, even for coun-
tries like Colombia—or Sri Lanka for that matter—seemingly “positive” mac-
rolevel evidence can disguise disturbing microlevel dynamics and ignore the
devastating and heterogeneous consequences for specific subsets of the popula-
tion, including those who reside in conflict-torn areas, those directly victimized,
and those displaced. In fact, recent microlevel evidence highlights how popula-
tions exposed to widespread episodes of violence suffer severely adverse, and in
some cases permanent, welfare effects. Such evidence has allowed us to observe
how violence imparts substantial negative effects on the livelihoods and behav-
iors of afflicted populations. As we will discuss, some of these effects vary accord-
ing to whether a specific population migrated or stayed behind. We also discuss
some of the channels through which these effects translate into a higher vulner-
ability to poverty. 3
11.3.1. Assets and Livelihoods

During conflict, households are vulnerable to severe losses in their stock of physi-
cal capital since their houses, lands, livestock, and appliances often are destroyed,
abandoned, or looted (Justino 2009). As a result, populations exposed to violence
have a lower capacity to cope and to generate income. They can then fall into per-
manent poverty. The extent of such asset losses, and its impact on persistent pov-
erty, varies according to the specific context—were conflicts specifically fueled
by land disputes and was violence targeted toward landowners, for instance?—
and to the segment of the population that one focuses on, stayers or leavers.
To understand the process and consequences of asset losses for the displaced,
we refer by way of example to various studies on forced displacement in Colombia.
Engel and Ibáñez (2007), for instance, find that civilians are forced to migrate
due to a combination of direct threats and violence directed toward households,
especially those with small or medium-sized land holdings. This is not surprising
since conflict dynamics in Colombia historically have been driven by disputes

over the use and control of land (Berry 2006). As a result, by 2006, land areas that
the displaced lost, abandoned, or sold at low prices are equivalent to 3.4 times as
much as the land areas awarded by the Agrarian Reform efforts of 1993 to 2002
(Ibáñez 2009).
The consequences of such asset losses are dreadful, as documented by a series
of studies that rely on microlevel data from a nationally representative survey of
displaced populations (Ibáñez and Moya 2010a, 2010b). The majority of the dis-
placed lost all of their productive assets throughout the process of displacement,
especially their lands, and only a few were able to recover their asset base over
time. As a result, household consumption and income levels fell, respectively, by
53 and 28 percent, placing the majority of the displaced below the extreme pov-
erty line. Moreover, the displaced have few mechanisms to cope with the shock
of violence and displacement, and thus rely on costly coping strategies that make
them, and their children, more vulnerable to poverty. As displaced households
are deprived of their main productive asset, livelihoods are compromised and
they seem destined for chronic poverty. In the presence of asset-based poverty
traps (Carter and Barret 2006), such losses can push victims below a critical asset
threshold and thwart their ability to recover in the future.
The experience of displaced populations in Colombia diverges from that of
Rwanda. In her study, Kondylis (2008) evaluates the impact of “villagerization”
initiatives by comparing asset holdings and levels of well-being of households
who were displaced and returned with those who stayed. She compares across
regions, with and without the policy intervention. Perhaps surprisingly, her
results indicate that returnees have higher access to land and higher levels of well-
being than the stayers. The results are not indicative of the circumstances of the
overall displaced population, however. Returnees are better off, in this case, most
likely as a consequence of their access to the villagerization intervention program
and other sources of aid, as well as because of the continuous adverse effects borne
by stayers during and after the genocide.
In fact, evidence from other studies highlights that populations that stay
behind suffer significant asset losses, although the channels by which these losses
occur are different from those that affect the displaced. In one study, Justino
and Verwimp (2013) identify the impact of the Rwandan genocide on poverty
dynamics using a unique panel dataset of households surveyed pre- and postwar.
Although there is not a clear distinction on whether the households in the panel
are stayers or migrants, it is safe to assume that most of them were stayers—bar-
ring the few who could have been surveyed in 1989–1990, migrated as a result of
the genocide, then returned, and were surveyed again in 2002. The results indi-
cate that the Rwandan genocide brought about massive asset losses, and those
households that resided in conflict-torn regions and lost homes and lands had a
higher risk of falling into poverty.
While the Colombian and Rwandan studies depict similar consequences for
the displaced, stayers lose assets for different reasons than do the displaced. In
Rwanda, for instance, stayers relied on distress land sales to cope with the shock
of violence and the disruption of employment, credit, and insurance markets.
Such behavior is consistent with findings of Verpoorten (2009), who highlights
how households used asset sales to smooth consumption during the genocide and
adjusted their asset portfolio afterward.4 While distress asset sales can be use-
ful to cope with the shock of violence, they permanently undermine households’
future ability to generate income and move out of poverty—eating the seed today
prevents the harvest tomorrow. In addition, the frequent use of land sales has gen-
eral equilibrium effects and depresses prices, thus deepening the vulnerability of
households residing in conflict or postconflict areas.
Asset losses are a result of costly coping and consumption-smoothing strate-
gies adopted by households during conflict, even if not directly exposed to vio-
lent shocks. The ability to generate income and to recover postconflict is similarly
undermined. For instance, Bozzoli and Brück (2009) find that postwar recovery
in northern Mozambique is endogenously linked to the wartime crop mix adopted
in response to asset losses. Thus, farmers who, postwar, retained their asset-poor
wartime mix of subsistence farming fared worse than those who, postwar, shifted
to cotton farming for export markets, as promoted by aid agencies.
Findings such as these carry powerful implications for postconflict policies.
Since policymakers, governments, international institutions, and nongovern-
mental organizations often direct their attention to the displaced, regardless of
whether or not they return, the nonmigrants—that is, the stayers—also suffered
severely, and yet may be ignored and be left even further behind.
While studies indisputably establish that violent conflict drives both leavers
and stayers into poverty through asset losses, it is possible that some population
subsets could benefit from conflict. Stayers, in particular, might gain access to
land by direct possession, low purchase prices, or an allocation from an armed
group. As discussed in section 11.2, stayers may be able to take advantage of eco-
nomic, social, and political alliances with armed groups, and progress as a result
(Verwimp 2005; Justino 2009; Justino and Verwimp 2013). However, this topic
seems unexplored and thus offers great potential for future research.
11.3.2. Risk Assessment and Livelihoods

Perceptions and assessment of risk also play a significant role in poverty dynam-
ics in conflict and affect migration decisions. Risk is a key determinant of eco-
nomic behavior since it imposes a cost on current investments. In the absence
of credit and insurance markets, investing in profitable but risky activities may
not be worthwhile, precisely because the returns are uncertain while the foregone
consumption needed to finance them is certain. Households may thus settle on
activities with unduly low levels of risk and return, or even increase present con-
sumption and not invest at all.
Consistent with this hypothesis, several studies find that violence-related risk
affects the activity choices, investment composition, and incomes and expendi-
tures of exposed populations. Among the displaced, the experience of violence
affects even their postviolence assessment of risk and consequent economic
behavior. For example, in a World Bank–guided Moving Out of Poverty study,
groups of displaced persons in Colombia received land as part of governors’ pro-
grams in six states (departamentos) but were reluctant to invest in these plots;
instead, they grew subsistence crops. When questioned, they stated that they
had learned that they could be displaced once, and this could happen yet again.
It thus did not make any sense to them to invest in their newly acquired plots
(Matijasevic et al. 2007). Direct exposure to violence and displacement led the
displaced to update their prior beliefs, magnify their assessments of the likeli-
hood of future shocks, and display higher levels of risk aversion. As we discuss in
more detail later on, the degree of the observed effects are driven by differences in
the experience of displacement, the severity of the exposure to violence, and the
resulting psychological consequences.
Just the mere exposure to violent environments, even in the absence of any
directly experienced victimization or displacement, can have similar effects on
risk assessments and behavior. In Uganda, Rockmore (2011, 2012) finds that vio-
lence-related risk perception induces changes in the composition of associated
livestock and crop portfolios, lowers per-capita household expenditure by 2 to
6 percent, and accounts for half of the overall conflict-related losses. In Colombia,
households residing in conflict regions respond to risk by lowering productive
investments and adjusting the allocation of land from perennial to subsistence, or
illegal, crops (Ibáñez, Muñoz, and Verwimp 2013; Arias, Ibáñez, and Zambrano
2014). Likewise, Bundervoet (2006) finds that, in conflict regions, assets lose their
value as informal saving and insurance mechanisms, and that both rich and poor
households resort to subsistence activities as a consequence of risk perceptions.
Overall, these studies suggest that, for leavers and stayers alike, conflict-related
risk has substantial effects on activity choices, farm productivity, investments, and
general behavior and can therefore result in long-lasting effects on poverty dynamics.
11.3.3. Human Capital

Violent conflict also increases vulnerability to poverty through its effect on
human capital, specifically on the physical and mental health, educational attain-
ment, and employment opportunities of migrants and stayers. In most cases,
the effects are more severe for children and compromise their future well-being,
thus constituting a mechanism for the intergenerational transmission of poverty
regardless of whether parents migrated or stayed behind.
Regarding employment opportunities of displaced populations, evidence has

shown severe constraints to join labor markets and a lack of demand for their agri-
cultural skills. The displaced, in particular, often migrate from rural to urban areas
where they are unable to find stable or quality employment. Three reasons explain
this: first, agricultural skills are not transferable to urban areas; second, the dis-
placed are discriminated against by urbanites and by potential urban employ-
ers; and third, urban labor markets are not able to absorb the massive inflow of
migrants. Consistent with these hypotheses, Ibáñez and Moya (2007) find that
in Colombia the rate of unemployment among the displaced increases from
1.7 percent in the places of origin to over 50 percent at reception sites in the first
year after migrating. Over time, their unemployment rates remain consider-
ably higher than those of poor and extremely poor urban populations. Similarly,
Kondylis (2010) finds that Bosnian displaced persons are not able to find jobs in
postwar Bosnia and Herzegovina, while Rodríguez et al. (2013) find suggestive
evidence on the discrimination patterns of employers in Colombia toward indi-
viduals fleeing conflict regions.
For stayers, the effects on employment opportunities depend on how labor and
product markets operate during conflict and postconflict periods. Unfortunately,
there is little evidence on the way in which conflict affects these markets. One
notable exception is the work by Fernández, Ibáñez, and Peña (2014), which finds
that populations residing in Colombian conflict regions increase off-farm labor
supply to mitigate the effect of violence shocks. Markets, however, are unable
to fully absorb the excess labor supply, leaving households with fewer options
to cope with the shock and uncertainty that characterize conflict regions. Such
effects can provide an explanation for the use of asset sales as a coping mecha-
nism, as previously discussed.
Violent conflict also directly affects school enrollment, educational attain-
ment, and test scores. Perhaps surprisingly, school enrollment can increase for
displaced children as a result of migration to urban areas where they have access
to government assistance programs and many more public schools. This is the case
in Colombia, where the displaced receive conditional cash transfers and have pref-
erential access to schools. School enrollment is thus higher for displaced children
than even for the urban and rural poor (Ibáñez and Moya 2007). Nonetheless,
despite higher rates of school attendance, displaced children lag behind same-
age peers in their educational attainment, possibly as a result of the interruption
caused by displacement. In addition, school drop-out rates are higher as displaced
households rely on child labor as a coping mechanism (Ibáñez and Moya 2010b).
For stayers, the evidence unequivocally shows negative effects of conflict on
education. In Cambodia (de Walque 2004), Colombia (Angrist and Kugler 2008;
Rodriguez and Sánchez 2012; Duque 2013), Rwanda (Akresh and de Walque
2011), Tajikistan (Shemyakina 2011), and Zimbabwe (Alderman, Hoddinott, and
Kinsey 2006), children residing in conflict regions have lower school enrollment
and higher drop-out rates, resulting in lower overall rates of educational attain-
ment. Some of these effects vary by gender, suggesting that households invest
more in the education of male children to preserve the household’s stock of
human capital (Shemyakina 2011); they also vary by schooling level, highlight-
ing that schooling is affected by household coping strategies and by the risk of
military recruitment of older children (Swee 2009). Finally, conflict also affects
children’s test scores through its effect on teacher absenteeism, temporary school
closures, and high principal turnover (Monteiro and Rocha 2013).
Overall, conflict generates severe, long-run effects on the well-being of younger
generations of both migrants and stayers. Yet, as for the case of asset losses, most
educational aid programs are targeted at displaced populations alone and seem
to ignore the circumstances of populations who remain in conflict regions. It is
worth noting, however, that studies that analyze the long-term effects of wars and
violence find ambiguous evidence in regard to adult education and earnings (see
Justino 2009 for a detailed discussion of this literature).
Large-scale violence also carries negative consequences for the physical health
of affected populations, although the literature does not necessarily distinguish
between leavers and stayers. Violent conflict, of course, causes severe injuries,
permanent physical disability, and death of family members of both migrants and
stayers. These human capital losses often thrust households into chronic, persis-
tent poverty, for instance, when widows are forced to assume the role of head of
household and primary economic provider (Berlage, Verpoorten, and Verwimp
2003; Brück and Schindler 2008; Verwimp and Bundervoet 2008; Justino and
Verwimp 2013).
In addition, there are considerable effects on children’s health, which vary
according to the age of exposure to violence; whether their household was a direct
victim or not; and, for the displaced, on the length of time spent at reception sites.
In Colombia, Ortiz (2012) finds that displaced children suffer higher likelihood
of chronic malnutrition than do children who remain in conflict-torn municipali-
ties. Evidence from Burundi, Rwanda, and Zimbabwe shows that children, even
if their family households were neither victimized nor displaced, are more likely
to experience stunting—lower height-for-age ratios—than do children from con-
flict-free regions (Bundervoet and Verwimp 2005; Alderman, Hoddinott, and
Kinsey 2006; Bundervoet, Verwimp, and Akresh 2009; Akresh and de Walque
2011; Akresh, Bundervoet, and Verwimp 2011; Agüero and Deolalikar 2012).
Related evidence from Colombia shows that residing in conflict areas with a
higher incidence of land mine explosions has a statistically significant negative
effect on children’s birthweight, an effect that is possibly explained by the impact
on mothers’ psychological stress (Camacho 2008).
Understanding the psychological consequences of violent conflict is impor-
tant since they can compromise the ability of both migrants and stayers to cope
and recover. Exposure to traumatic events, including violence and forced or
compelled displacement, brings about distressing psychological consequences

and emotional disorders, including post-traumatic stress disorder, major depres-
sion, and anxiety disorders (Doctors Without Borders 2010). Such disorders
overwhelm psychological coping mechanisms, produce severe emotional suffer-
ing, and can adversely influence social functioning and hinder task performance
(Mollica, Wyshak, and Lavelle 1987; Vinck et al. 2007). What is more, mental
trauma can persist and develop into chronic anxiety and depressive disorders,
especially in environments of continuing stress or violence (Kessler et al. 1995;
Goenjian et al. 2000; Yehuda 2002).
Although few studies differentiate between displaced households and stayers,
studies from Colombia find that the incidence of psychological disorders is higher
for the former and for direct victims than for populations remaining in conflict
areas. As might be expected, even the latter exhibit a higher incidence of mental
trauma than do residents of conflict-free regions (Doctors Without Borders 2006,
2010; Moya 2015). As in the case of the effects on labor, education, and physi-
cal health, the mental health consequences of violence shocks households’ stock
of human capital, both for migrants and stayers, and can thus have transcending
effects on poverty. Unfortunately, the implications of mental trauma for the eco-
nomic recovery of conflict-afflicted populations are largely unexplored. 5 Perhaps
this is why current policies often focus on repairing the material consequences
of violence and neglect the psychological consequences. Although there is an
emerging body of work that relates to the psychological dimensions of behavior,
more work is needed to understand the implications of violence-related mental
trauma on poverty dynamics.
As a whole, the research reviewed here suggests that the human capital of
migrants and stayers alike is severely affected by large-scale violence. Accordingly,
both groups face considerable hardships to cope with the shocks of violence and
recovery, and thus face high vulnerability to poverty that is also transmitted across
generations. Unfortunately, most policy frameworks fail to incorporate specific
programs to mitigate the human capital cost of populations exposed to violence,
especially when it comes to mental health–related programs. Even when available
they often focus on displaced populations, leaving out those who stayed behind.
11.3.4. Behavior
Finally, we discuss a body of literature that analyzes ways in which violent conflict
brings about changes in other attributes, attitudes, and behaviors, such as hope,
altruism, collective action, political participation, risk and time preferences, and
social capital, in conflict and postconflict countries (Bellows and Miguel 2009;
Blattman 2009; Cassar, Healy, and Kessler 2011; Voors et al. 2012; Moya 2015;
Bauer et al. 2014; Callen et al. 2014; Moya and Carter 2014). Behavior influences
key decisions over productive investments, asset accumulation, and schooling,
among others. Depending on the direction, magnitude, and persistence of the

effects of violence, exposed populations can experience obstacles to cope and
recover, in addition to those discussed already. Perhaps surprisingly, there is no
clear picture regarding the direction of such changes or the way in which they
promote or hinder the socioeconomic recovery of individuals, households, and
entire societies.
The seemingly ambiguous results are explained, in part, by the methodologi-
cal approach employed in most studies. In particular, difficulties in collecting
detailed individual-level data on violence exposure have driven researchers to
rely on violence data at regional, municipal, or district levels. This implies that it
is not possible to distinguish behavioral effects among those who were displaced
or stayed behind, or those who were directly exposed to violence and those who
were not, nor among those affected by violence at different times. Therefore, the
studies’ estimated effects result from data-weighted averages that possibly mask
individual-level effects as well as any temporal or permanent nature of such effects.
Recent evidence, however, highlights that behavioral responses and conse-
quences do in fact vary according to the severity of violence exposure, whether
the population was forced to migrate, and to the time elapsed since such shocks
occurred. In Colombia, for instance, Moya (2015) and Moya and Carter (2014)
find that displaced households are more risk-averse and hopeless than comparable
households that remained in conflict-torn regions but that had not been directly
exposed to violence. More importantly, they find that the levels of risk aversion
and hopelessness increase if the displaced had been exposed to more severe and
recent episodes of violence, and if they suffer from anxiety and depression dis-
orders. Such differential behavioral effects also arise among populations that
remained in conflict regions and did not migrate. For instance, in Afghanistan,
Callen et al. (2013) find that the combination of higher levels of district-level vio-
lence and cues to induce specific emotions induce a higher preference for certain
outcomes over uncertain but profitable ones, and that the effects are stronger in
districts with higher levels of recent violence. Similar trends are found by Bauer
et al. (2014) regarding changes in altruism and reciprocity among displaced and
non-displaced populations in Georgia and Sierra Leone, although the underlying
mechanisms are different.
Such findings lay the groundwork for future studies on the behavioral conse-
quences of large-scale violence. On the one hand, they highlight how the effects
vary according to whether or not the population was forced to leave, but also with
regard to the severity and recentness of the exposure to violence. On the other
hand, they provide evidence on the underlying mechanisms that explain such
behavioral changes, in this case psychological mechanisms related to the inci-
dence of mental trauma and the susceptibility to psychological priming, and in
this particular case to the recall of previous experiences of violence, which bring
about unconscious but predictable patterns of behavior. While the data necessary
to analyze heterogeneous behavioral effects of widespread violence is hard to get,

this work is essential so as to arrive at a better understanding of the way in which
economic, social, and political preferences change with the degree of exposure to
violence, whether such changes are permanent, and the channels through which
they occur.
Furthermore, it is important to understand how changes in individual prefer-
ences affect economic decisions, and if they compromise the ability of the pop-
ulation to cope and recover. From a theoretical perspective, if violence induces
higher levels of risk aversion and impatience and lowers the hopes and aspirations
of migrants and stayers, it will likely bring about worrisome patterns of economic
behavior, which can increase vulnerability to poverty. In this way, violent conflict
could create behavioral poverty traps, requiring different types of policy interven-
tions, for instance, those that pay more attention to mental health and behavioral
dimensions. But from an empirical perspective, and with the exception of the
work by Voors et al. (2012) in Burundi, such work has not yet been done and is a
topic of great relevance for future research.
11.4. Conclusion
We reviewed research from economics, political science, sociology, medicine,
and psychology with a specific purpose in mind: to distinguish how large-scale
violent conflict differentially affects those who are displaced as opposed to
those who stay behind. We need to identify why civilians are targeted by armed
actors in the first place, what are the strategies that civilians use to cope and to
survive in the wake of violent conflict, why some eventually end up migrating,
and what are the consequences for those who stay and for those who leave. This
knowledge is important to acquire, catalogue, and process and not just for the
sake of academic interest. Rather, it comes with profound policy implications as
it highlights how civilians are involved in conflict dynamics; how they adjust,
suffer, and in some cases even benefit from conflict; and the different channels
through which conflict can thrust those who stay and those who leave into long-
lasting and even permanent poverty. As such, distinguishing among causes and
consequences of staying in, or migrating away from, conflict regions can better
inform policymakers, donors, and aid agencies on appropriate programs and poli-
cies to protect and assist affected civilian populations.
However, little is known about how people living in conflict regions devise
strategies to minimize their risk of victimization and pursue their daily activities,
and how the chosen strategies ultimately determine the impact of armed conflict
on economic outcomes. For this reason, we examined the literature on violent
conflict and compared studies in which we could identify reasons why civil-
ians may become an integral part of conflict dynamics, the factors that motivate
civilians to stay or to leave, the strategies adopted to cope, and the consequences
for both groups. All this is just a first step to learn more about conflict-related
migration, but also a call to push the boundaries of our knowledge further.
We call for improvements in the methodological approaches used, including
those that we have used in our own studies. One concern is that researchers often
rely on aggregate-level violence data—homicide rates, attacks, or bombings at
district or municipal levels. By doing so, it is not possible to differentiate between
those who were directly victimized and those who were not, or between those
who migrated and those who stayed. This is problematic because pictures of aver-
age effects are difficult to understand and could be biased if the samples are not
representative or if the impact of conflict is nonlinear. Likewise, the data-average
approach ignores that civilians employ different strategies to minimize the risk of
victimization and to ensure survival, and that it is those strategies that determine
the eventual economic, social, and political consequences of violence.
Finally, it is important to further explore the channels and mechanisms
through which different populations can fall into poverty on account of vio-
lence. Postconflict policies often target direct victims, such as internally dis-
placed persons, and generally follow a one-size-fits-all approach. As a result, they
ignore the population that was compelled to stay in conflict regions and who also
face extremely difficult economic and social conditions. Thus, preexisting
inequalities may be deepened if people who stay in conflict regions cannot and
do not receive postconflict assistance. Moreover, a generic policy approach fails
to recognize that the different survival strategies that leavers and stayers adopt
may push some households even further into poverty traps from which they may
be unable to escape on their own. A better understanding of the determinants of
victimization and the decision to migrate, as well as of the coping strategies and
consequences of violent conflict, is thus crucial to permit researchers to inform
and improve policies and programs for conflict environments and postconflict
reconstruction.
Notes
1. For UNHCR statistics, see UNHCR (2015). Also see Ibáñez (2014).
2. For a detailed review of the economics of civil war, see Blattman and Miguel (2010).
3. We do not discuss the effect of conflict on product and financial markets, since the litera-
ture on these topics is scant. These effects, however, can bring about multiple market failures
and oblige the population, especially the one that stays behind, to rely on their own income
generation activities and informal risk management. They can therefore induce economic
environments with multiple regimes, where decisions are endowment sensitive, and bring
about a poverty trap.
4. Such behavior is different from that of households exposed to weather shocks, who do not
necessarily smooth consumption but may smooth asset holdings to maintain their future
productive capacity (see Fafchamps, Udry, and Czukas 1998; Kazianga and Udry 2006;
Carter and Lybbert 2012).
5. There is substantial evidence in psychology on negative and persistent effects of domestic
violence and early childhood adversities on cognitive and socioemotional development,
school enrollment, educational attainments, and income. While these results come from
highly vulnerable environments in developed countries, one imagines that the consequences
for children exposed to violent collective conflict is similar and even more disturbing. See
Tough (2012) for a good review of this evidence.
References
Adhikari, P. 2013. “Conflict-Induced Displacement, Understanding the Causes of Flight.”
American Journal of Political Science 57, no. 1: 82–89.
Agüero, J. M., and A. Deolalikar. 2012. “Late Bloomers? Identifying Critical Periods in
Human Capital Accumulation. Evidence from the Rwanda Genocide.” Presented at the
Ninth Midwest International Economics Department Conference, Minneapolis, April
20–21.
Akresh, R., and D. de Walque. 2011. “Armed Conflict and Schooling: Evidence from the 1994
Rwandan Genocide.” World Bank Policy Research Working Paper 4606. Washington,
DC: World Bank.
Akresh, R., T. Bundervoet, and Verwimp, P. 2011. “Civil War, Crop Failure and Child Stunting
Alderman, H., J. Hoddinott, and B. Kinsey. 2006. “Long-Term Consequences of Early Childhood
Malnutrition.” Oxford Economic Papers 58, no. 3: 450–74.
André, C., and J.-P. Platteau. 1998. “Land Relations under Unbearable Stress: Rwanda Caught in
Angrist, J., and A. Kugler. 2008. “Rural Windfall or a New Resource Curse? Coca, Income, and
Civil Conflict in Colombia.” Review of Economics and Statistics 90, no. 2: 191–215.
Arias, M., A. M. Ibáñez, and A. Zambrano. 2014. “Agricultural Production amid Conflict: The
Effects of Shocks, Uncertainty, and Governance of Non-State Armed Actors.” Documento
CEDE No. 8. Available at https://ideas.repec.org/p/col/000089/011005.html.
Azam, J. P., and A. Hoeffler. 2002. “Violence against Civilians in Civil Wars: Looting or Terror?”
Balcells, L. 2012. “Violence and Displacement in Civil War: Evidence from the Spanish War
1936–1939.” Barcelona GSE Working Paper No. 603. Available at www.iae.csic.es/investi-
gatorsMaterial/a1292083411archivoPdf86166.pdf.
Balcells, L., and A. Steele. 2012. “Warfare, Political Identities, and Displacement in Spain and
Colombia.” Households in Conflict Network Working Paper No. 124. http://www.hicn.
org/wordpress/wp-content/uploads/2012/06/HiCN-WP-124.pdf [accessed October
2014].
Bauer, M., A. Cassar, J. Chytilová, and J. Henrich. 2014. “War’s Enduring Effects on the
Development of Egalitarian Motivations and In-Group Biases.” Psychological Science 25, no.
1: 47–57.
Bellows, J., and E. Miguel. 2009. “War and Local Collective Action in Sierra Leone.” Journal of
Public Economics 93, nos. 11–12: 1144–57.
Berlage, L., M. Verpoorten, and P. Verwimp. 2003. “Rural Households under Extreme
Stress: Survival Strategies of Poor Households in Post-Genocide Rwanda.” A Report
for the Flemish Interuniversity Council and the Belgium Department of International
Cooperation under the Policy Research Program.
Berry, A. 2006. “Has Colombia Finally Found an Agrarian Reform That Works?” In J. Boyce, S.
Cullenberg, and P. Pattanaik, eds., Human Development in the Era of Globalization: Essays in
Honor of Keith B. Griffin. Portland, OR: Book News, 126–53.
Blattman, C. 2009. “From Violence to Voting: War and Political Participation in Uganda.”
American Political Science Review 103, no. 2: 231–47.
Bohra-Mishra, P., and D. S. Massey 2011. “Individual Decisions to Migrate during Civil
Conflict.” Demography 48, no. 2: 401–24.
Bozzoli, C., and T. Brück. 2009. “Agriculture, Poverty, and Postwar Reconstruction: Micro-
Level Evidence from Northern Mozambique.” Journal of Peace Research 46, 3: 377–97.
Brakman, S., H. Garretsen, and M. Schramm. 2004. “The Strategic Bombing of Cities in
Germany in World War II and Its Impact on City Growth.” Journal of Economic Geography
4, no. 2: 201–18.
Brück, T., and K. Schindler. 2008. “The Impact of Conflict and Fragility on
Households: A Conceptual Framework with Reference to Widows.” World Institute for
Development Economic Research Working Paper 2008/83. Helsinki: UN-WIDER.
Bundervoet, T. 2006. “Livestock, Activity Choices and Conflict: Evidence from Burundi.”
Households in Conflict Network Working Paper No. 24. http://www.hicn.org/wordpress/
wp-content/uploads/2012/06/wp24.pdf [accessed October 2014].
Bundervoet, T., and P. Verwimp. 2005. “Civil War and Economic Sanctions: An Analysis of
Anthropometric Outcomes in Burundi.” Households in Conflict Network No. 11. http://
www.hicn.org/wordpress/wp-content/uploads/2012/06/wp10.pdf [accessed October
2014].
Bundervoet, T., P. Verwimp, and R. Akresh. 2009. “Health and Civil War in Rural Burundi.”
Journal of Human Resources 44, no. 2: 536–63.
Callen, M., M. Isaqzadeh, J. Long, and C. Sprenger. 2014. “Violence and Risk
Preference: Experimental Evidence from Afghanistan.” American Economic Review 104, no.
1: 123–48.
Camacho, A. 2008. “Stress and Birth Outcomes: Evidence from Terrorist Attacks in Colombia.”
Carter, M., and C. Barrett. 2006. “The Economics of Poverty Traps and Persistent Poverty: An
Asset-Based Approach.” Journal of Development Studies 42, no. 2: 178–99.
Carter, M., and T. Lybbert. 2012. “Consumption versus Asset Smoothing: Testing the
Implications of Poverty Trap Theory in Burkina Faso.” Journal of Development Economics 99,
no. 2: 255–64.
Cassar, A., A. Healy, and C. Kessler. 2011. “Trust, Risk and Time Preferences after a Natural
Disaster: Experimental Evidence from Thailand.” University of San Francisco Working
Paper. http://www.alessandracassar.net/research.html [accessed June 2012].
Cerra, V., and S. C. Saxena. 2008. “Growth Dynamics: The Myth of Economic Recovery.”
Czaika, M., and K. Kis-Katos. 2009. “Civil Conflict and Displacement: Village-Level
Determinants of Forced Migration in Aceh.” Journal of Peace Research 46, no. 3: 399–418.
Davis, D., and D. Weinstein. 2002. “Bones, Bombs, and Break Points: The Geography of
Economic Activity.” American Economic Review 92, no. 5: 1269–89.
Deininger, K. 2003. “Causes and Consequences of Civil Strife: Micro-Level Evidence from
Uganda.” Oxford Economic Papers 55, no. 4: 579–606.
de Walque, D. 2004. “The Long-Term Legacy of the Khmer Rouge Period in Cambodia.” Policy
Research Working Paper No. 3446. Washington, DC: World Bank.
Doctors Without Borders. 2006. “Living with Fear: Colombia’s Cycle of Violence.”
Amsterdam: Doctors Without Borders.
Doctors Without Borders. 2010. “Three Times Victims: Victims of Violence, Silence and
Neglect.” Amsterdam: Doctors Without Borders.
Duque, V. 2013. “The Hidden Costs and Lasting Legacies of Violence on Education: Evidence
from Colombia.” Presented at the Population Association of America, 2013 Annual
Meeting Program.
Engel, S., and A. M. Ibáñez. 2007. “Displacement Due to Violence in Colombia: A Household
Level Analysis.” Economic Development and Cultural Change 55, no. 2: 335–65.
Fafchamps, M., C. Udry, and K. Czukas. 1998. “Drought and Saving in West Africa: Are
Livestock a Butter Stock?” Journal of Development Economics 55, no. 2: 273–305.
Fernández, M., A. M. Ibáñez, and X. Peña. 2014. “Adjusting the Labor Supply to Mitigate Violent
Shocks: Evidence from Rural Colombia.” Journal of Development Studies 50, no. 8: 1135–55.
Goenjian, A., A. Steinberg, L. Najarian, L. Fairbanks, M. Tashjian, and R. Pynoos. 2000.
“Prospective Study of Posttraumatic Stress, Anxiety, and Depressive Reactions after
Earthquake and Political Violence.” American Journal of Psychiatry 157, no. 6: 911–16.
Gottschang, T. R. 1987. “Economic Change, Natural Disasters, and Migration: The Historical
Case of Manchuria.” Economic Development and Cultural Change 35, no. 3: 461–90.
Grun, R. E. 2008. “Household Investment Under Violence—The Colombian Case.” Policy
Research Working Papers 4713. Washington, DC: World Bank.
Ibáñez, A. M. 2009. “Forced Displacement in Colombia: Magnitude and Causes.” Economics of
Peace and Security Journal 4, no. 1: 48–54.
Ibáñez, A. M. 2014. “Growth in Forced Displacement: Cross-Country, Sub-National and
Household Evidence on Potential Determinants.” In R. E. B. Lucas, ed., International
Handbook on Migration and Economic Development. Cheltenham, UK: Elgar, 350–87.
Ibáñez, A. M., and A. Moya. 2007 “¿Cómo deteriora el desplazamiento forzado el bienestar de
los hogares desplazados: Análisis y determinantes del bienestar en los municipios de recep-
ción.” Coyuntura Social 37: 29–62.
Ibáñez, A. M., and A. Moya. 2010a. “Do Conflicts Create Poverty Traps? Asset Losses and
Recovery for Displaced Households in Colombia.” In R. Di Tella, S. Edwards, and E.
Schargrodsky, eds., The Economics of Crime. Chicago: University of Chicago Press,
137–72.
Ibáñez, A. M., and A. Moya. 2010b. “Vulnerability of Victims of Civil Conflict: Empirical
Evidence for the Displaced Population in Colombia.” World Development 38, no. 4: 647–63.
Ibáñez, A. M., and C. Vélez. 2008. “Civil Conflict and Forced Migration: The Micro-
Determinants and the Welfare Losses of Displacement in Colombia.” World Development
36, no. 4: 659–76.
Ibáñez, A. M., J. Muñoz, and P. Verwimp. 2013. “Abandoning Coffee under the Threat of
Violence and the Presence of Illicit Crops: Evidence from Colombia.” Households in
Conflict Network Working Paper No. 150. http://www.hicn.org/wordpress/wp-content/
uploads/2012/06/HiCN-WP-150.pdf [accessed June 2014].
Justino, P. 2009. “Poverty and Violent Conflict: A Micro-Level Perspective on the Causes and
Duration of Warfare.” Journal of Peace Research 46, no. 3: 315–33.
Justino, P. 2011. “War and Poverty.” In M. Garfinkel and S. Skarpedas, eds., Oxford Handbook of
Economics of Peace and Security. Oxford: Oxford University Press, 676–705.
Justino, P., and P. Verwimp. 2013. “Poverty Dynamics, Conflict and Convergence in Rwanda.”
Review of Income and Wealth 59, no. 1: 66–90.
Society 11, no. 3: 243–85.
Kalyvas, S. N. 2006. The Logic of Violence in Civil War. New York: Cambridge University Press.
Kalyvas, S. N., and M. Kocher. 2007. “How ‘Free’ Is Free Riding in Civil Wars? Violence,
Insurgency, and the Collective Action Problem.” World Politics 59, no. 2: 177–216.
Kazianga, H., and C. Udry. 2006. “Consumption Smoothing? Livestock, Insurance and Drought
in Rural Burkina Faso” Journal of Development Economics 79, no. 2: 413–46.
Kessler, R., A. Sonnega, E. Bromet, M. Hughes, and C. Nelson. 1995. “Post-traumatic Stress
Disorder in the National Comorbidity Survey.” Journal of American Medical Association 52,
no. 12: 1048–60.
Kondylis, F. 2008. “Agricultural Outputs and Conflict Displacement: Evidence from a Policy
Intervention in Rwanda.” Journal of Development Economics 96, no. 1: 1–15.
Kondylis, F. 2010. “Conflict Displacement and Labor Market Outcomes in Post-War Bosnia and
Herzegovina.” Journal of Development Economics 93, no. 2: 235–48.
Korf, B. 2004. “War, Livelihoods and Vulnerability in Sri Lanka.” Development and Change 35,
no. 2: 275–95.
Lindley, A. 2010. “Leaving Mogadishu: Towards a Sociology of Conflict-Related Mobility.”
Journal of Refugee Studies 23, no. 1: 2–21.
Lozano-Gracia, N., G. Piras, A. M. Ibáñez, and G. Hewings. 2010. “The Journey to
Safety: Conflict-Driven Migration Flows in Colombia.” International Regional Science Review
33, no. 2: 157–80.
Matijasevic, T. M., L. Velásquez, C. Villada, and M. Ramírez. 2007. “Moving Out of
Poverty: Understanding Freedom, Growth and Democracy from the Bottom-Up.” National
Synthesis Report, Colombia, Centro de Estudios Regionales Cafeteros y Empresariales,
Manizales, Colombia.
Miguel, E., and G. Roland. 2011. “The Long-Run Impact of Bombing Vietnam.” Journal of
Development Economics 96, no. 1: 1–15.
Mollica, R., G. Wyshak, and J. Lavelle. 1987. “The Psychosocial Impact of War Trauma and
Torture on Southeast Asian Refugees.” American Journal of Psychiatry 144, no. 12: 1567–72.
Monteiro, J., and R. Rocha. 2013. “Drug Battles and School Achievement: Evidence from Rio
de Janeiro’s Favelas.” CAF Working Paper No. 5. Available at https://www.dartmouth.
edu/~neudc2012/docs/paper_260.pdf.
Morrison, A. R., and R. A. May. 1994. “Escape from Terror: Violence and Migration in Post-
Revolutionary Guatemala.” Latin American Research Review 29, no. 2: 111–32.
Moya, A. 2015. “Violence, Psychological Trauma, and Induced Changes in Risk Attitudes in
Colombia.” Universidad de los Andes. https://economia.uniandes.edu.co/files/pro-
fesores/andres_moya/docs/ViolenceRiskAversion_AndresMoya(1).pdf [accessed
September 2015].
Moya, A., and M. Carter. 2014. “Violence and the Formation of Hopelessness and Pessimistic
Prospects of Upward Mobility in Colombia.” National Bureau of Economic Research
(NBER). Working Paper No. W20463. Cambridge, MA: NBER.
Ortiz, K. 2012. “Desplazamiento Forzoso en Colombia: Evidencia sobre el Impacto en el
Desarrollo Nutricional durante la Primera Infancia.” Universidad de Los Andes: Bogotá,
Colombia.
Rockmore, M. 2011. “The Cost of Fear: The Welfare Effects of the Risk of Violence in Northern
Uganda.” Households in Conflict Network Working Paper No. 109. http://www.hicn.org/
wordpress/wp-content/uploads/2012/06/wp1091.pdf [accessed June 2013].
Rockmore, M. 2012. “Living Within Conflicts: Risk of Violence and Livelihood Portfolios.”
Households in Conflict Working Paper No. 121. http://www.hicn.org/wordpress/wp-
content/uploads/2012/06/HiCN-WP-121.pdf [accessed October 2014].
Rodríguez, C., J. Cárdenas, J. Oviedo, and S. Villamizar. 2013. “La Discriminación Racial en
el Trabajo: Un Estudio Experimental en Bogotá.” Documentos Dejusticia 7. Bogotá,
Colombia: Observatorio de Discriminación Racial (Universidad de los Andes) and Centro
de Estudios de Derecho, Justicia y Sociedad, Dejusticia. http://dejusticia.org [accessed
April 2015].
Rodriguez, C., and F. Sanchez. 2012. “Armed Conflict Exposure, Human Capital Investments,
and Child Labor: Evidence from Colombia.” Defence and Peace Economics 23, no. 2: 161–84.
Shemyakina, O. 2011. “The Effect of Armed Conflict on Accumulation of Schooling: Results
from Tajikistan.” Journal of Development Studies 95, no. 2: 186–200.
Stanley, W. D. 1987. “Economic Migrants or Refugees from Violence? A Time-Series Analysis
of Salvadoran Migration to the United States.” Latin American Research Review 22, no.
1: 132–54.
Steele, A. 2009. “Seeking Safety: Avoiding Displacement and Choosing Destinations in Civil
Wars.” Journal of Peace Research 46, no. 3: 419–30.
Swee, E. L. 2009. “On War Intensity and Schooling Attainment: The Case of Bosnia and
Herzegovina.” Households in Conflict Network Working Paper No. 57. http://www.hicn.
org/wordpress/wp-content/uploads/2012/06/wp57.pdf [accessed October 2014].
Tough, P. 2012. How Children Succeed: Grit, Curiosity, and the Hidden Power of Character.
New York: Houghton Mifflin Harcourt.
[UNHCR] United Nations High Commissioner for Refugees. 2015. Global Trends 2014.
Geneva: UNCHR.
Uzonyi, G. 2014. “Unpacking the Effects of Genocide and Politicide on Forced Migration.”
Conflict Management and Peace Science 31, no. 3: 225–43.
Verpoorten, M. 2009. “Household Coping in War- and Peacetime: Cattle Sales in Rwanda,
1991–2001.” Journal of Development Economics 88, no. 1: 67–86.
Verwimp, P. 2005. “An Economic Profile of Peasant Perpetrators of Genocide: Micro-Level
Evidence for Rwanda.” Journal of Development Economics 77, no. 2: 297–323.
Verwimp, P., and T. Bundervoet. 2008. “Consumption Growth, Household Splits and Civil
War.” Households in Conflict Network Working Paper No. 48. http://www.hicn.org/
wordpress/wp-content/uploads/2012/06/wp48.pdf [accessed October 2014].
Vinck, P., P. N. Pham, E. Stover, and H. M. Weinstein. 2007. “Exposure to War Crimes and
Implications for Peace Building in Northern Uganda.” Journal of American Medical
Association 298, no. 5: 543–54.
Voors, M., E. Nillesen, P. Verwimp, E. Bulte, R. Lensik, and D. van Soest. 2012. “Does Conflict
Affect Attitudes? Results from Field Experiments in Burundi.” American Economic Review
102, no. 2: 941–64.
Williams, N. E. 2008. “Betting on Life and Livelihoods: The Role of Employment and Assets in
the Decision to Migrate during Armed Conflict.” Population Studies Center (PSC). PSC
Research Report No. 09-679. Ann Arbor: University of Michigan.
Williams, N. E. 2013. “How Community Organizations Moderate the Effect of Armed Conflict
on Migration in Nepal.” Population Studies 67, no. 3: 353–69.
Wood, R. M. 2010. “Rebel Capability and Strategic Violence against Civilians.” Journal of Peace
Research 47, no. 5: 601–14.
World Bank. 2011. World Development Report. Washington, DC: World Bank.
Yehuda, R. 2002. “Post-Traumatic Stress Disorder.” New England Journal of Medicine 346, no.
2: 108–14.
Zetter, R., A. Purdekova, and A. M. Ibáñez. 2013. “Violence, Conflict, and Mobility: A Micro
Level Analysis.” In T. Brück, P. Justino, and P. Verwimp, eds., A Micro-Level Perspective on the
Dynamics of Conflict, Violence, and Development. Oxford: Oxford University Press, 206–28.
12
Media Persuasion, Ethnic Hatred, and

Mass Violence
A Brief Overview of Recent Research Advances
M a r i a Pet rova a n d Dav i d Ya nagi z awa-Drot t
12.1. Introduction
The study of modern-day mass killing is arguably incomplete without an account
of whether and how mass persuasion plays a role. There are at least two key theo-
retical arguments for why mass persuasion is central to a deeper understanding
of how mass killing comes about. First, for perpetrators’ potential participation
in, and execution of, mass killing, the calculus of their costs and benefits will be
shaped by their beliefs, which in turn are formed, in part, by mass media exposure.
Direct persuasion due to mass media exposure may increase hatred of, and violence
against, perceived enemies. Just as marketing campaigns can convince people to
buy products or services of a certain brand, and just as political campaigns can
convince people that a certain political candidate is worth voting for, mass media,
in principle, can convince people to participate in mass violence. Second, indi-
rect persuasion may influence behavior even in the absence of direct exposure. For
example, if mass media are able to shift social norms on what is deemed socially
acceptable, or if they induce contagion—from those directly exposed to others—
then mass persuasion may indirectly play a significant role in mass killing.1
However, the empirical study of the effect of mass media on mass killing faces
some nontrivial challenges. One key difficulty is to convincingly identify causal
relationships, as a simple correlation between consumption of mass media and
violent behavior can arise for many reasons unrelated to a genuine violence-induc-
ing effect of media consumption. On the supply side, elites in control may target
persuasion efforts toward audiences predisposed to violence. On the demand
side, preexisting animosity toward certain groups in society may drive an interest
in mass media that spreads further hatred and propaganda toward those groups.
274
M e d i a Pe r s u a s i o n , E t h n i c H a t r e d , a n d M a s s V i o l e n c e 275
For example, after controlling for various demographic characteristics, data from
predominantly Muslim countries suggests that exposure to CNN is associated
with less anti-Americanism in Muslim countries, while exposure to Al-Jazeera
is associated with higher levels of reported anti-Americanism (Gentzkow and
Shapiro 2004). This may well reflect a true direct-persuasion effect of watching
Al-Jazeera. But alternatively, and perhaps just as likely, it may simply reflect that
people with a predisposition to anti-Americanism prefer to watch Al-Jazeera to
begin with. Moreover, since the data seldom contain all the necessary variables
one would need, the traditional “selection-on-observables” strategy—which, to
determine the outcomes under study, requires controls for all preexisting beliefs
and preferences jointly correlated with media consumption—is vulnerable to
various biases. Therefore, a statistically clean identification of causality that leaves
little room for alternative explanations is the key to a deeper understanding of the
role of mass persuasion in times of conflict.
The chapter proceeds as follows. Section 12.2 discusses recent methodological
advances related to the study of media persuasion, addressing the fundamental
empirical challenge just outlined.2 Section 12.3 discusses emerging economic lit-
erature about media persuasion and conflict, and section 12.4 concludes.
12.2. Empirical Challenges and Strategies

Studies of media persuasion date back to the 1940s, inspired by the seemingly
effective mass persuasion campaigns organized by Joseph Goebbels in Nazi
Germany in the 1930s. Perhaps surprisingly, these early studies, based on US data,
did not find any profound media effects (Berelson, Lazarsfeld, and McPhee 1954;
Lazarsfeld, Berelson, and Gaudet 1944). Concluding that media do not carry a
significant independent influence on people’s behavior, they do find that media
strengthen people’s predispositions. However, given the fundamental empirical
challenge already outlined, it is unclear whether the estimates in these studies can
be given a causal interpretation.
In contrast, recent economic studies of media effects are based on the idea of
finding some source of exogenous variation in media exposure, or media content,
to ensure that self-selection in media consumption or supply-side factors are not
biasing the results in any direction. In this regard, the benchmark test for study-
ing causal effects is through the conduct of field experiments, and the study of
media is no exception. For example, Gerber, Karlan, and Bergan (2009) offered
a ten-week-long subscription, free of charge, to either the Washington Post or the
Washington Times to randomly selected residents of the Washington, DC, area.
They found that, as compared to a control group, residents assigned to receive the
Washington Post were 8 percent more likely to vote for a Democratic candidate in
an upcoming gubernatorial election.
Field experimentation is a great method for the study of media persuasion, but
it is rarely used in practice. First, implementation difficulties preclude research-
ers from randomizing media exposure. Second, it is not ethical to conduct field
experiments to study media persuasion in real-world circumstances if there is
a risk that media might trigger ethnic violence or genocide. In these circum-
stances, researchers are forced to use quasi-experimental variation or natural
experiments to study the effect of media on the behavior of interest. Relevant
techniques include differences-in-differences studies, instrumental variable
techniques, or regression discontinuity approaches. In the subsections that fol-
low, we summarize a selection of recent papers that have employed such meth-
ods in order to establish credible causal effects.
12.2.1. The Differences-in-Differences Approach

The basic idea behind the differences-in-differences (DiD) approach is simple.
While treatment group individuals who are exposed to mass media may share
key demographic and political characteristics in ways that differ from those unex-
posed to mass media (the control group)—such as income levels or preexisting
political beliefs—causality can be tested for if the selection of subjects into treat-
ment or control groups is such that in the absence of exposure, the difference in
outcomes would have remained the same over time on average. This assumption
of parallel trends is the key to establishing causal effects, if any exist. An example
of this approach is DellaVigna and Kaplan (2007), who study the impact of Fox
News on voting behavior in the United States. A central empirical challenge in
their context was that conservative political leanings will drive the demand for
Fox News, making it difficult to disentangle whether exposure truly shifts politi-
cal views rather than simply reflecting preexisting views. To get around this prob-
lem, the authors based their strategy on the presumption that the initial rollout
of Fox News was mostly determined by supply constraints of local cable compa-
nies, rather than by underlying trends in political preferences. The results show
that Fox News significantly increased the vote share for George W. Bush with
an estimated 0.5 percentage points. Placebo tests further showed that the avail-
ability of Fox News in 2000 was not correlated with preexisting voting trends for
1992–1996, indicating that unobservable trends in political views are unlikely to
bias the results and that the parallel trends assumption was arguably reasonable.
12.2.2. Instrumental Variables Approaches

and the Irregular Terrain Model
The basic idea behind the instrumental variables approach is that if a statisti-
cian can find one or more factors that drive media exposure but do not directly
affect the outcome under study through any other channel, then a clean, causal
estimate of the impact of media exposure is feasible. For example, within the con-
text of a New Deal relief program implemented during the expansion period of
AM radio in the United States, Strömberg (2004) studies whether the amount of
federal redistributive spending via the Federal Emergency Relief Administration
(FERA) increased when a larger share of voters listened to radio—and thus were
better informed about the existence of the federal program.
To deal with the problem that unobserved heterogeneity in demand-side fac-
tors (e.g., a general interest in politics) could jointly explain why some counties that
received more funds are also those counties where radio listening happened to
be popular, Strömberg exploits the fact that radio signal quality—and thus radio
ownership—depended on geographic determinants of radio propagation, in par-
ticular ground conductivity and the proportion of woodland. Strömberg’s idea was
that since these geographic factors were unlikely to directly influence relief spend-
ing, they can be used as instrumental variables to establish causality. His main
result shows that media indeed had a positive impact on spending, with a 1 percent
increase in radio penetration in counties leading to 0.61 percent higher per-capita
FERA spending.
A related approach, and one that has grown in popularity, is to measure the
signal strength of radio or television directly, using the so-called Irregular Terrain
Model (ITM) or some other algorithm for electromagnetic propagation. To our
knowledge, Olken (2009) was the first to employ this strategy in his study of the
impact of television and radio on social capital in Indonesia. It is based on a simple
idea. Consider two villages: for one of them a hill blocks the line of sight (the
signal line) between a transmitter and a receiver; for the other, there is no such
hill and the transmission is unimpeded by geography. Except for signal availabil-
ity, these villages could be similar in every other respect. The presence of the hill
between transmitter and receiver introduces a quasi-random source of exogenous
variation in media availability. This statistical approach is suitable in countries or
regions with significant topographic variation and, when feasible, is a powerful
approach for identifying causal effects. Olken (2009) concludes that the avail-
ability of TV and radio indeed leads to less participation in social life and lower
self-reported trust, but that it does not have any impact on the local quality of
governance or corruption.
Enikolopov, Petrova, and Zhuravskaya (2011) also use ITM and certain idio-
syncrasies of media resource allocation to study the effect of the presence of an
independent TV channel on voting for Unity, the party that helped Vladimir
Putin come to power in Russia. Specifically, they use the fact that prior to the
1999 Russian parliamentary elections approximately two-thirds of the popula-
tion had access to only one independent TV channel. Put differently, in 1999 two-
thirds of the population had access to both points of view (for and against Putin’s
party). But the remaining one-third had access only to one-sided media messages.
The authors conclude that the independent TV channel increased the combined
vote share for opposition parties by 6.3 percentage points and decreased the vote
share for Putin’s party by 8.9 percentage points. The findings were confirmed
both by aggregate-level studies and individual-level surveys. Overall, Enikolopov,
Petrova, and Zhuravskaya (2011) suggest that media effects in immature democ-
racies can be larger than those in mature democracies.
12.2.3. Identifying Effects of Specific Media Content

A completely different approach exploits the fact that media outlets operate under
constraints of limited space and time to deliver content that is deemed newswor-
thy. Whether a piece of news is reported or not—or is given prominent cover-
age—will depend, in part, on whether or not there is other newsworthy material.
For example, Eisensee and Strömberg (2007) investigate whether media cover-
age of non-US natural disasters increases the likelihood that US government
disaster relief funds are disbursed. The empirical challenge in this context was
that some disasters are inherently more likely to receive disaster relief for rea-
sons other than media coverage—depending for example on their severity or the
needs of the potential recipients—and these factors also drive whether the event
is deemed newsworthy. To get around this problem, Eisensee and Strömberg
(2007) exploit the timing of natural disasters, specifically whether or not they
occurred during a major sports event (e.g., the Olympic Games), in which case
news about the disaster tended to be crowded out. Since the timing and sever-
ity of natural disasters are unrelated to major sports events, news coverage is
arguably as good as randomly assigned. With this creative empirical approach,
the authors show that when there is less media coverage of natural disasters,
substantially less humanitarian aid is allocated as a result. A similar approach
was used by Durante and Zhuravskaya (2015) to show that Israeli attacks in the
Israeli-Palestinian conflict are more likely to happen when voters in the United
States are interested in other news (e.g., the Olympics or natural disasters). The
results in the latter paper suggest that international media attention can explain
the dynamics of conflict, at least in part.
An alternative approach to the study of media effects of particular content
is used in Snyder and Strömberg (2010). The starting point for their study was
the observation that political and media markets sometimes coincide, but in
a “jagged” way. Specifically, their key variable of interest is overlap—or con-
gruence—between media markets and US congressional districts, with the key
observation that congressional districts change over time because of redistrict-
ing. Thus, the local press covers “its” US House representatives less intensely
in some districts than in others for reasons unrelated to other determinants
of information, views, or political behavior. With this approach, Snyder and
Strömberg’s (2010) striking results show that a low degree of congruence (i.e., a
low degree of overlap between media and political markets) causes lower levels
of political knowledge in the population and lower turnout in local elections,

and that politicians representing these communities work less for their con-
stituencies. They also find that policy is affected: congressional districts with a
larger share of population covered by out-of-state media markets have less fed-
eral spending per capita.
12.3. Media Persuasion in Conflict Environments:

Empirical Evidence
The examples of empirical strategies mentioned in section 12.2 highlight how
recent methodological advances have deepened our understanding of the role
of media and media persuasion. However, these examples mostly cover noncon-
flict environments, and mostly in relatively advanced democracies. Much less is
known as to whether mass media is a powerful tool in situations of intrastate and
interstate mass violence, and how political elites may exploit persuasion methods
to achieve political goals. Little is known, too, about mechanisms, for example in
regard to the role of a population’s political predispositions or how social interac-
tion comes into play. In what follows we describe some recent empirical evidence
toward this end.
12.3.1. Persuasion as a Precursor to Hatred and Mass Violence

Yanagizawa-Drott (2014) studies the impact of propaganda during the 1994
Rwandan genocide, where an estimated 800,000 civilians—primarily of the Tutsi
minority—were killed during a period lasting little more than three months. He
investigates the government-backed radio station Radio Télévision Libre des
Mille Collines (RTLM) that led propaganda efforts to spread hate against the
Tutsi minority population, encouraging—even mandating—the Hutu majority
population to kill Tutsi.
Hutu extremists set up the radio station less than one year before the genocide
broke out in April 1994. (President Habyarimana was assassinated by unknown
forces on April 6 that year.) Using a talk-show format, mixed with contemporary
music, RTLM quickly became very popular. Shortly after the genocide was under
way, Hutu extremists managed to seize power, and the radio station became the
main mass-media tool for delivering messages to the population. Most impor-
tantly, RTLM called for the extermination of the Tutsi ethnic group, claiming
that preemptive violence against it was necessary for “self-defense.” Based on
its inflammatory content, it is clear that a key goal was to induce ethnic hatred
and violence against Tutsi. In fact, after the genocide, the International Criminal
Tribunal for Rwanda convicted the station founders for having instigated
genocide.
The station used propaganda techniques familiar from other contexts, spread-
ing fear and using dehumanizing language to describe Tutsi, often referring to
them as “cockroaches.” It painted a picture of Tutsi as constituting a political threat
against Hutu, and that all Tutsi should be considered part of a conspiracy against
Hutu. The political and institutional backdrop arguably played a role in shaping
the political messages and their perceived credibility. Tutsi had dominated politi-
cal life in precolonial times, ethnic clashes had occurred multiple times after inde-
pendence, and the country had recently experienced a civil war from 1990 to 1993
that had begun after a Tutsi rebel group led by Paul Kagame invaded the country
from Uganda (see c hapter 15 in this volume). Put simply, RTLM’s message to the
Hutu population was “kill or be killed.” Importantly, this message carried a dual
meaning. First, it called for preemptive violence against Tutsi as a matter of self-
defense. Killing Tutsi was an imperative. Second, and perhaps just as important,
it made clear that dissent carried significant risks and Hutu themselves would be
considered traitors if they did not partake in the efforts to exterminate the Tutsi
minority. This risk was real, as government forces and Hutu extremists did in fact
kill thousands of moderate Hutu during the genocide.
In his paper, Yanagizawa-Drott (2014) hypothesizes that listening to the sta-
tion could have affected violence via two broad mechanisms, direct and indirect
persuasion. First, direct persuasion means that some marginal listeners could have
been convinced that participation in attacks on Tutsi was preferable to nonpar-
ticipation. This mechanism is plausible given that the broadcasts contained not
only strong anti-Tutsi rhetoric that may have increased hatred but also informa-
tion about relevant tradeoffs: they made it clear that the government would not
punish participation in the killing of Tutsi citizens, but instead mandated such
behavior. Second, following a long tradition in the study of mass media, start-
ing with Lazarsfeld, Berelson, and Gaudet (1944) and Katz and Lazarfeld (1955),
social interactions could have played a crucial role. In particular, a direct persua-
sion effect could coexist or be reinforced with indirect persuasion. For example, one
would expect this to be the case if violence begets violence, leading to contagion.
(On identity, social groupings, and contagion, also see c hapters 21 and 22 in this
volume.) Toward this end, Yanagizawa-Drott (2014) notes a key element of radio
broadcasts: they are public. Everybody who listens knows that all the other listen-
ers receive the self-same messages. What everybody else is believed to think, and
the likelihood that they might support the violence and actively participate in it,
then is likely to shape any given listener’s perceived costs and benefits and will-
ingness to join (or not) in the attacks. Yanagizawa-Drott (2014) outlines a simple
model of strategic complementarities in violence where, in addition to any direct
persuasion effects, public broadcasts play a role in coordinating violence. (On
strategic complementarities, also see chapter 19 in this volume.) The model shows
that by functioning as a coordination device, propaganda can exhibit strong non-
linearities—scale effects—in the share of the population receiving the messages.
A given increase in listeners will have miniscule effects when a small share of the
population listens, but large effects arise once a critical mass do.
To test the hypothesis that the broadcasts fueled violence, Yanagizawa-Drott
(2014) used village-level data on participation in violence based on prosecutions
in local courts, together with measures of radio reception in villages using a ver-
sion of the ITM propagation model. To establish causality, he exploits the geo-
graphic scattering of hills and valleys throughout Rwanda, making local variation
in radio reception essentially as good as if they had been randomly assigned. The
main results show that RTLM’s hate messages increased participation in the vio-
lence, both in terms of local militia violence as well as participation by ordinary
citizens. The effects are significantly weaker in areas with higher primary educa-
tion levels and literacy rates, indicating that investments in education may miti-
gate people’s susceptibility to inflammatory propaganda in times of conflict. One
explanation for these heterogeneous effects could be that education increases
interethnic tolerance, decreasing anti-Tutsi predispositions among the more edu-
cated Hutu. While the data do not allow the author to test this specific mecha-
nism directly, it is consistent with Adena et al. (2015), who do have data to test
for the role of predisposition in the context of Nazi propaganda (see paragraphs
that follow). Interestingly, for Rwanda the data provide evidence of strategic com-
plementarities in militia violence, as the results shows both scale effects within
villages and spillover effects to nearby villages. Reception in any given village
increased militia violence not only in that village but also in nearby villages. In
fact, the spillovers had a greater aggregate effect on militia violence than did the
direct effects of radio signal reception. This result indicates that one channel by
which mass media can amplify mass violence is through coordination and the
triggering of contagion. Thus, both direct and indirect persuasion seem to have
mattered. When assessing the countrywide impact of RTLM on the overall level
of violence, the estimates suggest that approximately 10 percent of participation
in the genocide can be attributed to RTLM broadcasts.
Adena et al. (2015) study the impact of German radio before and after the Nazi
Party’s electoral victory in 1933. By combining panel ITM-based data on radio
signal availability with data on large changes in radio content and various pla-
cebo tests for earlier outcomes, they document several findings. First, they show
that radio content changed in 1929, when radio stopped being apolitical, and in
1933, when radio content started to be heavily supportive of Nazism. This con-
trasted to limited Nazi access to radio before. For example, during 1925 to 1932,
representatives of the Nazi Party, the National Socialist German Workers' Party
(NSDAP), spoke only four times on radio, and Adolf Hitler was not given a floor.
In contrast, in February 1933 alone, he spoke sixteen times on radio, and the
total number of appearances of Nazi politicians during that month was twenty-
eight. Anti-Semitism was a recurrent theme in radio broadcasts in 1933 and later
on as well, from 1937 onward. The authors find that for 1930 to 1932, the places
with radio access were less likely to vote for the NSDAP or Nazi-supporting can-
didates. Gaining influence over, and eventual control of, radio thus was an impor-
tant aspect of Nazi Germany. Whether this result suggests that limiting extremist
speech may potentially prevent a popular dictator from gaining public support is
another matter (briefly discussed in section 12.4).
Second, Adena et al. (2015) find that after Hitler was appointed chancel-
lor, exposure to radio had a positive effect on different indicators of Nazi sup-
port, such as voting for Nazis, joining the Nazi Party, or instances of anti-Jewish
discrimination.
Third, the paper investigates how German radio affected anti-Jewish violence in
the late 1930s. Specifically, it looks at three types of outcomes: deportations of Jews,
anti-Jewish letters to Der Stürmer (a prominent Nazi newspaper), and synagogue
destruction during the Night of Broken Glass (Kristallnacht). The basic finding is
that places with radio access then experienced more deportations and anti-Jewish
letters. But an important heterogeneity in these results is that these effects crucially
depended upon people’s predisposition to Nazi messages. Specifically, the measured
effects are stronger in locations in which people were historically predisposed
toward welcoming Nazi messages. Predisposition is measured by historical anti-
Semitism since the fourteenth century, by early (1924) votes for nationalistic par-
ties, or by historical land inequality. The paper also demonstrates that propaganda
could backfire in places with higher levels of tolerance (that is, without the history
of Jewish pogroms in the fourteenth century or with historically low degrees of
inequality) and with less anti-Jewish violence in locations with radio access but
there was negative predisposition toward Nazi messages.
In sum, there are two main lessons from the Adena et al. (2015) paper. First,
in a politically volatile environment, who controls mass media is important, as
is whether there are restrictions on extremist speech in place. Second, the per-
suasive power of radio propaganda depends on people’s predispositions toward
media messages—receptive or not. It appears that effective media campaigns
need already fertile ground for media persuasion to work. Media perhaps amplify
feelings more than they generate them in the first place.
Finally, the evidence by Yanagizawa-Drott (2014) and by Adena et al. (2015)
may also be relevant for policy debates on placing restrictions on mass media
and on how to prevent mass atrocities such as genocides. Restrictions placed
on mass media are often pitched against the values of freedom of speech and
expression. External intervention, and the responsibility by the international
community to protect civilians from repression by their own governments, is
commonly pitched against the respect for state sovereignty. The international
debate during the Rwandan genocide is illustrative in this regard. Leading up to
the genocide, the United Nations force commander for the peacekeeping inter-
vention had urged the international community to jam RTLM signals. He had
repeatedly asked for the capability to jam RTLM, but the request was denied.
Arguments against the measure were that it would violate Rwanda’s state sov-
ereignty and impinge on the fundamental human rights to free speech and free
press. Also, the United States government had estimated that jamming the sta-
tion would be costly, approximately USD four million in total. Since, at the time,
no robust historical evidence existed that such broadcasts could cause hatred
and violence, the monetary, legal, and political costs of jamming RTLM were not
counterbalanced by any clear benefits. The evidence by Yanagizawa-Drott (2014)
and by Adena et al. (2015) arguably tilts this policy calculus.
12.3.2. Postconflict Persuasion

DellaVigna et al. (2014) use a combination of methods to study the impact of foreign
radio on postwar nationalism and reconciliation: How do mass media affect post-
war beliefs and behavior? Their case is the Serbo-Croatian war, the deadliest military
conflict in Europe since World War II. Its atrocities against civilians are character-
ized as genocide by Easterly, Gatti, and Kurlat (2006). In the 2000s, Serbian public
radio still carried nationalistic, anti-Croatian content. The study by DellaVigna et al.
(2014) establishes several facts. First, according to survey data, many ethnic Croats
listened to Serbian radio despite it being hostile to them; second, in places where
Serbian radio was available in the 2000s, people were more likely to vote for extreme
nationalist parties and more likely to draw nationalistic graffiti; and, third, Croatian
subjects in a field-based laboratory experiment exhibited more anti-Serbian senti-
ment after listening to ten minutes of Serbian radio. The laboratory experiment in
particular sheds light on the mechanism, as even neutral (not nationalistic) Serbian
radio increased anti-Serbian attitudes of the subjects, and, probably, Serbian radio
served as a reminder about the war and the atrocities just a decade before.
This evidence indicates that radio aimed at inducing ethnic animosity and vio-
lence can be exploited by political elites. 3 And it begs the question of whether mass
media can be used for more benevolent goals, such as the fostering of interethnic
tolerance, trust, or cooperative behavior. Paluck (2009) and Paluck and Green
(2009) study the impact of a reconciliation radio soap opera in postgenocide
Rwanda that featured messages about reducing intergroup prejudice, violence,
and trauma. Over the course of one year, communities were randomly assigned to
receive broadcasts of either the reconciliation soap opera or a control soap opera
(aimed at health education), where group listening took place in public spaces.
Results from this experiment show that the reconciliation program had a positive
effect on social norms, such as a higher acceptance of intermarriage across ethnic
and religious groups; that it increased listeners’ willingness to express dissent; and
that communal problem-solving improved. Although it remains an open question
whether content like these soap operas can reduce ethnic violence, at the very
least these results indicate that persuasion can work to influence mechanisms that
presumably affect the ability of political elites to orchestrate mass violence.
12.4. Conclusions
This chapter provides an overview, if brief, of recent empirical literature on the
role of mass media in influencing political outcomes, with a focus on studies of
mass media effects before, during, and after mass killings. The chapter does not
aim to provide an overview of all recent literature on mass media effects, but
rather to selectively showcase techniques used by economists in recent years to
identify such effects in general, if any, and to illustrate more specifically how these
techniques are used in studies of mass media effects on mass violence and post-
war reconciliation.
The empirical bottom line is that mass media can play an important role in the
organization of mass killing, but this is not the only lesson. When propaganda is
aligned with population predispositions, persuasion appears especially effective.
Both Yanagizawa-Drott (2014) and Adena et al. (2015) suggest that community-
level beliefs and behaviors are important for propaganda affecting individual
participation in genocide. But Adena et al. (2015) also find that propaganda can
backfire. Beyond direct persuasion, the evidence from Rwanda is consistent with
data showing that the mass media functioned as a coordination device for vio-
lence and that spillover effects from social interactions were nontrivial. Overall,
this emerging literature points to common beliefs in society—preexisting or
induced by media exposure—being important prerequisites to mass killing. That
said, more evidence is clearly desirable to establish the generalizability of these
results and to arrive at a deeper understanding of the specific mechanisms driv-
ing them.
This chapter does not discuss new forms of media that spread information and
beliefs such as social media and cell phones/smart phones. It is an open ques-
tion, at least statistically, whether governments or rebel groups can use these new
technologies to ignite, fuel, and coordinate violent behavior. Some suggestive evi-
dence exists, however. Pierskalla and Hollenbach (2013) use new disaggregated
and geocoded data on violent conflict and find evidence consistent with cell-
phone coverage increasing violence in Africa. Two recent papers on the effects of
social media on coordinating political protests also suggest a coordination role
for organized violence. Acemoglu, Hassan, and Tahoun (2015) show that Twitter
activity predicted spikes in participation in the Tahrir Square protests in Egypt in
2011; and Enikolopov, Makarin, and Petrova (2015) find that social media pen-
etration led to a greater incidence of protests, and higher protest participation, in
Russia in 2011–2012. More research is clearly desirable here as well.
What policy conclusions, if any, can one draw from this sophisticated statisti-
cal research? Probably the main message is that simply turning a blind eye toward
autocratic regimes that freely disseminate inflammatory messages targeting eth-
nic minorities can and does allow them to fuel hatred and violence, ultimately
leading to significant human suffering. It is good to be able to empirically dem-

onstrate the existence of such effects. There is also some suggestive evidence that
restrictions on extremist speech may possibly help to prevent dictators from com-
ing to power in the first place. Such evidence strengthens arguments for preven-
tion and external intervention, at least at the margin. Finally, to our knowledge
there is no available evidence on whether and how mass media can be used benev-
olently—as a tool to prevent mass killing. Mass media effects may or may not act
symmetrically. Research shedding light on this issue is highly desirable.
Notes
1. The hypothesis that social interaction provides an indirect channel for persuasion effects on
behavior dates back to the two-step flow communication model by Lazarsfeld, Berelson, and
Gaudet (1944) and Katz and Lazarsfeld (1955).
2. The aim of this chapter is not to provide a complete survey of the literature. We focus on refer-
ences within economics. For a more extensive survey, see DellaVigna and Gentzkow (2010)
or Prat and Strömberg (2013).
3. Fergusson, Vargas, and Vela (2013) use a regression discontinuity design to study the impact
of newspaper scandals on electoral coercion in paramilitary-controlled areas in Colombia.
Specifically, they found that electoral coercion was more likely to happen after candidates,
sympathetic to paramilitary groups, were threatened to lose elections after corruption scan-
dals about their behavior got into the news. While their results do not speak directly to media
effects during and after mass killings, they highlight an important potential mechanism to
trigger violent events.
References
Acemoglu, D., T. Hassan, and A. Tahoun. 2015. “The Power of the Street: Evidence from Egypt’s
Arab Spring.” NBER Working Paper No. 20665. Cambridge, MA: National Bureau of
Economic Research.
Adena, M., R. Enikolopov, M. Petrova, V. Santarosa, and E. Zhuravskaya. 2015. “Radio and
the Rise of the Nazis in Pre-War Germany.” Quarterly Journal of Economics 130, no. 4:
1885–939.
Berelson, B., P. F. Lazarsfeld, and W. N. McPhee. 1954. Voting: A Study of Opinion Formation in a
Presidential Campaign. Chicago: University of Chicago Press.
DellaVigna, S., and E. Kaplan. 2007. “The Fox News Effect: Media Bias and Voting.” Quarterly
Journal of Economics 122, no. 3: 807–60.
DellaVigna, S., R. Enikolopov, V. Mironova, M. Petrova, and E. Zhuravskaya. 2014. “Cross-
Border Effects of Foreign Media: Serbian Radio and Nationalism in Croatia.” American
Economic Journal: Applied Economics 6, no. 3: 103–32.
DellaVigna, S., and M. Gentzkow. 2010. “Persuasion: Empirical Evidence.” Annual Review of
Durante, R., and E. Zhuravskaya. 2015. “Attack When the World Is Not Watching? International
Media and the Israeli-Palestinian Conflict.” Working paper. http://papers.ssrn.com/sol3/
papers.cfm?abstract_id=2566741 [accessed May 11, 2015].
Eisensee, T., and D. Strömberg. 2007. “News Droughts, News Floods, and U.S. Disaster Relief.”
Enikolopov, R., M. Petrova, and E. Zhuravskaya. 2011. “Media and Political Persuasion: Evidence
from Russia.” American Economic Review 111, no. 7: 3253–85.
Enikolopov, R., A. Makarin, and M. Petrova. 2015. “Social Media and Protest
Participation: Evidence from Russia.” Working paper. www.ieb.ub.edu/files/Enikolopov.
pdf [accessed May 11, 2015].
Fergusson, L., J. F. Vargas, and M. A. Vela. 2013. “Sunlight Disinfects? Free Media in Weak
Democracies.” CEDE Working Paper 2013–14. Universidad de los Andes, Bogotá,
Colombia. https://economia.uniandes.edu.co/component/booklibrary/478/view/46/
Documentos%20CEDE/54/sunlight-disinfects-free-media-in-weak-democracies
[accessed May 11, 2015].
Gentzkow, M., and J. M. Shapiro. 2004. “Media, Education and Anti-Americanism in the
Muslim World.” Journal of Economic Perspectives 18, no. 3: 117–33.
Gerber, A. S., D. Karlan, and D. Bergan. 2009. “Does the Media Matter? A Field Experiment
Measuring the Effect of Newspapers on Voting Behavior and Political Opinions.” American
Economic Journal: Applied Economics 1, no. 2: 35–52.
Katz, E., and P. F. Lazarsfeld. 1955. Personal Influence: The Part Played by People in the Flow of Mass
Communication. New York: Free Press.
Lazarsfeld, P. F., B. Berelson, and H. Gaudet. 1944. The People’s Choice: How the Voter Makes Up
His Mind in a Presidential Campaign. New York: Columbia University Press.
Olken, B. 2009. “Do TV And Radio Destroy Social Capital? Evidence from Indonesian Villages.”
American Economic Journal: Applied Economics 1, no. 4: 1–33.
Paluck, E. L. 2009. “Reducing Intergroup Prejudice and Conflict Using the Media: A Field
Experiment in Rwanda.” Journal of Personality and Social Psychology 96, no. 3: 574–87.
Paluck, E. L., and D. P. Green. 2009. “Deference, Dissent, and Dispute Resolution: An
Experimental Intervention Using Mass Media to Change Norms and Behavior in Rwanda.”
American Political Science Review 103, no. 4: 622–44.
Pierskalla, J. H., and F. M. Hollenbach. 2013. “Technology and Collective Action: The Effect of
Cell Phone Coverage on Political Violence in Africa.” American Political Science Review 107,
no. 2: 207–24.
Prat, A., and D. Strömberg. 2013. “The Political Economy of Mass Media.” In D. Acemoglu,
M. Arellano, and E. Dekel, eds., Advances in Economics and Econometrics. Vol. 2, Applied
Economics: Tenth World Congress. New York: Cambridge University Press, 135–87.
Snyder, J., and D. Strömberg. 2010. “Press Coverage and Political Accountability.” Journal of
Strömberg, D. 2004. “Radio’s Impact on Public Spending.” Quarterly Journal of Economics 119,
no. 1: 189–221.
Yanagizawa-Drott, D. 2014. “Propaganda and Conflict: Evidence from the Rwandan Genocide.”
PA R T T H R E E
CASE STUDIES I
13
“For Being Aboriginal”

Economic Perspectives on Pre-H olocaust Genocides
J u rge n Br au e r a n d R au l C a ruso
13.1. Introduction
The word “genocide” was coined in print in 1944 (Lemkin 1944) and codified in
international law in the United Nations (UN) Genocide Convention of December
9, 1948 (coming into force on January 12, 1951). The Convention defines genocide
as “any of the following acts committed with the intent to destroy, in whole or in
part, a national, ethnical, racial, or religious group, as such: (a) Killing members
of the group; (b) Causing serious bodily or mental harm to members of the group;
(c) Deliberately inflicting on the group conditions of life calculated to bring about
its physical destruction in whole or in part; (d) Imposing measures intended to
prevent births within the group; (e) Forcibly transferring children of the group to
another group” (United Nations 1951).
Samuel Totten and William Parsons (2013) point out in the very title of their
book, Centuries of Genocide, that once the idea of a new concept is born, one rec-
ognizes that it may apply to eras preceding its invention and naming. Yet, much
of the scholarly literature on mass atrocities, such as genocide, concerns itself
with the Holocaust and post–World War II instances. Indeed, the book in which
this chapter appears reflects this scholarly preference in that most of its mass kill-
ing cases and applications are of post-1948 vintage (e.g., Colombia, Indonesia,
Mexico, Pakistan, Rwanda, and Vietnam). In contrast, this chapter applies
economic concepts to pre-Holocaust cases of genocide, in fact to the five pre-
Holocaust cases included in the Totten and Parsons book (4th ed., 2013). These
are the cases of the Yana people in California, the Aborigines peoples of Australia,
the Herero and Nama peoples in then-German South-West Africa, Armenians
in the Ottoman Empire, and Ukrainians under Stalin. Specifically, two general
economic schemata recently proposed in the genocide literature are applied to
the pre-Holocaust case material in Totten and Parsons (2013). The first schema,
289
290 Case Studies I
by Anderton (2014), lays out six “interdependencies” (116) said to link econom-
ics and genocide. The second, by Brauer and Anderton (2014), discusses how
constrained optimization theory may apply to the study of genocide.
Anderton’s “interdependencies” schema states, first, that conflict, including
genocidal conflict, is a deliberate choice made against feasible alternatives. The
study of the conditions under which particular choices are made is a staple of
economics research. Second, prevailing economic conditions can affect the risk,
intensity, contagion, termination, and recurrence of genocidal conflict. Third,
genocidal conflict, once started, affects the performance of an economy, such as
in diverting and destroying economic resources, in disrupting trade, in displac-
ing people and disrupting the labor market, and in diminishing prospects for
postviolence recovery and redevelopment. Fourth, genocidal conflict requires a
“business model,” that is, an organizational setup, whereby for example labor is
recruited and trained, weapons and other deadly materials are acquired, a com-
mand and control system is built, and logistics and supply chains are established.
These topics fall within such subfields as managerial economics, business eco-
nomics, and the economics of industrial organization. Fifth, genocidal conflict
can be a mode of wealth appropriation. Examples include the accumulation of
territories that contain valuable natural resources, the capture of capital goods
and financial assets, and even the capture of cultural goods (e.g., through artifact
looting) and of victimized peoples themselves (e.g., forced labor). Finally, sixth,
genocidal violence deprives victims of the security of their property and person.
But security is a fundamental economic good without which sustained economic
productivity and prosperity cannot be expected.
Section 13.2 of this chapter briefly puts each of these six interdependencies
to a test with examples drawn from the pre-Holocaust cases narrated in Totten
and Parsons (2013). The hope is twofold: first, that genocide scholars may per-
haps more systematically perceive, recognize, and study the economic “field”
and “techniques” that Raphael Lemkin (1944) himself spelled out; and, second,
that economists and economic historians may feel encouraged to delve more
deeply into specific pre-Holocaust cases and to elaborate on them through an
economic lens.1
The second schema, by Brauer and Anderton (2014), takes one economic
theory—constrained optimization theory—and illustratively applies it in sec-
tion 13.3 to pre-Holocaust examples. This is a detailed elaboration of Anderton’s
(2014) first interdependency—genocide as a deliberate choice made against fea-
sible nongenocidal alternatives. The general idea is that perpetrators, victims, and
third parties (i.e., internal and external bystanders) each have specific objectives
(respectively, to victimize, to escape victimization, and potential intervention),
and that the feasible alternatives from which each chooses one specific course of
action are constrained by a limited pool of resources available to pay the cost of
the action (victimization, escape, or intervention). For perpetrators, victims, and
“For Being Aboriginal ” 291
third parties, respectively, Appendix Tables 13.A1, 13.A2, and 13.A3, are created
from the text material in Brauer and Anderton (2014). They lay out in some detail
who forms preferences (objectives) for genocide (or to escape from or to intervene
in genocide); what these preferences entail; and why, when, where, and how they
are formed. The tables also lay out detailed, but not case-specific, ideas in regard
to the monetary and nonmonetary resources available to pay costs of perpetrat-
ing, escaping, or intervening in genocide. They also make points in regard to fac-
tors that may enhance the productivity of conducting, escaping, or intervening
in genocide.
Again, the hope is twofold. We hope that genocide scholars will appreciate
the illustrative rereading of pre-Holocaust cases in light of the idea that spe-
cific genocidal choices made are directly shaped by resource and cost constraints
and that, had the constraints been more, or less, binding (for perpetrators, vic-
tims, or potential interveners), a nongenocidal choice may well have been made.
(This does not deny that noneconomic factors have their own importance; but
primarily they affect the formation of preferences rather than their execution.)
Rereading genocide history through the lens of optimization, given monetary and
nonmonetary resource and cost constraints, may help to more deeply or, at any
rate, differently understand cases of genocide. Likewise, for economists, the illus-
trative applications of constrained optimization theory may encourage them to
delve more deeply into particular aspects of genocidal decision-making, not just
in post-Holocaust cases as the literature already does but also in pre-Holocaust
cases, and thus to deepen the theoretical and empirical basis of applying (one)
economic theory to a larger set of cases of genocide (or of other mass atrocities).
Section 13.4 concludes the chapter.
13.2. Interdependencies
Anderton’s (2014) “independencies” between economics and genocide are not
meant to follow any chronology. They simply categorize and exemplify economic
aspects of genocide.
13.2.1. Genocidal Conflict as a Deliberate Choice

Genocide rarely appears as a fully formed choice. Typically, it is preceded by a
number of signposts. In a famous briefing paper prepared for US government offi-
cials, Gregory Stanton (1996) writes of eight stages of genocide: “The first stages
precede later stages, but continue to operate throughout the genocidal process.
Each stage reinforces the others. A strategy to prevent genocide should attack each
stage, each process. The eight stages of genocide are classification, symboliza-
tion, [discrimination], dehumanization, organization, polarization, preparation,
292 Case Studies I
[persecution], extermination, and denial.”2 Thus, in time, the choice of genocide

is arrived at. For each of the cases of the Yana, Aborigines, Herero and Nama,
Armenians, and Ukrainians, the literature summarized in the Totten and Parsons
(2013) case collection leaves no doubt about genocide eventually having been
chosen by perpetrators as a deliberate objective as against feasible, nongenocidal
alternatives of dealing with unwanted out-g roups.
For example, the Yana people were a population of some 3,000 Native
Americans, living at the northern end of California’s Central Valley. Their exter-
mination was total in the sense that at the beginning of the twentieth century only
one Yana survivor was reported to be alive and the group’s reproduction therefore
was no longer possible (also see, e.g., c hapters 4 and 17 in this volume). According
to the case author, Ben Madley, in the wake of the California gold rush from the
late 1840s to the 1880s, the “motives driving immigrants to destroy the Yana
changed over time, as did the organization of their killing operations” (18). 3 In the
end, however, the initial colonizers as well as later immigrants wanted to establish
an “Indian-free environment” (24) to appropriate land-based natural resources
for “ranching, hunting, and mining” (22). On occasion, their purpose might have
appeared to be defensive and retaliatory because of Indian raids against settlers
but, in time, the goal became Yana elimination for its own sake. This was govern-
ment tolerated and government cofinanced. For instance, the state’s first civilian
governor, Peter H. Burnett, declared that “a war of extermination will continue
to be waged … until the Indian race becomes extinct” (20). One can hardly be
clearer about the deliberateness of the intent and the genocidal choice made. In
support, California’s legislators appropriated USD 500,000 and USD 600,000
dollars in 1851 and 1852, respectively, to fund state-sanctioned Indian-hunting
campaigns staffed by militia volunteers. In 1858–1859 colonizers bent on institu-
tionalized killings received support from then-Governor John B. Weller. Again, a
state-supported militia, led by Adjutant General William C. Kibbe, was created.
Money was raised repeatedly to hire men to literally hunt Indians. Indian hunters
were compensated with an amount of cash per Indian scalp, and Kibbe himself,
for example, claimed the sum of USD 69,486 for a single expedition (29). The
incentive structure proved important, and personal greed partly motivated and
reinforced the campaign against the Yana. “Ultimately,” writes Madley, “the drive
to destroy the Yana became an ideology of total annihilation” (44). The choice no
longer was merely to confiscate Indian-held resources or to enslave Indians. The
choice made was to kill to the point of group extermination. Section 13.3 elabo-
rates on the cost and resource constraints that shaped this, and other, genocide
choices. That genocide becomes a choice, and that the process of becoming entails
opportunities for prevention, no longer is in dispute among genocide scholars. But
the conditions that help shape the making of this choice, and what may or may not
have been the feasible alternatives, still are not well examined and u nderstood,
certainly not from a choice-t heoretic perspective.
13.2.2. Economic Conditions Affecting Genocidal Conflict

Between 1904 and 1907 imperial Germany waged a war against the Herero popu-
lation that had revolted against colonial rule in German South-West Africa (now
Namibia). This war was shaped by specific economic conditions that deprived
Herero of much of their resource base. Dominik Schaller writes (in Totten and
Parsons 2013) that, initially aimed solely against Herero, the war eventually
embroiled Nama people as well. The latter, after at first supporting the settlers,
began guerrilla operations against the German army because of a growing fear
that commander Lothar von Trotta’s racist and explicit extermination policy
would, in time, be applied to Nama as well as to Herero (90). By the end of the war
about 60,000 Herero and 10,000 Nama had died (90). Long before the German-
Herero war and subsequent genocide commenced, colonists had hoped for the
decisive decline of the local population as self-reliant actors so as to gain cheap
labor, but they “were well aware that their military power and colonial infrastruc-
ture had not been sufficient. Therefore, settlers were afraid that a major African
uprising could hamper the positive political and economic development in the
colony” (90–91). Uprisings did happen, with the one on January 12, 1904 gen-
erally cited as the beginning of war between Herero and about 5,000 German,
English, and Boer colonists (94).
At the turn of the century, Herero numbered about 80,000 people in Central
Namibia. Traditionally pastoralists, by the mid-1800s they had become the
region’s predominant cattle herders, breeders, and traders. The Nama, num-
bering about 20,000, lived mostly in South Namibia. Also a cattle people, they
traded mostly with the Cape Colony further to the south. Northern Namibia,
then not much accessed by Europeans, was inhabited by about 450,000 Ovambo
people who lived on fishing, agriculture, and trade with Ovambos to the north,
in Portuguese Angola. Schaller describes the indigenous societies as “strong and
independent and … not at all ready to give up their self-sustaining economies”
(93).4 However, a cattle disease (Rinderpest) arrived in 1897 and resulted in a
90 percent loss of cattle. While the settlers managed to inoculate their own herds,
for the indigenous people loss of cattle meant loss of nutrition. Subsequent enfee-
blement then made them vulnerable to malaria, of which about 10,000 Herero
died. Bereft of labor and cattle, many survivors found themselves forced to sell
their labor services to the settlers. Economically weakened tribal chiefs sold large
swaths of tribal land to the settlers; this was in addition to huge land holdings they
had previously sold in exchange for recognition as tribal leaders by settler govern-
ments (especially Germany).
The sudden cattle deprivation shifted the economic balance between coloniz-
ers and natives dramatically. The purpose of the Herero revolt was, in part, to
forestall full economic expropriation (dispossession of all land and property).
Once German policy became an extermination policy in a later phase of the
294 Case Studies I
war, natives’ objectives shifted to pure physical survival. Military strategy and
geography made things worse in that Herero could flee only into “the water-
less Omaheke desert where they had to face death from starvation and exhaus-
tion. … The paths through the desert to British Bechuanaland were known to
the Herero as traditional trade routes [but] the capacity of the water holes was not
sufficient to ensure the survival of all the refugees” (90–91). Thus, as Anderton
(2014) claims, in addition to other factors, economic conditions can affect the
onset, course, and termination of genocide.
13.2.3. Genocidal Conflict Affects the Economy

If economic conditions can affect the onset, conduct, and termination of geno-
cide, the converse holds as well: genocide affects the economy. The genocide of
Australia’s indigenous peoples—A borigines and Torres Strait Islanders—took
place approximately between the late 1700s and the 1980s, an unusually extensive
period of about two hundred years. Australia’s federal government did not apolo-
gize formally to Australia’s indigenous peoples until the year 2008. In terms of
the UN Genocide Convention, it is undisputed, writes chapter author Colin Tatz
in Totten and Parsons (2013), that an attempt was made to eliminate Aborigines
as distinct peoples. Many were killed outright; serious physical and mental harm
was inflicted with deliberation; conditions of life were calculated to destroy its
members; children of pure and mixed Aborigines descent were forcibly trans-
ferred; and measures were taken to prevent births within the groups. Conspiracy,
incitement, attempt, and complicity to commit genocide all were orchestrated as
a matter of government policy. Whereas Aborigines populations numbered on
the order of 500,000 people in the late 1700s, by 1911 their numbers had fallen
to between 30,000 to 80,000 people. (According to the Australian Bureau of
Statistics, the populations recovered to 670,000 by mid-2011.)
In the initial phase of the genocide—the first one hundred years or so—its
effect on the developing economy of the various Australian colonies5 was to clear
land for settlement: “The land was treated as a wasteland but for flora and fauna,
of which the ‘natives’ were deemed a part” (Tatz in Totten and Parsons 2013, 56).
Land was deemed terra nullius—belonging to no one and therefore free for the
taking, a doctrine not overturned until a High Court decision, in 1992, recog-
nized Aboriginal title to traditional lands. Policy with regard to “natives” was first
colony-and then state-based until 1967, when a nationwide referendum resulted
in the transfer of relevant powers to the federal government.
The effect of the genocide on Aborigines communities was opposite to that
on the settlers’ economy. As settlers expanded north and south of Botany Bay
(roughly, today’s Sydney) and inland to Central and then Western Australia,
Aborigines were herded into remote reservations, cattle stations, and church
missions, most “splendidly secluded … foolishly selected … [and] almost
inaccessible domains [that] were not Aboriginal choices or places of their natu-
ral habitat. … Quintessentially nomadic hunter-gatherers became sedentary
and stationary” (Tatz in Totten and Parsons 2013, 64). Conditions were con-
structed, such as absence of title to land, or any other legal protection, such that
their economy, numbers, and communities collapsed. Where “actual remunera-
tion [for work] was paid, these monies went into state-r un trust funds, much of
which disappeared;” nor could Aborigines workers “join essentially racist trade
unions” (ibid.). The examples can be multiplied and, as in the Yana and South-
West African cases, show an unambiguous bifurcated effect favoring the settlers’
economy while decimating that of the indigenous peoples.
13.2.4. The “Business Model” of Genocide

Between 1915 and 1923, in the final phase of the dissolution of the Ottoman
Empire, its Armenian population suffered genocide. The Empire successfully
organized and carried out “business-l ike” plans to destroy the Armenian popula-
tion. According to Rouben Adalian in Totten and Parsons (2013), the number
of victims is estimated at 1.5 million people. The Armenian minority in Turkey
constituted an anomaly: over the span of more than fourteen centuries it never
assimilated and consistently kept its Christian identity and culture. Throughout
the late 1800s, Armenians already had been the target of episodic massacres by
military and paramilitary forces. By 1915, Ottoman society had experienced a
steady decline of its powers and economy. Average gross domestic product (GDP)
per capita was less than one-third of the European average, and the economy
was mostly at subsistence levels in rural areas. The Empire, buffeted by Russian,
Austro-Hungarian, British, and other interests, did not succeed to keep pace and
modernize. Its imperial wings were clipped as its soldiers lost numerous battles
and territories in Southeastern Europe, Western Asia, and North Africa. The
resulting internal turmoil led Ottoman authorities under the initially somewhat
liberal and tolerant Committee of Union and Progress (CUP) to withdraw into
Turkish nationalism, militarization, Islamization, centralization, and consolida-
tion of power, at the expense of non-Muslim and non-Turkish populations.
Seeking to effect a far-reaching elimination of minorities, to guarantee itself
full control of territory, and to confiscate Armenian properties, the state needed
an organized way to go about its efforts; it needed a “business model” to guide
command and control, logistics and supply, labor recruitment and training, and
so on. Some details are provided in section 13.3 but, in brief, many agencies and
ministries of the Ottoman state were involved in the implementation of the geno-
cide. A secret agency, the “Special Organization,” committed to mass murder, was
created. A secret police of 40,000 people was built up. An extermination plan was
created that involved, in large part, forced deportations to Syria and Mesopotamia
that, by design, then reduced to a simple process. Escorted columns of deportees,
296 Case Studies I
mostly women, children, the elderly, and infirm, were decimated through rob-
bery, exposure, privation, starvation, or direct attack, on their way to, or in, the
Syrian desert. This did not happen spontaneously, but was the outcome of an
organizational effort. Illustration 4 in section 13.3 will expand on this, especially
in regard to how the availability of Ottoman resources helped shape the specific
form that this genocide took.
13.2.5. Genocide as Wealth Appropriation

Stalin’s manufactured famine in east-central Ukraine and the North Caucasus
in 1932–1933 resulted in perhaps five to seven million victims. At the time, the
Great Depression badly affected the economies of the entire world. James Mace,
author of the Ukraine chapter in Totten and Parsons (2013), writes that the USSR
“seized with unprecedented force and thoroughness the 1932 crop and food-stuffs
from the agricultural population” (157). Ukrainian farms were forcibly collectiv-
ized; food production was first requisitioned, then confiscated and siphoned off.
But wealth appropriation goes beyond Lemkin’s narrowly understood economic
“field,” such as “forced impoverishment, expelling people from businesses and
occupations, manipulations of trade and finance, [and] appropriation of assets
and enslavement” (Anderton 2014, 125). Wealth appropriation in a broader sense
refers to the taking of any and all resources at the victims’ disposal, economic or
otherwise, and thus covers all of Lemkin’s eight domains over which genocide
is carried out (the political, social, cultural, [narrowly] economic, biological,
physical, religious, and moral fields). For example, Ukrainian victims, mostly a
peasantry, had an “incomplete social structure, [lacked their] own ruling class,”
and had no independent political clout (158). As for others in imperial Russia,
Ukrainians’ social and cultural assets already had been severely challenged (158).
With the collapse of the empire, in 1917, Russia’s new leaders sought to unify the
country under the ideological banner of a progressive urban proletariat to be set
against rural, traditionalist interests. A new consciousness was to be established,
neither Russian nor Ukrainian, but based on a self-understanding as an inter-
national proletariat united against exploitative capitalists. Rural and urban pro-
letarians of the various nations of the USSR were to unite against land-owning
rentiers. In a way, all peoples, and the wealth inherent in their ethnic, linguis-
tic, cultural, and other forms of diversity, were to be appropriated and abolished
as separately recognizable entities and cultures, to be reborn as a new nation of
proletarian workers and peasants.
This notion of “national nihilism” was based, in part, on Rosa Luxemburg’s
writings, writes Mace (in Totten and Parsons 2013, 161), and on Joseph Stalin’s
own article on “Marxism and the National Question,” penned in 1912–1913,
in which he had criticized the idea of permitting any cultural-national auton-
omy to persist. But under Lenin’s initial leadership of the USSR, following the
October Revolution of 1917, a pragmatic course was chosen in that non-Russian

nations and their local mores were recognized and accommodated. Lenin’s New
Economic Policy (NEP) of 1921 thus permitted, for example, private rural mar-
kets to persist, one avenue by which to help integrate diverse populations into the
new Soviet workers’ and peasants’ state (159). In 1923, in part to deal with “con-
tinued national resistance” to integration, Lenin fostered “indigenization, which
attempted to give non-Russian Soviet regimes a veneer of national legitimacy by
promoting the spread of the local language and culture in the cities, recruiting
local people into the regime, ordering Russian officials to learn the local language,
and fostering a broad range of cultural activities” (159; also see 161). Following
Lenin’s death in 1924 and the subsequent power struggle, Stalin replaced Lenin’s
NEP in 1927 with a program of forced collectivization of agriculture and pur-
sued a push toward urban-based industrialization. This required an increasingly
simplified ideology—wherein a stubborn petty bourgeois class of landowners
(the kulaks) was said to exploit other types of farmworkers and was accused of
withholding food supplies from the remainder of the USSR—and combined with
Stalin’s definition of a nation: “A nation is a historically constituted, stable com-
munity of people, formed on the basis of a common language, territory, economic
life, and psychological make-up manifested in a common culture. … It must be
emphasized that none of the above characteristics taken separately is sufficient to
define a nation. More than that, it is sufficient for a single one of these character-
istics to be lacking and the nation ceases to be a nation.”6 This definition proved
to be the basis for eliminating non-Russian national identities. Wealth appropria-
tion in this broad sense—as the total appropriation of any and all resources that
give rise to sustainable cultures, economies, and polities—is astonishingly far-
reaching. Appropriation also is an eminently economic topic.
13.2.6. Genocide as Deprivation of Security

as a Fundamental Service
Alongside health and education, the physical security of one’s property, person,
family, and community is fundamental to a well-ordered society. Depriving a
people of the protection of the law and giving immunity to perpetrators—or even
to incentivize them—creates insecurity. For example, under Stalin an internal
passport system was introduced that limited people’s freedom of movement.
Escape routes were blocked. In Australia, permission to leave government-cre-
ated Aborigines reserves needed to be sought from the colonial authorities, usu-
ally to work on cattle stations but without any labor rights, not even to the legal
minimum wage. Defined as “under legal guardianship, wards of the state, minors
in law, specifically denied civil rights, social welfare entitlements, and most of the
benefits inherent and explicit in the rule of law” (Tatz in Totten and Parsons 2013,
64), Aborigines could neither marry nor legally engage in sexual intercourse with
298 Case Studies I
non-A borigines. Nor could they vote in state or federal elections until the 1960s.
The point was segregation, privation, and—w ith the removal of their children—
genocide (64–65). The Yana, in California, had no protection of their property
or their lives under the law, and the hunting of Yana was financially incentivized
both by the government of California and the federal government of the United
States. The Herero and Nama, in South-West Africa, “were dispossessed and all
their land was officially seized by the colonial power … All Africans were com-
pelled to compulsory work” and colonial decrees “restricted the Africans’ free-
dom of movement gravely and forced them to carry a tiny identity badge around
the neck” (Schaller in Totten and Parsons 2013, 96). In Ottoman Turkey, the
state was “withholding from [Armenians] the protection of the state” (Adalian
in Totten and Parsons 2013, 119). The examples can be multiplied, all illustrating
conditions of pervasive insecurity and situations in which a part of the workforce
is converted from an economic asset to an economic liability as their productivity
and economic contribution to the country is reduced. This necessarily is a topic
for economics and economists.
In sum, even a cursory reading of the pre-Holocaust literature easily yields
ample illustrations for each of Anderton’s (2014) claimed “interdependencies”
between economics and genocide. His contention that economic considerations
may offer valuable additional insights into the process of genocide seems quite on
the mark and is well worth pursuing in greater detail in separate research efforts.
13.3. Constrained Optimization Theory

This section delves into the first of Anderton’s (2014) “interdependencies,” that
is, genocide as a deliberate choice. To be deliberate, a reasoning process has to
be involved and, in economics, reasoning processes often are examined with the
help of constrained optimization theory (COT). An example of a formal, that is,
mathematically specified, application of COT to cases of genocides and mass kill-
ings is provided in c hapter 6 in this volume. The theory makes specific behav-
ioral predictions and raises attendant policy issues for would-be interveners to
consider. A nonformal statement of COT is developed in Brauer and Anderton
(2014) for which this section provides several case illustrations drawn from the
pre-Holocaust era. In short, the hardly disputable idea is that to achieve a desired
end, even that of genocide, perpetrators must avail themselves of a resource pool
to defray the cost of achieving their objective.7 Logically speaking, there are but
two dimensions in thinking about genocide in terms of COT. First, if the general
objective concerns the removal of an undesired out-g roup, then the specific act
of killing is but one among a number of alternative options to achieve the gen-
eral objective. For example, the relocation of the out-g roup to territories of other
states is, in principle, an alternative to killing. Second, even if killing itself is the
objective, it would still be true that different forms of killing (shooting, gassing,
starving, and so on) impose different draws on the perpetrators’ resource base.
The theory predicts that as cost and resource constraints—or changes therein—
become binding, they seemingly force the hand of perpetrators as if to direct
them, given a specific objective they wish to achieve, into one form of genocide
rather than another. Perpetrators are hypothesized to attempt to optimize the
destruction, subject to cost and resource constraints.
The COT does not justify perpetrators’ objectives. Its point is not to under-
stand the beliefs upon which perpetrators act. Instead, the point is to understand
the calculus of the act itself and of how (changes in) cost and resource constraints
influence the act about to be undertaken or change the direction and sequence of
the acts undertaken. Note that COT also applies to victims who attempt to escape
victimization as well as to potential third-party interveners since they, too, act
under cost and resource constraints. One conclusion is that mass atrocities may be
viewed, analytically, as a life-or-death contest over costs and the resources needed
to defray them. If the cost is too high or the resource base too small, perpetrators
may remain hateful but have no means to act on their beliefs. Conversely, if the
cost is too high or the resource base too small, victims cannot hope to escape vic-
timization and third parties cannot intervene, whatever else they may feel, think,
or say, in private or in public. Also note that costs and resources include nonmon-
etary costs and resources and include the entire panoply of feelings, perceptions,
images, and attitudes on the basis of which humans form, maintain, or change the
objectives they wish to pursue (see Illustration 2 below; also see Boulding 1956;
Stigler and Becker 1977).
A total of forty-five possible combinations can be considered for detailed study
(see Table 13.1). Five pre-Holocaust cases, times three parties (perpetrators, vic-
tims, and third parties), times three analytic categories (objectives, costs, and
resources). For example, for the Yana case alone, one could examine the objectives
of perpetrators, of victims, and of third parties; the costs perpetrators, victims,
and third parties incur; and the resources available to perpetrators, to victims,
and to third parties. In each instance, the focus is on one party and one category.
The combinations would be multiplied when considering how each combination
may affect the productivity of each party or when considering interactions, feed-
back effects, and path dependencies among parties and categories. For reasons of
limited space, only five of the forty-five combinations are chosen in the following
illustrations (see Table 13.1), but hints at interactions are provided.
13.3.1. Illustration 1 (Yana)

Ben Madley, in his chapter on the Yana in Totten and Parsons (2013), suc-
cinctly summarizes the victims’ alternatives: “To protect themselves Yana
people had three choices. They could seek protection from the newcomers (by
Table 13.1 Five Illustrative Cases Covered (X) Out of 45 Possible Combinations
Combination Yana Aborigines Herero/Nama Armenians Ukrainians

1. Perpetrators’
objectives
2. Victims’
objectives
3. Th
ird parties’ X
objectives
4. Perpetrators’ X
costs
5. Victims’ costs X
6. Th
ird parties’
costs
7. Perpetrators’ X
resources
8. Victims’ X
resources
9. Th ird parties’
resources
becoming servants, concubines, wives, and laborers), fight them, or retreat into
the mountains. All three options were hazardous” (Madley in Totten and Parsons
2013, 19–20). Since Native Americans had virtually no rights under California
law, living among colonizers and immigrants could be (and was) capricious. Not
having legal recourse translates into not having at one’s disposal the resource
of law and its enforcement. Moreover, individual-by-individual “integration”
would, in any case, dilute the group as a recognizable, separate entity with which
individuals born to the group could identify. Regarding the second option, to
hope to engage, fight, and win against the settlers’ superior numbers and fire-
power with the bows and arrows at the Yana’s disposal implied an implausibly
high death-toll cost. The remaining option was withdrawal into the surrounding
mountains, by no means a pleasant choice, but a choice nonetheless—i n fact, a
constrained choice precisely in the optimization meaning of COT, to “make the
best of a situation, given the circumstances.” Had they then been left alone, the
Yana might have adapted, but “immigrants made mountain life increasingly dif-
ficult” (20). Ranching, hunting, and mining (by despoiling rivers and thereby
decimating the salmon fish stock upon which Yana relied) depleted the resource
base by which to survive. Instead of open fighting, Yana resorted to raids, rob-
bery, and arson of settlers’ homesteads (26). In an action-reaction pattern, or
so Madley describes the sequence of events, settlers responded in kind, raiding

Indian camps in turn. Their numbers, firepower, recourse to state financing (and,
later, to federal funding, 32) as well as immunity from the law for abusing or
killing Indians, overwhelmed the Yana. During one of General Kibbe’s Indian-
hunting campaigns, he endeavored to keep Yana on the run, hemming them in at
certain locations to deny them opportunity and time “to gather acorns or seeds
sufficient for winter, or to fish.” He thus prepared them for death by starvation
lest they surrendered to accept forced relocation onto reservations, “an agoniz-
ing choice” (31)—again precisely the point COT makes. True, fleeing ever fur-
ther from familiar hunting, harvesting, and fishing grounds and abandoning
camps and food stores did imply that the Yana had increasingly “little to lose”
(35), but they also had even less to fight with to effectively oppose their own
destruction. Madley rightly points out that “the state of California, the U.S. gov-
ernment, or immigrant communities could have negotiated with the Yana and
avoided genocide” and that “diplomacy might have generated a very different
outcome” (45). But by then both the mindset and the incentive structure (e.g.,
cash for scalps and legal immunity) made that unlikely, especially since a good
part of the final phase of the Californian annihilation campaigns (1860–1872)
occurred while the East and South of the country were embroiled in the US
Civil War (1861–1865) and its aftermath. The case well illustrates how victims’
resources constrain the set of feasible alternatives left to them.
13.3.2. Illustration 2 (Aborigines)

If the Yana illustration focuses on the resource constraint for the victims, the
Aborigines illustration centers on third-party preferences and objectives stem-
ming therefrom—a lbeit with a twist. While in that genocide’s first phase, many
individuals in Australia felt uninhibited to abuse Aborigines at will, for others
this behavior caused misgivings, and a movement to protect victims arose such
that “the colony of Queensland was pressed to introduce the world’s first statute
to protect a people from genocide: The Aboriginals Protection and Restriction
of the Sale of Opium Act of 1897” (Tatz in Totten and Parsons 2013, 58). With
it came state-concerted efforts at segregation into mission schools, welfare insti-
tutions, land reserves, and institutionalized child removals. Given Victorian-era
mores, this may have been well intended. An internal third party—t he colonies’
good-hearted, educated members of society of prominence and rank—g radually
asserted more political power over dealing with Australia’s Aborigines popu-
lations. A preference change took place but, according to Colin Tatz’s account
in Totten and Parsons (2013), this did not change the fact that the objective
remained genocide. What resulted was the conversion of seemingly random,
more or less spontaneous acts of disorderly genocide into an ordered, institution-
alized genocide. Aborigines peoples and cultures still were to be eliminated as
302 Case Studies I
distinctly identifiable entities. One group of perpetrators replaced another; one

form of genocide morphed into another.
For instance, “key state bureaucrats” held a national summit in Canberra in
1937. One of its leaders proposed a three-part plan: “full blood” Aborigines were
to be segregated into reserves; mixed children removed from their mothers; and
marriage control orchestrated such that “quarter-and half-blood Aboriginal
maidens would marry into the white community. It would then be possible to
‘eventually forget that there were ever any Aborigines in Australia’ ” (Tatz in
Totten and Parsons 2013, 62). In another example, until 1939, in the state of
New South Wales (which includes the city of Sydney), the frequently stated,
official reason for child removals from Aborigines parents—handwritten into
the record books—read “For being Aboriginal” (63). 8 Criminalization not for
“having done” something, but “for being” someone, goes to the heart of geno-
cide. Thus, while lives were saved (people not killed), child removals nonetheless
“destroyed much of the fabric of Aboriginal societies” (65). The “protection-
segregation era” (55), which lasted into the 1980s, further raises the point that
the UN Genocide Convention’s “intent to destroy” need not be restricted to
bad faith or intent (male fides). Even well-meaning—“good”—i ntent can lead to
genocide and is a punishable offense. The prevalence of a male fides interpreta-
tion of the Convention perhaps explains why child removals in particular con-
tinued on a grand scale even after Australia ratified, in 1949, the UN Genocide
Convention.
There is an interaction that emerges here between perpetrators’ mindset and
nonmonetary costs and resources. Once perpetrators’ mental, perceptual constraint
changed in the 1970s and 1980s, once their image of themselves and of their vic-
tims changed, genocidal actions directed against Aborigines peoples ceased rap-
idly. In the South African context, Kleinschmidt (1972) referred to this as the
need for “white liberation” from their own mindset. This might be viewed as a
“positive liberation” instead of a “negative liberation,” the latter restraining acts
of genocide but leaving the intent unchanged, the former changing the intent of
the perpetrators.9 The point refers to the cost of preference formation and, more
specifically, the cost of changing habitual, or habituated, even inculcated, lines of
thought and leads to a discussion, which cannot be pursued here, of the role of
upbringing, neurodevelopment, peer effects, socialization, education, and con-
formity versus resistance. (For an application, see c hapter 26 in this volume.)
13.3.3. Illustration 3 (Herero and Nama)

In section 13.2, the story of how economic conditions affected the genocide of
Herero and Nama peoples was told mainly from the victims’ point of view. The
story is continued here from the settlers’ point of view and as an illustration of
how binding cost and resource constraints affect the initiation and conduct of
genocide. Namibia’s climate was harsh, and transportation and communication

of the day (both to as well as within Namibia) were poor, dispersed, and difficult.
While Chancellor Bismarck’s government in Berlin did want colonies, it was not
keen to expend great sums on them. Prior to 1904, a divide-and-r ule policy that
played tribal leaders in Namibia against each other permitted the acquisition of
vast tracts of land even as the number of settlers was small, only about 4 percent of
the Central and South Namibian populations. When war broke out, Herero at first
prevailed. It took nearly half a year for imperial Germany to send 14,000 troops
and start its annihilationist campaign under von Trotta’s command. As a military
man, he sought a military sort of victory. “The Hereros are no longer German
subjects,” he wrote, and continued that “all the Hereros must leave the land. If
the people do not do this, then I will force them to do it with the great guns. Any
Herero found within the German borders with or without a gun, with or with-
out cattle, will be shot” (quoted from Schaller’s account in Totten and Parsons
2013, 89). But the campaign was not cheap. In monetary terms, the apparently
vast sum of 585 million Reichsmark was spent on troops, weapons, and logis-
tics, and almost 2,000 of von Trotta’s troops died from epidemics (90). Not only
but in part because of this monetary and manpower cost constraint, von Trotta
changed tactics, forcing people to flee into the desert and to perish there (89–90).
And yet, the settlers were not in favor of von Trotta’s policy of killing. What con-
cerned them was their quest for forced labor so as to lower their labor costs and
to improve the fortunes of the settler economy. Both private and public German
employers increasingly relied on forced labor to address a labor supply shortage.
Settlers contested von Trotta’s extermination order and lobbied the Kaiser’s gov-
ernment in Berlin to order the establishment of concentration camps to corral a
coerced workforce. These were in fact ordered up by then-Chancellor von Bülow
and built in the second phase of the war. Conditions were harsh, with a camp
death rate of 44 percent between 1905 and 1907, as the German army command
itself reported (percentage calculated from figures given on 95). In arguing why
von Trotta’s killing policy needed to stop, the German commissioner of settler
affairs in South-West Africa, Paul Rohrbach, commented in his 1953 biography
that “South-West Africa with natives was of much more value … than without”
(90). Cost and resource considerations (constraints) influenced the conduct of
the genocide.
In a third phase, following the war itself, captives were released into a politi-
cal, economic, and cultural vacuum. The colonial authorities instituted a policy of
tribal de-identification and proletarian re-identification, a cultural genocide and
repurposing of human beings for economic exploitation. A previously autono-
mous, self-reliant population was readied for an economic “harvest.” In a 1907
decree, freedom of movement was severely restricted for indigenous peoples and
the wearing of identity tags made mandatory. Additionally, settlers demanded
that all indigenous peoples be tattooed (Schaller in Totten and Parsons, 96).
304 Case Studies I
Still fearing uprisings, a deportation policy was considered to ship victims to the
then-German colonies of Cameroon and Papua New Guinea, but the outbreak of
World War I prevented its large-scale implementation (96).
13.3.4. Illustration 4 (Armenians)

The fourth illustration focuses on the resources a perpetrator can bring to bear
on the victims. As mentioned, the Ottoman Empire “withdrew the protection of
the state” from Armenian subjects: its cabinet made the relevant decisions and
charged its Ministries of the Interior and of War with overseeing the destruction.
In turn, they “instructed local authorities on procedure, the timing of deporta-
tion, and the routing of the convoys of exiles” (Adalian in Totten and Parsons,
126). Parliament “enacted legislation legalizing the decisions of the cabinet”
(126). The Turkish leaders had “at their disposal immense resources of power and
an arsenal of formal and informal instruments of coercion” (118). And “at every
level of the operation against the Armenians, party [i.e., CUP] functionaries
relayed, received, and enforced the orders of the government” (119). The primary
prong of attack consisted of deportation orders, implemented by removal by train,
horse-d rawn wagon, mules, or—for most victims—walking, either southeast
from the Anatolian plain or south from the Black Sea or southwest from Armenia,
both mountainous regions, toward the Syrian desert. The government made
“no provisions” to supply food or overnight shelter and “only one-quarter of all
deportees survived the hundreds of miles and weeks of walking” (120). Local gov-
ernments assisted in the deliberate neglect. Marauders, including Kurds (another,
but Muslim, minority), preyed on refugee caravans and engaged in looting, rap-
ing, kidnapping, and killing without penalty. Those who survived all the way to
Syria died there from the day’s heat or night’s cold or were assaulted by “sword
and bayonet” by the Special Organization’s killing units (120). The genocide was
highly orderly in that government provided for the underlying structure that was
“conceived with secrecy and deliberation and implemented with organization and
efficiency” (119). For example, joining Germany late in 1914 as a World War I axis
power, Turkey hoped to regain territories it had lost in the Russo-Turkish war in
the late 1870s. Fearing Armenian collaboration with (coreligionist) Russians, or
so Enver Pasha, the minister of war argued, able-bodied Armenian men were to
be “conscripted into the Turkish armies, [then] disarmed, forced into labor bat-
talions [for logistics support], and either worked to death or outright murdered”
(120–21). Older males “were summoned by the government and ordered to pre-
pare themselves for removal from their places of habitation” (121). They submit-
ted unwittingly, apparently not expecting subsequent imprisonment, torture, and
mass executions. To undermine resistance, “prominent leaders were specially
selected for swift excision from their communities” (121), about 250 of them in
an especially coordinated action in Constantinople on the night of April 23–2 4,
1915, which is now often taken as the nominal date of the beginning of the geno-
cide against Armenians. Women, children, the elderly, and the infirm went on the
death marches to Syria, herded along by soldiers (126). Extermination camps were
built on today’s borders with Iraq and Syria, and the killings in Syria were carried
out by killing units. But resources were not infinite. For example, bullets were
spared for the war effort by forming a Special Organization of convicted crimi-
nals to kill with “scimitars and daggers” (127). Armenians’ “abandoned goods”
were confiscated and auctioned off, profiting CUP officials and as a “means that
rewarded its supporters” (127). Still, set against the resources of the Ottoman
state as a whole, neither Armenians themselves nor third parties—embroiled as
they were in World War I—could or did offer any effective counterforce.
13.3.5. Illustration 5 (Ukraine)

This illustration highlights a few aspects of the impossibly high cost to victims
of escaping Stalin’s program of forced starvation. Since James Mace’s chapter in
Totten and Parsons (2013) is written primarily from the point of view of perpe-
trators’ actions, relatively little is said that directly pertains to the monetary or
nonmonetary costs victims’ faced to resist or to escape.10 But from the actions of
the perpetrators one can infer a few things about the potential cost of attempting
escape. Ukrainian victims were primarily rural populations with an underdevel-
oped class of intellectuals who might have articulated the experiences, feelings,
and thoughts of their compatriots. The famine “corresponded in time with a
reversal of official policies that had hitherto permitted significant self-expression
of the USSR’s non-Russian nations.” Now, “non-Russian national self-assertion
was labeled bourgeois nationalism and suppressed” (Mace in Totten and Parsons
2013, 157). Freedom of movement was eventually curtailed with the introduction
of an internal passport system (166, 172). Forced collectivization of farms was
introduced, using as enforcers city-based workers who were notionally integrated
into pan-Soviet aspirations. “Workers were sent from factories, and sometimes a
factory would be named ‘patron’ of a given number of villages; that is, the factory
would be assigned villages in which to enforce collectivization and seize foods”
(162–63). Accompanied by State Political Directorate (GPU) forces and hunting
dogs, these “tow brigades” would harass, manhandle, torture, and kill peasants
in order to search for, find, and confiscate food. Horses, tools, and farm imple-
ments were removed as well (175). Without passports, transport, or food, but with
close supervision of their activities, escape was beyond reach. To successfully
run away or obtain forged papers to travel (171) appear to be exceptional events.
Peasants whose farms had not yet been collectivized nonetheless were subject to
deliver on food quotas. Nonfulfillment meant fines and searches of their farms
(163). Children were encouraged to turn on their parents and report suspected
hoarding of food or seeds (167). Bits of precious metals and jewelry in peasant
306 Case Studies I
possession were extracted in exchange for food at an increasing number of state-

operated hard currency stores (166, 173). Eyewitness accounts report acts of
cannibalism by starving peasants to ensure physical survival (e.g., 172–73, 186).
Starvation greatly increased vulnerability to disease and a great many people
died of typhus, dysentery, and the like. Demoralization added weight to the psy-
chological cost of dealing with loss of lives and property. One eyewitness, then
a school child, recalls a Russian-language song that in the heyday of starvation
“would play every day, ten times a day” over the ubiquitous village loudspeakers
heralding the “joyous refrain of town and country: Our burdens have lightened,
our lives have gladdened” (184). Resistance met with virtually immediate reprisal
by state security forces (e.g., 172). While confiscated foodstuffs were exported to
elsewhere in the USSR, migration was prohibited and offers of food aid (imports)
were rejected, matching the official denial of a food crisis or of widespread starva-
tion. As Steven Rosefielde (2009, 259) writes: “Grain supplies were sufficient to
sustain everyone if properly distributed. People died mostly of terror-starvation
(excess grain exports, seizure of edibles from the starving, state refusal to provide
emergency relief, bans on outmigration, and forced deportation to food-deficit
locales), not poor harvests and routine administrative bungling.” All this com-
bined with the vagaries of weather (particularly bad in 1932–1933), with “rapid
linguistic and cultural Russification” in Ukrainian cities to which many peasants
had been transferred under Stalin’s industrialization drive (Mace in Totten and
Parsons 2013, 167), and topped by state “monopolies or near-monopolies of pro-
paganda, reward, and coercion” (167; also see chapter 12 in this volume). Starved
of resources, figuratively and literally, the cost of exit became correspondingly
impossible to pay, whatever the victims’ objectives. Victimhood might be defined
by the degree to which cost and resource constraints are binding and foreclose
any option to escape victimization, or what Lawrence Langer (1980), in a differ-
ent context, referred to as “choiceless choice.”
13.4. Conclusion
Much of the genocide literature covers the Holocaust and post-Holocaust cases,
and because of that this chapter has focused on pre-Holocaust genocides. Given its
claim of universal application, if economic theory is relevant for post-1940s geno-
cides, it should also be relevant for pre-1940s genocides. Using a sixfold schema
elaborated by Anderton (2014) and applying it in no more than a cursory manner to
the cases of the Yana, Aborigines, Herero and Nama, Armenians, and Ukrainians,
we find that all of Anderton’s contentions as to the relevance and potential con-
tribution of economics to help to more fully understand genocide appear correct.
A more specific investigation, focusing only on genocide as a behavior of choice,
likewise reveals that the economic concept of meeting objectives (“preferences”)
under cost and resource constraints is highly relevant and can be illustrated with
ease for each of the pre-Holocaust cases that this chapter has, however briefly,
examined.
To conclude, we offer eight impressions from our reading of the cases through
an economic lens. The first is that the case study writers predominantly seem
interested in questions of guilt and of justice. Documentation focuses on perpe-
trators’ objectives—however distorted and malevolent—and on the mechanics
of how and why the atrocities were committed rather than on how and why they
were carried out in one specific way rather than another. As a whole, the chapters
rarely directly consider any genocidal act, or escape therefrom, as the outcome
of constraints imposed by costs and resources. Authors tend to take feasibility as
given and emphasis is placed on a particular choice made, a behavior observed,
and not on what the alternatives, if any, were or on how any one choice came to be
made. This overstates things for the sake of argument, but an important implica-
tion is that one cannot hope to improve the score on genocide prevention unless
one more fully understands the constraints that may divide acts of genocide into
feasible and infeasible.
Second, as to the victims, and again overstating for the sake of argument, in
most cases the approach taken is about what is done to them and what is not done
for them: perpetrators are active, third parties may or may not choose to be active,
and victims are passive.11 Even the eyewitness accounts appended to each chapter
tend to emphasize victims as passive recipients of violence, not much exploring the
options they may or may not have had. To an economist, interesting questions
lie in exploring what the victims’ cost and resource constraints were and how
their (presumed) passivity came to be. The Ukrainian peasants who starved to
death during Stalin’s reign saw their predicament increased as an internal pass-
port system and travel restrictions were imposed. This limited their set of feasible
choices, but limited it to what extent? What were the remaining feasible options?
Just how did survivors survive? It is not clear to us that genocide survivor stories
have been parsed systematically, and comparatively, to come to a theory-d riven
and/or theory-creating general understanding of why survival occurs so that sur-
vivors’ experiences may be used to better assist future victims and limit the dam-
age done. Delving into resource and cost constraints just may permit one a deeper
understanding of the feasible (and infeasible) options victims may have had.
Third, our sample of five cases is small and not representative of the genocides
and mass atrocities literatures at large, but in our cases, third parties appear some-
what incidental. In all five cases, there was no lack of information about a geno-
cide happening: The German, English, and Boer settlers in South-West Africa
knew what was happening and wanted to change the course of genocide toward
their own liking, as did “well-meaning” people in Australia in regards to the
Aborigines populations. In the Armenian case, the news of genocide was wide-
spread throughout the world. But the relative capacity or incapacity of potential
308 Case Studies I
internal and external interveners appears not well-theorized or -explored, and

economists might make helpful contributions in this regard, as might of course
other scholars who, certainly in other contexts, have begun to think about
“enablers” and “disablers” of genocides and mass atrocities, whether internally or
externally organized.
Fourth, regarding preferences, costs, and resources, the cases pay most atten-
tion to the former, and mostly to those of perpetrators. But one other theme does
emerge strongly, namely, that of depriving victims of food sources and of compel-
ling them to deal with sheer day-to-day physical survival. Herero and Armenians
were driven into deserts, Ukrainians dispossessed of farm produce, and Yana
chased into food-scarce mountain regions. While probably beyond the logistical
capacity of the world at the time, today this suggests that a Convention be sought
with built-in triggers (thresholds) that, when tripped, precipitate automatic food
aid delivery from UN depots by UN forces (on this point, also see the eminently
practical chapter 26, as well as c hapter 24 on early warning and genocide predic-
tion, both in this volume).
Fifth, beyond perpetrators, victims, and third parties, and beyond prefer-
ences, costs, and resources, lie some other observations. From our reading it
would appear that among victims a spontaneous separation takes place. Victims
are not all equal; some have better opportunities than others. Diasporas seem
self-selected, for example, in that the relative geographic openness of a genocide-
perpetrating country matters. When victims can flee with relative ease to neigh-
boring countries—some Nama south to the Cape Colony, some Armenians north
into Russia—t hen those living near a border will generally have an easier time to
escape than those bottled up in the country’s interior. Their cost is lower; their
opportunity set is enlarged, exactly as argued theoretically in Brauer and Anderton
(2014). In contrast, when one border is more heavily guarded or sealed, then a
cross-country journey to another border becomes all the more hazardous, as illus-
trated by Herero trying to cross the desert. This seems too obvious to state, and
while the Holocaust literature no longer lumps victims into an undifferentiated
mass but recognizes differences among them, in the pre-Holocaust cases we
have read we see relatively little exploration of such themes, nor how they may be
exploited to prevent or at least mitigate genocide.
Sixth, in regard to the productivity with which acts of genocide and mass atroc-
ities are carried out, economies of scale and density are important. For the case of
the Yana, scale was not as relevant. Living in small bands, being relatively mobile,
and knowing their own lands better than their pursuers, killing took place one
“batch” at a time. In contrast, concentrated columns of Armenian people march-
ing in the open were far easier to attack.
Seventh, also decisive in facilitating the killing is the relative speed with which
victims and perpetrators can act. Speed depends on transport, communication,
and bureaucratic or organizational coordination. The set of feasible options for
victims can become extremely narrow because of timing. For the Yana, the geno-
cide wore on for decades and for the Aborigines for about two centuries. The
economic issue here was neither time nor agglomeration but, to the contrary, the
dispersed nature of the victims, albeit within well-defined boundaries, combined
with information and network economies. In the cases of Armenia and Ukraine,
the killing proceeded fast, in part because perpetrators could coordinate their
actions far more easily than victims could organize their escape.12 That genocide
can be both fast and slow, and can occur in both low-density and high-density
population contexts, should surprise, highlighting once more that genocide is
conditional, that it stands or falls on its feasibility.
Eighth, the economics of image (perceptions) and identity (preferences) is
highly relevant to genocide and genocide studies (Boulding 1956; Akerlof and
Kranton 2011; also see chapters 14 and 21 in this volume). The relative ease of
(self-)identification and (self-)labeling of victims facilitates genocide. Image and
identity speak to preferences and yet can be a resource (or a cost!). The poten-
tially strong call for the development of a supervening identity (“we,” “all of us,”
our “common humanity”) is generally not explored in the cases. The possibility
of preference resets is elaborated from a behavioral economics perspective in
chapters 6 and 26 in this volume. In our five cases here, if such calls had been
made, why were they not heeded?13 Neither perpetrators or victims, nor third
parties, are unitary actors. Each of “us” and each of “them” lives by constraints,
impelling each to action or inaction. What combination of strength of conviction
(preferences), cost, and resources procures one outcome over another? Unless
one explores this nexus, one must conclude—one hopes, incorrectly—that for
the victims there exists no exit.
There are other themes in, and tools of, economics that could be helpful to the
study and prevention of mass atrocities such as genocide. One thinks for exam-
ple of evolutionary game theory applications, such as public good coordination
failure within a game-t heoretic setup; or of identifying and studying the condi-
tions that shape the anticipatory, strategic behavior of perpetrators, victims, and
third parties; or of topics such as framing, cognitive bias, and reference-point
dependence in behavioral economics. Even our brief excursion demonstrates
that economists can no longer stay away from making the contributions their
science affords them to make to the study of genocides and other mass atrocities
and their prevention.
Acknowledgments
The authors acknowledge helpful comments made by Charles H. Anderton and
by three anonymous reviewers. The usual disclaimer, that no one but the authors
is responsible for any remaining shortcomings, applies.
Appendix
Table 13.A1(Perpetrators): Comparative Genocide Study Conceptual Matrix Based on Brauer and Anderton (2014)
Preferences Prices (costs) Resources

WHO? [Who held these preferences?] MONETARY MONETARY
1. Architects and rank-and-fi le (in-group/s) 1. F
ixed (genocide mgt; plant buildup) 1. S heer size of monetary resource pool
2. W
hom were they about? (out-g roup/s) [Any sunk cost?] and resource stream (cash-on-hand
3. A
re there any in-group opponents with 2. Variable (recruitment and training plus current inflow regime; GDP;
alternate preferences? of perpetrators; isolation of victims; natural resource stock)
WHAT? [What were these preferences about?] transport; holding areas; food, drink, 2. R esource augmentation (higher
4. Perceived threat to full political control shelter, hygiene provision; genocide tax rates on incomes and assets;
(exclusion of new or persistent rival) technology; information control; forced debt; debt repudiation; debt
5. Perceived threat to full territorial control security to control in-/out-groups; rescheduling; forced credit and
or territorial integrity border patrol/control costs; etc.) grants; theft and looting; depletion of
6. P erceived desire for ethnic/race purification 3. ( Im/explicit) permission of theft/ natural resources; money creation and
7. Desire for out-g roup elimination looting may reduce recruitment cost counterfeit currency; etc.)
(removal but not necessarily death) or NONMONETARY
extermination (death) [either as a means 4. Audience cost (e.g., domestic and
to an end (instrumental preferences) or as foreign moral disapproval; pressure
an end-i n-itself (malevolent preferences)] to cease and desist)
8. Resource confiscation/looting 5. Degree of credibility of intervention
6. Th
reats (e.g., of persecution/
prosecution)
WHY? [Why did these preferences arise?] PRODUCTIVITY-RELATED ISSUES NONMONETARY
9. Gradual or sudden changes in the internal 1. C onstrained optimization and 3. P hysical capital (e.g., stock of
or external socio-sphere (political, trade-offs plant, property, and equipment;
economic, and/or cultural, e.g., entry or 2. Dis/economies-of-scale infrastructure, weapons stock; etc.)
exit of third party) 3. P ath-dependence and interaction 4. Human capital (e.g., education levels;
10. G
radual or sudden changes in the effects with preferences genocide “entrepreneurship” and
biosphere 4. Complements in production; “reward” systems?)
WHEN? [When were preferences formed, substitutes in production 5. Social capital (e.g., sense and strength
shifted, or manipulated?] 5. I nduced substitution by threat of of in-group identity; loyalty, probity,
11. I deological antecedents (e.g., race-based (or actual) intervention rectitude of in-group members;
thinking and eugenics) 6. C ost-lowering devices (such sympathetic foreigners)
12. New local/provincial/national leaders as “self-identification” and
13. Historical and/or catalytic events “self-v ictimization”)
7. L earning-by-doing
WHERE? [Place/location?]
8. “Bundling” (e.g., enslavement
14. Locales, cities, regions, nations
and death)
HOW? [How were preferences formed and 9. Hyperspecialization and comparative
reinforced?] advantage in “production of
15. Relative ease of identification destruction”
16. Scope; leverage ratio of out/i n-g roups
17. “ Social engineering” of in-g roup
preferences by genocide architects
Table 13.A2(Victims): Comparative Genocide Study Conceptual Matrix Based on Brauer and Anderton (2014)

WHO? [Who held these MONETARY MONETARY
preferences?] 1. C ost of resistance (single or collective; 1. I ncome and assets (threatened by, e.g.,
1. O
ne or more out-g roups spontaneous or organized) punitive taxes; limited employment;
WHAT? [What were these 2. C ost of legal passage (e.g., exit visa; underpaid or forced labor; forced dissaving;
preferences about?] forfeiture of assets) or cost of clandestine asset surrender; confiscation, theft, looting)
2. Avoidance/evasion of passage (e.g., distance to safe border, NONMONETARY
identification as out-g roup bribing guards) 2. I nstitutions of family and community;
member and victim 3. C ost of communication and coordination reproductive ability
3. A void/evade submission to or 4. Cost placed on running one’s business (e.g., 3. R eal estate and productive hunting or
abuse by perpetrators restricted occupations, place of business) grazing land (incl. wells, herds, and other
4. Safe escape and wealth 5. Cost of forestalling expropriation resources; could be but often not thought of
transfer/h iding 6. C ost of daily survival (e.g., food, as monetized)
5. Physical a/o mental survival clothing, drink, shelter, medications, 4. Special skills and attributes (e.g., links
of self and family rationing, etc.) overseas; languages spoken; youth; current
6. S elf-sacrifice to save others NONMONETARY location, e.g., close to borders)
7. Preferences usually about 7. Restrictions placed on schooling, housing, 5. Relative strength of in-g roup leadership and
the self, not about the occupation, transport, movement, cohesion that may offer protection
targeted out-g roup [median association, communication, dress, etc.
preference] 8. Nonmonetary costs of passage
WHY? [Why did these 9. Search cost for substitutes 6. I D cards and passports (that may be subject
preferences arise?] 10. L
oss of trust within the out-g roup to confiscation or reissuance)
8. Basic human preference (declining out-g roup social capital) 7. I n-g roup social capital and political
for security of self, family, 11. Psychological burden of insecurity; organization (subject to deliberate
community exists => fighting eventual loss of “spirit” and will to live destruction)
for appropriate security PRODUCTIVITY-RELATED ISSUES
“market share”? [Perpetrators 1. A
voiding “singularity of victimhood”
monopolize security by 2. T errible trade-offs; uncommon choice =>
exclusion devices?] high “search cost” for optimal solutions
9. Perpetrators arise 3. H umans slow, can’t survive long without
10. L
ack of foresight to water and food; hence intervention
anticipate the needs to be fast and substantial => ease
coming danger of victimization relative to protection,
WHEN? [When were esp. for poor, incommunicado, immobile,
preferences formed, shifted, or young, aged, a/o illiterate/u nskilled; more
manipulated?] so when onslaught is rapid and high-scaled
11. S uddenly or gradually? 4. Production function for hiding/escape
Subtly or obviously? [Elements include ease of victim
WHERE? [Place/location?] identification, geography, resistance
12. L
ocales, cities, regions, networks, city/r ural settings, closeness of
nations friendly regions, nations, etc.]
HOW? [How were preferences
formed and reinforced?]
13. Nature and role of
self-identification
Table 13.A3(Third Parties): Comparative Genocide Study Conceptual Matrix based on Brauer and Anderton (2014)

WHO? [Who held these preferences?] MONETARY MONETARY
1. N
eighboring state/s [NS] (e.g., in their own 1. N
S: Cost of border patrol/control 1. S pecific budgets/f unds available
territorial integrity and political stability) (dep. on physical attributes of for equipment, troops, billeting,
2. N onneighboring state/s (e.g., unilateral borders) and logistics
intervener/s) 2. N S: Direct monetary/budgetary cost 2. G eneral size of global GDP
3. I nternational organizations, institutions, of hosting refugees 3. S pecific size of intervening
and communities [IOIC] (e.g., UN; NGOs; 3. C ost of intervention (troops, country GDP
foreign media interests; foreign individuals equipment, etc.) NONMONETARY
and businesses) 4. N NS/IOIC: Distance increases all 4. Audience costs (against
4. I nternal organizations, institutions, and costs of intervention perpetrators) [but audiences are
communities [INT] (e.g., resisters or active 5. I nformation and fickle and can change rapidly]
compliers) communication cost 5. Targeted (“smart”) sanctions, incl.
WHAT? [What were these preferences about?] NONMONETARY disrupted trade and investment,
5. Active opposition to genocide; open or 6. P otential conflict spillover on NS’s esp. when there is an “inclusive
implied complicity; outright disagreement own polity elite” (i.e., broad-based links to be
6. U pholding UN charter and genocide 7. Environmental costs of hosting large targeted)
convention; state sovereignty principle numbers of refugees 6. C redible threat of penalties against
7. Can-do-something or can-do-nothing 8. Domestic and global audience cost perpetrators (e.g., International
attitude 9. Transaction and coordination cost Criminal Court)
8. Neutrality or bystander (“wait-a nd-see”) (can be monetary)
9. Outside interest in NS’s territorial a/o polity 10. P opulation ignorance a/o apathy to
integrity lend intervention support
WHY? [Why did these preferences arise?] PRODUCTIVITY-RELATED ISSUES 7. L ong-term engagement with
10. S overeign control vs. universal human rights 1. U
nilateral, bilateral, or multilateral and development of potentially
11. Th
ird-party political, economic, cultural intervention; regional or global vulnerable societies to build up a
interests for antigenocide (or silence) multilateral intervention? diverse, dispersed resource pool;
preferences 2. P rivate vs. collective interests; development of institutions of law
12. A nti-(or pro-)genocide as preference derived amalgamation, bargaining of and order; land registries; open-
from (subordinate to) other preferences conflicting interests; UNSC P5 society culture [possible backlash
WHEN? [When were preferences formed, veto powers charge of “neocolonialism,” etc.]
shifted, or manipulated?] 3. C oordination, if any, with civil society 8. Genocide prevention strategy and
13. Peace of Westphalia; state formation; League a/o business parties and interests implementation
of Nations; United Nations 4. Principal-agent and free-/easy-r ider
14. R
esult of pre-, during, postcolonial period problem; bargaining over cost-sharing
WHERE? [Place/location?]
15. L
ocales, cities, regions, nations
HOW? [How were preferences formed and
reinforced?]
16. Nation-state sovereignty principle
17. “Responsibility-to-protect” concept
18. Moral hazard potential
19. Political, economic, cultural interests
316 Case Studies I
Notes
1. Lemkin’s eight “fields” of genocide are the political, social, cultural, economic, biological,
physical, religious, and moral (Lemkin 1944, 82–9 0).
2. The items in square brackets—d iscrimination and persecution—were added in a later
rendition to become “The 10 Stages of Genocide.” In the United States, at least, work
has progressed beyond identifying “stages” of genocides into areas of early warning and
recognition of genocide triggers and to develop capacities in terms of prevention, mitiga-
tion, and response. See, for example, materials available via the United States Holocaust
Memorial Museum (www.ushmm.org). Also see chapter 10 in this volume.
3. Unless otherwise noted, all page references are to the chapters in Totten and Parsons
(2013).
4. The chapter by Jon Bridgman and Leslie Worley in Totten and Parsons’ third edition (2008)
also reports an apparently well-f unctioning Herero economy, centered around land and
cattle, which was disrupted with the injection of colonial settlers into the region. Political,
economic, and cultural relations within, between, and among local tribes, colonists, and
the colonial government then changed. Economically adverse pre-revolt conditions arose,
such as co-opting chiefs into colonial governmental graces (with the attendant, corrosive,
and corruptive perks of office), large-scale sales of indigenous lands, the emergence of slave
labor, and a credit economy that indebted locals beyond their ability to repay. The imperial
government in Berlin, however, was at first less than fully supportive of building an eco-
nomic infrastructure in support of the settlers and the colony’s economic development. To
some extent, this left the settlers to fend for themselves. Bridgman and Worley write that
while “racial tension [was] a major factor,” and while “a purely economic explanation [for
the revolt of 1904] is too simple … [t]‌he naked economic exploitation of the natives was a
major reason for the rebellion” (22). We thank an anonymous reviewer for pointing out the
third edition chapter to us.
5. Australia’s six colonies federated into an independent nation-state only on January 1, 1901.
6. Quote taken from http://w ww.marxists.org/reference/a rchive/stalin/works/1913/03a.
htm#s1 [accessed September 30, 2014].
7. In economics, resources are any monetary or nonmonetary means to pursue an end; costs
are “opportunity costs,” namely opportunities forgone by selecting a particular course of
action. (For example, reading this chapter costs the opportunity of doing something else
with your time.)
8. After 1939, at least a removal hearing before a magistrate was required.
9. We are reminded of Abraham Lincoln who once asked: “Do I not destroy my enemies when
I make them my friends?” (http://en.wikiquote.org/w iki/A braham_ Lincoln [accessed
February 11, 2015]). Similarly, the October 1986 Reykjavik meeting between then-US
President Ronald Reagan and Soviet leader Mikhail Gorbachev initiated the “reset” of
mutual images about each other’s country. When the “preferences” changed, so did the
objectives to be pursued.
10. Little is mentioned of their resources, either, but both are mentioned to some degree in the
twenty pages of eyewitness accounts appended to Mace’s chapter.
11. For the Holocaust, specifically, considerable work has been done to understand victims’
options and a “myth of victim passivity” has been dispelled. But for pre-Holocaust cases of
genocide, and perhaps for post-Holocaust cases as well, that may not be true in equal mea-
sure. At any rate, in the specific cases under consideration in this chapter, relatively little is
said about victims’ activity.
12. On aspects of genocide organization, see chapters 8 and 25 in this volume.
13. For example, in the case of the Herero, the German missionary society as well some
newspapers and political parties did bring the issues to national attention and debate in
parliament.
References
Akerlof, G. A., and R. E. Kranton. 2011. Identity Economics: How Our Identities Shape Our Work,
Wages, and Well-Being. Princeton, NJ: Princeton University Press.
Boulding, K. 1956. The Image: Knowledge in Life and Society. Ann Arbor: University of
Michigan Press.
Brauer, J., and C. H. Anderton. 2014. “Economics and Genocide: Choices and Consequences.”
Seton Hall Journal of Diplomacy and International Relations 15, no. 2: 65–78.
Kleinschmidt, H., ed. 1972. White Liberation. Johannesburg: Spro-cas.
Langer, L. L. 1980. “The Dilemma of Choice in the Deathcamps.” Centerpoint: A Journal of
Interdisciplinary Studies 1, no. 1: 53–59.
Lemkin, R. 1944 [2008]. Axis Rule in Occupied Europe. 2nd ed. Clark, NJ: Lawbook Exchange.
Rosefielde, S. 2009. Red Holocaust. Milton Park, UK: Routledge.
Stanton, G. 1996. “The 8 Stages of Genocide.” Originally a 1996 briefing paper, later updated
to “10 stages.” See http://w ww.genocidewatch.org/i mages/8StagesBriefingpaper.pdf and
http://genocidewatch.org/genocide/tenstagesofgenocide.html [both accessed September
16, 2014].
Stigler, G. J., and G. S. Becker. 1977. “De Gustibus Non Est Disputandum.” American Economic
Review 67, no. 2: 76–9 0.
Totten, S., and W. S. Parsons, eds. 2013. Centuries of Genocide. 4th ed. London: Routledge.
14
Identity and Incentives

An Economic Interpretation of the Holocaust
R au l C a ruso
14.1. Introduction
The Holocaust was brutal and utterly inhumane. The deliberate murder of mil-
lions of Jews and other supposedly subhuman peoples became, in time, the ulti-
mate goal of an entire societal system, guided by Adolf Hitler and the National
Socialist German Workers Party (the Nazis). A gigantic predatory enterprise,
the Holocaust was shaped and engineered by complex institutional machinery.
Contrary to still widely held public opinion, a substantial number of purposeful
(“rational”) actors played specific roles in bringing about an outcome that can be
meaningfully interpreted in an economic conceptual framework. Specifically, this
chapter proposes an interpretation of the Holocaust along the lines of economic
theory and public choice: The formulation of economic policies and the provi-
sion of economic incentives contributed heavily to the genocide. In this respect,
the events of the Holocaust are congruent with Raphael Lemkin’s definition of
genocide as a “coordinated plan of different actions aiming at the destruction of
the essential foundations of life of national groups, with the aim of annihilating
the groups themselves” (1944, 79).
It is now widely accepted that a large proportion of Germans were aware of
the evolving genocide and took part in the Holocaust, from its very early stages
to its final realization (see, e.g., Gellately 2001; Glass 1997). This aspect of indi-
vidual participation lends itself to economic analysis, which takes as its point of
departure the purposeful (“rational”) behavior of agents that then coalesces into
an aggregate social outcome. That said, the foundational concept of this chapter
is that the Holocaust can be explained only by including nonmaterial, identity-
related and ideological, components in the utility functions of the actors involved
in the gigantic destruction (see Anderton 2014; also see c hapters 21 and 22 in this
volume). As explained in Akerlof and Kranton (2000), the inclusion of identity
318
Identity and Incentives 319
into the utility functions of individuals is payoff-relevant and generates externali-

ties affecting other persons’ behavior. Thus, Nazi anti-Semitism influenced the
behavior of large numbers of individuals, whether they were Nazi officials or not.
The relevance of Nazi identity, and of its social acceptance, is therefore crucial to
the understanding of individual and collective behaviors and choices, which may
appear to be “irrational” if considered only in light of material utility.
The creation and establishment of identity and related, hateful behavior, such
as that associated with Nazism, is a matter of political economy (Glaeser 2005).
Nazi identity, centered on the concept of racial supremacy, was reinforced by per-
vasive propaganda. Nazi political entrepreneurs devoted themselves to increas-
ing the supply of hatred against Jews and other minorities (on propaganda, see
chapter 12 in this volume). In Glaeser’s model, individuals are likely to lie about,
manipulate, and falsify information if they have little or no incentive to learn the
truth. Therefore, the study of the Nazi-supplied incentive structure also is cru-
cial to the understanding of why thousands of individuals chose to embrace the
Nazi identity and accepted the dehumanization of Jews as a social prescription
guiding their own identity and consequent behavior. The horrors perpetrated in
Auschwitz and other extermination camps would not have been possible with-
out the dehumanization facilitated by specific propaganda intended to instill and
enhance an obsession with racial supremacy and hatred in a large quota of the
population. Conversely, dehumanization would not have been sustainable in the
absence of incentives secured and implemented by complex institutional machin-
ery. Therefore, the line of reasoning followed in this chapter takes into consider-
ation both material and immaterial, identity-related, components of utility.
Material motivation was by no means trivial. The material dimension mat-
tered. Nazis were motivated by economic incentives (Brustein 1996), and
German citizens heavily benefited from Nazi policies (Aly 2007). It was a two-
way street: opportunistic individuals benefited from the Nazi’s organization of
the Holocaust and, to evoke participation, the Nazis created material incentives
in both public and private spheres. The interplay between material incentives and
identity is therefore the guiding principle of this chapter. In addition, the chapter
also suggests that the individual behavior of Nazis can be understood better by
taking into account some aspects of Hitler’s approach to public choice, in par-
ticular with regard to incentives emerging for both public bureaus and private
business organizations.
The chapter is organized as follows: section 14.2 explores the themes of iden-
tity and incentive structure. Section 14.3 discusses forced impoverishment and
the Aryanization (expropriation) of Jewish businesses, and section 14.4 is an eco-
nomic interpretation of forced labor and the role of extermination camps. Section
14.5 concludes with a distillation of the main points of the analysis. The hope,
in part, is twofold: first, that economists come to see how their approach may be
applied to understand, and ideally help to prevent, the horrible crime of genocide
320 Case Studies I
and other types of mass atrocities; and second, that genocide scholars and geno-
cide prevention policymakers include more theory-based economic analysis into
their tool kits of scholarship and policymaking.
Three limitations have to be highlighted. First, I do not explicitly analyze the
creation of ghettos and the abhorrent massacres perpetrated by Einsatzgruppen
(paramilitary mobile-death squads). The ghettos were designed as tempo-
rary measures before extermination took its final shape in the camps. The
Einsatzgruppen operated from late 1939 with Germany’s invasion of Poland.
They accompanied regular army troops as they advanced through Poland and
carried out massacres—usually executions by shootings—of the Polish intelli-
gentsia and other leaders and of Jews, Romani, prostitutes, and the physically
and mentally ill.1 Later, in 1941, the Einsatzgruppen accompanied the German
army in its invasion of the Soviet Union, to similar purpose. In contrast to
these mobile extermination forces, my interest lies in the stationary extermi-
nation camps (section 14.4). Second, the chapter is based almost totally on the
German scenario, even as the destruction of Jews and other victims took place
all over Europe. In focusing on Nazi Germany, however, the analysis does not
lose generality. Third, the literature on the Holocaust is vast and cannot be reca-
pitulated here. Readers are presumed to possess adequate ex ante knowledge of
the Holocaust. I use selected facts, data, and figures in order to remark on some
aspects of the whole line of reasoning. The goal here is to provide readers with an
economic interpretation of selected aspects of the Holocaust, and the approach of
the chapter has to be considered inherently conceptual.
14.2. Identity and Incentives

Economists have begun to include identity in models of behavior. The pioneering
work in this respect is Akerlof and Kranton (2000), who show how identity can
be modeled to enter utility functions of individuals. Identity is payoff-relevant
and generates externalities affecting other persons’ behavior. The general form
of an identity-based utility function for an individual, j, is Uj = Uj (aj, a-j, Ij),
where aj and a-j denote, respectively, the vectors of j’s actions and the actions of
others (who are not j, or -j). Ij denotes j’s self-i mage or identity. This self-i mage,
from which an individual derives utility or satisfaction, is, in turn, defined as Ij =
Ij (aj, a-j; cj, εj, P). In words, j’s identity depends on his or her own actions and that
of others as well as on social categories, cj, and on the extent to which his or her
own characteristics match with these categories, ε j. The identity also depends on
sets of prescriptions, P, or behaviors deemed appropriate for each of the asso-
ciated social categories. Each individual j maximizes utility by taking as given
the actions of others (a-j) and also of cj, ε j, and P. (Note that cj may be partially
determined by j’s choices.)
Several implications follow from this setup. First, utility payoffs from behav-
ior depend upon one’s own actions and on others’ actions; second, third par-
ties can shape persistent changes in these payoffs; third, individuals can choose
their identity; and fourth, self-image is one driver of behavior. Whenever an
individual identifies him-or herself with a social group, he or she will choose
actions that match with the prescriptions of that social group. If someone
chooses a different action, not congruent with the group, she or he would likely
see a loss of identity associated with that social group, leading to a decrease
in total utility. In our context, imagine a set of two social group categories, c1
and c2, say “Nazi” and “non-Nazi.” If the social category for an individual j is
“Nazi,” a corresponding set of behavioral prescriptions is associated with that
category. Among the expected appropriate behaviors might be that individual
j publicly manifests anti-Semitism, and j’s utility would increase as a result.
Symmetrically, if someone chose not to engage in public anti-Semitism, she or
he might not be perceived as a “true” Nazi, thus suffering a utility loss. Utility
gains and losses depend in part on how other people interpret j’s actions. An
individual suffers decreased utility if other people do not interpret his or her
actions as the behavior of a true Nazi.
Modeling identity this way implies several things. First, it necessarily implies
externalities. The actions of any one individual affect the behavior of others.
Individuals do choose their own identity—and choice of identity is the crucial
choice for the entire panoply of behaviors and personal well-being—but it is a
choice influenced by others. Second, the way an individual maps her or his own
social categories need not overlap with the social categories others assign to the
individual. Third, each individual may be mapped onto several categories (e.g.,
an individual j is both a Nazi and a woman). And fourth, adding identity to the
utility function carries remarkable implications, in particular when considering
how much effort is likely to be exerted when undertaking activities associated
with the Nazi mission. Thus, in a follow-on paper, Akerlof and Kranton (2005)
show that identity is likely to generate higher individual effort as they partici-
pate in organizations. In our context, individuals choose not only whether or not
to behave as true Nazis, they also choose the level of effort they exert. If choos-
ing between low and high effort, an identity-based choice is predictably that of
high effort. For example, Nazi brutality against Jews would have been predicted
to be higher if perpetrated by a committed insider of the SS, the Schutzstaffel (of
which the aforementioned Einsatsgruppen were a part). What is remarkable here
is that the interaction between identity and monetary incentives differs from
the traditional economic script: true Nazi individuals exert a high level of effort
irrespective of monetary incentive. Put more accurately, there could be cases
where identity and monetary incentives are complements rather than substitutes. In
the first case, the desire of money complements the ideal wishes, whereas in the
latter they can replace each other.
322 Case Studies I
14.2.1. Investing in Identity

Under an identity-based model of utility, one has a clear-cut rationale for investing
in identity, and the Nazi leadership can be conceived of as quite rationally devot-
ing a large amount of resources to propaganda. Needless to say, identity-based
behavior does not take shape overnight, and the Nazi leadership would have
had to consider several factors such as the historical roots and origin of German
anti-Semitism,2 the production and recurrence of hatred in the interwar period
through propaganda, and the persistence of beliefs in anti-Semitism and racial
supremacy. The traditional attitude against Jews constituted the backbone on
which the Nazi plan of destruction rested. For example, Voigtländer and Voth
(2012) find historical continuity of anti-Semitism in Germany stretching over
600 years. They show that Nazi violence against Jews, and political support for the
National Socialist German Workers’ Party (hereafter NSDAP), can be predicted
by considering the regions where anti-Jewish pogroms were carried on at the time
of the Black Death in 1348–1350. 3 While German anti-Semitism was not created
by Hitler, it certainly soared under his leadership of the NSDAP.
Akerlof and Kranton write that “[i]‌ndividuals may—more or less consciously—
choose who they want to be” (2000, 717). But the choice can be influenced by
propaganda, a pillar of the Nazi regime.4 Glaeser (2005) analyzed how political
entrepreneurs can increase the supply of hatred against a minority. In his model,
falsification and lies prevail when individuals have little or no incentive to learn
the truth. They do not update their beliefs if they do not have incentives to do
so. Taken together, then, it is not farfetched to claim that Germans, and other
Europeans who contributed to the Holocaust, chose to participate in the atroci-
ties in part because incentives were engineered both to invite people to join in the
Nazi ideology as well as to prevent individuals from challenging the prescriptions
of Nazi identity. Propaganda functioned to create Nazi beliefs and to make them
persistent.
14.2.2. Persistence of Identity: Moral Disengagement

and Organizational Structure
From identity per se, attention shifts to its persistence. A powerful engine of action
may be one’s need to support or to strengthen one’s self-i mage. That is, individuals
may choose actions to confirm their own identity not only with respect to oth-
ers but also with respect to themselves. In an early article, Akerlof and Dickens
(1982) apply to economics the theory of cognitive dissonance originally developed
by Festinger (1957). The basic idea is that people feel uncomfortable with con-
flicting, dissonant cognitions. Feasible reactions to dissonance are changing one
or more beliefs, acquiring new information to increase consonance, and/or reduc-
ing the importance of dissonant cognitions. Individuals may be expected to select
or to manipulate information in order to confirm the foundational beliefs of their

own identity. In our context, this could be crucial. Individuals likely confirmed
their beliefs about Nazi identity, racial supremacy, and dehumanization of Jews
in spite of the available information on inhumane massacres and the developing
Holocaust. The firm belief of being “on the right side of things” led Nazi Germans
and other perpetrators to confirm their beliefs, reducing dissonance.
Dissonance would originate from any challenge to the dehumanization of
Jews. Dehumanization was a fundamental component in the creation of moral
disengagement from the Holocaust. Moral disengagement is a psychosocial
mechanism by which moral self-sanctioning is selectively disengaged from one’s
inhumane conduct (Bandura 1999), with the consequence that when individuals
engage in inhumane behavior, they do not feel responsible for it—a characteristic
feature of Nazi Germany. Not surprisingly, moral disengagement was magnified
inasmuch as the victims were not, or no longer, considered as human beings. The
Nazis were so effective, in part, because the system was designed to produce moral
disengagement on a large scale. As noted, the state’s production of hatred shaped
beliefs of anti-Semitism and racial supremacy, thus favoring the dehumanization
of victims.
Moral disengagement alone does not suffice to explain the emergence of
large-scale destruction. Complementary to dehumanization was organizational
structure. From the very beginning, the predatory and exterminatory policy was
characterized by an increasing number of public agencies committed to carry on
tasks related to expropriation and eventually to extermination. Crucial to the
facilitation of large-scale moral disengagement was a hierarchical organizational
structure based on the diffusion of responsibility. At least since Milgram (1974),
it is widely accepted that when people interpret their actions as originating from a
legitimate authority, they do not feel personally responsible for their actions and
their consequences. This effect also occurs in the displacement of responsibility
when it derives from the division of labor (Bandura 1999, 2010). That is, fragmen-
tation of tasks and duties reinforces moral disengagement. Experimental findings
confirm that moral disengagement is more likely when a principal can hire an
agent to carry on immoral actions. 5
14.2.3. Discontinuities
Clearly, identity-related, incentivized payoffs are relevant and can help to explain
behaviors that otherwise might appear to be irrational, if one follows the tradi-
tional economic script. That said, one has to acknowledge that identity-based
utility functions may contain discontinuities. An extreme case that deserves to
be mentioned is that proposed by Bernholz (2001) and Hillman (2010). Their
approach highlights the role of supreme values in shaping behavior. Supreme val-
ues are characterized by a lexicographic ordering of preferences that does not
324 Case Studies I
allow for any tradeoff in objectives. That is, an individual prefers an amount of
one good, X, to any amount of another good, Y, and if offered several bundles of
goods, the individual will choose the bundle that offers the most X, no matter how
much Y there is to be had in any other bundle. When, for instance, the supreme
value concerns the utter destruction of a rival community, agents’ subsequent
behavioral choices are immutably shaped by that objective. Therefore, material
utility becomes essentially irrelevant. In the lexicographic ordering of Hitler and
other Nazi leaders, annihilation and destruction of Jews was the overriding objec-
tive; utility derived from other objectives appeared to be infinitesimally small.
This view sheds some light on certain crucial choices made that translated into
the Holocaust. Echoing Bernholz (2001, 35): “Why should Hitler devote scarce
transportation facilities and armed forces to transport people to Auschwitz,
which both were badly needed to support the struggling German armies?” The
rhetorical question clarifies the perceptual lens through which many Nazi choices
have to be interpreted.
Whether or not every perpetrator in the Holocaust had lexicographic pref-
erences can be questioned, of course, and is handled in theory by splitting the
group of perpetrators into two representative types, one whose utility function is
characterized by lexicographic ordering, and another whose utility function still
incorporates identity but without lexicographic ordering. For reasons of expo-
sition, the first group may be thought of as members of the SS or others highly
placed in the Nazi orbit. Led by Heinrich Himmler, the SS was widely regarded as
the embodiment of Nazi ideology. In light of the general identity model outlined
above, actions by SS members generated externality effects, influencing identity-
driven behavior of less-t han-f ully committed stalwarts of the system. In practice,
even a small group of fervent Nazis could have significantly influenced the behav-
ior of all Germans.
14.2.4. Material Incentives in the Public Sector

As noted, in addition to identity-d riven behavior, material incentives played
a significant role. The inhumane behavior of many Nazis was reinforced by
material incentives. Both public servants and private actors benefited from the
Nazi system. The system of incentives among public servants is a crucial topic.
Presumably Hitler did not want to build up an entirely efficient bureaucratic
structure. Given lexicographic preferences, any efficiency objective had, for him,
only an infinitesimally small value as compared to his primary objective of the
destruction of Jews and other minorities. Efficiency was of value, but not for its
own sake, and only inasmuch as it contributed to the destruction. Based on the
concept of vertical trust networks, Breton and Wintrobe (1986) elaborate on the
competitive nature within the Nazi bureaucracy.6 The approach relies on the idea
that superiors and subordinates trade with each other. Superiors seek informal
and uncodified services from subordinates who, in exchange, seek rewards such
as rapid promotion. Vertical trust networks support competitive and creative
efforts of superiors to make them stand out within the bureaucratic apparatus.
The contest among top bureaucrats was neither erratic nor unwanted. A leader
can inflame competition by means of apparent imprecision. The very lack of pre-
cise orders generates competition because different and even unconventional
responses may emerge from various individuals, branches, and agencies. In the
presence of vague goals—other than the final goal—superiors seek creative and
novel solutions to accomplish the overarching goal of the leadership. With vague-
ness comes a degree of arbitrariness but also of autonomy. Deliberately vague
leadership can be expected to be more effective when the bureaucracy is quite
large. Narratives on Nazi hierarchy confirm this idea. Hitler himself favored com-
petition among agencies and bureaucrats by issuing often informal and imprecise
orders. Bureaucratic entrepreneurship and competition were magnified in light of
the lexicographic preferences of the Nazi aristocracy: “It was widely appreciated
throughout the Nazi bureaucracy that, in the eyes of the political leadership, ‘solv-
ing’ the Jewish question had a priority that was second only to war, and possibly
not even second to that” (Breton and Wintrobe 1986, 911). This attitude was not
only Hitler’s. For example, Aly (2007) reports that the Nazi’s long-serving minis-
ter of finance, Lutz von Krosigk, launched a brainstorming contest among officers
to find innovative ideas for the expropriation of Jewish property.
Mixon, Sawyer, and Trevino (2004a, 2004b) present anecdotal and empiri-
cal evidence on the competitive aspects of Nazi bureaucracy. They highlight
that orders issued by Reinhard Heydrich at the Wannsee Conference in January
1942 (more on this later) were imprecise regarding the nature of the “solutions”
to be implemented and on the number of agencies to be involved. At Wannsee,
among the fifteen attendees, at least eleven branches of the Nazi hierarchy were
represented. Both the imprecision of the orders and the large number of agencies
involved appear as concrete examples of the attitude of top Nazi officials to favor
competition among bureaus for accomplishing the ultimate goal of the leader-
ship. In line with the theoretical approach of Breton and Wintrobe (1986), the
implementation of the “final solution” activated a number of career advancement
prospects within the Nazi hierarchy. This affected the final outcome: Mixon and
King (2009), for example, establish empirically that across Europe the number
of Jews killed was higher in countries where vertical trust networks were active
as compared to countries where the Holocaust was carried on mainly through
German coercion.
14.2.5. Material Incentives for the Private Sector

Complementary to incentives for bureaus and public servants, and just as crucial
to the Nazis, were incentives generated for private businesses. The Nazi economic
326 Case Studies I
system generated incentives for a plethora of private agents. In general, the Nazi
model for the economy can be described as a commanded private economy. Even
though the private property of business of most Germans was secure, the general
direction of the economy did not depend on the choices of private entrepreneurs.
As explained in Scheweitzer (1946, 5): “Profit as stimulus for private owners was
retained but greatly modified. Profit as a guide for the general direction of the
economy was suppressed and its place was taken by the economic plans of the
government.” Economic activity became ancillary to the general principles guid-
ing the regime. In particular, the Reich implemented a war economy, designed for
mobilization for a future war (Scherner 2013).
The Nazi economic policy mix included in its Four Year Plan after 1936 was
based on (1) massive military spending, (2) privatization, (3) mandatory cartel-
ization, (4) fixed profit margins and administered prices, and (5) Aryanization
(see section 14.3). The need for rearmament to prepare for a future war was the
main objective of Nazi economic policy. To carry out the rearmament, the gov-
ernment implemented a large-scale privatization policy (Buccheim and Scherner
2006; Bel 2010; Scherner 2013), which, additionally, turned out be a powerful
instrument to increase the support of, and web of relationships with, industri-
alists and various business sectors. For example, Voth and Ferguson (2008)
have recently shown that profitable firms were heavily interconnected with the
NSDAP. In addition, in 1936 a Cartel Law was approved, leading toward com-
pulsory cartelization, which took effective shape two years later (Newman 1948).
Such policy mix turned out to be beneficial, especially for top-income earners. As
reported by Dell (2005), between 1933 and 1938 the share of earnings for top-
incomes grew amazingly: more than 50 percent growth for the top percentile and
more than 150 percent for the top 0.01 percent.
Even for the less-a ffluent, consumption was ideologically influenced as well.
As Wiesen writes: “[C]‌onsumption was to serve a higher purpose, namely the
enrichment of the Volk during its struggle for global and racial dominance” (2011,
36). On the one hand, consumption was intended to improve people’s material
well-being; on the other hand, it was supposed to be based on the needs of the
state. Consumers were expected to take into account collective and racial aspects
when deciding which bundle of goods to purchase.
Apart from public servants and private businesses, one long-lasting concern
of the Nazi leadership was to keep the ideological consensus alive by providing
direct economic benefits to German citizens. The unifying theme of Brustein
(1996) and Aly (2007) is that economic benefits constituted the leading motive
for most Germans who embraced the Nazi experience. Brustein (1996), in partic-
ular, analyzes the early years of the Nazi Party, showing that affiliation with it fol-
lowed from rational economic decisions that emerged in certain social categories.
Aly (2007) highlights the direct benefits for German households from economic
policies, the dispossession of Jews, and the tolerated plundering of occupied
territories. Interestingly, the social unit that benefited most was the family: Aly
reports on family policies: “[F]‌a mily and child tax credits, marriage loans, and
home-f urnishing and child-education allowances were among the measures with
which the state tried to relieve the financial burden on parents” (2007, 38).7
In sum, economic models that combine material and immaterial (i.e., identity)
aspects of individual utility functions seem well worth the effort. They combine,
and more tightly link, factors that underlie human behavior—and in that may lie
future insights into how genocide comes about, or how it may be prevented.
14.3. Forced Impoverishment and Aryanization

The previous section included a sketch of the economic benefits accruing to
Germans from the pursuit of the Holocaust. This section examines an early phase
of the Holocaust that saw the forced impoverishment of Jews (mainly) by means
of predatory economic policies. To quote Lemkin: “The destruction of the eco-
nomic existence of a national group necessarily brings about a crippling of its
development, even a retrogression” (1944, 85). Economic impoverishment took
shape through two main channels, heavy taxation and dispossession of Jewish
assets and businesses, or “Aryanization.”
With regard to taxation, two main measures were applied. The first is usu-
ally referred to as the “flight tax.” Introduced in 1931, its original purpose was
to limit capital flight, but in 1934 it became the main instrument to dispossess
Jews who wished to leave Germany: Emigrants paid a tax equal to one-quarter
of their assets.8 Associated with the flight tax was a tight control on foreign
exchange. Emigrating Jews had to open a bank account with the Gold Discount
Bank (a branch of the Reichsbank), which charged a fee (or “discount”) for the
conversion of marks into foreign currency. The discount charge was 20 percent
in 1934, 68 percent in 1935, 81 percent in late 1936, 90 percent in June 1938, and
96 percent once the war had started (Feldman 2007). In late 1938, an additional
Atonement Tax of 20 percent on registered assets was imposed on all German
Jews. To emigrate, Jews needed to demonstrate the payment of both the flight tax
and the atonement tax. Taxation was so prohibitive that it prevented rather than
encouraged the emigration of Jews.
Even more effective as a policy of forced impoverishment was Aryanization, the
process of Jewish expulsion from economic life in Germany and the occupied ter-
ritories. It is widely accepted that Aryanization can be divided into two phases.
The first lasted from 1933 into 1938, with sales of Jewish businesses somewhat
“voluntary” in that Jews were allowed to bargain with potential buyers. One of the
effects can be seen in the decreasing number of small-scale businesses. According
to Stargardt (1944), between 1936 and 1938 the number of one-man handicraft
plants in Germany decreased by 153,390 units, predictable when considering that
328 Case Studies I
46 percent of the Jewish community was involved in small businesses (Hilberg

1985). In the second phase, after November 1938, the sale of Jewish businesses
became compulsory. In one of his many Nazi leadership roles, Hermann Göring
issued a decree on the exclusion of Jews from German economic life, dictating
for Jews to go out of business by January 1, 1939. Jewish enterprises were to be
put under the control of administrators through state agencies and economic
organizations.9
Conceptually, particularly for the first phase, Nazis and Jews can be mod-
eled as agents involved in a conflict over economic activities. Nazis committed
to expropriate Jewish businesses, whereas Jews intended to protect them. To
interpret the interaction between Nazi and Jew, one can refer to the copious lit-
erature on conflict economics (see, e.g., Hirshleifer 1988; Skaperdas 1992; Dixit
2004; Garfinkel and Skaperdas 2007).10 Based on general equilibrium models
of continuing conflict, this literature depicts noncooperative scenarios in which
rational agents struggle over the redistribution of potential income. In its sim-
plest form, for example, two rational agents both are in possession of positive
endowments and technological capabilities at a given point in time. Each agent
allocates resources toward productive and conflictual activities (“butter” and
“guns,” respectively).11 The chosen levels of resources invested in productive and
predatory activities determine the social outcome of the conflict, and the result-
ing social state is Pareto-inferior to a social state without conflict.
To apply conflict economics to the Nazi case, it is necessary to consider that
they had the advantage of taxing Jewish economic activity, thus imposing an
additional cost on the other party to the conflict. Borrowing from a model first
developed in Caruso (2012), which enriches the basic Hirshleifer-style model to
include taxation, one can depict an interaction between a predatory government
and another social group. Both agents allocate resources to contested or uncon-
tested production. Contested production is the fraction of economic activity that
is contested between government and the other group; uncontested production
is the fraction of economic activity that is safe from appropriation. In principle,
the government can either be benevolent or predatory, depending on taxation
and redistribution. Taxation and redistribution define the type of government
and they are treated as given parameters. In the context of this chapter, Nazis and
Jews had a conflictual interaction over a fraction of economic activity, namely the
businesses that were subject to Aryanization. In addition, the Nazi government
was predatory because it imposed a heavy fiscal burden on Jewish entrepreneurs.
In a simple general-equilibrium setting, the choice variables are the resources to
be allocated, by both agents, to unproductive conflict activities in the contested
business and in the productive sectors. The government chooses its optimal level
of conflict and production given its type. As taxation imposed on the other party
increases, investments in productive activities decrease. The model predicts
that the economy becomes impoverished because of the excessive amount of
resources devoted to conflict. In the context of Aryanization, the expropriation

efforts undertaken by Nazis, and Jewish resistance, diverted resources from pro-
ductive activities. In the long run, this turns out to be detrimental for the econ-
omy as a whole. All this may seem quite straightforward, but the point is that in
spite of appropriation—in spite of German businessmen being better off because
they were allowed to purchase Jewish businesses at low cost—t he whole of the
German economy nonetheless suffered from Aryanization. (Of course, Jews were
even worse off.)
14.4. Forced Labor and Extermination Camps

Coming to power in 1933, the Nazis established a large number of detention
facilities, so-called concentration camps. Inmates—a ll kinds of inmates, not just
Jews—were used as cheap labor for both SS-related businesses and for private
firms. Between 1939 and 1941, the supply of forced labor was not specifically
organized to benefit the war economy. But once the German war effort placed
significant demands on the arms industry, the camp system, under the control of
the SS, was expanded substantially to provide inexpensive labor to key firms.12
Overy (1988) reports in detail how the total mobilization for war created severe
labor shortages, dictating the need to rely on slave labor. However, as is detailed in
many studies (e.g., Herbert 2000; Spoerer and Fleischhacker 2002), the analysis
of forced labor under the Nazis has to take into consideration the ethnic identity
of inmates and their differential probability of survival, as a hierarchy of ethnici-
ties determined the set of rules and the conditions of life. At the top of this ranking
were inmates considered close to Germanic identity. At the bottom were Poles,
any Soviets, Gypsies, and Jews. The eventual destiny of the latter was destruction
in spite of any productive need. Because of the zero probability of intended sur-
vival, Ferencz (1979) coined the term “less than slave” laborers for them.
For economists, such distinctions offer perspectives for modeling. When con-
sidering forced labor with some positive probability of survival, a recent, insight-
ful theoretical model of labor coercion is of interest (Acemoglu and Wolitsky
2011). The model enriches the classic structure of a principal-agent relationship
in that the laborer (agent) has no wealth and the employer (principal) can choose
a level of coercion (a credible threat of punishment by using armed guards and
enforcers). The latter assumption is crucial because it removes voluntariness of
the laborer in the relationship with the employer. That is, under the credible threat
of violent punishment the laborer would of course “accept” terms of employment
that she or he would otherwise be expected to reject.
There are four main results of the model. First, coercion increases efforts of
the agent. Second, since more productive employers use more coercion to induce
higher effort from laborers, coerced laborers are better off with less efficient
330 Case Studies I
employers. (This contrasts with the traditional principal-agent model.) Third,

ex ante investments in coercion allow employers to avoid payments to induce
higher effort from the laborers. And fourth, coercion is always socially inefficient
and therefore detrimental.
In the slave labor literature, employers are assumed to have the incentive to
let slaves survive. This would be the rational (purposeful) choice of an employer
whose objective is to secure a maximum level of profit. In contrast, in the
Holocaust, the SS were committed to summarily killing Jews. Concentration
camps were joined by extermination camps. (Hitler appears to have made
up his mind on the complete extermination of Jews in December 1941—see
Longerich n.d., and chapter 17 in this volume—which led to the infamous
Wannsee Conference in January 1942 on the interagency coordination of the
“final solution to the Jewish question.”) But when firms’ labor demand became
compelling and the use of forced labor became a priority, “the firms had to
use all their influence and persuasion to get all the help they felt they needed.
The private companies were to pour millions of marks into the coffers of the SS
for the privilege of using the camp inmates” (Ferencz 1979, 24).13 In brief, a trade-
off emerged between the SS’s desire for extermination and firms’ desire for the
use of forced labor.14 For example, Reinhard Heydrich—until his death, in June
1942, considered to be among the most fearsome of the Nazi leaders (he chaired
the Wannsee conference)—stated that “although the relevance of economic con-
siderations is, of course, recognized, any attempt to postpone the question of
racial and ethnic culture until after the war must be firmly rejected” (Herbert
1993, 167). Germany had to be Judenrein (in literal translation: “clean of Jews”).
According to Goldhagen (1996, 291), in September 1942 Hitler did not authorize
SS-leader Heinrich Himmler and armaments minister Alfred Speer to transfer
Jewish forced laborers from camps in occupied territories back to Germany, a pol-
icy Hitler partially relaxed only in April 1944, when labor needs within Germany
had grown to pressing levels. Himmler himself, when asked to grant concessions
for skilled Jews, also was firm in ordering that Jews could be employed only in
large camps run by the SS “but even there—in keeping with the wishes of the
Fuhrer—Jews must vanish one day” (Herbert 1993, 175).
To effect the annihilation of Jews, large numbers of them were dispossessed,
deported, and murdered immediately.15 For example, in March 1943 when the
first transports of 2,757 Jews deployed to armament firms were registered in
Auschwitz, 1,689 were killed immediately (Herbert 2000). Earlier, between
late 1941 and 1943, the Aktion Reinhard (also spelled Reinhardt)16—designed to
murder all Jews residing in the Polish Generalgouvernement—was based on the
deporting of Jews to the purpose-built extermination camps of Belzec, Sobibor,
and Treblinka, resulting in the loss of perhaps 1.7 million human lives (Black
2011). The largest extermination camp was Auschwitz II-Birkenau—Auschwitz
consisted of a complex of camps—where between 1943 and 1944, an average of
6,000 Jews were gassed to death daily. The total death toll there was well in excess
of one million people. Their everyday treatment was shaped by inhumane brutal-
ity. Kept in extremely poor conditions, lack of hygiene and inadequate food sup-
plies led to a very high mortality rate. Herbert (1993) reports, for example, that in
Buchenwald the death rate rose from 10 percent in 1938 to 36 percent in 1941; in
the camp of Mauthausen, it was 76 percent in 1940.
Obsession with extermination is the most significant departure from the tra-
ditional script of rational economic behavior, in that the SS could have behaved,
had it chosen to do so, as a monopolist supplier of camp inmates to firms and
could have charged a high price from labor-seeking firms. Monopolistic pricing,
however, would have been inconsistent with the overriding objective of extermi-
nation. Therefore, under the assumption that extermination was beyond ques-
tion, the crucial point of interest in the economics of camps is how to determine
the optimal use of forced labor so as to secure both extermination for the SS and
an adequate supply of labor for German industry. In this scenario, a modeling
approach would consider inmates only as “consumable” productive inputs—raw
material—and not as human beings. In this end stage of the Holocaust, humans
were effectively considered as perishable inputs, the final outcome of the dehuman-
ization process.
The interaction between the SS and any given firm seeking forced labor
resources can then be modeled as a game (see the Appendix for technical
details). The agents are the SS, as the supplier of forced labor, and any one firm,
as demander. The variables to be solved for in the model are the labor fee to
be paid by the firm to the SS and the length of time a victim survives (dura-
tion). As monopolist, the SS chooses the fee, discounted over time, whereas
the private firm, because of the work-to-death arrangement, effectively deter-
mines the duration of survival. Since the victims are interpreted as paid-for
but perishable inputs, the firm wants to optimally balance the costly use of
forced labor with their certain death. Thus, the protocol is the following: (1)
the SS demands a fee for supplying inmates; (2) the private firm observes (and
accepts) the fee demanded and chooses the duration of the “useful life” of the
perishable input, the period over which labor is kept alive before having been
worked to death; and (3) the game ends and the payoffs are determined. It
turns out that optimal duration depends on an evaluation made by the firm
in regard to inmate health. The weaker are the prisoners to begin with, the
shorter their remaining “useful lives.” Conversely, the higher is the expected
contribution of forced labor to the firm, the longer the period of imprisonment
and forced work. Interestingly, the model predicts that the fee to be paid does
not increase over time. Instead, the optimal SS strategy is to set the current
fee equal to the past fee, meaning that fees charged for the use of enslaved
people are kept invariant (which is a testable proposition). In the model, the
SS has no incentive to raise the fee because a high fee is linked with firms’
332 Case Studies I
then-corresponding need for a longer duration of inmate survival in order to

“recoup” the fee. It follows that, rather than behaving as a prototypical profit-
maximizing supplier of workers (like a modern-d ay labor union or staffing
agency, say), the SS is predicted to keep the fee low so as to secure the speedy
extermination of its victims.
Of course, this sort of modeling is extremely distasteful, to say the least.
However, if one wishes to further understand the mechanics of the Holocaust
and, by extension, the mechanics of other genocides and mass atrocities, then
deriving model-based, empirically testable hypotheses is important. At that point, the
model likely includes the key variables driving the behavior under observation
and, with any luck at all, it is behavior that is amenable to policy intervention for
atrocity prevention or, at least, mitigation.
14.5. Conclusion
This chapter proposes an economic interpretation of certain aspects of the
Holocaust. Three main conclusions can be drawn. First, the study of the
Holocaust can be enriched if one considers economic incentives as comple-
mentary to motivations of ideology and racial supremacy. Among scholars of
the Holocaust, there is a separation between supporters of the supremacy thesis
as the main motivating engine of Nazi behavior and those who take economic
incentives and rationales into account. In reality, it is possible to combine, and
reconcile, these two only apparently diverging approaches. One channel to do
that lies in applying insights from recent developments in economics and public
choice with specific reference to the inclusion of identity into the utility func-
tions of atrocity perpetrators. Therefore, in this chapter I discuss how the inclu-
sion of identity and expressive components could have favored the emergence of
the Holocaust.
A second conclusion is that the analysis here strongly supports the idea that
the Holocaust could not have resulted in a net positive payoff for the perpetrators.
A recurring idea in the literature is that the Nazis might have benefited from the
genocide had the war turned out differently than it did (or might have benefited
the Nazis, even if they did lose the war, as Lemkin surmised). The Aryanization
model discussed in the chapter rejects this idea. In light of general equilibrium
approaches, the reliance on forced labor, and also the process of expropriation,
was detrimental to the German economy as a whole. Simply put, unproductive
and destructive activities were exceeding productive and constructive ones, a
setup no economic system can sustain over the long run (see Baumol 1990). In
brief, the economy of Nazi Germany would have collapsed in any case. Needless
to say, the latter proposition is not intended as an intellectual relief from the inhu-
mane horror perpetrated against millions of victims; it only reinforces the idea
that a traditional approach needs to be combined with a wider approach, such as

the inclusion of identity-d riven behavior.
And third, much work can be done examining internal conflict within the
Nazi regime, as illustrated with the example of forced labor demand and supply
in s ection 14.4. A better and more precise understanding of the drivers of such
conflict, and of their own underlying incentive structure, may provide insight
into crucial levers to affect behavioral change and thus assist policymakers in the
prevention of future atrocities.
Acknowledgments
This chapter benefited from conversation with Peter Bernholz, Arye L. Hillman,
and Michele Grillo. Special thanks go to Charles Anderton and Jurgen Brauer for
comments, critique, support, and patience. All errors are mine.
Appendix
The following model can be used to interpret the interplay between the SS and
firms with regard to the exploitation of less-t han-slave camp inmates. Let R and
r denote, respectively, the current fee and past fee charged by the SS for forced
labor, and let δ denote the time preference or discount factor. Utility for the SS is
a function of fees paid by the employer (the firms) and the discount factor, such
that current utility is given by U = U(R, δ). The SS is assumed to maximize an
objective function that spans the difference between the utility given by the fee
to be attained and an ideal reference utility that depends upon the past fee. The
objective function the SS wishes to maximize then is:
Z = (U ( R, δ ) − U ( r , δ ))2 = ( R δ δ − r δ δ )2 . (1)
The objective function of the labor-employing firm is a loss function to be

minimized. Let V be the expected contribution provided by the laborers. Let M
denote labor’s diminishing value once it had been seized by the SS, and assume
that the value of labor to firms decreases in the amount of the labor fee (R) and
in the duration of imprisonment, denoted by t ε(0,1). Imprisonment itself leads to
prisoner deterioration and decay, the degree of which is captured with a param-
eter, h. This narrative setup can be succinctly summarized as follows:
M(V , R , t , h) = V / t (1/ h) – R . (2)
On the left-hand side, the diminishing value of camp labor for firms (M) is
determined by four forces: workers’ productive contribution (V), the fee to be
paid for them (R), the length of time already imprisoned (t), and the health decay
334 Case Studies I
suffered during their time of imprisonment (h). When t and/or h increase, V

decreases and, with it, so does the value of forced labor to the firm; likewise, when
R increases. Elderly, sick, or injured prisoners exhibit a larger h. Ceteris paribus,
the larger is the value of h, the larger is the value of M, that is, the diminished value
of the victim to the firm. Therefore, a private firm minimizes the following loss
function:
2
  V  (3)
L = (V − M(V , R , t , h))2 = V −  1 − R  .
  t(h ) 
From the firm’s perspective, the variables to be solved for are the optimal fee
to be paid (R*) and the optimal duration of the imprisonment (t*). In the first
stage of the interaction between the SS and the firm, the SS demands a labor fee
by maximizing its payoff function, Z, with respect to R. At the time of action, r
and δ are exogenously given. The first-order condition for maximization of Z is
(2R2δ–1/δ)−(2R2δ–1rδ/δ) = 0. The second-order condition, Rδ(2δ-1) + rδ(1-δ) < 0,
holds if and only if δ < 1/2 . Solving, it turns out that the optimal current fee
demanded, denoted by R*, equals the past fee, R* = r. In the second stage, the pri-
vate firm observes (and accepts) the demanded fee. It now chooses the duration
of imprisonment by minimizing the loss function, L, with respect to t. The first-
order condition is given by [2Vt–(h+1)/h × (V+r)]/h−[2V2t–(h+2)/h]/h = 0. The second-
order condition for a minimum, ∂2L/∂t2 > 0, holds if and only if r < {[Vt−(1/h)(h+2)]/
(h+1)}−V. Solving, the optimal expected duration (and the consequent survival of
inmates) is t* = [(V+r)/V]−h .
Optimal duration is chosen by the employing firm: t* is decreasing in h and r,
and increasing in V. In words, the weaker are the prisoners (a high h), or the higher
the past fee (r), the shorter is their time of survival. In contrast, the higher is their
expected contribution (V), the longer they will survive. Distressingly, prisoners
needed to “earn their keep” (V), for the alternative is to be nearer to death. More
distressingly still, ill-considered external interference with the “value” of camp
labor (V) can hasten their death!
When the game ends, the loss inflicted on the private firm is L* = 0 (the loss is
minimized to zero). The optimal strategy of the labor-using firm is to accept the
fee the SS demands. This minimizes the loss of value of labor. At the same time, the
objective function of the SS is maximized when R* = r, which translates into a zero-
profit, Z* = 0 (the SS does not seek profit, but extermination). The optimal strategy
of the SS is setting the current demanded fee equal to the past fee; in fact, the SS
would charge a fee in any case. In practical terms, this means that fees for enslaved
Jews would have been kept invariant, a testable proposition. The value of h is cru-
cial. In light of the dehumanization the prisoners experienced, one would expect
large values of h, reducing their exploitable labor reserves at work and consequently
shorter survival times. The SS had no incentive to raise the current labor fee because
this would have been associated with a longer survival of inmates. Keeping the fee
low would increase their exploitation and secure their speedy extermination. As
mentioned in the main text, the SS did not seek to behave as a profit-maximizing
supplier of workers, such as a labor union or staffing agency might.
Notes
1. The operations of the Einsatzgruppen can be organizationally compared with the genocidal
operations of Ottoman Turks against Armenians between 1915 and 1923.
2. The first part of Goldhagen (1996) discusses the historical roots of German anti-Semitism
in detail. Hillman (2013) also discusses the foundations of anti-Semitism highlighting a
definition of anti-Semitic behavior made of (1) “big lies”; (2) demonization; and (3) denial
to Jews of the right of self-defense.
3. Jews were blamed for poisoning the victims of the Black Death (the bubonic plague).
4. On Nazi propaganda see, e.g., Welch (2004), Herf (2005), and Grabowski (2009). Related
to propaganda was also scientists’ legitimization of racial supremacy. On this see, e.g.,
Ehrenreich (2007) and Cornwell (2004).
5. The question is explicitly addressed in Hamman, Loewenstein, and Weber (2010). By
means of three experiments, they study a principal-agent relationship in which the prin-
cipal hires an agent to take immoral actions that the principal would be reluctant to take
directly. This appears to be a scenario favorable to the moral disengagement of both. In
fact, the principal may feel less responsible for wrongdoing while the agent may feel only
that he was just obeying orders.
6. On Nazi bureaucracy, also see Yehouda (2013) and Clegg (2009).
7. It is worth noting that, according to an interpretation by Aly and Heim (2003), these mea-
sures to benefit Germans could be considered within a broader development policy that
favored the organization of the Holocaust as a rational solution to implement superior
productivity in the modernized manufacturing sector, “solving” the problem of overpop-
ulation with its Malthusian consequences. For criticism of this approach, see Browning
(1992). The review is based on the original version of the essay by Aly and Heim, released
in 1991.
8. Hilberg (1985) reports the following revenues (in Reichsmarks): 1 million in 1932–1933,
45 million in 1935–1936, 70 million in 1936–1937, 81 million in 1937–1938, and 342 mil-
lion in 1938–1939.
9. Aryanization took place not only in Germany but also elsewhere in Europe. In Austria, the
newly created Property Transfer Bureau (Vermögensverkehrsstelle) handled 26,000 Jewish
enterprise transfers between 1938 and 1939 (Feldman 2007). It “aryanized” 5,000 of
them, and shut down the remaining 21,000. Between 1940 and 1942, the Slovak govern-
ment closed 9,987 and aryanized 1,910 out of about 12,000 Jewish firms (Aly 2007). Zakic
(2014) describes Aryanization in Serbia.
10. This way of modeling is perhaps best applied to large rather than small firms. Hilberg
(1985) reports that several Jewish firms were willing to stand up against Nazi expropria-
tion and this happened, in particular, with larger Jewish corporations.
11. Interestingly, it was Paul Samuelson who first applied the “butter” and “guns” labels to pro-
ductive and unproductive activities, and in coining the terms he had in fact Nazi Germany
in mind. In the 1970 edition of Economics, he wrote, “So let us assume that only two eco-
nomic goods (or classes of economic goods) are to be produced. For dramatic purposes, we
can choose the pair Adolf Hitler ranted about—g uns and butter” (1970, 18).
336 Case Studies I
12. There are no reliable figures on the total number of camps. Goldhagen (1996) reports
10,000. The US Holocaust Memorial Museum speaks of “about 20,000” (http://w ww.
ushmm.org/wlc/en/a rticle.php?ModuleId=10005144). The system was designed as a net-
work of main camps linked to subcamps or satellite camps. In 1944, between 60 percent
and 80 percent of the prisoners were in subcamps. The number of subcamps increased dra-
matically in the war years. Buchenwald, for example, had 88 satellite camps by the end of
war, in 1945. In general, up until late 1942, there were only around 80 subcamps in all.
One year later, the SS had set up 186 subcamps throughout the entire area controlled by
the Germans. By June 1944, there were 341 camps and only a few months later, in January
1945, the number had grown to at least 662 subcamps (Buggeln 2009).
13. On the relationship between firms and slave labor also see Roth (1980).
14. The trade-off between extermination and utilization of forced labor was not novel in
German history. Between 1904 and 1907 imperial Germany waged a genocidal war
against the native South-West African populations of Herero and Nama. At the time, some
German officials, obsessed with racist superiority, wanted to completely exterminate the
native inhabitants of the land. In contrast, public and private German employers relied on
forced labor to solve the problem of labor shortages. Consequently, many settlers contested
commander Lothar von Trotta’s extermination order and convinced the government in
Berlin to favor the establishment of forced labor concentration camps (see c hapter 13 in
this volume).
15. Jews were robbed of any valuable belongings, jewelry, and precious metals in particular.
With specific regard to precious metals and commodities, see Macqueen (2004) and
Banken (2006).
16. The historical origin of the code name is still disputed.
References
Acemoglu, D., and A. Wolitzky. 2011. “The Economics of Labor Coercion.” Econometrica 79, no.
2: 555–6 00.
Akerlof, G., and W. T. Dickens. 1982. “The Economic Consequences of Cognitive Dissonance.”
Akerlof, G., and R. E. Kranton. 2000. “Economics and Identity.” Quarterly Journal of Economics
115, no. 3: 715–53.
Akerlof, G., and R. E. Kranton. 2005. “Identity and the Economics of Organizations.” Journal of
Economic Perspectives 19, no. 1: 9–32.
Aly, G. 2007. Hitler’s Beneficiaries: Plunder, Racial War and Nazi Welfare State. New York: Holt.
Aly, G., and S. Heim. 2003. Architects of Annihilation: Auschwitz and the Logic of Destruction.
Princeton, NJ: Princeton University Press.
Anderton, C. 2014. “A Research Agenda for the Economic Study of Genocide: Signposts from
Bandura, A. 1999. “Moral Disengagement in the Perpetration of Inhumanities.” Personality and
Social Psychology Review 3, no. 3: 193–2 09.
Bandura, A. 2010. “Selective Moral Disengagement in the Exercise of Moral Agency.” Journal of
Moral Education 31, no. 2: 101–19.
Banken, R. 2006. “National Socialist Plundering of Precious Metals, 1933–1945: The Role of
Degussa.” Working paper. Institute of European Studies. University of California, Berkeley.
Baumol, W. J. 1990. “Entrepreneurship: Productive, Unproductive, and Destructive.” Journal of
Bel, G. 2010. “Against the Mainstream: Nazi Privatization in 1930s Germany.” Economic History
Review 63, no. 1: 34–55.
Bernholz, P. 2001. “Ideocracy and Totalitarianism: A Formal Analysis Incorporating Ideology.”

Public Choice 108, nos. 1–2: 33–75.
Black, P. 2011. “Foot Soldiers of the Final Solution: The Trawniki Training Camp and Operation
Reinhard.” Holocaust and Genocide Studies 25, no. 1: 1–9 9.
Breton, A., and R. Wintrobe. 1986. “The Bureaucracy of Murder Revisited.” Journal of Political
Economy 94, no. 5: 905–2 6.
Browning, C. R. 1992. The Path to Genocide: Essays on Launching the Final Solution. New York:
Cambridge University Press.
Brustein, W. 1996. The Logic of Evil: The Social Origins of the Nazi Party, 1925–1933. New Haven,
CT: Yale University Press.
Buccheim, C., and J. Scherner. 2006. “The Role of Private Property in the Nazi Economy: The
Case of Industry.” Journal of Economic History 66, no. 2: 390–416.
Buggeln, M. 2009. “Building to Death: Prisoner Forced Labor in the German War
Economy—t he Neuengamme Subcamps, 1942–1945.” European History Quarterly 39,
no. 4: 606–32.
Caruso, R. 2012. “Differentials in Property Rights in a Two-Sector Economy.” Revue d’Economie
Politique 122, no. 2: 257–78.
Clegg, S. 2009. “Bureaucracy, the Holocaust and Techniques of Power at Work.” Management
Revue 20, no. 4: 326–47.
Cornwell, J. 2004. Hitler’s Scientists: Science, War and the Devil’s Pact. New York: Penguin.
Dell, F. 2005. “Top Incomes in Germany and Switzerland over the Twentieth Century.” Journal
of the European Economic Association, Proceedings of the Nineteenth Annual Congress of the EEA
3, no. 2/3: 412–21.
Dixit, A. 2004. Lawlessness and Economics: Alternative Modes of Governance. Princeton,
Ehrenreich, E. 2007. “Otmar von Verschuer and the ‘Scientific’ Legitimization of Nazi Anti-
Jewish Policy.” Holocaust and Genocide Studies 21, no. 1: 55–72.
Feldman, G. D. 2007 “The Economics of the ‘Final Solution.’” Australian Journal of Politics and
History 53, no. 1: 57–67.
Ferencz, B. B. 1979. Less Than Slaves: Jewish Forced Labor and the Quest for Compensation.
Bloomington: Indiana University Press.
Festinger, L. 1957. A Theory of Cognitive Dissonance. Palo Alto, CA: Stanford University Press.
Garfinkel, M. R., and S. Skaperdas. 2007. “Economics of Conflict: An Overview.” In T. Sandler
and K. Hartley, eds., Handbook of Defense Economics. Vol. 2. Amsterdam: Elsevier, 649–709.
1: 45–86.
Gellately, R. 2001. Backing Hitler: Consent and Coercion in Nazi Germany. Oxford: Oxford
University Press.
Glass, J. M. 1997. “Against the Indifference Hypothesis: The Holocaust and the Enthusiasts for
Murder.” Political Psychology 18, no. 1: 129–45.
Goldhagen, D. J. 1996. Hitler’s Willing Executioners: Ordinary Germans and the Holocaust.
New York: Alfred A. Knopf.
Grabowski, J. 2009. “German Anti-Jewish Propaganda in the Generalgouvernment, 1939–1945:
Inciting Hate through Posters, Films, and Exhibitions.” Holocaust and Genocide Studies 23,
no. 3: 381–412.
Hamman, J. R., G. Loewenstein, and R. A. Weber. 2010. “Self-I nterest through Delegation: An
Additional Rationale for the Principal-A gent Relationship.” American Economic Review 100,
no. 4: 1826–4 6.
Herbert, U. 1993. “Labor and Extermination: Economic Interest and the Primacy of
Weltanschauung in National Socialism.” Past and Present 138, no. 1: 144–95.
Herbert, U. 2000. “Forced Laborers in the Third Reich: An Overview.” International Labor and
Working-Class History 58:192–218.
338 Case Studies I
Herf, J. 2005. “The ‘Jewish War’: Goebbels and the Antisemitic Campaigns of the Nazi
Propaganda Ministry.” Holocaust and Genocide Studies 19, no. 1: 51–80.
Hilberg, R. 1985. The Destruction of the European Jews. New Haven, CT: Yale University Press.
Hillman, A. L. 2010. “Expressive Behavior in Economics and Politics.” European Journal of
Hillman, A. L. 2013. “Economic and Behavioral Foundations of Prejudice.” In C. A. Small, ed.,
Global Antisemitism: A Crisis of Modernity. Leiden: Martinus Nijhoff, 51–67.
Hirshleifer, J. 1988. “The Analytics of Continuing Conflict.” Synthese 76, no. 2: 201–33.
Lemkin R. 1944. Axis Rule in Occupied Europe: Laws of Occupation, Analysis of Government, Proposals
for Redress. Washington, DC: Carnegie Endowment for International Peace.
Longerich, H. P. n.d. Hitler’s Role in the Persecution of the Jews by the Nazi Regime: Electronic Version.
http://web.archive.org/web/2 0090709111759/http://w ww.hdot.org/en/t rial/defense/
pl1/17 [accessed May 17, 2015].
MacQueen, M. 2004. “The Conversion of Looted Jewish Assets to Run the German War
Machine.” Holocaust and Genocide Studies 18, no. 1: 27–45.
Milgram, S. 1974. Obedience to Authority: An Experimental View. New York: Harper & Row.
Mixon, F. G., and E. W. King. 2009. “Coercion, Vertical Trust, and Entrepreneurism in
Bureaucracies: Evidence from the Nazi Holocaust.” Economics Bulletin 29, no. 2: 673–79.
Mixon, F. G., W. C. Sawyer, and L. J. Trevino. 2004a. “The Bureaucracy of Murder: Empirical
Evidence.” International Journal of Social Economics 31, no. 9: 855–67.
Mixon, F. G., W. C. Sawyer, and L. J. Trevino. 2004b. “Vertical and Horizontal Trust Networks
in Bureaucracies: Evidence from the Third Reich.” Constitutional Political Economy 15, no.
4: 371–81.
Newman, P. C. 1948. “Key German Cartels under the Nazi Regime.” Quarterly Journal of
Economics 62, no. 4: 576–95.
Overy, R. J. 1988. “Mobilization for Total War in Germany.” English Historical Review 103, no.
408: 613–39.
Roth, J. K. 1980. “Holocaust Business: Some Reflections on Arbeit Macht Frei.” Annals of the
American Academy of Political and Social Science 450, no. 1: 68–82.
Samuelson, P. A. 1970. Economics. New York: McGraw-H ill.
Scherner, J. 2013. “Armament in Depth or Armament in Breadth? German Investment Patterns
and Rearmament During the Nazi Period.” Economic History Review 66, no. 2: 497–517.
Scheweitzer, A. 1946. “Profits under Nazi Planning.” Quarterly Journal of Economics 61, no.
1: 1–25.
Skaperdas, S. 1992. “Cooperation, Conflict, and Power in the Absence of Property Rights.”
Spoerer, M., and J. Fleischhacker. 2002. “Forced Laborers in Nazi Germany: Categories,
Numbers and Survivors.” Journal of Interdisciplinary History 33, no. 2: 169–2 04.
Stargardt, A. W. 1944. “The Nazi-Fascist Economy.” Australian Quarterly 16, no. 4: 5–15.
Voigtländer, N., and H. J. Voth. 2012. “Persecution Perpetuated: The Medieval Origins of Anti-
Semitic Violence in Nazi Germany.” Quarterly Journal of Economics 127, no. 3: 1339–92.
Voth, H. J., and T. Ferguson. 2008. “Betting on Hitler: The Value of Political Connections in
Nazi Germany.” Quarterly Journal of Economics 123, no. 1: 101–37.
Welch, D. 2004. “Nazi Propaganda and the Volksgemeinschaft: Constructing a People’s
Community.” Journal of Contemporary History 39, no. 2: 213–38.
Wiesen, S. J. 2011. Creating the Nazi Marketplace: Commerce and Consumption in the Third Reich.
Cambridge, UK: Cambridge University Press.
Yehouda, S. 2013. “Beyond ‘Instrumental Rationality’: Lord Cromer and the Imperial Roots of
Eichmann’s Bureaucracy.” Journal of Genocide Research 15, no. 4: 379–9 9.
Zakic, M. 2014. “The Price of Belonging to the Volk: Volksdeutsche, Land Redistribution
and Aryanization in the Serbian Banat, 1941–4” Journal of Contemporary History 49, no.
2: 320–4 0.
15
The Economics of Genocide in Rwanda

W i l l a F r i e dm a n
15.1. Introduction
Since the one hundred days in 1994 during which around one million people were
killed in Rwanda, numerous explanations have been presented to make sense of
what happened. They cover a broad range of specific hypotheses about what cre-
ated the motivation for leaders and participants to kill and what made the speed
and intensity of the violence possible. In this chapter, I focus on four widely dis-
cussed ideas about why Rwanda was “ready” for genocide in early 1994. The first
is that there was a Malthusian crisis. The population density of Rwanda was more
than three hundred people per square kilometer, and a high birth rate meant that
pressure on the land was growing steadily. Many have argued that this increas-
ing population pressure brought onto a comparatively tiny area of arable land
may have made the genocide inevitable, or at least likely. A second idea is related
to poverty and poor economic prospects. Reduced economic opportunity from
collapsing international prices of key exports and high unemployment yielded
a mass of potential recruits with low opportunity costs of joining the violence.
The third idea is inequality. There were two ethnic groups with very different eco-
nomic opportunities. Horizontal inequality—inequality between groups—has
been shown to contribute to an increased likelihood of violence in other contexts.
The rhetoric of the disadvantaged rising up to overthrow those with historically
accumulated economic access was strong. Perhaps interethnic inequality brought
about the genocide. The fourth idea is that the phenomenal organization of soci-
ety made it possible for an incredibly efficient, decentralized, and deadly program
of killing to commence. While this explanation cannot hold without considering
factors spurring the desire for destruction, its adherents point to the many ways in
which certain strengths of government and society generated devastation.
A brief note on what is not discussed is warranted. Even as the chapter pro-
ceeds to lay out four broad economic ideas to explain the violence, this should not
be taken to exhaust the universe of explanatory ideas and schemes that have been
339
340 Case Studies I
offered—indeed, not even all economic ones. For example, the role of Rwanda’s
colonial experience, the perceptions of an outside security threat, and the shoot-
ing down of the president’s airplane, which served as the immediate trigger that
unleashed the violence on April 6, 1994, have all provided fodder for numerous
books and scholarly articles. But in the interest of focus and brevity, this chapter
addresses only four of these main ideas.
15.2. Malthus
Given the popularity of Jared Diamond’s book Collapse (2005), his chapter on
Rwanda may be the most widely read explanation of what happened there. With
a pregenocide population of nearly eight million people, and with more than
90 percent of the economically active population engaged in agriculture, Rwanda
faced considerable land scarcity prior to the 1994 genocide. Diamond explains
how Rwanda’s conditions fit well with Malthus’s hypothesis of violence induced
by a rising population under a stagnant agricultural sector and argues that the
1994 genocide was indeed evidence in support of Malthus’s thesis: intensifying
land pressure can, in time, lead to an outbreak of unspeakably vicious violence.
Country-specific scholarship confirms that Rwanda did experience food
production-related problems. For example, based on fieldwork conducted in
1992, Olson (1994) concluded that the country no longer was self-sufficient in
food production. Yet even as food imports increased, farm households experi-
enced decreased food availability, in part because increased farming intensity,
combined with reduced fallowing, led to deteriorating soil quality. And even
the food that was produced yielded increasingly lower nutritional value (also see
Percival and Homer-Dixon 1998). At the same time, agricultural families had lit-
tle access to alternative sources of income, as off-farm employment opportunities
were very limited.
Andre and Platteau (1998) show that land disputes and changing social rela-
tionships, because of underlying changes in land arrangements, were leading to
insecurity in the years before the genocide. Originally collected to investigate
changes in land tenure rules, their work is based on a highly detailed pregeno-
cide dataset for a small area in the north of the country. After the genocide, the
authors then visited refugee camps to establish what had happened to each of
their respondents in their original fieldwork. They found that members of the
richer, more well-off farm households were more likely to have been killed (but
also that “troublemakers” were killed, which Andre and Platteau argue is evidence
that the genocide was taken as a chance to settle old scores). In their pregenocide
dataset, they noted an increase, from 1988 to 1993, in the number of households
that owned less than one-fourth of a hectare of land, an insufficient amount to
produce enough food to feed a family. Specifically, while 36 percent of survey
T h e E c o n o m i c s o f G e n o c i d e i n R wa n d a 341
respondents owned this little land or less in 1988, 45 percent did in 1993. They
also point out population density had equalized across the entirety of the coun-
try, evidence that there was no longer room for the population to expand. Des
Forges (1999, 11) writes: “Of the nearly 60 percent of Rwandans under the age of
twenty, tens of thousands had little hope of obtaining the land needed to establish
their own households or the jobs necessary to provide for a family.” In a sample
of 352 individuals interviewed before the genocide and then tracked thereafter,
Verwimp (2005) finds that households with perpetrators did have less land per
adult than households without. Perpetrators also were more likely to have worked
outside their own farm, possibly a sign that they felt the land scarcity more than
others. But Verwimp did not find that any given household’s low land productiv-
ity was associated with whether or not one of its members became a perpetrator.
Based on detailed data, including levels of violence and population density
measures taken from 1,294 sectors, Verpoorten (2012) finds much higher death
rates in areas with higher population density and lower access to new land for
young men. (A “sector” is a small administrative unit, with about 5,000 people, on
average.) Still, she rightly points out that population density is not necessarily an
exogenous factor, that is, the population density observed at the time of the geno-
cide was itself the outcome of pregenocide population mobility. Mobility, or lack
of mobility, itself needs to be explained before accepting the Malthusian thesis.
15.2.1. Regulated Limited Mobility

Rwanda’s pregenocide legal system restricted land transfers and residential and
occupational mobility across the country. As well, rules designed to support the
agricultural sector limited off-farm employment options. Moreover, government
policy in regard to food self-sufficiency, evidenced by a relatively low degree
of exports, may have further limited occupational opportunities (Andre and
Platteau 1998; Boudreaux 2009).
Land sales were severely limited by law, but there is evidence that they hap-
pened anyway. Officially, the land market was heavily regulated, to the point
of outlawing most buying and selling (Boudreaux 2009). And yet, Andre and
Platteau (1998) find that land transfers occurred nonetheless, often in very formal
ways, with witnesses to support the validity of a transfer even if these transactions
were not technically legal. Still, they acknowledge that a more open system of
legal land transfer and titling may have reduced the severity of the conflict. Yet
they also point out that many of the conflicts (e.g., between fathers and sons) or
forms of marginalization and exclusion from the land market (e.g., of women)
would not have been alleviated by a more accommodative legal system alone.
In his detailed account of the history of Rwanda and the lead-up to the geno-
cide, Prunier (1995) points to strict laws as well in regard to the population’s free-
dom to move. Identity cards stated one’s legal residence, and permission to move
342 Case Studies I
was often denied if one did not have a prearranged school or job placement. The
degree of enforcement, however, is open to debate. For example, Olson (1994)
argues that although the rules were strict, there is strong evidence that many peo-
ple not only moved but moved to places with relatively better farming opportuni-
ties. For 1978 and 1991, she looks at population growth in each commune, relative
to what would be expected from reported birth and death rates alone, and finds
people moving to areas with poor but potentially underexploited farm land.
In this regard, Rwandan emigration provides an important additional caveat to
claims about limited population mobility (Yanagizawa-Drott 2006). Still, while
some of these individuals moved to neighboring countries to pursue educational
opportunities, many were refugees from frequent outbreaks of violence within
Rwanda. All in all, though, the key idea here is that even if moving between loca-
tions for improved economic opportunity—a useful tool in limiting the negative
consequences of geographically specific reductions in employment prospects—
did take place to some degree, it still may have been limited enough to exacerbate
the effects of land scarcity on well-being. Thus, even if diluted, the Malthusian
hypothesis may still have some validity to it.
15.2.2. Skepticism about Malthusian Claims

The Malthusian explanation comes with significant problems and shortcomings.
One might expect, for instance, that areas of high population density and popula-
tion pressure would have seen greater levels of violence. But for the most part, this
was not the case (Verwimp 2005). As mentioned, population pressure is not easy
to measure. It is not an exogenous factor; it is likely that high population areas
are more population dense because they are better suited for crop production
and thus can sustain a greater population to begin with. A dry area or one with
poor soil quality can produce insufficient food for its population even with very
low population density. (There is also an economies of scale argument here: of
course, mass murder takes place where people amass, in high-population-density
areas.) All this makes the measurement of the relationship between violence and
population pressure difficult. Another counterpoint to the Malthusian popula-
tion density argument is that while Rwanda’s population density was high for an
agricultural society, a number of other countries with even greater density and
poverty did not experience genocidal violence (Uvin 1998).
The Malthusian story also suggests that violence would have been higher in
areas that were economically worse off, yet this is not consistently found to have
been the case. Details are provided in section 15.3, but as an example, using lit-
eracy, wealth, and density, McDoom (2014) finds no evidence that community
deprivation is associated with earlier onsets of violence. Likewise, Verwimp
(2005) finds that households with perpetrators did not have less land than house-
holds without perpetrators (although they did have less land per adult).
In a different line of reasoning, Percival and Homer-Dixon (1998) argue

that there were too few genocide participants to support a Malthusian-type
hypothesis.1 Instead, they place greater reliance on the role of genocide leader-
ship, arguing that genocide architects were less likely to have felt the pinch of any
economic deprivation. They also note that the violence began in areas with rela-
tively low population density before it then spread to the south where the popula-
tion density was the highest.
In sum, while Rwanda’s population density was high, while families that once
had been reasonably secure in their ability to feed themselves were no longer able
to do that, while young people saw their own prospect of obtaining enough land to
feed a family or access alternative sources of income diminished, and while there
existed clear pressures on land ownership and limits on population movement,
the details of the empirical evidence of where the violence started, how it spread,
and where it was the most pronounced does not well match the Malthusian-based
storyline. This is not to deny that insufficient land in an agricultural society with a
growing population was an important factor in building frustration that may have
contributed to so many people participating in the violence, but it does not appear
to have been the only reason.
15.3. Poverty, Vertical Inequality, and Poor Prospects

Lack of economic well-being can generate dissatisfaction and frustration with the
existing system, perhaps enough to stimulate participation in violence to obtain
material gain of food or income more generally. Economic conditions in the years
leading up to the genocide certainly had created a situation in which increasingly
large numbers of individuals had few economic opportunities and little hope of
attaining economic success. Economists now routinely distinguish between ver-
tical and horizontal inequality, the former referring to income or asset classes
irrespective of any ethnic or other distinct composition of the members in each
class. In contrast, horizontal inequality refers to distinct groups, such as Hutu and
Tutsi, being clustered into corresponding income and asset classes so that conflict
may arise not merely between “rich” and “poor” but of “rich Tutsi” as against “poor
Hutu,” giving conflict a special hue. Section 15.3 deals with vertical inequality;
section 15.4 addresses horizontal inequality, a concept first promoted by Frances
Stewart (Stewart 2002).
15.3.1. Weakening Prospects for Advancement

A common theme in studies of violence is that low opportunity costs facilitate par-
ticipation in destructive activities. Without alternative means of earning a living,
potential recruits may be far easier to recruit than otherwise. While in the 1970s,
344 Case Studies I
Rwanda’s “economy did better than others in the same region, with a net increase
in gross national product in relation to population … [its] prosperity was both
fragile and superficial. The mass of the people stayed poor and faced the prospect
of getting only poorer” (Des Forges 1999, 46). Indeed, two important interna-
tional events in the 1980s and early 1990s deeply hurt employment prospects and
reduced Rwandan’s economic well-being dramatically: world market prices for
coffee and tin—t he country’s key exports—fell dramatically. The International
Coffee Organization suspended country quotas in 1989, and Vietnam became a
significant producer and exporter of coffee in the early 1990s. Consequently, the
world supply of coffee beans increased and export prices plummeted. In 1985,
coffee exports earned Rwanda USD 144 million; by 1993, this had declined to
just USD 30 million. World coffee prices had fallen before, in the early 1980s, but
tin export earnings had mitigated the adverse economic impact. Yet the world tin
market was changing as well: competition from new producer countries, com-
bined with a decline in international tin demand, gradually eroded world tin
prices from 1980 onward. For Rwanda, tin production eventually became unprof-
itable. Thus, in the years leading up to 1994, the country’s two primary exports
became far less valuable as a source of private and public employment and reve-
nue. (For the information in this paragraph, see Prunier 1995; Andre and Platteau
1998; Uvin 1998; Verwimp 2005; and Boudreaux 2009.)
15.3.2. Poverty and Recruitment

Des Forges (1999, 382) writes that recruitment was likely easier among those
with the fewest economic opportunities: “Many of these zealous killers were
poor, drawn from a population 86 percent of whom lived in poverty, the highest
percentage in the world. They included many young men who had hung out on the
streets of Kigali or smaller commercial centers, with little prospect of obtaining
either the land or the jobs needed to marry and raise families.” Prunier (1995) also
claims that the Interahamwe and Impuzambugambi militias recruited largely from
the least well-off. For those with few alternatives, joining the militias offered the
prospect of material gain:
Authorities offered tangible incentives to participants. They delivered

food, drink, and other intoxicants, parts of military uniforms and small
payments in cash to hungry, jobless young men. They encouraged cul-
tivators to pillage farm animals, crops, and such building materials as
doors, windows and roofs. Even more important in this land-hungry
society, they promised cultivators the fields left vacant by Tutsi victims.
To entrepreneurs and members of the local elite, they granted houses,
vehicles, control of a small business, or such rare goods as television sets
or computers. (Des Forges 1999, 10)
Similarly, Prunier (1995, 248) writes:
There was of course also an element of material interest in the killings,

even in the countryside. The killers looted household belongings and
slaughtered the cattle. Meat became very cheap, and grand feasts were
held, as if in celebration of the massacre. … Villagers also probably had
a vague hope that if things settled down after the massacres they could
obtain pieces of land belonging to the victims, a strong lure in such a
land-starved country as Rwanda. But greed was not the main motiva-
tion. It was belief and obedience.
Empirically, however, the evidence is mixed on whether higher levels of vio-

lence were associated with worse economic conditions among Hutu. For example,
Friedman (2013) finds higher levels of participation in areas with higher levels
of Hutu unemployment. But Verwimp (2005) finds that while households with
genocide perpetrators had less land per adult than households without perpetra-
tors, he also finds that, on the whole, the perpetrator households were wealthier
because of a greater likelihood of off-farm income. McDoom (2014) does not find
a relationship between the overall level of deprivation in a commune and the tim-
ing of the onset of violence. However, his analysis does not split the measures
of economic deprivation between Hutu and Tutsi, so if inequality between the
groups was an important factor in this regard, it was not considered. The next
section therefore turns to a consideration of horizontal inequality.
15.4. Horizontal Inequality

Horizontal, and specifically interethnic, inequality was a central theme of the
Rwandan genocide organizers’ rhetoric. They frequently pointed to the ways in
which the Tutsi population had maintained economic advantages, even after the
government became Hutu-dominated immediately following the country’s inde-
pendence. In popular accounts, one group—t he Tutsi—was historically advan-
taged, having been granted a privileged economic position by Belgian colonizers,
while the other—t he Hutu—a lways received less than it was due. This section
discusses levels of intergroup inequality, the rhetoric about inequality—used
to galvanize the population during the genocide—and the degree to which the
evidence supports inequality as a driver of the onset of and participation in the
genocide.
It is difficult to pin down the exact meaning, let alone the numbers, of differ-
ent types of interethnic inequality. For example, employment in a modern econ-
omy differs from the meaning of employment in an overwhelmingly agrarian
economy. Thus, in the 1991 census, even as more than 95 percent of the relevant
346 Case Studies I
population report some form and level of employment, only 7.4 percent of (self-
identified) Hutu, and 11.3 percent of (self-identified) Tutsi, report that they
worked for someone else (Friedman 2013). The 3.9 percentage point difference
is appreciable, yet pales relative to the 95 percent number. There were differences
in human capital attainment: 51.7 percent of Hutu were literate as compared to
67.3 percent among Tutsi. Hutu were slightly more likely than Tutsi to report that
they owned any land (92.8 percent and 90.5 percent, respectively), but informa-
tion about the size of the land holdings is unavailable. Asset ownership was higher
among Tutsi (Friedman 2013), but the magnitude and meaning of any difference
to Hutu is difficult to express as very little is known about assets that contribute
to income, notably animals and land.
15.4.1. The Rhetoric of Inequality

The organizers of the genocide—in encouraging the population to join in the
slaughter—rarely failed to claim that Tutsi had taken more than their fair share.
Des Forges (1999, 88) writes:
The propagandists said the Tutsi had infiltrated the economy—at one
point Kangura [a magazine in Rwanda] claimed that 70 percent of the
rich in Rwanda were Tutsi—monopolized credit at the banks, and
won a disproportionate share of the highly coveted import and export
licenses. In a clear effort to divert the resentment otherwise directed
towards Hutu from Habyarimana’s region, propagandists argued that it
was Tutsi, not other Hutu, who occupied the jobs which southern Hutu
wanted and failed to get. They also accused the Tutsi of having taken a
disproportionate share of places in secondary school and university and,
because of their educational advantages, of having dominated the profes-
sions and government.
Prunier (1995) concurs, writing about how having being treated as inferior, the
Hutu developed hostility to even the poorest Tutsi. Interethnic inequality did
exist, to some degree. Prunier (1995, 232) writes:
In fact there was no contradiction between the ethnic and the social
aspects of the killings since, in Kigali at least, the Tutsi tended to be bet-
ter off than the Hutu. Political power had been in Hutu hands for thirty-
five years but, thanks to the Belgian social and educational favouritism
towards the Tutsi for the forty years before that, the Tutsi community
was still able to do well for itself socially and economically. This did not
only mean the big Tutsi businessmen; it also meant that most of the local
personnel in foreign embassies and in NGOs and international agencies
were Tutsi, that there were many Tutsi in the professions and even that
the best and highest-priced bar girls, the ones to be encountered in the
big hotels, were Tutsi. Social envy came together with political hatred to
fire the Interahamwe bloodlust.
The story of this inequality had roots from long before independence.
The definitions used during colonization to assign ethnicity meant that the
distinctions became inextricably linked with inequality. Thus, stories of classi-
fication into ethnic groups are reported whereby those with more than ten cows
were assigned to be Tutsi while those with fewer became Hutu (Prunier 1995;
Mamdani 2002). What may have been degrees of vertical inequality, based on
arbitrary differences in livestock ownership, in a Belgian colony became hori-
zontal “interethnic” inequality in Rwanda. Vansina (2012) discusses this in more
detail, also pointing out the various ways by which individuals could change eth-
nic assignment simply by changing the type of work they did or the number of
cows they owned.
To what extent the two groups divided has been heatedly debated since the
genocide. A few points of agreement can be mentioned. They share a language
(Kinyarwanda) and a religion (Christianity), and they live in the same areas. For
example, the proportion of Tutsi in a commune is never more than 40 percent.
Intermarriage rates may be open to interpretation but in 1991, 28.6 percent of
Tutsi were married to Hutu and 2.5 percent of Hutu were married to Tutsi. These
numbers had not increased from the previous generation when 26.8 percent
of Tutsi were married to Hutu and 2.4 percent of Hutu were married to Tutsi.
Despite these high levels of integration, there is evidence that the segregation that
did exist mattered in the genocide: McDoom (2014) finds that segregation sped
up the onset of violence during the genocide.
As mentioned, inequality was a central theme in genocide propaganda, and
the importance of propaganda in the violence cannot be overstated: “As authori-
ties played on popular fears and greed, some people picked up their machetes
and came readily” (Des Forges 1999, 10; also see chapter 12 in this volume). In a
groundbreaking set of court cases, the leaders of the primary radio station during
the genocide (RTLM) were accused of incitement to genocide and crimes against
humanity and given life sentences. But the question remains, to what degree did
actual inequality fuel violence?
15.4.2. Richest Were Attacked

The flip-side of the argument that the poor attack the rich is that relative wealth
may attract violence but may also afford opportunities to avoid, mitigate, or alto-
gether escape harm. Be that as it may, resentment, jealously, and opportunity to
expropriate lootable assets are frequently mentioned in accounts of the genocide
348 Case Studies I
(e.g., Prunier 1995; Gourevitch 1998; Des Forges 1999; Mamdani 2002; Cramer
2003; Straus 2006).
While I have seen no claims that any of these amounted to a direct motiva-
tion for participation in the genocide, many sources point to the indirect effect of
the possibility of looting in stoking incentives to participate.2 Des Forges (1999,
1166) mentions looting: “Among those who did carry out genocide, actors partici-
pated in many ways: from the national leaders who aimed to extirpate the Tutsi
down to the level of ordinary people who showed no taste for violence but wanted
only to enrich themselves through pillage.” Prunier (1995, 250) writes: “Although
it seems that few people were killed purely for robbery, there was nevertheless
a strong element of social envy in the killings, and in the rural areas this could
work at a very simple level. In the vivid words of a survivor, ‘the people whose
children had to walk barefoot to school killed the people who could buy shoes for
theirs.’ ” In some cases, the expropriation of goods was more complicated than
simple theft. Prunier (1995, 257) writes, “Some people were denounced by their
colleagues who wanted their jobs or killed by people who wanted their property,
while others were saved by unknown Hutu disgusted by the violence.”
Many accounts report that the well-off were consistently attacked, even—in
some cases—when these were Hutu. Thus, in the sample studied by Andre and
Platteau (1998), which includes only one Tutsi woman and is otherwise entirely
Hutu, the richest were the most likely to be killed. In a sample from southern
Rwanda, Verwimp (2005) found that older and wealthier people were more likely
to have been killed. Using data from the entire country, de Walque and Verwimp
(2010) found urban and highly educated people more likely to have died.
Jealously and expropriation are not the only explanations: “And as in many
such situations, intellectuals were also a target: journalists, professionals and uni-
versity people were highly suspect because they thought too much and as such
were probably not good citizens, even when they were Hutu. Those who lived on a
university campus and doctors were targeted” (Prunier 1995, 249). One survivor
talked of how teachers were particularly targeted:
Since there were not many schools open to Batutsis, due to the admis-
sions quotas in each district, we teachers would have the pupils sit in a
circle in the shade of tall leafy trees and we would improvise lessons right
there in the dust. In the Bugesera, the authorities and the administration
were Bahutu, as were the soldiers, the mayors, and those who controlled
the purse strings. So, as soon as a Batutsi caught on to some learning, he
became an instructor and taught school to Batutsi children. That is how
we teachers became very poorly regarded by the authorities, who were
clearly jealous. They did not dare silence us directly, but the moment any
killings began, teachers appeared high on the list, on the pretext that
they had ties to the inkotanyi. The inkotanyi were the Batutsi rebels,
the underground force in Burundi that launched attacks on Rwanda.

Whenever the inkotanyi attacked the Bahutus, the army would go kill
Batutsis as punishment. (Hatzfeld 2006, 67–68)
15.4.3. Was Inequality a Factor?

There is reasonably consistent evidence that genocide participation was higher
among Hutu who had fewer economic opportunities. There also is consistent
evidence that wealthier Tutsi, and Hutu, were commonly targeted. Still, a find-
ing that the richest were attacked is a finding about the intensive rather than the
extensive margin: wealth may determine who was attacked rather than deter-
mining whether an attack happens on a wealthy or on another person (Andre and
Platteau 1998). 3 For example, Andre and Platteau (1998) find that “troublemak-
ers” also were attacked, and that wealthy Hutu were the most likely persons to
have been killed, neither of which fact fits the rhetoric of interethnic inequality.
Violence levels were also high in areas with low inequality. Prunier (1995)
remarks that interethnic economic inequality in rural areas—unlike in Kigali—
was low. Friedman (2013) finds no evidence of greater violence in areas with
higher levels of well-being among Tutsi, in employment or education. One sur-
vivor reports that violence was high in her area, despite the story of Tutsi wealth
being a myth:
Hutus also say that we had too many cows. That wasn’t true. My parents
kept no cattle. Our neighbors raised no cattle, and their families were
larger and more needy. Cows at the market are there to be bought by
anyone with the money to buy them. The truth is that Hutu don’t like the
company of cattle. When a Tutsi spots a herd of cows in a grove of trees,
he sees good fortune. A Hutu, when he runs into some cows, he sees only
trouble and trampling hoofs. (Claudine Kayitesi, quoted in Hatzfeld
2006, 203)
Stories of scores being settled, as with the “troublemakers” in the sample of

Andre and Platteau (1998), may provide an alternative explanation for the empiri-
cal finding that wealthy citizens were more likely to be attacked. Prunier (1995,
231) writes:
This social aspect of the killings has often been overlooked. In Kigali the
Interahamwe and Impuzamugambi had tended to recruit mostly among the
poor. As soon as they went into action, they drew around them a cloud
of even poorer people, a lumpenproletariat of street boys, rag-pickers, car-
washers and homeless unemployed. For these people the genocide was
the best thing that could ever happen to them. They had the blessings of
350 Case Studies I
a form of authority to take revenge on socially powerful people as long as

these were on the wrong side of the political fence. They could steal, they
could kill with minimum justification, they could rape and get drunk for
free. This was wonderful. The political aims pursued by the masters of
this dark carnival were quite beyond their scope.
15.5. Organization
Whereas the discussion of vertical and horizontal inequality focused on indi-
vidual participants in genocide, even if as members of a group, this section
focuses on the organizers. By all accounts, the genocide was conducted in a
mind-boggling “efficient” manner. An enormous number of people were very
quickly mobilized and made capable of killing their neighbors. Beyond any
explanation of why, this demands a discussion of how. Rwandan government and
society were, and continue to be, well known for a high level of organization. A
highly functioning, decentralized system of local to regional government hier-
archy was in place, organizing community work long before the genocide, and
by many accounts, this system played an important role in facilitating the rapid
spread of violence. In addition to the government systems in place, scholars
have pointed to a prevailing culture that demanded obedience to authority,
which may have made it easier for individuals to follow their leaders even when
the orders were suspect.
15.5.1. State Capacity

It has been said that the speed and organization with which the violence spread
is evidence that the genocide was preplanned, not spontaneous. Prunier (1995,
242) writes that the
efficiency in carrying out the killings proves that these had been planned
well in advance. But the particularly chilling quality of that efficiency
is that, as in other genocides, it would not have been enough had it not
been for two other factors: the capacity to recruit fairly large numbers of
people as actual killers and the moral support and approbation of a large
segment, possibly a majority of the population.
Using an impressive dataset documenting which communes had loyalist burgo-

masters and which were led by opposition, McDoom (2014) finds that violence
began earlier in loyalist areas. He also finds that violence spread faster in areas
with greater state capacity (defined as being a shorter distance from the capital).
This is corroborated by recent work that uses the placement of roads and the
timing of rainfall, a variable used to exploit the variation in an exogenous variable

to connect to the presence of militiamen to recruit in rural areas: rainfall made
roads impassable and would delay recruiters (Rogall and Guariso 2013). And,
indeed, the author finds statistical evidence that levels of participation increased
when recruiters were enabled to visit.
This evidence is consistent with the findings of McDoom (2013a), showing
that not only were perpetrators more socially connected with other perpetra-
tors but also were more connected generally than nonparticipants. In a second
related paper, McDoom (2013b) shows that geographic proximity to other
perpetrators—i n the same neighborhood or household—is also associated with
a greater likelihood of participation, arguing that these close contacts encour-
aged those around them to join in participating. Relatedly, Fujii (2009) analyzes
detailed interviews with genocide participants and finds that social networks
played an important role in bringing people into the violent action. Similarly,
Hintjens (1999) discusses the role that the preexisting organizational structure
of government in rural areas played in facilitating the genocide. Thus, regular
community work programs, umuganda, were used to help organize the killing
during the genocide. Verwimp (2005) finds that perpetrators were overrepre-
sented among the educated in his sample of 352 people, another form of state
capacity (through schooling). Friedman (2013) likewise finds higher levels of
violence in areas with higher levels of education among Hutu. One potential
explanation for these findings is that education leads to a greater ability to mobi-
lize, both on the part of the local leaders of the violence as well as among those
to be mobilized for killing.
15.5.2. Radio as Coordination

The use of propaganda, specifically through radio communication, has been
widely discussed as an important means of pushing the violence forward. While
early work (Straus 2007) found only a weak relation between radio broadcasts
on the timing of violence onset, Yanagizawa-Drott (2013, 2014) exploits detailed
geographic data on the placement of radio towers and the topography of the
countryside to estimate the relation between where and when broadcasts could
be heard and the subsequent participation in the genocide. He finds that greater
levels of violence arose in those communes that had a higher degree of access to
broadcast messages. Interestingly, he argues that radio increased the violence by
coordinating participants, rather than by galvanizing them. He proposes a model
in which a lack of coordination can be a barrier to participation, to be alleviated
by a public signal. He finds a big impact of radio on participation in the violence
in areas where both a lot of people listened to radio and the Hutu majority was
large, facts that he argues demonstrate this mechanism of coordinating poten-
tial participants. He also found that radio mattered most in areas with relatively
352 Case Studies I
low levels of education. (For further analysis of media issues in genocides, see
chapter 12 in this volume.)
In the absence of motivation (willingness), organization and obedience (abil-
ity) are not dangerous. That said, with motivation, a weaker, less well-organized
state would not have been able to stage such an efficiently violent movement.
15.6. The Aftermath of Genocide

The short-and long-r un aftermaths of the genocide on Rwanda’s economy, pol-
ity, and culture are difficult to pull together. There is no question that the awful
experience has had and will continue to have broad effects on individuals and the
future of the country. Nonetheless, some researchers have delved into this terri-
tory. Exploring very specific aspects of the differential execution, experience, and
effects of the violence, these can help to understand certain facets of the genocide,
and perhaps help to understand how humanity copes with disaster, but should not
be taken as a summary of this genocide’s overall effects.
Many of the effects of the genocide follow directly from what has already been
discussed. For example, as the more educated were more likely to be attacked,
a long-r un decline in the stock of human capital follows, particularly in areas
that saw high levels of violence (de Walque and Verwimp 2010). In addition,
adults who were of school age during the genocide experienced long-lasting
reductions in their educational attainment (Akresh, Verwimp, and Bundervoet
2011). Verpoorten and Berlage (2007) find that households in which a member
was murdered or imprisoned suffered adverse long-term consequences in their
economic well-being. Justino and Verwimp (2013) find that even those who suf-
fered lesser shocks (a house destroyed, loss of land) experienced long-r un con-
sequences and a higher risk of poverty.
Returning to the Malthusian theory—t hat land pressure encourages violence,
which, when carried out, then eases the pressure—Rogall and Yanagizawa-Drott
(2013) find that surviving households in areas with greater levels of violence did
show higher levels of assets and of consumption in 2000. Following Yanagizawa-
Drott (2013, 2014), they estimate this using exogenous variation in violence lev-
els induced by radio access. However, these households also experienced higher
fertility, and their children report relatively low cognitive ability, so the authors
speculate that these gains could be short-l ived.
Men were disproportionately more likely to be killed, which resulted in skewed
sex ratios. La Mattina (2013) argues that areas with more dramatically skewed sex
ratios show lower female bargaining power and increased the incidence of domes-
tic violence. Finnoff (2012) also finds a positive correlation between genocide
intensity at the provincial level and intimate partner violence.
15.7. Conclusion
This chapter summarizes four important themes surrounding the economic
causes of the Rwandan genocide: Malthusian population density, poverty and
horizontal inequality, vertical inequality, and organization. Also highlighted are
a selection of economic consequences of the genocide, which are still in process
as Rwanda develops. In isolation, none of the four economic causes is sufficient
to explain the violence. The Malthusian hypothesis seems compelling given the
pregenocide conditions in Rwanda. There was a weak, and weakening, economic
system; pressure on scarce land increased due to high and rising population den-
sity; and limited alternative economic opportunities existed but, as discussed, the
fit of theory to empirical data is not strong. Other countries were just as densely
populated, and areas with higher population pressure within Rwanda do not
appear to have generated earlier or more intense violence. Poverty and horizontal
inequality were hugely important parts of the rhetoric of the genocidaires as well as
of the subsequent scholarly discussion. Moreover, interethnic (vertical) inequal-
ity did exist; Hutu with the fewest economic opportunities were most likely to
attack, and richer and elite individuals among the Tutsi were typically targeted.
Still, poorer people also were attacked, as were “troublemakers” and rich Hutu
as well as rich Tutsi. Hence, the evidence for horizontal and vertical inequality is
not entirely conclusive. Finally, while the organizational structures in place in the
country were quite instrumental in facilitating the rapid spread of the violence,
ability to conduct violence does not explain willingness.
The views on economic sources of genocide raised in this chapter, albeit not
definitive, do raise important issues for future scholarship. In particular, the
legacy of social domination and social envy rooted in Rwanda’s colonial history
suggests, in part, a genocidal impetus along class lines and not ethnic lines alone.
Although research on civil war and genocide are paying new and deserved atten-
tion to issues of horizontal (e.g., interethnic) inequality, it would be premature to
discount the potential importance of vertical inequality for Rwanda and perhaps
for other cases of genocide as well.
Notes
1. The number of people killed, and the number who participated, are both widely debated.
Straus (2004) used a broad range of empirical evidence to estimate that there were approx-
imately 200,000 perpetrators. This exercise was undertaken before 2005 when lists of
accused were drawn up based on the testimonies of victims and witnesses in each sector
as part of the Gacaca tribunals. In these records, more than 800,000 people are accused
of participating in the genocide in some form, about 300,000 of them accused of property
crimes and about 500,000 accused of organizing, killing, or attempted killing.
354 Case Studies I
2. As mentioned, according to the Gacaca records, about 300,000 individuals were accused
of property crimes, a category in which many of the female participants are listed.
3. A margin, in economics, refers to a set of constraints conceptualized as a border. An inten-
sive margin refers to a change in one or more constraints so that the border shrinks (e.g., the
removal of one worker reduces the amount of output that may be produced); conversely, an
extensive margin refers to a change in one or more constraints so that the border expands
(e.g., the addition of one worker increases the amount of output that may be produced).
References
Akresh, R., P. Verwimp, and T. Bundervoet. 2011. “Civil War, Crop Failure, and Child Stunting
Andre, C., and J.-P. Platteau. 1998. “Land Relations under Unbearable Stress: Rwanda Caught in
Boudreaux, K. 2009. “Land Conflict and Genocide in Rwanda.” Electronic Journal of Sustainable
Development 1, no. 3: 86–95.
Cramer, C. 2003. “Does Inequality Cause Conflict?” Journal of International Development 15, no.
4: 397–412.
De Walque, D., and P. Verwimp. 2010. “The Demographic and Socio-Economic Distribution of
Excess Mortality during the 1994 Genocide in Rwanda.” Journal of African Economies 19, no.
2: 141–62.
Des Forges, A. 1999. “Leave None to Tell the Story”: Genocide in Rwanda. New York: Human Rights
Watch. http://w ww.hrw.org/reports/pdfs/r/r wanda/r wanda993.pdf [accessed April
14, 2015].
Diamond, J. 2005. Collapse: How Societies Choose to Fail or Succeed. Rev. ed. New York: Penguin.
Finnoff, K. 2012. “Intimate Partner Violence, Female Employment, and Male Backlash in
Rwanda.” Economics of Peace and Security Journal 7, no. 2: 14–2 4.
Friedman, W. 2013. “Local Economic Conditions and Participation in the Rwandan Genocide.”
Households in Conflict Network. Working Paper No. 160. http://w ww.hicn.org/.
Fujii, L. A. 2009. Killing Neighbors: Webs of Violence in Rwanda. Ithaca, NY: Cornell University Press.
Gourevitch, P. 1998. We Wish to Inform You That Tomorrow We Will be Killed with Our Families: Stories
from Rwanda. New York: Farrar, Straus, & Giroux.
Hatzfeld, J. 2006. Life Laid Bare: The Survivors of Rwanda Speak. New York: Other Press.
Hintjens, H. M. 1999. “Explaining the 1994 Genocide in Rwanda.” Journal of Modern African
Studies 37, no. 2: 241–86.
Justino, P., and P. Verwimp. 2013. “Poverty Dynamics, Violent Conflict, and Convergence in
Rwanda.” Review of Income and Wealth 59, no. 1: 66–9 0.
La Mattina, G. 2013. “Civil Conflict, Sex Ratio and Intimate Partner Violence in Rwanda.” Tech
nical Report, Boston University (Working Paper Series). http://ssrn.com/abstract=2186916
or http://d x. doi.org/10.2139/ssrn.2186916 [accessed December 4, 2013].
Mamdani, M. 2002. When Victims Become Killers: Colonialism, Nativism, and the Genocide in
Rwanda. Princeton, NJ: Princeton University Press.
McDoom, O. S. 2013a. “Anti-Social Capital: A Profile of Rwandan Genocide Perpetrators’
Social Networks.” Journal of Conflict Resolution 58, no. 5: 866–94.
McDoom, O. S. 2013b. “Who Killed in Rwanda’s Genocide? Micro-Space, Social Influence,
and Individual Participation in Inter-Group Violence.” Journal of Peace Research 50, no.
4: 453–67.
McDoom, O. S. 2014. “Predicting Violence within Genocide: A Model of Elite Competition and
Ethnic Segregation from Rwanda.” Political Geography 42, no. 1: 34–45.
Olson, J. M. 1994. “Demographic Responses to Resource Constraints in Rwanda.” Department

of Geography and the Center for Advanced Study of International Development.
Lansing: Michigan State University.
Percival, V., and T. Homer-Dixon. 1998. “Environmental Scarcity and Violent Conflict: The
Case of South Africa.” Journal of Peace Research 35, no. 3: 279–98.
Prunier, G. 1995. The Rwanda Crisis: History of a Genocide. New York: Columbia University
Press.
Rogall, T., and A. Guariso. 2014. “The Escalation of Violence: Armed Groups and Civilian
Perpetrators.” http://w ww2.warwick.ac.uk/fac/soc/economics/events/2 014/3/phd_
conference_ 2 014/t horsten_rogall.pdf [accessed April 14, 2015].
Rogall, T., and D. Yanagizawa-Drott. 2013. “The Legacy of Political Mass Killings: Evidence
from the Rwandan Genocide.” http://w ww.hks.harvard.edu/fs/dyanagi/R esearch/
Legacy_of_Rwandan_Genocide.pdf [accessed April 14, 2015].
Stewart, F. 2002. “Horizontal Inequalities: A Neglected Dimension of Development.” Queen
Elizabeth House. Working Paper No. 81. Oxford: University of Oxford. www3.qeh.ox.ac.
uk/pdf/qehwp/qehwps81.pdf [accessed April 17, 2015].
Straus, S. 2004. “How Many Perpetrators Were There in the Rwandan Genocide? An Estimate.”
Straus, S. 2006. The Order of Genocide: Race, Power, and War in Rwanda. New York: Cambridge
University Press.
Straus, S. 2007. “What Is the Relationship between Hate Radio and Violence? Rethinking
Rwanda’s Radio Machete.” Politics and Society 35, no. 4: 609–37.
Uvin, P. 1998. Aiding Violence: The Development Enterprise in Rwanda. West Hartford,
CT: Kumarian Press.
Vansina, J. 2012. Le Rwanda ancien: le royaume nyiginya, Paris: Karthala Editions.
Verpoorten, M. 2012. “Leave None to Claim the Land: A Malthusian Catastrophe in Rwanda?”
Verpoorten, M., and L. Berlage. 2007. “Economic Mobility in Rural Rwanda: A Study of the
Effects of War and Genocide at the Household Level.” Journal of African Economies 16, no.
3: 349–92.
Verwimp, P. 2005. “An Economic Profile of Peasant Perpetrators of Genocide: Micro-L evel
Yanagizawa-Drott, D. 2006. “Malthus in Rwanda? Scarcity, Survival and Causes of the
Genocide.” https://ideas.repec.org/p/h hs/g unwpe/0 201.html [accessed April 14, 2015].
Yanagizawa-D rott, D. 2013. “Propaganda vs. Education: A Case Study of Hate Radio in
Rwanda.” In J. Auerbach and R. Castronovo, eds., The Oxford Handbook of Propaganda
Studies. New York: Oxford University Press, 378–94.
Yanagizawa-Drott, D. 2014. “Propaganda and Conflict: Theory and Evidence from the Rwandan
Genocide.” Quarterly Journal of Economics 129, no. 4: 1947–94.
16
Peace and the Killing

Compatible Logics in the Democratic Republic of the Congo
Zoë M a r r i age
16.1. Introduction
Mass atrocities in the Democratic Republic of the Congo (DR Congo or Congo),
and principally the widespread destruction of civilian life through war and dis-
placement, started in 1998 with the onset of the second Congo war, a conflict that
involved most of the countries in the Great Lakes region of central Africa. The
first Report of a United Nations (UN) expert panel on the illegal exploitation of
natural resources and other forms of wealth of the DR Congo was published in
2001. The panel detailed profits derived from looting by the Rwandan, Ugandan,
and Burundian armies and their allied militias, and named companies deemed
to be in violation of Organization for Economic Cooperation and Development
(OECD) guidelines (UNSC 2001). The import of the report was that regional
military and international economic actors were part of the dynamics of the war,
and that their activities were sustained by economic gain.
Policy implications followed, harnessing the notion that addressing the eco-
nomic functions of violence could throw its mechanisms into reverse and lead
to peace. The Global and All-Inclusive Agreement on the Transition of the
Democratic Republic of Congo of late 2002, agreed between the Congolese
government, militia forces, political opposition, and civil society, and funded by
northern donors, rested on the supposition that economic incentives could be
reordered through an elite political settlement and the liberalization of the coun-
try’s mineral trade. What ensued, though, was not sustainable development and
security fueled by the exploitation of the country’s mineral resources; instead, the
mass atrocities persisted and profits continued to accrue to political elites, those
of neighboring countries, their armies and allied militias, and foreign companies.
In examining the persistence of mass atrocities, this chapter moves beyond a
focus on economic incentives for decision making in its discussion of the second
356
Pe a c e a n d t h e K i l l i n g 357
Congo war and the terms of the peace. It takes its cue from the reggae band The
Congos who, referring to the betrayal of the Messiah, sang, “For thirty pieces of
silver they sold Jah Rasta, and why did they do that?” This lyric foregrounds the
transactional basis of economics in benefiting both parties. It also captures that,
while economic returns are instrumental, they do not fully explain behavior. The
answer, “For thirty pieces of silver,” would be facile. The lyric is used to frame an
analysis of how economics is shaped by power and violence.
The chapter presents the case that a bargain was struck between northern
donors and elite Congolese politicians in the peace agreement that provided
unmonitored aid and facilitated the rapid, unregulated liberalization of assets.
These processes reinforced a political economy that had been ordered through
war and excluded the majority of the population from political and economic
returns. Donors focused on high-value goods and high-profile politicians to effect
a transition to a form of peace that reduced military activity in the region and
responded to the emergence of China as a dominant force. These political aims
were more significant to the global north than the economic gains on mining
(albeit they were connected), and justified the expenditure made to the region
through aid. By bringing together the politics of economic incentivizing with a
discussion on the nature of contemporary mass atrocities, including genocide, the
chapter generates conclusions that have significance beyond the Congolese case.
16.2. The War Machine

Regarding collective violence, Mbembe theorizes a “war machine,” which places
mass atrocities in a political context and facilitates an investigation into how key
actors contribute to mass atrocities. Mbembe writes:
War machines are made up of segments of armed men that split or merge
with one another depending on the tasks to be carried out and the cir-
cumstances. Polymorphous and diffuse organizations, war machines are
characterized by their capacity for metamorphosis. Their relationship to
space is mobile. Sometimes, they enjoy complex links with state forms
(from autonomy to incorporation). (Mbembe 2003, 32)
The concept of a war machine does not rest on explaining only the moment
of death, but is concerned with the set of processes that makes life irrelevant and
death the predictable outcome. The metaphor is helpful in maintaining elements of
functionality and agency while acknowledging also a complexity and interplay of
parts: politics incorporates diverse actors into its processes through threat or
opportunity. As such, Mbembe’s theorization allows connections to be made
among economics, political science, and genocide studies. The conceptualization
358 Case Studies I
of a machine moves away from a sole focus on economic gain (as static and iso-
lated) to a perspective on relationships and interaction.
16.2.1. Rwanda, Uganda, and Their Militias

While all the countries in the central African region had strategic interest in
Congo’s fortunes and became embroiled in the conflict, the territorial assault
of the Congo wars of the late twentieth century was led by Rwanda and Uganda
(Clark 2002). During the first Congo war (1996–1997), Rwanda had security con-
cerns over the former Rwandan regime and allied militia forces, which, having led
the 1994 genocide, were exiled in eastern Congo and were largely concentrated
in militarized camps. In 1996, the Rwandan army destroyed these camps and dis-
persed the inhabitants. The principal armed group responsible for the genocide
in Rwanda was originally known as the Interahamwe; it regrouped as the Army
for the Liberation of Rwanda (ALiR) during the first Congo war, and became the
Democratic Forces for the Liberation of Rwanda (FDLR) in 2000.
In addition to military achievements, Rwanda and Uganda made significant
economic gains in the first war and, opening hostilities again in 1998, Rwanda
revealed an ambitious agenda by its attempt to take the Congolese capital,
Kinshasa. Failing to do so, Rwanda and its militias settled for militarily enforced
economic exploitation of the east of the country, occupying territory as far as
Kisangani, while Uganda and its militias occupied the north. Disputes between
Rwanda and Uganda, and the military-economic complexes established by the
invading forces, occasioned extreme violence on Congolese civilians.
Rwanda and Uganda’s military and economic operations interacted with atrophied
Congolese governance. Mbembe posits that the war machine is formed in “direct rela-
tion to the erosion of the postcolonial state’s capacity to build the economic underpin-
nings of political authority and order” (Mbembe 2003, 33). In the absence of coherent
Congolese state activity, and further undermining the possibility of economic con-
solidation, the occupying forces used violence and promoted local insecurity to gain
control of mines, roads, forests, and airports. Rwanda and Uganda’s control of min-
ing activity was founded on prewar networks of transborder trade and connected
opportunistically with the “informalized” economics and politics of eastern Congo
(Jackson 2006, 430; Vlassenroot, Perrot, and Cuvelier 2012, 7).
Rwanda did not engage the FDLR, ignoring information about their location
and focusing instead on areas that offered economic returns. Both Rwanda and
Uganda saw massively increased production or exports of minerals that were
not extensively mined in either of these countries (APPG 2002, 11). Table 16.1
details the gains made in the first (1996–1997) and second (1998–2002/2003)
Congo wars.
The economic conditions were given by the interaction between the incapac-
ity of the Congolese state, the military strength of Rwanda, and the international
Table 16.1 Ugandan and Rwandan Exports and Production, 1995–2 001
Panel A: Uganda Exports and Production
Gold Coltan Niobium Diamond
Exports Production Exports Production Exports Exports

(tons) (tons) (tons) (tons) (US$) (US$)
1995 3.09 0.0015 – 1.824 – –
1996 5.07 0.003 – – – –
1997 6.82 0.0064 2.57 – 13,000 198,302
1998 5.03 0.0082 18.57 – 580,000 1,440,000
1999 11.45 0.0047 69.50 – 782,000 1,813,500
2000 10.83 0.0044 – – – 1,263,385
2001 2,539,000
Panel B: Rwanda Production and Exports

Gold Cassiterite Coltan Diamond
production production production exports
(tons) (tons) (tons) (US$)
1995 1 247 54 –
1996 1 330 97 –
1997 10 327 224 720,425
1998 17 330 224 16,606
1999 10 309 122 439,347
2000 10 437 83 1,888,036
2001 1,245,000
Note: Although listed as “production,” Rwanda did not have appreciable quantities of any of
these resources (Jackson 2002, 526).
Source: APPG 2001, 19 (for Panel A) and 22 (for Panel B).
market. From the array of profits made, it was the trade in columbite-tantalite
(coltan) that became emblematic of the war economy and the use of coltan in
mobile phones, games consoles, and computers linked the pillage in Congo to
the purchase of luxury consumer products elsewhere. The coltan spike in 2000
saw prices increase tenfold, with Rwanda making between USD 150 and USD
200 million that year as the Rally for Congolese Democracy (RCD), the largest
Rwandan-sponsored militia, imposed a monopoly on the buying and selling of
coltan (Jackson 2002, 526; Reid 2006, 79).
360 Case Studies I
16.2.2. Northern Markets and Aid

The 2001 UN Panel of Experts report focused on illegal exploitation in eastern
Congo and recommended sanctions against Rwanda, Uganda, and Burundi;
a second report included findings from government-controlled areas. The reports
recorded that in the first year of the second Congo war the invading armies pil-
laged stocks and thereafter established the means of extraction. Further reports
followed,1 reiterating the message that the pillage was unacceptable according to
international norms.
Two factors weakened the UN Panel’s influence of “naming and shaming”
individuals, companies, and military structures, and implicated northern donors
in the war machine. The first was that the various economies in Congo were ruled
by violence; that was something regional armies were prepared to invest in but the
UN was not, and still less so the pressure groups or donor agencies. Second, there
was no robust political action to substantiate the claims that the trade in conflict
minerals was unacceptable. On the contrary, Rwanda and Uganda received tacit
approval for their local management of the war economies, as there was minimal
effort to monitor the activities of companies based in the global north (Marriage
2013a, 32).
In addition to their inability or unwillingness to regulate economic activity,
donors backed Rwanda and Uganda with aid throughout the two Congo wars and
afterward. Improvements in the creditworthiness of Rwanda and Uganda led to
debt relief resulting in a 50 percent fall in the service payments made by both coun-
tries. The United States, United Kingdom, European Union, and the Netherlands
provided the majority of aid and held up Rwanda and Uganda as models for devel-
opment (Reid 2006, 82). This stood in contrast to the fact that aid relations had
been cut with Congo in the early 1990s and the World Bank had declared the coun-
try bankrupt in 1994, so there were no bilateral or multilateral efforts to support
the Congolese government during the occupation by Rwanda and Uganda.
The aid money given to Rwanda and Uganda shored up the trade partnership
with a political alliance as donors became vested in the development successes of
the recipient countries. As the economics were regional, so too were the politics,
and the area in eastern Congo became the “central periphery” (Jackson 2006,
427) around which military and economic power was arrayed. The consonance
between aid and trade relationships meant that northern support or oversight rec-
reated conditions for violence to be inflicted by regional military actors.
16.2.3. Implication of the Congolese

The involvement of the Congolese in the mass atrocities was an intricate element
of the war machine. Mbembe surmises that “the lowest form of survival is kill-
ing” (Mbembe 2003, 36). He theorizes with regard to the African continent that
“coercion itself has become a market commodity … the vast majority of armies
are composed of citizen soldiers, child soldiers, mercenaries, and privateers”
(Mbembe 2003, 32). In Congo, estimates ran to 300,000 combatants with only a
fraction of these being salaried members of the army, and child recruitment being
widespread (ICG 2006b).
The Congolese population was implicated in the war economies through the
militarization of the area under occupation, which incorporated local economic
transactions into the mechanisms of the war. Forced recruitment of fighters
and the formation of self-defense units exposed young men and boys to height-
ened levels of aggression, and increased the violence, including sexual violence,
inflicted on others (Amnesty International 2003). Patterns of violence fed into
one another as displacement from agricultural land and loss of livestock linked
up with the risky enticement of mining or trading in harsh and dangerous con-
ditions. Local militias attacked civilians for revenge, for scavenging, or in bat-
tles over mineral access, and civilians paid taxes to the armed groups that were
destroying them.
As combatants or victims, the systematic use of violence dissolved boundar-
ies between trade, taxation, and thieving. The economy had been informalized
through Mobutu’s rule (MacGaffey 1991), but the time-honored practice of
“fending for yourself ”—the primary means of survival in Congo—was radi-
cally changed during the second war “from perceptions of the heroic to percep-
tions of criminal domination by ‘foreigners’ and ‘Congolese traitors’ ” (Jackson
2002, 515). “The difference today,” wrote the British All-Party Parliamentary
Group (APPG) with reference to the wartime economy, “is that it appears to be
‘exploitation by force’ of not only mineral resources but also livestock, crops and
labor” (APPG 2002, 28).
As the population folded in on itself, it was controlled by the symbolic terms of
“life force atrocities” (von Joeden-Forgey 2012, 91; see also chapter 17 in this vol-
ume) that invert or destroy human relations. The incorporation of violence into
decisions about where to live and how to gain access to food meant that the pro-
cesses of the war economy were entwined at the level of individual and household
survival. The use of force to control trade, extort taxes, and expropriate goods
made it central to the mechanics of economics, not a byproduct of them, and
excluded from economic and political viability those who did not have access to
the means of violence.
16.3. Peace: For Thirty Pieces of Silver

The war did not reach a military conclusion: Elite belligerents had settled
into operational patterns that were profitable and tolerably low-risk for them.
The Global and All-Inclusive Agreement, signed in late 2002 and finalized in
362 Case Studies I
April 2003, was designed and implemented by northern powers and reflected
their s trategic interests. It coaxed elite belligerents from the region to comply
minimally with northern demands for a formal close of hostilities and opening
of markets (see chapter 7 in this volume). Northern tolerance for instability in
developing countries had diminished following the attacks on the United States
in 2001, and increasing Chinese investment on the African continent threatened
the privileged access to Congo’s mineral resources that the United States and
Europe had enjoyed (Marriage 2013b, 140–41). The northern pressure toward
the peace agreement had a strong economic rationale, evident in the moves to
reorder the economic incentives for the belligerents and, more ambitiously, to use
the postwar moment to liberalize the economy.
Joseph Kabila, who had taken over the presidency upon his father’s murder
in January 2001, remained in post at the head of the transitional government,
which took office in 2003. The heads of the most powerful insurgent groups—
the RCD and the Movement for the Liberation of Congo (MLC, Uganda’s
primary militia force)—were granted vice-presidential positions, alongside
representatives of the nominally civilian opposition: one former member of
the RCD and one from the president’s party. The withdrawal from the frontline
of the war by the Rwandan and Ugandan armies was the signal concession fol-
lowing the signing of the peace, and it allowed the country to be formally polit-
ically united. The appointment of vice presidents institutionalized predatory
taxation, and the withdrawal of troops was rewarded with two new sources of
revenue, one from aid money and the other from the spoils of the unregulated
process of liberalization.
16.3.1. Continuing Pillage and Taxation

Despite the formal withdrawal of foreign troops, fighting persisted in the east
of Congo, increasing in some areas. The Kivus, the provinces directly bordering
Rwanda, continued to be annexed by Rwandan-sponsored militias. Meanwhile,
two new theaters of conflict opened. In Ituri district, the Ugandan withdrawal
led to contestation between rival Congolese groups, the Hema and Lendu, in a
violent conflict that took on a dynamic of its own. In Northern Katanga, erstwhile
government-a llied local militias, the Mai Mai, turned against the army assess-
ing that the terms of the peace had been detrimental to them (Vlassenroot and
Raeymaeker 2004; ICG 2006a). Insecurity had been a key mechanism in the war
for extraction, and the ongoing violence in the east foreclosed the possibility of
state regulation and maintained the conditions in which militias could sustain
the war economies despite the formal end of hostilities.
The resurgence or realignment of armed groups was accompanied by the for-
mation of new militias. In 2006 Laurent Nkunda formed the National Congress
for the Defense of the People (CNDP), which claimed to protect Tutsi groups in
eastern Congo and wreaked extraordinary violence on others. Shortly after the
CNDP’s demise in 2009, the M23 (an insurgent movement named after the failed
peace agreement of March 23, 2009) took over, comprising many of the same per-
sonnel as the CNDP and espousing a similar mission.2 In a show of strength, the
M23 took the provincial town of Goma in November 2012 but, failing to precipi-
tate political concessions, it withdrew a few days later, while maintaining its influ-
ence over mineral extraction in North Kivu.
In Kinshasa, the members of the Transitional Government used their time
in office to multiply the forms of extractive taxation imposed on the population.
Taxation was one of the most profitable nonmineral revenues during the war as
belligerents captured state institutions, and it increased during the transition as
each vice president fought his corner (Global Witness 2006; Laudati 2013, 35).
The imposition of taxes without the provision of services meant that the popula-
tion was another resource exploited by those with the means of violence. Oxfam
reported that “in 2012, government soldiers, armed rebels, police and civilian
authorities are all vying for the right to exploit local communities and extort
money and goods from them, pushing people further into poverty and undermin-
ing their efforts to earn a living” (Oxfam 2012, 1). The years of pillage, lack of ser-
vice provision, and war meant that further taxation critically threatened people’s
efforts to survive in conditions of generalized destitution.
16.3.2. Pillage of Aid

Northern donors had stopped providing aid to Congo in the early 1990s as
Mobutu had worn through their patience and the country’s strategic significance
dropped with the end of the cold war. The reengagement of aid donors at the turn
of the twenty-fi rst century was rapid and uncoordinated: USD 10 billion of bilat-
eral and multilateral aid was injected between 2002 and 2006 to bankroll the
peace process and host the presidential elections (Stearns and Wrong 2006).
A key expenditure of donor reengagement was of that of buying out the bel-
ligerents by establishing political institutions and opportunities that were more
profitable for them than military engagement. The approach was flawed as there
was no way to limit the profits taken, and as the belligerents retained the possibil-
ity of returning to war, they could act as spoilers at any moment (HRW 2005). The
refusal of the vice presidents to meet each other meant that no common purpose or
strategy was built within the government, and for each officeholder defaulting was
the rational option. Through ongoing violence and political reversals, donor pres-
sure was imposed chiefly through the Commission to Accompany the Transition
(known by its French acronym, CIAT). This was charged with the task of keeping
the transition on track, and it accepted all compromises in achieving this goal.
The opportunities offered by aid funds and the lack of regulation set up a
disastrous political dynamic: the vice presidents had little chance of succeeding
364 Case Studies I
in the elections and free rein with aid money. Englebert and Tull refer to a “cor-
ruption binge” by elite politicians alongside a “lenient approach towards corrup-
tion” on the part of the donors (Englebert and Tull 2008, 123). Preparations for
the elections were undermined by the inauspicious political and economic cir-
cumstances, but delaying involved prolonging the corruption of the transition.
Donors provided USD 546.2 million for the presidential elections of 2006, which
constituted 90 percent of the total costs (Minani Bihuzo 2008, 105). These funds,
too, were “literally pillaged” (Ngoma-Binda, Madefu Yahisule, and Mombo 2010,
119), reinforcing the political economy domestically by enriching the elite politi-
cians at the expense of investment in the population.
16.3.3. Unregulated Liberalization

The oversight of ongoing violent extraction in the east and the pillage of aid funds
in Kinshasa took place within the rapid and haphazard process of liberalization,
first of the economy in 2001, and the following year of the mineral trade. The lib-
eralization was formulated by donors, with the policy being drafted by the World
Bank, and it stemmed from the neoliberal assumption that the country’s massive
wealth could be harnessed to promote development and security (DFID 2008).
Resistance to liberalization in Kinshasa accompanied a de facto acceptance of its
processes, leading to “quasi-schizophrenia” as officials accommodated the donors’
policy despite the political impossibility of operationalizing it (Moshonas 2012).
Some initiatives were mounted to enhance the economic environment for lib-
eralization. The Lutundula Report, which was published in 2005, reviewed the
terms of the concessions sales made during the war and in the immediate post-
war era, producing a highly sensitive document (Lutundula Commission 2005).
The report was never discussed in parliament, though, and opaque sales of the
huge parastatal companies Gécamines and MIBA (mining copper and diamonds,
respectively) reproduced wartime economies in the patterns of personnel who
benefited and the clandestine nature of negotiations. During the transition, con-
cessions sales were made with no transparency or competition, and served to
enrich elite politicians (le Carré and Stearns 2006).
Taking place in a country with no functional political or legal institutions, the
liberalization was conducted without assessment of the value of assets. By the end
of the transition, control of more than 75 percent of the country’s copper mines
had been distributed outside of the country, many to companies registered in the
British Virgin Islands, with links to the city of London (Stearns 2011, 320–21;
Trapido 2015, 17). During the same period of time, the world food price crisis in
2007 prompted a rush on African land by multinational companies competing for
relatively inexpensive assets. Between 2008 and 2011, half of Congo’s agricultural
land was sold, while an estimated 70 percent of the population lacked access to
adequate food (BBC 2012).
16.3.4. Outcomes
The impact of transition and liberalization on the macroeconomic situation of the
region was positive according to the World Bank’s data and, after an uncertain
economic episode in the late 1990s, gross domestic product (GDP) per capita rose
steadily in Congo, as well as in Rwanda and Uganda (see Figure 16.1). But the rise
in GDP took place against a backdrop of continued elevated mortality. The most
widely cited figures on the deaths during and after the second Congo war are
taken from the International Rescue Committee’s (IRC) findings, first published
in 2000 (IRC 2000, 2008). This is the work that earned the second Congo war
the reputation of being the most deadly conflict since World War II. The results
of the IRC’s surveys are notable both for the high excess mortality during the war
and for the continuation of these figures after the peace had been signed. Between
April and July 2004, Congo had a 40 percent higher crude mortality rate than the
sub-Saharan regional average (Coghlan et al. 2006). 3
Alongside the deaths, violence has been meted out through extensive rapes
and other forms of sexual assault, committed mainly by combatants on civilians
and continuing after the peace was signed (HRW 2002; Bartels et al. 2013). This
violence falls within the UN Genocide Convention’s terminology of “causing seri-
ous bodily or mental harm to members of the group” and “imposing measures
intended to prevent births within the group” (Sharlach 2009, 181). Rape under-
scores the gendered element of the violence: “Genocide [is] a crime committed
GDP per
capita 700
(current
USD) Rwanda
600
Uganda
500
DRC
400
300
200
100
0
2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
Year
Figure 16.1 World Bank figures: GDP per capita (current USD). Figure adapted from
data available at the World Bank at http://data.worldbank.org/i ndicator/N Y.GDP.
PCAP.CD/countries/CD-RW-UG?display=graph [accessed July 22, 2014].
366 Case Studies I
predominantly by men” (von Joeden-Forgey 2012, 79), as the majority of the

perpetrators are part of male institutions such as the army or militia forces.
The IRC’s estimate of 5.4 million excess deaths between 1998 and 2007
was broadsided by the 2009–2010 Human Security Report (HSR). The HSR’s
authors critiqued the retrospective methodology, assessed that the data had
been collected in areas that were not random or representative, and gauged that
Congo’s extremely low level of development should be factored into the calcu-
lation. The HSR authors amended the excess mortality figure of 5.4 million to
1.5 million: 678,000 for the first year and 863,000 for the period of 2001 to 2007
(HSRP 2010, 131).4
The debate over how many excess deaths resulted from the war largely pivots
on where the baseline for normal death rates is set (Lambert and Lohlé-Tart 2008;
Kapend, Hinde, and Bijak 2013). The IRC uses the average mortality rate for sub-
Saharan Africa of 1.5 per 1,000 per month. The IRC’s detractors argue that this is
too low: HSRP takes a 2 per 1,000 per month baseline, which is what IRC finds
in western Congo. Wherever the baseline is set, the IRC’s figures are likely to be
robust relative to each other (although the HSRP reserved specific criticism for
the first survey), with the implication that high mortality rates continued after
the peace had been signed. (On the demography of genocide, see chapter 4 in this
volume.)
Two observations arise from the discussion. First, Congo gives reason to chal-
lenge the priority given to “event-based” emergencies over “process-based or
institutional ones” (Jones 2012, 145) in the study of genocide. Second, the vic-
tims are unmissed: the IRC’s analysis produces a figure that is 270 times higher
than that of Lambert and Lohlé-Tart, but no other data has been gathered on a
comparable scale that could corroborate or refute the positions. These lessons
indicate the need to enquire into the intertwined question of the continuation or
generation of new mechanisms of violence after the peace had been signed, and
into the processes by which the population remained excluded, even through a
process of democratization.
16.4. The Mechanisms of Betrayal

Emmanuel Mbi, the World Bank’s country director in Congo, claimed that “the
early provision of peace dividends, in the form of concrete actions that reach the
population, is critical to sustain the momentum for peace” (quoted in Heyes and
Burge 2003, 29). While his claim is consistent with the World Bank’s stated pol-
icy, in reality the success of the agreed-upon peace did not depend on distribut-
ing dividends; instead, it required actively excluding the population so that they
could not place political or economic demands on the processes. The momentum
for the peace did not come from, and was not maintained by, the population.
The donors’ economic rationale that informed the peace betrayed the popula-
tion on three interrelated counts: it was a bargain made exclusively with high-
profile politicians, it disproportionately regarded high-value goods, and it was
operationalized by strategically overlooking violence. As a result, the population
was marginalized from the economic and political order that emerged. Moreover,
because the peace did not galvanize the bargain between the members of the tran-
sitional government, the intransigence and acquisitiveness of these politicians
forestalled economic reform and with it the possibility of the market acting as a
driver for development.
16.4.1. High-Profile Politicians

Raeymaekers (2013) argues that in processes of state-making, political legiti-
macy is forged through the acceptance of the business community, not the popu-
lation more widely, and the implications of this are exacerbated when business
is conducted through violence. Drawing on Tilly, Raeymaekers examines “how
symbiotic relationships between violent rulers and (trans)national capitalists in
wartime may acquire some level of legitimacy in the context of post-war states.”
He finds that “international peacebuilding practice appears to lead to a grow-
ing integration of wartime accumulation networks in post-war government”
(Raeymaekers 2013, 602).
In Congo the integration of wartime accumulation through accommodation
of high-profile politicians reinforced the existing political stratification, extend-
ing elite politicians’ powers over the population through the consolidation of
their position. Despite mounting evidence of illegal activity, and some public
denouncement of this, no criminal charges were brought regarding irregularities
in the liberalization process (ICG 2005, 17). None of the men in presidential or
vice-presidential positions had achieved their positions through popular support
or a vote, meaning that they did not have a constituency to answer to with regard
to the provision of goods or services after the peace had been signed. Their power
depended instead on control of economic and military resources, and the weak-
ness of state institutions meant that even after the formal end of the war, corrup-
tion remained “systemic,” with estimates that 75 percent of exports were leaving
the country informally (Aust and Jaspers 2006, 50–53).
16.4.2. High-Value Goods

Throughout the liberalization process, the primary donor concern to integrate
the Congolese economy into the international market resulted in a focus on high-
value goods: mining, forestry, and agricultural land. This focus fit the “solution”
of liberalization that was offered in the donor-sponsored peace, and was linked
to measurable economic returns and macroeconomic stability. The liberalization
368 Case Studies I
strengthened the international stratification by imposing a northern steer on

political institutions and their interaction with the global economy, and exercis-
ing a degree of control on the distribution of Congo’s high-value economic assets.
The focus on the macroeconomic function of high-value goods did not stem
the violence that directed their trade. The implementation of the liberalization
meant that the population was squeezed both by the policy and by its contor-
tion. The intended outcome of the liberalization was the formalization of the
economy, which implied sloughing off artisanal miners, who populated the active
workforce. Redundancy schemes, such as that funded by the World Bank for
Gécamines workers, were practically unworkable on account of the institutional
malaise. Contortions in the liberalization meant that mining that was not sold
to foreign companies was recycled into the control of the army or other armed
groups (Global Witness 2010).
The inattention to the links with local-level violence and economics meant that
the relative success at macroeconomic conditions distracted from the continu-
ities on the ground. Daily needs were provisioned by low-value goods, particu-
larly hemp and charcoal, and violence continued to define these trading practices
(Laudati 2013, 37). These markets were not subject to reform or support, and
without capital assets, the distressed population had no political or military lever-
age to gain access to the economic arrangements after the war. A “negligible” sum
was spent on social services (Aust and Jaspers 2006, 34), and Jackson notes that
there was no “rebuilding of infrastructure to re-l ink severed territories” (Jackson
2006, 437).
16.4.3. Violence Overlooked

Throughout Congo’s transition and into the era of democracy, politics was con-
ducted through violence (Marriage 2013b, 102–4). In Kinshasa the stand-off
between Joseph Kabila and Jean Pierre Bemba, his closest contestant in the first
presidential elections, was amplified by a military threat as both maintained per-
sonal armies. These armies clashed at the time the results of the first round of
the presidential elections were announced in August 2006, and again the follow-
ing March, when several hundred people were killed on the streets of the capi-
tal. This violence served as a reminder to the population that the political realm
was the reserve of the militarily powerful. It drew some spoken disapprobation
from donors, but there was no political mechanism for stopping it, and it was
overlooked in the push toward the elections.
In the east of the country, hostilities were underpinned by Rwanda’s claim that
it was threatened by the FDLR. As the group dwindled, this claim became more
difficult to maintain internationally. Joint offensives between the Congolese and
Rwandan armies and the UN force against the few thousand fighters had little mil-
itary impact and brought catastrophic consequences for the population, who were
the victims of reprisal attacks (Oxfam 2009; Marriage 2012). Rwanda’s security
claims provided some cover for continued economic activity, but its expansionist
ambitions during the war had destroyed the credibility of the claim that its opera-
tions in Congo were commensurate with concerns about threats at the border.
Rwanda’s security claims were facilitated by the political interests of northern
donors. Having invested heavily in the East African region, both financially and
ideologically, northern donors maintained aid relations with Rwanda, in spite of
the evidence of its military and economic activity in Congo. The donors’ enthu-
siasm stemmed largely from the progress that Rwanda had made in liberalizing
its economy and toward the Millennium Development Goals (DFID 2011). In
addition, Rwanda’s contribution to the African Union and UN peacekeeping mis-
sion in Darfur protected it from criticism from donors who were not prepared to
commit troops.
The situation became increasingly contentious and politically untenable,
though. In 2012, even as some donors suspended funds following reports that
Rwanda was supporting the M23, others did not, and the M23 disbanded the fol-
lowing year when much of the aid had been reinstated (Marriage, forthcoming).
The erratic attempt to discipline Rwanda through the suspension of aid money
showed some donor acknowledgment of their responsibility, alongside a failure
to deal with this responsibility in a concerted way.
16.4.4. The Fall Guy

Mbembe employs the term necropolitics to explore the “subjugation of life to the
power of death” (Mbembe 2003, 39). For him, “biopower,” which is frequently
cited in critiques of neoliberalism (Dillon and Reid 2001; Duffield 2008), is insuf-
ficient as a concept to grasp the mechanisms of violence in distributing power.
Theorizing necropolitics, Mbembe makes the case that “weapons are deployed in
the interest of maximum destruction of persons and the creation of death-worlds,
new and unique forms of social existence in which vast populations are subjected
to conditions of life conferring upon them the status of living dead” (Mbembe
2003, 40).
The indulgence of Rwanda’s claims about the FDLR threat passed the cost of the
peace onto the population as victims of predatory taxation, lack of services, state
violence, and ongoing hostilities between militia forces in eastern Congo. The
population was politically and economically devastated: they could not resist the
Rwandan-supported militias and they could not force their voice at the negoti-
ating table. The collusion between the elite politicians and their international
funders arranged sales of the country’s assets on the global market and with them
the population’s stake in the country’s infrastructure.
The economic rationale of the peace agreement commoditized the country’s
mineral wealth, and with it the country’s human capital, valuing the population
370 Case Studies I
at zero or less on the global market: the population needed to be cleared from
the mineral and agricultural land. This valuation is evidenced by the divergent
estimates of the number of people killed and the fact that the loss of human capi-
tal is not presented as a cause of the country’s underdevelopment. They were, in
Mbembe’s terminology, the “living dead,” and were irrelevant to the country’s
political and economic development and security.
16.4.5. The Nature of the Genocide

According to the UN Convention on the Prevention and Punishment of the
Crime of Genocide, allocating the marker of genocide requires identifying two
factors: the intent of the perpetrators and a victim group. Observing how people
were killed indicates the nature of the intent, and detecting who has been killed
and what they had in common with other victims establishes what form the group
takes. The point is not to manipulate the category of genocide, but to harness the
processes that resulted in the convention in order to investigate contemporary
violence.
Thompson and Quets argue that “an actor performs an act … with intent if
there are foreseeable ends or consequences” (quoted in Fein 2009, 51). The issue
of intent is elaborated in case law. Following “the ad hoc tribunals for the former
Yugoslavia (ICTY) and Rwanda (ICTR), international law now allows the issue of
‘intent’ to be inferred from actions on the ground” (Totten and Bartrop 2009, 34).
Rwanda sent 20,000 troops to eastern Congo and supported a comparable num-
ber of RCD fighters, while Uganda sent 10,000 troops and supported the MLC
(UNSC 2001, 27). Intent in Congo can be inferred from the use of violence to
impose military and economic control over a large area of foreign territory.
Intent is not limited to the Rwandan and Ugandan armies and their sponsored
militias. Quigley’s The Genocide Convention: An International Law Analysis (2006)
problematizes “a situation in which an actor lacks actual knowledge but is aware of
a high probability of the existence of a fact and fails to inquire into circumstances
that would have made him aware of the existence of the fact” (quoted in Jones
2012, 135). “Willful blindness” and “reckless disregard” have also been included
in the definition of intent (Jones 2012, 147). In the case of Congo, the violation
of sovereignty by Rwanda and Uganda was indisputable, and the UN Panel of
Experts’ reports from 2001 onward meant that no UN member state could claim
ignorance of the slaughter in eastern Congo and the plunder that was intertwined
with it. The notion of “reckless disregard” indicates that there may be elements
that are not fully known, but that this does not eliminate the culpability of those
with the power to direct the machinery.
The long-a rming of violent economies during the war and the disruption of
the transition distanced northern governments and their populations from the
outcomes of their trade and aid policy. Feierstein, examining how people commit
genocide, notes “the metaphysical category of absolute evil distracts attention

from our everyday experience, leaving us safe from the distress that might be
caused by examining the genocidal potential latent in every modern society, and
all its members” (Feierstein 2012, 29). The northern participation in the econom-
ics of genocide has been blurred by the attention given to proximate actors and
by the fact that successes were made through the apparent “failure” of policy
on liberalization and democratization: the ways in which these processes were
appropriated by elite politicians. These apparent failures protected the ideology
supporting the policy from scrutiny as the fault was ostensibly at implementation.
As far as the group is concerned, Lieberman notes that “genocide research has
always struggled to describe victims” (Lieberman 2012, 14; also see c hapter 13 in
this volume), and in Congo, conventional genocide signifiers are not illuminat-
ing. The victims of mass atrocities are not identifiable with reference to the tags
of national, ethnic, racial, or religious groups that are established in the 1948 UN
Genocide Convention. Ethnic cleavages in Congo have largely been rent by vio-
lence rather than preexisting it, and the atrocities have been committed against
many groups.
Instead, the group of victims is characterized by its political space: violence has
been inflicted on people who fall outside state protection and have little economic
or military capacity and no legal or political recourse. The process of “othering”
has been recognized as a prelude to mass atrocities in other contexts, as groups are
rendered different or polluting in preparation for extermination (Feierstein 2012,
23–28). In Congo, the transition “othered” people who did not fit the demands of
the neoliberal policy. These were people who had no assets or skills for the market,
and they were dealt with through political and economic isolation that kept them
outside structures of power or security.
In accordance with ongoing definitional debate (see c hapters 1 through 3 in
this volume), genocide can usefully be conceptualized as “existential killing”
(Lieberman 2012, 3), as the “elimination of the continuity of a people” is funda-
mental to its operation (Charny 2009, 36). In Congo, the peace agreement was
built on the political economy that had been forged by violence; it is “existential
killing” in that it threatened to erase a section of the population from the political
and economic order of Congo, making death the predictable outcome. The recon-
figuration of political space in the wake of the peace agreement refined the gains
made through violence in the bureaucratized framework of a neoliberal market
that was continuously reinforced through violence.
16.5. And Why Did They Do That?

Kodi asserts that “the former warlords and their external partners accepted the
All-I nclusive Agreement because it … left their privileges intact” (Kodi 2007, 9).
372 Case Studies I
The enquiry, “Why did they sell Jah Rasta for thirty pieces of silver?” is not about
questioning economic rationality. Rather, it is asking how the circumstances
came around for this sale (and purchase) to be made. In the case of Congo, how
were power and violence used to frame the economic transaction on which the
peace agreement depended?
At first sight, the northern intervention in Congo is puzzling, as the speed of
the transition meant that the ruinous outcome was foreseeable. The transitional
government did not gain domestic legitimacy, and the political processes were
trammeled by a lack of commitment from the belligerents. The democratization
enticed them to further rapaciousness, and the liberalization was broadsided by
irregularities. The obvious weaknesses of the rapid and unmanaged processes
indicate that their rationale cannot be taken at face value. The policies make
more sense when understood as tactical operations that promote a larger strategic
project.
Northern donors paid off the belligerents because they could not wait, and had
no reason to, for democracy or economic growth to take place through processes
of inclusive development (many northern donors have a similarly ruthless disre-
gard for the victims of neoliberal development in their own countries). The collu-
sion between the northern donors and domestic elites provided the belligerents
with opportunities to consolidate their domestic networks and an internationally
recognized legitimacy that they could not have generated unaided. The pace and
extent of the reengagement with Congo contrasted sharply with the disinterest
demonstrated by aid donors when fighting was at its height. While aid was sham-
bolic in achieving its stated aims during and after the transition, the costs were
borne not by the donors but by the Congolese population.
The peace agreement shifted the locus of power from the Rwanda-Uganda axis
of violent governance to a dispersed neoliberal market that was governed, even if
not wholly controlled, by the global north. This shift in power reduced the scale of
the fighting and went some way toward staving off the competition by China for
primary control of Congo’s resources. For Rwanda and Uganda the peace agree-
ment was acceptable to the extent that it offered a political transition that retained
their military and economic advantage.
Chinese investment on the continent went from under USD 5 million at the
beginning of the 1990s to over USD 6 billion by 2005. In 2008, a Sicomines
deal between Chinese companies and the Congolese government sealed USD
9.25 billion of investment (Matti 2010, 408). The Chinese involvement was
directed by demand for raw materials, particularly copper, to support China’s
booming economy. For the donors, the peace and transition was about much
more than gaining mining concessions. It was about the way business was done,
and thus it constituted an ideological and political response to China more than
an economic one (which would frame China simply as another competitor within
an unchallenged system).
The peace was not only compatible with killing, it necessitated it. For the
collusion between the regional elites and northern donors to function, it was
imperative to exclude the population of Congo, who would otherwise have
derailed the process by slowing it and bringing disparate priorities to the table.
Mbembe, in his discussion of necropolitics, analyzes the “syntheses between
massacre and bureaucracy, that incarnation of Western rationality” (Mbembe
2003, 23). The peace strengthened political hierarchies nationally and interna-
tionally by privileging foreign capital, and regional and domestic elites through
the use and strategic oversight of violence. Liberalization and democratization
have contributed to the war machine, prolonging the killing rather than
protecting against it.
In 2013, a Peace, Security and Cooperation Framework was signed by repre-
sentatives from Congo, all nine neighboring countries, and South Africa with
explicit recognition of the “recurring cycles of conflict and violence” (UN 2013).
Around twenty armed groups, including the FDLR, remain operational, and
Oxfam noted two years later “how little progress has been made towards building
legitimate and credible state authority in many parts of eastern Congo” (Oxfam
2015, 3). Incentivizing belligerents ring-fences the domestic arena and downplays
its broader political context. Analyzing how the political context is ruled by power
and violence, though, explains how the economic incentives are formulated and
conceptualizes these as products of the political system, rather than simply as the
determinants of it.
16.6. Conclusions
During and after the war in Congo, the majority of the victims died from malnu-
trition and infectious diseases as a result of displacement, disruption of supply
routes, and political alienation. The liberalization, the support given through aid,
and the political and economic exclusion of the population through the transition
contributed to the necropolitics that consigned the population to the status of
“living dead.” This interplay of activity casts light on other forms of mass vul-
nerability that result from complex political interactions. The deaths arising, for
example, from pandemics or climate change occur without any party being in full
control of how interests coalesce. Inferring the intent of actors from their “actions
on the ground” and “willful blindness” allows for an understanding of intent to
be built that incorporates a political analysis with respect to the distribution of
benefits and costs incurred.
The seeming lack of a policy alternative to neoliberal policy is not a mitigating
but an aggravating factor as far as donor intervention is concerned. The domi-
nance of economic theorizing in conflict and postconflict development environ-
ments limits the scope for policy formulation and the range of perspectives that
374 Case Studies I
are included in setting the agenda. The political category of the group presents
a challenge to conventional genocide definitions and opens a line of analysis
with regard to dispersed and procedural atrocities. The “othering” performed
by violence-assisted neoliberal policy produces victims who can be geographi-
cally disparate: they share a political and economic identity with respect to
their vulnerability and relation to domestic state institutions and processes of
global capital.
Mbembe’s critique of the incarnation of Western rationality highlights the
way in which the demands of the market alienate and problematize those who
do not conform, and simultaneously highlights how contentious it is to unpick
the processes that make mass death “inevitable.” The implication of the analy-
sis of the war machine and its component parts of intent and groups is that
protecting against mass atrocities is complex and diffuse, requiring the dis-
mantling of war machinery that maintains incentives and opportunities for
violence.
Acknowledgments
The author would like to express her gratitude to Charles Anderton, Jurgen Brauer,
and three anonymous reviewers for their generous comments that contributed
greatly to this chapter.
Notes
1. http://w ww.un.org/sc/committees/1533/egroup.shtml (accessed July 17, 2014).
2. The M23 took their name from the peace agreement signed on March 23, 2009. It was the
perceived failure of the agreement that resulted in the mobilization of the group.
3. Excess mortality is understood as the deaths that would not have occurred if there had
not been a war. The first two surveys were carried out in the east of Congo, and the fol-
lowing three were nationwide. The results were cumulative, rising by around a million
deaths per year, with a final tally of 5.4 million excess deaths between 1998 and 2007
(IRC 2008). According to the last two IRC surveys, which were conducted after the
peace had been signed, less than 2 percent of excess deaths were attributable to direct
violence but deaths were concentrated in the areas in which direct violence was most
intense. The IRC team assessed: “Regression analysis suggested that if the effects of
violence were removed, all-c ause mortality could fall to almost normal rates” (Coghlan
et al. 2006, 44).
4. The HSR’s skepticism chimed with other assessments: Kapend, Hinde, and Bijak (2013)
calculated that between 1.6 and 1.9 million excess deaths had taken place between 1998
and 2007. Lambert and Lohlé-Tart (2008) estimated that 200,000 people had been killed
as a result of the war. Regarding child mortality, the IRC calculated that 47 percent of
victims of the war were children (IRC 2008, 26). The Demographic and Health Survey
(DHS) in 2007 focused on under-five mortality, which it rated as slightly over half the
excess mortality the IRC had found for the period 2006–2 007 (DHS 2007, 191).
References
Amnesty International. 2003. “The Democratic Republic of the Congo: Our Brothers Who Help
Kill Us: Exploitation and Human Rights Abuses in the East.” London: Amnesty International.
[APPG] All Party Parliamentary Group. 2002. “Cursed by Riches: Who Benefits from Resource
Exploitation in the Democratic Republic of the Congo?” London: All Party Parliamentary
Group on the Great Lakes Region and Genocide Prevention.
Aust, B., and W. Jaspers. 2006. “From Resource War to ‘Violent Peace’: Transition in the
Democratic Republic of Congo (DRC).” Paper 50. Bonn: Bonn International Center for
Conversion.
Bartels, S., J. Kelly, J. Scott, J. Leaning, D. Mukwege, N. Joyce, and M. van Rooyen. 2013.
“Militarized Sexual Violence in South Kivu, Democratic Republic of Congo.” Journal of
Interpersonal Violence 28, no. 2: 340–58.
BBC. 2012. “Analysis: Land Grab or Development Opportunity?” London: BBC http://w ww.
bbc.co.uk/news/world-a frica-17099348 [accessed August 9, 2013].
Charny, I. W. 2009. “The Definition of Genocide.” In S. Totten and P. R. Bartrop, eds., The
Genocide Studies Reader. London: Routledge, 36–4 0.
Clark, J. F. 2002. The African Stakes of the Congo War. New York: Palgrave Macmillan.
Coghlan, B., R. J. Brennan, P. Ngoy, D. Dofara, B. Otto, M. Clements, and T. Stewart. 2006.
“Mortality in the Democratic Republic of Congo: A Nationwide Survey.” The Lancet 367,
no. 9504: 44–51.
[DFID] Department for International Development. 2008. “Democratic Republic of Congo.
Country Plan.” London: Department for International Development.
[DFID] Department for International Development. 2011. “Operational Plan 2011–15. DFID
Democratic Republic of Congo.” London: Department for International Development.
[DHS] The Demographic and Health Survey. 2007. “The Democratic Republic of the
Congo: Key Findings.” Ministry of Planning and Ministry of Health. Macro International
Inc. Calverton, Maryland, USA.
Dillon, M., and J. Reid. 2001. “Global Liberal Governance: Biopolitics, Security and War.”
Millennium 30, no. 1: 41–6 6.
Duffield, M. 2008. “Global Civil War: The Non-I nsured, International Containment and Post-
Interventionary Society.” Journal of Refugee Studies 21, no. 2: 145–65.
Englebert, P., and D. M. Tull. 2008. “Postconflict Reconstruction in Africa: Flawed Ideas about
Failed States.” International Security 32, no. 4: 106–39.
Feierstein, D. 2012. “The Concept of ‘Genocidal Social Practices’.” In A. Jones, ed. New Directions
in Genocide Research. London: Routledge, 18–36.
Fein, H. 2009. “Defining Genocide as a Sociological Concept.” In S. Totten and P. R. Bartrop,
eds., The Genocide Studies Reader. London: Routledge, 44–56.
Global Witness. 2006. “Digging in Corruption: Fraud, Abuse and Exploitation in Katanga’s
Copper and Cobalt Mines.” Washington, DC: Global Witness.
Global Witness. 2010. “‘The Hill Belongs to Them’: The Need for International Action on
Congo’s Conflict Minerals Trade.” London: Global Witness.
Heyes, K., and R. Burge. 2003. “Coltan Mining in the Democratic Republic of Congo: How
Tantalum-Using Industries Can Commit to the Reconstruction of the DRC.” Cambridge,
UK: Fauna and Flora International.
[HRW] Human Rights Watch. 2002. “The War within the War: Sexual Violence against Women
and Girls in Eastern Congo.” New York: Human Rights Watch.
[HRW] Human Rights Watch. 2005. “Democratic Republic of Congo. Elections in Sight: Don’t
‘Rock the Boat’?” New York: Human Rights Watch.
[HSRP] Human Security Report Project. 2010. “The Human Security Report 2009/10: The
Causes of Peace and the Shrinking Costs of War.” Vancouver: Human Security Centre,
Simon Fraser University.
376 Case Studies I
[ICG] International Crisis Group. 2005. “The Congo’s Transition Is Failing: Crisis in the Kivus.”
Africa Report 91. Nairobi and Brussels: International Crisis Group.
[ICG] International Crisis Group. 2006a. “Katanga: The Congo’s Forgotten Crisis.” Africa
Report 103. Nairobi and Brussels: International Crisis Group.
[ICG] International Crisis Group. 2006b. “Security Sector Reform in the Congo.” Africa Report
No. 104. Nairobi and Brussels: International Crisis Group.
[IRC] International Rescue Committee. 2000. “Mortality Study, Eastern Democratic Republic
of Congo (April–May 2000).” New York: International Rescue Committee.
[IRC] International Rescue Committee. 2008. “Mortality in the Democratic Republic of Congo.
An Ongoing Crisis.” New York: International Rescue Committee.
Jackson, S. 2002. “Making a Killing: Criminality and Coping in Kivu War Economy.” Review of
African Political Economy 29, nos. 93–94: 517–36.
Jackson, S. 2006. “Borderlands and the Transformation of War Economies: Lessons from the
DR Congo.” Conflict, Security and Development 6, no. 3: 425–47.
Jones, A. 2012. “Genocide and Structural Violence: Charting the Terrain.” In A. Jones, ed., New
Directions in Genocide Research. London: Routledge, 132–52.
Kapend, R., A. Hinde, and J. Bijak. 2013. “The DR Congo Conflict (1998–2 004): Assessing
Excess Deaths Based on War and Non-War Scenarios.” Paper presented at International
Union for the Scientific Study of Population (IUSSP) meetings, Busan, South Korea.
Kodi, M. 2007. “Anti-Corruption Challenges in Post-Election Democratic Republic of Congo.”
Chatham House Report. London: Chatham House.
Lambert, A., and L. Lohlé-Tart. 2008. “Overmortality in Congo (DRC) during the 1998–2 004
Conflicts: An Estimate of Excess Deaths Scientifically Based on Demographic Methods.”
Belgium: Association pour le Développement de la Recherche Appliquée en Sciences
Sociales.
Laudati, A. 2013. “Beyond Minerals: Broadening ‘Economies of Violence’ in Eastern Democratic
Republic of Congo.” Review of African Political Economy 40, no. 135: 32–50.
le Carré, J., and J. Stearns. 2006. “Getting Congo’s Wealth to Its People.” Boston Globe. December
22, 2006. http://w ww.boston.com/news/g lobe/editorial_opinion/oped/a rticles/2 006/
12/22/getting_congos_wealth_to_ its_ people/.
Lieberman, B. 2012. “From Definition to Process: The Effects and Roots of Genocide.” In
A. Jones, ed., New Directions in Genocide Research. London: Routledge, 3–17.
Lutundula Commission. 2005. “Assemblee Nationale Commission Spéciale Chargée de l’Examen
de la Validité des Conventions a Caractere Economique et Financier. Conclues pendant
les Guerres de 1996–1997 et de 1998.” Kinshasa: République Démocratique du Congo.
www.congonline.com/documents/R apport_ Lutundula_ pillage_ 2006.pdf [accessed
May 6, 2015].
MacGaffey, J. 1991. The Real Economy of Zaire: The Contribution of Smuggling and Other Unofficial
Activities to National Wealth. London: James Currey.
Marriage, Z. 2012. “The Case of the FDLR in DR Congo: A Facade of Collaboration?” In
A. Giustozz, ed., Post-Conflict Disarmament, Demobilization and Reintegration: Bringing State-
Building Back In. London: Ashgate, 87–98.
Marriage, Z. 2013a. “Compliance Versus the Ragged Threat: Problem-Solving Security in the
Democratic Republic of Congo.” Civil Wars 15, no. 1: 19–36.
Marriage, Z. 2013b. Formal Peace and Informal War: Security and Development in Congo. London
and New York: Routledge.
Marriage, Z. Forthcoming. “Aid to Rwanda: Unstoppable Rock, Immovable Post.” In T.
Hagmann and F. Reyntjens, eds., Aid and Authoritarianism in Africa: Development without
Democracy. London: Zed Books.
Matti, S. 2010. “Resources and Rent Seeking in the Democratic Republic of the Congo.” Third
World Quarterly 31, no. 3: 401–13.
Mbembe, A. 2003. “Necropolitics.” Public Culture 15, no. 1: 11–4 0.
Minani Bihuzo, R. 2008. “1990–2 007. 17 ans de transition politique et perspectives democra-
tique en RDC.” Kinshasa: CEPAS/RODHECIC Médiaspaul.
Moshonas, S. 2012. “Beyond the Governance State: Aid Relations and State Reforms in the
Democratic Republic of Congo.” PhD Thesis. Bristol: University of Bristol.
Ngoma-Binda, P., J. O. Madefu Yahisule, and L. M. Mombo. 2010. “République Démocratique
du Congo. Démocratie et participation à la vie politique: une évaluation des premiers pas
dans la IIIème République.” Une étude d’AfriMAP et de L’Open Society Initiative for
Southern Africa. Johannesburg: Open Society Initiative for Southern Africa.
Oxfam. 2009. “DR Congo: Civilian Cost of Military Operation Is Unacceptable.” Oxford,
UK: Oxfam.
Oxfam. 2012. “Commodities of War: Communities Speak Out on the True Cost of Conflict in
Eastern DRC.” Oxfam Briefing Paper. Oxford, UK: Oxfam.
Oxfam. 2015. “Secure Insecurity.” Oxfam Briefing Paper. Oxford, UK: Oxfam.
Raeymaekers, T. 2013. “Post-War Conflict and the Market for Protection: The Challenges to
Congo’s Hybrid Peace.” International Peacekeeping 20, no. 5: 600–617.
Reid, T. 2006. “Killing Them Softly: Has Foreign Aid to Rwanda and Uganda Contributed to the
Humanitarian Tragedy in the DRC?” Africa Policy Journal 1 (Spring): 74–94.
Sharlach, L. 2009. “State-R ape: Sexual Violence as Genocide.” In S. Totten and P. R. Bartrop,
eds., Genocide Studies Reader. London: Routledge, 180–92.
Stearns, J. 2011. Dancing in the Glory of Monsters: The Collapse of the Congo and the Great War of
Africa. New York: Public Affairs.
Stearns, J., and M. Wrong. 2006. “Struggle for a Functioning Congo.” Financial Times,
August 3. http://w ww.ft.com/cms/s/2/f 28ddbfc-2315-11db-8 48d- 0 000779e2340.
html#axzz3ZNa0RPlG [accessed May 6, 2015].
Totten, S., and P. R. Bartrop, eds. 2009. The Genocide Studies Reader. London: Routledge.
Trapido, J. 2015. “Africa’s Leaky Giant.” New Left Review 92 (March–A pril): 1–23. http://
newleftreview.org/I I/92/joe-t rapido-a frica-s-leaky-g iant.
[UN] United Nations. 2013. “A Framework of Hope: The Peace, Security and Cooperation
Framework for the Democratic Republic of Congo and the Region.” Addis Ababa: Office of
the Special Envoy of the Secretary-General for the Great Lakes Region of Africa.
[UNSC] United Nations Security Council. 2001. “Report of the Panel of Experts on the
Illegal Exploitation of Natural Resources and Other Forms of Wealth of the Democratic
Republic of the Congo.” S/2 001/357. April 12. New York: United Nations Security
Council.
Vlassenroot, K., S. Perrot, and J. Cuvelier. 2012. “Doing Business Out of War: An Analysis of
the UPDF’s Presence in the Democratic Republic of Congo.” Journal of East African Studies
6, no. 1: 2–12.
Vlassenroot, K., and T. Raeymaekers. 2004. “The Politics of Rebellion and Intervention in
Ituri: The Emergence of a New Political Economy?” African Affairs 103, no. 412: 385–412.
von Joeden-Forgey, E. 2012. “Genocidal Masculinity.” In A. Jones, ed., New Directions in Genocide
Research. London and New York: Routledge, 76–95.
17
Gender and the Genocidal Economy

E l is a von Joe de n-F orge y
17.1. Introduction
Gender influences genocide in multifarious ways—from the construction of
identities of the groups involved, to the decision-making processes among victim
groups as they face genocide, to the patterns of destruction pursued by perpe-
trators. It is difficult to fully grasp the crime of genocide without taking gender
into account. So much of what perpetrators do in their efforts to destroy a group
involves undermining the group’s ability to reproduce itself biologically, cultur-
ally, and historically, all highly gendered processes. On a practical level, the rea-
sons for perpetrators’ attention to the reproductive capacities of a group are not
hard to understand: if one destroys a group’s capacity to reproduce, one destroys
the group as such. A closer look at the relationship between genocidal ideology
and gendered practice, however, suggests that there are forces beyond the merely
practical at play. Gendered patterns of destruction shine a light on aspects of
genocidal ideology that are tied to concepts of reproduction not only in a bio-
logical and social sense but also in a metaphysical one. This chapter examines the
gendered nature of a key facet of most genocidal ideologies: namely, the belief in
a highly gendered, metaphysical, zero-sum battle between perpetrator and tar-
get groups, both of which are conceptualized as generative units. I refer to this
gendered, cosmic, zero-sum framework as the “genocidal economy” and examine
how it regulates many of the techniques of destruction pursued by genocidaires,
particularly with reference to generative forces and the reproductive realm.
17.2. Gender and the Genocidal Economy

Genocidal ideologies are particularly gendered ways of thinking. They prob-
lematize as a core unit of analysis the human reproductive realm and often
quite self-consciously craft strategy and propaganda around their beliefs about
378
Gender and the Genocidal Economy 379
reproduction and its relationship to power. The group basis of the crime of geno-
cide emanates from this core preoccupation with reproduction, since genocid-
aires see power in metaphysical terms involving generative forces and generative
units. In the genocidal economy generative units exist in a constant metaphysi-
cal battle that revolves around certain binaries: light/dark, good/evil, worthy/
unworthy, pure/polluted. Perpetrators of genocide view their rival generative
units not only as enemies, but as cosmic enemies: eternal, preternaturally pow-
erful, unrelentingly destructive, and embroiled in an invisible war that is about
securing access to the source of life itself. Thus, when leaders of genocide imagine
their enemies, they imagine them within what Jacques Semelin calls the “imagi-
naire of death,” by which he means a logic in which the destruction of the “other”
is necessary to the survival of the leaders’ chosen group (Semelin 2007, 13–17).
In fact, Semelin argues, the destruction of the enemy group serves not simply to
ensure the survival of the chosen group but also to purify, and hence strengthen,
that group. There exists, in other words, a zero-sum relationship between the
weakening of one group through destruction and the strengthening of another
through purification. A gain on one side is a loss on the other, and vice versa.
This rigid, binary, zero-sum ideological dynamic is at the heart of the genocidal
economy, and is quite in contrast to the bulk of academic economics with its
focus on specialized production and exchange via mutually beneficial trade, that
is, positive-sum interactions.1
Despite its metaphysical groundings, the genocidal economy is evidenced in
material policies, strategies, speech acts, and atrocities that suggest its existence.
Genocidal economies will be more or less extreme, based on the calculation of
the required level of loss within the target group for the perpetrator group to gain
fundamentally and in the long run. In what Adam Jones (2010, 187), borrowing
from Leo Kuper, refers to as “root-and-branch genocide,” the zero-sum logic of the
genocidal economy is total. The targeted group must lose everything, every trace
of existence, in order for the perpetrator group to advance. This chapter ties
together facets of the genocidal economy by looking at policies and strategies
across various cases of genocide and cases of genocidal violence. Certain bodies
and certain spaces are prime sites for the expression of the logic of the genocidal
economy and call attention to deeper connections among policies, strategies, and
acts during genocidal processes that are not well understood.
17.2.1. Raphael Lemkin, Gender, and the Genocidal

Economy
The gendered nature of the genocidal economy is very clear in the works of Raphael
Lemkin, the Polish Jewish jurist who coined the term “genocide” and pushed for
the crime’s recognition in international law. In his thinking, the gendered nature
of genocidal destruction appears most directly in what he called the “biological”
380 Case Studies I
aspects of genocidal strategies, that is, those techniques that undermine the abil-
ity of groups to reproduce themselves physically. Although he did not address
gender directly, Lemkin specifically included among these techniques measures
intended to decrease the birth rate (such as marriage controls and bans and the
criminalization of biological reproduction), as well as the separation of the sexes
and “the undernourishment of the parents” (Lemkin 1944, 28). Lemkin’s defini-
tion of the biological realm recognized the importance of the zero-sum dynamic.
Genocidal practices aimed at the biological realm consisted of two interrelated
parts: first, policies aimed at compromising the reproductive capacity of the tar-
get group; and second, policies aimed at strengthening the reproductive capacity
of the perpetrator group. One can compare this with his identification of the two
phases of genocide, as laid out in Axis Rule in Occupied Europe: “One, destruction
of the national pattern of the oppressed group; the other, the imposition of the
national pattern of the oppressor” (Lemkin 1944, 79). Throughout Lemkin’s writ-
ings, he recognized the zero-sum economy inherent in all of the eight genocidal
techniques he identified—political, social, cultural, economic, biological, physi-
cal, religious, and moral—a lthough he did not analyze them or their interrela-
tionships in light of this economy.
Many of the quotations from Nazi officials that Lemkin highlights emphasize
the zero-sum and metaphysical reproductive logic of the genocidal economy. He
cites an article written by Nazi Minister of Labor Robert Ley, which was pub-
lished in early 1940 in the party organ Der Angriff. Ley wrote that “[a]‌lower race
needs less room, less clothing, less food … than a higher race. The German
cannot live in the same fashion as the Pole and the Jew … More bread, more
clothes, more living room … these our race must have or it dies” (Jewish Black
Book Committee 1946, 203). His statement neatly sums up the Nazi approach to
the economy, and the genocidal approach to economics more generally. Just as
perpetrators construct identities that they believe are in zero-sum conflict, they
construct models of sustainability that are also zero-sum: the preferred group can
only be sustained so long as the sustainability of victim groups is being under-
mined (Deng n.d.). Important here is that the constant undermining of the target
group is necessary to perpetrator sustainability. Plunder is not time-l imited; it is
an integral part of the economy and must be constantly recapitulated. The victim
group needs less while the perpetrator group needs more. Without more, which the
perpetrator group gains at the expense of victim groups, the perpetrator group
will die. In other words, the continued existence of the perpetrator group is con-
tingent upon the continual impoverishment of target groups. The struggle is an
existential one.2
The specific resources mentioned by Ley are food, clothing, and Lebensraum
(“living room”)— a ll resources necessary to the sustenance of basic life.
Embedded within this construct is a particular approach to sustainability that
is dependent on the continual plunder of victim groups. However, this logic
of sustainability extends well beyond the perpetrators’ treatment of the mate-

rial resources—land, food, and clothing—noted by the Nazi minister of labor.
Genocidal ideologies also betray obsessions with symbols of the existential
strength of groups; symbols, then, are also limited resources to be exploited
by genocidaires for their own benefit. Indeed, genocidal processes show us
that perpetrators understand “sustainability” quite broadly to include all
sorts of symbolic and invisible (immaterial) resources, such as security, gen-
erative (reproductive) power, historical agency, and human dignity. In each of
these spheres, perpetrators feel they have more (or even enough) only to the
degree that their targets have less.
17.2.2. The Genocidal Economy of Atrocity

A prominent way in which perpetrators attempt to destroy their targeted groups
is through “life force atrocities” (von Joeden-Forgey 2010). Life force atrocities
are rituals of cruelty aimed at destroying not only the physical life of members of
the targeted group but also the source of that life. The means that perpetrators use
to effect this destruction include the desecration of symbols of the group’s past,
present, and future power; the undermining or inversion of structures of order
in the target group; and the rupture of community and familial bonds that make
group life possible. Life force atrocities tend to fall into one of two patterns: inver-
sion rituals, where proper hierarchies and social mores are intentionally inverted
in order to destroy the bonds between individuals; and ritual desecrations, where
important symbols are intentionally debased and defiled. Life force atrocities are
gendered atrocities in that they instrumentalize peoples’ positions within the
generative hierarchy in order to inflict maximum damage to the group. Common
examples of life force atrocities include coerced sexual acts between family mem-
bers, public rapes, rape and torture of family members in front of each other, the
evisceration of pregnant women, the mutilation of male and female genitalia and
women’s breasts, the violent separation of family members, targeted humilia-
tions of family and community leaders, and all forms of torture and desecration
aimed at religious and cultural symbols as well as at infants and young children.
Widespread sexualized violence, which characterizes all genocides, is itself a life
force atrocity of enormous proportions.
Life force atrocities usually take place when perpetrators find people in fam-
ily or small community groups. These groups are often used as stand-ins for the
larger group being destroyed. These smaller groups can be seen as the reproductive
realm of the group and they comprise the sociological space of the biological realm
in genocide. Like biological reproduction itself, assaults on the reproductive
realm during genocide are highly gendered. The zero-sum economy of genocide
operates on the reproductive realm in several ways, with disastrous consequences
for the victims. Perpetrators seek to enrich their own reproductive realm and
382 Case Studies I
h istorical power through the destruction of the victim group’s reproductive insti-
tutions, including its symbols, values, and other cultural resources. Perpetrators
view group reproduction in this sense as both biological and cultural in nature,
and they target each space by subjecting both visible and invisible symbols to ritu-
alized forms of abuse. Anything that is seen to be a source of reproductive and
historical power, such as family bonds and social relationships, becomes a target.
Men and women, boys and girls, are targeted in very specific ways based on their
social roles in order to exert maximum damage to the community. The aim of
inflicting this damage is not only to destroy the target group, but also to enrich
the perpetrators, materially, existentially, and cosmically.
In classic genocidal processes, like those we have seen during the Holocaust
as well as during the genocides against Armenians in the Ottoman Empire and
Rwandan Tutsi, the enrichment/impoverishment dynamic of the genocidal
economy is very direct. The perpetrator’s group is “enriched” in various ways
through the mass murder of the target population. Enrichment goes far beyond
the gains derived from plunder; indeed, it involves the zero-sum relationship that
is believed to exist between the generative strength of each group. This is why,
within the ideological systems behind these “total genocides,” the impoverish-
ment of the historical power of the victim group requires their (near-total) anni-
hilation. Any cohesive continuation of the group, no matter how small, would still
be a potentially lethal drain on perpetrator vitality. Discussing Nazi ideology,
David Patterson sums up the metaphysical nature of totalizing genocidal think-
ing in the following way:
From the standpoint of the Nazi Weltanschauung . . . “race” is not about

color or physiognomy or anything else that meets the eye. No, it is a
metaphysical category: race means soul . . . In Nazi thinking, race is tied
to the nature of thought, which, in turn, determines the substance of
one’s being. Inasmuch as Nazi thought is expressed in Nazi teaching,
it cannot tolerate the thought expressed in Jewish teaching. And since
Jewish teaching lies in Jewish blood, every Jew must be eliminated. For
the slightest trace of Jewish thinking—of Judaism—can infect both the
Aryan Geist and the Aryan blood. There lies the existential threat of the
Jew. (Patterson 2013, 164)
Alon Confino (2014) recently argued, along similar lines, that the Nazi destruc-
tion of Jews was about destroying all vestiges of the religious and moral order
that over time had given birth to the modern world, and which the Nazis saw
as posing an existential threat to the purity of the Aryan race. Because of these
existential framings, total annihilation enriches the perpetrator body politic by
removing groups that constitute a cosmic drain, thereby supposedly forging an
unobstructed path for the birth of a new utopia (Weitz 2003).
Gendered life force atrocities are one of the most effective means perpetra-
tors have devised to impoverish victim groups, enrich themselves, and thereby
attempt to deal a decisive blow to the target group’s life force. These atrocities
use coded terror to perform the zero-sum narrative of the genocidal economy
in potently symbolic ways. For this reason, the type of violence that has so often
been called “gratuitous and excessive” is anything but. It is in fact the perfect
expression of the genocidal logic, so much so that each total genocide of the
twentieth century can been seen as a life force atrocity writ large. Sometimes
perpetrators announce this ultimate goal quite directly, as when they invoke the
God of their victims. “We are fighting against you and against your God,” the
Nazis are reported to have said in Mlawa, Poland, while defiling sacred objects
in the synagogue. “Death to all of you! Let your God show whether he can
help you!” (Apenszlak 1943, 227). The perpetrators’ fight against the invisible,
ultimate source of a victim group’s life is one reason that religious leaders and
religious institutions are treated with particular cruelty and contempt during
genocide, even when the genocide is being committed against groups that are not
defined by religious criteria or religious criteria alone.
17.3. The Impoverishment/Enrichment Dynamic

Lemkin placed particular importance not simply on the material aspects of eco-
nomic destruction, such as the impoverishment and death of group members, but
also on the effects of a genocidal economy on the reproductive and cultural life of
groups. “The lowering of the standard of living,” he argued, “creates difficulties
in fulfilling cultural-spiritual requirements. Furthermore, a daily fight literally
for bread and for physical survival may handicap thinking in both general and
national terms” (Lemkin 1944, 85). He further notes that “[t]‌he undernourish-
ment of the parents, because of discrimination in rationing, brings about not only
a lowering of the birthrate, but a lowering of the survival capacity of children born
of the underfed parents” (Lemkin 1944, 86). His attentiveness to these seemingly
smaller details of genocidal patterns is reflected in his insight—only recently
taken up by genocide scholars—t hat “[t]he social structure of a nation being vital
to its national development, the occupant also endeavors to bring about such
changes as may weaken the national spiritual resources” (Lemkin 1944, 83).
At various points in Axis Rule, Lemkin notes the zero-sum logic that under-
girded Nazi economic decision-making. He speaks about the impoverishment/
enrichment dynamic that existed all over Nazi-occupied Europe. For example,
in discussing Nazi economic policies toward non-Jewish Poles, he writes: “[T]‌he
purpose of the occupant was to shift the economic resources from the Polish
national groups to the German national group. Thus the Polish national group
384 Case Studies I
had to be impoverished and the German enriched” (Lemkin 1944, 85). Although
Lemkin never draws out this insight analytically, it clearly informed his under-
standing of many of the Nazis’ crimes. Writing specifically on biological tech-
niques of genocide, Lemkin notes that “[f]oremost among the methods employed
for this purpose [depopulation of victim groups] is the adoption of measures cal-
culated to decrease the birthrate of the national groups of non-related [i.e., non-
“Aryan”] blood, while at the same time steps are taken to encourage the birthrate
of the Volksdeutsche living in these countries” (Lemkin 1944, 86). Because any
Nazi policies aimed at increasing the birth rate of preferred groups was tied to the
biological destruction of other groups, Lemkin viewed discriminatory, pronatal-
ist policies of occupying forces as an element of the crime of genocide (Lemkin
1944, 92).
Lemkin’s understanding that perpetrators seek, in all ways possible, to
strengthen their group at the expense of their target groups allowed him to iden-
tify subtleties in the crime of genocide that traditionally had been overlooked or,
when recognized, were understood as discrete abuses. Lemkin recognized that
the Nazis were waging a total war where, as the Nazis often described it them-
selves, “the nation, not the state, is the predominant factor. … In this German
conception, the nation provides the biological element for the state” (Lemkin
1944, 80). He summed up the resulting reasoning of the Nazi leadership in the
following way:
The enemy nation within the control of Germany must be destroyed, dis-
integrated, or weakened in different degrees for decades to come. Thus
the German people in the post-war period will be in a position to deal
with other European peoples from the vantage point of biological superi-
ority. Because the imposition of this policy of genocide is more destruc-
tive for a people than injuries suffered in actual fighting, the German
people will be stronger than the subjugated peoples after the war even if
the German army is defeated. In this respect genocide is a new technique
of occupation aimed at winning the peace even though the war itself is
lost. (Lemkin 1944, 81)
Lemkin here proposes a very interesting view of the war aims of genocid-
aires. Wars can be lost so long as they are successful in sufficiently debilitat-
ing the victim groups such that they will no longer pose an existential threat
in the future. In other words, genocide is a crime that is about historical and
metaphysical power.
Because the perpetrators of genocide believe their strength to be tied to the
imposition of physical, spiritual, and cosmic weakness on their target groups,
they gain through these rituals an intense and immediate sense of their own
existential prowess and generative vitality. Consider, for example, the words of
John Girumuhatse, a perpetrator of the 1994 genocide against the Rwandan
Tutsi. He told the American journalist Philipp Gourevitch that killing during the
genocide was a pleasure that made him feel stronger:
For me, it became a pleasure to kill. The first time, it’s to please the gov-
ernment. After that, I developed a taste for it. I hunted and caught and
killed with real enthusiasm. It was work, but work that I enjoyed. It
wasn’t like working for the government. It was like doing your own true
job—like working for myself … I was very, very excited when I killed.
I remember each killing. Yes, I woke every morning excited to go into the
bush. It was the hunt—t he human hunt. … The genocide was like a fes-
tival. At day’s end, or any time there was an occasion, we took a cow from
the Tutsis, and slaughtered it and grilled it and drank beer. There were
no limits anymore. It was a festival. We celebrated. (Gourevitch 2009)
Killing and the desecration of sacred spaces, social norms, affinities, and loyal-
ties are, in this sense, weapons used to transfer generative resources and histori-
cal strength from the victim group to the perpetrators. The genocidal economy
can help us understand the common presence of cruel “jokes” during genocide.
During the Nazi occupation of Warsaw, for example, a witness reported that
“[t]‌he home of a Jewish woman lawyer was entered one evening while she was
entertaining some women friends and a young couple. All the women were forced
to strip and dance on the table, and the young couple to cohabit in the presence of
all the others” (Apenszlak 1943, 29). Here the desecration of sex and the body, the
coerced and violent pornography directed at Jews, gave the perpetrators a sense
of their own vitality. Often such scenes are repeated at massacre sites and other
spaces of mass death, and they are usually accompanied by perpetrator laughter.
Sexualized violence is so prevalent in genocides in part because it offers a
means of accessing—and attempting to destroy—t he institutions, symbols, and
physical means of group reproduction. Even during the Holocaust, a case where
sexual relations between perpetrators and members of the victimized Jewish
community were criminalized, Jewish women were often subjected to sexual-
ized violence by German policemen, SS-men, and soldiers at every stage of the
destruction (Goldenberg 2010; Hedgepeth and Saidel 2010). A common torture
used against Jewish women was killing children before raping and killing their
mothers. Babies and small children would be killed in front of their mothers in
deliberately cruel ways, by being impaled on bayonets, used for target practice,
or smashed against walls and trees. One witness who testified at Nuremberg said
they saw the SS take “a baby from its mother’s breast and [kill] it before her eyes
by smashing it against the barracks wall,” and another “saw babies taken from
their mothers and killed before their eyes: they would take a baby by one foot
386 Case Studies I
and step on the other, and so tear the baby apart” (Jewish Black Book Committee
1946, 384). Such scenes are repeated throughout genocides.
Apart from destroying the immediate physical symbols of community cohe-
sion and generative power, genocidaires also target any signs of love and affec-
tion, however indirect, in their efforts to strengthen themselves at the expense
of victim groups. The Khmer Rouge, for example, abolished the family and out-
lawed all demonstrations of filial love, on pain of death (Kiernan 2006, 195). The
Nazis, as we know, criminalized pregnancy and childbirth in the ghettos, concen-
tration camps, and death camps (Amesberger 2010; Ben-Sefer 2010). Both the
Khmer Rouge and the Nazis sought to control all marriages and claim control
over children born even to their own groups. Even small indicators of love could
be grounds for atrocity. One survivor of the Buchenwald concentration camp, for
example, told of a middle-aged Jewish man who had received a letter from his
wife. According to the prisoner, “[h]‌e was an educated man of a sensitive type
and unable always to maintain the degree of self-control and outward hardness
which is demanded.” So, the man “moaned as he read” the letter. For this, he was
punished by being tied to a tree and left there for fourteen hours. After he had lost
consciousness, two guards untied him, jumped on his body, and stamped on him
with their boots (Jewish Black Book Committee 1946, 265).
17.4. Resource-R ich and Resource-Poor Genocides

17.4.1. Resource-R ich Genocides
Thus far we have treated intangibles like generative power, dignity, and love as
resources that genocidaires attempt to appropriate from their victims through the
use of extremely cruel atrocities that instrumentalize these resources in order to
drain them from the victim group. These depredations are common features of all
genocides, in part because they send such strong messages to victim communi-
ties, the outside world, and perpetrators themselves about what the perpetrators
are doing. But they are also common because they perform the appropriation of
the one thing that is most valued in any genocidal process: generative power, and
especially generative power gained through draining the victim community of its
invisible vitality. Such an interpretation can help us understand why perpetrators
often feel so strengthened by killing.
Perpetrators of genocide also seek to appropriate the physical resources associ-
ated with generative power—especially through slave labor and sexual slavery.
One way to look at classic cases of genocide is as “resource-r ich” genocides, in that
the perpetrators, having control over states and armed forces, and representing
majority populations, did not view themselves to be in need of population replen-
ishment from their victim groups, nor did they view their home societies to be
dependent upon their victim groups for basic needs. For the most part, they saw
themselves as biologically independent of their victims. Of course, this did not

mean that resources extracted from targeted populations were not used. Quite
to the contrary: the economy of Nazi-occupied Europe was based on plunder
and slave labor (Aly 2006). The expropriation of Armenian land and businesses
was a windfall for the Young Turks (Üngör and Polatel 2011). In Rwanda, Hutu
perpetrators freely availed themselves of the property and land of Tutsi neigh-
bors whom they killed (Hogg 2010; Burnet 2012). The existential independence
of the perpetrator group from its victims simply means that, from an existential
viewpoint, targeted groups could theoretically be completely annihilated without
harm to the perpetrator’s group; in fact, ideologically speaking, targeted groups
in these cases had to be annihilated in order for the perpetrator group to reach its
full potential.
One aspect of Nazi expansion that did betray a sense of biological vulnerability
and interdependence with respect to some of its targets was the effort to divide
conquered peoples into those who could be Aryanized and those who could not.
However, the Nazis believed that their principal targets, the Jews, constituted such
a totalizing threat that they had to be eliminated in toto. In the Nazi pantheon of
races, the children of certain groups targeted for partial destruction, such as Slavs,
were sometimes kidnapped because they had supposedly Aryan features and
could strengthen the Aryan race (Lower 2007). Those groups, such as Northern
Europeans and others whose blood could be appropriated, would become part
of the Thousand Year Reich, replenishing Germany with proper biological stock,
with males to be used as soldiers and females as birth-g ivers. These policies were
all determined by the particular brand of Nazi racism, which taught that good
biological stock anywhere would strengthen the bad biological stock and hence
pose a future threat to the Third Reich. Heinrich Himmler framed the relational
dynamic literally in terms of blood resources: “Either we win over good blood
that we can use for ourselves and give it a place in our people or—gentlemen, you
may call this cruel, but nature is cruel—we destroy this blood. But we cannot
answer for it to our sons and ancestors if we leave this blood on the other side, thus
letting our enemies gain able leaders and able commanders” (Jewish Black Book
Committee 1946, 246).
In a similar vein, but for different ideological reasons, it was commonplace
for perpetrators to treat certain Armenian girls and young women as exploitable
resources to be taken into their households as sex slaves and domestic laborers
(Miller and Miller 1999). As women and girls were considered to be weaker carri-
ers of ethnic and religious identity, they were sometimes given the option of forced
conversion and enslavement (Derderian 2005; Bjørnlund 2009; Ekmekcioglu
2013). A common practice during the destructive process was to kill and torture
men and boys, rape and kill women, and distribute the surviving young girls to
Turkish and Kurdish families. A different sort of biological appropriation during
a genocidal process are the policies pursued by A. O. Neville during the interwar
388 Case Studies I
period in Australia. Neville, who was Chief Protector of the Aborigines, believed
in a policy of “absorption,” where “half-caste” children would be kidnapped
from their families, specially trained in residential schools, incorporated into
white society as domestics and menial laborers, and their “colour” slowly “bred
out” through the generations. The rest of Aboriginal society, the “full blooded”
Aborigines, were expected to die out according to the natural law of history, lead-
ing to a complete erasure of Aborigines from the earth and the creation of a fully
white Australia (Moses 2004). It is estimated that 20,000 to 25,000 children were
kidnapped from their families between 1910 and 1970 (Manne 2010).
17.4.2. Resource-Poor Genocides

Despite the appropriation of victims as slave laborers, concubines, and proper
material for forced marriage and breeding schemes, none of the perpetrating
states in the resource-rich genocides was reliant on the victim populations to
replenish resources and ensure the reproduction of the perpetrator group. The
situation is different with “resource-poor” genocides, where armed forces do not
have access to the considerable economic power of a state and often have neither a
stable territorial base nor a clear population group that they can count on to repro-
duce themselves. Examples of such genocidal processes are the wars led by the
Lord’s Resistance Army (LRA) in Central and East Africa and the Revolutionary
United Front (RUF) in Sierra Leone. In these cases, the perpetrator group does
not represent a preexisting historical configuration, and perpetrator identity
groups are synonymous with the fighting forces. Because of this, they are reliant
upon the surrounding civilian communities for their own reproduction. As in the
total genocides, the perpetrators in these cases rely on plunder to fuel their war
machines and sweeten the pot for their foot soldiers. However, unlike those cases,
they rely on biological resources from their victim groups in order to survive, and
they extract these resources from their target groups in ways that generally do not
discriminate among ethnic, national, or religious groups.
The common genocidal pattern in these resource-poor, nonstate cases is to
appropriate and transfer reproductive power from the civilian world to the fight-
ing cadres through kidnapping male and female children while laying waste to
the civilian communities from which the captives come. In these cases, ideolo-
gies pivot around the construction of a dichotomy between the fighting force on
the one hand and the civilian world on the other. Human Rights Watch described
the tactics of the Lord’s Resistance Army in the following way: “In towns, the
rebels loot trading posts and steal medicines from small health clinics. In the
bush, they loot compounds, beating and often killing the adults, and abduct-
ing many of the children. They burn huts when they leave, and steal everything
edible or useful. The small rebel bands then reunite, and march together back
across the Sudanese border” (HRW 1997, 14). Between 1986 and 2006, the LRA
kidnapped between 54,000 and 75,000 people, 25,000 to 38,000 of whom were
children. Although the LRA has historically drawn its members from the Acholi
communities in northern Uganda from which the fighting force originated, it
also abducts people irrespective of ethnic or national affiliation and incorporates
them into the LRA as porters, soldiers, sex slaves, and domestic servants. Women
and girls are in fact expected to become the “long-term sexual partners” of LRA
soldiers and to bear children for them. These “families” are sometimes kept in
encampments behind the war zones and can be seen as efforts to replenish the
fighting force in conditions not conducive to stable population growth (Pham,
Vinck, and Stover 2008).
The LRA emerged in northern Uganda after President Yoweri Museveni came
to power in 1986. Museveni, who was from the south, faced armed opposition in
the north, where many government soldiers fled after he took power. The Acholi
national group feared loss of political representation and faced direct persecu-
tion by Museveni’s forces. Since 1986 Museveni has presided over policies in the
north that have been coercive and violent; some Acholi characterize these policies
as genocidal, particularly with reference to the “protected villages” established
ostensibly to guard the Acholi from LRA depredations (Dolan 2009, 6). These
villages are frequently referred to as “concentration camps” and have rendered
Acholi communities powerless and vulnerable to both the LRA and government
troops, whose record of atrocity and mistreatment of civilians is similar to the
LRA’s, including recruitment of child soldiers, extrajudicial executions, rape, tor-
ture, beatings, forced displacement, and looting (Human Rights Watch 2003a;
Dolan 2009, 107–58). Due to the behavior of Ugandan government forces, the
LRA has found some reluctant support among the Acholi.
Scholarship on the LRA has emphasized the rational aspects of its organiza-
tion, strategy, and goals, pointing out that it emerged from legitimate political
and economic grievances and fears that were widespread in northern Uganda
in the 1980s (Dolan 2009). The studies have tended to focus on the legitimate
grievances of Acholi people, out of which LRA leader Joseph Kony arose, as well
as the functional aspects of the LRA’s terror tactics and forced recruitment, par-
ticularly in terms of the contribution these make to the larger, ill-defined, goals
of the LRA. The LRA’s goals include discrediting the Museveni regime, exercis-
ing some power within the government of Uganda, and, increasingly, mere self-
preservation (ICG 2004).
While terrorizing its own soldiers, as well as civilians, certainly serves direct
and pragmatic ends for insurgency groups that count on mobility as part of their
long-term strategy, there is much more behind LRA atrocity than its functional
aspects. Terrorizing civilians and abductees can also be viewed as the result of a
particular approach to the existential crisis brought about within Acholi commu-
nities by the victory of Museveni over the short-l ived reign of the Acholi general
Tito Okello. In other words, apart from serving pragmatic demands, the use of
390 Case Studies I
terror and atrocity by the LRA has an existential and metaphysical dimension that
places the LRA squarely in the realm of genocidal cosmologies.
The LRA is perhaps best known for what has been called its “bizarre” spiri-
tualism (Fisher 2011, headline). Joseph Kony, who inherited his role from Alice
Lakwena’s Holy Spirit Movement, leads the LRA based on the commands of a
rotating group of spirits who possess him several times a day and issue orders
that cannot be contravened. The messages sent by the spirits are directed toward
the creation of a utopian and pure “New Acholi” community out of the polluted
world outside the LRA’s control. Since the LRA is not only concerned with the
old Acholiland, and instead operates across borders in Central and East Africa,
the New Acholi community can draw inductees from any community overrun
by LRA forces. Inductees go through grueling rituals in order to become part of
the LRA community. Frequently these involve killing children and other LRA
members who have disobeyed the rules. Like other genocidal systems, the LRA
enforces strict rules about gender relations and claims full control over the repro-
ductive actions of its members. The sexes are rigidly separated; Kony and his
senior officials make all decisions about sexual relations and “marriages,” thereby
fully controlling the “family life” that the LRA seeks to create as a utopian alterna-
tive to family life within the civilian world (Titeca 2010). The LRA functions as a
system of continuous terror for its members, whose lives are viewed as disposable.
The genocidal economy of the LRA worldview expresses itself in the types
of atrocities soldiers are ordered to commit by spirits speaking through Joseph
Kony. The common pattern of attack on communities is to loot property, rape,
kill adults, abduct children and teenagers, and burn down whatever is left.
Occasionally, adults are used as short-term porters and then let go; children are
usually kept longer. When parents are killed, it is generally in front of the captive
children, and sometimes children are forced to kill their parents and other family
and community members (HRW 2010). The sorts of atrocities that foot soldiers
(often captive children) are forced to commit are similar in logic to those com-
mitted by soldiers during classic, root-and-branch genocides. These are often life
force atrocities that go beyond the simple and functional terrorizing of civilian
communities; as in other cases of genocidal violence, they betray a core genocidal
ideology that finds its logic within the genocidal economy. The LRA feeds off of
civilian communities both practically (foodstuffs, medicines, weapons, soldiers,
sex slaves) and metaphysically (dignity, generative and historical power). The
looted resources, including male and female children, are brought back to the
fighting force to replenish its biological stock and strengthen its generative power,
ensuring its future.
In an effort to counteract what has been called the “New Barbarism” approach
to African conflict in Western journals and newspapers, several scholars have
pointed out the rational and functional aspects of spiritualism in African wars
more generally. Kristof Titeca has specifically analyzed the LRA’s “spiritual order”
as part of a rational response to postcolonial crises in Uganda (Titeca 2010, 61). In

his view, while Kony and the LRA may be immensely destructive and cruel, they
are not any more barbaric or illogical than similar forces that operate through
discourses more easily decoded by Western observers. This is true, as the LRA
is not committing atrocities that have not been seen in genocidal conflicts in the
Western world. The logic on display is the logic of the genocidal economy, which
sets up resources—material and metaphysical—in a zero-sum dynamic between
perpetrators and their targets, however their targets are defined. Studies that
overemphasize the purely functional aspects of atrocity in order to familiarize it
can risk overlooking the important genocidal logic that ties many of the atroci-
ties together in a meaningful way, one connected to the genocidal worldview of
the fighting force, which is seeking to enrich itself through the destruction of the
target group.
The genocidal economy practiced by the LRA shares similarities with other
genocidal insurgencies, such as the RUF in Sierra Leone and, more recently, Boko
Haram in Nigeria. The RUF in particular also abducted children to serve as child
soldiers and sex slaves, and used terror as a means of ensuring obedience among
its recruits and encouraging self-identification with the RUF fighting forces. The
RUF pattern of attack was similar to the LRA’s: destroy the civilian world through
killing, torture, rape, mutilation, pillage, and burning while appropriating what
would sustain and grow its own force. There is a great deal of evidence that the
RUF’s terror campaign was aimed at destroying in whole the civilian worlds it
came upon. The RUF used not only life force atrocities but also specific poli-
cies to undermine the victim groups’ reproductive abilities. These include mass
rape of women of all ages, ethnicities, religions, and regions, and specialized tor-
tures such as “virgination,” where the RUF explicitly sought out virgins to rape.
Sometimes RUF nurses would be required to do a medical exam to find virgins
within an occupied population (HRW 2003b; Jones 2010, 92–130).
The point of all this violence was complex. The leader of the RUF, Foday
Sankoh, had trained in Libya with the Liberian Charles Taylor, who was to help
the RUF get off the ground and invade Sierra Leone from Liberia in the early years
of the war. Sankoh’s movement was initially formed to overthrow the corrupt
Sierra Leone government; it gained initial support from the many people who
had been left behind politically and economically. But the RUF quickly lost its
ideology, rooted in Libya’s then-leader Muammar al-Gaddafi’s Green Book—as
well as any popular support it enjoyed—as the original RUF leadership was killed
and it became an army of captives who were forced to commit horrific atrocities
on civilian communities without distinction (Richards 2006, 655). One of the
strategic goals of the rebel invasion was to gain control of Sierra Leone’s diamond
mines, which were to finance the war effort both in Sierra Leone and in Liberia.
Once this was accomplished, the RUF seems to have worked to preserve itself
and increase its own strength through abduction and the creation of a new social
392 Case Studies I
order based on allegiance to the RUF. As Myrian Denov has pointed out in her
study of child soldiers in the RUF, “[u]‌ltimately through the creation of this new
social order, there is a gradual movement from one world view to another—f rom
a civic world based on principles of humanity, civic associations, empathy and
caring, to a world of torture based on inhumanity, rigid hierarchies, detachment
and cruelty” (Denov 2010, 103). As in the LRA, children abducted into the RUF
were often forced to kill family and community members. This is, as the litera-
ture points out, a means of severing children’s ties with their old lives and old
worlds. It is also a symbolic expression of the zero-sum relationship the RUF had
with the civilian world. By destroying the family units within communities that
it overran in Sierra Leone through killing, torture, and abduction, it literally and
metaphysically enriched its own generative potential and historical force.
17.5. Conclusion
In our efforts to define genocide, it is important that we recognize the central
importance of the zero-sum genocidal economy in determining what perpetrators
do. Sometimes this economy is laid out in well-developed (albeit illogical and often
internally inconsistent) ideologies, as was the case with the National Socialist
regime in Germany. Sometimes the zero-sum economy can be gleaned from the
statements of perpetrators. And sometimes it can best be seen in the stark, on-the-
ground realities of committed atrocities. In all cases, the genocidal economy is
ultimately concerned with strengthening the preferred group at the expense of the
targeted group or groups. This strengthening is metaphysical because it is about
generative forces: Perpetrators feed off of target groups and seek to drain them in
order to ensure their own historic victory over them. Ideology and worldview will
dictate how much killing is required to effect an asymmetry that the perpetrators
consider to be sustainable over the long run, but both total and partial genocides
follow a similar script written by the genocidal economy that is driving the violence.
The implications of understanding genocide according to an invisible geno-
cidal economy are numerous. It helps give a deeper dimension to the twin pro-
cesses of mass killing and mass plunder, which so often go hand-in-hand (Semelin
2007, 165). While mass plunder may seem to be a practical consequence of mass
murder, they both share the same origin in the metaphysical logic of the genocidal
economy. Taking resources and killing bodies both work to enrich the generative
power of the perpetrators. In a similar vein, the genocidal economy demonstrates
how biological and cultural destruction are different sides of the same coin: Both
target generative powers within the out-g roup in order to enhance generative
power of the in-g roup. As Raphael Lemkin often argued, they cannot easily be
distinguished. In fact, they often overlap, as is the case with religious leaders and
other important symbolic and institutional figures. The genocidal economy calls
attention to the fact that sexualized and reproductive violence during genocide
is part and parcel of the genocidal process rather than the result of other factors.
In genocide, rape follows a logic of perpetrator enrichment, whether or not each
individual rapist fully comprehends the deeper forces at work. Along these lines,
the genocidal economy also complicates our thinking about intent and how it is
shared and refracted through military and political hierarchies as well as across
thousands of individual perpetrators. Once the genocidal economy is structuring
relations between perpetrators and victims, it becomes a force in and of itself.
It is interesting to think about what this means for genocide prevention.
Clearly one goal of prevention will have to be figuring out how to stay the develop-
ment of genocidal economies in peacetime contexts. In the long term, we need to
identify those social processes that have the greatest likelihood of creating stark
existential crises that could lead to deep-seated anxieties about generative power.
This is especially true within groups that have enjoyed a certain degree of histori-
cal privilege, which may experience perceived assaults on their generative power
more acutely and be more prone to react with violent, zero-sum imageries. In the
short term, it is valuable to think about how we might identify the existence of a
genocidal economy in speech acts, propaganda, and early forms of symbolic and
material violence so that we can address looming genocidal processes more effec-
tively (see chapter 12 in this volume). In the final analysis, a better understanding
of the genocidal economy will rely on our ability to link it to other economies, the
moral economies of peacetime, and the ways they exacerbate, or serve to mediate,
yawning divides in the life chances of individuals and groups.
Acknowledgments
I thank the editors of this volume and the anonymous reviewers for their helpful
comments on a first draft of this chapter.
Notes
1. Human interaction can be negative-sum, zero-sum, or positive-sum. As indicated in the
main text, economists ordinarily concern themselves with positive-sum games. They also
have extensively dealt with negative-sum games, especially in the context of war, and ref-
erences to this literature are found throughout this volume. In contrast, the “genocidal
economy” is one that views all interaction between an in-g roup and out-g roup as zero-
sum: What one group gains, the other loses. The zero-sum economy described in this chap-
ter is not about Schumpeterian “creative destruction,” which ultimately is positive-sum,
but about—to coin a phrase—appropriative destruction: the genocidal perpetrator appro-
priates the very generative power of the victim group and destroys it in the process. In this
regard, readers may find Kenneth Boulding’s The Economy of Love and Fear (1973) of inter-
est, in which he describes three economies: the economy of love (a one-sided beneficial
394 Case Studies I
giving without expectation of return), the economy of exchange (voluntary and mutually
beneficial), and the economy of fear (one-sided and coerced, an appropriation). I am grate-
ful to Jurgen Brauer for the phrase “appropriative destruction” as a contrast to “creative
destruction.”
2. Several chapters in this volume address the economics of individual and group identity in
theoretical, empirical, and otherwise applied ways (see, e.g., chapters 13, 14, 21, and 22).
Another valuable source is Boulding’s The Image (1956).
References
Aly, G. 2006. Hitler’s Beneficiaries: Plunder, Racial War, and the Nazi Welfare State. New York:
Metropolitan.
Amesberger, H. 2010. “Reproduction under the Swastika: The Other Side of the Nazi Glorification
of Motherhood.” In S. M. Hedgepeth and R. G. Saidel, eds., Sexual Violence against Jewish
Women during the Holocaust. Waltham, MA: Brandeis University Press, 139–55.
Apenszlak, J., ed. 1943. The Black Book of Polish Jewry: An Account of the Martyrdom of Polish Jewry
under the Nazi Occupation. New York: American Federation of Polish Jews.
Ben-Sefer, E. 2010. “Forced Sterilization and Abortion as Sexual Abuse.” In S. M. Hedgepeth
and R. G. Saidel, eds., Sexual Violence against Jewish Women during the Holocaust. Waltham,
MA: Brandeis University Press, 156–73.
Bjørnlund, M. 2009. “‘A Fate Worse Than Dying’: Sexual Violence during the Armenian
Genocide.” In D. Herzog, ed., Brutality and Desire: War and Sexuality in Europe’s Twentieth
Century. New York: Palgrave Macmillan, 16–58.
Boulding, K. E. 1956. The Image. Ann Arbor: University of Michigan Press.
Boulding, K. E. 1973. The Economy of Love and Fear. Belmont, CA: Wadsworth.
Burnet, J. 2012. Genocide Lives in Us: Women, Memory and Silence in Rwanda. Madison: University
of Wisconsin Press.
Confino, A. 2014. A World without Jews: The Nazi Imagination from Persecution to Genocide. New
Haven, CT: Yale University Press.
Deng, F. n.d. “In the Shadow of the Holocaust.” The Holocaust and the United Nations Outreach
Program. Discussion Paper Series 1(7). http://w ww.un.org/en/holocaustremembrance/
docs/paper7.shtml [accessed May 9, 2015].
Denov, M. 2010. Child Soldiers: Sierra Leone’s Revolutionary United Front. Cambridge, UK:
Derderian, K. 2005. “Common Fate, Different Experience: Gender-Specific Aspects of the
Armenian Genocide, 1915–1917.” Holocaust and Genocide Studies 19, no. 1: 1–25.
Dolan, C. 2009. Social Torture: The Case of Northern Uganda, 1986–2006. New York: Berghahn.
Ekmekcioglu, L. 2013. “A Climate for Abduction, a Climate for Redemption: The Politics of
Inclusion during and after the Armenian Genocide.” Comparative Studies in Society and
History 55, no. 3: 522–53.
Fisher, M. 2011. “The Bizarre and Horrifying Story of the Lord’s Resistance Army.” Atlantic
Monthly, October 17, 2011. http://w ww.theatlantic.com/i nternational/a rchive/2 011/10/
the-bizarre-a nd-horrifying-story-of-t he-lords-resistance-a rmy/2 46836/ [accessed May
9, 2015].
Goldenberg, M. 2010. “Sex-Based Violence and the Politics and Ethics of Survival.” In
M. Goldenberg and A. Shapiro, eds., Different Horrors, Same Hell: Gender and the Holocaust.
Seattle: University of Washington Press, 99–131.
Gourevitch, P. 2009. “The Life After: Fifteen Years after the Genocide in Rwanda, the
Reconciliation Defies Expectations.” The New Yorker, May 4, 2009. http://w ww.newyorker.
com/magazine/2 009/05/0 4/t he-l ife-a fter [accessed May 9, 2015].
Hedgepeth, S. M., and R. G. Saidel, eds. 2010. Sexual Violence against Jewish Women during the
Holocaust. Waltham, MA: Brandeis University Press.
Hogg, N. 2010. “Women’s Participation in the Rwandan Genocide: Mothers or Monsters?”
International Review of the Red Cross 92, no. 877: 69–102.
[HRW] Human Rights Watch. 1997. “The Scars of Death: Children Abducted by the Lord’s
Resistance Army in Uganda.” New York: Human Rights Watch.
[HRW] Human Rights Watch. 2003a. “Abducted and Abused.” New York: Human Rights Watch.
[HRW] Human Rights Watch. 2003b. “We’ll Kill You If You Cry: Sexual Violence in the Sierra
Leone Conflict.” New York: Human Rights Watch.
[HRW] Human Rights Watch. 2010. “Trail of Death.” New York: Human Rights Watch.
[ICG] International Crisis Group. 2004. “Northern Uganda: Understanding and Solving the
Conflict.” Africa Report 77. Brussels: International Crisis Group.
Jewish Black Book Committee. 1946. The Black Book: The Nazi Crime against the Jewish People.
New York: American Book/Stratford Press.
Jones, A. 2000. “Gendercide and Genocide.” Journal of Genocide Research 2, no. 2: 185–211.
Jones, A. 2010. War Is Not Over When It’s Over. New York: Henry Holt.
Kiernan, B. 2006. “External and Indigenous Sources of Khmer Ideology.” In O. A. Westad and
S. Quinn-Judge, eds., The Third Indochina War: Conflict between China, Vietnam and Cambodia.
London: Routledge, 187–2 06.
Lemkin, R. 1944. Axis Rule in Occupied Europe. Washington, DC: Carnegie Endowment for
International Peace.
Lower, W. 2007. Nazi Empire Building and the Holocaust in Ukraine. Chapel Hill: University of
North Carolina Press.
Manne, R. 2010. “Comment: Keith Windschuttle.” The Monthly, February 2010. https://w ww.
themonthly.com.au/nation-reviewed-robert-manne- comment-keith-w indschuttle-2 256
Miller, D. E., and L. T. Miller. 1999. Survivors: An Oral History of the Armenian Genocide. Berkeley:
University of California Press.
Moses, A. D., ed. 2004. Genocide and Settler Society: Frontier Violence and Stolen Indigenous Children
in Australian History. New York: Berghahn.
Patterson, D. 2013. “The Nazi Assault on the Jewish Soul through the Murder of the Jewish
Mother.” In M. Goldenberg and A. Shapiro, eds., Different Horrors, Same Hell: Gender and the
Holocaust. Seattle: University of Washington Press, 163–75.
Pham, P. N., P. Vinck, and E. Stover. 2008. “The Lord’s Resistance Army and Forced Conscription
in Northern Uganda.” Human Rights Quarterly 30, no. 2: 404–11.
Richards, P. 2006. “An Accidental Sect: How War Made Belief in Sierra Leone.” Review of African
Semelin, J. 2007. Purify and Destroy: The Political Uses of Massacre and Genocide. New York: Columbia
University Press.
Titeca, K. 2010. “The Spiritual Order of the LRA.” In T. Allen and K. Vlassenroot, eds., The Lord’s
Resistance Army: Myth and Reality. London: Zed, 59–73.
Üngör, U. U., and M. Polatel. 2011. Confiscation and Destruction: The Young Turk Seizure of Armenian
Property. London: Continuum International.
von Joeden-Forgey, E. 2010. “The Devil in the Details: ‘Life Force Atrocities’ and the Assault on
the Family in Times of Conflict.” Genocide Studies and Prevention: An International Journal 5,
no. 1: 1–19.
von Joeden-Forgey, E. 2011. “Genocidal Masculinity.” In A. Jones, ed., New Directions in Genocide
Research. New York: Routledge, 76–94.
Weitz, E. 2003. A Century of Genocide: Utopias of Race and Nation. Princeton, NJ: Princeton
University Press.
PA R T F O U R
CASE STUDIES II
18
On the Logistics of Violence

Evidence from Stalin’s Great Terror, Nazi-O ccupied
Belarus, and Modern African Civil Wars
Y u r i M . Z h u kov
18.1. Introduction
Logistics make organized violence possible. One cannot kill without the means to
reach a target. Without transport and open lines of communication, combatants
cannot easily deploy their forces, reload their weapons, refuel their vehicles, repair
their equipment, feed their troops, evacuate their wounded, or send detainees to
camps. The same logistical constraints that apply to warfare extend to violence
against civilians—whether intentional or a result of collateral damage. As this
chapter illustrates through micro-level data from historical and contemporary
cases, disruptions to military logistics force combatants to slow their tempo and
divert resources away from fighting. As logistical challenges mount, a combatant
loses the capacity to repress, kill, and destroy on a massive scale.
Research on genocide and violence against civilians regularly cites the impor-
tance of logistics, but only infrequently studies them directly. In contrast to the
dedicated treatment the topic has received in military theory (Jomini 1862),
doctrine (Joint Chiefs of Staff 2013), history (Guerlac 1986; van Creveld 2004),
policy analysis (Eccles 1991; Foxton 1994; Owen and Mueller 2007), and opera-
tions research (Kress 2000, 2002; Baker et al. 2002), social scientists have treated
logistics mostly as an ancillary factor in civil conflict rather than as a phenomenon
of primary theoretical interest.
Recent empirical conflict research has examined the impact of road density
on violence (Murshed and Gates 2005; Bellows and Miguel 2006; Buhaug and
Rod 2006; Raleigh and Hegre 2009), the use of roadside bombs by insurgents
(McFate 2005; Townsley, Johnson, and Ratcliffe 2008), the diffusion of violence
through transportation networks (Zhukov 2012), and government efforts to iso-
late centers of rebel activity (Toft and Zhukov 2012). With few exceptions, this
399
400 Case Studies II
literature has avoided probing the conditions under which logistics might affect
violence against civilians and how these logistical considerations relate to the
pursuit of local popular support.
If logistical constraints indeed make violence against civilians more costly,
research on this topic should be of great value in predicting the location, timing,
and scale of atrocities. Recent evidence suggests that these constraints apply to
great powers as well as to local militias, and that the importance of logistics has
not declined over time. Zhukov (2015a) shows, with archival data, that attacks
against railroad networks made German forces kill fewer civilians in World War
II. Using evidence from a more recent case, Rogall (2014) employs an interaction
of transportation infrastructure and rainfall as an instrumental variable for mobi-
lization during the Rwandan genocide. Such studies offer a dark contrast to recent
work on the positive economic effects of low transport costs (Donaldson 2010;
Banerjee, Duflo, and Qian 2012). The same roads that lead to prosperity, this new
research suggests, can also lead to a much darker place.
This chapter offers a more direct look at the logistics of political violence, with
an emphasis on violence against civilians during intrastate conflict.1 In section
18.2, I distinguish between two types of supply systems—a reliance on local
resources obtained from within a conflict zone, and external resources shipped
in from outside—and the relative prevalence of these systems in government
and rebel armed forces. In section 18.3, I consider the implications of external
resources for the quality and quantity of violence, particularly the use of force
against civilians. I argue that more extreme forms of violence call for a greater
reliance on external resources, but also create vulnerabilities that opponents can
exploit. In section 18.4, I examine how disruptions to logistics affect the behav-
ior of armed groups. Section 18.5 considers the empirical basis for these claims.
Using data on contemporary and historical conflicts in Africa (as well as a hand-
ful of countries elsewhere), I show that where logistical costs are high—due to
infrastructure or sabotage—we see less violence overall, and less violence against
civilians in particular.
18.2. The Art of the Logistically Feasible

In his Precis de l’Art de la Guerre, the French general and military theorist Antoine-
Henri Jomini defined logistics as the “practical art of moving armies” (Jomini
1862, 15). If strategy “decides where to act,” then logistics “bring the troops to
this point” (Jomini 1862, 51). Core tasks include moving and sustaining forces;
managing their inventory of fuel, food, medicine, clothing, construction materi-
als, ammunition, and spare parts; organizing the supply chain; protecting roads,
bridges, and waterways; and ensuring that field equipment is in serviceable condi-
tion (Joint Chiefs of Staff 2013, II-1-6). Logistics determine the scope of what is
On the Logistics of Violence 401
feasible, and can have profound implications for both strategy and tactics, shaping
the amount and type of violence that we observe.
18.2.1. Local and External Resources

The effect of logistics on the production of violence depends on the type of supply
system an armed group uses—one that relies mostly on locally obtained resources
or those shipped in from outside (Leites and Wolff 1970, 76; Kress 2002, 29).
In the first case, the army recruits its personnel and procures its supplies from
the immediate geographical area in which it is deployed. This requires a local
resource base with sufficient carrying capacity to support the force, as well as
the cooperation or acquiescence of the local population. In the second case, an
army supports its operations through a logistical network, which connects supply
depots with distribution centers, and home bases with forward-deployed units.
The external option requires, first and foremost, open and protected lines of com-
munication. The greater one’s reliance on external resources, the greater the logis-
tical burden.
Most combatants rely on some bundle of local and external resources. The
balance depends on static factors like infrastructure and resource endowments,
as well as dynamic factors like popular support and the actions of adversaries.
A reliance on local support emerges where there is little infrastructure to receive
outside supplies, or where use of this infrastructure is prohibitively expensive.
A reliance on external support is likely where the local population is uncoopera-
tive or where key materiel cannot be locally obtained.
18.2.2. Government Logistics

In confrontations between governments and their nonstate opponents, the
incumbent generally has the heavier logistical tail. Because rebel armies typically
organize their forces into small light infantry units, many of the basic resources
needed to produce violence—recruits, food, water, and clothing—can be locally
acquired through voluntary donations, taxation, and various forms of coercion
(Mao 1966, 111). Even firearms, body armor, and explosives can sometimes be
locally purchased or looted from police stations, prisons, commercial and indus-
trial enterprises, and private individuals.
For regular armies and state security forces, requisitions from the immedi-
ate neighborhood are too insufficient and uncertain to keep soldiers well fed and
stocked. As field armies grew in size in early modern Europe, mass foraging and
plunder quickly exhausted local supplies, making it difficult to occupy any one
piece of territory for a significant period of time. To ensure that provisions kept
flowing, militaries developed sophisticated networks of supply convoys, garri-
sons, and magazines (van Creveld 2004, 16, 41–42). Government supply needs
402 Case Studies II
expanded in the twentieth century as armies became mechanized, their fleets of

vehicles requiring massive quantities of fuel to remain mobile.
18.2.3. Rebel Logistics

Although rebels generally rely on local support more than governments do, this
reliance is not absolute. Some groups can insulate themselves from the whims of
popular support by extracting revenue from lootable natural resources such as
diamonds (Bellows and Miguel 2009). Yet the insulation is only partial, as extrac-
tion often requires local labor (Weinstein 2007, 173). Alternatively, groups may
solicit support from external patrons, like foreign governments, ethnic diasporas,
charities, and volunteers (Saideman 2002; Salehyan, Gleditsch, and Cunningham
2011; Bakke 2014).
Few groups are fully self-sufficient, but some receive more external support
than others. Cross-national evidence has shown that groups with links to an eth-
nic diaspora are more likely to attract foreign support (Saideman 2002; Gleditsch
2007). Other studies have pointed to a more complex set of explanations of exter-
nal support, like initial rebel strength and the existence of a transnational con-
stituency (Salehyan, Gleditsch, and Cunningham 2011) or simple cost savings
(Salehyan 2010).
An emerging consensus in cross-national literature on civil conflict is that rebel
groups with greater access to external support are more indiscriminate in their use
of force (Weinstein 2005, 2007; Wood 2010, 2014; Salehyan, Siroky, and Wood
2014). Where rebels receive external support, they are less reliant on the local
population, less likely to seek cooperative bargains with local civilians, and less
vulnerable to local backlash (Beardsley and McQuinn 2009). Recent evidence has
shown that groups reliant on external support are able to operate even where it
is very costly for the local population to support them (Toft and Zhukov 2015). Not
surprisingly, external support for rebels tends to make civil wars longer (Balch-
Lindsay and Enterline 2000; Cunningham 2006; Fearon and Laitin 2007).
As rebels receive more support from outside, their logistical requirements
increase. Much like governments, rebels become dependent on the efficient deliv-
ery of supplies through logistical networks (Galula 1964, 40; Buhaug and Gates
2002). In South Vietnam, the Viet Cong created a complex system of roads, rivers,
fuel pipelines, and porters—collectively known as the Ho Chi Minh Trail—to
transport manpower and supplies from the north. In Gaza, Palestinian militant
groups have used underground tunnels to smuggle arms, food, and money from
Egypt, and to smuggle fighters and weapons into Israel (Piven 2014). Recent
research on the Caucasus and Afghanistan has shown that rebel violence can
spread through road networks (Zhukov 2012) and tends to cluster around major
lines of communication (O’Loughlin and Witmer 2011; O’Loughlin et al. 2010).
18.2.4. Sources of Logistical Costs

The costs of utilizing external resources depend on what is being transported and
how. Among the most significant drivers of costs is physical distance (Boulding
1962; Sprout and Sprout 1965; Starr 1978; Schutte 2014). As distances increase
between logistical bases and conflict zones, armies divert more resources to non-
combat duties like escort and supply chain management (Cederman, Buhaug,
and Rod 2009, 503), and more investment becomes necessary to keep fighting.
Costs also depend on cargo type and means of transportation. Fuel and ammu-
nition weigh more than food and spare parts, and are also in greater demand.
Trains are cheaper to operate than trucks. Almost every means of transport is
cheaper than air. Railways and highways have a greater capacity than unpaved
roads, where truck breakdowns are common and movement is slow (Dunnigan
2003, 499–500). The most basic mode of transportation—by foot—is least effi-
cient, since porters and packhorses move slowly, carry little, and require regular
food and rest.
18.3. Logistics and the Scale of Violence

Where the costs of acquiring external resources are low—due to robust infra-
structure, short distances, or a lack of enemy sabotage—v iolence can be more
intense and often more indiscriminate. To a greater extent than other types of
violence, mass killings and arrests are difficult to sustain through local resources
alone. The scale and type of such violence is therefore highly sensitive to disrup-
tions in the logistics network.
Logistical cost and complexity increase with an operation’s scale. An individ-
ual detention requires little more than two or three armed personnel, a transport
vehicle, and a jail cell. To detain multiple individuals from a single neighborhood
requires a cordon team to regulate entry and exit to the area; a search team to con-
duct reconnaissance, check documents, interrogate suspects, and make arrests;
and a third team to transport and process the prisoners. Such operations cannot
easily reach a massive scale, ensnaring hundreds and thousands of people in a
short period of time, without a surge in logistics, as reinforcements arrive to assist
the local personnel, as additional detention facilities become necessary to hold
the prisoners, and as the number of people requiring transportation, food, and
medicine grows.
Air strikes and artillery shelling present different sorts of logistical problems.
While these forms of violence do not require the direct insertion of personnel into
the location of the fighting, and, as such, are attractive when ground transporta-
tion options are limited, airfields and firing positions still require significant sup-
plies and open lines of communication. A single sortie, one flight by one aircraft
404 Case Studies II
on one mission, requires hundreds of hours of labor and dozens of tons of supplies
to launch. A single artillery piece may fire up to five hundred shells in a single day,
consuming some twenty tons of ammunition, or two to six times the carrying
capacity of a typical military truck (Dunnigan 2003, 107–8, 509).
The economies of scale needed for a systematic campaign of mass violence
requires extensive coordination and logistical infrastructure. It is difficult if not
impossible to incarcerate, enslave, or kill tens of thousands of people, much less the
tens of millions ensnared by Soviet, German, and Chinese democides (Rummel
1994), without resource mobilization and sustainment on a colossal scale.
To scale the violence and sustain it for an extended period of time, a heavily
armed and mechanized combatant needs access to potentially thousands of tons
of fuel, munitions, and spare parts each day. Even combatants reliant on small
arms and edged weapons, like militias in many parts of the developing world,
need to ensure that personnel are present in sufficient numbers and are ade-
quately fed and equipped to keep fighting. If the violence is expected to last more
than a month, unit rotations will be necessary to relieve tired troops and maintain
morale.
There is yet another reason why large-scale violence places heavier demands
on external resources: local personnel can be reluctant to use force against their
neighbors and co-v illagers, particularly where doing so puts their families at risk
of retaliation. To suppress a local uprising, governments often rely on military and
police units from other parts of their country (Hassan 2015), further increasing
the logistical burden.
18.4. Disruptions to Logistics

If large-scale violence requires large-scale resources, can disruptions to a combat-
ant’s supply network reduce such violence? Armed groups can seek to interrupt
their opponents’ local or external sources of support. Local disruptions include
mass killings, population resettlement, and other efforts to reduce or eliminate a
combatant’s civilian base of support. External disruptions include interdictions,
sieges, blockades, ambushes, and various forms of sabotage aimed at raising the
costs of outside support.
18.4.1. Disruption of Local Support

Attacks on local sources of support come in two varieties. First is the use of coer-
cion to compel local residents to either not cooperate with the opponent (Kalyvas
2006) or leave the area (Steele 2009). As members of the local population defect
or flee in large numbers, a combatant gradually becomes unable to extract signifi-
cant resources from his local support base.
A second category of local actions includes brute force campaigns of mass

killing (Valentino, Huth, and Balch-Lindsay 2004) and population resettlement
(Zhukov 2015b), which operate by physically eliminating local civilians or relo-
cating them elsewhere. If the logic of the first category is one of deterrence, rais-
ing the costs of “bad” behavior through threats, intimidation, and selective force,
then the second adopts the logic of resource denial: a combatant cannot extract
much support from someone who is either dead or in a concentration camp.
Efforts to disrupt local resources, somewhat ironically, often require signifi-
cant external resources to implement. The deterrence, physical elimination, and
detention of local civilians are all tasks that require personnel, weapons, and
supplies.
The types of tactics combatants employ against civilians depend on what
resources are readily available. If there are no means to transport heavy artillery
within range of the targeted locality, artillery shelling will not be part of a com-
batant’s tactical choice set. If it is prohibitively costly to transport prisoners to a
detention facility far away, the prisoners are less likely to be exiled, relative to other
options. The more logistically costly a given technology of violence becomes, the
less likely combatants are to use it.
18.4.2. Disruption of External Support

Attacks on external supply networks can constrain a combatant’s behavior
through several mechanisms. First is the direct disruption of vital supplies like fuel
and ammunition, which can gradually render a force immobile and ill-equipped.
A combatant can forestall this outcome by rerouting traffic, stockpiling supplies,
and reducing daily consumption to overcome short-term shortages, but such
countermeasures are not without costs of their own. Second, supply disruptions
divert military resources away from offensive operations, and toward emergency
management, engineering, and defense. Units committed to rebuilding bridges,
tunnels, and railroads, escorting convoys, and monitoring ambush points are
ones that cannot simultaneously be used for patrols, sweeps, and other efforts
to find and kill the enemy. Third, sufficiently frequent disruptions can change
the structure and capacity of a support network, forcing traffic to use expensive
detours, creating choke points and congestion, and exposing the supply chain to a
new set of vulnerabilities. If fewer people and materiel can move from point A to
point B in a set interval of time, the tempo of operations will inevitably decline.
In all three cases, attacks on supply lines make it more costly for the opponent to
operate and force him to make some unpleasant choices.
Examples of external supply disruptions abound on both the government
and rebel sides. The Russian army employed siege warfare extensively during
the Caucasus Wars of 1816–1864, to contain Imam Shamil’s forces in fortified
406 Case Studies II
mountain settlements, where supplies of food and water would grow increasingly
scarce (Baddeley 2005 [1908], 323). A century later, government forces in Malaya
used an extensive system of checkpoints and rail and road traffic inspections to
enforce a food denial policy aimed against the guerrillas’ supply chain (Komer
1972, 59–60). In Algeria in the 1950s, French forces devoted great resources to
seal the border from Tunisia and Morocco, in a campaign so disruptive that the
Front de Libération Nationale (FLN) eventually buried most of their automatic
weapons for lack of ammunition (Galula 1964, 30).
These practices continue in contemporary conflicts. In the Syrian city of Homs,
government troops cut off supply routes, along with electricity, telecommunica-
tions, and water, to the rebel-controlled neighborhoods of Old City and Khalidiya.
As one activist described it to a journalist, “The only thing they haven’t blocked is
the air we breathe” (Barnard 2013). Similar efforts could be observed during the
Serb siege of Dubrovnik, the Croat siege of Bihac, and the Serbian blockade of
Sarajevo (Waxman 1999; Andreas 2011). To take a more recent example, the pri-
mary military objective of Israel’s 2014 ground incursion into Gaza was to close
and destroy Hamas’s network of 500 underground tunnels (Piven 2014).
Efforts to disrupt an opponent’s supply chain can cause widespread devasta-
tion. During Major General William T. Sherman’s March to the Sea in the US
Civil War, Union forces heavily bombarded, evacuated, and razed Atlanta, to
eliminate what was then a major Confederate manufacturing center and railway
hub (Waxman 1998, 376–77). A century later, some 500,000 civilians died of star-
vation during Nigeria’s blockade of Biafra in 1967–1970 (de St. Jorre 1972, 412).
Not surprisingly, rebels have also sought to exploit the vulnerabilities that the
government’s heavy logistical tail creates. T. E. Lawrence observed that Turks’
long supply lines exposed them to ambushes and blockades by Faisal’s irregular
forces during the Arab Revolt of 1916–1918 (Lawrence 1920, 55–56). In subse-
quent decades, Chinese communists blockaded Suchow in 1948, the Viet Cong
conducted a siege of Khe Sahn during the 1968 Tet Offensive in Vietnam, and
in 1975 the Khmer Rouge laid siege to Phnom Penh. More recently, the Forces
démocratiques de libération du Rwanda (FDLR) used road blockades to isolate
areas of North and South Kivu.
18.5. Empirical Evidence

The following section compares the claims made in this chapter against the
empirical record, using disaggregated data on contemporary and historical civil
conflicts. To evaluate the claim that violence against civilians is decreasing in
logistical costs, I use district-level data from fifty-eight contemporary civil con-
flicts in Africa (and a small number of countries elsewhere). To assess the claim
that logistical costs shape tactical choices in violence against civilians, I use
archival data on Stalin’s Great Terror. To show that logistical sabotage reduces
the intensity of violence against civilians, I use data on Soviet partisan efforts to
disrupt German supply lines during World War II.
18.5.1. Violence against Civilians

in Contemporary Civil Wars
Although the mechanization of warfare has expanded the technologies of violence
available to combatants, most mass killings today are still carried out with small
arms and melee weapons.2 Unlike the resource-intensive armored formations and
field artillery units that dominated conventional warfare in the twentieth century,
today’s lightly equipped death squads and militias have a much lighter logistical
burden, and rely on foraging at least as much as on hauling fuel, food, and ammu-
nition from elsewhere. Contemporary civil conflicts in Africa and elsewhere
should then pose a hard test for the proposition that higher logistical costs reduce
the intensity of violence.
If logistical constraints can be shown to reduce violence against civilians in
areas where logistics should matter the least, in irregular conflicts like Rwanda
and the Democratic Republic of Congo, we can have greater confidence that simi-
lar patterns obtain in less extreme circumstances. The Armed Conflict Location
and Event Data (ACLED) project catalogs the violent activities of governments,
opposition groups, political parties, and militias, in primarily an African set of
states, from 1997 to 2010 (Raleigh et al. 2010). The ACLED includes data on over
80,000 events from 58 conflicts in 57 countries, disaggregated between violence
directed against armed political actors and violence against civilians. 3
For each event, I classified its initiators into government and rebel categories,
and extracted the subset of actions classified by the ACLED as “violence against
civilians.”4 To aggregate these individual incidents of violence into consistent
units of space and time, I used the district-week as a level of analysis. Districts are
second-order administrative divisions, lower than a province or governorate, but
above a village or town. They are politically relevant as centers of local govern-
ment power. Figure 18.1 shows, for the African cases only, the spatial distribution
of the resulting event counts, pooled across all actors.
As previously discussed, I expect violence against civilians, by either side, to
be more intense where logistical costs are low. I capture these logistical costs with
two covariates: road density, or kilometers of primary and secondary paved roads
per square kilometer of area (Defense Mapping Agency 1992), and the physical
distance from each district center to the national capital. I expect violence against
civilians to be increasing in road density for both groups, and violence by govern-
ment forces to be decreasing in distance from the hub of a country’s political and
military power. I also include standard control variables for linguistic diversity,
land cover, rough terrain, urbanization, and population density. 5
408 Case Studies II
Violence vs. civilians

0
(1, 10)
(10, 50)
(50, 100)
(100, 250)
(250)
Figure 18.1 Violence against civilians, 1997–2010. District-level event counts from
ACLED data (Africa only). Government and rebel-i nitiated actions included.
Analysis of the data indeed suggests that violence by both government and
rebel forces is more intense where logistical costs are low. Table 18.1 shows the
results of quasi-Poisson regressions at the district-week level, with conflict and
year fixed effects. The number of attacks against civilians is significantly higher
in districts with high local road density. This intensity is also higher closer to the
capital, but only for government troops.
Figure 18.2 shows simulations of these results. In districts with just 3 meters of
road per square kilometer of area (1st percentile), the models predict an average
of 0.44 (95 percent confidence interval: 0.34 to 0.57) incidents of government
violence against civilians and 0.08 (0.05 to 0.13) incidents of rebel violence per
week. In a district with the same population, terrain, and other characteristics,
but much higher road density (430 m/k m 2, or 99th percentile), the predicted
levels of violence increase more than threefold for the government, to 1.87 (1.16
to 3.05), and fivefold for rebels, to 0.45 (0.20 to 1.03). The government’s ability
to generate violence against civilians also decreases at longer distances from the
capital city. For the rebels, many of whom do not depend on resources from the
capital, this distance has no significant effect on violence.
These results offer two important insights on the general relationship between
logistics and anticivilian violence. First, both governments and rebels rely on
access to transportation infrastructure to generate and scale violence against
Table 18.1 Determinants of Violence against Civilians, 1997–2 010: Quasi-Poisson Regression with Conflict and Year Fixed Effects
Dependent Variable:
Gov’t Violence vs. Civilians Rebel Violence vs. Civilians
(1) (2) (3) (4) (5) (6)

Road density (per sq. km) 2.030*** 3.380*** 3.269*** 5.106*** 3.957*** 3.500***
(0.54) (0.61) (0.61) (0.76) (1.03) (1.03)
Distance to capital (km) –1.213*** –0.567*** – 0.678*** –0.895*** –0.1 –0.21
(0.12) (0.15) (0.15) (0.19) (0.26) (0.26)
Population density – 0.6 0.54 0.56 2.609*** 3.453*** 3.360***
(0.42) (0.41) (0.4) (0.41) (0.56) (0.56)
Built-up areas 1.378*** 0.573*** 0.623*** – 0.07 – 0.5 –0.44
(0.16) (0.19) (0.19) (0.43) (0.45) (0.44)
SD (elevation) –2 .437*** 0.745** 0.672** 0.420* –0.713** –0.754***
(0.21) (0.31) (0.31) (0.22) (0.3) (0.29)
Open terrain –0.622*** 0.08 0.06 0.352** 0.05 0.01
(0.08) (0.1) (0.09) (0.15) (0.19) (0.19)
Number of languages 0.435** 0.13 0.1 1.745*** 1.053*** 1.118***
(0.17) (0.19) (0.2) (0.23) (0.24) (0.23)
W Gov’t violence 0.319*** 0.074** 0.068**
vs. civilians (t-1) (0.03) (0.03) (0.03)
(Continued)
Table 18.1 (continued)
Dependent Variable:
Gov’t Violence vs. Civilians Rebel Violence vs. Civilians
(1) (2) (3) (4) (5) (6)

Gov’t violence 0.095*** 0.061*** 0.057***
vs. civilians (t-1) (0.01) (0.01) (0.01)
W Rebel violence 0.443*** 0.269*** 0.221**
vs. civilians (t-1) (0.09) (0.09) (0.09)
Rebel violence 0.447*** 0.401*** 0.373***
vs. civilians (t-1) (0.02) (0.02) (0.02)
Constant –0.154** –2 .106*** –1.880*** –2 .972*** –2 .497*** –2 .402***
(0.07) (0.14) (0.17) (0.13) (0.2) (0.33)
Conflict fixed effects N Y Y N Y Y
Year fixed effects N N Y N N Y
Observations 11046 11046 11046 11046 11046 11046
AIC 35445.02 30826.01 30648.8 28522.49 27773.41 27609.14
Note: *p<0.1; **p<0.05; ***p<0.01.

GOVT violence vs. civillians
GOVT violence vs. civillians

2.5
0.7
1.5
0.5
0.5 0.3
0 0.2 0.4 0.6 0 400 800 1400
Road density (per sq. km) Distance from capital (km)
REBEL violence vs. civillians
REBEL violence vs. civillians

0.8 0.18
0.6 0.14
0.4
0.10
0.2
0.06
0 0.2 0.4 0.6 0 400 800 1400
Road density (per sq. km) Distance from capital (km)
Figure 18.2 Predicted levels of violence against civilians. Parameter values from Model
2 used for government violence, Model 5 for rebel violence. All other variables held
constant at median values.
civilians. Although rebels generally have a lighter military footprint and lower
level of mechanization than modern state armies, they are just as vulnerable to
traditional logistical constraints. Second, these constraints have a powerful
impact on violence even in low-intensity African (and other states’) civil conflicts,
where one might expect a relatively limited use of tanks and armored vehicles to
minimize the importance of conventional military logistics.
18.5.2. Stalin’s Mass Resettlements

Among the other claims made in this chapter is that logistical costs, all else equal,
should explain variation in tactics. One of the most extreme and common forms
of violence used by modern governments against civilians is population reset-
tlement (Dobby 1952, 163–6 4; Galula 1964, 82). Governments have forcibly
uprooted and relocated civilian populations in almost a third of all counterinsur-
gency campaigns since the Napoleonic era, including over two dozen cases since
the end of the cold war (Zhukov 2015b). In three decades of Joseph Stalin’s rule in
the Soviet Union, authorities forcibly resettled twelve million people. These oper-
ations generally revolved around one overarching goal—to undermine the local
popular support base of the regime’s political opponents, real or imagined.
Why did the Soviets rely heavily on resettlement in some areas, but used other
forms of mass repression elsewhere? Due to the wealth of archival data currently
412 Case Studies II
Resettled
Not resettled
Rail network
Figure 18.3 Soviet rail network and mass resettlement. Points represent 2.65 million
arrests by Soviet secret police, 1917–1959.
available on the actions of the Soviet secret police, Stalin’s resettlement policy
offers an opportune test of tactical substitution.
The Soviet resettlement campaign ranks among the most logistically taxing
enterprises ever mounted by any government against its citizens. To relocate
21,000 households during a single resettlement operation in 1941, the Soviet
Union required 10,000 police to round up the families, 636 freight cars to trans-
port them, and another 194 passenger cars to carry the 1,300 enlisted personnel,
195 officers, and 260 medics accompanying the human cargo to its destination.6
Despite its impressive scale, this operation accounted for less than one percent of
all Soviet citizens forcibly resettled by Stalin’s regime (Pobol’ and Polyan 2005).
To evaluate whether low logistical costs made resettlement more likely, I use
data on 2.65 million arrest records from 1917–1959, collected from Russian
and other post-Soviet archives (Memorial 2014), along with origin-destination
distances between 618 major Soviet rail junctions from a geo-referenced Soviet
military map (Military-Topographical Directorate of the General Staff of the
Red Army 1945). Figure 18.3 shows the geographic distribution of the arrest
records, the sentences issued (resettlement or no resettlement), and the railroad
network used to transport resettled arrestees to their destination.7 The data show
that 33.6 percent of the arrestees received a sentence of forcible resettlement. The
remainder received sentences including local incarceration (15.6 percent), execu-
tion (3.7 percent), property confiscation (1.6 percent), compulsory medical treat-
ment (0.2 percent), or travel bans (less than 0.1 percent).8
To measure the logistical costs of resettlement, I use the locations of 618 major
Soviet railroad junctions from 1945 and the travel distances between them.9
I matched arrests and junctions by minimum geographic distance, such that each
arrest is matched to its nearest rail junction, and each junction is matched with the
set of arrests to which it is most proximate. In so doing, I calculated aggregated
arrest statistics at the junction level, and appended nearest-junction attributes to
the arrest data. Table 18.2 reports summary statistics at both levels of analysis.
Table 18.2 Summary Statistics: Soviet Mass Terror Data, 1945–1959
Junction-Level Range Median Mean Std. Dev.

(N = 618)
Total arrests [0, 69374] 1581.5 3730.41 6354.18
Resettled (sum) [0, 52835] 271 1259.94 3511.32
Resettled [0, 0.976] 0.21 0.27 0.22
(proportion)
Rail distance to [0.402, 149.63 204.8 218.56
nearest Gulag (km) 1839.74]
Proportion ethnic [0, 0.783] 0.18 0.22 0.17
Russian
Proportion farm [0, 0.633] 0.11 0.14 0.12
workers
Betweenness [0, 89518.253] 5101.88 10164.02 12999.3
centrality
Betweenness [0, 0.014] 0 0 0
centrality (rescaled)
Arrest-level Range Median Mean Std. Dev.
(N = 2,305,394)
Resettled [0, 1] 0 0.34 0.47
Distance to nearest [0, 22347.139] 53.07 110.14 299.13
rail junction (km)
Rail distance to [0.803, 185.14 254.7 357.52
nearest Gulag (km) 22562.83]
Ethnic Russian [0, 1] 0 0.24 0.43
Farm worker [0, 1] 0 0.14 0.35
Betweenness [0, 89518.253] 9358.83 16295.48 18215.27
centrality
Betweenness [0, 0.014] 0 0 0
centrality (rescaled)
414 Case Studies II
Table 18.3 Regression Results for Soviet Mass Terror Data: Right-Hand-Side
Variables Rescaled between 0 and 1
Level of Analysis
Junction Arrest
Dependent Variable
Arrested Resettled Resettled Resettled

(count) (count) (dummy) (dummy)
Poisson Logit
Model 7 Model 8 Model 9 Model 10

Distance to nearest –23.17***
junction (0.30)
Rail distance to –2 .42*** –2 .64*** –17.74***
nearest Gulag (0.01) (0.02) (0.21)
Centrality 1.35*** 1.47*** 0.40*** 0.33***
(0.00) (0.01) (0.01) (0.01)
Ethnic Russian –0.39*** –2 .13*** –1.59*** –1.60***
(0.00) (0.01) (0.01) (0.01)
Farm worker –1.00*** –2 .46*** – 0.72*** –0.70***
(0.00) (0.01) (0.01) (0.01)
Regional dummies Y Y Y Y
Time trend N N Y Y
Observations 611 611 1835584 1834322
Log Likelihood –1209711 –681719.6 –984478.5 –983578.8
Akaike Inf. Crit. 2419460 1363477 1968997 1967198
Note: *p<0.1; **p<0.05; ***p<0.01.
These data suggest that the scale and nature of Stalin’s mass resettlement
depended strongly on railroad infrastructure. Table 18.3 reports the results of four
regression models. Models 7 and 8, at the junction level, are Poisson event count
models regressing, respectively, the number of people arrested or resettled on the
travel distance, by rail, to the nearest Gulag; the centrality of the junction in the
network; and local demographic and economic characteristics. Models 9 and 10,
at the arrest level, are logit regressions of individual resettlement, a dummy vari-
able, on the distance to the nearest rail junction, the distance from that junction
to the nearest Gulag camp, the centrality of that junction, the arrestee’s demo-
graphic and economic attributes, and regional dummies.
In each model, the intensity and probability of resettlement is decreasing in

distance from the rail network and Gulags, and increasing in centrality. All else
equal, a person living 10 kilometers from a rail station was 6.8 percent (6.6 to
6.9) more likely to receive a sentence of resettlement, to a Gulag labor camp or a
special settlement, than someone living 100 kilometers away. The probability of
resettlement was higher still if the rail junction occupied a central position in the
Soviet long-d istance rail network, or if the railway distance to the nearest Gulag
camp was relatively short. Victims of repression who lived in less logistically
accessible areas were more likely to receive other types of sentences, including
local incarceration and confiscation of property.
In contrast to the common view of Stalin’s regime as an omnipotent Leviathan
that sent its enemies to Siberia on a whim, the quantity and quality of Soviet
repression was highly dependent on logistical costs. Archival data on mass arrests
in the Soviet Union show that the scale of repression, and the likelihood of popu-
lation resettlement, was higher in locations relatively accessible by rail, and prox-
imate to preexisting labor camps and “special settlements.” Logistics, it seems,
were among the very few constraints on Stalin’s totalitarian power.
18.5.3. Railroad Sabotage in World War II

A third claim in need of some empirical validation is that efforts to disrupt a
combatant’s logistics should decrease violence against civilians. One of the most
extensive campaigns of this type was the Rail War waged by Soviet partisans
in German-occupied Belarus. Between 1941 and 1944, Belarusian partisans
derailed 11,128 trains carrying German personnel and supplies to the eastern
front (Bryukhanov 1980, 251). In response, German forces conducted mass
reprisals against the local population, razing over 9,000 villages and summarily
executing the civilians who lived there (State Archives of the Republic of Belarus
2014). In part due to the brutality of these reprisals, Belarus lost a larger share of
its population, 25 percent, than any other country in World War II.
Western historians have attributed much of the blame for this devastation to
the partisans themselves. As Snyder (2011, 36) writes, “[T]‌he Germans shot so
many civilians in part because Soviet partisans deliberately provoked reprisals.”
Recent research in political science has challenged this assertion (Zhukov 2015a).
To see if railroad sabotage inflamed or suppressed German reprisal killings,
I use archival data from the State Archives of the Republic of Belarus (2014),
and a series of historical maps from Gamov et al. (2013). The data include 8,526
Belarusian villages destroyed by German forces in 1941–1944, along with statis-
tics on dates, numbers of civilians killed, houses destroyed, and prewar popula-
tion numbers.10 The data also include information on thirty-t hree major World
War II–era railroad junctions in Belarus, and the number of partisan-caused
derailments on the routes connecting them (Figure 18.4).11 I matched villages
416 Case Studies II
Village destroyed by Germans

Rail station
<100 convoys attacked
>500 convoys attacked
People killed
1 <10 <100 <500 >500
Figure 18.4 Partisan railroad.
and junctions by minimum geographic distance, such that each village is asso-
ciated with the nearest rail junction, and each junction with the set of villages
to which it is closest. Summary statistics at both levels of analysis are shown in
Table 18.4.
Despite the popular perception of Soviet partisans as purposefully provoking
German retaliation against civilians, archival evidence suggests that their sabotage
of railways had the opposite effect. Table 18.5 reports the results of several regres-
sion models. Models 11–13, at the junction level, are Poisson event count models
regressing the number of destroyed villages—and elsewhere, the total number of
civilians killed and houses destroyed—in the vicinity of the junction, on the num-
ber of trains derailed by partisans and the betweenness-centrality of the junction.
Models 14 and 15 are also Poisson models regressing the number of civilians killed
per village—or, separately, houses destroyed—on the number of train derailments
at the closest junction, and a set of provincial dummies. In each model, partisan
train derailments have a strong, negative relationship with German reprisals.
All else equal, the Germans destroyed significantly fewer villages in areas
where the partisans derailed more trains (Table 18.6). In the villages they
did destroy, they killed a significantly smaller proportion of the population
and destroyed fewer houses. A high-r isk junction in the rail network (1,030 train
derailments, or 90th percentile) saw 49 percent fewer nearby villages destroyed
than a low-r isk junction (9 derailments, 10th percentile), even after controlling
for the centrality of the junction in the rail network. German-attacked villages
near high-r isk junctions saw 20 percent fewer civilian deaths and 17.5 percent
fewer houses demolished than otherwise similar villages near a low-r isk section
of the network.
Table 18.4 Summary Statistics: Belarus Data
Junction-Level Belarus Range Median Mean Std. Dev.

Data. N = 33.
Villages destroyed by [0, 1053] 138 189.3 235.19
Germans
People killed by Germans [0, 19242] 3447 4708.39 4673.51
Houses destroyed by Germans [0, 33244] 5791 8301.97 9155.42
Prewar population [0, 162083] 37187 46251.58 48526.57
Prewar number of houses [0, 43389] 8259 11237.48 11978.64
Trains derailed by partisans [0, 1548] 252 449.09 486.85
Rails destroyed by partisans [0, 47729] 9632 14797.7 14472.15
Routes closed by partisans [0, 3] 1 0.73 0.84
Betweenness centrality [0, 559.333] 9 117.12 163.83
(rescaled)
Village-Level Belarus Range Median Mean Std. Dev.
Data. N = 7967.
People killed by Germans [0, 2060] 5 24.64 82.51
Houses destroyed by [0, 1257] 27 43 54.72
Germans
Prewar population [0, 5500] 152 239.89 305.39
Prewar number of houses [0, 1773] 36 58.36 73.75
Trains derailed by partisans [0, 1548] 149 341.89 417.82
Rails destroyed by partisans [0, 47726] 9632 12044.63 10423.92
Routes closed by partisans [0, 2] 1 0.73 0.72
(rescaled)
What explains the dampening effect of train derailments on reprisals?

The overwhelming majority of cargo passing through Belarus was slated for
use on the front lines against the Red Army. Units stationed within Belarus
were lower on the pecking order for these supplies, and their capacity to punish
the population was only indirectly affected. The larger impact of the sabotage
was to divert German resources to defensive duties. In high-r isk sections of the
network, German forces built pillboxes, laid tripwires, created minefields, lev-
eled forests, established permanent garrisons, and conducted regular patrols.
418 Case Studies II
Table 18.5 Regression Results for Belarus Data: Right-Side Variables

Rescaled between 0 and 1
Level of Analysis
Junction Junction Junction Village Village
Dependent Variable
Villages People Houses People Houses

Destroyed Killed Destroyed Killed Destroyed
Model 11 Model 12 Model 13 Model 14 Model 15

Trains –1.02*** – 0.25*** –0.20*** –0.34*** –0.29***
derailed (0.06) (0.01) (0.01) (0.01) (0.01)
Population 2.10*** 6.19***
(prewar) (0.01) (0.01)
Houses 2.61*** 2.83*** 5.20***
(prewar) (0.04) (0.01) (0.01)
Centrality 0.93*** 0.31*** 0.33***
(0.05) (0.01) (0.01)
Constant 4.28*** 7.62*** 7.88*** 2.77*** 3.73***
(0.03) (0.01) (0.00) (0.01) (0.01)
Regional N N N Y Y
dummies
Observations 33 33 33 7967 7967
Log –1531.04 –39698.19 –38826.33 –233899.5 –123364.3
Likelihood
Akaike Inf. 3070.08 79404.37 77660.67 467815.1 246744.7
Crit.
Note: *p<0.1; **p<0.05; ***p<0.01.
So great was the demand for German personnel to protect the supply lines that
many units were diverted from the front to these rear areas (Bryukhanov 1980,
29–3 0, 49).
Although a more comprehensive look at interdiction lies beyond the scope of
this brief survey (but see Zhukov 2015a), a brief look at the data suggests that the
partisans’ Rail War is more likely to have reduced the German’s violence against
local civilians, than to have increased it.12 If violence against civilians is decreas-
ing in logistical costs, one way to reduce such violence—as the partisans found—
is to increase these costs.
Table 18.6 Simulation Results for Belarus Data. Counterfactual Is an Increase

in Local Train Derailments from 9 (10th Percentile) to 1,030 (90th
Percentile)
Model Dependent Variable Percent Change

Model 11 Villages destroyed –48.8 (95% CI: –52.4,–45.1)
Model 12 People killed –15.4 (95% CI: –16.5,–14.3)
Model 13 Houses destroyed –12.3 (95% CI: –13.1,–11.4)
Model 14 People killed –20.1 (95% CI: –21,–19.1)
Model 15 Houses destroyed –17.5 (95% CI: –18.2,–16.8)
18.6. Conclusion
Mass violence can only occur where it is logistically feasible. Cheap and uninter-
rupted access to external resources is of course not sufficient for governments and
rebels to begin committing atrocities. But where external resources are difficult
or impossible to obtain, mass violence will be difficult or impossible to produce.
The preceding discussion suggests that more systematic efforts to incorporate
logistics into theoretical and empirical models of conflict may help us account for
much previously unexplained variation in violence. Uninterrupted flows of exter-
nal resources may help us understand why some combatants can produce high lev-
els of violence despite a lack of local popular support. Interruptions to these flows
through border closings, ambushes, or sabotage can explain why such violence
sometimes fails to occur, despite compelling incentives to escalate. A group’s rela-
tive reliance on local and external resources may reveal why some actors are more
sensitive to these disruptions than others. Above all, logistics can help us predict
how much violence a group can conceivably generate and, by extension, whether
and how that group’s capacity for violence may be curtailed.
The implications of such research for the study and prevention of genocide
are significant. The preceding analysis has shown that logistical costs constrain
violence against civilians not only during conventional mechanized warfare in
highly industrialized societies but also in dozens of lower-intensity conflicts
in (mostly African) developing countries. Logistics matter for both heavily armed
governments and lightly equipped rebels. Logistics can even shape the behavior
of regimes otherwise unrestrained by institutional checks on their totalitarian
power, such as Hitler’s Germany and Stalin’s Soviet Union.
If the sabotage or interdiction of external resources indeed reduces civilian
suffering, states and organizations seeking to prevent mass atrocities should focus
their efforts on disrupting supply lines and starving the perpetrators of resources
needed to keep killing. Because supply routes are more vulnerable and sparsely
420 Case Studies II
defended than enemy positions, a strategic focus on interdiction is also less costly
for opposition groups and third parties to implement than direct military engage-
ment. A deeper understanding of the logistics of mass killing can uncover ways to
make civil conflicts less deadly to civilians.
Notes
1. A conflict is intrastate if the parties to the dispute are state and nonstate actors competing
for sovereignty (i.e., supreme, independent authority over a body politic) in a common geo-
graphical area. Violence in such conflicts is political if it is part of an organized campaign to
compel loyalty or deter opposition to those perpetrating it. This definition includes most
violence in civil wars, revolutions, armed rebellions, as well as one-sided mass killings and
state terror. It excludes unorganized violence like riots and looting, political nonviolence
like protests, and nonpolitical violence due to criminal activity.
2. The author is indebted to an anonymous reviewer for this observation.
3. The conflict zones include Afghanistan, Algeria, Angola, Benin, Bosnia and Herzegovina,
Burkina Faso, Burundi, Cambodia, Cameroon, Central African Republic, Chad, Republic
of Congo, Djibouti, Democratic Republic of the Congo (first and second Congo wars),
Egypt, Equatorial Guinea, Eritrea, Ethiopia, Gabon, Gambia, Ghana, Guinea, Guinea-
Bissau, Haiti, Côte d’Ivoire, Kenya, Kosovo, Laos, Lesotho, Liberia, Libya, Macedonia,
Madagascar, Malawi, Mali, Mauritania, Morocco, Mozambique, Myanmar, Namibia,
Nepal, Niger, Nigeria, Pakistan, Rwanda, Senegal, Sierra Leone, Somalia, South Africa,
Sudan, Swaziland, Tanzania, Togo, Tunisia, Uganda, Zambia, and Zimbabwe. (Non-
African countries in italics.)
4. The government group includes the military, police, intelligence agencies, and other fed-
eral, regional, and municipal security services subordinate to the executive branch, as well
as militias and paramilitary forces affiliated with the government or ruling party. The rebel
category includes any armed opposition group seeking to challenge the government’s
monopoly on the use of force, locally, regionally, or nationwide, including organized
insurgencies and terrorist organizations, revolutionary movements, paramilitary wings of
opposition parties, secessionist groups, local self-defense units, and ethnic militias outside
the government’s control. Because each country, conflict, and time period featured spe-
cific constellations of combatants, I created a custom actor dictionary for each conflict,
which supplemented (or supplanted) these general classes of actors with the names of spe-
cific organizations and individuals.
5. I acquired data on the number of distinct ethnic groups within each district from the
Georeferencing of Ethnic Groups data, a digital version of the Soviet Atlas Narodov Mira
(Weidmann, Rod, and Cederman 2010). I used the US Geological Service’s (USGS)
Global Land Cover Characteristics database to calculate the proportion of a district’s land
covered by open terrain (Loveland et al. 2000). I used the US National Oceanographic and
Atmospheric Administration’s ETOPO5 5-m inute gridded digital elevation model to cal-
culate the standard deviation of elevation in each district (NOAA 1988). I used the USGS
Global GIS database to calculate the number of unique built-up areas within each district
(Hearn et al. 2005), and used the Gridded Population of the World raster dataset to calcu-
late average local population density in 1990, 1995, or 2000, depending on the start year
of each conflict. Finally, I include a time-lagged, row-normalized spatial lag of violence
against civilians, using a queen’s case border contiguity matrix.
6. State Archive of the Russian Federation (GARF), Fond 9479, Opis 1, Delo 62, List 72–73.
7. Of the arrest records in Memorial (2014)’s “Victims of Political Terror in the USSR”
project, I was able to geocode 2.3 million (87 percent) to the municipal or district level,
using Google Maps API and Yandex.Maps API. At the republican level, 81.3 percent of the
arrests occurred in Russia, 5.6 percent in Ukraine, 5.1 percent in Belarus, 4.6 percent in
Kazakhstan, and less than 1 percent in each of (from most to least) Kyrgyzstan, Moldova,
Lithuania, Latvia, Uzbekistan, Azerbaijan, Georgia, Estonia, Armenia, Turkmenistan, and
Tajikistan.
8. The arrestees’ most common professions were in agriculture (13.9 percent), heavy indus-
try (1.6 percent), service sector (1.3 percent), and forestry (1.2 percent). The most fre-
quently arrested ethnic groups were Russian (24 percent), Polish (3.5 percent), Belarusian
(3 percent), Ukrainian (2.3 percent), German (2 percent), Jewish (1.7 percent), and Tatar
(1.2 percent).
9. Based on the structure of the rail network, I calculated centrality scores for each junction,
using betweenness centrality as the measure: Betweenness centrality(i)= Σi ≠ j ≠ k (νjk(i))/(νjk),
where ν jk is the total number of shortest paths from junction j to junction k and ν jk(i) is the
number of those paths that pass through i. This statistic can be interpreted as the number
of times a rail junction acts as a bridge along the shortest path between two other junc-
tions. I rescaled this measure by (N-1)(N-2)/2 to ensure that it is bounded between 0 and
1, where N = 618 is the total number of junctions in the network.
10. Of these villages, I was able to geocode 7,967 (93 percent) to the district or municipal level,
using Google Maps API and Yandex.Maps API. At the provincial level, 252 of the locations
were in Baranovichi, 23 in Belostok, 330 in Brest, 718 in Gomel’, 992 in Minsk, 1,564 in
Mogilev, 258 in Pinsk, 658 in Polessie, 533 in Vileysk, and 2,601 in Vitebsk voblasts.
11. Based on the structure of the rail network, I calculated betweenness-centrality scores for
each junction, as defined in n. 9.
12. Zhukov (2015a) conducts a more extensive analysis of these data and finds a heteroge-
neous effect in partisan actions: Attacks against German garrisons increased reprisal
killings, while attacks against the rail network reduced them, consistent with the results
shown here.
References
Andreas, P. 2011. Blue Helmets and Black Markets: The Business of Survival in the Siege of Sarajevo.
Ithaca, NY: Cornell University Press.
Baddeley, J. F. 2005 [1908]. The Russian Conquest of the Caucasus. London: Longmans, Green.
Baker, S. F., D. P. Morton, R. E. Rosenthal, and L. M. Williams. 2002. “Optimizing Military
Airlift.” Operations Research 50, no. 4: 582–6 02.
Bakke, K. M. 2014. “Help Wanted? The Mixed Record of Foreign Fighters in Domestic
Insurgencies.” International Security 38, no. 4: 150–87.
Balch-Lindsay, D., and A. J. Enterline. 2000. “Killing Time: The World Politics of Civil War
Duration, 1820–1992.” International Studies Quarterly 44, no. 4: 615–42.
Banerjee, A., E. Duflo, and N. Qian. 2012. “On the Road: Access to Transportation Infrastructure
and Economic Growth in China.” Working paper available at http://w ww.nber.org/
papers/w17897.
Barnard, A. 2013. “Syria Attacks Rebel-Held Area in New Push to Retake City.” New York Times,
June 29. http://w ww.nytimes.com/2 013/0 6/30/world/m iddleeast/s yria-attacks-rebel-
held-a rea-i n-new-push-to-retake-city.html?_r =0.
Beardsley, K., and B. McQuinn. 2009. “Rebel Groups as Predatory Organizations: The Political
Effects of the 2004 Tsunami in Indonesia and Sri Lanka.” Journal of Conflict Resolution 53,
no. 4: 624–45.
Bellows, J., and E. Miguel. 2006. “War and Institutions: New Evidence from Sierra Leone.”
American Economic Review 96, no. 2: 394–9 9.
422 Case Studies II
Bellows, J., and E. Miguel. 2009. “War and Local Collective Action in Sierra Leone.” Journal of
Public Economics 93, no. 11: 1144–57.
Boulding, K. 1962. Conflict and Defense. New York: Harper.
Bryukhanov, A. I. 1980. V shtabe partisanskogo dvizheniya [On the Staff of the Partisan Movement].
Minsk: Belarus’.
Buhaug, H., and S. Gates. 2002. “The Geography of Civil War.” Journal of Peace Research 39, no.
4: 417–33.
Buhaug, H., and J. K. Rod. 2006. “Local Determinants of African Civil Wars, 1970–2 001.”
Political Geography 25, no. 3: 315–35.
Cederman, L.-E ., H. Buhaug, and J. K. Rod. 2009. “Ethno-Nationalist Dyads and Civil War.”
Cunningham, D. E. 2006. “Veto Players and Civil War Duration.” American Journal of Political
Science 50, no. 4: 875–92.
Defense Mapping Agency. 1992. “Development of the Digital Chart of the World.” Washington,
DC: US Government Printing Office.
de St. Jorre, J. 1972. The Brothers’ War: Biafra and Nigeria. Boston: Houghton Mifflin.
Dobby, E. H. G. 1952. “Resettlement Transforms Malaya: A Case-H istory of Relocating the
Population of an Asian Plural Society.” Economic Development and Cultural Change 1, no.
3: 163–89.
Donaldson, D. 2010. “Railroads of the Raj: Estimating the Impact of Transportation
Infrastructure.” Working paper available at http://w ww.nber.org/papers/w16487.
Dunnigan, J. F. 2003. How to Make War: A Comprehensive Guide to Modern Warfare in the Twenty-
First Century. 4th ed. New York: Harper Collins.
Eccles, H. E. 1991. Logistics in the National Defense. Arlington, VA: Headquarters, U.S. Marine
Corps. [Originally published in 1959 by Stackpole Co., Harrisburg, PA.]
Fearon, J. D., and D. D. Laitin. 2007. “Civil War Termination.” Working paper available at http://
sites.lsa.umich.edu/polisci-c pw/w p-content/uploads/sites/2 23/2 015/0 2/d avid_ l aitin.
pdf.
Foxton, P. D. 1994. Powering War: Modern Land Force Logistics. London: Brassey’s.
Galula, D. 1964. Counterinsurgency Warfare. New York: Praeger.
Gamov, V. I., D. A. Migun, S. F. Shimukovych, and I. I. Yanushevych. 2013. “Velikaya
Otechestvennaya Voyna sovetskogo naroda (v kontekste Vtoroy mirovoy voyny). Lektsiya
7: Partizanskaya in podpol’naya bor’ba na okkupirovannoy territory Belarusi.” [“Great
Patriotic War of the Soviet People (in the context of the Second World War). Lecture
7: Partisan and underground struggle in the occupied territory of Belarus.” University
Lecture.] Available at http://w ww.docme.ru/doc/533276/lekciya-7.
Gleditsch, K. S. 2007. “Transnational Dimensions of Civil War.” Journal of Peace Research 44, no.
3: 293–309.
Guerlac, H. 1986. “Vauban: The Impact of Science on War.” In P. Paret, G. A. Craig, and F.
Gilbert, eds., Makers of Modern Strategy from Machiavelli to the Nuclear Age. Princeton,
NJ: Princeton University Press, 64–9 0.
Hassan, M. 2015. “The Strategic Shuffle: Ethnic Geography, the Internal Security Apparatus, and
Elections in Kenya.” Working Paper available at http://scholar.harvard.edu/m hassan/pub-
lications/strategic-shuffle-ethnic-geography-i nternal-security-apparatus-a nd-elections.
Hearn, P. P., T. M. Hare, P. Schruben, D. Sherrill, C. LaMar, and P. Tsushima. 2005. “Global
GIS.” Computer file. U.S. Geological Survey & American Geological Institute. Available at
http://pubs.er.usgs.gov/publication/ds62B.
Joint Chiefs of Staff. 2013. “Joint Logistics.” Joint Publication 4-0. Joint Staff. Defense Technical
Information Center. www.dtic.mil/doctrine/new_ pubs/jp4_0.pdf. [accessed April
29, 2015].
Jomini, A.-H . 1862. The Art of War. Philadelphia: J. P. Lippincott.
Komer, R. W. 1972. The Malayan Emergency in Retrospect: Organization of a Successful

Counterinsurgency Effort. Santa Monica, CA: RAND.
Kress, M. 2000. “Flexibility in Operational-L evel Logistics.” Military Operations Research 5, no.
1: 41–5 4.
Kress, M. 2002. Operational Logistics: The Art and Science of Sustaining Military Operations.
Berlin: Springer.
Lawrence, T. E. 1920. “The Evolution of a Revolt.” Army Quarterly and Defence Journal 1, no.
1: 55–69.
Leites, N., and C. Wolff, Jr. 1970. Rebellion and Authority. Santa Monica, CA: RAND.
Loveland, T. R., B. C. Reed, J. F. Brown, D. O. Ohlen, Z. Zhu, L. W. M. J. Yang, and J. W. Merchant.
2000. “Development of a Global Land Cover Characteristics Database and IGBP DISCover
from 1 km AVHRR data.” International Journal of Remote Sensing 21, nos. 6–7: 1303–30.
Mao, T.-T. 1966. Basic Tactics. Trans. by Stuart R. Schram. New York: Praeger.
McFate, M. 2005. “Iraq: The Social Context of IEDs.” Military Review 85, no. 3: 37–4 0.
Memorial. 2014. “Zhertvy politicheskogo terrora v SSSR.” [Victims of Political Terror in the
USSR.] http://l ists.memo.ru/.
Military-Topographical Directorate of the General Staff of the Red Army. 1945. Skhemy
avtomobil’nykh dorog SSSR [Automobile Road Layout of the USSR]. Moscow: Voennoe
izdatel’stvo Narodnogo Komissariata SSSR.
Murshed, S. M., and S. Gates. 2005. “Spatial-Horizontal Inequality and the Maoist Insurgency
in Nepal.” Review of Development Economics 9, no. 1: 121–3 4.
NOAA. 1988. “Data Announcement 88-MGG-02, Digital Relief of the Surface of the Earth.”
National Geophysical Data Center, Boulder, CO.
O’Loughlin, J., and F. D. Witmer. 2011. “The Localized Geographies of Violence in the North
Caucasus of Russia, 1999–2 007.” Annals, Association of American Geographers 101, no.
1: 178–2 01.
O’Loughlin, J., F. D. Witmer, A. M. Linke, and N. Thorwardson. 2010. “Peering into the Fog
of War: The Geography of the WikiLeaks Afghanistan War Logs, 2004–2 009.” Eurasian
Geography and Economics 51, no. 4: 472–95.
Owen, R. C., and K. P. Mueller. 2007. Airlift Capabilities for Future U.S. Counterinsurgency
Operations. Santa Monica, CA: Rand Corporation.
Piven, B. 2014. “Gaza’s Underground: A Vast Tunnel Network That Empowers Hamas.” Al Jazeera,
July 23. http://america.aljazeera.com/articles/2014/7/23/gaza-undergroundhamastunnels.
html [accessed April 29, 2015].
Pobol’, N. L., and P. M. Polyan, eds. 2005. Stalinskie Deportatsii, 1928–1953 [Stalin’s Deportations,
1929–1953]. Moscow: Materik.
Raleigh, C., and H. Hegre. 2009. “Population Size, Concentration, and Civil
War: A Geographically Disaggregated Analysis.” Political Geography 28, no. 4: 224–38.
Raleigh, C., A. Linke, H. Hegre, and J. Karlsen. 2010. “Introducing ACLED: An Armed Conflict
Location and Event Dataset.” Journal of Peace Research 47, no. 5: 651–6 0.
Rogall, T. 2014. “Mobilizing the Masses for Genocide.” Working paper available at https://w ww.
hhs.se/contentassets/8f81d097190942ce97b901d5cef1d4ab/rogall_ jmp.pdf.
Rummel, R. J. 1994. Death by Government. New Brunswick, NJ: Transaction Publishers.
Saideman, S. M. 2002. “Discrimination in International Relations: Analyzing External Support
for Ethnic Groups.” Journal of Peace Research 39, no. 1: 27–50.
Salehyan, I. 2010. “The Delegation of War to Rebel Organizations.’’ Journal of Conflict Resolution
54, no. 3: 493–515.
Salehyan, I., K. S. Gleditsch, and D. E. Cunningham. 2011. “Explaining External Support for
Insurgent Groups.” International Organization 65, no. 4: 709–4 4.
Abuse: A Principal-A gent Analysis of Wartime Atrocities.” International Organization 68,
no. 3: 633–61.
424 Case Studies II
Schutte, S. 2014. “Geography, Outcome, and Casualties: A Unified Model of Insurgency.” Journal
of Conflict Resolution 59, no. 6: 1001–1028.
Snyder, T. 2011. “Hitler vs. Stalin: Who Killed More?” New York Review of Books. March 10: 35–
36. http://w ww.nybooks.com/a rticles/a rchives/2 011/mar/10/h itler-vs-stalin-who-
killed-more/ [accessed April 29, 2015].
Sprout, H., and M. Sprout. 1965. The Ecological Perspective on Human Affairs. Princeton,
Starr, H. 1978. “‘Opportunity’ and ‘Willingness’ as Ordering Concepts in the Study of War.”
International Interactions 4, no. 4: 363–87.
State Archives of the Republic of Belarus. 2014. “Belorusskie derevni, unichtozhennye v gody
Velikoy Otechestvennoy voyny. 1941–1944 gody.” [Belarusian Villages Destroyed during
the Great Patriotic War. The Years 1941–1944.] http://db.narb.by/.
Steele, A. 2009. “Seeking Safety: Avoiding Displacement and Choosing Destinations in Civil
Wars.” Journal of Peace Research 46, no. 3: 419–30.
Toft, M. D., and Y. M. Zhukov. 2012. “Denial and Punishment in the North Caucasus: Evaluating
the Effectiveness of Coercive Counterinsurgency.” Journal of Peace Research 49, no.
6: 785–800.
Toft, M. D., and Y. M. Zhukov. 2015. “Islamists and Nationalists: Rebel Motivation and
Counterinsurgency in Russia’s North Caucasus.” American Political Science Review 109, no.
2: 222–38.
Townsley, M., S. D. Johnson, and J. H. Ratcliffe. 2008. “Space Time Dynamics of Insurgent
Activity in Iraq.” Security Journal 21, no. 3: 139–4 6.
Guerrilla Warfare.” International Organization 58, no. 2: 375–4 07.
van Creveld, M. 2004. Supplying War: Logistics from Wallenstein to Patton. New York: Cambridge
University Press.
Waxman, M. C. 1999. “Siegecraft and Surrender: The Law and Strategy of Cities as Targets.”
Virginia Journal of International Law 39: 353–423.
Weidmann, N. B., J. K. Rod, and L.-E . Cederman. 2010. “Representing Ethnic Groups in
Space: A New Dataset.” Journal of Peace Research 47, no. 4: 491–9 9.
Weinstein, J. M. 2005. “Resources and the Information Problem in Rebel Recruitment.” Journal
of Conflict Resolution 49, no. 4: 598–624.
Weinstein, J. M. 2007. Inside Rebellion: The Politics of Insurgent Violence. Cambridge, UK: Cambridge
University Press.
Research 47, no. 5: 601–14.
Wood, R. M. 2014. “From Loss to Looting? Battlefield Costs and Rebel Incentives for Violence.”
International Organization 68, no. 4: 979–9 9.
Zhukov, Y. M. 2012. “Roads and the Diffusion of Insurgent Violence: The Logistics of Conflict
in Russia’s North Caucasus.” Political Geography 31, no. 3: 144–56.
Zhukov, Y. M. 2015a. “External Resources and Indiscriminate Violence: Evidence from
German-Occupied Belarus.” Paper presented at International Studies Association annual
meeting, New Orleans, LA.
Zhukov, Y. M. 2015b. “Population Resettlement in War: Theory and Evidence from Soviet
Archives.” Journal of Conflict Resolution 59, no 7: 1155–85.
19
Strategic Atrocities
Civilians under Crossfire—T heory
and Evidence from Colombia
J ua n F. Va rg a s
19.1. Introduction
Over 3,100 civilians died in 92 massacres in Algeria between August 1996 and
December 1998.1 The war between the government and various Islamist insur-
gencies (notably the Armed Islamic Group) spanned the period 1991 to 2002 and
left a remarkably high civilian death toll. The legacy of massacres in Colombia is
equally distressing: between 1988 and 2005 Colombian guerrillas killed 1,200
civilians in about 200 massacres while over 6,100 died in just under 1,000 mas-
sacres perpetrated by right-w ing militias.2 This pattern of terror is by no means
inherent to these two countries. Targeting civilians was the main strategy of the
Peruvian militias (Rondas Campesinas) to recover rural strongholds under the
control of the Shining Path in the mid-1980s and the 1990s. Before a cease-fi re was
declared in October 2002, the Sudan People’s Liberation Army targeted civilians
all across southern Sudan to punish alleged supporters of Karthoum-backed mili-
tias (Johnson 1995). During Museveni’s rule in the 1990s, the Lord’s Resistance
Army (LRA) alienated the local population in northern Uganda by massacring
civilians. Later, to recover lost territories, Museveni’s National Revolutionary
Army used the same strategy, killing alleged LRA supporters (Berkeley 2001).
What determines whether and to what extent civilians are targeted in civil
war? This chapter explores theoretically and empirically the relationship of mili-
tary empowerment and civilian victimization in contexts of territorial contesta-
tion among armed groups in a civil war. Most civilian killings are deliberately
planned by both state and nonstate actors (Eck and Hultman 2007). The num-
ber of civilians killed intentionally and directly in internal armed conflicts is about
half the number of total deaths in combat. 3 War-induced famine and disease are
likely to hit civilians further both during the conflict and in the postconflict stage.
425
426 Case Studies II
The deaths of noncombatants represent an enormous human tragedy, and also a

severe long-term economic cost of civil war as it erodes the labor force of a coun-
try. In addition, the deliberate killing of civilians nurses hatred and desire for
revenge, which may trigger cycles of violence.
Civilians are targeted to create fear, spawn collaboration, and consolidate con-
trol of contested territories. Neither rebels nor an incumbent ruler can be success-
ful without the support of civilians. The gain in allegiance of the population is a
primary objective of armed groups in civil war. Contesting parties seek to form a
support network of locals to secure the provision of food, shelter, supplies, infor-
mation, and recruits. The effectiveness of such a network determines the groups’
success in securing contested areas.4
The strategic rationale of targeting the civilian network of the enemy is to
weaken its power by damaging its local base. Petersen and Liaras (2006) argue
that fear and terror are effective weapons often used to bend the enemy’s objec-
tives. Azam and Hoeffler (2002) show that in sub-Saharan Africa terrorizing
civilians is often used to substitute for actual combat. Because they generate
widespread fear, civilian killings not only reduce the support of the enemy but
also increase compliance with the perpetrator. So the degree to which different
armed groups receive support from the population is endogenous to the dynam-
ics of conflict (Kalyvas 2006). On the one hand, armed groups constantly attempt
to secure popular collaboration and deter support to rival groups. On the other
hand, irrespective of their true preferences, most people tend to collaborate with
the group that maximizes their survival opportunities. Indeed, when caught at a
juncture of violence, people naturally tend to put survival considerations and the
protection of their family and property before their political preferences. Thus,
the control of a specific territory by one group often leads to compliance from
local communities. 5 Civilians are then compelled to participate in combat for
strategic considerations rather than ideology. Greater compliance from the civil-
ian population in turn helps in consolidating control, making collaboration and
control mutually reinforcing.
Collaboration is a zero-sum game. It must be exclusive and is enforced with
violence. Defection is severely punished. Information leaks, for example, are often
punished with execution. Killing defectors creates widespread fear, which deters
others from doing so. In addition, noncompliance with a group is equated with
treason and also punished. Neutrality is not an option for the civilian population
who often, and not without risk, prefer to flee the contested region.6 Households
who choose to stay must show allegiance to one group or the other. Civil war is a
polarizing process.
This chapter develops a game-t heoretic model in which two armed groups fight
for the control of a strategically important territory. Local civilians are compelled
to decide which of the contesting groups to support. This decision depends on
the combination of material payoffs and coercion offered by each group to the
S t ra t e g i c A t r o c i t i e s 427
civilians as well as on an idiosyncratic political preference that every individual

has toward one of the armed groups. One obvious result of this simplified frame-
work is that when the majority of the population has strong a priori preferences
toward one group, compliance is so massive that such a group is likely to easily
consolidate territorial control.7
I derive an expression for the equilibrium number of civilians killed by either
armed group. The numeric size of civilian deaths depends on the equilibrium size
of the support networks of the armed groups and the military capacity of each
of them. A testable comparative static is the effect on the number of killings of
a change in the balance of power of the two armed contestants. I rationalize the
conditions under which such change in the conflict environment generates an
increase in the death toll.
Because the model describes a subnational phenomenon that cannot be con-
trasted with cross-national variation, I test the model’s predictions using micro-
level conflict data across circa 1,000 Colombian municipalities for the period
1988–2005.
The remainder of the chapter is organized as follows. Section 19.2 presents the
analytical framework and the main testable predictions. Section 19.3 discusses
the case study of Colombia, a country that has experienced civil war for over fifty
years and where about one-third of all deaths are civilians killed directly and
intentionally. I exploit longitudinal variation across Colombian municipalities to
test the model. Section 19.4 concludes.
19.2. The Model

I develop a model to study the pattern of civilian killings given their strategic
role regarding the formation of networks for the provision of supplies, shelter,
and information to armed groups. In the model, armed groups fight for the con-
trol of a strategically important territory, and it is the allegiance of the civilian
population that determines the relative success in achieving territorial control.
Indeed, as recognized by Russell (1974), no insurgency can be successful with-
out the support of civilians and no incumbent can retain power without it.
19.2.1. The Setup of the Model

Consider a territory of some strategic value, V, without state presence, which is
disputed by two armed groups, the rebels and the militia (henceforth R and M,
respectively).8,9 The probability that group j secures control over the contested
territory is determined by the size of its civilian support network: Nj, j ∈ {R, M}.
Compliance by the civilian population is enforced through the use of vio-
lence: Each armed group targets the supporters of its enemy. The probability (pj)
428 Case Studies II
that a given civilian, i, gets killed by group j given her support to j’s enemy (-j) is
determined by the relative bellicose effort of j and -j. Let r be the fighting effort
of R, and m that of M, with r, m ≥ 0. The probability that civilian i is killed by the
rebels given her support to the militia is then determined by a standard contest
success function of the form:10
ïìï λr ; {r , m} > 0
ï λr + γm
pR = ïí , (1)
ïï 1 ; {r , m} = 0
ïïî 2
where λ>0 is the effectiveness of the fighting effort of the rebels and γ>0 is that of
the fighting effort of the militia. The effectiveness parameters λ and γ can be inter-
preted as the “power” of R and M, respectively. In this model I assume that λ and γ
are exogenous. Note that pR is an increasing and concave function of the ratio: λr/
γm. In turn, the probability that civilian i is killed by the militia given her support
to the rebels, pM, is equal to 1-pR.
Armed groups simultaneously choose their fighting effort to maximize the size
of their support network and thus the probability of controlling the contested ter-
ritory and appropriating the prize V. Hence the objective function of group j is
max Nj(r , m)V − c j(ej) (2)

ej ,
where
ì r; j = R
ï
ej = ï
ím; j = M
ï
ï
î
and cj (ej) is the cost for group j of its fighting effort, ej.
There is a continuum of mass 1 of civilians, each of whom must decide which
group to support. Civilian i’s payoff from supporting the rebels is linear and addi-
tive in her expected material compensation and in an idiosyncratic (nonnegative)
private payoff, σ iR, representing the individual’s bias toward supporting the reb-
els.11 This can be represented with the utility function
U Ri = (1 − pM )y + σ Ri , (3)
where pM is the probability of being killed by the militia—who target i because of

her support to the rebels—and y is i’s income, which is assumed exogenous and
the same for every civilian, and which can be thought of as the average compensa-
tion to peasants in the contested (rural) territory.
Similarly, individual i’s utility from supporting the militia is
U iM = (1 − pR )y + σ iM , (4)
where pR is the probability of being killed by the rebels—who target i because of

her support to the militia—and σMi is the nonnegative idiosyncratic reward derived
from supporting the militia.
For completeness, consider a third option whereby i does not support either
armed group. In this case, the utility of not taking part in the conflict is given by:
U Ri , M = (1 − pR )(1 − pM )y + σ R i, M .
This formulation captures the idea that neutrality is a very risky strategy in civil
war. Collaborators to one armed group are automatically seen as noncollabora-
tors of the rival group and hence targeted by the latter (Kalyvas 2006). In this
sense, the choice of neutrality implies becoming a target of both armed groups.
For simplicity, I assume that the neutrality alternative is strictly dominated by
supporting either armed group for every civilian i. This is equivalent to assuming
that σ Ri, M cannot take extraordinarily high values. In other words, the private
utility of not taking part in the conflict is limited by survival considerations.
Opposition to all armed actors is of course observed in reality. Among other
reasons, in order to avoid the risk inherent to not aligning with any armed group,
a great number of civilians usually flee from territories under dispute. According
to the UN High Commissioner for Refugees (UNHCR 2006), in the 2000s the
country with the largest number of internally displaced people was Colombia.
Iraq and Sudan ranked second and third, respectively. The Colombian nongov-
ernmental organization Consultoría para los Derechos Humanos y el Desplazamiento
(CODHES) estimates that between 1988 and 2004 over three million people had
internally migrated. The model I present here abstracts from this phenomenon
and implicitly deals with the situation of civilians who choose not to flee, and stay
after displacement has taken place. The fact that in the model those who stay are
forced to take a side in the conflict is consistent with the empirical observation
that collaboration must be exclusive.12
The relative importance of pj and σ ij, j = {R, M} depends on the stage of the civil
war at which fighting takes place. The average civil war lasts about sixteen years
(Fearon 2004). Initial ideological convictions may be replaced by the accumu-
lated hatred created by long-lasting conflict. As a result, the relative importance
of σ ij may be offset over time by an increase in coercion.
Given equations (3) and (4), civilian i will support the rebels if
(1 - pM )y > (1 - pR )y + σ i which, using (1) and the fact that pM = 1—pR, can
be written as
æ λr − γm ö÷
σ i < yçç ÷, (5)
çè λr + γm ÷÷ø
430 Case Studies II
where
ìï> 0; σ iM > σ Ri
ïï
σ i = σ iM − σ Riïí= 0; σ iM = σ Ri .
ïï
ïïî< 0; σ iM < σ Ri
The σi parameter is crucial in the analysis since it gives the private component
to (5) and hence allows individuals to differ in their optimal decisions. Notice
that in the absence of σi, for any given set of material compensations and coer-
cion parameters, every civilian would support the same armed group. To focus
on the more interesting case in which civilian support is divided between the two
groups, let σi be distributed across civilians according to the probability density
function f(σ i).
Define σ = (pR - pM )y. From inequality (5) it follows that any civilian i
for whom σ i <  σ will support the rebels, otherwise she will support the militia.
Hence, the fraction of civilians who align with the rebels (NR) is
σ
NR = ∫−∞ f (σ i) dσ i , (6)
while a fraction NM = 1−NR aligns with the militia.

Note that f(σ i) depends on how attached the local population is to the cause of
one group or the other. In an extreme case, for a particular civilian i, σ i would be
sufficiently negative (positive) that i would strongly support the rebels (the mili-
tia). In order to make i switch sides, the enemy would have to supply extremely
high threats {r, m}.
Polk (2007) argues that the very presence of foreigners who attempt to control
a specific region stimulates a strong sense of group cohesion among natives that
often materializes in strong insurgency movements. To Polk, the Vietnam expe-
rience is the ultimate example of how, regardless of the number of soldiers and
civilians killed, the amount of money spent, and the power and sophistication of
the weapons, foreigners cannot militarily defeat an insurgency that is supported
by the majority of the people, except perhaps by genocide. Other examples range
from the Spanish guerrillas against the French in the early nineteenth century to
the Mau Mau rebellion in Kenya in the 1950s.
In sharp contrast with the corner solution in which a foreign power is ulti-
mately defeated by a cohesive resistance, in most instances σ i is likely to be suffi-
ciently small for the majority of the population. Only a small share of civilians will
be tightly attached to one party or the other. In fact, the majority of people tend
to be only weakly committed to any one specific group. That is, the average non-
combatant does not support any cause with sufficient conviction to be willing to
make big sacrifices (including death) for the cause. In this case of weak inherent
preferences, military considerations can be more important than actual political
preferences at driving actions of what group to show allegiance to. In fact, irre-
spective of preferences, equilibrium behavior in terms of support to a given group
can change.
Switching sides is common in civil war. For instance, entire Algerian communi-
ties in the early 1990s defected from the Armed Islamic Group of Algeria (known
by its French acronym as GIA) to join the militias (Kalyvas 1999). Widespread
fear created by the Rondas Campesinas led to massive desertions of insurgents
from the Shining Path in Peru during the 1980s. The Zapatista revolution in early-
twentieth-century Mexico was only able to overthrow Porfirio Diaz when 26,000
men deserted from the constitutional army and joined the rebels (Wolf 1973). In
Colombia, 46 percent of the 316 FARC members demobilized by 2002 stated that
they joined the guerrilla force because a salary was promised, or simply because
of fear. Only 12 percent of the subjects claim to have joined the guerrilla for ideo-
logical reasons (Pinto, Vergara, and Lahuerta 2002).13
19.2.2. Civilian Deaths

In this reduced-form framework with an inherent risk of death—civilians who
decide to support one group are targeted by the other—both the rebels and the
militia end up killing civilians. This is consistent with the observation that in civil
wars featuring local contests for territorial control, civilian communities are stra-
tegically targeted by all armed actors. In the model, the total number of civilian
killings is then the sum of killings by the rebels and the militia: KT = KR + KM. This
can be written
KT = pR NM (σ) + pM NR (σ). (7)
This equation summarizes the magnitude and structure of civilian deaths in

an environment of armed struggle for the control of a valuable stateless territory.
I now characterize the equilibrium values of KR and KM as functions of the exog-
enous parameters of the model, notably the fighting effectiveness of the armed
groups, λ and γ.
19.2.3. Timing of the Game

Let G be the game described above. The timing of actions is as follows:
1. Every civilian stationed in the contested territory decides whether to align

with R or M.
432 Case Studies II
2. R and M simultaneously choose their fighting efforts, respectively r and m, to

maximize compliance of civilians with their own group, and hence the prob-
ability of achieving territorial control.
3. All the killing of civilians by armed groups takes place given compliance levels
(NR and NM) as well as killing probabilities pR and pM.
19.2.4. Equilibrium
This is a one-shot sequential game with perfect information. I solve the game by
backward induction starting with the optimal fighting effort chosen by R and M
given the compliance shares NR and NM. In the second stage, armed group j maxi-
mizes the payoff function given by equation (2) where Nj depends on the critical
value σ and f(σ i), the distribution across the civilian population of the net compo-
nent of political preferences on civilians’ payoff. Recall that σ i can take positive as
well as negative values: A civilian i who is ideologically biased toward the rebels
has σ i less than 0, while σ i > 0 reflects bias toward the militia.
I make the following simplifying assumption: Let f(σ i) be uniformly distrib-
uted in the support éê− 1 , 1 ùú with density φ. Notice that the density, φ, is inversely
êë 2ϕ 2ϕ úû
related to the strength of ideology in the population. Larger values of φ shorten
the support of the distribution, which becomes more concentrated around the
mean (σ i = 0) and so individuals are less “ideological” and more responsive to
incentives (which in this case are given by coercion). With this functional form
for f(σ i), equation (6) becomes:
1 æ λr − γm ö÷
NR = 1 − NM = + ϕyçç ÷.
2 çè λr + γm ÷÷ø (8)
Substituting (8) into (2) and assuming for simplicity a linear cost of effort such
that cj(ej) = ej, the first-order conditions (FOCs) for the rebels and the militia,
respectively, are
Vyϕλ éê æ λr − γm ö÷ù
çç
1 − ÷ú = 1
(λr + γm) êêë çè λr + γm ÷÷øúú
û
and
Vyϕγ éê æ λr − γm ö÷ù
çç
1 + ÷ú = 1.
(λr + γm) êêë çè λr + γm ÷÷øúú
û
The FOCs can be expressed as reaction functions of each group’s fighting effort
with respect to the enemy’s effort. In the case of the rebels’ effort, we have
1
r= [(2Vyϕλγm)1/2 − γm] .
λ
By symmetry, the reaction function of the militia effort is

1
m= [(2Vyϕλγr)1/2 − λr] .
γ
It follows that the equilibrium fighting efforts r* and m* are given by
2 Vyϕγλ
r * = m* =
(γ + λ)2
Note that equilibrium efforts are strictly positive; r=m  = 0 is not an equi-
librium in pure strategies. To see this, notice from (1) that  r=m  = 0 yields
pR = pM = ½. For any ε arbitrarily small but positive, setting ej = ε makes pj dis-
continuously jump from pj = ½ to pj = 1. Thus ej = 0 is strictly dominated and

r=m  = 0 cannot be an equilibrium.
One interesting fact about the equilibrium efforts is that these are the same
for the two groups in spite of the fact that their effectiveness per unit of effort
is not necessarily the same (λ ≠ γ). This result is equivalent to the strong form
of Hirshleifer’s (1991) paradox of power, whereby weaker contestants choose a
higher bellicose effort relative to stronger ones in order to offset the advantage of
their enemy.
Turning to the first stage of the game, r* and m* determine the equilibrium kill-
λ γ
ing probabilities by both armed groups. These are p R* = and p *M = .
λ+γ λ+γ
Each civilian takes these as given and, for every value of her idiosyncratic-prefer-
ence parameter, she decides what group to support. In equilibrium, inequality (5),
æλ − γö
which defines whether civilian i supports R or M, can be written as σ i < yçç ÷.
è λ + γ ÷÷ø
Hence the equilibrium compliance shares are
1 æ λ − γ ö÷
N R* = 1 − N M
* = + ϕyçç ÷
2 çè λ + γ ÷÷ø. (9)
The compliance share of group j is increasing in its power and decreasing in

the power of its enemy. This happens because, for a given net idiosyncratic com-
ponent of utility σ i, the probability of being killed by a group given compliance to
its enemy is increasing in the power of the group. Thus, there is a fear effect driven
by survival considerations whereby more powerful groups are able to increase the
size of their support network.
434 Case Studies II
From equation (7), it follows that the number of civilians killed in equilibrium is
æ λ ö÷ æ γ ö÷
KT* = çç ÷N * + çç ÷N , (10)
çè λ + γ ÷÷ø M çè λ + γ ÷÷ø R*
with the equilibrium compliance shares given by (9). Note that the magnitude
and structure of civilian deaths is determined in equilibrium by the parameters
of the model, namely the power of both armed groups (λ and γ), the average
compensation of peasants (y), and the relative importance of incentives vis-à-v is
ideology among the civilian population φ. Of these, the most interesting compar-
ative statics comes from shifts in λ and/or γ. I now turn to a formal analysis of the
implications of such shifts in the balance of power of the contesting groups and
derive testable implications. Before that, I briefly discuss the empirical relevance
of studying changes in the balance of power of fighting actors in the context of
armed conflict.
19.2.5. Empowerment
There are many factors that can alter the balance of power between armed groups
in a civil war. Foreign intervention is perhaps the most notorious. A foreign
power may intervene by providing financial aid or military support to one group.
The involvement of the Soviet Union and the United States in Africa and Latin
America during the cold war era is an example. While the USSR gave military
and financial support to communist insurgencies fighting in most cases against
authoritarian regimes, the United States backed incumbents in their struggle to
contain such insurgencies. Zaire’s then-president Mobutu Sese Seko is a telling
example of a ruthless dictator backed by successive American administrations
because of the country’s strategic value in its anticommunist campaign in cen-
tral Africa. There are also numerous examples of foreign states altering the bal-
ance of power, other than in the cold war context. The governments of Rwanda,
Uganda, and Zimbabwe allegedly financed armed movements in the DR Congo
at least since the fall of Mobutu in 1996. More recently, the Colombian con-
flict has increasingly been shaped by the participation of international actors
and both the government and the rebels have benefited from external support.
While the Colombian government is the largest recipient of US military aid in
the Western hemisphere, the Chavez/Maduro regime in neighboring Venezuela
has been accused of protecting and financing the Revolutionary Armed Forces of
Colombia (FARC).
Intervention from abroad can also take the form of donations from diaspo-
ras. Examples abound and range from the support of the Tamil diaspora in North
America to Sri Lanka’s Tamil Tigers to the support of the Albanian diaspora
in Europe to the Kosovo Liberation Army. For many years the main source of
finance of the Eritrean People’s Liberation Front was its huge diaspora (World
Bank 2003). Irish Americans were suspected of contributing to the campaign of
the Irish Republican Army (IRA) in Great Britain.
The balance of power in civil war can also be altered by the presence of natural
resources, through several mechanisms. First, fluctuations in the value of natural
resources used to finance armed groups, like petroleum and raw diamonds, will
shape the groups’ fighting power. Second, insurgent organizations in Angola, DR
Congo, Liberia, and Sierra Leone have acquired nonnegligible resources by sell-
ing the future rights on the war booty (Ross 2005). Finally, armed groups can
boost their power by merging into unified armies, thereby sharing intelligence
and taking advantage of military economies of scale. The large upsurge of the
late 1990s in militia activity in Colombia documented by Restrepo, Spagat, and
Vargas (2004) originated in the collusion of a large number of militias from differ-
ent parts of the country under an umbrella organization.
How is the security of the civilian population affected by shifts in the relative
balance of power of armed groups? The model proposed in this chapter helps shed
light on this question. Because local rebels and militias strategically target civil-
ians, the consequences of empowering one of them may be deadly. But it also may
not be. In the model, increased military capacity boosts the ability of generating
widespread fear by killing the civilian infrastructure of the enemy. This direct
effect of military empowerment maps into higher tolls of civilian deaths. However,
an empowered army that has built a reputation of executing enemy supporters is
more frightful, and so induces some civilians to shift their support toward it. Thus
there is an indirect fear effect of empowerment that works in the opposite direction
and saves lives. Studying the conditions under which one effect dominates the
other is important for the design of policies aimed at protecting civilians.
The ultimate effect on civilian deaths depends on the size of the direct effect
relative to the fear effect. I examine the conditions under which one effect domi-
nates the other by looking at equation (10). Without loss of generality, assume
that γ increases exogenously relative to λ, that is, there is an episode of empower-
ment of M relative to R. The total change in civilian deaths due to an increase in
¶K
the power of the militia, T , can be decomposed into the change in rebel victims,
¶K R ¶γ
, and the change in civilians killed by the militia, ¶K M . The first component
¶γ ¶γ
can be written as
∂K R ∂N ∂p
= pR M + NM R . (11)
∂γ ∂γ ∂γ
Using the equilibrium values for p R* and N*M it can be seen that the first compo-
nent on the right-hand side (RHS) of this equation is positive since p R* is positive
436 Case Studies II
and ¶N*M = 2ϕyλ > 0. This is because a more powerful militia attracts more
¶γ out(λ of
compliers + γfear.
)2 Indeed, dying becomes more likely for rebel supporters.
The second component on the RHS of equation (11) is, however, negative since
∂pR* λ
N*M > 0 but =− < 0 which happens because a greater fighting effec-
∂γ (λ + γ )2
tiveness of the militia makes it less likely to be killed by the rebels; see the contest
success function defined in equation (1). The net effect of militia empowerment
on the victimization of civilians by the rebels then depends on whether the fear-
driven increase in NM offsets the drop in pR. Hence we have:
Lemma 1. The effect of an increase in the power of the militia on the number
of civilians killed by the rebels is:
¶K R
• > 0 if pR ∂NM > −NM ∂pR , or γ < λ. So KR increases as long as the
¶γ ∂γ ∂γ
power of the militia is less than the power of the rebels.
¶K R ∂N M ∂pR
• < 0 if pR < −N M , or γ > λ. So KR decreases as long as the
¶γ ∂γ ∂γ
power of the militia is greater than the power of the rebels.
¶K R ∂N M ∂pR
• = 0 if pR = −N M , or γ = λ. So KR remains unchanged as long
¶γ ∂γ ∂γ
as the power of the militia is equal to the power of the rebels.
Proof. See Appendix.

In turn, the effect of an increase in the power of the militia on the number of
their own killings of civilians, ¶K M , can be written as
¶γ
∂K M ∂N ∂p
= pM R + NR M . (12)
∂γ ∂γ ∂γ
Using the equilibrium values for p *M and N*R, it can be seen that the first
component of the RHS of equation (12) is negative since p *M is positive and
∂NR* 2ϕy
=− < 0. This is because some rebel compliers shift sides and sup-
∂γ (λ + γ )2
port the empowered militia. The second component on the RHS is, however, pos-
∂p *M λ
itive since N*R > 0 and = > 0, which happens because the militia
∂γ
2 (λ + γ )
becomes more lethal. The net effect of militia empowerment on their own vic-
timization of civilians then depends on whether the fear-d riven decrease in their
target pool, NR, offsets their higher killing probability, pM. Hence we have:
Lemma 2. The effect of an increase in the power of the militia on their civilian
killings is:
• ¶K M > 0 if −p ∂NR < N ∂pM , or γ < λ. So the direct effect of militia

¶γ M R ∂γ ∂γ
empowerment dominates and KM increases.
¶K M ∂NR ∂pM
• < 0 if −pM > NR , or γ > λ. So the fear effect of militia
¶γ ∂γ ∂γ
empowerment dominates and KM decreases.
¶K M ∂NR ∂pM
• = 0 if −pM = NR , or γ = λ. So the two effects cancel out and
¶γ ∂γ ∂γ
KM remains unchanged.
Proof. See Appendix.

Combining lemmas 1 and 2, we have:
Proposition. The effect of an increase in the power of the militia on the total
number of civilians killed is:
• If γ < λ then KT increases because both KR and KM increase.

• If γ > λ then KT decreases because both KR and KM decrease.
• If γ = λ then KT remains unchanged because both KR and KM do so.
¶K ¶K M
The fact that ¶γR and ¶γ are positive if and only if the rebels are relatively
strong, γ < λ, implies that greater power of the rebel group is associated with a
larger parameter space for which empowerment of the militia will result in a
higher civilian death toll. This result is not obvious a priori. Intuitively, this hap-
pens because a bigger λ implies that civilians who support the militia are more
likely to be killed, and hence the indirect effect of an increase in γ—whereby the
support to the empowered militia increases—w ill be weakened by the fact that
the rebels themselves are also more effective so militia collaborators will be tar-
geted by a more powerful enemy. Thus, the weakening of the indirect effect is due
to the fear that the rebels generate among civilians. In the next section, I test this
prediction using longitudinal data of conflict-related violence in Colombia.
19.3. Empirical Evidence: Empowerment

of Colombian Militias
19.3.1. Background
19.3.1.1. Colombia’s Conflict
Colombia’s civil war involves rebel insurgencies, government forces, and illegal
militias. By most accounts the civil war has lasted for over four decades. Scholars
identify its origin in La Violencia, a period of intense violence between the two
traditional political parties, from 1946 to 1966. Insurgent groups were formed in
the early 1960s as peasant self-defense organizations originally aligned with the
Liberal Party. Two of them survive today as the main guerrilla organizations: the
FARC and the National Liberation Army (ELN) with about 20,000 and 4,000
combatants, respectively. Allegedly, the main objective of these groups is taking
438 Case Studies II
over political power, but their actions have increasingly relied on terrorism. For
instance, the two most important sources of finance for rebel groups from the
early 1990s onward are the illicit drug business and the kidnapping of civilians.
Drugs are a major source of finance especially for the FARC, which is known to
tax coca crops and to control the production, processing, and export of cocaine
and heroin. In terms of bellicose activity, the most common guerrilla actions
involve disruption of the economic infrastructure (e.g., attacks on oil pipelines),
attacks on government military positions, and bombings and roadblocks.
The other major active armed actors in the conflict are illegal militias, known
as paramilitary forces. They are said to have had over 12,000 members at the peak
of their strength. The first militias were organized by the country’s armed forces,
the military, during the late 1970s thanks to a law that permitted the formation
of civilians in armed self-defense organizations, which were encouraged to fight
against the insurgents. Subsequently, rural elites formed private armies that
emerged on a widespread scale during the 1980s when drug lords became land-
owners and faced extortion from the guerillas. These militias were banned in 1989
but kept operating in the shadows, after which the Colombian conflict technically
became three-sided. However, in recent years the vast majority of the fighting has
involved the guerillas against the military. Paramilitaries try to avoid direct com-
bat with either guerrilla or government forces. Rather, the militia specializes in
selective killings of civilians whom they presume to support the rebels. Seven out
of ten civilians killed in Colombia from 1988 to 2005 were victimized by armed
militias, often with the alleged acquiescence of the military. Over 70 percent of all
uncontested attacks carried out by militias have been massacres, with incursions,
checkpoints, and kidnapping taking up the slack (Restrepo and Spagat 2004).
19.3.1.2. Colombian Militias

In 1997, several disparate local militias joined in an umbrella alliance, called the
United Self-Defense of Colombia (AUC), which contributed substantially to the
dramatic expansion of conflict activity during the late 1990s. Two-thirds of the
7,000 civilians killed by paramilitary groups from 1988 to 2005 died in one-third
of the time, from 1997 to 2002, that is, during the lifespan of the AUC. At the
same time, Colombia witnessed a rapid geographic expansion of militia presence.
The year 2000 was the peak of paramilitary activity with attacks in 120 munici-
palities, four times the average geographic incidence count in the pre-AUC period,
1988–1997. This is consistent with the idea that civilians are killed as means of
consolidating control over new strongholds. In the words of AUC leader Carlos
Castaño: “I made of this conflict a high intensity war that now involves the people
it must involve: the hidden allies of the guerrilla” (quoted in Aranguren 2001, 116).
Indeed, the AUC publicly claimed that at least two-t hirds of the guerrilla
members were civilian supporters rather than proper combatants (Aranguren
2001). The organization also argued that while human rights constraints prevent
the Colombian military from involving civilians in the conflict, the survival
success of the rebels is determined by the capacity of coercing rural communi-
ties into supporting them. According to AUC rhetoric, an effective counterin-
surgency strategy must give priority to blocking these guerrilla supporters.
Taking advantage of a presidential transition, in 2002 the AUC leadership
estimated the organization had enough leverage to cut a good deal on an even-
tual peace process (see Romero 2003). In December that year, the AUC com-
mand unilaterally declared a cease-fi re as a gesture to foster negotiations with the
administration of President Alvaro Uribe. Negotiations started in January 2003
and lasted about three years, ending with a massive demobilization of militia
combatants in 2006 and 2007.14
19.3.2. Data
The conflict dataset used in this chapter was first introduced by Restrepo, Spagat,
and Vargas (2004). Since 2005, it is being maintained by CERAC (the Spanish
acronym for the Conflict Analysis Resources Center), a Colombia-based think-
tank. It is an event-based conflict dataset on Colombia covering the period 1988–
2005. For every event the dataset records its type, date, location, perpetrator, and
victims involved. The dataset is described thoroughly in Restrepo, Spagat, and
Vargas (2004) and in Dube and Vargas (2013). Here I provide a succinct account
of the data collection process.
The dataset is constructed on the basis of events listed in the annexes of peri-
odicals published by two Colombian human rights nongovernmental organiza-
tions (NGOs): CINEP and Justicia y Paz. Most of the event information in these
annexes comes from two primary sources, a network of priests from the Catholic
Church—w ith representation in almost every municipality in Colombia—and
over twenty-five newspapers with national and local coverage. The inclusion of
reports from the Catholic priests, who are often located in rural areas that are
unlikely to receive press coverage, broadens the municipality-level representa-
tion. Based on these sources, the resulting data includes every municipality that
has ever experienced a conflict-related action (either a unilateral attack or a clash
between two groups). A stringent data-related regime guarantees its quality and
representativeness. As a first step, a large number of events is randomly sampled and
compared against the original source, to check for correct coding from the annexes
into the dataset. Second, a different random sample is looked up in press archives
to confirm whether incidents should have been included in the annexes. This step
checks the quality of the raw information provided by the NGOs, which turns out
to be quite high. Third, the largest events associated with the highest number of
deaths are carefully investigated in press records. Finally, without double-coding,
the dataset is complemented with additional events provided in reports by human
rights NGOs and by Colombian government agencies.
440 Case Studies II
I use several variables from the CERAC dataset throughout the empirical
analysis that follows. These include (1) the number of civilian deaths, which is
the dependent variable; (2) the number of combatants killed; (3) the number of
massacres of civilians;15 and (4) the number of attacks by the rebels, which I treat
as the baseline proxy of rebel power, λ (see Table 19.1). Other proxies of rebel
power are a dummy variable for the presence of rebel fronts (obtained from the
Colombian Ministry of Defense), a dummy variable for the presence of coca crops
in 1994 (from the Colombian National Police Department), and a dummy vari-
able for whether a municipality is a strategic stronghold of FARC, the country’s
largest rebel group (from Giraldo, Lozada, and Muñoz 2001)—see Table 19.1.
The log of population is used as a scale control, and it is taken from Departamento
Administrativo Nacional de Estadística (DANE), Colombia’s Census Bureau.
Table 19.1 Descriptive Statistics
Type Mean Standard Source

Deviation
1. Conflict Variables
1.1. Intensity Measures
Civilian casualties count 0.61 3.3 CERAC
Number of massacres count 0.06 0.34 CERAC
Combatant casualties count 0.01 0.31 CERAC
1.2. Proxies of Rebel

Power
Attacks by rebels count 0.61 1.78 CERAC
Presence of FARC dummy 0.48 0.5 Colombian Army
fronta
Presence of ELN dummy 0.23 0.42 Colombian Army
fronta
Presence of coca crops dummy 0.06 0.23 National Police
in 1994 Department
Municipality is FARC dummy 0.07 0.25 Giraldo, Lozada,
strategic strongholdb and Muñoz (2001)
2. Municipality Characteristics (time-varying)

Log of population Ln(count) 9.66 1.05 DANE
a
FARC and ELN are the two largest rebel groups.
b
See text for discussion.
19.3.3. Empirical Analysis

19.3.3.1. Benchmark Results
Recall that an empowerment episode triggers two opposite forces: an indirect effect
and a direct effect. I have argued that, everything else equal, when the militia gets
more powerful, then the greater the power of the rebels the weaker is the indirect,
or fear, effect, which operates by preventing rebel supporters from switching sides.
Hence civilian deaths will be higher in places where rebels are stronger. Table 19.2
shows evidence in this respect. I regress the number of civilians killed on the interac-
tion between λ (proxied by the number of rebel attacks in a given municipality-year)
and a dummy variable that takes the value of 1 from 1997 to 2002, representing the
period when militias colluded into the AUC.16 The model predicts that the lethal-
ity of the rebels should be positively associated with civilian killings, especially after
the formation of the AUC (which shifted the relative balance of power in favor of
the militia). In other words, an increase in the power of the militia worsens civilian
security, especially in places where the rebels are more powerful. I therefore expect
the coefficient on the interaction between the proxy of λ and a dummy for the militia
collusion period to be positive and statistically significant.17
Column 1 of Table 19.2 reports estimates from an Ordinary Least Squares
(OLS) regression of the number of civilian deaths on the interaction of the number
Table 19.2 Benchmark Results
Dependent Variable: Civilian Casualties

Ordinary Least Squares
(1) (2) (3)
Rebel power x militia 0.620*** 0.136*** 0.484***

empowerment period (0.139) (0.052) (0.108)
Time-varying controls yes yes yes

State-level time trends yes yes yes
Municipality fixed effects yes yes yes
Year fixed effects yes yes yes
Observations 13,740 13,740 13,740
Notes: Panel-robust standard errors are in parentheses. Regression disturbance terms are
clustered at the municipality level. The (time-varying) log of population (coefficient estimates not
reported) is used as a scale control.
* Significant at the 10% level.
** Significant at the 5% level.
*** Significant at the 1% level.
442 Case Studies II
of rebel attacks and the six-year period of militia empowerment (1997–2002). The
noninteracted version of the two variables is not included since the regression
model includes municipality and time fixed effects, which control, respectively,
for any time-invariant municipal characteristics that may be correlated with civil-
ian victimization, and for any arbitrary aggregate shock that may affect victimiza-
tion.18 In addition to the two-way fixed effects I include state-specific time-trends.
This makes the test of the model a very stringent one because the roughly 1,100
municipalities are part of just thirty-t wo states. The trend controls for serial cor-
relation over time and across municipalities in the same state. To account for the
time-varying municipal scale I control for the (log of) municipal population.19
Results (statistically significant at the 99 percent level of confidence) show that
the power of rebels as measured by the number of unilateral attacks by gueril-
las is positively associated with the number of civilian deaths during the militia
empowerment period.20 Columns 2 and 3 break down the number of civilian
deaths between civilians killed by the rebels (column 2) and civilians killed by the
militia (column 3). The coefficients on the interaction of interest are positive and
statistically significant in both cases. This is consistent with the model’s proposi-
tion that the increase in the total number of civilian deaths (KT) is due both to an
increase in rebel killings (KR) and an increase in militia killings (KM).
19.3.3.2. Accounting for the Count Nature of the Data

Applied econometricians disagree on the relative benefits of fitting linear models
relative to nonlinear ones. On the one hand, some argue that count data, as used
in this chapter to measure the outcome of interest, are highly non-normal and
hence not well estimated by linear regression (e.g., Winkelmann 2000). On the
other hand, Angrist and Pischke (2008) emphasize that given uncertainty about
the true data-generating processes, (incorrect) nonlinear models result in poten-
tially much larger biases than OLS, which is the best linear approximation to any
data-generating process.
I explore the robustness of the benchmark results to fitting nonlinear count
data models. The Poisson model is usually incorporated to account for the data-
generating process that produces counts.21 However, this approach has two
potential problems. First, the Poisson distribution sets the population mean equal
to the variance. But it is often the case that count variables are overdispersed, that
is, they have a variance greater than the mean. Here, the variance of the depen-
dent variable (the municipality-year count of civilians killed) is about six times
larger than the mean.22
The second problem of the Poisson model is that count data may be highly
left-skewed, having “excess zeros.” Consider the process that could lead to a
count being zero. An armed group may be stationed in a municipality but abstain
from targeting the civilian population. Another municipality may lack an armed
presence altogether and hence present zero killings. In the latter case zero killings
are a certain outcome, thus the number of zeros may be inflated, and killings in
municipalities free of armed groups cannot be explained in the same way as kill-
ings in regions with armed presence. Here, in 90 percent of the 12,420 observa-
tions, the dependent variable is zero, so the data is indeed highly left-skewed.
This discussion suggests that the Poisson assumptions are probably not
met and different models for count data analysis should be considered. Hence,
Table 19.3 fits a (fixed effects) Negative Binomial distribution, which can be
regarded as a generalization of the Poisson with one additional, ancillary param-
eter that allows for the variance to be greater than the mean. The model estimated
in column 1 produces an estimate of such a parameter that confirms the existence
of overdispersion and validates the Negative Binomial over the Poisson. The coef-
ficient on the interaction of interest is positive and statistically significant at the
1 percent level.
The complication of excess zeros is corrected in column 2 by fitting a zero-
inflated Negative Binomial model with two otherwise separate models. A Logit
is used to predict the cases in which zero is a certain outcome, and a Negative
Binomial distribution fits the counts having noncertain zeros. For the former,
I use as predictors both the municipal population and a battery of time-invariant
municipality-specific variables. The Vuong test (not reported) suggests that the
Table 19.3 Results from Distributions for Count Data
Dependent Variable: Civilian Casualties

Negative Binomial Model
(1) (2)
Rebel power x militia 0.056*** 0.044**
empowerment period (0.009) (0.020)
Time-varying controls yes yes
Municipality fixed effects yes no
Year fixed effects yes no
Excess-zeros correction no yes
Observations 4,995 12,420
Notes: Standard errors are in parentheses. Regression disturbance terms of column (2) are clus-
tered at the municipality level. Time-invariant controls used in the Logit model underlying the zero-
inflated negative binomial model of column 2 include as regressors the log of population as well as the
poverty rate, average education, health conditions, institutional quality, whether the municipality is
urban or rural, and geographic characteristics like average rainfall, average temperature, and altitude.
444 Case Studies II
zero-inflated model is significantly better than the standard Negative Binomial

of column 1. Finally, the coefficient of interest turns out to be robust to this gen-
eral model that accounts for both overdispersion and excess zeros and it is indeed
positive and statistically significant, this time at the 95 percent level of confi-
dence. As for its marginal effect, a municipality going from the mean number of
rebel attacks (0.6) in the pre-AUC period to one additional rebel attack during the
empowerment of the militia (1997–2002) will have, on average, double the num-
ber of civilians killed.
19.3.3.3. Other Proxies of Rebel Power

Proxying the power of an armed group that fights for territorial control with
the number of attacks it carries out on such territory may be problematic. It could
be argued that the opposite is in fact true: The more contested a territory is, the
more bellicose activity there will be from contesting groups, but the consolida-
tion of territorial control is associated with a rather peaceful period ex post. Table
19.4 acknowledges such concern and looks at the robustness of the benchmark
OLS fixed-effects model presented in Table 19.2 (column 1) to the use of different
proxies of λ. Columns 1 and 2 use a dummy variable for whether a given munici-
pality is reported (by military intelligence) to have, respectively, a FARC or ELN
front operating in a municipality, independently of whether there has been any
recent active bellicose activity by it or not. The idea is that, everything else equal,
rebels hold relatively more power in places where they are reported to have a
permanent basis. The results suggest that, with 99 percent confidence, districts
where there is a FARC (ELN) front had an average of 0.6 (0.7) additional civilian
killings in the period 1997–2002.
Columns 3 and 4 of Table 19.4 look at two other proxies of λ. The coefficient
of interest in column 3 is the interaction between the dummy for the militia
empowerment period and a dummy for whether a municipality grew any coca
plants in 1994 according to a survey conducted that year by the National Police
Department. This is arguably highly correlated with locations where rebels have
been traditionally powerful. Indeed, after the Medellin and Cali drug cartels
were dismantled in the early to mid-1990s, the rebels took over control of the
production-trafficking chain. The militia was not involved in the coca business
until well after the formation of the AUC in 1997 (Aranguren 2001). The interac-
tion term is positive and statistically significant, so the baseline results are also
robust to the third alternative measure of rebel power, suggesting that, everything
else equal, municipalities that had grown coca in 1994 had 0.7 additional guerrilla
killings in the period 1997–2002. In turn, column 4 focuses on the effect of rebel
power proxied by an indicator of municipalities that, according to the thorough
study of rebel territorial control in Colombia by Giraldo, Lozada, and Muñoz.
(2001), are strongholds of the FARC, the largest rebel group in Colombia. These
Table 19.4 Robustness to Measures of Rebel Power
Dependent Variable: Civilian Deaths

Ordinary Least Squares
Proxy for FARC ELN Coca in FARC Strategic

Rebel Power: Front (1) Front (2) 1994 (3) Stronghold (4)
Rebel power x militia 0.587*** 0.702*** 0.739* 1.335*
empowerment period (0.213) (0.230) (0.442)* (0.791)
Time-varying controls yes yes yes yes
Time trends yes yes yes yes
Municipality fixed yes yes yes yes
effects
Year fixed effects yes yes yes yes
Observations 13,740 13,740 13,740 13,740
Notes: Panel-robust standard errors are in parentheses. Regression disturbance terms are clus-
tered at the municipality level. The log of population (coefficient estimates not reported) is used as
a scale control, and it is time-varying. Rebel power measures are: the presence of a FARC company
in column 1; the presence of an ELN company in column 2; whether there were illegal coca crops
grown in a given municipality in 1994 in column 3; whether a given municipality is part of a FARC
strategic stronghold according to Giraldo, Lozada, and Muñoz (2001) in column 4.
are municipalities that (1) belong to the strategic strongholds of the rebel group—
mainly located in the states of Caqueta and Meta—where the rebel command is
thought to be located; (2) secure access to strategically important roads and rivers
(in Caqueta, Cundinamarca, Huila, and Guaviare); and (3) ensure a steady source
of financing for the rebel group. Column 4 suggests that these municipalities have
had on average 1.3 additional civilians killed in the period of the empowerment of
the militia relative to the pre-1997 period.
The baseline results, then, are robust to using a heterogeneous set of proxies for
rebel power, in addition to time-varying scale controls, state-specific time trends,
and year-and municipality-specific fixed effects, which account respectively for
time and municipality unobserved heterogeneity not captured by the controls.
19.3.3.4. Additional Robustness Checks

Table 19.5 reports additional robustness checks. Column 1 checks whether
the results are robust to changing the dependent variable. Instead of looking at
the number of civilians killed, these columns report the effect of the interaction
446 Case Studies II
between λ and the militia empowerment period on the number of civilian mas-
sacres. The effect is positive and statistically significant, suggesting that civilians
were massacred more often in places where the rebels were more powerful.
Columns 2 and 3 perform falsification tests. A placebo empowerment period,
one that also lasts six years like the true AUC lifespan (1997–2002), but covers a
period prior to 1997, is used instead to interact with the proxy of λ. The coefficient
of the resulting interaction between this time dummy and the number of guerrilla
attacks on the number of civilians killed is statistically insignificant.
Further, the model predicts an association between the power of one of the
contesting groups and the number of civilians killed by the other. While combat-
ants die as a result of clashes with contesting illegal armed groups or government
forces, civilians are targeted with the specific objective of consolidating territorial
control. Indeed, after controlling for municipality-specific and year-fi xed effects,
state time trends, and time-varying characteristics of the municipality, a different
Table 19.5 Additional Robustness Checks
Ordinary Least Falsification Tests

Squares
Dependent Placebo Dependent Variable:

Variable: Number Empowerment Killing of Combatants
of Massacres (1) Period (2) (3)
Rebel power 0.061*** –0.044 0.000
x militia (0.011) (0.106) (0.004)
empowerment
period
Time-varying yes yes yes
controls
Time trends yes yes yes
Municipality yes yes yes
fixed effects
Year fixed effects yes yes yes
Observations 13,740 13,740 13,740
Notes: Panel-robust standard errors are in parentheses. Regression disturbance terms are
clustered at the municipality level. The (time-varying) log of population (coefficient estimates not
reported) is used as a scale control. Column 1 looks at the robustness to using the number of militia
massacres of civilians as the dependent variable. Columns 2 and 3 report results from falsification
tests. Column 2 uses a placebo empowerment period for the militias. Column 3 looks at militia
killing of combatants as the dependent variable.
dependent variable (namely the number of combatants killed) is not associated in

a statistically significant way with the interaction of interest.
19.4. Conclusion
Most civil wars witness the killing of noncombatants by both state and nonstate
parties. The objective behind this practice seems to be the weakening of the
enemy by eliminating its civilian support network and to take military advantage
of it. I capture this idea in a model where civilians stationed in a contested terri-
tory are killed by the party they do not comply with. In this context, I examine
under what circumstances the empowerment of one of the groups will result in
more or fewer civilian deaths.
Two opposite forces are captured by the model. On the one hand, assuming
that more power translates into a greater killing capacity, there is a direct effect
whereby more civilians will die as a result of the greater killing capability. On
the other hand, this same mechanism dissuades some civilians from supporting
the enemy (the fear effect), so it is not clear whether the total number of civilians
killed increases or decreases. The model predicts that greater power will result in
more civilian killings only if the enemy is itself powerful enough.
Using an event-based dataset that permits exploiting the subnational variation
of the Colombian armed conflict, I find empirical support for this prediction. The
empowerment of illegal right-w ing militias resulted in higher killing of civilians in
places where the rebels are more powerful. This result is robust to various econo-
metric specifications, sets of controls, measures of power, and dependent variables.
Future work is needed to enrich the model with a more complex economic
environment that allows testing predictions on the pattern of killings in places
that vary in terms of their characteristics. For instance, it is possible to add to
the model a term that captures the baseline share of the population that sup-
ports one group over the other and then tests nuanced predictions that this may
give, using preempowerment electoral data at the local level to compute the
share of people who voted for left-w ing parties. Another extension to the theo-
retical model could be adding a production economy to the environment so that
both the income received by citizens and the value of capturing a municipality
is endogenized.
Acknowledgments
I am grateful to Ernesto Dal Bó and James Robinson. I thank Robert Bates,
Alexandra Hartman, Ron Smith, and participants at the Fourth Annual HiCN
Workshop, and seminars at CEDE-Universidad de los Andes and Universidad del
Rosario. I thank Hend Alhinawi for excellent research assistance.
448 Case Studies II
Appendix
Proof of Lemma 1
∂NM ∂pR
Equation (11) is positive as long as pR > −NM . Substituting in the equi-
∂γ ∂γ
librium values for p*R and N M
* and their derivatives with respect to γ, we obtain
the following condition:
∂K R λ(6ϕy − 1)
>0⇔γ< . (A1)
∂γ 1 + 2ϕy
Further, note that KR(γ = 0) has to be equal to 0: if the militia has no power at
all, it will get no supporters; everybody will support the rebels who then end up
killing no one. Since KR(γ = 0) = ½ - φy, the condition KR(γ = 0) = 0 imposes the
following restriction on the parameters: 2φy = 1. Substituting into (A1) we have:
∂K R
> 0 ⇔ γ < λ.
∂γ
Proof of Lemma 2
¶NR ¶pM
Equation (12) is positive as long as -p < NR . Substituting in the equi-
M ¶γ ¶γ
librium values for p *M and N*R and their derivatives with respect to γ, we obtain
the following condition:
∂K M λ(1 + 2ϕy)
>0⇔γ< =λ (A2)
∂γ 6ϕy − 1
where the right-hand-side equality comes from the restriction KR(γ = 0) = 0 ⇔

2φy = 1 discussed in the proof of Lemma 1. <
Notes
1. Kalyvas (1999) reports all massacre events that occurred in Algeria in that twenty-n ine-
months-long period. Each event is described in terms of its date and location and includes
the number of people killed.
2. CERAC is the Spanish acronym for the Conflict Analysis Resources Center, a Colombian
think tank. It maintains an event-based dataset on the Colombian conflict; see www.cerac.
org.co. More about the dataset will be explained later in the chapter.
3. Lacina and Gleditsch (2005) introduce a longitudinal dataset on battle deaths. The
Uppsala Conflict Data Program has a dataset on civilian casualties in civil war for the
period 1989–2 005; see http://w ww.pcr.uu.se/research/UCDP/.
4. As pointed out by an IRA combatant: “Without the community we were irrelevant. We

carried the guns and planted the bombs, but the community fed us, hid us, opened their
homes to us, turned a blind eye to our operations” (cited in Kalyvas 2006, 91). The same
point is also emphasized by Carlos Castaño, head of the Colombian paramilitary mili-
tias: “Since we could not combat [the rebels] where they were, we chose to neutralize the
people who brought to their camps food, medicine, messages, liquor, prostitutes, and these
types of things. And we realized that we could isolate them and that this strategy would
give us very good results” (quoted in Kirk 2003, 152).
5. According to Cenarro (2002), during the Spanish civil war many leftists living in regions
dominated by right-w ing militias ended up supporting them. In Colombia, collaboration
with the FARC and ELN guerrillas is almost completely a rural phenomenon.
6. On the decision to stay or to leave violence-a fflicted areas, see c hapter 11 in this volume.
7. This corner solution can be illustrated by the fact that most foreign powers find it extremely
difficult to subjugate colonized territories for long. Polk (2007) argues that because most
human beings are territorial, they are seldom willing to accept foreign rulers.
8. The choice of these two actors over the more obvious pair Rebels and Government is that in
most civil wars civilians are not targeted by the government directly, but by illegal militias
instead (although often with the acquiescence and support of the government). However,
this is just a choice of notation. The two armed groups described in this model could just
as well be two rebel groups. This has indeed been the case in specific areas of countries like
Congo, Angola, and Colombia.
9. The model abstracts from the reason why achieving control over the territory (and there-
fore appropriating the prize V) is important. Some such reasons may include the existence
of a valuable natural resource, the control of which would help in financing the armed
struggle; the necessity of consolidating a safe haven for the cultivation of drugs or for the
illegal transportation of arms and supplies; or the strategic proximity of an important city
or enemy’s camp.
10. A contest success function (CSF) is a technology that translates the resources invested
by two or more contenders into the probability of winning a contest (or, alternatively, the
share of the prize that each contender will appropriate). Hirshleifer (1991) was among the
first to use a CSF in conflict models.
11. As in a standard probabilistic-voting model (Lindbeck and Weibull 1987; Persson and
Tabellini 2000), σ R i can be interpreted as parameterization of the ideological bias of indi-
vidual i toward group R. For an application of this approach to the context of the tradeoff
between democracy and fighting, see Chacón, Robinson, and Torvik (2011). This parame-
terization of individual allegiance is also consistent with Petersen’s (2001) account of how
ordinary people became involved in resistance movements in Eastern Europe. According
to Petersen, there is wide variation in patterns of civilian support toward armed groups.
The level of support varies from neutrality to sympathetic feelings to the provision of infor-
mation to full involvement with the group.
12. Horton (2004) describes how in the Nicaraguan civil war, the Contras repressed sym-
pathizers of the Frente Sandinista. As a result some peasants abandoned their farms.
Some others decided to stay and comply with the Contras, withdrawing support from
the Sandinista project. Horton (2004, 175) quotes a peasant from Quilali town: “If you
behaved well you wouldn’t have problems [with the contras]. If not, it was a mess.”
13. However, it is likely that this number is lower-bound because of selection issues. Arguably,
the most ideological of the combatants of a rebel group are underrepresented among those
who demobilize.
14. Although the number of massacres dropped significantly, militia killings of civilians
did not stop. Since 2003, several massacres have taken place. However, the government
argues that most of these violations were carried out by splinter militia groups and AUC
dissidents.
450 Case Studies II
15. Massacres are defined as single killing events resulting in the deaths of at least four people.
16. This proxy of λ may be problematic. In particular, rebel attacks may be negatively related
to their power if such attacks occur in places where they seek to gain control through vio-
lence. Table 19.2 deals with these concerns and shows the robustness of the baseline results
to a number of potential proxies for the unobservable power measure.
17. For conciseness, in the reported tables I only include the coefficients of interest. Estimated
coefficients on control variables have the expected sign and vary in significance. These are
available from the author by request.
18. Results from a Hausman test (not reported) suggest one cannot accept the null hypothesis
that the unobserved heterogeneity is uncorrelated with the covariates. That is, it seems
that the data-generating process is best described by a fixed-effects model instead of a
random-effects model.
19. The specifications also account for the fact that stochastic disturbances are likely to be cor-
related over time within a given municipality, or may have covariances that differ across
regions. Such potential problems of serial correlation and heteroskedasticity, respectively,
are taken care of by clustering the errors at the municipal level. Hence, the reported stan-
dard errors in this table (and all subsequent tables) are panel-robust.
20. Note that I do not make any claim of causality since the power of rebels is likely to be
endogenous. Despite the fact that the fixed-effects approach deals in part with such endo-
geneity by controlling for unobserved characteristics that may be affecting both civilian
casualties and the power of the rebels, I interpret the econometric results as associations.
Nevertheless, these associations are informative, especially since they are consistent with
the predictions of the theoretical framework.
21. Fixed-effects Poisson results are available from the author upon request.
22. These are, however, the unconditional mean and variance and their comparison only sug-
gests whether overdispersion is likely to be present. More formally, in the regression set-
ting one can test whether the conditional mean and variance are significantly different
from each other or not.
References
Angrist, J., and J. Pischke. 2008. Mostly Harmless Econometrics: An Empiricist’s Companion.
Princeton, NJ: Princeton University Press.
Aranguren, M. 2001. Mi Confesión: Carlos Castaño Revela sus Secretos. Bogota: Oveja Negra.
Azam, J. P., and A. Hoeffler. 2002. “Violence against Civilians in Civil Wars: Looting or Terror?”
Berkeley, B. 2001. The Graves Are Not Yet Full: Race, Tribe and Power in the Heart of Africa.
New York: Basic Books.
Cenarro, A. 2002. “Matar, Vigilar y Delatar: La Quiebra de la Sociedad Civil Durante la Guerra
y la Posguerra en España (1936–1948).” Historia Social 44: 65–86.
Chacón, M., J. Robinson, and R. Torvik. 2011. “When Is Democracy an Equilibrium?
Theory and Evidence from Colombia’s La Violencia.” Journal of Conflict Resolution 55, no.
3: 366–9 6.
Dube, O., and J. F. Vargas. 2013. “Commodity Price Shocks and Civil Conflict: Evidence from
Colombia.” Review of Economic Studies 80, no. 4: 1384–1421.
Fearon, J. 2004. “Why Do Some Civil Wars Last So Much Longer Than Others?” Journal of Peace
Research 41, no. 3: 275–302.
Giraldo, F., R. Lozada, and P. Muñoz. 2001. Colombia: Elecciones 2000. Bogota: Pontificia
Universidad Javeriana.
Hirshleifer, J. 1991. “The Paradox of Power.” Economics and Politics 3, no. 3: 177–2 00.
Horton, L. 2004. “Constructing Conservative Identity: Peasant Mobilization Against
Revolution in Nicaragua.” Mobilizatio: An International Quarterly 9, no. 2: 167–80.
Johnson, D. H. 1995. “The Sudan People’s Liberation Army and the Problem of Factionalism.” In
C. Clapham, ed., African Guerrillas. Bloomington: Indiana University Press, 53–72.
Society 11, no. 3: 243–85.
Kalyvas, S. N. 2006. The Logic of Violence in Civil War. Cambridge, UK: Cambridge
University Press.
Kirk, R. 2003. More Terrible Than Death: Massacres, Drugs, and America’s War in Colombia.
New York: Public Affairs.
Lacina, B. A., and N. P. Gleditsch. 2005. “Monitoring Trends in Global Combat: A New Dataset
of Battle Deaths.” European Journal of Population 21, nos. 2–3: 145–65.
Lindbeck, A., and J. Weibull. 1987. “Balanced-Budget Redistribution as the Outcome of Political
Competition.” Public Choice 52, no. 3: 273–9 7.
Persson, T., and G. Tabellini. 2000. Political Economics: Explaining Economic Policy. Cambridge,
MA: MIT Press.
Petersen, R. D. 2001. Resistance and Rebellion: Lessons from Eastern Europe. Cambridge,
Petersen, R., and E. Liaras. 2006. “Countering Fear in War: The Strategic Use of Emotion.”
Journal of Military Ethics 5, no. 4: 317–33.
Pinto, M. E., A. Vergara, and Y. Lahuerta. 2002. “Diagnóstico del Programa de Reinserción
en Colombia: Mecanismos para Incentivar la Desmovilización Voluntaria Individual.”
Archivos de Economia. Working Paper No. 211. National Planning Department, Colombia.
Polk, W. R. 2007. Violent Politics: A History of Insurgency, Terrorism, and Guerrilla War from the
American Revolution to Iraq. New York: Harper Collins.
Restrepo, J., and M. Spagat. 2004. “Civilian Casualties in the Colombian Conflict: A New
Approach to Human Security.” Unpublished manuscript. Royal Holloway, University
of London. Available from: http://personal.rhul.ac.uk/u hte/014/HS%20in%20
Colombia%20Civil%20Conflict.pdf.
Restrepo, J., M. Spagat, and J. F. Vargas. 2004. “The Dynamics of the Colombian Civil
Conflict: A New Data Set.” Homo Oeconomicus 21, no. 2: 396–428.
Romero, M. 2003. Paramilitares y Autodefensas 1982–2003. Bogota: Universidad Nacional.
Ross, M. 2005. “Booty Futures.” Unpublished manuscript. University of California–L os
Angeles. Available from: http://w ww.sscnet.ucla.edu/polisci/faculty/ross/bootyfutures.
pdf.
Russell, D. E. H. 1974. Rebellion, Revolution and Armed Force. New York: Academic Press.
UNHCR. 2006. 2005 Global Refugee Trends. Available at: http://w ww.unhcr.org/statistics.
Wolf, E. 1973. Peasant Wars of the Twentieth Century. New York: Harper & Row.
Winkelmann, R. 2000. Econometric Analysis of Count Data. Berlin: Springer-Verlag.
World Bank. 2003. Breaking the Conflict Trap: Civil War and Development Policy. Washington,
DC: World Bank Policy Report.
20
From Pax Narcótica to Guerra Pública

Explaining Civilian Violence in Mexico’s Illicit Drug Wars
N e i l T. N. F e rguson, M a r e n M . M ic h a e l se n, a n d Toph e r
L . Mc Doug a l
20.1. Introduction
On December 9, 2010, 2,000 federal police surrounded the village of El Alcalde,
Michoacán, during a festival sponsored by La Familia Michoacana, a local drug
trafficking organization (DTO). The strike aimed to capture its leader, Nazario
Moreno González, but instead resulted in a two-day shootout that claimed at
least eleven lives. González was assumed dead, but he later resurfaced as leader
of a new, even more brutal cartel. The shootout also marked a turning point in
patterns of local drug violence. Beforehand, La Familia’s violence had principally
focused on “business-as-usual targets,” like the police and military (Finnegan
2010). Afterward, the civilian death toll rose significantly, peaking in 2011 with
almost 2,500 homicides in a state with a population of just 4.5 million people.
Remarkably, the acceleration of violence in Michoacán has been tame compared
with some states. Across Mexico, homicides tripled from around 8 per 100,000
inhabitants in 2007 to 24 per 100,000 in 2011, representing the largest upsurge
in violence in any Latin American country over the last two decades (Heinle,
Ferreira, and Shirk 2014, 24). Perhaps most troubling, however, the civilian death
toll has risen in line with the general upsurge in violence.
We contend that Mexico exhibits a number of features commonly associated
with cases of mass killing (Fein 1993; Markusen and Kopf 1995; Straus 2001).
With estimates of up to 100,000 dead from the mid-2000s to the mid-2010s, there
is no question that the severity of violence in Mexico matches that of other mass
killing scenarios, and we present strong evidence that a significant number of the
dead are civilian noncombatants. However, DTOs, rather than Mexican state
forces, have perpetrated a majority of this violence, linking to a wider debate as
to whether agents of the state must perpetrate violence in order for episodes of
452
F r o m Pa x N a r c ó t i c a t o G u e r ra P ú b l i c a 453
mass murder to count as mass killing. Ulfelder and Valentino (2008) define mass
killing as that “perpetrated by states against noncombatant[s]‌”1 but include casu-
alties from the Colombian conflict between guerilla groups, paramilitary groups,
and the government. The Uppsala Conflict Data Program (UCDP) classifies the
“drug war” in Mexico as a “nonstate conflict,” although the reality of the situation
is more complicated. Journalistic evidence suggests that representatives of local
and federal law enforcement have acted in their official capacities on behalf of the
cartels (Finnegan 2010) and that cartels have acted at the behest of state agents,
as in the notorious massacre of forty-three students near Iguala in September
2014 (Thomson Reuters 2015). Given the intercalation of government and crimi-
nal interests observed in Mexico, violence there matches even the most stringent
definitions of mass killing.
In this chapter, we then ask: What accounts for this dramatic increase in vio-
lence? And specifically, why have so many civilians been killed? Beginning with
the wider question, journalistic explanations focus almost exclusively on the
disruption of DTO operations by the Calderón administration’s “war on drugs,”
positing that before 2006, and certainly before Mexico’s democratization process,
government authorities and cartels coexisted more or less cooperatively. In this
pax narcótica, law enforcement turned a blind eye to drug trafficking operations
and even to business-as-usual levels of violence. The DTOs, in turn, maintained
strict command structures constraining wider violence.
To explain the shift from pax narcótica to more open conflict, we employ a
formal economic model with a specific focus on the civilian death toll. In line
with recent civil conflict literature (see, e.g., Ferguson forthcoming), we view the
upsurge in violence in Mexico as the result of two parallel conflicts. The first takes
place between the cartels, vying for territory and power; the second takes place
between the government, seeking to destroy DTOs, and DTOs seeking to survive
and maximize profit. We posit that combatants usually die as a result of inter-
and intracartel violence, and are presumably of relatively little consequence to
the Mexican government. We further argue that the conflict between the cartels
and the government is the predominant source of violence against noncombatant
civilians. Violence against civilians is of much greater concern and occurs, at least
in part, to damage the government’s credibility.
To model the government-cartel conflict, we explicitly ignore intercartel
violence. This ensures that we focus on the question of violence against non-
combatant civilians. We introduce a single representative cartel that seeks to
maximize profits, subject to market equilibrium prices and quantities. Noting
work such as Rios (2012), we introduce two layers of government—national and
subnational—each of which aims to maximize rents, subject to legitimacy and
security concerns. Given the cooperative nature of the pax narcótica, we model the
relationship among the three strategic players in a cooperative framework. The
players thus face the choice to be involved in a peaceful grand coalition or feasible
454 Case Studies II
subcoalition, or to operate as a singleton. This model explains the incentives that

sustained the pax narcótica, those that led to the “war on drugs,” and those that
resulted in a massive spike in violence against civilians.
Focusing on three cases of specific interest—the grand coalition, a couplet
coalition involving national and subnational government against the cartel,
and a couplet of subnational government and the cartel against the national
government—we show the incentives (1) under which cooperation breaks down,
(2) under which a crackdown on the cartel would be most effective, and (3) that
seem to have driven the acceleration in violence. In doing so, we show the effec-
tiveness of violence against civilians as a strategic weapon that encourages both
layers of government to cooperate with the cartel. Such cooperation, however, is
more likely to break down when governments underestimate the violent capacity
of the cartel.
When cooperation does break down, we show that a coalition of both layers of
government engaging in a coordinated crackdown provides the best opportunity
to inflict meaningful damage on the cartel. We further show civilian-d irected
violence can be used as a tool to coerce cooperation from both layers of govern-
ment, and can also be used as a tool to ensure the formation of couplet coalitions
where the layers of government do not coordinate their crackdown. In this case,
any attempted crackdown is implicitly weak, ensuring that conflict retains a high
strategic value, with credible threats of violence against targets relevant to the
subnational government incentivizing it to cooperate with the cartel. Put differ-
ently, cartel violence against civilians residing in a particular subnational juris-
diction drives a wedge between the subnational and the national government, in
favor of the cartel. This builds on the weak position of a subnational government
operating as a singleton and, accordingly, its easy corruptibility.
Finally, we show the importance of the national government’s preference
over centralization in influencing this outcome: the higher this preference, the
weaker is the subnational government’s singleton position and, thus, the easier
it is to corrupt. In turn, this generates a moral hazard for outside agents attempt-
ing to incentivize the national government to crack down on the cartel, as high
rewards to the national government for a crackdown increase its preferences
over centralization. This implies an important role for international governments
in supporting Mexico’s subnational governments and ensuring meaningful
decentralization.
The chapter proceeds as follows. Section 20.2 provides a short history of the
drug-related conflict in Mexico, a discussion of then-President Calderón’s strategy
to fight DTOs, an overview of recent levels of violence, and evidence of increased
violence against noncombatants. In Section 20.3 we discuss relevant literature, both
in explaining the sharp increase in violence in Mexico since 2006 and in discussing
formal models that explain the behavior of organized crime groups. We present our
own formal model and results in section 20.4 and conclude in section 20.5.
20.2. Background
20.2.1. History of Drug Trafficking and the Pax
Narcótica in Mexico
The beginning of drug trafficking in Mexico dates back to the country’s revolu-
tion in the early twentieth century, when it was largely restricted to domestic pro-
duction and minimal international exchange. It was only in the 1940s in Ciudad
Juárez, one of the most violent cities in Mexico today, that international traffick-
ing became more prominent (Astorga and Shirk 2010). Since then, drug-related
activities in Mexico have experienced three peaks: a boom driven by increasing
marijuana demand from the United States in the 1960s; an increase in interna-
tional demand for cocaine in the 1980s; and, in the 2000s, as a result of supply-
side constraints in South America. Until the 1980s, Mexico mainly transited,
rather than produced, drugs. Production took place almost entirely in Colombia
and other South American countries until joint US and Colombian efforts against
production and trafficking led Colombian criminals to cooperate with Mexican
cartels. Along with joint US and Caribbean counternarcotics operations in the
1980s and 1990s, this led to a shift of the drug-trafficking business to Mexico
(Seelke et al. 2011; Castillo, Mejia, and Restrepo 2013). This period coincided
with the seventy-year-long one-party rule of the Institutional Revolutionary
Party (PRI) (Magaloni 2006). The PRI cooperated with DTOs by agreeing to
unofficial rules allocating territories to illegal organizations—t he pax narcótica—
which was relatively stable during the last two decades of PRI administration in
the 1980s and 1990s (Lindau 2011).
With credible democratization occurring in the late 1990s, the PRI lost the
presidency in 2000 to the National Action Party (PAN), led by Vicente Fox. This
entailed a loss of national control over DTO activities, allowing them to enlarge
territories and expand political power (Lindau 2011). Moreover, faced with
increasing corruption scandals, the new government saw the necessity to demon-
strate its legitimacy, leading to the “war on drugs” declared by Felipe Calderón,
Fox’s successor. In this “war,” two broad parallel conflicts take place: one between
Mexican DTOs fighting turf wars, and one between the DTOs and the Mexican
government’s forces.
20.2.2. Calderón’s Strategy and Security Spending

in the 1990s and 2000s
Upon Calderón’s inauguration in December 2006, he immediately acted on his
campaign promise to fight drug-related crime by deploying thousands of soldiers.
In fact, during his sexenio (six-year term), he stationed 45,000 troops annually;
more than double his predecessor’s commitment (Grayson 2013). This uncon-
ditional tactic reaped early benefits like the arrest of “kingpin” Víctor Magno
456 Case Studies II
Number of personnel, desertions 200000 50000
150000 37500
Military budget
In mio. pesos
100000 25000
50000 12500
0 0
1980 1990 2000 2010
Army personnel Navy personnel
Number of desertions Military budget
Figure 20.1 Desertions, military spending, and composition of Mexico’s armed

forces. Source: Grayson (2013); authors’ graphical illustration.
Escobar. Such arrests constituted Calderón’s main approach to combating DTOs,

in addition to confiscation of assets and weapons, destruction of drug crops, and
deployment of security forces to reassert control of territory. This strategy was
backed by a huge increase in government spending on armed forces, primarily
through large increases in the salaries of combat troops (see Figure 20.1). Monthly
salaries increased from about 4,300 pesos in 2006 to 10,800 pesos in 2012 (El
Economista 2012), which accordingly bolstered the armed forces’ loyalty and
lowered the number of desertions (Moloeznik 2013). This budget increase, how-
ever, coincided with the world financial crisis of 2008–2009. Federal income
decreased by 6.9 percent in six months, with corresponding decreases in trans-
fers to subnational government institutions. By the end of 2009, about 80 percent
of Mexico’s 2,500 municipalities were insolvent (Lindau 2011). Nevertheless,
despite this crisis, some one hundred “drug kingpins” were arrested or killed dur-
ing the Calderón administration (Grayson 2013).
20.2.3. Current Levels of Violence

Following a secular downward trend, annual homicide rates in Mexico dramati-
cally rose after 2007, increasing from 8.1 to 23.7 per 100,000 inhabitants in five
years. Figure 20.2 plots a simulation of expected violence in Mexico, based on a
statistical model linking Mexican violence to violence in other North and Central
American countries (the synthetic control rate, or SCR; dashed line), which con-
trasts with the observed rate of homicides in Mexico (solid line). The figure shows
25
20
Homicide Rate (Annual per 100k)
15
10
1995 2000 2005 2010 2015

Year
Observed rate Synthetic control rate (SCR) 95% CI for SCR
Figure 20.2 Annual intentional homicide rates per 100,000 people in Mexico, 1998–
2011. Note: Vertical line is the start, in 2006, of the Calderón presidency.
Source: Instituto Nacional de Estadística y Geografía (INEGI). Calculations by the authors.
500
Mean
Maximum
400
300
200
100
0
Chihuahua (1)
Sinaloa (2)
Durango (3)
Guerrero (4)
Baja California (5)
Nayarit (6)
Tamaulipas (7)
Neuvo Leon (8)
Sonora (9)
Michoacan (10)
Morelos (11)
Coahuila (12)
Colima (13)
Oaxaca (14)
Zacatecas (15)
Jalisco (16)
Mexico (17)
Distrito Federal (18)
Quintana Roo (19)
San Luis Potosi (20)
Veracruz (21)
Tabasco (22)
Guanajuato (23)
Chiapas (24)
Aguas Calientes (25)
Hidalgo (26)
Campeche (27)
Puebla (28)
Baja California Sur (29)
Tlaxcala (30)
Queretaro (31)
Yucatan (32)
Figure 20.3 Mean and maximum homicide rates per 100,000 people for Mexican states,
2006–2012. Source: INEGI. Calculations by the authors.
that violence in Mexico has increased dramatically since the mid-2000s, and has
done so against the expected trend. Locally, intentional homicides peaked in
Ciudad Juárez with more than 3,000 drug-related killings in 2012 (Figure 20.3).
The state of Chihuahua, which includes an important north-south crossroads for
458 Case Studies II
2007
(0, .05)
(.05, .5)
(.5, 1)
(1, 20)
2011
(0, .05)
(.05, .5)
(.5, 1)
(1, 20)
Figure 20.4 Geographic distribution of intentional homicide rate (per 1,000) in

Mexican municipalities, 2007 and 2011. Source: Sistema Nacional de Información en Salud
(SINAIS). Calculations by the authors.
the drugs trade, has a post-2006 average of 162 homicides per 100,000 inhabit-
ants. Violence has also spread geographically, as shown in Figure 20.4. From 2007
to 2012, the number of Mexican municipalities experiencing at least 1 intentional
homicide per 1,000 inhabitants grew from less than 1 percent to more than 5 per-
cent, while the number of municipalities with at least 1 per 10,000 almost doubled
from 19 percent to 35 percent.
Numbers alone fail to convey the often-brutal nature of DTO violence.
Corpses are frequently deformed, decapitated, burned, and displayed in
public—for example, hung from bridges or dumped in the street. Severed heads
have been placed in front of schools or rolled onto dance floors. At times, these
displays are accompanied by narco-messages from a perpetrator DTO to other
DTOs, to the public, journalists, or security personnel, causing fear and menac-
ing the everyday life of locals. Such cruelty, and the frequency of such events, has
fueled scholarly interest in the wide-ranging consequences of the war against
drugs on Mexico’s population. 2 Massacres, too, have become common during
the war on drugs, with multiple individuals killed in a single event. Primary
data on such massacres are unavailable, but summing the number of intentional
homicides per region that occurred in the same type of location (e.g., street,
school, commercial area, etc.) provides some insight. 3 Since no incontestable
definition exists of what constitutes a massacre—human rights groups’ defini-
tions change over time (Ballesteros et al. 2007)—we adopt definitional thresh-
olds of five and ten homicides. As shown in Figure 20.5, and as with the overall
homicide rate, we see a dramatic increase after 2007, with massacres of at least
500
400
Number of massacres
300
200
100
0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
Year
5 or more killings 10 or more killings
Figure 20.5 Number of massacres in Mexico, 2002–2011. Source: SINAIS. Calculations by

the authors.
460 Case Studies II
five (ten) increasing from 5 (2) to 500 (100) in four years. Given that such inci-
dents have attracted widespread news coverage, it is likely that they also have
become part of DTOs’ toolkits, generating attention and intimidating and influ-
encing citizens, police, politicians, and rival DTOs.
20.2.4. Evidence of Civilian Victims

As with massacres, it is impossible to find reliable numbers of homicide victims
not involved in the drug business, yet many reports in media and other outlets
suggest a significant share of victims are civilians. To suggest the extent of this
problem, we pursue two strategies. First, we document a number of cases in which
civilians have been victims. Second, we provide an account of the socioeconomic
characteristics of victims. This splits fatalities into victims that are typical of
drug-related violence (male, unemployed, uneducated, young) and those that are
atypical (e.g., females, older males, and employed and/or educated individuals).
Using typical and atypical as approximate splits of the combatant/civilian status
of a victim, we show that fatalities of atypical targets have increased dramatically,
along with fatalities of typical targets.
Source materials drawn from cases reported in newspapers, blogs, and human
rights groups’ reports argue that civilian deaths occur in the Mexican drug war
for one of two principal reasons: “collateral damage” and intentional killings by
DTOs. There is also some evidence of a third type of civilian victim: those killed
by security forces, largely because of adverse incentives to present evidence of
captured or killed criminals. Although such falsos positivos were a serious problem
during the civil war in Colombia, no strong evidence exists that this is an equally
serious problem in Mexico. Human Rights Watch (2011), however, notes several
cases of torture committed by security forces seeking information about DTO
leaders. Where extrajudicial killings have occurred, they are unlikely to be made
public, and investigations into such cases are rare (Human Rights Watch 2011).4
Nonetheless, such human rights violations are few compared to the killing, kid-
napping, and extortion of civilians by DTOs. The latter include, for example, the
following: (1) shooting at three buses of maquiladora workers in Ciudad Juárez on
October 27, 2010 (Archibald 2010); (2) documented loss of family members by
Mexican refugees (Hernandez 2012); (3) 52 people killed in a casino in Monterrey
(BBC News Online 2011a); (4) 7 civilians murdered near Cuernavaca (BBC
News Online 2011b); (5) 8 people killed and 100 injured in Morelia (El Universal
2008; García Tinoco 2012); (6) 32 people killed in Veracruz, September 11, 2011
(Martinez 2012); (7) Mexico’s Human Rights Commission confirms at least 45
innocent victims killed by Mexican armed forces from 2007 to 2012 (Martinez
2013); (8) 5 civilians, among them three children, killed when soldiers opened fire
on a truck at a checkpoint (Moloeznik 2013); and (9) 30 mayors killed between
2006 and 2013 (Grayson 2013, xiv), although noninvolvement with criminal
activity was not established.
15,000
Number of homicide victims

10,000
5,000
0
98
99
00
01
02
03
04
05
06
07
08
09
10
11
12
19
19
20
20
20
20
20
20
20
20
20
20
20
20
20
Typical victim Atypical victim
Figure 20.6 The number of typical victims, defined as male, aged between 15 and 50,
and with less than secondary education. Atypical victims are those who do not possess
this set of characteristics. Source: SINAIS. Calculation by the authors.
The most common conception of an individual involved with a DTO in Mexico

is a low-educated male, fifteen to fifty years old. Broadly speaking, this conception
is supported by the data in Figure 20.6, which shows that these individuals consti-
tute a much higher proportion of victims than their underlying distribution in the
population would suggest. We call these individuals “typical victims,” while those
who do not possess all these characteristics comprise the group of “atypical vic-
tims.” This includes all women, males less than fifteen years of age or over fifty, men
with at least secondary education, and so on. Before 2007, more than half of all
intentional homicides targeted typical victims. Except for 2011, this pattern is rep-
licated every year through 2012. As with our earlier victim analyses, Figure 20.6
shows a large year-over-year increase in the number of victims since 2007, yet the
composition of these victims does not change significantly over time. The number
of atypical victims increases in line with the overall increase, suggesting a signifi-
cant spillover effect of DTO violence to civilian victims. Although we cannot say
with certainty that atypical victims are uninvolved in the drugs trade, this is likely
the case, implying significant and proportional increases in civilian victims.
20.3. Previous Literature

20.3.1. Literature on the Determinants of Increased
Homicide Rates in Mexico
Over a dozen studies have analyzed the determinants of increased violence dur-
ing the Calderón administration. All of them suggest that some combination of
462 Case Studies II
underlying factors and Calderón’s military approach caused the acceleration in

violence since 2007. The studies fall into two broad groups, the first highlighting
the role of Mexican politics and institutions, and the second focusing on aggre-
gate supply shocks.
Among the first, Vilalta (2014) suggests that systemic imbalances in law
enforcement between Mexican states can explain the escalation in violence.
Lessing (2012) shows that crackdowns, when carried out unconditionally, lead to
more violence in a corrupt environment. In contrast, Dell (forthcoming) shows
that homicides significantly increased after close municipal elections won by PAN
candidates. She states that a PAN victory weakens local DTOs and attracts vio-
lence by rivals seeking to control the territory. Similar arguments are proposed by
Ríos (2012, 2014), who elaborates more on the democratic process in which the
government’s interest in law enforcement increases. Snyder and Duran-Martinez
(2009) view violence as the manifestation of a breakdown of a relatively peace-
ful, state-sponsored protection racket. Such changes in power, particularly at sub-
national levels, have led to an interruption of the pax narcótica because the new
government is concerned about its legitimacy. The vulnerability of this strategy is
not only visible in the resulting increase in homicide rates. In the 2011 gubernato-
rial elections in Michoacán, approximately fifty individuals at different institu-
tional levels withdrew their candidacy due to criminal threats (Wilkinson 2011;
Aguirre and Herrera 2013, 222).
The second strand of literature suggests that aggregate supply shocks con-
tribute to the rise of violence in Mexico. For example, the lapse of the US fed-
eral assault weapons ban has been shown to be a significant driver of homicides
in Mexico (Chicoine 2011; Dube, Dube, and García-Ponce 2013). McDougal
et al. (2014) support this, demonstrating that upward of 2 percent of realized
demand for firearms in the United States is purchased for smuggling to Mexico.
Dube, García-Ponce, and Thom (2014) link changes in maize (i.e., corn) prices
to increased violence, through changes in the opportunity costs of growing
certain crops, while Castillo, Mejía, and Restrepo (2013) suggest that cocaine
supply constraints caused by seizures in Colombia triggered greater violence
in Mexico. All of these studies state that such shocks are important underly-
ing factors in the escalation of homicide rates when the war on drugs began
in 2006.
It also bears mentioning that both strands include studies highlighting long-
run structural explanations for the rise of the cartels. Davis (2007) suggests
that in the wake of the Mexican Revolution, the state outsourced law enforce-
ment functions to local militias, compromising its Weberian monopoly on the
legitimate use of force. Other explanations emphasize structural changes to the
Mexican economy precipitated by the 1994 passage of the North American Free
Trade Agreement (NAFTA) and subsequent social dislocation and migration
(e.g., Mercille 2011; Redmond 2013).
20.3.2. Previous Models of Gang Violence

Numerous works attempt to explain organized criminal violence via formal
models. A recurring tension, however, is that some models imply that gangs
thrive where governments are weakest (Skaperdas and Syropoulos 1995;
Skaperdas 2001), while others argue that some minimum threshold of govern-
ment capacity is necessary to provide the public goods required for business
(e.g., de Groot and Shortland 2010). We argue that the latter model fits the
Mexican case best, not least because Mexico is home to a strong and relatively
well-sourced federal government. Grossman (1991) develops a model of insur-
rections that, nevertheless, has certain applicable features for predicting gang
activity, showing that the relative strengths of strategic players are a function
of the choices made by the citizenry. The inapplicability of this model to the
study at hand, however, is that rebelling, while illegal, is an imprecise proxy
for DTO activity, as the only payout for rebelling is gained by overthrowing
the government and usurping its revenues, not engaging in illicit economies as
cartels often do. 5
Akerlof and Yellen (1993) model the interactions between three predefined
groups of actors: police, gangs, and community. Gangs maximize profits by
minimizing costs, which are determined by legal penalties, police monitor-
ing expenditures, and community cooperation with the police. Gangs, thus,
wish to convince the community to keep knowledge of its activities secret by
keeping violence low or maintaining a credible threat of retaliation. Profits
rise in crime levels until a threshold is reached, beyond which the community
begins to report crime, leading to reductions in profits. Despite intuitive find-
ings, such models inherently preclude the kind of cooperative grand coalition
among strategic players that would result in the pax narcótica in Mexico. For
that reason, we invoke a cooperative model (e.g., Aivazian and Callen 1981,
McDougal and Ferguson 2012), which looks at the incentives of each group to
cooperate and the incentives to form couplets of particular relevance, should a
grand coalition arrangement fail.
20.4. A Model of Incentives for Cooperation

and Violence
20.4.1. Model Intuition
Our model builds upon the literature, following the idea that a corrupt—or at
least corruptible—government is essential to the explanation of the increase in
violence. We also follow the notion that a breakdown between different layers
of government fuels violence, although we parse this in terms of the incentives
and preferences that the national government holds in regard to centralization
of power and resources. A national government with increasing preferences or
464 Case Studies II
incentives to centralize can result in a weakened and potentially corruptible sub-

national government.
We augment this idea with the suggestion that, broadly, violence targeting
civilians has a direct cost to government and thus is of strategic value to the
cartels. This does not imply that all violence against civilians by DTOs is carried
out to cost the government, but offers a rationale for DTOs to engage in behavior
that may otherwise seem irrational (see, e.g., Kalyvas 2006). This line of argu-
ment suggests that increasing civilian victimization is a result of an interaction
between the government and the cartels. In contrast, as discussed earlier, the
increase in typical victims is more likely indicative of interactions between rival
cartels. In our model, we seek to explain increasing civilian victimization in terms
of relationships between different layers of government and also between these
layers of government and the cartels.
We incorporate the basic logic that all strategic players involved in the war on
drugs act according to private economic incentives, which can, in turn, lead to a
breakdown of cooperation between local and national layers of government. Such
a breakdown not only creates adverse incentives, especially for subnational gov-
ernment, but also helps to foster the type of violent atmosphere seen in Mexico
in the past decade. Given the nature of potential coordination between strategic
players in Mexico and the cooperation inherent in the pax narcótica, we argue that
the strategic relationships are analogous to a cooperative game, where violence is
influenced by if, and how, any grand coalition breaks down.
20.4.2. Model Setup

Consider two levels of rent-seeking government: the national (N) and the sub-
national (S). These government players must coordinate among themselves
and with a representative DTO, referred to as “the” cartel (C), to choose their
optimal strategic action. The national government maintains a monopoly on
the capacity to raise rents but hands some of this capacity to the subnational
government, which can use it with greater allocative efficiency. The cartel sells
drugs at market equilibrium prices and quantities. We first consider a grand
coalition, akin to the pax narcótica, where all three players cooperate. Implicit
in cooperation among the players is that neither layer of government cracks
down on the cartel, nor that the cartel engages in violence over and above a
business-a s-u sual level. 6 Thus, the payoff function in the grand coalition
(GC) is:
ΠGC = R + (α − 1)R s + PGC , (1)

where ΠGC refers to overall payoff flow available to the grand coalition (GC). On
the right-hand side of the equation, R stands for the total amount of rents raised
by the national government and R s represents national to subnational transfers,
augmented by an efficiency parameter, α > 1. That is, in transferring rents from the
national to the subnational level, efficiencies are gained that increase the overall
payoff. (R s is associated with a normalized cost of 1 to the national government but
is spent by subnational government.) PGC represents the profits of the cartel, with
the GC subscript referring to equilibrium prices and quantities of the drug market
in the grand coalition (GC) condition. In the GC, any redistribution of equation
(1) profits among the players is feasible. Thus, a series of bribes are likely to act
as lubricants. This means that the cartel is capable of paying the government to
turn a blind eye to business-as-usual violence and, perhaps more controversially,
that the government can bribe the cartel to remain peaceful. Equation (1) shows
the essential structure of the pax narcótica in which all three players cooperate.
As both layers of government are willing to tolerate illegal profit-raising actions
of the cartel, there are no costs associated with open conflict between any of the
players.
In situations outside of the GC, however, neither level of government has a
direct incentive to tolerate illegal activities. This implies that a feasible couplet
coalition exists, where the two layers of the government work together to crack
down on the cartel, which we denote National/Subnational (NS). This crack-
down is met with violent resistance from the cartel, which imposes costs on both
layers of government. The costs of this violence, however, are traded off against
improved government legitimacy, which could take the form of local electoral
benefits or increased international standing bringing in greater external invest-
ment and support from foreign governments, for example. As the purpose of the
crackdown is to reduce the capabilities of the cartel, the cartel’s violent response
is tempered as a result. This implies that the governments do not, directly, benefit
from cracking down but garner some material gain in the form of legitimacy and
through reducing cartel violence. The payoff function of the NS coalition is:
ΠNS = R + (α − 1)R s + (1 − cC)C + L − V (2)
where cC is a constant relative marginal cost of crackdown of scale C, and where C

itself is additively separable, such that C = CN + CS. The marginal benefit of C is nor-
malized to 1, with cC a relative cost to this return. By implication, cc < 1 ∀ C > 0, as
any devotion to C when cC > 1 would result in a loss, L, where L = LN + LS is the accrued
benefit to the governments from improved legitimacy. L is, thus, also additively sepa-
rable, comprising benefits that accrue to each level of government separately. By the
same token, V = VN + VS, where V is the violence perpetrated by the cartel against
targets relevant to each level of government. In sum, the right-hand side of equation
(2) states that the payoff to the NS coalition is comprised of the rent-related payoff to
466 Case Studies II
the national and subnational governments, as before, plus the net benefit (benefits
minus costs) from the crackdown on the cartel, plus legitimacy benefits, minus vio-
lence costs stemming from retaliatory violence.
Two subsequent couplets occur, which sees the cartel form a coalition with one
layer of government or the other. In the first of these, the national government forms
a coalition with the cartel (NC). As such, it abandons links with the subnational
government and essentially cuts off the transfer of rents. Given the nature of the
military forces available to the national and subnational governments, respectively,
we assume that the national government is capable of defending itself and its car-
tel partner against a subnational government crackdown. While it is intuitive as to
why the cartel would like to form a couplet with the national government, this may
be infeasible. The cartel may then attempt to form a coalition with the subnational
government (SC). In this setting, the subnational government may be susceptible to
corruption, as the national government withholds rent transfers. At the same time,
while suffering from a crackdown by the national government, the cartel benefits
from avoiding the added impacts associated with coordinated government action.
The payoffs for the NC and SC couplets are written, respectively, as follows:
Π NC = R + PNC + (1 − c V )VS (3)
and
Π SC = PSC − C N + (1 − c V )V N , (4)
where cV is the relative constant marginal cost of cartel violence. As with cc, this
assumption implies that c V < 1 ∀ V > 0.
Finally, it is possible that at least one player would have no incentive to form
any kind of coalition at all. The national government, for example, stands to gain
both from maintaining a monopoly over rents and from improved legitimacy,
should it not form a coalition with either of the other players. The payoffs from
these singletons are written for the national government, subnational govern-
ment, and the cartel, respectively:
Π N = R + L N − (1 − cC)C N − V N (5)
ΠS = L S − (1 − cC)C S − VS (6)
ΠC = PC − C + (1 − c V )V . (7)
These equations, particularly equation (6), show the vulnerability of the

subnational government to corruption. As a singleton, it is entirely reliant on
legitimacy from outside for income, yet still suffers the damaging effects of
cartel violence. At the same time, however, equation (5) shows that, particu-
larly when the benefits of legitimacy are large, the national government has
strong incentives to act as a singleton and to cut off transfers to the subnational
government.
20.4.3. The Grand Coalition

Given the payoffs defined in equations (1) to (7), we turn our attention to the con-
ditions under which the three situations of major interest occur. First, we consider
the grand coalition (GC) as analogous to the pax narcótica and look at the incen-
tives under which all three players cooperate. Second, we look at a situation where
both layers of government cooperate (NS) in their bid to bring down the cartel.
Given the coordinated crackdown associated with this couplet, we argue that it is
analogous to a best-case scenario, where cartel capabilities and violence are mini-
mized. As such, this couplet essentially reflects the way toward “peace.” Finally,
following Rios (2012), we look at a situation where cooperation between the two
layers of government breaks down and the effect this has on violence. Given the
easy corruptibility of the subnational government, this implies that an SC coali-
tion could be formed, as the subnational government can be incentivized to enter
any coalition, given the vulnerable outcome if it acts as a singleton.
Start with the grand coalition. For the grand coalition to hold, each player’s
individual outcome in it (denoted N, S, and C for the national government, sub-
national government, and cartel, respectively) must be greater than it would be in
any other couplet coalition or as a singleton. This implies that the following set of
conditions must hold simultaneously:
N + S ≥ R + (α − 1)R S + L + (1-c C) C−V (8)
N + C ≥ R + P NC+ (1−c V)VS (9)
S + C ≥ PSC −CN + (1−c V)V N (10)
N ≥ R + L N − (1−c C)CN −V N (11)
S ≥ L S − (1−c C)CS −VS (12)
C ≥ PC − C + (1−c V)V, (13)
where N+S+C = ΠGC = R+(α–1)RS+PGC, and reflects the payoff in the grand coali-
tion to the national government, subnational government, and cartel, respec-
tively. By collecting the terms on the left-hand side of equations (8) to (13) and
noting that N+S+C = R+(α–1)RS+PGC, we are able to rewrite this set of conditions
as a single condition:
468 Case Studies II
Condition (1): 2(α –1)RS + 3PGC ≥ 2L –2cVV + CS –2cCC + PNC + PSC + PC.
Condition (1) shows that the essential structure of the grand coalition is rather
intuitive. Importantly, it also shows the strategic value of cartel violence (V) as a
means to influence government behavior—as the right-hand side of Condition
(1) decreases, the likelihood of the grand coalition prevailing grows. Thus, as vio-
lence grows, so does the likelihood of the grand coalition, implying that a high
credible threat of violence from the cartel can outweigh any legitimacy benefits by
imposing such significant costs on the government that a corrupt peace remains
preferable. Furthermore, should we relax the assumption that crackdowns affect
the market structure, such that PGC=PNC=PSC=PC, then, ceteris paribus, the grand
coalition is more likely to prevail when the allocative efficiency “premium” (α)
of the subnational government is high. Perhaps more importantly, it also implies
that the grand coalition is more likely to fail, should the rent transfer to the subna-
tional government, RS, decrease.
Allowing the structure of the drug market to vary with the scale of the crack-
down, however, provides more in-depth understanding of the intuition of the
model and details the complexity of the situation. Here, two features of the
drug market become important determinants in shaping the players’ incentives
to form a grand coalition. First is the relationship between the crackdown and
consequent reductions in drug supply, which will affect drug pricing; second is
the demand side’s sensitivity to price changes, technically called the “price elas-
ticity of demand.” Under certain conditions, a government-led crackdown can
have the perverse effect of increasing cartel profits. This can increase the cartel’s
incentive to defect from the grand coalition, which in turn could spark violence.
The alternatives, of course, also are important. Profits could rise in a situation
where the market is price elastic and the supply of drugs is positively related to the
scale of the crackdown. From this, we generate the first and second hypotheses of
the model with proofs available in the notes for these and subsequent hypotheses.
Hypothesis 1: Ceteris paribus, if market supply decreases with the scale

of the crackdown and if demand is price inelastic, profits available to the
cartels increase when the government cracks down, making a collapse of
the grand coalition more likely.7
Hypothesis 2: Ceteris paribus, the greater the level of violence that the
cartel can credibly threaten, the higher the likelihood that the grand
coalition will hold.8
We simulate these outcomes in Figure 20.7. In each of the four diagrams, the
shaded area shows the proportion of the strategy space in which the grand coalition
prevails. The dashed line reveals the payoff from the grand coalition, and the dot-
ted line the outcomes from the other situations. Where the dashed line exceeds the
P
A
Y
O
F
F
ALPHA LEGITIMACY
P
A
Y
O
F
F
VIOLENCE NON-GC PROFITS
Figure 20.7 Simulations of the grand coalition (GC), varying clockwise from top
left: the allocative efficiency of the subnational government; the legitimacy premiums
of the national and subnational governments; the relative profits between the grand
coalition and other scenarios; and the credible threat of violence from the cartels.
dotted line, the grand coalition outcome yields a superior payoff. Thus, in the top
left of the figure, we vary α, the subnational efficiency parameter, holding all other
parameters constant and at neutral values. On the top right, we vary L, the govern-
ment legitimacy parameter, in the same manner.9 When subnational efficiency, α, is
high or when legitimacy, L, is at low to medium levels, the grand coalition outcome
prevails. On the bottom left, we vary the level of violence (V) credibly threatened
by the cartel. Perhaps surprisingly, as stated in Hypothesis 2, this shows a willing-
ness to form a grand coalition that is increasing in violence. This suggests that in
order to influence government behavior, violence can be used as a credible threat.
As the cartel’s capacity for violence grows, the greater the threat it poses to govern-
ment. This implies that the more violent the cartel can be, the greater the opportu-
nity it has, through violence, to influence the government’s own strategic behavior
and tilt it toward a grand coalition outcome. Alternatively, this also implies that,
should the government underestimate the violent capacity of the cartel, it will more
readily engage in a crackdown from which it does not materially benefit. Finally,
we vary the relationship between the impact of the crackdown on cartel profits in
the bottom right of the figure. On the left-hand side of the x-a xis, we start with a
470 Case Studies II
situation where profits are decreasing in the crackdown, while the right-hand side
of the x-a xis depicts a situation where profits are increasing in the crackdown.10
Figure 20.7 elucidates the important effects of the structure of the drug mar-
ket on the breakdown of the grand coalition and, accordingly, on any subsequent
upturn in violence. An incorrect understanding of the economics of a crackdown,
then, can lead to undesired violence, particularly given the possibility of the for-
mation of couplets. Given the essential black market nature of the drug market, it
is easy to understand how knowledge could be imperfect. Similarly, it also shows
the important strategic value of cartel violence. Figure 20.7 shows that this vio-
lence is an important determinant of the incentives of the government to cooper-
ate with the cartel.
In the next two subsections, we further elucidate these effects by focusing
on two other coalitions. The first is the situation where the national and subna-
tional governments work together to undertake a strong crackdown against the
cartel (section 20.4.4). In the second, we look at the situation where the sub-
national government and the cartel cooperate (section 20.4.5). These feasible
couplet coalitions are important as they, respectively, indicate the “optimal crack-
down” and vulnerability of the subnational government to corruption, and the
link this vulnerability has to widespread violence.
20.4.4. The NS Couplet

In this couplet, the national and subnational governments work together to
jointly crack down on the cartel. This would imply an integrated approach from
both layers of government, with the capacity to significantly damage the cartel
and to curtail its violent activities. On the flip side, a failure to integrate, as in Rios
(2012), suggests significant costs. In this model, we depict the NS couplet coali-
tion as the model outcome most likely to minimize violence and, in the long term,
lead to defeat for the cartels. As such, discussing the incentives under which this
outcome can occur remains important. As with the grand coalition, the NS coali-
tion will prevail if and only if both the national and subnational governments are
better off in this coalition than in any other feasible situation. Thus, following the
approach highlighted in equations (8) to (13) and noting that neither the national
nor subnational governments will enter any coalition unless that coalition offers
them a payoff that, at least, matches their outcome as a singleton, we can write the
condition for the NS couplet.
Condition (2): 2[(α –1)R S + L –cCC –V] ≥ PGC + PNC + PSC –3PC + CS
Condition (2) provides a similar set of comparative statics to those of the grand
coalition, only this time it is increasing legitimacy that is associated with a positive
probability that the NS coalition will prevail. Importantly, the willingness of the
national government to transfer rents to the subnational government, given by RS

on the left-hand side of the inequality, is positively associated with the probability
that this couplet will prevail. Exogenous decreases in these transfers, as happened
during the financial crisis in 2009, reduce the probability of this couplet coali-
tion forming. In combination with the same effect of RS on the likelihood of the
grand coalition prevailing, reductions in the capacity and desire of the national
government to transfer rents to the subnational government is a major barrier to
cooperation between the layers of government.
The effect of violence, too, changes and now makes the formation of this coali-
tion less likely. These outcomes are intuitive: increasing legitimacy, ceteris pari-
bus, directly incentivizes the governments to crack down on the cartel. At the
same time, increasing violence has a comparable but opposite effect, increasing
the costs to both layers of government from cracking down.
Hypothesis 3: The probability of an NS coalition is decreasing in the

scale of cartel violence.11
Hypothesis 3 is important as it shows a dual effect of cartel violence: Not only

does cartel violence positively contribute to the desirability of forming a grand
coalition, but it also hampers the desirability of the coalition capable of inflict-
ing the greatest damage on the cartel. From this, our first main conclusion is
drawn, that a major source of the violence seen in Mexico is aimed at influencing
the strategic behavior of the multiple levels of government. Given the direct and
damaging impact of cartel violence on government, both directly and through
associations of state failure, violence targeted at civilians is likely to have a high
strategic value in this regard.
20.4.5. The SC Couplet

We now turn our attention to the second couplet coalition of interest, the SC cou-
plet, where the relatively weak subnational government is protected from cartel
violence.
Both within the context of our model and events in Mexico as from the mid-
2000s, the possibility of a subnational government-cartel (SC) couplet is not only
an important potential coalition but also one in which violence has the potential
to rise drastically. Here, the cartel avoids the joint backlash of both layers of gov-
ernment, allowing it to perpetrate even higher levels of violence. By preventing
the two layers of government from working together, the cartel avoids a large-
scale and more dangerous crackdown. Not only does this directly benefit the car-
tel, as any crackdown is smaller than it would be otherwise, it also suggests that
cartel violence is tempered less than in any situation where the cartel operates as
472 Case Studies II
a singleton. At the same time, the subnational government is protected from the
cartel’s violence, incentivizing its potential acquiescence.
Following the methodology for defining Conditions (1) and (2), we can write
the condition for the SC couplet.
Condition (3): 3PSC + [2(1 –cV) –3]V N + VS –cVV ≥ 2(α –1)R S + L +

L S –3LN + CS + 2cCCN
Condition (3) reveals two results of profound interest. First, the relationship
between legitimacy and the payoff outcome is deeply complex (the L + LS –3LN
terms) and relies, in part, on who stands to gain what. High national government
legitimacy, LN, increases the probability that it has an incentive to cut off the
subnational government, through stopping transfers of rents, and to operate as a
singleton. But legitimacy accruing to the subnational government, LS, reduces the
ease with which it can be corrupted by the cartel. Second, Condition (3) shows
that the effect of violence is also much more complex than before (the various
V terms) and is dependent on whom the cartel would target and at what cost.
Ceteris paribus, increasing relative violence against the subnational government
increases the probability that an SC coalition will prevail, while high violence
against the national government increases the probability that an SC coalition
will not be formed. From these results, we derive two final hypotheses.
Hypothesis 4: Ceteris paribus, the more the national government stands

to gain legitimacy from a crackdown, the more likely it is that the SC
coalition prevails.12
Hypothesis 5: Ceteris paribus, should the cartel credibly threaten a high
level of violence against the national government, the SC coalition is less
likely to prevail. Should the cartel credibly threaten a high level of vio-
lence against the subnational government, however, the SC coalition is
more likely to prevail.13
These findings should not, necessarily, be surprising and are certainly linked.
In situations where gains from legitimacy to the national government are high,
the incentive for the national government to form a singleton grows. In this set-
ting, the national government jointly benefits from improved legitimacy and
from cutting off rent transfers to subnational government. This creates a trade-off
between these income gains and cartel violence. In situations where income gains
grow, relative to the effects of violence, the desirability of the singleton grows
in tandem. In this situation, subnational government is left in a weak position,
however, as it suffers significant loss of transferred rents. Should the subnational
government itself remain in a singleton, it gains in legitimacy terms but loses
out both from cartel violence and from the costs necessary for the crackdown.
In these situations, subnational government is susceptible to corruption as such

large losses of income make it relatively “cheap” for a cartel to incentivize the
subnational government to join an SC coalition, particularly when it can credibly
threaten large-scale violence against the subnational government, should it not
cooperate.
At the same time, however, should the cartel also credibly threaten large-scale
violence against the national government, the calculus inherent in the trade-offs
change. In this situation, ceteris paribus, the negative effects of violence are grow-
ing relative to increases in income. In these situations, the national government
is more inclined to cooperate with the subnational government, in a bid to ward
off the worst ravages of violence. Should this not happen, however, and the SC
couplet occur, a weaker and less effective crackdown results, which in turn allows
violence to grow relative to that in either the grand coalition or the NS couplet.
These results again show the importance of strategic violence by the cartel. On
the one hand, general increases in the net level of violence can incentivize both
layers of government to cooperate in the grand coalition. On the other hand, spe-
cific targeting of violence toward groups more relevant to one layer of government
than to the other can also influence which coalitions are formed in a situation
where the grand coalition fails. This elucidates the strategic depth of violence in
Mexico. Not only is widespread violence, itself, a strategic technique the cartel
can use to influence government behavior, but the specific targeting of such vio-
lence can be used to split coordinated government crackdown efforts. Moreover,
our model shows how private government incentives (national or subnational)
can make each layer, if unwittingly, complicit in such outcomes.
20.4.6. Summing Up the Model

In conclusion, our model shows the complexity of the interactions between differ-
ent layers of government and DTOs in Mexico. Perhaps unsurprisingly, one of the
most important determinants of the outcomes following a crackdown relate to
the structure of the drug market. In certain situations, a crackdown may increase
DTO profits, suggesting that the war on drugs could be an inefficient way to tackle
the problem. It is not necessarily clear, however, that the impact of a crackdown
on supply, or on the price elasticity of demand, in the drug market varies over
time. As such, while these economic factors are clearly important drivers of the
outcomes, they do not necessarily explain why the pax narcótica broke down, or
why violence has increased so dramatically since 2006.
We therefore also explore other incentives of the players. We show that, out-
side of the grand coalition, a feasible couplet coalition can be formed between the
national and subnational governments (NS), which maximizes the crackdown
and helps to minimize cartel violence. In situations where such a coalition could
be damaging to cartel profits, however, cartels have an incentive to prevent it
474 Case Studies II
being formed. High levels of cartel violence are shown to be an effective strategic
weapon toward achieving this end. If this violence is high enough, then, regard-
less of the target, it increases the probability that the two levels of governments
will move back toward the pax narcótica indicated by the grand coalition payoff
outcome. Furthermore, we show how changing the target of violence can lead to
the formation of alternative couplet coalitions, should the cartel be incapable of
generating enough violence to form (or restore) the grand coalition.
Here, violence targeted toward victims more relevant to the subnational than
the national government increases the probability that the subnational govern-
ment will form a coalition with the cartel (the SC outcome). In such a situation,
the substitution of violence to targets relevant to the national government allows
subnational government to reap any economic rewards of association with the
cartel without unduly suffering from its violence. These incentives leave subna-
tional government in a weak and corruptible state. Moreover, violence remains
high. The lack of a coordinated crackdown from both layers of government allows
a high amount of localized violence to go unchecked, facilitating further cartel
violence. By the same token, any change in the willingness or the capacity of the
national government to transfer rents to the subnational government creates a
similar set of adverse incentives. In 2009, for example, the transfers from the
Mexican national government to the municipal governments dropped signifi-
cantly as a result of the financial crisis. Our model shows the potentially dam-
aging effects of such reductions in RS, as it reduces the incentive of subnational
government to cooperate, in any way, with the national government.
The model thus shows the necessity to reinforce cooperation incentives
between different layers of government, not only in reducing violence in the short
term but also in winning the longer-term war on drugs. These results fit with those
of, for example, Rios (2012), who suggests that different party loyalties at different
layers of government create coordination problems, which can result in increased
violence. Our model augments this argument, suggesting that coordination prob-
lems relate not just to different political parties but also to the economic prefer-
ences and incentives of all layers of government, as well as the capacity of each. In
turn, this implies a major role for the international community in supporting sub-
national layers of government in Mexico. In our model, such support is analogous
to an exogenous increase in LS, which functions as a deterrent to the formation
of both the grand coalition and the SC coalition. Providing such external sup-
port decreases the latent corruptibility of subnational government while, simul-
taneously, laying the groundwork for a coordinated crackdown from all layers of
government. Perhaps most importantly of all, the model elucidates the potential
moral hazard of unilateral support offered to the national government, analogous
to an exogenous increase in LN. While such policies can encourage national gov-
ernment to crack down on the cartel, they have the undesirable consequence of
incentivizing it to cut off subnational government, fostering the very weaknesses
that can lead to an SC coalition and the unchecked increase in cartel violence
witnessed since 2006.
20.5. Conclusion
In this chapter, we discuss the nature of violence in Mexico and show that vio-
lence against civilians increased from the declaration of the war on drugs by
President Felipe Calderón in 2006 until 2011. To explain this phenomenon, we
create a model that derives the incentive sets for various levels of Mexico’s gov-
ernment and Mexican drug cartels. It shows how government actions and cartel
activities interrelate, and how these interactions might have driven the move from
a long-term pax narcótica to open warfare—a guerra pública.
We suggest that these incentives relate, most strongly, to the strength, and
costs, of coordination between different layers of government and the strategic
use and impact of DTO-d irected violence. A national government with strong
preferences for centralization, for example, can lead to weak subnational levels of
government, which makes the latter more susceptible to DTO-initiated corrup-
tion. The effects of such corruption, too, offer a rationale for large-scale increases
in violence. In this setting, where the DTO and subnational levels of government
engage in de facto cooperation, the DTO maintains significant capacity to engage
in violence, without this violence being tempered by a coordinated crackdown
by national and subnational government. Similarly, while international govern-
ments can influence the decision of the national government to crack down,
the incentives used to ensure such behavior can lead to moral hazard, with the
national government increasingly favoring centralization at the cost of subna-
tional government.
We show, further, that cartels in Mexico use violence, or the threat thereof, as
a strategic weapon to influence government behavior. In situations where DTOs
can credibly threaten large-scale violence, the government’s incentives to remain
in the pax narcótica grow. Specific targets of violence also are important. When
cartels target victims more relevant to subnational than to national government,
the pax narcótica may break down, and subnational government has an incentive
to cooperate with the cartels. Alternatively, when DTO violent capacity is low, or
perhaps underestimated by the government, a crackdown becomes more likely. In
this context, we conclude broadly that violence by Mexican DTOs has increased,
at least in part, due to the strategic value it plays in influencing the behavior of
government, particularly at subnational levels. It follows that the strategic and
symbolic importance of civilian casualties show how these outcomes explain the
large increases in atypical, as well as typical, victims. The effects of this strategy
can significantly influence the actions of subnational levels of government, which,
in turn, can further foster the violent atmosphere in Mexico.
476 Case Studies II
This model obviously does not attempt to describe how diversification of

cartel revenue streams into licit value chains—such as the Knights Templars’
forays into the extractive and agricultural sectors (USA Today 2014)—m ight
negatively affect civilian welfare. An intuitive argument could be made that, to
the extent that cartels engage in licit businesses, they may wish to intimidate
civilian participants in those revenue streams into keeping secrets, accepting low
pay, or paying for “protection.” However, the Knights Templars seem, for the
time being, to be more the exception than the rule among cartels, which still
derive the bulk of their revenues from narcotics trafficking. However, this line of
thinking might inform a future extension of our model. Likewise, our approach
does not attempt to model the actions of local communities. To the extent that
self-defense forces have arisen in parts of Mexico to expel cartels, it may make
sense that cartels have started to target civilians actually or potentially associ-
ated with them.
Acknowledgments
We are grateful for excellent research assistance from Ieva Sriubaite. We would
like to thank four anonymous reviewers for their valuable critiques and Jurgen
Brauer and Charles Anderton for their guidance and insight. All remaining errors
are our own.
Notes
1. Other definitions abound. See, e.g., Fein (1993), Markusen and Kopf (1995), and Straus
(2001).
2. See, e.g., the consequences on economic performance by Robles, Magaloni, and Calderón
(2013) and Ajzenman, Galiani, and Seira (2014); on migration by Rios (2014); and on
human capital by Michaelsen and Salardi (2014).
3. Mexico’s geostatistical areas (regions) are divided into 32 states, about 2,450 municipali-
ties, and circa 300,000 localities.
4. Human Rights Watch (2011) investigated a series of extrajudicial killings (24), cases of
torture (217), and forced disappearances (39) of civilians by Mexico’s security personnel.
These figures are based on research conducted in the five states Baja California, Chihuahua,
Guerrero, Nuevo León, and Tabasco, which are among the most affected, between 2006
and 2010 and only provide a glimpse of the number of civilian victims by the military. By
2013, HRW documents 250 disappearances since 2007. In 149 of these, HRW found evi-
dence of involvement of the military, sometimes in conjunctions with DTOs.
5. A large literature on war economies argues, however, that rebel groups may wage war pre-
cisely in order to engage in illicit markets (Duffield 1999; Naidoo 2000; Nietschke 2003;
Pugh and Cooper 2004; Cockayne 2010; Shortland 2011).
6. This implies that there is a certain amount of drug-related violence the government is will-
ing to tolerate. This excludes violence against targets deemed important to the govern-
ment, such as military personnel, political personnel, and civilians.
7. Proof: Under these conditions, and the sensible assumption that C > CN > CS > 0, it fol-
lows that PC > PSC > PNC > PGC. Thus, ceteris paribus, the right-hand side of Condition
(1) increases relative to the left-hand side in a crackdown as PNC + PSC + PC > 3PGC. This, in
turn, implies that the grand coalition is less likely to prevail than in any alternative situa-
tion, PNC + PSC + PC ≤ 3PGC.
8. Proof: The term –2cVV on the right-hand side of Condition (1) is and remains negative
under the logical assumption that cV </ 0. Thus, as violence increases, the left-hand side
of Condition (1) decreases relative to the right-hand side, increasing the likelihood that the
grand coalition will prevail.
9. For descriptive ease, these graphs are depicted under the assumptions that
PGC=PNC=PSC=PC=1 and with the parameterizations R S=0.5, cV=1, C=1.25, CS=0.5, cC=1
and where, when not varied, α=1.5, L=1.5, and V=0.5.
10. These simulations are conducted using the same parameterizations as described in the
previous note.
11. Proof: V appears with a negative sign and only on the left-hand side of Condition (2). An
increase in V, therefore, comparatively reduces the left-hand-side value of Condition (2),
leading to a decrease in the probability that the payoffs from an NS couplet will be suitably
large. For all levels of V ≥ 0, the negative effect on the left-hand side ensures that returns
from the NS coalition are decreasing in violence.
12. Proof: The right-hand side of Condition (3) is increasing in L and L S but decreasing in LN.
Given our earlier assumption that L = LN + L S, we can disaggregate these differential legiti-
macy effects into 2L S –2LN. In situations where LN > L S, the probability that the subnational
government will cooperate with the cartel is increasing in legitimacy. However, in all cases
where LN ≥ LS this probability is still increasing in national government legitimacy, LN.
13. Proof: Taking the earlier assumption that V = V N + VS, we can rewrite the violence com-
ponents of Condition (3) as [–2cV –1]V N + (1 –cV)VS. Thus, in all situations, the net impact
of violence against the national government reduces the left-hand side value of Condition
(3), decreasing the probability that it will hold. Given our earlier assertion that 0 < cV ≤
1, however, the effect of violence on the subnational government on the left-hand side of
Condition (3) is positive, increasing the probability that the condition will hold.
References
Aguirre, J., and H. A. Herrera. 2013. “Institutional Weakness and Organized Crime in
Mexico: The Case of Michoacán.” Trends in Organized Crime 16, no. 2: 221–38.
Aivazian, V. A., and J. L. Callen. 1981. “The Coase Theorem and the Empty Core.” Journal of Law
and Economics 24, no. 1: 175–81.
Ajzenman, N., S. Galiani, and E. Seira. 2014. “On the Distributed Costs of Drug-R elated
Homicides.” Working Paper Series. Washington, DC: Center for Global Development.
http:// w w w.cgdev.org/ p ublication/ d istributed- c osts- d r ug- r elated- h omicides-
working-paper-364.
Akerlof, G. A., and J. L. Yellen. 1993. “Gang Behavior, Law Enforcement, and Community
Values.” In H. J. Aaron, T. E. Mann, and T. Taylor, eds., Values and Public Policy. Washington,
DC: Brookings Institution Press, chap. 7.
Archibald, R. C. 2010. “Civilians Falling Victim to Mexico Drug War.” New York Times, October
29. http://w ww.nytimes.com/2 010/10/29/world/a mericas/29mexico.html?_ r =4&
Astorga, L., and D. A. Shirk. 2010. “Drug Trafficking Organizations and Counter-Drug Strategies
in the U.S.-Mexican Context.” Center for U.S.-Mexican Studies. San Diego: University of
California, San Diego.
Ballesteros, A., J. A. Restrepo, M. Spagat, and J. F. Vargas. 2007. “The Work of Amnesty
International and Human Rights Watch: Evidence from Colombia.” NGO Monitor. http://
www.ngo-monitor.org/data/i mages/File/evidencefromcolumbia_ feb2007.pdf.
478 Case Studies II
BBC News Online. 2011a. “Mexico Drugs Crisis: Monterrey Shocked by Casino Attack.” BBC
News, August 27. http://w ww.bbc.co.uk/news/world-latin-a merica-14693074.
BBC News Online. 2011b. “Mexico Holds Suspect over Juan Francisco Sicilia Murder.” BBC
News, May 25. http://w ww.bbc.co.uk/news/world-latin-a merica-13550487.
Castillo, J. C., D. Mejia, and P. Restrepo. 2013. “Illegal Drug Markets and Violence in Mexico: The
Causes beyond Calderón.” Economics Department. Universidad de los Andes. Bogotá,
Colombia. http://cddrl.fsi.stanford.edu/sites/default/fi les/143.illegaldrug.pdf.
Chicoine, L. 2011. “Exporting the Second Amendment: U.S. Assault Weapons and the Homicide
Rate in Mexico.” Department of Economics. University of Notre Dame. South Bend, IN.
http://w ww.econ-jobs.com/research/32941-E xporting-t he-S econd-A mendment-US-
Assault-Weapons-a nd-t he-Homicide-R ate-i n-Mexico.pdf.
Cockayne, J. 2010. “Crime, Corruption and Violent Economies.” In J. Cockayne, ed., Ending
Wars, Consolidating Peace: Economic Perspectives. International Institute for Strategic Studies.
Adelphi Series. London: Routledge, 189–218.
Davis, D. 2007. Policing, Regime Change, and Democracy: Lessons from the Case of Mexico.
London: Crisis States Research Center.
de Groot, O. J., and A. Shortland. 2010. “Gov-a rrrgh-nance: Jolly Rogers and Dodgy Rulers.”
Economics of Security Working Paper Series. Berlin: DIW Berlin. http://w ww.diw.de/six-
cms/detail.php?id=diw_01.c.361841.de.
Dell, M. Forthcoming. “Traﬃcking Networks and the Mexican Drug War.” American Economic
Review.
Dube, A., O. Dube, and O. García-Ponce. 2013. “Cross-Border Spillover: U.S. Gun Laws and
Violence in Mexico.” American Political Science Review 107, no. 3: 397–417.
Dube, O., O. García-Ponce, and K. Thom. 2014. “From Maize to Haze: Agricultural
Shocks and the Growth of the Mexican Drug Sector.” Working Paper Series. Vol.
355. Washington, DC: Center for Global Development. http://w ww.cgdev.org/p ub-
lication/m aize-h aze-a gricultural-s hocks-a nd-g rowth-mexican-d rug-s ector-w orking-
paper-355.
Duffield, M. 1999. “Globalization and War Economies: Promoting Order or the Return of
History?” Fletcher Forum of World Affairs 23, no. 2: 21–36.
El Economista. 2012. “A Soldados, Aumento Salarial de 150%.” El Economista, February 19.
El Universal. 2008. “Atentados en Morelia: EU ve ‘narcoterrorismo en cobardes ataques.’”
September 18. http://w ww.eluniversal.com.mx/nacion/162452.html.
Fein, H. 1993. Genocide: A Sociological Perspective. London: Sage Press.
Ferguson, N. T. Forthcoming. “Just the Two of Us? Civil Conflicts, Pro-State Militants and the
Violence Premium.” Terrorism and Political Violence.
Finnegan, W. 2010. “Silver or Lead: The Drug Cartel La Familia Gives Local Officials a
Choice: Take a Bribe or a Bullet.” The New Yorker. May 31, 38–51.
García Tinoco, M. 2012. “Morelia recuerda a víctimas de ataque del 15-S.” Excelsior, September
15. http://w ww.excelsior.com.mx/2 012/0 9/15/nacional/859191.
Grayson, G. W. 2013. “The Impact of President Felipe Calderon’s War on Drugs on the Armed
Forces: The Prospects for Mexico’s ‘Militarization’ and Bilateral Relations.” Carlisle,
PA: US Army War College.
Grossman, H. I. 1991. “A General Equilibrium Model of Insurrections.” American Economic
Review 81, no. 4: 912–21.
Heinle, K., O. R. Ferreira, and D. A. Shirk. 2014. “Drug Violence in Mexico—Data and Analysis
through 2013.” Justice in Mexico Project. University of San Diego, California.
Hernandez, D. 2012. “Mexican Drug War’s Innocent Victims: ‘They Tried to Kill Me and My
Kids.’” The Guardian, August 13. http://w ww.theguardian.com/world/2 012/aug/13/
mexican-d rug-war-i nnocent-v ictims.
Human Rights Watch. 2011. World Report 2011. New York: Human Rights Watch.
Lessing, B. 2012. “The Logic of Violence in Criminal War: Cartel-State Conflict in Mexico,
Colombia, and Brazil.” PhD dissertation, University of California, Berkeley.
Lindau, J. D. 2011. “The Drug War’s Impact on Executive Power, Judicial Reform, and Federalism
in Mexico.” Political Science Quarterly 126, no. 2: 177–2 00.
Magaloni, B. 2006. Voting for Autocracy: Hegemonic Party Survival and Its Demise in Mexico.
Markusen, E., and D. Kopf. 1995. The Holocaust and Strategic Bombing: Genocide and Total War in
the Twentieth Century. Boulder, CO: Westview.
Martinez, C. 2012. “Mexico: The Killing of Innocents by Cartels, Police and Death Squads.”
Borderland Beat, May 28. http://w ww.borderlandbeat.com/2 012/05/mexico-k illing-of-
innocents-by-cartels.html.
Martinez, P. 2013. “En un sexenio, el Ejército mató a 45 civiles inocentes.” Animal Politico.
January 28. http://w ww.animalpolitico.com/2 013/01/en-u n-sexenio-el-ejercito-mato-a-
45-civiles-i nocentes/#axzz2tykAvO3D.
McDougal, T. L., and N. T. Ferguson. 2012. “Land Inequality and Conflict Onset: Cooperative
Game Theoretic Implications for Economic Policy.” Post-2015 Global Thematic Consultation
on Inequalities. http://w ww.worldwewant2015.org/node/2 83321.
McDougal, T. L., D. A. Shirk, R. Muggah, and J. H. Patterson. 2014. “The Way of the
Gun: Estimating Firearms Traffic across the U.S.-Mexico Border.” Journal of Economic
Geography 15, no. 2: 297–327.
Mercille, J. 2011. “Violent Narco-Cartels or US Hegemony? The Political Economy of the ‘War
on Drugs’ in Mexico.” Third World Quarterly 32, no. 9: 1637–53.
Michaelsen, M. M., and P. Salardi. 2014. “The War on Drugs in Mexico and Primary School
Performance.” Department of Economics. Ruhr University Bochum. https://editorialex-
press.com/cgi-bin/conference/download.cgi?db_name=NEUDC2013&paper_id=284.
Moloeznik, M. P. 2013. “Organized Crime, the Militarization of Public Security, and
the Debate on the ‘New’ Police Model in Mexico.” Trends in Organized Crime 16, no.
2:177–9 4.
Naidoo, S. 2000. “The Role of War Economies in Understanding Contemporary Conflicts.”
Global Dialogue 5, no. 2: 1–10.
Nietschke, H. 2003. “Transforming War Economies: Challenges for Peacemaking and
Peacebuilding.” Paper presented at the International Peace Institute. December 31,
New York.
Pugh, M., and N. Cooper. 2004. War Economies in a Regional Context: Challenges of Transformation.
Boulder, CO: Lynne Rienner.
Redmond, H. 2013. “The Political Economy of Mexico’s Drug War.” International Socialist Review
90. http://isreview.org/issue/9 0/political-economy-mexicos-d rug-war.
Ríos, V. 2012. “How Government Structure Encourages Criminal Violence: The Causes of
Mexico’s Drug War.” PhD dissertation. Harvard University, Cambridge, MA.
Ríos, V. 2014. “Security Issues and Immigration Flows: Drug-V iolence Refugees, the New
Mexican Immigrants.” Latin American Research Review 49, no. 3: 199–217.
Robles, G., B. Magaloni, and G. Calderón. 2013. “The Economic Costs of Drug-Trafficking
Violence in Mexico.” Working Paper Series. Stanford, CA: Stanford University. http://
cddrl.fsi.stanford.edu/povgov/publications/t he_economic_costs_of_d rugtrafficking_
violence_ i n_ mexico.
Seelke, C. R., L. S. Wyler, J. S. Beittel, and M. P. Sullivan. 2011. “Latin America and the
Caribbean: Illicit Drug Trafficking and U.S. Counterdrug Programs.” Report R41215.
Washington, DC: Congressional Research Service.
Shortland, A. 2011. “The Puntland Pirate Economy.” Paper presented at the Jan Tinbergen
European Peace Science Conference, Amsterdam, Netherlands.
Skaperdas, S. 2001. “The Political Economy of Organized Crime: Providing Protection When
the State Does Not.” Economics of Governance 2, no. 3: 173–2 02.
480 Case Studies II
Skaperdas, S., and C. Syropoulos 1995. “Gangs as Primitive States.” In G. Florentini and S.
Peltzman, eds., The Economics of Organized Crime. Cambridge, UK: Cambridge University
Press, 61–82.
Snyder, R., and A. Duran-M artinez. 2009. “Does Illegality Breed Violence? Drug Trafficking
and State-Sponsored Protection Rackets.” Crime, Law, and Social Change 52, no.
3: 253–73.
Straus, S. 2001. “Contested Meanings and Conflicting Imperatives: A Conceptual Analysis of
Genocide.” Journal of Genocide Research 3, no. 3: 349–75.
Thomson Reuters. 2015. “Iguala Massacre: Mexico Says 43 Students Were Murdered on Cartel
Orders.” http://w ww.cbc.ca/news/world/iguala-massacre-mexico-says- 43-students-were-
murdered-on-cartel-orders-1.2934374.
Ulfelder, J., and B. Valentino. 2008. “Assessing Risks of State-Sponsored Mass Killing.” Working
Paper. http://d x.doi.org/10.2139/ssrn.1703426.
USA Today. 2014. “Drug Cartel Makes More Mining than Trafficking.” USA Today. March
17. http://w ww.usatoday.com/story/news/world/2 014/03/17/cartel-dealing-i ron-ore/
6512167/.
Vilalta, C. 2014. “How Did Things Get So Bad So Quickly? An Assessment of the Initial
Conditions of the War against Organized Crime in Mexico.” European Journal on Criminal
Policy and Research 20, no. 1: 137–61.
Wilkinson, T. 2011. “Mexican Drug Traffickers Undermine Elections.” Los Angeles Times,
November 14. http://a rticles.latimes.com/2 011/nov/14/world/la-fg-mexico- michoacan-
elections-2 0111114.
21
Long-Term Economic Development in the

Presence of an Episode of Mass Killing
The Case of Indonesia, 1965–1966
S. M a nsoob M u r sh e d a n d Moh a m m a d Zu l fa n Ta djoe ddi n
21.1. Introduction
Indonesia is sadly among the list of nations that have experienced episodes of
mass killings. Following its independence from Dutch rule in 1949, Indonesia
was ruled until 1998 by two strongmen: first by the charismatic leader Sukarno
(until 1966), then by Suharto (1966–1998).1 The mass killing around which this
chapter is structured—t he politicide of 1965–1966—was carried out during the
transition of power from Sukarno to Suharto, which took place via a coup d’ état.2
Sukarno had been a staunch nationalist and a leader in the global movement of
nonaligned nations, and so he was of necessity suspect to the West. The coup and
subsequent pogrom eliminated a left-w ing challenge to his rule and helped to
consolidate Suharto’s grip on power.
Section 21.2 sketches the lead-up to these events in Indonesia and describes
how they unfolded in different parts of the country. In this context, it has to be
borne in mind that the 1960s were a high point in the cold war, and the “domino
theory” fed angst about the spread of communism in East Asia. Section 21.3 out-
lines a model of why and how individuals may participate in an episode of mass
atrocity (with model details deferred to an Appendix). Section 21.4 discusses the
pretext, context, and aftermath of the politicide within the context of Indonesia’s
overall economic development, all the way through the end of Suharto’s regime in
1998 (augured by the 1997 Asian financial crisis and resulting in its own episodes
of mass killings in Aceh and East Timor, respectively). Section 21.5 concludes.
Anderton (2010) develops a rational choice model where genocide (or politi-
cide) is a strategy chosen by an aggressor group in a game of power, based on
expected payoffs. In the model, expected returns from eliminating a persistent
rival are compared with the expected cost of concessions or compromise. The
481
482 Case Studies II
expected returns can incentivize mass killing as a dominant strategy, the attrac-
tiveness of which can rise with imperfect information, indivisibilities, enforce-
ment costs of peaceful-sharing agreements, and the long shadow cast by the
disutility of the antagonist’s future existence. This choice-t heoretic framework for
the aggressor group’s leadership is as it should be, but it still raises the question as
to why individual perpetrators of genocide within the aggressor group choose to
participate in the first place. Anderton (2010) sheds light on this issue by catego-
rizing individuals in the aggressor group as hard-liners (eager to commit geno-
cide), bystanders, or resisters (to mass murder). Individual choices are based on
payoffs and costs imposed by stronger members of the group. But there could be
other—deep, innate—behavioral factors that drive individual decisions to par-
ticipate in genocide, and we sketch these in an identity-based behavioral model in
section 21.3 of this chapter (again, with details deferred to the Appendix).
Economics and politics are in practice inseparable, and the run-up to mass
murder always has economic underpinnings, as does its aftermath. In the case
we are interested in, serious economic mismanagement, a stagnant economy, and
poverty bordering on famine characterized the period between 1957 and 1966 in
Indonesia. It would be fair to say that per capita gross domestic product (GDP)
was lower in Indonesia than in most sub-Saharan African economies at the time
(see Myrdal 1968). This raises the question of the distributional and macroeco-
nomic consequences for Indonesia of the politicide of the mid-1960s: How did
an episode of mass killing “play out” in terms of economic development in the
decades following the event?
In the aftermath of mass killing, per capita income can rise in agrarian econo-
mies characterized by surplus labor provided there is not much infrastructural
destruction, as is shown to have been the case in Europe from the fourteenth to
the seventeenth centuries (Voigtländer and Voth 2013; also see chapters 4 and 5
in this volume). But will a country’s leadership always have an interest in grow-
ing the economy following a climactic experience of mass murder? The question
becomes more complex in the presence of natural resource rents such as crude oil
revenues, which began to accrue to Indonesia even as its economy in the 1960s
remained dominated by agriculture, particularly in the densely populated island
of Java. In resource-r ich economies, a state’s leadership can have an incentive to
diversify the country’s economic base in the face of volatile resource rents. If suc-
cessful, leaders can avoid the so-called Dutch disease, become less dependent on
natural resource exports, and make the economy-at-large grow more broadly and
evenly (Dunning 2005). 3 Occasionally, leaders may even have a direct interest
in reducing regional disparities and poverty. These factors appear to have been
present in Suharto’s postpoliticide “New Order” regime. We describe and analyze
these factors in some detail in section 21.4, which ends with a brief, speculative
counterfactual analysis of where Indonesia might be today, economically, had
there been no regime change in the mid-1960s.
L o n g -Te r m E c o n o m i c D e v e l o p m e n t 483
In short, we present a theory of motivation to participate in violent collective

action based upon the utility individual participants derive from group identity,
and also examine how a hate message sent by self-seeking leaders can spur peo-
ple into action, all in the context of Indonesia’s poverty and economic decline at
the time of the event. We then look at the politicide’s economic aftermath for the
duration of Suharto’s three-decade-long regime.
21.2. The Indonesian Politicide of 1965–1966

The Indonesian mass killing of communists, communist sympathizers, and sus-
pected communists took place mainly between October 1965 and March 1966.
At the time, this six-month-long period was the second-most momentous moment
in Indonesian history, the first being the war of independence in the late 1940s.
(The third would be the fall of Suharto in May 1998.) With regard to fatalities of
the mass killing, most scholars cite figures in the region of half a million deaths,
with estimates varying between a hundred thousand to two million (Cribb 1990,
2001). The genocide and mass killing datasets reviewed in c hapter 3 of this vol-
ume report fatality estimates for Indonesia ranging from 375,000 to 750,000
(Easterly, Gatti, and Kurlat 2006). The episode spelled the end of the Indonesian
Communist Party (Partai Komunis Indonesia, PKI). The killing was nationally
orchestrated by the army and supported by different elements in each locality,
mainly Muslims and nationalists. It was concentrated in East and Central Java,
where the PKI had its strongest base. Large-scale massacres also occurred in Bali,
North Sumatra, and West Java. Killing on a smaller scale erupted in the outer
islands of Sumatra, Kalimantan, Sulawesi, and Nusatenggara. Unfortunately, the
mass killing itself has received far less academic scrutiny than did its pretext, to
which we now turn.
21.2.1. The Pretext

The pretext for the communist purge was the 30 September Movement, a one-day
affair that took place in the Indonesian capital, Jakarta. In the early morning on
October 1, 1965, six high-ranking army generals and a lieutenant were abducted
and murdered by a group of conspirators within the Indonesian army. Consisting
of young, progressive officers led by a lieutenant colonel, the group announced
that they had acted to safeguard President Sukarno from a coup planned, they
said, by a right-w ing council of generals. By the early evening of that day, the
Movement was put down by the army under the leadership of the senior surviv-
ing army general, Major General Suharto, and the mass killing ensued. It was the
start of Sukarno’s loss of grip on the presidency and Suharto’s eventual seizure of
state power. The assassination of six army generals was more than simply a pretext
484 Case Studies II
as it had a lingering, transformative effect on Indonesian politics for decades to

come, even to the present day.
The Movement could be seen as an internal army affair, a form of mutiny. The
group of left-leaning young officers received political support from the chairman
of the PKI, who secretly discussed the issue with a handful of the party’s polit-
buro members. The 30 September Movement was not a policy of the communist
party as it had never been discussed in the party’s central committee (Roosa
2006). Nonetheless, the PKI as an institution was blamed by the army for puppet-
mastering the Movement; therefore, or so it seemed to follow, the party had to
be destroyed down to its roots. This was the rationale for the ensuing mass kill-
ing. Roosa (2006) concludes that the Movement was in fact a mutiny: a purge of
the army’s high command by lower-ranking military officers. If the Movement
was not a coup, the reaction to the Movement by the army under the leadership
of General Suharto, who eventually seized power from Sukarno in March 1967,
could be seen as a coup. They used the Movement as a pretext to break the power
of both the PKI and Sukarno.
The head-to-head collision between the PKI and the army was rooted in the
years preceding the Movement, dating back to the start of Sukarno’s so-called
Guided Democracy in 1957. Although Indonesia was one of the initiators of the
Non-A ligned Movement, Sukarno showed his increasingly anti-Western policy
and moved Indonesia toward a Hanoi-Peking-Moscow alliance. Sukarno was
supported by two competing dominant powers around him, the largely anticom-
munist army and the Indonesian Communist Party, the PKI.
In 1965, the PKI was at the peak of its political performance. It was the largest
and the most organized political party in the country with a very strong cadre
system and grassroots base. It secured fourth position with 16 percent of the vote
in the 1955 national elections, won a majority of the Javanese votes in the 1957
regional elections (even though with only 27 percent of the vote), and enjoyed
Sukarno’s political support. Sukarno was at the height of his grand idea to bring
together three dominant political forces: Nationalist, religious (Islamist), and
communist. The PKI was the largest communist party outside the Soviet Union
and China. At the time, many believed that the PKI would win the plurality of
votes if elections were held again (Roosa 2006).
The army was the only viable force that challenged the growing PKI influence
nationwide. Sukarno, who was not able to fully control the army, used the PKI to
keep a balance. The army was aware that Sukarno was not on their side but did not
directly challenge him because of his very strong political base. Instead, it waited
for the right moment to do so, and the 30 September Movement provided the
perfect opportunity.
Indonesian society and politics in the mid-1960s were deeply polarized
between pro-and anticommunists. With such a macrosetting for the mass killing,
Roosa (2006, 224) concludes that “we are dealing with a boxer who not only
knocks out his opponent in the ring but goes on to attack all of that boxer’s fans
in the stadium, then hunts down and attacks his opponent’s fans throughout the
country, even those living far away who had not even heard about the match.”
This boxing analogy may be spurious as Roosa tries to suggest that the politics of
Guided Democracy was as normal as, say, US politics. Instead, the army-versus-
PKI conflict was a contest for the soul of Indonesia, a struggle in which both sides
were vicious and ruthless. It had its roots in ideologically irreconcilable differ-
ences of nationalists, Islamists, and communists that brought Indonesian par-
liamentary democracy to an end in 1957 when Sukarno introduced the Guided
Democracy.
21.2.2. The Mass Killing

The mass killing of suspected communists was centrally orchestrated by the army,
where the army’s Special Force unit (Resimen Para Komando Angkatan Darat,
RPKAD) played a central role. It began by quickly putting down the Movement
in Jakarta on the evening of October 1, 1965. The mass killing throughout the
country then followed. Although orchestrated by the army, the role of civilians
was crucial to the massive scale of the killing. The civilians who participated in
the killings were usually members of militias who had received training from the
military, along with weapons, vehicles, and assurances of immunity. The mass
killing was accompanied by mass detentions of suspected communists, mostly
without trials. In the mid-1970s, it was estimated that around one million were
detained for alleged involvement with the PKI (Friend 2003).
In Central Java, the RPKAD played a leading role in organizing anticommu-
nist violence. Its commander, Colonel Sarwo Edhie Wibowo, joined his troops in
the capital city of Central Java, Semarang, on October 19. On that night, RPKAD
detained more than 1,000 people and encouraged youths from various religious
groups (Nahdatul Ulama, Muhammadiyah, and the Catholic Party) and a nation-
alist group (Indonesian National Party, PNI) to attack communist and Chinese
buildings in Semarang. The pattern of the RPKAD operation—mass detention
and rioting by anticommunist groups against PKI and ethnic Chinese—was
soon repeated in towns throughout Central Java and Yogyakarta. Due to limited
military personnel to carry out the mission, RPKAD trained and armed religious
and nationalist militias. The operation in Central Java was closely monitored
by General Suharto, who toured the province in mid-November (Jenkins and
Kammen 2012).
The story of the mass killing in East Java is a little different, although here,
too, the army played a key role. Much of the violence was carried out by mem-
bers of the traditionalist Muslim mass organization, Nahdatul Ulama (NU)
486 Case Studies II
(Fealy and McGregor 2012). The violence was the outcome of heightened
polarization between Islamists and communists in Indonesian society. East
Java was the strongest base of the NU, but it faced a growing challenge from the
PKI that posed a big threat to the socioeconomic standing of NU elites. While
the PKI represented landless farmers, peasants, and the poor, the NU was
backed by big landowners and merchants. Fealy and McGregor (2012) argue
that political and socioeconomic factors were more important than religion,
although, among NU followers, religion was often used as a central justifica-
tion of the killing.
By the end of 1965, it is estimated that around 100,000 people had been mur-
dered and 70,000 detained in Central Java, while some 200,000 people had been
massacred and some 25,000 detained in East Java (Kammen and McGregor
2012). In Bali, supporters of the nationalist PNI party conducted most of the
killing. The arrival of the Special Force in December 1965 coincided with the
rapid intensification of the killing in the province. Between December 1965 and
February 1966, the mass killing took the lives of some 80,000 Balinese represent-
ing 5 percent of the island’s population (Robinson 1995). Bali suffered the highest
per-capita rate of killing in the events of 1965–1966. In West Kalimantan, the
army used civilian paramilitaries to proscribe communists, creating an intereth-
nic pogrom: ethnic Dayaks were mobilized to hunt down ethnic Chinese in rural
areas where around 100,000 were expelled and some 3,000 killed (Davidson and
Kammen 2002). Elsewhere, numbers of casualties were far lower than in East
Java, Central Java, and Bali. In North Sumatra, it is estimated that some 15,000
were killed and another 15,000 detained, with some anti-Chinese elements driv-
ing the killing as in West Kalimantan. In West Java, less than 10,000 were killed
and 10,000 detained. A few thousand were killed in South Sulawesi (Kammen
and McGregor 2012).
The politicide of (suspected) communists was the main part of the overall
attack on the Indonesian left (including Sukarno as well as his social base and
ideals), but it should also be viewed from the perspective of a larger process of
reintegrating Indonesia into the capitalist world economy: The PKI’s destruc-
tion was welcomed and supported by the West (Simpson 2008). Even more
generally, the mass killing must also be situated within the global polarization
in the context of the cold war between the United States–led Western capital-
ist bloc and the Soviet-led communist bloc. The basic battle took place in the
newly decolonized, developing world, and Indonesia was very much a part of
it. In Indonesia, as in other newly independent countries in post–World War
II Asia and Africa, left-leaning nationalistic policies were a dominant feature.
The reaction by the West was often to encourage Western-oriented military
dictatorships to take power. Thus, Suharto’s rise to power in Indonesia was
similar to the stories, for instance, of Mobutu in then-Z aire and of Pinochet in
Chile (Schmitz 2006).
21.3. Identity and Motivation to Participate

in Genocide
In this section and in the Appendix we provide a choice-theoretic framework
grounded in the economics of identity that we believe is relevant for under-
standing aspects of mass atrocity, including the mass killings in Indonesia. Our
objective is to shed light on why individuals may take part in nonconformism,
extremism, and even mass murder. We emphasize that we posit a spectrum of
activity, of which violence is an extreme aspect. Our work builds on the seminal
contribution by Akerlof and Kranton (2000) on the economics of identity in gen-
eral, and on Murshed (2011) on modeling conflict-related identity issues specifi-
cally. Similar work, but with a focus on why individuals choose to become suicide
bombers, can be found in Iannaccone (2006) and Wintrobe (2006).
We begin with an individual for whom identity is an important aspect of the
actions that he or she takes, including, in an atrocity context, actions to support
or resist a leadership elite that seeks to destroy an out-g roup. Assume that an
individual, i (call him “Bob”), has an identity or self-image, Ii. This identity will
depend, in part, on many types of actions Bob can take, ai. The bold-font nota-
tion indicates a range, or vector, of actions such as buying ethnic food, attending
a political rally, or responding to a call for prayer. Also assume that these actions
are taken in the context of Bob’s self-i mage as a member of a principal group (the
in-g roup), such as an ethnic group, a religious organization, or a political party.
Bob’s group-related identity not only depends on his own actions but also on the
multitude of actions taken by similar-m inded people in the group, aj. The letter j
represents all the other people (j = 1, …, n; i ≠ j) who, like Bob, tether their iden-
tity to the same principal group.
Now consider how Bob’s identity can be affected by a hate parameter, μ. The
introduction of this hate parameter has similarities with Boulding’s (1956) con-
cept of image. Boulding regards image to be the basis of all behavior. Image,
including self-i mage, is always subject to messages, akin to signals, which can be
absorbed and internalized or else lead to changes in the image that, on occasion,
can be quite dramatic or revolutionary. In our context of mass atrocity, the hate
parameter is subject to manipulation by a leadership elite. Taken together, these
ideas can be summarized in mathematical shorthand by writing i’s identity as
Ii = Ii(a i, aj, μ), (1)
that is, the individual’s own actions (ai), the actions of others in the principal group
(aj), and the effect of a hate parameter on the formation of the self-i mage (μ).
We now postulate that an individual’s group-related utility or life satisfaction,
Uig , can be written as
488 Case Studies II
Uig = Uig [a i, a j, Ii(a i, a j, μ)] . (2)
In words, an individual i obtains group-related satisfaction directly through his

own and others’ actions—ai and aj—and also indirectly through the effects of how
such actions, and the hate parameter, affect and may modify i’s identity, Ii.
Next, we extend Akerlof and Kranton (2000) by permitting an individual to
possess multiple identities (Sen 2008). Specifically, assume that our individual
(Bob) has some other important identity and that Bob also takes actions that are
tied to this other, additional identity. For example, while Bob’s principal group
might be a religious organization, he also derives value from his identity as a citi-
zen in the state in which he lives (he is a “good citizen”). Let k i symbolize the
vector of actions that i (Bob) takes that are tied to his second (additional) iden-
tity. We postulate that Bob’s utility from such actions in relation to the additional
identity are represented by the function Uio(k i ). Assume further that Bob’s princi-
pal group and other identity utilities are additive and separable. This leads to the
following multiple-identity utility function for individual i:
Uitotal = Uig [a i, a j, Ii(a i, a j, μ)] + Uio(k i) . (3)
where Uitotal symbolizes i’s total utility or life-satisfaction from actions taken that
relate to the principal group as well as actions taken that relate to other, additional
identities.
The model in (3) contains as a special case the standard utility function of neo-
classical economics because the actions taken can simply represent the regular
goods (e.g., food, clothing) that an individual purchases, and even the regular
goods that others purchase, which may have spillover benefits (e.g., network exter-
nalities) and/or spillover costs (e.g., noise pollution) to i. But the economics of
identity incorporates much more than such regular economics. In (3), Bob’s prin-
cipal group identity and his second (additional) identity affect the overall utility
level that Bob achieves. The consumption of typical goods (e.g., food, clothing),
but also of nontypical goods (e.g., political rallies, actions to enhance self-i mage
in social contexts, hatred), occur in the context of identity. As in regular econom-
ics, Bob, as a utility maximizer, will be interested in the goods he consumes; but,
additionally, he is a social being and presumably cares about his self-i mage, about
fitting into his principal group, and about peer pressure and other identity-related
concerns. This expanded utility-maximizing framework can have important
implications for understanding mass atrocity. For technically interested readers,
in the Appendix we formally derive key results related to mass atrocity based on
the model begun here. Meanwhile, we turn to an intuitive description of how key
economic aspects of identity manifested themselves in the Indonesian mass kill-
ings of 1965–1966.
21.3.1. Principal Identity and Conformity to Genocide

It is important to note that the utility function in (3) applies to an individual, spe-
cifically to individual i (whom we have called “Bob”). Therefore, each individual
whom we might consider in a principal-identity grouping (e.g., Suharto politi-
cal supporters, Muslims, communists) would have their own utility function,
which would specify how their own actions, the actions of others, and the hate
parameter affect each individual’s self-i mage and utility from being aligned with
g
the principal group. This is captured by the Ui function, the group-related func-
tion. Each individual whom we consider would also receive utility from a second
identity group such as being a citizen of the Indonesian state. This is captured by
the Uio function.
Consider now, by way of example, Indonesian Muslims of the mid-1960s who
see their Muslim group as their principal identity. Some individuals from this
group can be assumed to be staunchly anticommunist and others less so, includ-
ing some who may not be anticommunist at all. Put differently, within the Muslim
group there exist varieties of sympathies and antipathies toward outsiders such as
communists. Following Akerlof and Kranton (2000), it is possible to show that
some individuals of the in-g roup derive disutility from nonconforming actions
(or beliefs) of other in-g roup members. Moreover, if the cost of inflicting pain on
“errant” members is low, as compared to the pain so inflicted, individuals from
the principal group will exert effort to bring members who have strayed from
some ideal-t ype in-g roup behavior back into the fold (see Akerlof and Kranton
2000). For example, the cost of ostracizing someone may be smaller than bearing
the pain of being ostracized. Further, conflict entrepreneurs within the in-g roup
(e.g., Muslims) who seek mass murder against an out-g roup (e.g., communists)
can manipulate individuals’ utility functions to foster group conformity to their
confrontational goals. It can be shown that principal group members would tend
to conform to ever more confrontational, and even murder-supporting, group
behavior when individuals from the principal group live in close proximity to
one another, suffer from poverty, and have low human capital (educational)
endowment. These, indeed, are the conditions in which many Muslims lived in
Indonesia in the mid-1960s. Under such conditions, leaders from the Muslim
group supporting atrocity may use the identity-oriented utility functions of indi-
viduals from the group to overcome intragroup mistrust and the collective action
problem described by Olson (1965). Group grievances become individual grievances
that individuals then act on. This, at the extreme, can induce genocidal acts simi-
lar to choices made about suicide bombing as outlined in Wintrobe (2006).
We now turn to the role of hatred, represented by the parameter μ in (3). It
denotes a broad spectrum of like or dislike of the “other” that can range from lik-
ing the other (μ < 0) to indifference (μ = 0) to mild dislike (μ is low) to extreme
hatred (μ is very high); and the last of which can induce homicidal behavior.
490 Case Studies II
A high μ implies more confrontational attitudes against members of the out-

group including, at the extreme, genocide or politicide. Following Glaeser
(2005), we can think of μ as originating in a signal sent out by a leader, say a
politician or a group of military officers. As a way of consolidating his hold on
power, a leader may encourage others to eliminate opponents by deliberately
sending out a false signal. Its attractiveness to individuals in the in-g roup will
depend on their need for scapegoats and their own personal life experiences
with people from the out-g roup. Not all such signals will be believed. For exam-
ple, some hate-mongering politicians may be mistrusted; the better educated
among the public may discount part of the message; and those with greater
knowledge of the “other,” based upon personal interaction, for example, may
similarly disregard such signals. Nevertheless, if, in conjunction with identity-
driven incentives for conformity, a sufficiently large number of individuals from
the in-g roup, the critical mass, believe such signals, and if they allow them to
shape their self-i mages and utilities from being part of the principal group, then
even false signals can be effective in fostering support for mass atrocity. For
the case of the Indonesian politicide of the mid-1960s, we postulate that some
Muslim leaders sent out signals that, under conditions such as those identified
before (e.g., poverty, low education, close proximity of group members, low cost
of enforcing conformity, and the like), were effective in consolidating identity-
based behavior within the in-g roup such that extreme actions, including mass
murder, were tolerated and even endorsed against an out-g roup.
21.3.2. Principal Identity and Resistance to Genocide

Although not a focus of this chapter, the same economics of identity theory
described intuitively in the previous subsection, and laid out formally in the
Appendix, can be “turned on its head” and used to analyze how a principal iden-
tity group can become resistant to genocide. We can imagine, for example, how a
group might have entrepreneurial leaders who send out signals of nonhate (μ = 0)
or even of out-g roup support (μ < 0) and in which conformity within the in-g roup
is fostered along lines in which its individuals have incentives not to designate
members of the out-g roup as “other.” Such leadership and identity incentives
can be “subsidized” by potential third-party interveners seeking to prevent mass
atrocity (e.g., by providing counter-hate signals, economic development assis-
tance, and interactions between in-and out-g roup members). We suggest this as a
line of enquiry for future research (also see c hapters 12, 19, and 22 in this volume).
21.4. The Aftermath of Mass Atrocity

An instance of a general trend in the developing world, Indonesia’s early
postindependence era was often characterized by political instability, violence,
poverty, and poor economic growth (van Zanden and Marks 2012). The coun-
try was preoccupied with politics, not economics (on the sequence of politics
and economics, see c hapter 28 in this volume). According to Tan (1967, vii),
the heart of the problem lay with President Sukarno’s inability to grasp the
saliency of sound economic management: “Sukarno was accustomed to live
with the conviction that, economically illiterate as he was, the President should
not be held responsible for the economic wellbeing of the nation. Such respon-
sibility should rest with his team of economic advisers.” In January 1967, a
cabinet minister frankly declared that “one of Sukarno’s sins is his ignorance
of economics” (Tan 1967, viii).
Indonesia’s postindependence years through to the mid-1960s can be divided
into two periods. The first is the era of parliamentary democracy (1949–1957),
wherein Sukarno largely played the role of symbolic solidarity-maker while
the executive was headed by a prime minister. The second is the era of Guided
Democracy (1957–1965), wherein Sukarno turned into a dictator.
21.4.1. Parliamentary Democracy (1949–1957)

The newly independent Indonesia inherited a colonial economic structure, a
classic dual-economy system. It consisted of modern, capitalist economic enter-
prises run by Dutch colonists, mainly in the form of plantation agriculture and
of shipping and trading companies, and a traditional, subsistence-based peasant
economy that operated alongside a weakly developed indigenous merchant class
(Boeke 1953). The two were hardly linked except from the perspective of labor
supply.
Economic performance, however, was tolerable during 1949–1957; total GDP
and per capita GDP annually grew at 5.5 percent and 2.9 percent, respectively
(see Table 21.1). The dual economy largely continued during this period, where
the modern colonial capitalist enterprises significantly contributed to growth.
While the government tried to expand its role in the modern economic sector,
the successive prime ministers from different parties during the period were very
cautious about any policies to nationalize foreign (colonial) enterprises owing to
potential damage the policy might have.
21.4.2. Guided Democracy (1957–1965) and the Economic

Context of Mass Atrocity (1965–1966)
The situation significantly changed in 1957 when Sukarno started to become a dic-
tator with his vision of a Guided Democracy (and Guided Economy). During the
1950s, a series of measures were taken to reduce the Dutch predominance in the
Indonesian economy, culminating in the unilateral nationalization of all Dutch
enterprises in 1957–1959, and followed by the takeover of British and American
economic interests in 1963–1965. In fact, between 1957 and 1965, nearly all
492 Case Studies II
foreign-owned enterprises were taken over by the state (Gibson 1967). These
state-owned enterprises were run by military officers not familiar with running
commercial enterprises efficiently. Parallel to the effort to reduce the remaining
colonial economic role, measures to limit the economic roles of ethnic Chinese
people were also part of the creation of the national economy. This ultranational-
ist economic policy of nationalization destroyed the productive capacity of the
already thin modern economic sector of the country, akin to Robert Mugabe’s
move to nationalize European-owned plantations in Zimbabwe.
These two measures, nationalization of Dutch enterprises and limiting the busi-
ness dominance of ethnic Chinese, significantly contributed to the deterioration
of the Indonesian economy during the course of Sukarno’s Guided Democracy as
it lacked an off-setting emergence of indigenous Indonesian entrepreneurs. The
first affirmative policy of the Benteng (fortress) program in the 1950s, aimed at
promoting indigenous Indonesian entrepreneurs, was unsuccessful. Indonesian
economic policies in the 1950s could be characterized as socialist-leaning eco-
nomic nationalism as commonly found in most of the newly decolonized coun-
tries in Asia and Africa after World War II.
Writing in the 1960s, although his main focus was India, Gunnar Myrdal
was pessimistic about Indonesia’s development prospects (Myrdal 1968).
He believed that traditional power structures were likely to endure and that
chances for economic takeoff were slim. This was largely because government
was unable to impose the discipline needed to implement development plans.
Myrdal concluded that democracy might not be the best system to achieve
the desired development progress and that authoritarian regimes might do it
better. In fact, Sukarno’s authoritarian turn to Guided Democracy, after the
country’s initial experiment with Western-style liberal democracy during the
early 1950s, was in line with Myrdal’s assessment, albeit with an anti-Western
tone and rather narrow nationalistic orientation. In contrast, Suharto’s
regime, which ruled the country after the mass killing, while a continuation
of Sukarno’s authoritarian style, had a clear Western economic orientation. It
was open to foreign capital, investing mainly in natural resource exploitation,
and, perhaps more importantly, it harnessed the ethnic Chinese-based entre-
preneurial talent pool.
Politically, Sukarno’s Guided Democracy preserved the unity of the country.
In the early 1960s, a series of regional rebellions were put down and Sukarno
successfully integrated West Papua, the last remaining former Dutch colony in
Southeast Asia, into Indonesia. However, the period was a total economic failure
as shown in Table 21.1. The decline in GDP growth and the decline in growth rates
across all sectors during 1957–1966 are dramatic. In 1965, Indonesia counted
among the world’s poorest economies, with a highly dominant agricultural sec-
tor and a very small manufacturing industry (see Table 21.2). Moreover, between
Table 21.1 Sectoral Real GDP Growth (Annual Average Percent), 1949–1957
and 1957–1966
Sector 1949–1966 1949–1957 1957–1966

Agriculture 2.5 2.4 2.7
Manufacturing 5.1 10.9 0.3
Oil and gas 8.4 12.9 4.5
Trade 3.8 5.9 2.1
Transport 2.4 7.3 –1.7
Government 0.4 6.8 –4.9
Other services 3.0 4.5 1.8
Total services 3.0 5.5 0.9
Total GDP 3.5 5.5 1.8
Non oil/gas GDP 2.2 4.3 0.4
GDP per capita 1.0 2.9 – 0.6
Source: van Zanden and Marks 2012, 151, Table 7.5.
Table 21.2 Indonesia in 1965 in a Comparative Perspective
GDP Per Capita (1990 Trade Agriculture

International Dollars) (as % of GDP) (as % of GDP)
Indonesia 983 24 53
South Korea 1,436 27 36
Malaysia 1,804 80 29
Taiwan 1,810 41 24
Thailand 1,308 35 32
South Asia 763 21 40
Sub-Saharan 1,099 43 39
Africa
Latin America 3,709 36 21
1953 and 1967, the central government ran budget deficits each year. Tables 21.1
and 21.2 indicate how economic decline and malaise correlated with mass atroc-
ity. As in the Armenian and Rwandan cases, catastrophic economic conditions
seem to accompany increased risk of mass atrocities.
494 Case Studies II
21.4.3. The Post-1966 Period

After totally eliminating the political left from Indonesian sociopolitical life,
Sukarno’s grand idea of accommodating the country’s three dominant social
forces— nationalist, religious (Islamist), and communist— was further dis-
mantled even as it kept an artificial façade of democracy. When Suharto finally
took power, informally in March 1967 and formally in March 1968, his military-
centered authoritarian regime was fully aware of continuing political weaknesses
embedded in the regime. Thus, nationalists and Islamists were grouped into two
new parties, and Suharto’s regime created another, Golongan Karya (GOLKAR).
Henceforth, only these three political parties (GOLKAR, nationalist, and
Islamist) were allowed to contest elections, with GOLKAR enjoying full support
of the military and state bureaucracy and designed to win every election. This out-
come was ironic, given the support that Islamist and nationalist militias provided
to the army in conducting the mass killing.
The capitalist- oriented military dictatorship concealed within a pseudo-
democracy did offer the country a modicum of political stability. Although com-
pelling the country to accept authoritarian rule, Suharto in return delivered a
broad-based increase in socioeconomic prosperity across the archipelago, a situ-
ation that Liddle (1999) labeled a Hobbesian bargain. Between the late 1960s
and mid-1990s, average per capita income more than quadrupled, the poverty
rate dropped from 70 percent to only 13 percent, infant mortality dropped from
159 to 49 per thousand live births, the adult illiteracy rate fell from 61 percent to
14 percent, and inequality, measured by the Gini coefficient of expenditure, was
broadly stable, varying between 32 and 35, which is a low level of inequality by
international standards (UNDP 2001).
The relatively remarkable socioeconomic performance of Suharto’s military
dictatorship, despite its highly corrupt nature, merits further explanation. First, it
reflects the overall superiority of an outward-looking liberal capitalist economic
system vis-à-v is an inward-looking socialist central-planning approach to devel-
opment. The former largely characterized economic policies in the context of the
East Asian miracle, while the latter can be associated with the poor economic
performances of pre-1979 reformed China and pre-1991 reformed India. Second,
authoritarian political stability, albeit at the expense of political rights, anchored
the economy under a strong state and without serious political turbulence. This
seems to echo Myrdal’s preference for an (effective) authoritarian regime allow-
ing for economic gains. Third, Suharto’s decision to appoint capable technocrats
to run the economy played a major role in the regime’s economic successes. The
technocrats were able to manage the economy in relative isolation from day-to-
day politics. Fourth, the crude oil boom in the 1970s provided fresh financing
for development, which helped to avoid the Dutch disease. Fifth, after usurping
power, Suharto’s main economic priority was to implement stabilization and
Table 21.3 Indonesia’s Sectoral Economic Growth (Annual Average Percent),

1967–1996
Sector 1967–1996 1967–1972 1972–1980 1980–1986 1986–1996

Agriculture 4.1 5.2 4.7 3.3 3.5
Manufacturing 10.8 12.7 10.3 9.1 11.3
Oil and gas 3.7 16.5 4.8 –3.7 1.6
Trade 7.6 8.9 7.8 4.4 8.9
Transport 7.1 5.8 7.3 6.4 8.1
Government 10.7 20.5 17.5 6.2 3.7
Other services 6.8 5.7 5.7 5.6 8.9
Total services 7.4 7.7 8.0 5.4 8.0
Total GDP 6.9 10.2 6.8 3.8 7.6
Non oil/gas GDP 7.4 8.0 7.3 5.3 8.4
GDP per capita 5.2 6.5 4.7 2.8 6.5
rehabilitation programs aimed at providing guidelines for Indonesia’s economic

recovery and specific policies on a balanced budget, the balance of payments,
rehabilitation of physical infrastructure, and agricultural development. These
were helped by foreign investment, foreign aid, and the overall international ori-
entation of the economy. Growth performance in the earlier phase of Suharto’s
New Order proved to be the most impressive (see Table 21.3). In comparative
perspective, the New Order was regarded as an economic success story (World
Bank 1993; ADB 1997).
21.4.4. Diversification and Redistribution in

the Post–Mass Atrocity Economy
The post-1996 trend, hinting at a possible economic supremacy of an authoritar-
ian style of government, should not, however, be taken as a generic pattern true
of all Western-backed military dictatorships. Telling examples are the contrast-
ing experiences of Suharto’s Indonesia and Mobutu’s Zaire (Dunning 2005).
In Indonesia, after eliminating the political left and sidelining Islamists and
nationalists, Suharto deliberately opted to pragmatically align with the minor-
ity ethnic-Chinese population for their private-sector entrepreneurial skills. This
community was economically powerful but politically weak and could not fea-
sibly challenge Suharto’s leadership. Providing them with incentives and other
public goods to develop the import-competing sector resulted in economic
496 Case Studies II
diversification away from the country’s natural resource–export dependence. In

contrast, in Zaire, Mobutu did not opt for economic diversification as he feared it
would strengthen political opposition.
It has to be borne in mind that natural resource rents are subject to sharp
“boom” and “bust” behavior over the course of a business cycle. This implies that
economic growth rates, as well as government revenue and patronage rents for
the elite, are volatile. With diversification, Indonesia managed to lower this risk.
It also avoided many of the features of Dutch disease ordinarily associated with
resource-dependent economies (Warr 1986). For example, it devalued its cur-
rency as early as November 1978 to maintain the competitiveness of the domes-
tic, import-competing sector at a time when most oil-rich economies allowed
their real exchange rate to appreciate, with the consequence that the interna-
tional competiveness of their non-oil-tradable sector eroded. Warr (1986) points
to other macroeconomic policy interventions Indonesia took in order to maintain
international competitiveness. These policies meant that when crude-oil revenues
diminished, the economy had achieved a degree of export diversification and
would not face a growth collapse. Moreover, the oil boom of the 1970s and later
economic diversification helped to maintain buoyant government expenditure,
particularly on public goods like health and education.
The success of economic diversification is evident in Table 21.4. As the econ-
omy industrialized, during the heyday of the New Order period (1975–1995),
the role of the primary sector, including oil, sharply declined in the sectoral
Table 21.4 Structural Change, 1975–1995 (Percentages)
Sectoral Composition of Value Sectoral Composition of Exports

Added
Sector 1975 1980 1985 1990 1995 1975 1980 1985 1990 1995
Primary 27.7 20.6 22.2 16.7 11.6 6.0 6.7 6.1 2.3 1.1
Oil, gas, mining 20.5 26.3 14.2 14.6 9.8 73.9 70.8 40.6 27.9 17.3
Petroleum 0.6 0.3 5.0 3.2 2.0 1.0 6.8 23.7 14.4 7.5
refinery
Manufacturing 10.9 11.1 13.0 19.1 24.6 9.4 7.4 17.9 38.4 51.1
Electricity, gas, 0.3 0.3 0.4 0.6 0.6 0.0 0.0 0.0 0.0 0.0
water
Construction 5.0 5.0 6.6 5.8 6.7 0.0 0.0 0.0 0.0 0.0
Finance and 2.4 2.0 2.6 3.8 4.1 0.0 0.2 2.3 3.0 3.3
insurance
Other services 32.6 34.4 36.0 36.2 40.6 9.7 8.1 9.3 14.0 19.7
Total 100 100 100 100 100 100 100 100 100 100
Source: Jacob 2005, 429–30, based on input-output tables.

composition of economic value-added. A similar story can also be observed in the

changing pattern of exports during this time period. Furthermore, the contribu-
tion of revenue from oil and gas in the state budget, which had increased from
9 percent in 1967 to a peak of 62 percent in 1981–1982, declined to around just
20 percent in 1996–1997 (van Zanden and Marks 2012).
Diversification was good for the economy, and Suharto hoped to gain political
leverage from his economic achievements. The growth in the Indonesian econ-
omy greatly reduced poverty, especially as it was combined with a conscious effort
10
8
Regional Gross Domestic Product
Annual Growth of Per Capita
6
(RGDP), 1976–96
0
0 100 200 300 400 500 600 700 800 900 1000
–2 Initial Per Capita Regional Gross Domestic Product
(RGDP), 1976 (in 1973 prices)
–1
100
80
60
Reduction in Poverty Rate,
40
1976–96 (in percent)
20
0
0 20 40 60 80 100
–20
Initial Poverty Rate, 1976 (in percent)
–40
–60
–80
–100
Figure 21.1 Interprovincial regional convergence, 1976–1996. Top panel: Average annual
per capita growth of real GDP 1976–1996 against initial per capita real GDP in 1976 (measured in
000’s of Indonesian rupiah, 1973 prices). Bottom panel: Average reduction in poverty rates 1976–1996
(percent) against initial poverty levels (as measured by head count ratio [HCR] 1976 [percent]).
Source: Tadjoeddin 2014, 50–51; figs. 3.2 and 3.3.
498 Case Studies II
to lower regional disparities (see Figure 21.1). Tadjoeddin (2014) demonstrates

a clear pattern of income convergence across provinces during 1976–1996. The
initial income levels of 1976 correlate negatively with the annual growth of per
capita regional gross domestic product (RGDP) for 1976–1996. This means that
poorer provinces grew faster than richer ones over the course of two decades.4
Hill (2000, 235) notes that “there is no case of a high-income province growing
much faster than the national average, or conversely of a poor province falling
sharply behind.” These convergences were achieved through deliberate fiscal
equalization policies.
The main instrument of these policies was Inpres (Instruksi Presiden, or
Presidential Instruction). Introduced in the early 1970s, Inpres primarily used
resource windfalls for the socioeconomic development of poorer regions, espe-
cially on Java, which has the highest population concentration (about 70 percent).
Investing in human capital through education and health spending was the key
feature of Inpres, and it also played a similarly crucial role in the early economic
success of East Asian economies (World Bank 1993). Development expenditure
on education and health as a proportion of the overall development expenditure
in the state budget was on the rise (UNDP 2001). The outcome is clear: across dis-
tricts, regional inequalities in human development–related indicators are much
lower than that of per capita RGDP (Figure 21.2).
To achieve interregional convergence may have been aimed at mollifying
rebel tendencies in some regions but eventually resulted in the “rage of the rich.”
This is due to dissatisfaction on the part of oil-and gas-producing regions for not
receiving a “fairer,” that is, higher, share of their own resource rents because of the
New Order’s equalization policies. These regions articulated heated secessionist
Real PC RGDP 0.277
Real PC RGDP* 0.110
Purchasing power 0.073
Years of shooling 0.032
Literacy 0.008
Life expectancy 0.002
Human Development Index 0.003
0.00 0.10 0.20 0.30
Figure 21.2 Regional inequalities of selected indicators, 1996 (Theil-L index

measure #). Notes: # Theil-L index in this case is a measure of interdistrict inequality of the relevant
indicators. Higher index means higher interdistrict inequality.
* Without oil and gas and thirteen richest districts.
Source: Tadjoeddin 2014, 52, fig. 3.4.
sentiments in the wake of Suharto’s own fall from power in 1998, a problem that
since then has been largely resolved by the country’s moves toward democratiza-
tion and decentralization (Tadjoeddin 2014).
Suharto’s economic achievements, in the end, turned out to be politically
costly. The drive for socioeconomic development and economic diversification
created societal bases of power outside the control of political elites (Dunning
2005). These independent bases of power facilitated challenges to the political
power of state incumbents, especially during economic downturns. The more
egalitarian and growing middle class that resulted from three decades of con-
tinuous, broad-based socioeconomic progress became Suharto’s nemesis in 1998.
The challenge did not come from Islamists or nationalists, the two groups that he
feared earlier. Instead, Suharto became the victim of his own (economic) success.
It is interesting to speculate what Indonesia might look like today if the pat-
tern of its stagnant economy of 1950–1966, where it recorded average annual per
capita GDP growth of only 1.1 percent, had continued after 1967. Figure 21.3
depicts this counterfactual: Indonesia today (that is, in 2010) might be poorer
than Vietnam or Nigeria.
21.5. Conclusions
We sketched Indonesia’s buildup to the 1965–1966 politicide of communists
and communist sympathizers. One of our contributions is to explain why ordi-
nary individuals, not belonging to the elite, might wish to participate in the act
of murder. We maintain that participation in mass murder cannot be explained
5,000
4,500
4,000
3,500
Indonesia
3,000 Indonesia
2,500 (counterfactual)
Vietnam
2,000
Nigeria
1,500
1,000
500
0
67
70
73
76
79
82
85
88
91
94
97
00
03
06
09
19
19
19
19
19
19
19
19
19
19
19
20
20
20
20
Figure 21.3 Real per capita GDP, 1967–2010 (in 1990 international
dollars). Source: Calculated from Maddison Project Database (http://www.ggdc.net/maddison).
500 Case Studies II
on pecuniary grounds only, but has to be situated in innate behavioral factors

that also govern individual actions, which we encapsulate in a theoretical model
based on the economics of identity. But even innate behavior does not occur in a
socioeconomic vacuum. One must scrutinize the political economy at large, lead-
ing up to mass murder. Indonesia experienced economic stagnation and decline
in an economy that was poor to begin with, which served to intensify existing
hatreds and polarization. The mass murder of communists certainly aided the
consolidation of the new regime under Suharto, even as his ascension to power
also has to be seen in the light of cold war politics of the time, in which the United
States in particular favored autocratic, right-of-center military rulers, particularly
in the context of East Asia where the battle against the alternative ideology—
communism—was being waged most fiercely.
For Indonesia, polarization (ideological or ethnic), combined with economic
stagnation, domestic political opportunism, and geopolitics that favored the
contender to power, increased the risk of mass atrocity. Broad-based economic
growth combined with policies of redistribution that diminish political and eco-
nomic polarization may reduce risks of mass atrocity.
As described, the broad-based economic growth strategy adopted by Suharto,
as well as his policies to lower the country’s reliance on natural resource–based
exports, led to economic growth in Indonesia, along with poverty reduction and a
diminution of regional disparities. Importantly, these policies cannot be inferred
to be the consequence of politicide; rather the politicide was a step in the consoli-
dation of power by the new regime, which sought to obtain legitimacy by offering
better economic conditions within an authoritarian social contract.
The contrasting paths taken by Suharto in Indonesia and Mobutu in Zaire
are worth highlighting and remembering. Economic growth and broad-based
socioeconomic progress resulted from Suharto’s three decades of rule, but his
rule ended in a democratic transition when a rising middle class pushed for an
endogenous process of democratization, akin to Lipset’s (1960) moderniza-
tion thesis, albeit with some level of atrocious violence occurring in Aceh and
East Timor. By contrast, the no-d iversification and no-development choices
followed by Mobutu in Zaire led the country directly into full-blown civil war
upon his departure. As hypothesized by Collier and Hoeffler (2004), the avail-
ability of large numbers of impoverished young men and presence of lootable
natural resources, along with an ailing economy, made civil war highly fea-
sible. Either way, as pointed out in the literature, periods of regime transition
are risk factors for mass atrocities.
To conclude, over three decades of authoritarian rule in Indonesia brought
about broad-based economic progress. But the authoritarian contract that sus-
tained the regime became untenable; the contract lacked credible commitment
without a transfer of political power to the emergent new middle class. This mir-
rors Lipset’s (1960) modernization theory of endogenous democracy, which
states that, with higher levels of income, the pressure for democracy becomes
inexorable. Yet, as Przeworski and Limongi (1993) argue, in some countries the
democratic transition occurs at a relatively low level of per capita income, as seems
to be the case in Indonesia, as compared to other countries in the region such as
South Korea and Taiwan.
Appendix
In the model below, there are two groups: “Muslims” who, facilitated by a
military government, might perpetrate politicide on their ideological opponents,
“communists.” As models go, obviously these are ideal-t ypes and henceforth we
drop the quotation marks. We model individual Muslim motivation to take action
against communists, based on prior beliefs and current signals, and move from there
to Muslim collective action. Whereas the notation in the main text was adapted to
correspond to Akerlof and Kranton (2000), here we use our own notation.
Following Akerlof and Kranton (2000), individuals obtain utility from their
identity and from the behavior demanded by that sense of belonging. Thus, an
individual member, r, of a Muslim group derives utility (Ur) from identity-related
actions in the following manner:5
Ur = Us(sr , s j , Ir) + Uo(k r) ⋯ j ≠ r (1)
where Ir = Ir(sr , s j , μ ).
The parameter s refers to a member’s actions related to his primary identity,
Muslim. Utility, however, comes from two components. Utility Us stems from
actions sr and utility Uo stems from other, non-Muslim identity-related actions
kr, that is, acting as a “good Indonesian citizen” in general. The former is like a
club good, and the latter is similar to a private good. The two components enter
individual utility in an additive and separable fashion. Thus, unlike Akerlof and
Kranton (2000), in our model an individual is permitted to possess multiple
identities (Sen 2008), which is an innovation of this model. In addition, the indi-
vidual not only derives utility from a vector of his own actions, sr, but also from
similar actions of other, like-m inded individuals belonging to the Muslim group,
denoted as sj, and, above all, from his own identity (Ir) that, in turn, depends
on the actions (sr, sj) just described as well as on another parameter, namely,
that of hatred of communists, denoted as μ. Similar to Boulding (1956), image,
including self-image, is subject to change via messages, akin to signals, which
can weaken or strengthen the image and, on occasion, can be quite dramatic or
revolutionary.
502 Case Studies II
Sr (with a capital S) denotes the total endowment of possible individual actions

divided up among Muslim-related and non-Muslim-related actions:
Sr ≤ sr (μ) + k r . (2)
It is postulated that the attractiveness of inputs into Muslim-identity behavior,

sr, rises with μ: the more hatred of communists, the more Muslim identity–based
behavior emerges.
Again following Akerlof and Kranton (2000), it is also possible to model how
individuals derive disutility from nonconformity by other group members and
how to deal with nonconformity. Thus, individuals of a group will exert effort to
bring members who have strayed from ideal group behavior back to the fold if the
costs of doing so are low as compared to the pain inflicted on the errant member.
Specifically, in a cooperative game, if Muslim group member j suffers disutility
(Ij) from non-Muslim-based behavior (kr) by member r, the first individual (j)
may lure back the errant individual (r), provided that the cost of doing so to j (cj)
is not too large and is less than the loss inflicted (lr) on the deviant group member
r. This is because if the loss imposed on the deviant member, r, is too great then
that member may not conform to group behavior at all, choosing permanent exit.
This setup requires the condition:
c j < I j < lr . (3)
This condition is more likely to hold among members of poor rather than rich
communities, or among those suffering from widespread poverty, low human
capital (education) endowments, or living in close proximity. Moreover, the
Muslim group may use the behavior implied in (3) to resolve mutual mistrust and
to overcome the collective action problem described by Olson (1965). Thus, col-
lective or group grievances become individual grievances. At the extreme, this
can induce genocidal acts, similar to choices made about suicide bombing as out-
lined in Wintrobe (2006), such that non-Muslim-based actions by Muslims tend
to zero, kr = 0.
We now turn to the determination of hatred of communists, μ. The parameter
μ denotes a spectrum of like (μ less than zero) to hatred (μ very much greater than
zero), which in the extreme can induce homicidal behavior. A higher μ implies
more confrontational attitudes, including at the extreme genocide. Following
Glaeser (2005) we can think of μ as originating in a signal sent out by a leader. To
consolidate the sender’s hold on power, the message may deliberately be sent as a
false hate message to encourage recipients (Muslims) to eliminate opponents. Its
attractiveness to Muslims will depend on the communities’ need for scapegoats
and on their own personal life experiences of and with communists. Not all these
signals will be believed. For example, a hate-mongering leader may be mistrusted,
or better-educated Muslims may discount part of the message, or people with

personal knowledge of communists may similarly disregard the signal. There is
a cost, z, to individuals of verifying the veracity of the signal, though, through a
search process. Let φ be the probability that a leader is sending out a false message
and that the communist group is largely innocent. Then, 1–φ is the probability
that they are not innocent, and this will therefore impose a net cost, in the form
of the hatred parameter μ, on members of the Muslim community. Individuals
update their Bayesian prior for this in the following manner:
φ
. (4)
φ + (1 − φ)μ
The prior may be updated subject to the aforementioned search cost, z, and exoge-
nous events like riots or acts of terror (close to home) perpetrated by communists.
The Muslim public is composed of two types: a high-cost type (indexed by h),
which suffers both greater potential damage (μ) and also has higher search costs
(z), and a low-cost type (subscript l), which suffers less disutility from a potential
communist threat and has lower search costs of finding the truth (because of edu-
cation, say). The former type may include the less educated, the more socioeco-
nomically disadvantaged, and those who would like to find scapegoats for their
poverty and vulnerability. In general:
Vi = yi − μ i(z)(1 − φ) − z(i) ⋯ i = h, l (5)
⇒ μ hz(1 − φ) > z h ⋯ for high-cost types and
⇒ μlz(1 − φ) < z l ⋯ for low-cost types,
where V represents expected (pecuniary) utility and y is the income of individu-

als of type i = h,l, diminished by μ and z costs. Maximization of this expected
utility with respect to search, z, leads to the conditions described in the second
and third lines of (5), respectively. The high-cost-t ype individual suffers both a
greater perceived loss from communists (μ h) and has a higher cost of verification
of the signal (zh), all the more so if the search cost of verifying the signal entails
an earlier, lumpy, fixed cost in education, say. These individuals are more likely
to abandon the search for truth in favor of the hate message, effectively setting
φ = 0. Even the low-cost-t ype individual (who will engage in the search for truth)
may at times randomize the probability of φ around 0 or 1, if, say, the equilibrium
φ ≈ 1/2 in (5). Also, after major incidents, like a communist uprising or a famine,
all individuals from the Muslim community may set φ to zero for some time,
effectively tarring all communists and quasi-communists with the same brush.
If enough Muslims believe the signal then collective action against communists
becomes more likely.
504 Case Studies II
Muslim group behavior is arrived at after summing the choices regard-

ing sr from individual utility maximization described in (1), subject to
constraints (2): n
∑ sr = e. (6)
r=1
For collective action (like a club good) to take place via the adoption of the
group strategy (e), a critical threshold of aggregate Muslim identity–based actions,
∑sr, must be chosen. Not all individuals will engage in Muslim identity–based
actions, and not all actions (sr) are violent. In order to keep the analysis tracta-
ble, we do not specify an exact tipping point where actions turn violent, let alone
genocidal, choosing to focus on a continuum instead. The forging of collective
action requires high enough values of e. Condition (3) must also hold so that it is
not too costly to deter non-Muslim-identity-based actions through cooperative
games. Also, at high enough values of μ, condition (3) becomes more relaxed, as
more self-enforcing and sincere Muslim identity–based behavior takes place via
(2). Note that e also denotes a continuum of peaceful actions, here viewed as the
converse of violence.
The Muslim group objective or utility function, R, takes the following form:
R = π(a , e)RP + (1 − π )(⋅)RC − E(e), (7)
where
RP = Y R + T + pF R(μ) (8)
RC = Y R + T + cF R (μ); FμR > 0
T(A)
e=
F R(μ)
c > p > 0, c + p = 1, π a , e > 0.
Superscripts p and c refer to states which are either more peaceful or more con-
frontational, with probabilities π and (1 –π), respectively. The probability of peace
increases with effort e by the dissident group and/or action a on the part of the
communists, π e, π a > 0, as outlined below, but with diminishing returns such that
π ee, π aa < 0. Both group strategies are a hybrid of accommodation and aggression.
RP and RC describe Muslim group payoffs in the two states, with utility greater in
peaceful states. Utility is derived from income (YR) and transfers (T) obtained
from the state. Strategic choices surround e (effort with regard to peace) obtained
from (6). Fighting, FR, is greater when the parameter µ rises, implying greater
hatred for communists. This can happen if there are exogenous events increasing
poverty, food shortages, communist bids to seize power, or encouragement from

the state toward hatred. E describes the aggregate cost function for undertaking
e, composed of the psychic costs of “capitulation” and the total costs of inducing
Muslim identity-based behavior in (3), with both Ee and Eee > 0. Note that as e rises
there is more peace with the communists; a decline in e defines greater militancy.
Collective group behavior, via group strategy e, is akin to a club or associa-
tional good (Cornes and Sandler 1996). A club good is excludable in nature; only
those who subscribe or contribute can partake in it. Yet it is voluntary because,
unlike the case of a nonexcludable pure public good, individuals do not have to
participate in the club. With club goods, membership and provision are insepa-
rable. Differentiating the dissident group’s strategic variable (e) in (8) yields
1 T(•)
de = dT − dμ . (9)
F R (μ) FμR2
The first term on the right-hand side of (9) is positive, and thus e rises with trans-
fers, T, but falls with hatred, μ.
The Muslim group will maximize (7) with respect to e, equating its marginal
benefit to marginal cost:
π e[RP(•) − RC(•)] = Ee . (10)
Turning to the communist group (and ignoring individual communist member

behavior, which can be argued to be similar to members of the Muslim group), the
utility of the group is given by:
G = π(a , e)GP + (1 − π )(•)GC − C(a) (11)
where
GP = Y G + pF G(μ) + T (12)
GC = Y G + cF G(μ) + T
T
a= .
F G(μ)
GP and GC refer to exogenous payoffs to the communists in the two states, with GP
>GC due to conflict-induced losses of endowments and transfers.
C refers to the cost of undertaking peaceful actions a by the communists,
Ca > 0. These costs consist of pecuniary and nonpecuniary elements; the first
because of the loss of “power,” the latter because of a political cost by alienating
506 Case Studies II
those in the group opposed to accommodation. Communists maximize (11) with

respect to a:
πa[GP (•) − GC (•)] = Ca . (13)
Equations (10) and (13) form the basis of the reaction functions for either side,
obtained by totally differentiating them with respect to a and e. Thus:
da Eee + πeeRC(•) − RP(•) ≥ ≥

= 0 if πae 0 (14)
de / RR πae[RP(•) − RC(•)] ≤ ≤
and
da πae[GP (•) − GC (•)] ≥ ≥
= 0 if πae 0 . (15)
de / RG Caa + πaa[GC (•) − GP (•)] ≤ ≤
Note that π ae = π ea by symmetry.

The reaction functions are positively sloped if π ae > 0, implying that they are
strategic complements. In other words, they represent a tit-for-tat strategy on the
part of both antagonists: If one side behaves more peacefully, then its opponent
does the same, and vice versa. This is the standard assumption in the literature on
conflict. It means that increases either in fighting or in peaceful efforts made by
one side are matched in the same direction by the other side. In our model, how-
ever, we allow for the possibility that π ae < 0, so that the choice variables are strate-
gic substitutes, and the reaction functions then slope downward (see Figure 21.4).
This can only occur because the strategy space is defined in terms of peace. Thus,
if one side behaves more peacefully, because it feels weaker or is on the defensive,
it increases the utility of both parties, and the other side may free-r ide on this
expected response by reducing their own peaceful action. Yet note that free-r iding
behavior does not necessarily lead to a rise in the equilibrium level of conflict, as
the side raising its peaceful efforts may compensate more than proportionately for
the group lowering its peaceful actions.
An increase in poverty-i nduced or government-s ponsored hatred of com-
munists will shift the reaction function of the Muslim group (R 0R to R 1R )
along the reaction function of the communist group (R 0G ), and the equilib-
rium point will move from A to B in Figure 21.4. Here, communists behave
more peacefully, or defensively; the Muslim group does the converse, and a
politicide of communists could ensue. Should the cost of peaceful behavior
rise exogenously for communists as well, their reaction function could move
downward, with a new equilibrium point at C. Politicide occurs when e = 0,
but the strategic value of a is not necessarily equal to zero, implying a corner
solution.
Communist Peaceful
Actions, a
R
R0
R1R
B
G
R0
R1G Muslim Group
0 Behavior, e
Figure 21.4 Strategic interaction. The figure shows the levels of peace (a and e) of the two
groups with an initial strategic equilibrium at A at the intersection of reaction functions R 0G and R 0G.
An increase in poverty experienced by the Muslim group and/or new government–sponsored hatred
of communists causes the Muslim reaction function to shift to R 1R , leading to a decrease in Muslim
peaceful actions and an increase in communist peaceful (or defensive) actions at strategic equilibrium
at B. Should the cost of peaceful behavior rise exogenously for the communists as well, their reaction
function could move down, with a new strategic equilibrium at C.
Notes
1. Sukarno was Indonesia’s president from 1945 to 1966. However, he was a dictator only
during the period of Guided Democracy (1957–1966). Indonesia experienced a war of
independence during 1945–1949 when the effectiveness of the central government was
largely in question. The country adopted a parliamentary democracy system during 1950–
1957, with executive power held by prime ministers and the presidency largely assigned a
symbolic role. Obviously, this changed with the transition to dictatorial rule in 1957.
2. The UN Genocide Convention refers to the destruction, or attempted destruction, of
distinct national, ethnic, racial, or religious groups, excluding political groups from the
definition. For the latter, the literature uses the term “politicide”: the mass murder of oppo-
nents specifically drawn from a political grouping. The killing of communists, and sus-
pected communists, in Indonesia during 1965 and 1966 fits the definition of politicide. For
a general discussion of terms and data, see chapters 1, 2, and 3 in this volume.
3. Dutch disease refers to an overvalued currency and other mechanisms that can signifi-
cantly hamper the development of economic sectors other than natural resource extrac-
tion, often leading to growth failure (see Warr 1986).
4. This is based on overall per capita GDP including oil and gas. If oil and gas are excluded,
interregional provincial inequality of per capita GDP is much lower, but still shows a
slightly increasing trend during 1980–2 000 (Milanovic 2005).
5. The utility function that follows is in terms of behavioral actions, not consumption.
508 Case Studies II
References
[ADB] Asian Development Bank. 1997. Emerging Asia: Changes and Challenges. Manila: Asian
Development Bank.
Akerlof, G., and R. Kranton. 2000. “Economics and Identity.” Quarterly Journal of Economics 115,
no. 3: 715–53.
Boeke, J. H. 1953. Economics and Economic Policies of Dual Societies. New York: Institute of Pacific
Relations.
Boulding, K. E. 1956. The Image: Knowledge in Life and Society. Ann Arbor: University of
Michigan Press.
Collier, P., and A. Hoeffler. 2004. “Greed and Grievance in Civil Wars.” Oxford Economic Papers
56, no. 4: 563–95.
Cornes, R., and T. Sandler. 1996. The Theory of Externalities, Public Goods, and Club Goods. 2nd ed.
Cribb, R. 1990. The Indonesian Killings of 1965–1966: Studies from Java and Bali. Monash Papers
on Southeast Asia No. 21. Clayton, Victoria: Monash University Centre of Southeast Asian
Studies.
Cribb, R. 2001. “How Many Deaths? Problems in the Statistics of Massacre in Indonesia
(1965–1966) and East Timor (1975–1980).” In I. Wessel and G. Wimhöfer, eds., Violence in
Indonesia. Hamburg: Abera, 82–98.
Davidson, J. S., and D. Kammen. 2002. “Indonesia’s Unknown War and the Lineages of Violence
in West Kalimantan.” Indonesia 73(April): 53–87.
Dunning, T. 2005. “Resource Dependence, Economic Performance, and Political Stability.”
Easterly, W., R. Gatti, and S. Kurlat. 2006. “Development, Democracy and Mass Killings.”
Fealy, G., and K. McGregor. 2012. “East Java and the Role of Nahdlatul Ulama in the 1965–6 6
Anti-Communist Violence.” In D. Kammen and K. McGregor, eds., The Contours of Mass
Violence in Indonesia, 1965–1968. Singapore: NUS Press, 104–30.
Friend, T. 2003. Indonesian Destinies. Cambridge, MA: Belknap Press.
Gibson, J. 1967. “Foreign Enterprise and Production Sharing.” In T. K. Tan, ed., Sukarno’s Guided
Indonesia. Brisbane: Jakaranda Press, 89–101.
1: 45–86.
Hill, H. 2000. The Indonesian Economy. Cambridge, UK: Cambridge University Press.
Iannaccone, L. 2006. “The Market for Martyrs.” Interdisciplinary Journal of Research on Religion 2
(Article 4).
Jacob, J. 2005. “Late Industrialization and Structural Change: Indonesia, 1975–2 000.” Oxford
Development Studies 33, nos. 3–4: 427–51.
Jenkins, D., and D. Kammen. 2012. “The Army Para-Commando Regiment and the Reign of
Terror in Central Java and Bali.” In D. Kammen and K. McGregor, eds., The Contours of Mass
Violence in Indonesia, 1965–1968. Singapore: NUS Press, 75–103.
Kammen, D., and K. McGregor, eds. 2012. The Contours of Mass Violence in Indonesia, 1965–1968.
Singapore: NUS Press.
Liddle, W. 1999. “Indonesia’s Unexpected Failure of Leadership’” In A. Schwarz and J. Paris,
eds., Politics of Post-Suharto Indonesia. New York: Council on Foreign Relations, 16–39.
Lipset, S. 1960. Political Man: The Social Bases of Politics. New York: Doubleday.
Milanovic, B. 2005. “Half a World: Regional Inequality in Five Great Federations.” Policy
Research Working Paper 3699. Washington, DC: World Bank.
Murshed, S. M. 2011. “The Clash of Civilizations and the Interaction between Fear and Hatred.”
International Area Studies Review 14, no. 1: 31–4 8.
Myrdal, G. 1968. Asian Drama: An Inquiry into the Poverty of Nations. Harmondsworth,
UK: Penguin.
Olson, M. 1965. The Logic of Collective Action. Cambridge MA: Harvard University Press.
Przeworski, A., and F. Limongi. 1993. “Political Regimes and Economic Growth.” Journal of
Economic Perspectives 7, no. 3: 51–69.
Robinson, G. 1995. The Dark Side of Paradise: Political Violence in Bali. Ithaca, NY: Cornell
University Press.
Roosa, J. 2006. Pretext for Mass Murder: The September 30th Movement and Suharto’s Coup D’Etat in
Indonesia. Madison: University of Wisconsin Press.
Schmitz, D. F. 2006. The United States and Right-Wing Dictatorships, 1965–1989. Cambridge.
Sen, A. K. 2008. “Violence, Identity and Poverty.” Journal of Peace Research 45, no. 1: 5–15.
Simpson, B. 2008. Economists with Guns: Authoritarian Development and U.S.-Indonesian Relations,
1960–1968. Palo Alto, CA: Stanford University Press.
Tadjoeddin, M. Z. 2014. Explaining Collective Violence in Contemporary Indonesia: From Conflict to
Cooperation. Basingstoke, UK: Palgrave Macmillan.
Tan, T. K. 1967. Sukarno’s Guided Indonesia. Brisbane: Jakaranda Press.
UNDP (2001). Indonesia National Human Development Report 2001. Toward a New
Consensus: Democracy and Human Development in Indonesia. Jakarta: UNDP.
van Zanden, J. L., and D. Marks. 2012. An Economic History of Indonesia 1800– 2010.
London: Routledge.
Voigtländer, N., and H. J. Voth. 2013. “Gifts of Mars: Warfare and Europe’s Early Rise to Riches.”
Journal of Economic Perspectives 27, no. 4: 165–86.
Warr, P. 1986. “Indonesia’s Other Dutch Disease: Economic Effects of the Petroleum Boom.”
In J. P. Neary and S. van Wijnbergen, eds., Natural Resources and the Macroeconomy.
Oxford: Blackwell, 288–323.
Wintrobe, R. 2006. “Extremism, Suicide Terror and Authoritarianism.” Public Choice 128, nos.
1–2: 169–95.
World Bank. 1993. The East Asian Miracle: Economic Growth and Public Policy. Oxford, UK: Oxford
University Press.
22
Economic Foundations of Religious

Killings and Genocide with Special
Reference to Pakistan, 1978–2012
Pa rt h a G a ngopa dh yay
22.1. Introduction
Since 1956, some forty-t hree genocides have taken place resulting in the deaths of
at least fifty million civilians (Political Instability Task Force 2010). As of 2008,
such episodes of violence have also caused the displacement of another fifty mil-
lion people (UNHCR 2009). Only relatively recently have economists begun to
think about mass killings, and this chapter contributes to the nascent field. In
particular, it focuses on the rational foundations of identity-based mass killings in
heterogeneous societies (also see c hapter 21 in this volume).1 Not only is mass kill-
ing obviously expensive in lives lost, social ties and trust disrupted, and property
destroyed, it is also expensive to rectify the damage done and reconstitute society.
Evidently, individual decisions to engage in mass killing are distressingly com-
mon; they are also inherently complex to analyze, in part because this analysis
requires information on social interactions and feedback mechanisms that lead
individuals to participate in the killings.
In this chapter, I create a new set of economic models to understand what econ-
omists would call “equilibrium levels” for mass killings that one social group may
have an incentive to unleash on another. (In economics, equilibrium just means that
a social system has come to a position of rest. When the underlying forces that cre-
ated the system in the first place no longer change, then the system is settled and at
rest, even if the equilibrium is terrifying in its consequences.) I try to understand
how ordinary members of a social group, hereafter called “citizens,” can become so
involved in the mayhem as to carry out atrocities against other citizens.
The microfoundation of my theoretical analysis is rooted in Gangopadhyay and
Chatterji (2009), in which social interactions are a part of agents’ objective func-
tions that determine their level of utility or satisfaction. An important concept
510
Economic Foundations of R eligious Killings and Genocide 511
in the analysis is that of social capital, as well as the idea that social interactions
can result in and entail anti-social capital. Social capital embodies the “influence
which the characteristics and behaviors of one’s reference groups have on one’s
assessments of alternative courses of behavior” (Durlauf 1999, 2). Bowles (1999)
spells out attributes such as trust, commitment, adherence to social norms, and
retribution to norm violators that together constitute social capital, in conso-
nance with Putnam’s initial idea (Putnam 1993). These attributes dictate relation-
ships among people in which a course of actions is chosen. In Gangopadhyay and
Chatterji (2009), anti-social capital exists when an agent displays a lack of trust and
commitment toward others. For example, an agent may display sheer hostility
and even cause economic harm to a minority as long as his reference groups allow,
tolerate, and possibly reward such displays (see also Gangopadhyay 2009). These
attributes are called “anti-social capital” since they typically open up chinks in a
social order and create insider-outsider conflicts. In contrast to Gangopadhyay
and Chatterji (2009), in this chapter religious killings are undertaken even if the
reference group opposes such actions.
The plan of the chapter is as follows. For context, section 22.2 briefly surveys
the relevant literature. Section 22.3 develops a dynamic model of social interac-
tions to explain mass killings, and section 22.4 examines the evolutionary sta-
bility of a mass killing equilibrium. By way of testing the theory, section 22.5
examines the empirical foundations of seemingly religiously motivated mass kill-
ings (hereafter referred to as “religious killings”) in Pakistan during the period
1978–2012. Section 22.6 concludes.
22.2. Literature Survey

22.2.1. Genocide as a Form of Mass Killing:
Features and Causes
The term genocide, an amalgam of the Greek genos (race) and the Latin cide (kill-
ing), was coined in 1944 by Raphael Lemkin (1944). Due to Lemkin’s persis-
tence, the United Nations (UN) General Assembly adopted the UN Genocide
Convention in 1948. Henceforth, genocide was recognized as a crime under inter-
national law. Article 2 of the Convention defines the crime as any of a number of
specific acts undertaken “with the intent to destroy, in whole or part, a national,
ethnical, racial, or religious group” (United Nations 1951). Among scholars, it is
not uncommon to include mass killings on grounds of political ideologies, that
is, politicide, either in the definition of genocide itself or under the umbrella term
of mass atrocities, which also includes genocides. (On terms, definitions, and data,
see chapters 1 through 3 in this volume.)
The history of genocides can be summarized in terms of four salient fea-
tures. First, genocide shows remarkable invariance throughout the political his-
tory of humankind. In the twentieth century alone, episodes of genocide have
512 Case Studies II
occurred in then-German southwestern Africa (Namibia), the Ottoman Empire,

the Soviet Union, Germany, Maoist China, Indonesia, Bangladesh, Burundi,
Afghanistan, Angola, Cambodia, East Timor, Bosnia, Guatemala, Uganda, Iraq,
Rwanda, Sudan-Darfur, and Palestine. Second, the scale of genocide is shock-
ing: Valentino (2004) estimates that between 60 million to 150 million people
perished in twentieth-century genocides alone. Third, and contrary to public per-
ception, genocide is rarely an irrational act driven by spontaneously expressed
hatred or momentary loss of civic sense. Instead, scholars find a disturbing com-
monality: as a rule, genocide is carefully planned to result in a well-orchestrated,
systematic execution of a targeted group. Fourth, although genocide is triggered
by political leaders, the killings are carried out by ordinary people. In addition, a
silent majority of bystanders implicitly or explicitly condones the mass killings.
Perpetrators and bystanders are the mothers, fathers, and children of our soci-
eties. By far the vast majority of them do not suffer from any form of psychopathol-
ogy (Waller 2007). Instead, standard motives behind genocide appear to fall into
one of four categories (Chalk and Jonassohn 1990): (1) to eliminate a perceived
threat; (2) to terrorize a perceived enemy; (3) to acquire wealth; or (4) to impose
a set of beliefs. Valentino (2004) adds two more: (1) dispossessive mass killing, to
remove a group from political power; and (2) coercive mass killing, to eliminate a
group that supports rebellion.
Economists will want to explain genocide as a form of mass killing in terms of
costs and benefits, as I explain in the following subsection.
22.2.2. Economic Theory of Mass Killings

To scapegoat people on the basis of ethnic, linguistic, political, or other identi-
ties creates enemies and leads one social group to turn against another. It is well
established that conflict between groups, combined with self-serving interests,
can add fuel to inflame group violence. Discrimination, intolerance, and initial,
limited amounts of violence can change the attitudes of individuals and groups to
label others as enemies and can thereby affect the evolution of social dynamics and
end in the mass killing of members of a target group. Certain economic, cultural,
political, and social characteristics make such developments likely. For example,
it is widely observed that the passivity of bystanders allows group violence to
unfold. To help prevent such violence one will need to understand the forces that
first initiate the violence against a group and, then, how a community is carried
forward to sustain its attempt to annihilate a target group.
Attempts at constructing an economic theory of mass killing are often based on
a model of war (e.g., Fearon 1995; Skaperdas 2006) such that conflict is assumed
to be a costly lottery or a costly bargain between two warring groups. Each group
has economic and military power to engage in fighting, which can escalate into
the complete annihilation of one of the groups. But a priori it may not be clear
which group will win (that is the lottery aspect). Alternatively, both groups can
settle on peaceful coexistence (the bargain aspect). If peaceful conflict resolu-
tion cannot be reached, then mass killings may occur via one of four avenues
(Anderton 2010). First, an incumbent group perceives a threat to its power that
emanates from the (eventual) victim group, and the latter is unable to credibly
commit to not challenging the power of the former; second, the conflict is about
an indivisible object like historically significant land or the racial “purity” of ter-
ritory; third, the incumbent has a strong political bias for mass killing because
“othering” the out-g roup conveys political and economic gains to the incumbent;
and, fourth, the victim group is a source of strong political rivalry such that the
incumbent becomes weary of such a persistent source of contestation.
Extending the costly lottery model, Esteban, Morelli, and Rohner (2015)
develop a model of rebellion with two groups that compete over nonexcludable
public goods such as natural resources. The availability of natural resource rents
provides incentives for each group to annihilate the other. As a result, mass kill-
ings are more likely in societies with (1) a large stock of natural resources in the
economy, (2) significant constraints on rent-sharing, (3) low productivity of capi-
tal and labor, and/or (4) low state capacity. This implies that mass killing is more
common in the context of intrastate wars than in the context of interstate wars.
22.2.3. Social Evolution of Intolerance in

Heterogeneous Societies
Only recently have economists turned their attention to study links between
intolerance and violent conflict, exploring in particular the opportunity cost
of hawkish behavior in a heterogeneous society (Gangopadhyay 2009; Gupta
2012).2,3 What creates intolerance toward target groups in a heterogeneous soci-
ety? In Chapter XIII of Leviathan, Hobbes suggests that humans are quarrel-
some by nature. He then suggests three main causes of intolerance and targeting.
First is the underlying theme of competition for limited resources, which moti-
vates people to invade other people’s lands. Second, to feel safe, mutual distrust
ironically induces them to attack each other. Third, people lock horns to achieve
glory—invasion for reputation. In his view, the existence of a social contract, if
enforced, banishes serious intolerance, violent conflict, and possible mass killings
from organized societies. However, enforcement requires them to be sufficiently
low cost; otherwise, it may not be possible to keep the sources of conflict at bay,
which can then cascade into mass killings as an end-state of group violence.4
Will rational, purposeful agents engage in peace or engage in killings?
Harsanyi’s (1961) doctrine argues that equally well-informed and rational citi-
zens will come to share the conclusion of how the end-game of conflict will play
out: if conflict is costly, and if it can end up in mass killings, then intelligent citi-
zens will shy away from mass violence. This is akin to two chess masters who, at
514 Case Studies II
some stage in the match, can predict that the end game will be a draw; they there-
fore decide to declare a draw before the match actually reaches the end game. But
when such ideal-t ype behavior is not possible, society tackles the enforcement (of
self-control) problem by erecting customs and social norms that influence indi-
vidual and group behavior. What specific action a person ends up choosing is
influenced by existing customs and social norms. 5 Thus, these norms, and how
persons interpret them, need to feature in the analysis of conflict.
In the context of group violence, any one action can take on different social
interpretations: a suicide bomber may be one group’s martyr, yet a criminal for
another. Serious problems can arise when an action, and its social reception and
subsequent consequences, is not so clear-cut. In a heterogeneous society, social
capital plays an important role in eradicating group violence. Social capital typi-
cally highlights those attributes, or virtues, in society that convert individual
people into a community (Putnam 1993; Bowles 1999; Durlauf 1999). These
attributes dictate the relationships among people wherefrom a course of action
is chosen and group violence is shunned. In contrast, the model presented in
section 22.3 explains the dynamics of individual behavior whereby society can
evolve either to shun or to embrace collective violence. The major innovation in
this chapter lies in the argument that customs and social norms allow multiple
social interpretations of the same action any one citizen takes. As a result, any
one action can lead to multiple possible outcomes. The uncertainty this entails
then can seriously impinge on the actual, overall social outcome. In the model
presented in section 22.3, the social contract is not fully enforceable and its rules
are subject to interpretation by society’s power-brokers. In addition, customs
and norms, which fill out the gaps in the social contract, likewise are subject to
interpretation.
22.3. Exogenous Social Interactions and the

Social Dynamics of Mass Killings
In economic theory, mass killing is predicated on the pecuniary and nonpecu-
niary costs an action incurs relative to the benefits it offers to its architects and
perpetrators. If the cost exceeds, or can be raised to exceed, the expected benefit,
then mass killing will not take place, as it is not rational to engage in an action
whose cost is larger than its benefit. Thus, society settles on a mass-k illing-free
equilibrium of social interactions. Conversely, mass killings are predicted to
occur if the cost to architects and perpetrators is less than the expected benefit.
In this section, I provide the rational foundations for the possibility that citi-
zens of a diverse society endogenously select an expected cost-benefit ratio in favor
of mass killing. The dynamics are modeled in terms of social interactions among
the agents of a multiethnic society. Social interactions take place whenever a citi-
zen’s payoff is a function not only of his own action but also of the actions of other
citizens (Townsend 1983). An extensive literature exists on the relevance of social

interactions for economic behavior, and in a wide range of contexts. For example,
Benabou (1993) considers social interactions in neighborhoods with regards to
education and crime. Diamond (1982) introduces social interactions in thick
market externalities in trading.6 The literature has shown that social interactions
can have a wide range of effects on equilibrium properties and that they can lead
either to conformity of behavior or to polarized actions (Bernheim 1994). Instead
of generating a unique equilibrium, social interactions can cause a multiplicity
of equilibria (Cooper and John 1988). More generally, Cooper and Haltiwanger
(1996) show how social interactions influence the dynamics or time-path of a sys-
tem. The model that follows shows how social interactions can generate chaotic
dynamic behavior, with far-reaching consequences for the economics of intoler-
ance and mass killing.
Assume that a representative citizen of group m receives utility Um from par-
ticipating in acts of intolerance—including killings, Tm—undertaken by group m:
U m = 1 2 ATm2 − ωB(Tm − α 0T A )Tm , (1a)
where B = 1 for (Tm –TA) > 0 and B = 0 for (Tm–TA) < 0, ω is the probability that
(Tm –TA) > 0, TA is the average social opinion about the likely size of killing that is
deemed tolerable to group m, and α 0 is the degree of lenience or tolerance to mass
killing.
In the model, social interactions are part of the citizen’s objective function as
his utility depends on TA, the opinion of others. In this sense, social interactions
are exogenous. Banerjee (1992) offers a model of social interactions in a sequen-
tial decision-making framework in which a special form of social interactions,
herd behavior, arises endogenously. As opposed to herd behavior, the model pre-
sented here highlights only exogenous social interactions. It also captures the idea
of anti-social capital, as discussed in section 22.1. To repeat, anti-social capital
arises when a citizen displays a lack of trust and commitment, is hostile, and/or
causes economic harm to a member of a rival group so long as his or her own refer-
ence group allows, tolerates, and possibly rewards his or her efforts.
The first-order condition of utility maximization yields:
Tm = [α0(ωB) / (2pB − A)]T A , (1b)
and the second-order condition requires that
2ωB − A > 0. (1c)
The representative citizen from group m believes that his act of participa-
tion in intolerance and killing alters neither the average opinion nor the degree
516 Case Studies II
of lenience. Thus, social interactions are exogenous, and a level of intolerance

(including killings) of Tm is chosen such as to maximize Um, and not to influence
TA and/or α 0 . However, a citizen from group m does hold the view that “others as
a group” can shape both TA and α 0 . From this, the following belief functions are
postulated:
Postulate 1: A citizen from group m believes that society shows increasing intol-
erance to mass killing by others as a group. For citizen m, this is modeled with
TA as a decreasing function of Tj, where j denotes the rest of the society (that is,
society other than group m). The functional relationship is
T A = T Max − εTj , for ε > 0, (1d’)
where TMax represents the maximum level of mass killing beyond which soci-
ety displays zero tolerance. Any intolerance beyond TMax by anyone in group m
attracts a serious and costly penalty to the perpetrator. This is where the social
contract “bites,” once the seriousness of the offense becomes very high. Now, date
this relationship as follows. Citizen m believes that the average opinion, TA, at
time t+1 depends on acts of killing by others at time t:
T A(t + 1) = T Max − εTj(t ). (1d)
Postulate 2: Citizen m believes that his own choice of killing does not affect social
lenience toward killing. However, social lenience depends on the acts of intoler-
ance of others. Thus, the time path of α 0 is
α t0+ 1 = τTj(t ), where τ > 0. (1e)
In similar fashion, one can describe the dynamics of Tj (t+1).

Theorem 1: Assuming that there are two groups of citizens, the dynamics of Ti
and Tj are as follows:
Ti(t + 1) = [(mωB) / (2ωB − A)]T MaxTj(t )[(1 − εTj(t ) / T Max)]. (2a)
Tj(t + 1) = [(mωB) / (2ωB − A)]T MaxTi(t )[(1 − εTi(t ) / T Max)]. (2b)
With a suitable change of variables, one obtains
k 0 = Tε
i /T
Max (2c)
k1 = Tjε / T Max. (2d)

The dynamics can be restated as
k 0(t + 1) = [(mωB) / (2ωB − A)]k1(t )(1 − k1(t )) = χk1(t )(1 − k1(t )) (3a)
k1(t + 1) = [(mωB) / (2ωB − A)]k 0(t )(1 − k 0(t )) = χk 0(t )(1 − k 0(t )) (3b)
χ = [(mωB) / (2ωB − A)]. (3c)
Proof: The proof, being simple, is omitted.

These dynamics are well recognized in the physical and economic sciences
as those of the quadratic map (May 1976; Feigenbaum 1978; Benhabib and
Day 1980, 1982; Day 1983, 1994). One can expect interesting local and global
bifurcation phenomena. Based on this line of research, I offer the following
theorem that characterizes the behavior of the dynamic system laid down in
(3a) and (3b).
Theorem 2: The dynamic system (3a) and (3b) and the first and second iterates
of these difference equations have four fixed points: E1 = (0,0), E2 = ((χ–1) /χ,
(χ–1) /χ), E3 = (k0*, k1*), and E4 = (k1*, k0*), where
k 0 * = (χ + 1 − (χ + 1)(χ − 3) / (2χ ) ) (4a)
k1 * = (χ + 1 + (χ + 1)(χ − 3) / (2χ ) ). (4b)
The stability properties of these fixed points are given by the following:
• E1 is a stable node if 0<χ<1 and is the unique Nash equilibrium. For χ>1,
E1 loses its stability, and for χ = 1 we have a transcritical bifurcation and E2
becomes stable for 1<χ<3.
• For χ>3 a stable two-period cycle is generated.
• For 3 < χ < (1 + 6 ), both E3 and E4 are stable. Which one gets established
depends on the initial condition.
• For χ > (1 + 6 ), levels of intolerance evolve through a cycle of infinite peri-
ods. The levels of intolerance are within the relevant bounds but never repeat.
For a higher order, these levels of intolerance may look like a random process
but they are fully deterministic.
Proof: The fixed points are derived from the quadratic map and their first iter-
ates. One then differentiates (3a) and (3b) with respect to k1 and k0 and forms the
Jacobian. The stability properties are arrived at by evaluating the Jacobian at these
fixed points. QED.
518 Case Studies II
22.4. Equilibrium Selection: The Replicator

Dynamics
I introduce the possibility that social behavior can be copied, or replicated, by citi-
zens over time in an evolutionary manner. Citizens have an incentive to update
their strategies if current strategies fail to yield an average return. This leads to
the possibility of equilibrium self-selection. In the model, there are two possible
equilibria: first, the mass-k illing-f ree equilibrium; and, second, the mass-k illing
equilibrium. The critical question regards the selection of the equilibrium by
ordinary citizens. Following Taylor and Jonker (1978), it is presumed that indi-
vidual citizens are programmed to play pure strategies. At any point in time, we
interpret the mixed strategy, h, as a population state, each component hi repre-
senting the population share of citizens programmed to play strategy i. In what
follows, replicator dynamics are presented as an ordinary differential equation,
and the stability properties of this differential equation are examined. It is well-
known in the literature that those stationary states that are dynamically unstable
do not correspond to the overall Nash equilibrium behavior. Assume that each
citizen has a utility function arising from intolerance as given by (4a). We define
h1 as the proportion of the population programmed to choose intended levels of
mass killing greater than TA, that is, the average level of propensity to undertake
mass killing. Define h2 (= 1–h1) as the proportion of the population programmed
to choose T less than the average TA. Table 22.1 summarizes the static game: Two
citizens meeting at random each can choose either strategy S1 or strategy S2. For
each citizen i, S1 = Ti such that Ti = TA+d > TA and S2 = Ti = TA–d ≤ TA, where d > 0.
In Table 22.1, H = ATi2/2(Ti–TA)Ti/2, A1 = A(TA+d)2/2, B1 = A(TA–d)2/2, and A2 =
ATA2/2. Note that A1>A2>B1>H1.
Based on Table 22.1, the expected return from strategy S1 is given as EU(S1, S):
EU(S1, S) = h1B1 + (1 − h1)H . (5a)
Similarly, the expected return from strategy S2 is given by EU(S2, S)
EU(S2 , S) = h1B1 + (1 − h1)A 2 . (5b)
Table 22.1 The Static Game
Citizen II
S1 S2
S1 A1, A1 H, B1
Citizen I
S2 B1, H A 2, A 2
The population average payoff/utility is EU(S, S)
EU(S , S) = (1 − h1)B1 + (1 − h1)2 A 2 . (5c)
Equation (5c) gives the expected utility or payoff to a citizen drawn at random
from the population. Note that this is the payoff that the mixed strategy h earns
when played against itself.
The replicator dynamics are given as
dh1 / dt = h1[EU(S1, S) − EU(S , S)] (6a)
dh1 / dt = [h22A1 − (1 − h1)h1H − h1(1 − h1)B1 − (1 − h1)2 A 2 ]. (6b)
Equation (6a) states that the growth rate of the population share using strategy
1 is equal to the difference between the strategy’s current payoff (in biology,
reproductive fitness) and the current average payoff. Equation (6b) is described
in Figure 22.1. By differentiating the EU(S, S) function we get the separatrix h1*:
h1* = (A 2 − B1) + (A 2 − H) / [2{(A1 − H) + (A 2 − B1)}]. (6c)
For h1 < h1*, the replicator dynamics converge to h1 = 0 and for h1 > h1*, the rep-
licator dynamics converge to h1 = 1.
Note that h1 is the proportion of the population having mass killing prefer-
ences, or tendencies, above the average such preference in society. The message of
this analysis is that the i nitial value of h1 determines the end state of the dynamics charac-
terizing the game. The analysis indicates a bipolar world. First, note that mass kill-
ings will gradually decline and finally disappear in societies for which the initial
dh1/dt
A1
A2
dh1/dt<0 dh1/dt>0
h1
0 h*1 h1 = 1
Figure 22.1 Quadratic map of replicator dynamics.

520 Case Studies II
value of h1 is bounded below h1*. In such societies, peace is gradually established

as more and more people voluntarily choose to shun violence after observing the
behavior of others. Second, mass killings will gradually rise and then explode in
societies for which h1 > h1* since more and more people come to choose mass kill-
ing from their social interactions. As a result, h1* acts as a trap-door: As long as
the value of h1 is contained below h1*, society will gradually evolve to banish mass
killing.
22.5. Economics of Religious Mass Killings: The

Case of Pakistan, 1978–2012
The theoretical model explains how the dynamics of mass killings are influenced
by three critical elements: First, the underlying game of interactions among
agents in society; second, the economic structure or factors that affect the payoff
from mass killings; and third, how strategies spread in a population. In what fol-
lows, I try to understand these factors empirically as they apply to the creation
and perpetuation of religious mass killings in Pakistan, 1978 to 2012.
22.5.1. Macro-versus Microlevel Analysis

When focused on the macrolevel, mass-k illing-related research faces serious limi-
tations (see c hapter 9 of this volume). First and foremost, it is well recognized
that in most societies mass killings are a rare event. As a result, any dataset will
have a large frequency of zero values. The universe of cases is small, and any infer-
ences made would call for cautious interpretation. Second, comparisons across
nation-states are restricted by unit heterogeneity, meaning that the unit of analy-
sis—countries—shows a wide range of differences, or heterogeneity. Third, in
the case of quantitative cross-national analyses, the exact causal mechanisms are
not easy to establish. While correlates of mass killings, including genocide, can be
established with some confidence, proffered reasons for why and how they matter
are on less secure ground (see chapter 10 in this volume). Fourth, in the case of
comparative historical analyses, most studies suffer from selection bias because of
their exclusion of negative cases, that is, cases in which mass killings could have
occurred but did not.
One promising approach to address some of these limitations is through disag-
gregation. By moving from macro-to microlevel analysis, a potentially larger set
of units to compare becomes available. These units enjoy greater homogeneity
(intrasociety similarity) and consequently have fewer differences for research-
ers to control for statistically. Importantly, disaggregation also usually involves
variation in the outcome variable of interest. From the social dynamics of mass
killings, as laid out in sections 22.3 and 22.4, killings of minorities by a majority
in every society start off as a rare event before cascading into routine incidents.
In time, the trend can explode into large-scale killings. Hence, it is important to
track the dynamics of violence over a long period of time, however small their
initial magnitude and frequency. That is what this section does.
The focus of the empirical analysis here is whether one can establish economic
underpinnings to killings of Shiite Muslims in Pakistan, a country where Sunni
Muslims constitute the population’s majority.7 For the 1978–2012 time period,
I assembled disaggregated data on Shiite killings across Pakistan’s five major
provinces, killings claimed by Pakistani Sunni-related groups. Recent incidents
indicate that an end to such killings is unlikely, and so they may yet escalate to
large-scale killings. The methodology employed is a microeconometric study
of subnational episodes of Shiite killings by Sunnis in Pakistan, killings whose
occurrence varies over time and regions (see King 2004).
It is hypothesized that the killings are predicated on three important economic
factors. First, positive economic development tends to reduce the intensity of
conflict between the groups, which lowers the incidence of persecutions and kill-
ings. Second, killings tend to decline as the productivity of land improves, which
increases the opportunity cost of organizing and engaging in killings. This is
especially important since poverty in most South Asian nations today is predomi-
nantly rural poverty. Third, an improvement in rural employment for unskilled
workers lowers the number of killing incidents because the opportunity cost of
killing rises with greater employment opportunities (and vice versa).
22.5.2. Methodology: The Equations of Killing

The empirical model captures the killing across five Pakistani provinces, mea-
sured by the number of Shiite Muslims killed in a given year. Because data on
killings are discrete (i.e., not continuous), a count model is used. But in many
provinces in many years, no killing incidents at all are reported, resulting in many
zero counts in the dataset. To take account of excess zeroes, variants of the stan-
dard Poisson model are employed, namely a zero-inflated count model and a neg-
ative binomial panel regression model.
22.5.1.1. Standard Poisson Model

The Poisson regression model for panel data is described in detail by Cameron
and Trivedi (1998). The dependent variable Yit, killings of Shiites in Pakistan, var-
ies across provinces (i=1,…,n) and over time (t=1,…,t1). The variable is assumed
to have a Poisson distribution with parameter λit, which depends on a vector of
exogenous variables, Xit, according to the log-l inear function
λ it = e δi + βX , (7a)
522 Case Studies II
where δ i is the fixed effect. One way to estimate this model is to conduct a conven-
tional Poisson regression by maximum likelihood estimation, including dummy
variables for all individuals (less one) to directly estimate the fixed effects. An
alternative random effects model will also be used to capture intercept heteroge-
neity (Hausman, Hall, and Griliches 1984).
22.5.1.2. The Zero-Inflated Poisson (ZIP) Equation

Since the dataset has a large frequency of zero values, a zero-inflated Poisson
(ZIP) model of cross-sectional analysis is applied to capture the effects, if any, of
economic factors in determining the killing of Shiites in Pakistan. The ZIP model
has been shown to be a useful alternative to the standard Poisson distribution in
the presence of a large frequency of zero values (Mullahy 1986). It is based on a
finite mixture model of two distributions, combining an indicator distribution for
the zero case and a standard count distribution (Mullahy 1986; Lambert 1992),
and it usually provides a good data fit. The ZIP distribution has two parameters, φ
and λ, and has the following probability function:
ïìϕ + (1 − ϕ)e−λ , for n = 0
Pr[N=n] = ïí , (7a’)
ïï Ω , for n = 1, 2 … .
î
where
æ ö e λ λn
Ω = çç1 − φ÷÷÷ .
çè ÷ø n ! (7b)
One can also undertake the panel data modeling by treating the zero-inflated
component as an individual parameter, add random effects to the mean param-
eter of the Poisson distribution, or even use them together. This is addressed in
the next subsection, as it is widely held that negative binomial modeling is more
robust in this regard than is ZIP (Allison 1996). The distribution of ZIP can be
expressed as:
Pr[Yi = 0] = V1 (7c)
Pr[Yi = a > 0] = V 2 , (7c’)
where
e−λ iλ i a
V1 = ϕi + (1 − ϕi) (7c’’)
n!
e−λ iλ i a
V 2 = (1 − ϕi)
n! (7d)
and
′
λ i = e(X i β) and φi = Φ(X ′iγ + ϵi), (7e)
and where e is the exponential transformation, ‘denotes a transpose, β and γ are

vectors of parameters, Φ is the cumulative distribution function of the standard
normal, and X is a vector of regressors.
22.5.1.3. The Negative Binomial Equation

One potential problem with a Poisson panel regression is its assumption that
the mean of the count variable, Yit, is equal to its variance. If the mean is less than
the variance, as in fact it is for our case (see Table 22.2), the ideal estimator will be
given by the negative binomial model (Cameron and Trivedi 1998). In addition,
the Poisson model also implies the postulate that, in our case, observed killings
occur randomly and independently across provinces and time. The negative bino-
mial model for killings (Yit) overcomes these problems and can be written as the
following mass function F(∙):
V (λ it + Yit ) θ i Yit 1 λ it (7a”)

F((Yit )|λ it θ i) = ( ) ,
V (λ it )V (Yit ) (1 + θ i) 1 + θ i
y it
where V(∙) is the gamma function and θi is a constant over time for province i. In
Ln(λ it ) = βXit , (7d’)
λit is decomposed as a function of the covariates, and the likelihood function is

minimized by the appropriate choice of β. Xit is the covariate for the i-t h province
at time t. An alternative model with random effects will capture intercept hetero-
geneity (Boucher, Denuit, and Guillen 2009).
22.5.2. Data and Results

A nonspatial dataset on Shiite killings in Pakistan, 1978 to 2012, is available (see
below). For the spatial aspect of the study, five regions are considered. They are
Sindh, Punjab, Balochistan, Khyber Pakhtunkhwa (KPK; formerly the North-
West Frontier Province), and the Federally Administered Tribal Areas (FATA).
The regional location of each killing was then added to the killing dataset. In addi-
tion, data on various economic factors were collected by region across different
points in time.
524 Case Studies II
Three fundamental questions are then examined. First, how do patterns of reli-
gious killings vary over time and space in Pakistan, and are they driven by spatial
variations in employment and levels of economic development? Second, how do
the data help one to decipher the effect of economic development on religious
killings in Pakistan? Third, does this empirical work assist us in determining
which factors, if any, may thwart the disturbing upward trend of religious violence
against Shiite Muslims in Pakistan?
22.5.2.1. Dataset
The dataset is created by collecting publicly available information. Incidents
of religious killings are taken from the “Shia Genocide Database: Killings in
Pakistan from 1963 to 31 May 2013” webpage. Each incident has been confirmed
from English-language newspapers published in the provinces.8 The economic
data is taken from various issues of official publications of the Federal Bureau of
Statistics (Economics Wing) of Pakistan.9
The variables are chosen to reflect the specific situation of Pakistan, estab-
lished from the author’s interactions with Pakistani expatriates, both Shiite and
Sunni. There are three critical issues here. First, the Shiite community has been
living in Pakistan, mostly peacefully, for more than a millennium. It is widely held
that this minority group originated primarily in Iran and modern-day Kurdistan
and then intermingled with the local (mostly Brahui) people to create the modern
Pakistani-Shiite community. Thus, some beliefs are held about genetic differences
between Shiite and Sunni communities in Pakistan. The factual veracity of these
beliefs cannot easily be confirmed, but beliefs nonetheless influence behavior.
Second, the Sunni majority in Pakistan receives widespread financial support
from some of the oil-rich, Sunni-majority Gulf nations to stir up violence, persecu-
tions, and killings in the Shiite community in Pakistan. Violence directed against
Shiite Muslims are criminal acts, of course, and Pakistan’s entire administration
and law enforcement agencies seek to deter such incidents. However, the killings
tend to be carried out by locals who are called “ultra-poor” in South Asia. They
commit these heinous crimes, forbidden in Islam, because they are “angry” about
their own economic plight. The Shiite minority is used as a scapegoat, and it is
an easy target to boot. The ultra-poor may also be viewed as having relatively low
opportunity costs as well as being desperate to earn a living for their families. Thus,
to understand the dynamics of religious killings, it is of great importance to exam-
ine the dynamics of rural (extreme) poverty in the various regions of Pakistan.
Third, the literature has considered levels of economic development in order
to understand economic determinants of conflict. Yet it has consistently ignored
the ultra-poor in this regard. Bringing them into the limelight, statistically, we are
able to test whether extremes of poverty are an important determinant of Shiite
killings in Pakistan.
22.5.2.2. Economic Development and Killings: The Need for New Variables
In Pakistan, a major force of economic transformation in rural areas is the
(ongoing) agrarian revolution, or Green Revolution. This revolution requires rea-
sonably large-scale farm operations that can afford it to adopt a package of new
production techniques for the cultivation of “golden” crops like wheat, rice, and
cotton. The ultra-poor cannot afford mechanization, or land for that matter, at
the required scale and increasingly are pushed toward alternative, less profitable
crops for their own economic survival (e.g., barley). Thus, to understand the mate-
rial position of the ultra-poor, and their incentives for killing religious minorities,
one must shift one’s attention away from the “golden” crops of the modern(izing)
portion of the agricultural markets and, instead, turn toward the less profitable,
traditional crops. For each region, therefore, I consider three relevant variables,
namely, total land area used for barley cultivation (BA), the output level of barley
produced (BP), and the yield per hectare in barley cultivation (BY).
Any improvement in BA, BP, or BY enhances the welfare of the ultra-poor,
which, in turn, leads to lower incentives for and intensities of Shiite killings, as
we will see momentarily. In addition to crop production, one must consider the
opportunity cost of committing crime. This cost is larger if employment pros-
pects in the rural sector are improving. To capture those prospects, a proxy is
used, namely, the percentage of cultivated sugarcane crushed (SCR). Crushing of
sugarcane is a low-skill activity and mostly employs the ultra-poor. One expects
that the killings variable will bear an inverse relation to SCR: if SCR is high, then
job opportunities for the ultra-poor are high, and the opportunity cost of commit-
ting violence against Shiites is high as well. Table 22.2 provides summary statis-
tics for the relevant variables.
22.5.2.3. Barley, Sugarcane, and Shiite Killings: Results in Brief

Tables 22.3 to 22.6 present the econometric results. They may be summarized
as follows. For ten of the twelve model variations (7 in Table 22.3; 1 in Table 22.5;
and 2 in Table 22.6), the coefficient for sugarcane crushing (SCR)—t he proxy for
employment options of the ultra-poor—is statistically significant, indicating that
sugarcane crushing is systematically related to the killings. Specifically, in each
of the ten models, the coefficient bears the expected negative sign, meaning that
more sugarcane crushing is related to fewer killings. Given the theoretical prior,
this counts as statistical evidence in favor of the hypothesis that the killings are
reduced when the opportunity cost of killing is high, that is, when employment in
the sugarcane crushing industry is high (or unemployment is low) for low-skilled
workers. This suggests that creating rural jobs for low-skill, ultra-poor people may
help to advance more peaceable relations between Shiites and Sunnis.
Turn now from sugarcane to barley, and define positive economic develop-
ment for the ultra-poor as an increase in the yield (or productivity) of barley
526 Case Studies II
Table 22.2 Summary Statistics
Variables (X) Mean STD Min Max Definition of Variables

Killings 31.54 91.4 0 800 Number of Shiites killed
by Sunni Terrorists in four
regions of Pakistan per year*
SCR 50.34 22.2 6.93 98.32 % Sugarcane Crushed in the
region
BA 33.06 25.4 1.8 173.6 Barley Cultivation Area (in
‘000 hectares) in the region
BP 26.02 22.1 1.2 112.5 Barley Output (in ‘000 tons)
in the region
BY 733.84 311.1 257.6 1399 Barley Yield (in kg/hectare)
in the region
Observations 170 170 170 170
Group 5 5 5 5
Notes: Summary statistics are constructed by the author.

* Variables are defined for region i (i = 1,2, …, 5) at date t (t = 1,2, …, 34).
cultivation (BY). Again as expected, the empirical results show that an increase
in BY is statistically associated with fewer Shiite killings (the relevant coefficients
carry negative signs in eleven of the twelve models). Similarly, in all eleven of the
models that include the BA variable—t he total area (in hectares) under barley
cultivation—the coefficient is statistically significant and negative: more land
planted in barley lowers the number of Shiite killings. Thus, quite apart from the
productivity of barley cultivation, access to land itself (to produce barley to meet
the livelihood needs of the ultra-poor) becomes an important policy variable to
help decrease Shiite killings. (If one uses BA as an exposure variable and treats
BP, the actual tons of barley production, as an explanatory variable—Model 6 in
Table 22.3—t he results are not altered: higher barley production levels are statis-
tically associated with fewer killings.) If one excludes the Federally Administered
Tribal Areas (FATA) region from the analysis, as FATA has very little by way of
agrarian activities, the results are unchanged (Model 4 in Table 22.3).
Some of the model variations in Table 22.3 include linear (Models 2, 4, 5, 6) and
nonlinear (Model 3) time trends. All of the linear trend coefficient estimates show
statistically significant increases in killings over time (from 1978 to 2012). The
coefficient for the nonlinear trend in Model 3, however, is not statistically signifi-
cant, suggesting that there is no evidence of a decline in killings over time.
Instead of the random effects specifications in Models 1, 2, 3, 4, and 7
in Table 22.3, Models 5 and 6 are fixed effects specifications. Both produce
Table 22.3 Standard Poisson Panel Model of Religious Killings in Pakistan
Dependent Killings
Variable
Panel Poisson Estimation
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7

Random Effect Random Effect Random Effect Random Effect Fixed Effect Fixed Effect Random Effect
(1) (2) (3) (4) (5) (6) (7)

Constant –14.15 (–13.45*) 7.56 (7.77*) 7.78 (7.78*) 6.86 (9.19*) 0.66 (9.63*)
Ln BA –1.27 (–36.75*) –0.26 (–3.91*) –0.27 (–4.00*) –0.24 (–6.81*) –0.28 (–4.35*) –0.20 (–2 .94*)
Ln BP –0.47 (–11.83*)
Ln BY 3.47 (29.91*) –1.25 (–7.51*) –1.29 (–7.53*) –1.18 (–9.84*) –1.11 (–6.68*) – 0.09 (– 0.53) –1.45 (–9.43*)
SCR –0.015 (–9.65*) –0.009 (–5.63*) – 0.01 (–5.43*) –0.01 (–6.81*) – 0.01 (–5.62*) –0.014 (–8.57*) –0.009 (–5.60*)
Exclude FATA No No No Yes No No No
(Continued)
Dependent Killings
Variable
Lag (Killings) 0.0016 (18.96*)
t 2 – 0.0002 (– 0.99)
T 0.11 (38.12*) 0.12 (11.70*) 0.11 (4.31*) 0.11 (38.33*) 0.12 (52.38*)
Exposure Variable BP BP BP (BA) BP
Log Likelihood –8327.1 –8328.3 –7171.6 –6752.7 –7135.1 –7135.1 – 6986.66
Groups 5 5 5 4 5 5 5
Observations 170 170 170 165 170 170 165
Wald Chi 2 335.6 2311.44 4116.89 4283 4103.99 4828.99 4413.12
Notes: Z values are given in parentheses.

*Significant at 1%.
Table 22.4 Estimation of Random-Effects Variance Term, σ
Likelihood Test for σ = 0 Coefficient Standard Error

Ln σ –2 .12 0.72
Σ 0.11 0.083
Chibar2 66.7
Prob > Chibar2 0.01
Notes: Since 66.7 > 0, the random effects (RE) model is chosen instead of the pooling model.
This is true for all RE models.
Table 22.5 Zero-Inflated Poisson (ZIP) Model of Religious Killings in Pakistan
Model 1 Zero-inflated Cross-sectional S.E.

Poisson Model
Constant 18.16 (61.87*) 0.29
Ln SCR –0.99 (–3 4.51*) 0.02
Ln BA 0.60 (–25.27*) 0.023
Ln BY –1.68 (–47.61*) 0.035
Inflate constant 0.33 (2.14*) 0.15
Exposure variable BP
Log likelihood –4730.84
LR chi (3)
2
3065.68
Group 5
Observations 170
Notes: This is a cross-sectional model; panel characteristics are ignored; z-values in parentheses.
congruent results. Note that in the random effects models, it is assumed that the
exponentiated random effects are distributed as gamma, with mean one and vari-
ance σ. The regression in Table 22.4 therefore includes a likelihood-ratio test for
σ = 0, a test that formally compares the pooled estimator (Poisson) with the panel
estimator. The test suggests that the panel estimation is statistically meaningful
since the estimated variance of the error term, σ, of the random effects model is
different from zero. Finally, Tables 22.5 and 22.6 show the results when using the
ZIP model and the negative binomial model, respectively. The results conform to
those of the standard Poisson models in Table 22.3.
530 Case Studies II
Table 22.6 Negative Binomial Model of Panel Regression of Religious Killings in Pakistan
Killings Dependent
Model 1 Model 2 Model 3 Model 4

Random Effect Random Effect Fixed Effect GEE** Population
Average
(1) (2) (3) (4)

Constant 2.10 (0.67) 0.20 (0.94) 2.37 (3.13*) 15.65 (0.827)
Ln BA –1.19 (–6.53*) –1.22 (–5.71*) –0.27 (–4.00*) – 0.25 (– 6.64*)
Ln BY –0.67 (2.14*) –0.84 (–2 .37*) –1.29 (–7.53*) –1.88 (–2 6.77*)
SCR/Ln CR – 0.013 (– 0.88) –2 .37 (–5.63*) –0.22 (–0.43) –0.65 (–6.64*)
Lag (SCR) No No 0.02 (2.55*) Yes
Lag (Killings) 0.035 (4.26*) 0.0034 (4.10*) 0.015 (18.66*)
Log r 0.42 0.58
Log s 5.56 5.83
R 1.53 1.79
S 261.29 396.20
Exposure BP BP
Variable
Log Likelihood –4 61.46 – 4 64.33
Groups 5 5 5 5
Observations 170 170 170 170
Wald Chi 2
81.68 75 1329.89
Notes: This is a panel data model. Z values are given in brackets.

*Significant at 1%.
GEE: Generalized Estimating Equation.
22.6. Conclusion
In sum, we have five main empirical findings. First, in virtually all model and
regression variations, a statistically significant inverse relation is found between
the area under barley cultivation (BA) and Shiite killings: just increasing the bar-
ley area under cultivation itself would be expected to reduce the number of kill-
ings of Shiites in Pakistan. Second, the same finding of an inverse relation applies
to barley yields (BY): the higher the yields, the lower the number of killing inci-
dents. This suggests that policies to improve the productivity of barley cultivation
may be helpful in reducing the number of killing incidents. Third, the relation
between sugarcane crushing (SCR) and the killings is inverse and statistically
significant in ten out of twelve specifications. This suggests that increases in sug-
arcane production, and hence in the need to employ low-skilled labor, could help
to reduce killings as well. Fourth, the effect of lagged values of SCR on current
killings is ambiguous (insignificant results are not reported). Fifth, we have sta-
tistically significant evidence to show inertia and persistence of the killings: past
killings influence current killings.
My primary goal in this work is to understand the economics of religious
killings, which is a form of collective violence, from social interactions. Social
interactions take place whenever an individual return is a function not only
of one’s own action but also the actions of other citizens in the model. My
theoretical work makes three major contributions. First, it shows that (posi-
tive or negative) income shocks from economic development can interact with
social interactions to create the possibility that citizens in a diverse society can
endogenously select the cost relative to the benefit of mass killings. From the
proposed feedback between social interactions and economic changes I charac-
terize rich dynamics involving religious killings. These can be volatile and, ex
ante, unpredictable. Subtle changes in economic factors and social perceptions
can have dramatic effects on the incentives to undertake large-scale killings. In
contrast to my model, traditional economic theory of mass killings proposes
that mass killings arise from a costly lottery or costly bargain between two war-
ring groups. Each group has economic and military powers to engage in fights
that can escalate into attempts for complete annihilation of one group. Both
groups can also settle for peaceful coexistence. From Anderton (2010) we know
that mass killings will take place, as per the economic theory of genocide, if a
peaceful resolution cannot be reached. The main weakness of the traditional
model is not that it cannot explain mass killings if the rival group does not pose
any immediate threat, which is often the case, but rather that subtle changes in
the conditions can quickly aggregate to a mass killing outcome. My theoretical
model fills this important gap in the theory of mass killings for cases when there
is no explicit threat from a minority group like the Shiite Muslims in Pakistan
to the majority Sunni community. Second, from the model one can argue that
adverse, local economic shocks can incentivize citizens to form small (warring)
groups, which move spatially in search of resources for survival, and that can,
in turn, precipitate mass killings. Finally, events of mass killings can create a
culture of killing to “resolve” economic problems. A culture of killing can per-
petuate beyond short-term shocks and beget further mass killings even in better
economic times. The theory thus highlights how temporary economic shocks
can critically alter the dynamics of interpersonal and intergroup relations such
that mass killings can recur regularly as economic conditions undergo minor
changes. The empirical section ascertains the relevance of economic factors,
hitherto not recognized in the literature, for creating and perpetuating religious
killings in Pakistan.
532 Case Studies II
Notes
1. On mass killings, see Anderton (2010); on heterogeneity, see Alesina and La Ferrara
(2002, 2003). A diverse ethnic mix gives rise to variety in abilities, experiences, and
cultures. These are beneficial and can lead to innovation and creativity (Alesina and La
Ferrara 2000; Berman 2000; Laitin 1995; Greif 1993). They can also lead to great costs
in the form of discrimination, misgovernance, racism, open and tacit conflicts, preju-
dices, terrorism, and civil unrest (Alesina et al. 2003; Pratt 2002; Alesina and La Ferrara
2002; Easterly 2001; Collier 2000; Gangopadhyay and Nath 2001a, 2001b; La Porta
et al. 1999).
2. Gupta (2012) writes: “In a recent work Gangopadhyay (2009) shows the link between
intolerance and violent conflicts. Gangopadhyay’s model … posits violent conflict as
a form of intolerance. Intolerance by one group begets intolerance by the other group;
beyond a point violent conflict perpetuates itself. It is interesting to see in Gangopadhyay’s
paper that for such an occurrence, potential marginal penalties should not be too
high” (21).
3. The literature is still evolving. Gupta (2012) develops a two-stage game to examine how
the opportunity cost of the leader of a national protest movement, and the intrinsic hawk-
ishness or pacifism of an occupier, determine the nature of the movement against occupa-
tion. In the model, the protest leader can actively convert the populace to protest while
the occupier chooses how much to punish the protest leader and other protesters for their
actions. An interesting finding is that under certain circumstances leaders with greater
opportunity cost of leading protests may be more active, as compared to leaders with
lower opportunity costs. Further, the former may be able to lead a movement with more
mass support.
4. A social contract is to prevent the vulnerable from being molested by the powerful
(Rousseau 1964). In contract social, Rousseau popularized this idea of the social contract,
recognized as a major difference between the human and nonhuman worlds. Here lurks
Rousseau’s famous paradox: In sacrificing rights, nothing is really given up. Rousseau’s
solution to the paradox is that group members be both legislators and subjects, and that
they undertake their civil burdens diligently to express the true interests of society in voic-
ing its “general will.” Rousseau’s solution does not necessitate contract enforcement by
an omnipotent and omniscient state since citizens, driven by their civil duties, ensure its
enforcement. Hobbes, in Chapter XIII of Leviathan, realizes that it is not an easy task to
protect the vulnerable from the powerful in any society because the powerful will will-
fully take on this civil burden. Hobbes’s suggestion is to create a “common power” in the
social contract “to keep all in the awe.” Today, it is widely recognized that there is a need to
enforce social contracts by legislative mandate.
5. An example may be helpful: Consider the wage bargaining problem as outlined in Akerlof
(1980) in which union leaders are bound by members’ normative expectations to hold out
against management whose position, in turn, makes concessions equally unacceptable to
their stakeholders. Akerlof ’s idea is akin to the market in gifts, which is governed by norms
of gift giving: what is appropriate to give, to whom, and on which occasion. Typically,
these norms are cast in iron, which uniquely determine individual actions wherefrom a
social outcome evolves, given a well-defined and enforceable penalty mechanism.
6. Thick market externalities occur when sufficient geographic clustering of business enter-
prises allows networking interactions that reduce costs and magnify profits, which benefit
businesses but also consumers. Due to clustering, thick markets have many buyers and
many sellers who gain from agglomeration caused by large trading volumes.
7. Both Sunni and Shia are branches of Islam. Both groups are Muslim, bound by the Quran
and the five pillars of Islam—one God, charity, fasting, hajj, and ritual prayers. Their main
difference lies in terms of power: Sunnis believe that Muslim leaders can be chosen from
those who are qualified for the job; Shiites believes leaders must be drawn from the proph-
et’s dynasty.
8. Data on the killings of Shiites in Pakistan come from the following source: http://world-
muslimcongress.blogspot.com/2 013/0 6/shia-genocide-database-k illings-i n.html.
9. See Government of Pakistan (1977–2 012) in the references.
References
Akerlof, G. 1980. “A Theory of Social Custom of Which Unemployment May be One
Consequence.” Quarterly Journal of Economics 94, no. 4: 749–75.
Rationality of Race Murder.” Defence and Peace Economics 21, no. 5: 459–86.
Alesina, A., and E. La Ferrara. 2000. “Participation in Heterogeneous Communities.” Quarterly
Journal of Economics 115, no. 3: 847–9 04.
Alesina, A., and E. La Ferrara. 2002. “Who Trusts Others?” Journal of Public Economics 85, no.
2: 207–3 4.
Alesina, A., and E. La Ferrara. 2003. “Ethnic Diversity and Economic Performance.” National
Bureau of Economic Research (NBER). Working Paper. [Revised November 2004.]
Cambridge, MA: NBER.
Alesina, A., A. Devleeschauwer, W. Easterly, S. Kurlat, and R. Wacziarg. 2003. “Fractionalization.”
Allison, P. D. 1996. “Fixed-E ffects Partial Likelihood for Repeated Events.” Sociological Methods
and Research 25, no. 2: 207–22.
Banerjee, A. V. 1992. “A Simple Model of Herd Behavior.” Quarterly Journal of Economics 107, no.
3: 797–817.
Benabou, R. 1993. “Workings of a City: Location, Education and Production.” Quarterly Journal
of Economics 108, no. 3: 619–62.
Benhabib, J., and R. H. Day. 1980. “Erratic Accumulation.” Economics Letters 6, no. 2: 113–17.
Benhabib, J., and R. H. Day. 1982. “A Characterization of Erratic Dynamics in the Overlapping
Generations Model.” Journal of Economic Dynamics and Control 4, no. 1: 37–55.
Berman, E. 2000. “Sect, Subsidy and Sacrifice: An Economist’s View of Ultra Orthodox Jews.”
Quarterly Journal of Economics 115, no. 3: 905–5 4.
Bernheim, B. D. 1994. “A Theory of Conformity.” Journal of Political Economy 102, no. 5 :841–77.
Boucher, J. P., M. Denuit, and M. Guillen. 2009. “Number of Accidents or Number of Claims? An
Approach with Zero-I nflated Poisson Models for Panel Data.” Journal of Risk and Insurance
76, no. 4: 821–4 6.
Bowles, S. 1999. “‘Social Capital’ and Community Governance.” Focus 20, no. 3: 6–10. http://
www.irp.wisc.edu/publications/focus.htm#F20:3 [accessed April 11, 2015].
Cameron, A. C., and P. K. Trivedi. 1998. Regression Analysis of Count Data. Cambridge,
Chalk, F., and K. Jonassohn. 1990. The History and Sociology of Genocide: Analysis and Case Studies.
New Haven, CT: Yale University Press.
Collier, P. 2000. “Ethnicity, Politics and Economic Performance.” Economics and Politics 12, no.
3: 225–4 6.
Cooper, R., and J. Haltiwanger. 1996. “Evidence on Macroeconomic Complementarities.”
Review of Economic and Statistics 78, no. 1: 78–93.
Cooper, R., and A. John. 1988. “Coordinating Coordination Failures in Keynesian Models.”
Day, R. H. 1983. “The Emergence of Chaos from Classical Economic Growth.” Quarterly Journal
of Economics 98, no. 2: 202–3 4.
Day, R. H. 1994. Complex Economic Dynamics. Cambridge, MA: MIT Press.
534 Case Studies II
Diamond, P. A. 1982. “Aggregate Demand Management in Search Equilibrium.” Journal of

Durlauf, S. 1999. “The Case against Social Capital.” Focus 20, no. 3: 1–5. http://w ww.irp.wisc.
edu/publications/focus.htm#F20:3 [accessed April 11, 2015].
Esteban, J., M. Morelli, and D. Rohner. 2015. “Strategic Mass Killing.” Journal of Political
Economy 123, no. 5: 1087–132.
Easterly, W. 2001. “Can Institutions Resolve Ethnic Conflicts?” Economic Development and
Cultural Change 49, no. 4: 687–706.
Fearon, J. D. 1995.” Rationalist Explanations for War.” International Organization 49, no.
3: 379–414.
Feigenbaum, M. J. 1978. “Quantitative Universality for a Class of Non-Linear Transformations.”
Journal of Statistical Physics 19, no. 1: 25–31.
Gangopadhyay, P. 2009. “Economics of Intolerance and Social Conflict.” Economics of Peace and
Security Journal 4, no. 2: 22–32.
Gangopadhyay, P., and S. Nath. 2001a. “Bargaining, Coalitions and Local Expenditure.” Urban
Studies 38, no. 13: 2379–91.
Gangopadhyay, P., and S. Nath. 2001b. “Deprivation and Incidence of Urban Public Services.”
Review of Urban and Regional Development Studies 13, no. 3: 207–2 0.
Gangopadhyay, P., and M. Chatterji. 2009. “An Economic Study of Ethnic Heterogeneity and
Its Implications for Conflict and Peace.” In P. Gangopadhyay and M. Chatterji, eds., Peace
Science: Theory and Cases. Bingley, UK: Emerald, 67–106.
Government of Pakistan. 1977–2 012. “Crops Area and Production by Districts.” Various Issues.
Federal Bureau of Statistics, Islamabad, Pakistan.
Greif, A. 1993. “Contract Enforceability and Economic Institutions in Early Trade: The
Maghribi Traders’ Coalition.” American Economic Review 83, no. 3: 525–4 8.
Gupta, R. 2012. “The Effect of Opportunity Cost and Hawkishness in Protests in Occupied
Regions.” Defence and Peace Economics 23, no. 1: 17–49.
Harsanyi, J. 1961. “On the Rationality Postulates Underlying the Theory of Cooperative Games.”
Hausman, J. A., B. H. Hall, and Z. Griliches. 1984. “Econometric Models for Count Data with
Application to the Patents R and D Relationship.” Econometrica 52, no. 4: 909–38.
King, C. 2004. “The Micropolitics of Social Violence.” World Politics 56, no. 3: 431–55.
La Porta, R., F. Lopez-de-Silanes, A. Shleifer, and R. Vishny. 1999. “The Quality of Government.”
Journal of Law, Economics and Organization 15, no. 1: 222–79.
Laitin, D. 1995. “Marginality: A Microperspective.” Rationality and Society 7, no. 1: 31–57.
Lambert, D. 1992. “Zero-Inflated Poisson Regression with an Application to Defects in
Manufacturing.” Technometrics 34, no. 1: 1–14.
May, R. M. 1976. “Simple Mathematical Models with Very Complicated Dynamics.” Nature
261: 459–67.
Mullahy, J. 1986. “Specification and Testing in Some Modified Count Data Models.” Journal of
Econometrics 33, no. 3: 341–65.
Political Instability Task Force. 2010. Genocides Dataset. http://g lobalpolicy.gmu.edu/pitf/.
Pratt, A. 2002. “How Homogeneous Should a Team Be?” European Economic Review 46, no.
7: 1187–1207.
Putnam, R. 1993. “The Prosperous Community: Social Capital and Public Life.” American
Prospect 13: 35–42.
Rousseau, J. J. 1964. The First and Second Discourses. Ed. R. D. Masters. New York: St. Martin’s.
Skaperdas, S. 2006. “Bargaining versus Fighting.” Defence and Peace Economics 17, no. 6: 657–76.
Taylor, P. D., and L. B. Jonker. 1978. “Evolutionary Stable Strategies and Game Dynamics.”
Mathematical Bioscience 40, nos. 1–2: 145–56.
Townsend, R. M. 1983. “Forecasting the Forecasts of Others.” Journal of Political Economy 91, no.
4: 546–88.
UNHCR. 2009. Statistical Year Book 2008. Geneva: UNHCR.
Valentino, B. 2004. The Final Solution: Mass Killing and Genocide in the 20th Century. Ithaca,
Waller, J. 2007. Becoming Evil. 2nd ed. New York: Oxford University Press.
23
Understanding Civil War Violence

through Military Intelligence
Mining Suspects’ Records from the Vietnam War
R e x W. Dougl a s s
23.1. Introduction
Civil wars blur the line between civilians and combatants. This is the fundamental
problem for governments that must separate rebels from innocents and for civil-
ians wanting to remain neutral and safe. Military and police forces expend enor-
mous resources attempting to identify and eliminate specific individuals fighting
for violent groups. What do those programs look like from the inside? How do
they pick their targets? How effective are they, and how are their costs distributed
across both civilians and rebel supporters?
Targeting programs are necessarily secretive, which makes government records
hard to come by. When available, they tend to consist of unwieldy, unstructured,
and often undigitized intelligence dossiers, which make quantitative analysis an
expensive proposition.1 Because of these problems, nearly all existing studies of
targeting programs are either qualitative in nature (Moyar and Summers 1997;
Comber 2008; Natapoff 2009) or depend on data developed from nongovern-
mental sources like interviews, surveys, and news reports (Ball, Tabeau, and
Verwimp 2007; Silva, Marwaha, and Klingner 2009).2 When declassified govern-
ment records are available, they tend to be at the event level, like attacks (Berman,
Shapiro, and Felter 2011; Biddle, Friedman, and Shapiro 2012) or air operations
(Lyall 2014) without details on the victims. The rare exceptions are peacetime
police records like stop-and-f risk data from New York, which provide information
on both demographic details about the suspects and details of the altercations
(Gelman, Fagan, and Kiss 2007).
I analyze an electronic database of civilian targeting efforts created during the
Vietnam War. The data are extensive, covering 73,712 individual rebel suspects
536
U n d e r s t a n d i n g C i v i l Wa r V i o l e n c e t h r o u g h M i l i t a r y I n t e l l i g e n c e 537
from the perspective of the government’s police and military operations. The
database contains detailed information on both the victims and the operations
targeting them. 3 At issue are two central questions.
The first is descriptive: How exactly does a civilian targeting program work
in practice? All civilian targeting programs are secretive, but the Phoenix
Program (Phung Hoang) is uniquely surrounded in historical controversy.
During and immediately after the war, critics and proponents debated whether
it was a broad intelligence and policing effort or simply a punitive assassina-
tion program (Colby and McCargar 1989). Since the war, the discussion has
turned to whether the Phoenix Program achieved its aims of neutralizing high-
ranking targets (Thayer 1985). More recently, it has been asked why the pro-
gram caught some suspects while letting others escape, and what implications
that might have for civilians deciding whether to join rebellions (Kalyvas and
Kocher 2007).
The second is theoretical: How should we conceptualize civilian targeting? Is
there a simpler topology we can use to classify violence against civilians in terms
of the kinds of victims or the kinds of methods employed? Much of the work on
civilian targeting either explicitly or implicitly disaggregates along the severity of
targeting, treating killings as distinct from arrests because they are theoretically
different or easier to document. Others divide targeting along how the victims
are selected, particularly whether they were individually singled out or targeted
as part of a larger group (Kalyvas 2006). Is there a principled and data-d riven way
to categorize and describe civilian targeting?
Providing an answer to both questions requires a highly inductive and multi-
variate approach. The overall analytical strategy is familiar in the machine-learning
literature. I start with a database I did not create and for which I have partial, incom-
plete documentation. Using clues like patterns of missingness, I determine that
there are related groups of attributes and records drawn from completely different
subpopulations of suspects. I then show how to prioritize and reduce the over forty
available attributes for each suspect to a manageable core group of key facts. I finally
perform dimensionality reduction on those facts, finding useful and easy to under-
stand dimensions on which both suspects and targeting tactics varied—providing
an answer to both how we should conceptualize civilian targeting and how the
Phoenix Program operated.
In the course of conducting the analysis, I intend the chapter to serve as a
primer for handling large, multivariate, found conflict data. An increasing number
of scholars work with observational data that they did not create, and over which
they had no control. They spend extraordinary effort on model specification for a
suspected empirical data-generating process, but typically pay little attention to
the institutional data-generating process underlying the reporting and recording
of facts. The Phoenix Program database provides an example of how working with
found data is at best an adversarial relationship. This poses unique problems for
538 Case Studies II
inference, but I demonstrate some of the growing number of tools that help to
tackle these kinds of problems.
The road map of the chapter is as follows. Section 23.2 provides an empirical
background for civilian targeting during the Vietnam War. Section 23.3 provides
a broad overview of the data created to track that targeting. A detailed examina-
tion of the database reveals undocumented heterogeneity between different kinds
of records, groups of related attributes, and a ranking of attributes’ importance,
pinpointing exactly where to start the analysis.
Section 23.4 develops a taxonomy of victims. Dimensionality reduction on a
set of key attributes reveals two key differences between suspects: (1) priority,
those who the government wishes it could target and those it targets in practice;
and (2) severity, how violent was the suspect’s final fate, ranging from voluntary
defection, to arrest or capture, and, at the extreme, death.
Section 23.5 develops a taxonomy of tactics. Dimensionality reduction just on
those observations where the government carried out an operation (killings and
arrests) reveals differences among tactics along: (1) priority, but also premedita-
tion, how specifically the individual was targeted prior to the arrest or killing;
and (2) domain, whether a suspect was targeted in operations with some connec-
tion to policing and intelligence resources, or whether the suspect was targeted by
third-party forces and then retroactively reported.
Finally, section 23.6 concludes with broader implications for security studies
and suggests avenues for future research.
23.2. The Empirical Background: War in South

Vietnam and the Phoenix Program
In the later years of the Vietnam War, the South Vietnamese government went on
the offensive against the nonmilitary members and supporters of the rebel oppo-
sition. While politically effective, the 1968 Tet Offensive was a military setback
for irregular forces that shifted the main military threat from the organized rural
insurgency to the incoming conventional forces from North Vietnam. The gov-
ernment took advantage of this shift by pushing out into rural areas, projecting its
power with a wide range of police, militia, military, and special forces units. With
the expansion of government control and a flood of funds, logistical support, and
US advisers, the government set out to map, monitor, and police its civilian popu-
lation on an industrial scale. Goals were set out to dismantle the political opposi-
tion with widespread arrests, killings, and induced defections.
The Phoenix Program was created in 1968 to coordinate this initiative—
which was, in reality, fractured across dozens of separate intelligence and police
initiatives across South Vietnam. Its substance as an institution included national
guidelines that influenced targeting goals and reporting, as well as physical offices
containing advisers who coordinated and distributed information on suspects.

While the program was not responsible for directly acting against suspects, it was
instrumental in bringing together and documenting all of the scattered existing
efforts (Moyar and Summers 1997). It was in operation until it collapsed during
the Eastertide invasion by North Vietnam at the end of 1972.4
The process by which the program collected information was scattered and
disjoint. Intelligence and Operations Coordinating Centers (IOCCs) main-
tained Lists of Communist Offenders and sometimes detailed maps of hamlets,
with names of occupants and photographs. 5 Military units like the First Infantry
Brigade, Fifth Infantry Division (mechanized) sometimes formed special teams
from its military intelligence detachment to coordinate with and make up for
weak IOCCs in their area, maintaining their own card files in addition to the local
List of Communist Offenders.6 Additionally, village, district, and province chiefs
often maintained their own parallel intelligence nets and records.7
In January 1969, the Office of the Secretary through the Advanced Research
Projects Agency (the predecessor to the Defense Advanced Research Projects
Agency or DARPA) began a project to develop a country-w ide information
reporting system to coordinate and allow comparative analysis of IOCC infor-
mation.8 Biographical data on suspects and neutralizations were entered into the
national database at district and province IOCCs. Enumerators were trained with
coding guidelines for converting dossiers and neutralization reports into a stan-
dardized format.9 The data were then passed up to the Phung Hoang Directorate
and from there to the National Police Command Data Management Center,
where they were recorded using punchcards before being entered onto magnetic
tape with an IBM360 mainframe.
23.3. Overview of the Targeting Database

The United States preserved an electronic copy of the final targeting database
called the National Police Infrastructure Analysis Subsystem II (NPIASS-I I).10 It
contains records from July 1970 to December 1972. This is the later period of the
war, following the Tet Offensive, the Republic of Vietnam (i.e., South Vietnam
or GVN) and the US counterattack, and the period of US withdrawal. It contains
data on 73,712 individuals (rows) that I refer to as “suspects” and who serve as the
unit of observation. It contains 45 attributes (columns) that provide information
on each suspect about their biography, job, details of operations targeted against
them, and their final disposition.11 The attributes are of mixed types includ-
ing numerical, nominal, dates, and nested lookup codes such as locations (e.g.,
region=>province=>district=>village) and rebel jobs (e.g., National Liberation
Front=>Liberation Woman’s Association=>Personnel). The structure of the data-
base is outlined in Table 23.1.
540 Case Studies II
Table 23.1 Structure of the National Police Infrastructure Analysis

Subsystem II (NPIASS-I I) Database
Record Type Status Outcome

Neutralization Killed (15,438)
Record (48,074) Neutralized
Suspects Captured (22,215)
(49,756)
(73,712) Biographical Record Defector (12,103)
(25,638) At-Large (23,943) At-Large (23,943)
Notes: The unit of observation (rows) is the individual suspect. The potential attributes (col-
umns) are available in blocks depending on the kind of record and outcome.
The first-order task with found data such as these is to verify the structure of
the database. Available codebooks often do not document important details, and
when they do, the documentation is often at odds with how the database was used
in practice. I take a two-pronged approach. First, I point to archival evidence about
the genesis and day-to-day use of the program. In particular, I located detailed cod-
ing instructions for a precursor database, the VCI Neutralization and Identification
Information System (VCINIIS), which appears to share most if not all of the same
properties of the final NPIASS-I I database.12 Second, I apply a machine-learning
approach to the structure of the database overall, not just the tabular values of indi-
vidual variables. Patterns of missing values, the meaning of different variables, and
heterogeneity among different kinds of observations are all targets of inquiry.
One revelation from the documentation of the predecessor system is that the
database combines two different kinds of records: those entered while a suspect was
still at large (a biographical record) and those entered after a suspect had already
been killed, captured, or defected (a neutralization record). The two kinds of
records resulted from different worksheets, with separate text examples and coding
guidelines. Neutralization records appear to have recorded the Phoenix Program
as it actually happened on the ground. Two-thirds of suspects, 48,074 (65 percent),
were entered in as neutralization records. Biographical records were a kind of grow-
ing wish list where analysts bothered to digitize information from the much larger
pool of suspects with dossiers, on blacklists, or reported in the Political Order of
Battle. A third of suspects, 25,638 (35 percent), were entered this way.
The second important division in the database is along the fate of each suspect,
as still at large, killed, captured, or defector (rallier). All biographic records start
off as at-large suspects. In rare cases (6.6 percent), some were updated to show the
suspect had later been neutralized, although there are sufficient irregularities to
suspect that some of these are actually coding errors.
The Phoenix Program was not, at least predominately, an assassination pro-
gram. The dominant form of neutralization was arrests/captures (45 percent).
A smaller share (24 percent) of neutralizations were suspects that turned
themselves in, defecting. Killings made up only a third of neutralizations docu-

mented by the database (31 percent).
23.3.1. Example Narrative and Coding

To illustrate the kinds of information available (and the kinds of facts that are
omitted), I provide the following comparison of a suspect’s record in the database
with their detailed interrogation report. The record was deanonymized by manu-
ally comparing data fields against reported details in declassified interrogation
reports from the Combined Military Interrogation Center (CMIC).13 Below is a
brief narrative of a suspect’s life, career, circumstance of defection, and immedi-
ate aftermath, drawn from his interrogation report, and with facts corroborated
by the NPIASS-I I database italicized in the text.
In 1929, a man named To Van Xiem was born in Thai Binh Province (North
Vietnam). He attended four years of village school and worked on his parents’
farm until January 1950 when he joined the Viet Minh. He was a probationary
member beginning June 1950 and became a full party member in 1951. Over the
next sixteen years he was promoted or reassigned to multiple roles within North
Vietnam until February 25, 1966, when he and fifty-six other civilian political cad-
res infiltrated South Vietnam, arriving in Tay Ninh Province. In July of that year,
he was assigned as a cadre of the Ca Mau Province Party Committee Farmers’
Association Section. In March 1967 he was reassigned to the same position in
Soc Trang Province. In September 1969 he became a member of the Current
Affairs Section of the Farmers’ Association Section, living in Ba Xuyen Province. He
was a Buddhist, middle-class farmer, married twice with several children. Citing
increasing hardships, limited rations, and almost total government control of the
province, he defected at Xuyen District, Ba Xuyen Province on March 21, 1971. He was
previously unknown to security forces, not listed on any blacklist, and his identity
was confirmed by confession. Four days later, March 25, 1971, his neutralization
was entered into the national database. He was assigned a VCI serial number 41-
100825, indicating a new, previously unlisted neutralization record. He was interro-
gated on July 25, 1971 by the Combined Military Interrogation Center (CMIC)
in Saigon where he was assigned a CMIC number of 0297-71. Details from his
interrogation produced a thirty-page report that documented the names and
details of fifty-t wo other individuals as well as a number of regional organizations.
23.3.2. The Attributes

The full list of forty-five attributes is provided in Table 23.2. Attributes are
arranged into groups, determined using a combination of descriptions from
the codebooks and an unsupervised clustering method described fully later on.
I have further aggregated the groups into four broad concepts: attributes that
542 Case Studies II
Table 23.2 Attributes for Each Suspect in the NPIASS-I I Database
Almost Always Biographical Neutralization Captured/Defector

Available Record Record
Serial Number At-Large‡ Killed/Capt./Defect. Detention Facility
Job Birth Place Action Force
Bio Proc. Date ID Source Arrest Level
Echelon Bio Info Date Neut. Process. Date Arrest Serial
Sex Dossier Loc. Neut. Action Date Arrest Year
Black List Neut. Location
Party Member. Photo Sent. Process. Date
Area of Operation Prints Specific Target Sentence Date
Priority A/B* Arrest Order Operation Level Sentence Code
Record Updates Address IOCC Involvement Sentence Location
Confirmation
Age
Release Process. Date
Release Action Date
Release Code
Release Location
Arrest Forwarding
Forw. Process. Date
Forw. Action Date
Forw. Location
*Imputed from official position Greenbook.

‡ Mutually exclusive and so merged with Killed/Capt./Defector.
are typically always available, attributes typically only available for biographi-
cal records, attributes typically only available for neutralization records, and
attributes only available for neutralization records of suspects who defected or
were captured.
23.3.3. Grouping Attributes and Records

The codebooks provide descriptions of each attribute but they omit important
details such as how missing values are handled. Every observation has at least
some missing attributes, and nearly 50 percent of cells are empty. Some of the
patterns are self-explanatory, for example, there will be no information about sen-
tencing if the suspect is still at-large. In other cases, missingness is more subtle, as
in cases where information on the suspect’s age is sometimes missing for suspects
who were killed in the field without questioning. In all cases there appears to be a
combination of missing attributes at random and undocumented structure. This
ambiguity and apparent latent structure suggests applying a machine-learning
approach to learning how attributes are related to one another.
I frame this as a blockclustering problem where the task is to simultaneously
find r groups of attributes and k groups of observations that are similar in terms
of missingness. Let an I ´ Q binary matrix, X NA, represent the missingness for
each individual i and attribute q, shown on the left in Figure 23.1. The task is to
decompose this matrix into a version sorted by row and column into homogenous
blocks, X NA, shown on the right side of Figure 23.1, and a smaller r by k binary
matrix of row and column clusters.
I employ an unsupervised biclustering algorithm, the Bernoulli Latent Block
Model (Govaert and Nadif 2003).14 The model is fit with a wide range of possible row
and column cluster counts, and the final model is selected with the best fit according
to the integrated complete likelihood (ICL) (Biernacki, Celeux, and Govaert 2000).
The structure of missingness in the NPIASS-I I database is best explained by
11 clusters of attributes and 9 clusters of observations (ICL = −260339.9). For
substantive reasons discussed next, I further split off dossier-related attributes as
a separate cluster bringing the total to 12. These are the low-level groups of attri-
butes shown in Table 23.2.
Original Data Co-Clustered Data
Legend
0
1
Figure 23.1 Block clustering of missing values in NPIASS-I I into 11 groups of

attributes and 9 groups of observations. True values (white) indicate a missing value.
544 Case Studies II
In almost every case, the method has recovered known groups of variables as
detailed in the codebook. In a few cases it has correctly identified an attribute as
belonging to a different group despite a misleading original variable name. The six
record clusters recover the undocumented split between biographical records and
neutralization records, the existence of neutralizations with additional follow-up
information about sentencing and release, and the existence of a small number of
biographical records that were updated with neutralizations. Since these record
clusters might have further substantive implications, I include “record cluster” as
an additional attribute in the analysis below.
Why devote special care to analysis of missingness? In this instance, motivation
comes from a discovered detail with major substantive implications for the only
other recent study to use this database. Kalyvas and Kocher (2007) ask whether
the strength of the evidence against suspects is related to the likelihood that they
will be caught. They estimate the likelihood of being neutralized as a function of a
dossier attribute called “confirmed” which takes the value of true when a suspect is
identified by three different sources or one irrefutable source. Counterintuitively,
they find that confirmed suspects, with presumably more evidence against them,
were actually much less likely to be neutralized than unconfirmed suspects. If
true, they conclude that this would have a troubling implication that the guilty
can be much safer than the innocent from government targeting in an insurgency.
In fact, this counterintuitive result turns entirely on a technical error. Their
key independent variable “confirmed” along with a few other dossier-related attri-
butes were stored as a 0/1 bit flag in the original IBM360 system. When they were
converted to modern formats, 0 values were converted to “No” when they should
really have been converted to missing values. Confirmations may have played a
role in neutralization, but they were only recorded for suspects who were still at
large. The desired comparison is just not possible. Further, the database records
many other details about the government’s interest in and information about spe-
cific suspects that all suggest the opposite conclusion. Specifically targeted sus-
pects suffered more dangerous outcomes than low-level incidental members, a
point discussed in detail over the next sections.
From a data analysis perspective, this is a tricky mistake to catch, an undocu-
mented difference in record types plus an undocumented imputation of values.
However, it pops out in this analysis of missingness since there appear to exist
at least two different kinds of records mixed together and all of the other dossier
attributes tend to be missing for neutralization records. It also pops out in the
analysis of attribute importance and interaction that I turn to next.
23.3.4. Variable Selection

Each suspect in the database has potentially over forty known facts about them. If
one were forced to describe an observation to someone else, which fact would one
start with?15 What is the single-most important fact about a suspect? The second-
most important fact? The third-most important? And so on. Typically these deci-
sions are made on an ad hoc basis given the researcher’s theoretical interests.
Here, the focus is, in part, on learning the structure of the database, and so we
need a principled definition of what makes an attribute important and a method
for ranking attributes on that dimension.
I frame this as an unsupervised learning problem, where the task is to learn
rules and relationships between attributes that could be used to distinguish a real
observation from a synthetic, randomly shuffled version. The only way to tell a
real observation from a randomly generated one is to learn patterns of regularity
and structure among attributes. In this conception, an attribute is important if it
conveys a great deal of information about what other values a suspect’s attributes
will take. The most important fact is the one that provides the most information
for inferring other facts. The least important fact is the one that provides com-
pletely unique but orthogonal or potentially random information.16
The classifier I use for this task is an unsupervised random forest.17 Random
forests are an ensemble method that combines the predictions of many individual
base learners. The individual learners in this case are fully grown binary deci-
sion trees, each fit to a different random subset of attributes and random subset
of observations (Breiman 2001).18 In the supervised case, cut points for covari-
ates are selected to separate observations into increasingly homogeneous groups
on some outcome variable. In the unsupervised case, the random forest learns
to identify a genuine observation from a synthetic scrambled version (Shi and
Horvath 2006). This method works for both categorical and continuous variables
and is nonparametric, so there is no need for prior knowledge of an underlying
functional form.
The most useful information for this learning task is provided by strong and
regular relationships between variables, so each decision tree will tend to select
variables with multiple strong interactions earlier in the process, toward the root
of the tree. Therefore, I measure the importance of a variable as the average dis-
tance from the root node to its maximal subtree (the earliest point in the tree that
splits on the variable) (Ishwaran et al. 2010). The interaction of two variables is
captured by the depth of their second-order maximal subtree (the distance from
the root node of one variable’s maximal subtree within the maximal subtree of
the other), as trees tend to split on one variable and then soon split on a related
variable. I define a symmetric distance between two variables as the sum of their
second-order maximal subtree depths. This distance is small when both variables
tend to split close to the root, soon after one another, and large if either splits late
in the tree or far from the other.19 This approach provides two remarkable pieces
of summary information shown in Figure 23.2.
The first piece of summary information is a ranking of variables in terms of
how much information they convey about the entire dataset. The answer to the
Attribute Interaction Strength (Hierarchical Clustering)
Overall Attribute Importance (Rank)
Release Location (40)
Release Action Date (38)
Release Processing Date (39)
Arrest Serial Number (32)
Sentence Code (12)
Sentence Location (36)
Sentence Action Date (35)
Sentence Process. Date (37)
Arrest Year (33)
Age (15)
Record Update Count (14)
VCI Serial Number (10)
Echelon (5)
Party Membership (3)
Area of Operation (11)
ID Source (6)
Black List (9)
Job (13)
Action Force (7)
Row Cluster (4)
Neutralization Location (17)
Neutralization Action Date (18)
Neutralization Process Date (20)
Operation Level (19)
Detention Facility (8)
Kill/Capt./Rally/At Large (1)
IOC Involvement (22)
A or B Priority (16)
Sex (2)
Specific Target (23)
Arrest Level (31)
Forward Location (41)
Arrest Forward Code (42)
Release Code (44)
Bio Information Date (28)
Dossier Location (27)
Birth Location (30)
Bio Process Date (29)
Record Type (21)
Dossier Confirmation (24)
Dossier Address (25)
Dossier Arrest Order (26)
Dossier Photograph (34)
Dossier Fingerprints (43)
2 1 0
Merge Height
Figure 23.2 Clustering of attributes by strength of interaction with merges selected to

minimize Ward’s distance (dendrogram). Rank order importance of each attribute in
terms of average maximal subtree depth in an unsupervised learning task (unsupervised
random forest). Smaller rank means an attribute was selected sooner in the random
forest construction and is thus more informative overall.
question “Which fact should we start with?” is definitively the fate of the sus-
pect: still at-large, killed, captured, or defector. No other single attribute implies
as much about the remaining details as that one. The next most important is
the suspect’s gender, followed by whether they were a party member, the kind
of record as estimated by the blockclustering above, the suspect’s echelon, the
source of information used to ID the suspect, and so on.
The second is a pairwise distance between attributes in terms of how much
information their interaction conveys about the entire dataset. Hierarchically
clustering attributes on that distance reveals what appear to be two mostly uncon-
nected data-generating processes: one related to the creation of biographical
records and dossiers, and another related to the neutralization of suspects. The
method has, without prompting, correctly recovered the undocumented differ-
ence between neutralization record attributes and biographical record attributes.
Justifying earlier concerns, the dossier attribute “confirmed” is flagged as being
closely related to other administrative details of dossiers and not the core demo-
graphic attributes of suspects or the empirical process of targeting.20
The clustering also pinpoints the place to start the analysis: a core group of
twenty-one highly related and informative attributes relating to the neutralization
and demographics of suspects. They are flanked by tangential groups of attributes
relating to the sentencing of a suspect, the release of a suspect, and the details of
dossiers for biographical records. There may be interesting structure within these
other groups of attributes, but they are mostly orthogonal to the core outcomes of
interest and so can be safely set aside for future work.
Having selected a core group of attributes, the next question is whether they
can be further summarized by a simpler topology. The next two sections unpack
these attributes and tackle dimensionality reduction with respect to two themes,
namely, the kinds of victims and the kinds of government operations. For that
analysis, I weed the list further to just eleven nominal demographic and neutral-
ization attributes. I set aside dates and locations. I exclude five attributes about the
administrative aspects of the dataset. And I single out two attributes, with a large
number of categories, for detailed analysis: the job of the suspect and the govern-
ment actor responsible for the neutralization.
23.4. The Victims of Targeting

Who were the Phoenix Program’s victims? The program was charged with dis-
mantling nonmilitary rebel organizations in South Vietnam.21 A full breakdown
of suspect counts by organization, demographic attributes, and final status is
shown in Table 23.3.
Broadly, the political opposition to the Republic of Vietnam was organized
into three groups. Political authority, command, and resources flowed from
548 Case Studies II
Table 23.3 Cross-Tabulation of Demographic Properties against Outcome
Outcome
All At Large Killed Captured Defector

% % % %
All 73,699* 32 21 30 16
Organization PRP 54,977 31 19 34 15
NLF 7,147 33 13 28 26
Communist 11,315 39 33 13 15
Orgs
Sex Male 56,257 36 25 22 16
Female 16,910 20 6 57 17
Party Full Member 22,443 48 29 13 10
Membership 29,470 37 19 31 12
Unknown
Probationary 7,042 17 26 37 20
Member
Nonmember 14,743 6 10 51 33
Echelon Hamlet 9,858 25 24 24 26
Village 44,513 31 20 33 15
District 12,917 42 23 22 13
City/Prov/ 6,400 33 16 34 17
Reg/COSVN
List Most Wanted 24,285 65 21 9 4
List
Target List 15,687 40 18 30 12
Most Active 8,136 23 15 51 11
List
Unknown 25,588 0 24 43 33
A or B A Priority 43,645 42 25 18 15
B Priority 29,790 18 15 47 19
Age [0,25] 15,500 19 14 47 21
(25,35] 14,858 37 22 25 16
(35,55] 27,225 35 14 31 20
(55,100] 3,742 22 5 56 17
Note: * The number of records is 73,712 but 13 records had to be discarded from analysis, leav-
ing 73,699.
North Vietnam into South Vietnam through the communist political apparatus,
the People’s Revolutionary Party (PRP). Indigenous popular support and partici-
pation was organized into the subordinate National Liberation Front (NLF), also
called the Viet Cong. Together they constructed alternative administrative insti-
tutions referred to as Communist Authority Organizations, such as the People’s
Revolutionary Government (PRG), as well as a number of political organizations
designed to involve civilians outside of the Communist Party.
Together, and with overlapping and changing roles and capabilities, these
three organizations embodied foreign authority, popular participation, and polit-
ical institutions. Four-fi fths of neutralizations were against PRP positions, with
fewer directed toward more indigenous NLF and Communist Organization posi-
tions. This is consistent with both the priorities of the program and the timing in
the war; the post-Tet phase was more externally driven by North Vietnam.
Members of or supporters performing active roles in these organizations were
collectively known as the Viet Cong Infrastructure (VCI).22 VCI were grouped
into Class A VCI, which were full or probationary PRP members or in leader-
ship and command roles, while Class B VCI were trained but voluntary members.
Where appropriate, roles were replicated at multiple levels of governance called
“echelons,” including hamlet, village, district, province, city, capital, regional, and
national levels.
The broad pattern is one of a program that targeted large numbers of low-level
suspects, a portfolio of targets that was bottom-heavy. Half of neutralizations
were B-level voluntary/support positions, and a little more than half were previ-
ously unknown to security forces. About a fifth of neutralizations were full party
members, and another fifth of neutralizations were at district or higher echelons.
It is unclear, however, whether this is disproportionate to the number of actual
rebels holding the positions. If the program selected targets uniformly from the
rebel population, these rates are probably proportional to the share of those posi-
tions of all rebel members during this period of the war.
There is a strong relationship between the demographics of suspects and
methods of targeting. In brief, more important suspects were fewer in number
but more directly targeted, either by having a file while at-large, or by being killed
in an operation. Low-level suspects were less likely to have a file at-large, and
were much more likely to be swept up in an arrest or to walk in off the street as a
defection.
The cross-tabulation in Table 23.3 shows this in terms of a single comparison
between demographics and the suspect’s final status. Targeting was gendered,
with female suspects much less likely to be killed or to be targeted as at-large.
Known party members were more likely to be targeted at-large, or killed, while
suspects known not to be party members were much more likely to simply be
arrested or defect. The lower the suspect’s echelon, the more likely they were sim-
ply arrested or defected and the fewer targeted while at-large.
550 Case Studies II
The same is true for the level of prior suspicion against the suspect. More prior
suspicion is associated with more severe outcomes. Previously unknown suspects
tended to be arrested in the field or defectors who were not killed.23 Moving up the
ladder of suspicion to the most active list, the target list, and the most wanted list
increases the chances that the suspect was killed by an operation or targeted at-
large.24 The same pattern holds true for the A or B priority of the suspect’s position.
Visualizing the underlying pattern in just the bivariate case is already some-
what overwhelming. Extending the analysis to the multivariate case and devel-
oping a simplifying taxonomy is the task of dimensionality reduction that I turn
to next.
23.4.1. A Taxonomy of Suspects

Is there a simpler topology for understanding differences between victims? I frame
this as a dimensionality reduction problem where nominal values for each of the
categorical variables are mapped to common latent dimensions. The method I use
for this estimation is Multiple Correspondence Analysis (MCA). This method is
a multivariate technique analogous to Principal Components Analysis but for
unordered categorical data (Lê, Josse, and Husson 2008).
Let an I ´ Q matrix represent the values for each individual i and attribute q
with K q possible values for each attribute and K total possible values. This matrix
is then converted to an I ´ K disjunctive table (dummy variables for each level of
each variable). Rarely used categories disproportionately influence the construc-
tion of these dimensions, so I suppress both rare and missing values.25
I use a variant of the algorithm, called Specific Multiple Correspondence
Analysis, that correctly calculates partial distance between points when rare and
missing values are dropped (Roux and Rouanet 2009).26 It decomposes the dis-
junctive table into principal axis-representing latent dimensions, points for indi-
viduals in that reduced space, and points for each attribute value in the same space.
The result is a geometric interpretation of the originally categorical data where
suspects and attributes are all now projected into a smaller number of continuous
dimensions.
The core variation of the dataset is well summarized by a few latent dimen-
sions, with the first two principal axes accounting for 74 percent of total variation
(inertia).27 They are summarized in Table 23.4. The first dimension (56 percent)
reflects a clear demographic concept of the suspect’s importance to the targeting
program. At one extreme are unimportant, previously unknown, low-level vol-
unteers, often caught in large raids. At the other extreme are high-level, full party
members that are on the most wanted list, but usually remain at-large.
The second dimension (18 percent) reflects the method of neutralization
used. At one extreme are killings, sometimes targeting specific individuals,
Table 23.4 The First Two Dimensions of Demographic and Operations

Attributes
Dimension 1 (56%) Ctr. Coord. Dimension 2 (18%) Ctr. Coord.

(+) At-Large 13.7 1.18 Killed 14.4 1.21
Most Wanted List 10.5 1.03 Order of Battle for ID 13.3 1.45
Full Party Member 5.6 0.78 Specific Target 8.8 1.05
A Priority 4.0 0.47 Result of IOCC 7.7 0.71
Information
Subsector/District 6.9 0.57
Level Op.
Directed by IOCC 5.4 0.98
(–) Nonspecific Target 8.3 –0.79 Defector 10.9 –1.19
Captured 8.2 –0.95 Confession for ID 7.1 –0.98
Unknown (No List) 6.2 –0.77 No IOCC Involvement 3.0 –0.70
B Priority 5.8 – 0.69
Subsector/District 5.4 – 0.63
Level Op.
Result of IOCC 4.3 – 0.66
Information
Non–Party Member 4.0 –0.81
Agent/I nformer 4.0 –0.78
for ID
Female 3.6 –0.73
Age [0,25] 3.0 – 0.69
Notes: Estimated with Multiple Correspondence Analysis for all suspects. Contribution
and coordinates of specific values shown for attributes with above average contribution to each
dimension.
in operations directed by the local intelligence office. At the other extreme are
defections, which required no previous effort by an intelligence office, where the
identity was confirmed by the suspect’s own confession. In between lie arrests
that share aspects of both kinds of targeting.
This provides a clean language for describing civilian targeting in terms of
just two concepts. First, there are suspects the government wishes it could target
and those that it actually targets in practice. Second, of those it targets, there is a
spectrum ranging from suspects that tend to defect, suspects that tend to be
arrested or captured out in the field, and suspects that tend to be assassinated (or
552 Case Studies II
1.5
ID.Source: Phung Hoang Political Order of Battle
Status: Killed
Target: Specific
1 Shape
IOCC: Directed by DIOCC/PIOCC
Age
ID.Source: Captured Document Echelon
ID.Source
IOCC: Result of DIOCC/PIOCC Information
IOCC
0.5 Op.Level: Subsector/District List
Party: Probationary Member
ID.Source: Agent/Informer Op.Level
2. Dimension 'Severity of Outcome' (18%)
Status: Captured Party: Full Member Party

List: Target List Priority
Echelon: Hamlet List: Most Wanted List
Target: Non-specific Echelon: Village
List: Most Active List
Priority: A Sex
0 Sex: Male Status
Op.Level: Sector/Province Age: [25,35]
Age: [55,100] Target
Age: [0,25]
Age: [35,55]
Sex: Female Echelon: District Size
Party: Membership Unknown 10000
Priority: B
List: Unknown 20000

Status: At Large
−0.5 Party: Non-Member 30000
Op.Level: Other 40000
ID.Source: Other Source
Echelon: Province
50000
IOCC: No DIOCC/PIOCC Involvement
−1 ID.Source: Confession
Status: Defector
−1.0 −0.5 0.0 0.5 1.0
1. Dimension 'Priority of Suspect' (56%)
Figure 23.3 Attribute map (all suspects; n = 73,712). Attribute values for all suspects
projected into two dimensions with Multiple Correspondence Analysis.
who die in battle but the program claims credit). A map of each value in the two
dimensions is shown in Figure 23.3.
For studies that use counts of rebel or civilian deaths as a dependent variable,
this shows that those raw counts could be driven by changes in at least three dif-
ferent underlying dynamics: (1) the intensity of the war, growing or shrinking the
size of the target list or the number of operations looking for suspects not on a list;
(2) changes in effectiveness, neutralizing more or fewer known targets already on
the list; or (3) changes in tactics, using more killings than arrests or more defec-
tions preempting killings, and so on. Shifts along any of these dimensions could
produce changes in total body counts or the portfolio of observed violence (e.g.,
ratios of civilian to military deaths). There is currently little in the way of theoreti-
cal expectations for how interventions should interact with each of these dimen-
sions, much less how those interactions should aggregate into changes in total
observed levels of violence.
23.4.2. The Jobs Held by Suspects

As an external check of validity, I compare the estimated position of each suspect
along the dimensions of targeting to a description of the job they held. If the dimen-
sions are correct, and useful, then jobs with similar functions ought to be more
similar to each other in terms of targeting. I find that suspects with similar jobs, as
described by third-party sources, do in fact have similar demographic attributes
and similar targeting behavior by the government. If the underlying data were
faked or entered in error, they were at least doing it in a consistent and creative way.
Each suspect is tagged with one of 485 specific jobs, coded according to a stan-
dardized official schema called the Greenbook. Each job is nested within increas-
ingly large departments called elements, subsections, sections, and the three
main branches. I focus on the section level of aggregation. The location of each
section along the dimensions of targeting is estimated by including the section
attribute as a noncontributing covariate in the Multiple Correspondence Analysis
introduced earlier. A map of sections along the two dimensions of targeting is
shown in Figure 23.4.
0.4
People's Council Standing Committee
Guerilla Unit
Liberation Workers' Ass'n Military Affairs
Area Administrative Officials
0.2 Missing
Political Struggle
Patriotic Teachers' Ass'n
2. Dimension 'Severity of Outcome' (16%)
Action Arrow Team

Peoples Revolutionary Committee
Land Distribution
Security Political Officers

Cadre Affairs
Specialized
Finance-Economy
Organization Section
Special Action
Liberation Youth Ass'n Nflsv Central Committee
0.0
Rear Service Central Wounded and Dead Soldiers Ass'n
Liberation Women's Ass'n Nflsv Secretariat
Commo-Liaison
Social Welfare Relief Ass'n
Military Proselyting
Propaganda, Culture And Indoctrination Provisional Revolutionary Government
Investigation
Administration (Party Office)
−0.2
Civilian Proselyting
Liberation Farmers' Ass'n Patriotic Buddhist Ass'n
Civic Action
Front
−0.4 Liberation Labor Ass'n
−1 0 1
1. Dimension 'Priority of Suspect' (56%)
Figure 23.4 Attribute map (rebel organizational sections). Map of rebel sections
projected into the two dimensions of suspect importance and severity of outcome.
554 Case Studies II
Most of the variation across sections is on the first dimension of priority (hori-
zontal axis). At one extreme, on the far left, are low-level logistics-related sections
like the Commo-Liaison and Rear Service sections, where suspects were rarely tar-
geted at-large and mostly swept up as arrests. At the other extreme, on the right, are
high-level leadership positions like the NLF Central Committee and the Provisional
Revolutionary Government where almost every suspect was most wanted but still
at-large. Along the second dimension of severity of outcome (vertical axis), some sec-
tions were likely to be specifically targeted or killed, such as Guerilla Units, Military
Affairs Section, or Area Administrative Officials. At the other extreme, some groups
were much more likely to defect, such as the Medical Section, the Frontline Supply
Council, or the Western Highlands Autonomous People’s Movement.
Next I cluster the sections according to their distance along the targeting
dimensions.28 The 30 sections with 100 or more suspects are described in Table
23.5. They are arranged hierarchically using Ward’s method by their proximity in
biographical dimensions estimated with MCA above (Ward 1963). Each section
is provided with a brief description based on their functions as outlined by US
intelligence (Combined Intelligence Center, Vietnam 1969).
The clustering recovered groups of sections with similar functions, arranged
into roughly four themes. The fighting-t hemed cluster contains four large sections
including the Guerrilla Unit, Military Affairs, Cadre Affairs, and the Liberation
Farmers Association. All operated at the low village or hamlet level and were high
risk in terms of the chance of being killed.
A leadership-t hemed cluster includes seven small groups with positions that
operated at the village, district, or higher echelons, and were more likely to be
targeted while at-large or killed if neutralized. This includes the NLF Secretariat,
the NLF Central Committee, Area Administration Officials, and so on.
There are two village administration-themed clusters. The first leans toward
higher priority and more violent outcomes. It includes Political Officers, the
Security Section, the Party Office, the Culture and Indoctrination, and the Finance-
Economy Section. This captures the main justice, propaganda, and tax collection
infrastructure at the district and village levels. The second administration-themed
cluster includes village-level Womens’ and Youth associations, the Military and
Civilian Proselyting sections, and sections responsible for medical and food pro-
duction. This captures very local leadership and organization at the village level.
A final logistics-t hemed cluster includes three sections, Special Action,
Commo-Liaison, and Rear Service. Suspects in these sections were much more
likely to be captured, of very low priority in terms of not being party members, not
on a wanted list, B-level positions, and women.
This set of clusters provides a way to think about targeting the different com-
ponents of an insurgency: a fighting component, a leadership component, a day-
to-day administration component, and a logistical tail. Both the demographics
Table 23.5 Organizational Sections with Over 100 Suspects

Section N Description Org.
Liberation Farmers' Ass'n 2,230 Mass Org. Local NLF
Cadre Affairs 2,007 Intel, Proselytizing PRP
Guerilla Unit 4,593 Armed local forces Org
Military Affairs 1,671 Coordinate Guerrillas PRP
People's Council 447 Administration Org
Area Adm. Officials 1,406 Administration Org
Peoples Rev. Comm. 2,202 Administration PRP
Organization Section 188 Administration PRP
Specialized 126 NLF
Nflsv Secretariat 216 NLF Leadership NLF
Nflsv Central Committee 594 NLF Leadership NLF
Political Officers 337 PRP Leadership PRP
Liberation Workers' Ass'n 137 Mass Org, Urban NLF
Land Distribution 663 PRP
Security 6,990 Intel., Police, Justice PRP
Party Office 163 Administration PRP
Culture-Indoctrination 2,210 Propaganda PRP
Finance-Economy 7,845 Logistics, Taxes, Food PRP
Liberation Youth Ass'n 1,050 Mass Org, Youth NLF
Action Arrow Team 2,665 Mobile Security Org
Political Struggle 443
Liberation Women's Ass'n 2,654 Mass Org. Women. NLF
Military Proselyting 5,848 Turn GVN soldiers PRP
Medical 3,387 Public/Civil health PRP
Civilian Proselyting 1,538 Party Recruiting PRP
Frontline Supply Council 681 Logistics. PRP
Production 1,375 Rear Production PRP
Special Action 1,425 Sappers PRP
Commo-Liaison 9,425 Logistics, Routes PRP
Rear Service 3,037 Logistics, Military PRP
Note: Hierarchical clustering with Ward’s distance shown in dendrogram on the left.
of suspects and the targeting methods of the government vary systematically

across these different components. This is important for studies that use vio-
lence over time as a dependent variable. If any of these groups change in size
or in level of activity, it will change the aggregate body count in potentially
unpredictable ways.
556 Case Studies II
23.5. The Methods of Targeting

Killings and arrests required the government to launch an operation, and defec-
tions required a receiving government actor or office. Which government actors
conducted those operations and what methods did they use? When a suspect is
neutralized, the details of the circumstance or operation leading to their neu-
tralization were recorded. The details of each neutralization are cross-tabulated
against outcomes in Table 23.6.
The tabulations show a program with a large base of incidental arrests and kill-
ings in the course of regular operations, topped with a sizable number of direct
planned strikes against specific targets. A full third of killings and captures were
suspects targeted by an operation, often an ambush along a route or a raid. About
a third came from operations directed by an IOCC. As noted before, more than
half of killed or captured suspects were already previously on a blacklist.
The identity of a suspect had to be confirmed at the time of neutralization.
For previously unidentified suspects, the source of ID at the time of neutraliza-
tion may have been the source that led them to be a suspect in the first place.
For suspects already on a blacklist or listed in the Political Order of Battle, there
Table 23.6 Properties of Operations across Neutralization Outcomes
Outcome
All Killed Captured Defector

% % %
All 49,774 31 45 24
Source Agent/I nformer 16,296 37 56 7
Captured Document 4,428 44 45 11
Confession 11,633 7 42 52
Order of Battle 9,942 50 41 9
Other Source 7,249 22 31 47
Level DTA/Other/Region 3,835 30 43 27
Sector/Province 6,880 27 58 15
Subsector/District 33,296 37 50 13
IOCC No Involvement 9,637 29 46 25
Result of Info 24,127 38 52 9
Directed by 8,785 38 59 3
Specific Nonspecific 32,467 31 49 20
Specific 12,630 43 49 8
was some unobserved process by which evidence was collected, leading to the
initial suspicion. The majority was either identified by another civilian (an agent
or informer) or was said to have confessed. Others were confirmed by material
evidence like documents captured on their person or identifying them by name.
Some were confirmed against descriptions in the Political Order of Battle. About
12 percent were identified as “other,” explained in a written comment on the back
of the worksheet and not recorded here.
23.5.1. A Taxonomy of Targeting Operations

Neutralizations can also be well summarized by a simpler typology as shown in
Table 23.7. I fit the same Specific Multiple Correspondence Analysis model as
before to just the subset of observations resulting in arrest or killing. The first
principal axis accounts for 59 percent of the variation and, as before, reflects the
Table 23.7 First Two Dimensions of Demographic and Operations Related

Attributes
Dimension 1 (59%) Ctr. Coord. Dimension 2 (15%) Ctr. Coord.

(+) Most Wanted List 13.3 1.37 No IOCC 19.0 1.34
Involvement
Full Party Member 10.9 1.11 Other Source for ID 14.1 1.59
A Priority 7.8 0.66 Sector/Province 8.0 0.96
Level Op.
Killed 7.8 0.73 Other Level Op. 7.7 1.57
Order of Battle 6.9 0.90 Province Echelon 3.7 1.19
for ID
Specific Target 5.5 0.71 Unknown (No List) 3.2 0.36
(–) B Priority 7.9 –0.67 Result of IOCC 9.6 – 0.6

Information
Female 5.8 –0.76 Captured 5.3 –1.0
Document for ID
Captured 5.4 –0.51 Subsector/District 4.7 – 0.3
Level Op.
Age [0,25] 3.3 –0.61 Most Active List 3.0 – 0.6
Confession for ID 3.0 –0.76 Target List 2.9 – 0.5
Notes: Estimated with Multiple Correspondence Analysis for observations resulting in killing
or capture only. Contribution and coordinates of specific values shown for attributes with above
average contribution to each dimension.
558 Case Studies II
level of priority of the suspect. At one extreme are low-level incidental captures
and at the other killings against priority targets. To a lesser degree it also captures
the level of premeditation, as high-priority targets were more likely to be the spe-
cific target of an operation and the target of an operation planned and directed by
an IOCC.
The second principal axis accounts for 15 percent of the variation and reflects
the domain of the operation. On one end are operations that found VCI and
reported them retroactively to the intelligence infrastructure for documentation.
These operations typically had no IOCC involvement and were carried out by
more conventional forces at province or sector levels. At the other extreme are
operations carried out by Phoenix-related forces against targets known about
beforehand. These operations were at the subsector or district level, against
village-echelon-level targets that were on the most active or target black lists, and
benefited from information provided by the IOCC. The map of each attribute
value along these two dimensions is shown in Figure 23.5.
ID.Source: Other Source

1.5 Op.Level: Other
IOCC: No DIOCC/PIOCC Involvement
Echelon: Province
1 Size
Op.Level: Sector/Province 10000
20000
Shape
2. Dimension 'Domain' (15%)
Age
0.5 Party: Non-Member Echelon
Echelon: District
ID.Source
ID.Source: Confession
List: Unknown Target: Specific IOCC
IOCC: Directed by DIOCC/PIOCC Party: Full Member List
Status: Captured Op.Level
Priority: B List: Most Wanted List
Age: [55,100]
Age: [35,55] Echelon: Hamlet Party
0 ID.Source: Agent/Informer Sex: Male Priority: A
Sex: Female Priority
Age: [0,25] Target: Non-specific Age: [25,35]
Status: Killed
Sex
Party: Membership Unknown Status
Echelon: Village
Target
Op.Level: Subsector/District ID.Source: Phung Hoang Political Order of Battle
List: Target List
−0.5
IOCC: Result of DIOCC/PIOCC Information
List: Most Active List Party: Probationary Member
−1 ID.Source: Captured Document
−1 0 1
1. Dimension 'Priority of Suspect/Premeditation' (59%)
Figure 23.5 Attribute map (killed or captured; n = 37,653). Attribute values for killed
and captured only, projected into two dimensions with Multiple Correspondence
Analysis.
23.5.2. The Perpetrators of Targeting

There were sixteen different organizations reported as the government actor
in the neutralization of suspects. As an additional external check of validity,
Figure 23.6 shows the position of the actors along the two dimensions of priority
and domain. If the underlying data are accurate and well summarized by these
dimensions, then actors with similar functions ought to be similar to each other
in terms of victims and tactics.
A clustering of actors along the dimensions of operations is shown in
Table 23.8. On paper, the official Phoenix forces were the National Police, the
Provincial Reconnaissance Units, Rural Development Cadre, Civilian Irregular
Defense Group, and the Armed Propaganda Team (APT). In practice, the tent
poles of the Phoenix Program were the Popular Forces, Regional Forces, and
Special Police who made up two-t hirds of all killings and captures.
The analysis reveals four small clusters. The first cluster includes urban and
suburban police organizations that were much more likely to arrest than kill. This
is likely both because they were in areas of greater government control, where
Chieu Hoi Cadre
U.S. Forces
0.5
2. Dimension 'Domain' (15%)
Special Police (SP)
ARVN Main Forces FWMAF Other than U.S.

Military Security Service (MSS)
National Police (NP)

Civilian Irregular Defense Group (CIDG)
Peoples Self Defense Force (PSDF)
0.0
Provincial Reconnaissance Unit (PRU)
National Police Field Force (NPFF)
Regional Forces (RF)

Popular Forces (PF)
Armed Propaganda Team (APT)

−0.5
Rural Development Cadre (RD)
−0.5 0.0 0.5

1. Dimension 'Priority of Suspect/Premeditation' (59%)
Figure 23.6 Attribute map (government actors). Government actors responsible for
killings and captures projected into the two dimensions of importance and domain.
560 Case Studies II
Table 23.8 Government Actors Responsible for Neutralizations

Actor N Description
Special Police (SP) 5,375 Urban Police
National Police (NP) 3,604 Urban/Suburban Police
Military Security Service (MSS) 561 Counter-intelligence
Provincial Reconnaissance Unit (PRU) 3,190 Mobile Special Forces
National Police Field Force (NPFF) 1,065 Mobile Rural Policing
Army of the Republic of Viet Nam (ARVN) 1,919 Regular GVN Military
Armed Propaganda Team (APT) 310 Mobile Cultural Team
Regional Forces (RF) 14,356 District/Province Paramilitary
Popular Forces (PF) 5,195 Hamlet/Village Paramilitary
Civilian Irregular Defense Group (CIDG) 146 Irregular Militia
Peoples Self Defense Force (PSDF) 111 Irregular Militia
Free World Military Assistance Force 308 Allied, South Korea
U.S. Forces 899 Regular U.S. Military
Other 539
Notes: Grouped according to similarity on operation properties estimated with Multiple

Correspondence Analysis. Dendrogram shows hierarchical clustering using Ward’s method.
contestation was less violent overall, and because they were more likely to docu-
ment their arrests than less institutionalized forces.
The second cluster includes mobile forces that operated in areas of weaker
government control but still had close ties to the Phoenix Program in terms of
intelligence sharing and reporting. Provincial Reconnaissance Units (PRU), for
example, were specially designed units for the Phoenix Program who had the
highest rate of neutralizations per fighter (Thayer 1985, 210).
The third cluster contains paramilitary forces. Regional Forces were respon-
sible for routes and intersections, while Popular Forces were responsible for
village and hamlet defense. Their neutralizations were the most violent and
numerous. This is because of where they operated (in more contested areas),
their high numbers (having more manpower in more places than police forces),
and their tactics (equipped and trained for defense and attack rather than regu-
lar policing).
A final cluster represents conventional forces: the US and Free World Military
Assistance forces (primarily South Korean). These forces reported few suspects
to the national database, likely because of parallel reporting mechanisms and also
because they were engaged in larger, more conventional fighting.
These clusters provide a way to think about targeting as a function of differ-
ent kinds of government forces: regular police, expeditionary forces, paramili-
tary forces, and conventional forces. Each employs different tactics, in a different
environment, with a different portfolio of victims, and different incentives and
capabilities to report back statistics. This is particularly relevant to the study of
aggregate levels of violence. Over the course of a conflict, the size and level of
activity of these four groups will change. Forces are raised, units move around
the country, and police and paramilitary forces are extended to newly secured
communities. Each of these will affect the amount of violence committed and the
number of casualties reported in a given area over a given period of time.
23.6. Conclusion
The Phoenix Program provides an unusually clear view of a large wartime govern-
ment targeting effort. In the aggregate, it provides an example of a typically mixed
targeting program. Most processed neutralizations were of low-priority targets,
while occasionally the program had the intelligence, or good luck, to launch oper-
ations against high-profile targets. This pattern is a result of the fundamental fea-
ture of civil war, an inability to easily separate rebels from neutral civilians.
That said, based on the portion of targeting recorded in the national database,
the government did track and target a large number of suspects with verifiable
links. In the same way that policing is primarily about deterring illegal behavior
through small risks of punishment, the Phoenix Program offered a credible risk
to rebels who might have otherwise operated openly or civilians who were on the
fence about joining.
There is a strong connection between a suspect’s demographics, their position
in rebel organizations, the kind of government actor that would target them, the
methods that would be used, and their ultimate fate. A combination of dimen-
sionality reduction and clustering suggests a few simplifying ways to describe that
connection. Suspects vary in priority to the government in terms of who it wishes
it could target and who it targets in practice. The outcomes for suspects vary in
severity ranging from voluntary defection to death in the field. Operations vary
in the priority of the suspect, arresting large numbers of unimportant suspects in
sweeps, and launching premeditated operations to kill more important suspects.
Operations also vary across domains. Some operations are carried out far away
from intelligence and police infrastructure and are underreported in official sta-
tistics. Other operations are carried out in clear view, often using previous intel-
ligence and regularly reporting back for inclusion in official statistics.
The data also reveal clear organizational differences between different gov-
ernment actors and different rebel sections. Rebel sections have functions that
fall into groups related to fighting, high-level leadership, low-level administra-
tion, and logistical support. Government actors fall into groups of regular police,
expeditionary forces, paramilitary forces, and conventional forces. Each kind of
organizational subdivision has a distinct signature in terms of types of civilians
involved and the types of targeting methods employed.
One motivation for moving toward bigger (wider) data in conflict studies is
that they reveal these underlying dimensions, organizational types, and pro-
cesses, which typically get relegated to an error term. These results should be a
562 Case Studies II
cautionary tale for analysis based on raw event counts, often drawn from news-
papers or similarly shallow reporting. Targeting in the Vietnam War was very
high-d imensional. An analyst could reach dramatically different conclusions
about outcomes by truncating the sample to just killings, by omitting information
about the suspect’s position or the government actor committing the violence, or
by missing important details about how the institution created records and aggre-
gated them into a final dataset.
With data this detailed, the analysis provided here is just the tip of the iceberg.
I have shown ways to identify and explore the main sources of variation in a data-
base, but there are many other places in the data to look for interesting structure.
There are three that come immediately to mind. The first is exploring the spa-
tial and temporal structure of data. The Vietnam War varied from province to
province and often from village to village, which should have clear implications
for how civilians were treated. The second is how neutralizations were related to
one another. I have treated neutralizations as isolated events, but in reality they
were often part of larger operations. There is enough detail on locations and tim-
ing to aggregate individual observations into a larger event-level analysis. Finally,
the structure of rebel organizations is an entire field of study on its own, and the
detailed data on rebel jobs, locations, and demographics can provide a remarkable
map of Viet Cong and North Vietnamese organization across South Vietnam.
Acknowledgments
I thank David Madden, Josh Martin, Walter Fick, and Roxanna Ramzipoor
for research assistance, and the United States National Archives staff, particu-
larly Richard Boylan and Lynn Goodsell, for generously sharing their time and
expertise. I am grateful for comments from Erik Gartzke, Joanne Gowa, Kristen
Harkness, Stathis Kalyvas, Chris Kennedy, Matthew Kocher, Alex Lanoszka,
John Lindsay, David Meyer, Kris Ramsay, Tom Scherer, Jacob Shapiro, members
of the Empirical Studies of Conflict Group, the Yale Program on Order, Conflict,
and Violence, the UCSD Cross Domain Deterrence Group, and two anony-
mous reviewers. This research was supported, in part, by the US Department of
Defense’s Minerva Research Initiative through the Air Force Office of Scientific
Research, grant #FA9550-09-1- 0314.
Notes
1. Two prime examples include the East German Stasi files, which came into public stew-
ardship; and the Guatemalan National Police Archive, which is now in the charge of gov-
ernment and international humanitarian agencies as part of a truth and reconciliation
effort (Aguirre, Doyle, and Hernández-Salazar 2013). The records are comprehensive but
unstructured and will require a major effort to analyze once all of the raw documents are
digitized (Price et al. 2009).
2. For some of the issues and methodology of working with retrospective sources, see Price
and Ball (2014) and Seybolt, Aronson, and Fischhoff (2013).
3. With few exceptions, the database and most declassified intelligence products from the
Vietnam War have remained unused. This is partially because working with found data is
technically challenging, requiring extensive cleaning and documentation—a nd partially
because the tools for such data are only now gaining popularity in the social sciences.
4. The program began out of a regional coordination effort called Intelligence Coordination
and Exploitation for the attack on the VCI (ICEX) in July 1967. By 1968, Phung Hoang
Committees were established in 44 provinces and 228 districts.
5. Example members of a typical district IOCC team included Village Chiefs, Deputy Village
Chiefs for Security, Village Military Affairs Commissioners, Village National Police
Chiefs, Popular Forces Platoon Leaders, Hamlet Chiefs, and more.
6. Military Assistance Advisory Group. Vietnam Lessons Learned No. 80: US Combat
Forces in Support of Pacification, 29 June 1970. Saigon, Vietnam: Headquarters, US Army
Section, Military Assistance Advisory Group, 1970-0 6-29. http://cgsc.contentdm.oclc.
org/u?/p 4013coll11,1524.
7. “Phung Hoang Review,” December 1970, 11, from VIETCONG INFRASTRUCTURE
NEUTRALIZATION SYSTEMS (VCINS), n.d., Folder 065, US Marine Corps History
Division Vietnam War Documents Collection, The Vietnam Center and Archive, Texas
Tech University. Accessed April 17, 2015, http://w ww.vietnam.ttu.edu/v irtualarchive/
items.php?item=1201065056.
8. “Org and Mission,” April 1969, Phung Hoang Directorate, Records of the Office of Civil
Operations for Rural Development Support (CORDS), General Records, 1967–1971;
Record Group 472.3.10. National Archives at College Park, College Park, MD. ARC
Identifier: 4495500.
9. Reference Copy of Technical Documentation for Accessioned Electronic Records,
National Police Infrastructure Analysis Subsystem (NPIASS) I and II Master Files, Record
Group 472 Records of the US Forces in Southeast Asia. Electronic Records Division, US
National Archives and Records Administration, College Park, MD.
10. United States Military Assistance Command Vietnam/ Civil Operations Rural
Development Support (MACORDS) National Police Infrastructure Analysis Subsystem
II (NPIASS II), 1971–1973. File Number 3-3 49-79-9 92-D. Created by the Military
Assistance Command/Civil Operations and Rural Development Support-R esearch and
Analysis (MACORDS-RA). US Military Assistance Command/Civil Operations Rural
Development Support.
11. The total number of attributes is higher if multipart attributes are disaggregated or if low-
to no-variance attributes are included.
12. “VCI Neutralization and Identification Information System (VCINIIS) Reporting
and Coordination Procedures.” Folder: “1603- 03A Operational Aids, 1969,” ARC
Identifier: 5958372, Administrative and Operational Records, compiled 05/ 1967–
1970, documenting the period 1966–1970, HMS Entry Number(s): A1 724, Record
Group 472: Records of the US Forces in Southeast Asia, 1950–1976. National Archives at
College Park, College Park, MD.
13. “VCI of the Soc Trang Province Party Committee,” March 1971, Folder 11, Box 09,
Douglas Pike Collection: Unit 05—National Liberation Front, The Vietnam Center and
Archive, Texas Tech University. Accessed April 9, 2015. http://w ww.vietnam.ttu.edu/v ir-
tualarchive/items.php?item=2310911004.
14. Implemented in the R package Blockcluster (Bhatia, Iovleff, and Govaert 2014).
15. Put another way, suspects are situated in some high-d imensional space where there is more
underlying structure than we could ever hope to completely document. What structure
should we prioritize as the most dominant or interesting in the data?
564 Case Studies II
16. Note that this is a reversal of the typical variable selection process, where the goal is to bet-
ter explain some outcome by removing redundant information to produce a smaller num-
ber of uncorrelated explanatory variables. In this multivariate setting, there is no single
outcome and the redundancies are the details of interest.
17. Implemented in the R package randomForestSRC (Ishwaran and Kogalur 2014).
18. I employ 1,000 trees, trying seven variables at each split, minimum of one unique case at
each split, and fully grown trees with no stopping criteria. Splits with missing values are
first determined using nonmissing in-bag observations, and then observations with that
attribute missing are randomly assigned to a child node.
19. Summing the second-order maximal subtree depths is a stronger test of interaction and is
a novel innovation so far as the author is aware.
20. This is all the more amazing because the variable is incorrectly imputed with values for the
majority of rows in the dataset. The method has correctly identified the subset of rows for
which the variable takes on meaningful values and has grouped it with related variables
accordingly.
21. With captured documents and defector reports, GVN and US intelligence analysts
mapped those organizations in great detail (Conley 1967, 165).
22. Military personnel serving in organizational roles, e.g., on Military Affairs Committee,
could qualify as VCI.
23. By definition, a previously unknown suspect (not on a blacklist) did not have a biographi-
cal record (was not targeted at-large).
24. Note that the data speak to the probability of being under suspicion given already being
targeted. Estimating changes in the risk of being targeted as a function of suspicion would
require additional information about the population of rebels overall.
25. This is another motivation for carefully studying missingness in the database. The main
source of variation in the database is technical, the difference between different kinds of
records. I manually suppress missing values and purely administrative variables so that the
estimated components reflect only the substantive empirical variation between attributes.
26. Implemented in the R package soc.ca.
27. In total fourteen dimensions account for 100 percent of variation.
28. I calculate Euclidean distance on the first three dimensions, which account for over 80 per-
cent of the variation.
References
Aguirre, C., K. Doyle, and D. Hernández-Salazar. 2013. From Silence to Memory: Revelations of the
AHPN. Guatemala, Policía Nacional, Archivo Histórico and Eugene: University of Oregon
Libraries.
Ball, P., E. Tabeau, and P. Verwimp. 2007. “The Bosnian Book of Dead: Assessment of the
Database (Full Report).” HiCN Research Design Notes 5. Households in Conflict
Network. http://w ww.hicn.org/wordpress/w p-content/uploads/2 012/0 7/rdn5.pdf
[accessed December 30, 2015].
Berman, E., J. Shapiro, and J. Felter. 2011. “Can Hearts and Minds Be Bought? The Economics of
Counterinsurgency in Iraq.” Journal of Political Economy 119, no. 4: 766–819.
Bhatia, P., S. Iovleff, and G. Govaert. 2014. “Blockcluster: An R Package for Model Based
Co-Clustering.” Journal of Statistical Software. https://hal.inria.fr/hal-01093554/fi le/
BlockCluster.pdf [accessed December 30, 2015].
Biddle, S., J. Friedman, and J. Shapiro. 2012. “Testing the Surge: Why Did Violence Decline in
Iraq in 2007?” International Security 37, no. 1: 7–4 0.
Biernacki, C., G. Celeux, and G. Govaert. 2000. “Assessing a Mixture Model for Clustering with
the Integrated Completed Likelihood.” IEEE Transactions on Pattern Analysis and Machine
Intelligence 22, no. 7) 719–25.
Breiman, L. 2001. “Random Forests.” Machine Learning 45, no. 1: 5–32.

Colby, W., and J. McCargar. 1989. Lost Victory: A Firsthand Account of America’s Sixteen-Year
Involvement in Vietnam. Chicago: Contemporary Books.
Comber, L. 2008. Malaya’s Secret Police 1945–60: The Role of the Special Branch in the Malayan
Emergency. Singapore: Institute of Southeast Asian Studies.
Combined Intelligence Center, Vietnam. 1969. “VCI Functional Element Description.”
February. Microfilmed from the holdings of the Library of the U.S. Army Military History
Institute Carlisle Barracks, Pennsylvania. Bethesda, MD: University Publications of
America.
Conley, M. 1967. Communist Insurgent Infrastructure in South Vietnam. Washington, DC: Center
for Research in Social Systems, American University.
Gelman, A., J. Fagan, and A. Kiss. 2007. “An Analysis of the New York City Police Department’s
‘Stop-a nd-Frisk’ Policy in the Context of Claims of Racial Bias.” Journal of the American
Statistical Association 102, no. 479: 813–23.
Govaert, G., and M. Nadif. 2003. “Clustering with Block Mixture Models.” Pattern Recognition
36, no. 2: 463–73.
Ishwaran, H., and U. Kogalur. 2014. Random Forests for Survival, Regression and
Classification (RF-SRC), R package version 1.6. http://CRAN. R-project. org/package=
randomForestSRC.
Ishwaran, H., U. Kogalur, E. Gorodeski, A. Minn, and M. Lauer. 2010. “High-Dimensional
Variable Selection for Survival Data.” Journal of the American Statistical Association 105, no.
489: 205–17.
Kalyvas, S., and M. Kocher. 2007. “How ‘Free’ Is Free Riding in Civil Wars? Violence, Insurgency,
and the Collective Action Problem.” World Politics 59, no. 2: 177–216.
Lê, S., J. Josse, and F. Husson. 2008. “FactoMineR: An R Package for Multivariate Analysis.”
Journal of Statistical Software 25, no. 1: 1–18.
Lyall, J. 2014. “Bombing to Lose? Airpower and the Dynamics of Violence in Counterinsurgency
Wars.” SSRN Scholarly Paper ID 2422170. Social Science Research Network, Rochester,
NY (August).
Moyar, M., and H. Summers. 1997. Phoenix and the Birds of Prey: The CIA’s Secret Campaign to
Destroy the Viet Cong. Annapolis, MD: Naval Institute Press.
Natapoff, A. 2009. Snitching: Criminal Informants and the Erosion of American Justice.
New York: New York University Press.
Price, M., and P. Ball. 2014. “Big Data, Selection Bias, and the Statistical Patterns of Mortality in
Conflict.” SAIS Review of International Affairs 34, no. 1: 9–2 0.
Price, M., T. Guberek, D. Guzmán, P. Zador, and G. Shapiro. 2009. “A Statistical Analysis of
the Guatemalan National Police Archive: Searching for Documentation of Human Rights
Abuses.” JSM Proceedings, Section on Survey Research Methods. American Statistical
Association. https://h rdag.org/w p-content/uploads/2 013/02/JSM-GT-estimates.pdf.
Roux, B., and H. Rouanet. 2009. Multiple Correspondence Analysis. Thousand Oaks, CA: Sage.
Seybolt, T., J. Aronson, and B. Fischhoff. 2013. Counting Civilian Casualties: An Introduction to
Recording and Estimating Nonmilitary Deaths in Conflict. Oxford: Oxford University Press.
Shi, T., and S. Horvath. 2006. “Unsupervised Learning with Random Forest Predictors.” Journal
of Computational and Graphical Statistics 15, no. 1: 118–38.
Silva, R., J. Marwaha, and J. Klingner. 2009. Violent Deaths and Enforced Disappearances during
the Counterinsurgency in Punjab, India: A Preliminary Quantitative Analysis. A Joint Report by
Benetech’s Human Rights Data Analysis Group and Ensaaf, Inc.
Thayer, T. 1985. War without Fronts: The American Experience in Vietnam. Boulder,
CO: Westview Press.
Ward, J. 1963. “Hierarchical Grouping to Optimize an Objective Function.” Journal of the
American Statistical Association 58, no. 301: 236–4 4.
PA R T F I V E
TOWARD PREDICTION
AND PREVENTION
24
Economic Risk Factors and Predictive

Modeling of Genocides and Other
Mass Atrocities
C h a r l e s R . Bu tc h e r a n d Be n ja m i n E . Gol dsm i t h
24.1. Introduction
The ability to predict genocides and other mass atrocities is an obvious public
good. Knowing where and when the next atrocity onset is likely to occur can
improve the efficacy of short-and long-term prevention efforts. Even the public
dissemination of lists of at-r isk countries can potentially reduce the chances of
atrocity onsets and their severity, if this information also increases the chances
that potential perpetrators are “named and shamed” (DeMeritt 2012; Krain
2012). A number of predictive models have been developed, often with the sup-
port of national governments.1 With publicly available data and relatively well-
known statistical techniques, these models can predict atrocity onset generally,
and genocide and politicide onset specifically, with a degree of accuracy that com-
pares favorably with, for example, predictions of civil wars (Ward, Greenhill, and
Bakke 2010).
In this chapter we survey the role that economic factors have played in pre-
dicting genocide onsets, although some attention is given to other mass atrocities
as well. Genocide is a contested and normatively burdened concept, which sig-
nificantly overlaps with other terms such as politicide, mass killing, mass atrocity,
and crimes against humanity. Abstracting the United Nations (UN) Genocide
Convention’s definition, Strauss (2012) defines genocide as intentional, group-
selective, and group-destructive violence. Harff and Gurr (1988), and others, dis-
tinguish between genocide and politicide, the latter targeting a politically defined
group (a group left out of the UN’s definition). We direct readers to chapters 1
through 3 of this volume for a detailed discussion of definitional issues, history,
and relevant data sources. For our purposes, we focus mostly on genocide and
569
570 Towa r d Pr ediction a nd Pr ev ention
politicide—especially in their mass killing forms—but use all these terms some-
what interchangeably, often employing the word “genocide” as a stand-in for all
these types of crimes.
What role have economic factors played in these predictions? Can indica-
tors of wealth and poverty, for example, help determine which states are at risk
of genocide? Does the economy give off signals of impending genocide? While
case study research points to economic conditions as a driver of genocide (Strauss
2007, 482), we find that economic variables play an ambivalent role in forecasting.
The utility of economic factors varies and depends on the specific definitions of
genocide used, the spatial and temporal domain over which forecasts are made,
and the modeling strategies employed. No single economic variable emerges as a
consistent and robust predictor of genocide onset, although trade as a percentage
of gross domestic product (GDP) and measures of income come closest to playing
this role. These mixed findings in the predictive modeling of genocide reflect, to an
extent, mixed findings in studies of correlational and causal modeling of genocide.
For reasons that will become clear shortly—and without implying that there is no
analysis in prediction or forecasting—we refer to the latter two as analytical studies.
In section 24.2, we discuss similarities and differences between predictive
and analytical studies. The latter are examined in section 24.3, with emphasis on
the role of economic risk factors that may contribute to genocide. The former are
examined in section 24.4, focusing on the role of economic variables in genocide
forecasting. Given the limited usefulness, to date, of economic variables in suc-
cessfully predicting genocide, section 24.5 explores how prediction studies may
make better use of economic variables in the future, hoping to improve prediction
accuracy in years to come. In particular, we focus on time-sensitive indicators of
economic change such as rapid closing-off of the economy from foreign interac-
tions, conditional indicators such as the interaction of economic inequality with
new data on politically relevant ethnic groups, better specification of functional
forms, the role of investment risk as a signal of mass killing onset, and the pros-
pects of forecasting mass atrocities committed by rebel groups, in addition to gov-
ernments. Section 24.6 concludes.
24.2. Differences Between Predictive and Analytical Models

Most quantitative studies of genocides are designed to test empirical implications
of causal arguments. This is often done by generating a theory of why genocide
occurs, developing testable hypotheses that derive from that theory, and finding
empirical measures that are then included in statistical models. The search then
begins for additional, potentially confounding, variables. Those are variables one
would expect to be causally related to both the dependent variable—genocide—
and a particular explanatory variable of interest. Of course, relevant measures
Economic R isk Factors and Predictive Modeling of Genocides 571
must be found as well to enable the researcher to statistically control for, or

exclude, the influence that confounding variables may have on the hypothesized
causal relationship under study.
Such statistical models are then interpreted in terms of statistical significance
(through interpretation of t-scores or p-values, for example), substantive signifi-
cance, and measures of goodness-of-fit (see Valentino, Huth, and Balch-Lindsay
2004; Easterly, Gatti, and Kurlat 2006; Eck and Hultman 2007).2 Together, these
statistics provide an estimate of how likely it is that there is no causal relation-
ship between any one of the explanatory variables and the dependent variable;
or, if there does exist a likely causal relationship, how much a unit change in the
explanatory variable increases or decreases the probability of genocide occurring,
and how much worse (or better) a particular statistical model fits the actual data
when potential explanatory variables are added or removed from it.
Although not entirely separate enterprises, the goals and processes of hypoth-
esis testing in causal models are, in practice, distinct from those of forecasting
models. Both are elements of scientific advancement, but hypothesis testing has
tended to be preferred over prediction in the social and political sciences (Schrodt
2014). If causal models and datasets were flawless, the two enterprises would con-
verge, and a causal model that perfectly explains the dependent variable or out-
come variable (in our case, genocide) would then also provide the best forecasting
tool. Given the complexity of the topic and uncertainty surrounding theory, defi-
nitions, and the uneven quality of data, it is to be expected that models designed
with hypothesis testing in mind, and assessed mainly in terms of the statisti-
cal significance of key explanatory variables of interest, will not always provide
the best forecasting tools (Ward et al. 2010). Thus, while the strength of causal
models can be assessed by putting them to a forecast performance test—g iven
the assumption that the correct inclusion of a causal variable in a model should
improve its predictive accuracy—forecasting metrics are probably more appro-
priate for assessing overall model fit to the data and, in particular, for develop-
ing models that will forecast events of interest outside of the data sample used to
construct them in the first place. In economics, where forecasting applications
are more common and have a longer history than in other social science fields,
it is typically assumed that for such models “the primary focus is on forecasting
future values of a time series process and not necessarily on estimating causal or
structural economic models” (Wooldridge 2009, 645).
Statistical significance does not mean, it must be noted, that an explanatory
variable efficiently discerns between cases that are likely to see genocide and those
that are not. Conversely, a variable may have a high correlation with the probabil-
ity of an outcome occurring, but this does not mean that the variable, as specified,
is optimal for sorting between high-and low-r isk genocide cases, especially when
considered in conjunction with other variables in a model. Perhaps most radically,
from the perspective of standard social science practice, explanatory variables
included in theory-based causal models might entirely neglect the potential utility
of correlated but spurious or theoretically inconsistent variables even if they are
powerful and more efficient predictors of genocide than are the causal variables.
Inferences regarding causality—whether changes in values of an explanatory
variable can be said to cause changes in value in the dependent genocide variable—
tend to rely on a common statistical threshold, such as a p-value of 0.05, which is
widely acknowledged as an arbitrary standard. While this standard may (or may
not) serve as a useful rule of thumb for assessing whether a hypothesized cause
of genocide thus has garnered empirical support, p-values are not a metric of the
predictive power of any given variable within any given model. In fact, variables
that turn out to be most useful for genocide forecasting may be neither intuitive,
nor causal, nor all that easy to identify with non-forecasting models in the first
place. In a word, statistical significance (in hypothesis testing) and substantive
significance (in forecasting) are different things.
An additional issue is that, statistically speaking, causal models actually are
sets of correlations. When, in the natural sciences, correlations reliably recur over
repeated controlled experiments and when certain statistical metrics are met, the
coefficient associated with a specific explanatory variable, X, then is interpreted as
“a one-unit changing of X causes so-and-so many units of change in Y,” where Y is
the outcome variable of interest. Exceptions and methodological advances not-
withstanding, in much of the social sciences, it is usually far harder to make such
statements regarding causality with dead-certain confidence. Thus, correlates of
war, correlates of peace, or correlates of genocide are possibly the preferable terms (see
chapter 10 in this volume). Since our interest in this chapter is in regard to fore-
casting, we lump both correlation and causation under the joint rubric of “analyti-
cal studies,” and we refer to “causal” only when the context requires it.
Forecasting performance is typically measured by a different set of metrics.
Simply put, the forecaster wants to know how well predictions made match the
actual data for the observed outcome variable—genocide, in our case—and more
so for data that the predictive model has not previously encountered (either by
applying the model to a new case or by applying it to future time periods). This
stark utilitarian criterion can be distinguished from approaches that seek to esti-
mate the probability of an event’s occurrence in the future. Nonprobabilistic yes/
no predictions can be enhanced and optimized based on selecting a threshold
probability above which a future event will be expected to occur, even if the same
model’s forecasted probabilities do not on average correlate with a greater likeli-
hood of the event (Ratcliff 2013).
The process of generating forecasting models is not the same as that for statis-
tical inference in analytical studies. Typically, forecasters are less interested in
the theoretical relationship between variables (although this will often drive what
variables they initially expect will be good at forecasting) than in their ability
to distinguish events from nonevents. An analogy comes from medicine. While
chest pain is correlated with heart attacks, it is a symptom of them, not a cause.
Chest pain is a good variable for forecasting heart attacks, but it does not tell us
(a lot) about why they happen. Of course, causal variables should also help us
predict events or values on the units we are interested in. Many, but not all, of
the variables that genocide scholars use for forecasting thus are derived from the
theoretical literature. But if forecasting is the goal, it cannot be ignored that there
may be noncausal (or else, poorly understood) variables that are very good at pre-
dicting the likelihood of genocide. One example from our own research is the
presence of peacekeeping forces, which we find is associated with increases in the
chances of genocide occurring. Clearly, we do not believe that peacekeepers cause
genocide. The presence of peacekeepers, however, may provide a signal that a situ-
ation of internal instability is very serious, and we would expect these situations
to be at a higher risk of genocide. 3
The ability of a model to make correct predictions of discrete events such as
genocides can be expressed in a contingency table (or confusion matrix). For
binary (yes/no) outcomes, this is a two-column-by-two-row (2 x 2) table that
records actual occurrences and nonoccurrences of a type of event, cross-tabulated
with the predicted occurrences and nonoccurrences. It yields two types of cor-
rect predictions (true positives and true negatives) and two types of incorrect
ones (false positives and false negatives). There are many ways to summarize and
compare forecasting performance based on these four categories, ranging from
simple indicators like percentages correctly predicted given a certain probability
(e.g., 0.5) or rank (e.g., top-ten) threshold, to more complex ones including Brier
scores or F-scores. The true positive rate is often termed sensitivity or recall,4 and
the true negative rate specificity. False positives are called fall-out and are equal to
(1—specificity). The percentage of all correct predictions (positives and negatives)
is termed accuracy. F-scores are readily adapted to give greater weight to true posi-
tive or true negative outcomes depending on the researcher’s needs, while Brier
scores focus on the calibration of the probabilities across all predicted outcomes
(Mason 2004; Sokolova, Japkowicz, and Szpakowicz 2006; Ratcliff 2013). 5
Perhaps the most commonly used indicator of forecasting accuracy is the
Receiver Operating Characteristic (ROC) graph and the Area Under the ROC
Curve (AUC) statistic. The ROC analysis plots the rate of true positives on the
vertical axis against the rate of false positives on the horizontal axis (that is, sensi-
tivity against fall-out, or 1—specificity) (Fawcett 2006). The closer the ROC curve
lies to the upper-left-hand corner of the graph, the better the model is at classify-
ing true positives to false positives. The ROC curve is related to a summary sta-
tistic, the AUC value, which ranges from 0 to 1, with 1 being perfect p rediction.
Figure 24.1 provides an example from the genocide forecasts of Goldsmith et al.
(2013).
One advantage of ROC analysis is that the researcher does not have to specify
a cutoff point for when an event is “likely” or “predicted.” Initially, it might seem
1.00
0.75
True Positive Rate
0.50
0.25
0.00
0.00 0.25 0.50 0.75 1.00
False Positive Rate
Area under ROC curve = 0.8878
Figure 24.1 Example ROC curve in genocide prediction, from Goldsmith et al. (2013).
that a probability of one half (0.5) is an intuitive threshold marker for when an
event is likely to happen, but there are some reasons why one might not be satis-
fied with this. First, models of rare events (and within the category of rare events,
genocides are especially rare) often understate probabilities (King and Zeng
2001a). Second, despite numerical indicators, it is to some extent a qualitative
decision as to when an event is likely, or when one might want to predict its occur-
rence. For example, if instead of one-half, there is an estimated one-in-ten chance
of genocide occurring in a particular year, one might still wish to classify this as
a high-r isk case. In contrast, ROC analysis assesses the predictive performance
of the model along the full range of possible thresholds for making a prediction,
obviating the need to specify a cutoff point.
The ROC analysis can be used to assess overall model performance; it can also
be used to assess the effect that individual variables have on the forecasting perfor-
mance of the model. The change in the AUC score can be observed when variables
in the model are added or removed, indicating improved or worsened predictive
performance.6
There are drawbacks to ROC analysis, one of which is that models with
relatively high AUC scores can still be bad at predicting the events that one is
interested in. This problem is especially relevant to predicting genocides. This is
because a prediction of no genocide in any country in any year would be correct
over 99 percent of the time for country-year data. The AUC statistic is not sen-
sitive to this rare event (or imbalanced data) problem (although the confidence
interval around the AUC will be wider in such cases; see He and Garcia 2009).
This further illustrates that whether or not a forecast is “good” depends to an

extent upon a qualitative judgment made with reference to the existing literature
and with respect to the relative benefits and costs of correctly predicting events
and correctly predicting nonevents.
A reasonable approach to evaluate whether progress is being made is compara-
tive and relative weighing of AUC scores for genocide-forecasting models against
those existing in the literature. But this requires that the underlying data and
model structures are comparable. We see genocide forecasts as “good” when they
are able to produce relatively short lists of at-r isk states that have a high true posi-
tive rate—t hat is, are able to get genocide predictions right when they are likely to
occur. The cost of predicting a genocidal event where none occurs is, we believe,
smaller than not predicting a genocidal event where one does occur. It is for this
reason that we believe that correctly predicting genocide onsets (true positives)
should be weighed more heavily than correctly predicting nongenocide onsets
(true negatives). In other words, we suggest a higher tolerance for false positives
than for false negatives.
The distinction between in-sample and out-of-sample forecasts is also relevant.
In-sample forecasting means that predictions are made and accuracy assessed
with reference to the data that the model was fitted to or trained on. Fitting a
statistical model to data on every country in the international system from 1974
to 2012 and then asking how good that model was at predicting genocides from
1974 to 2012 would be in-sample forecasting. Out-of-sample forecasting occurs
when one makes predictions and assesses accuracy on data that the model was
not trained on, or has not yet encountered. Fitting a statistical model to data from
1974 to 1987 and then using that model to predict and assess for 1988 to 2012 is
out-of-sample forecasting. The ROC graphs and AUC statistics can be produced
for both types of forecasts. In general, out-of-sample forecasting is a harder test of
a model and more closely reflects the practical constraints of forecasting. When
predictions are made for data that the model has been fitted to, there is the risk of
overfitting to idiosyncratic aspects of the sample data and therefore of inflating
the predictive accuracy of the model (King and Zeng 2001b). Of course, if one
wants to predict future genocides, one does not have the luxury of knowing what
will happen in these years.
Finally, there are practical differences between sensitivity or accuracy in fore-
casting models and analytical models. Most importantly, forecasting usually
depends on publicly available data that are regularly updated. Thus, some vari-
ables that are included in analytical studies are not practical for forecasting pur-
poses because they are only available for a particular period of time, or are very
expensive to collect for all units (e.g., countries) and time periods (e.g., years) up
to the present (assuming forecasting of future events is the aim). The extent of
popular support for a guerrilla insurgency, which is central in Valentino et al.’s
(2004) work, might be one example, when up-to-date data to predict events of the
next time period cannot be accessed. As it happens, this general data constraint
highlights why fully exploiting the potentially predictive role of economic factors
for forecasting genocide is important: a wide range of economic data is regularly
collected and updated by national governments and intergovernmental institu-
tions such as the UN, the International Monetary Fund (IMF), and the World
Bank and are available for most countries and for relatively long periods of time.
24.3. Role of Economic Risk Factors in Analytical Models

In general, analytical models employing economic variables have focused on
poverty and trade openness as drivers of genocide, with little convergence on the
role of these variables. Easterly, Gatti, and Kurlat (2006), for example, find that
income (or development) is related to the onset of mass killing episodes. Generally,
while it was richer countries in the nineteenth and early twentieth centuries that
were at higher risk of mass killing onset, after 1945 it is the poorer states that have
been at higher risk.
Besançon (2005), Montalvo and Reynal-Querol (2008), Querido (2009),
Kathman and Wood (2011), Esteban, Morelli, and Rohner (2015), Uzonyi (2014),
and Anderton and Carter (2015) test measures of GDP per capita against Harff’s
genocide data and the Uppsala Conflict Data Program’s (UCDP) one-sided
violence dataset. Montalvo and Reynal-Querol (2008), Ulfelder and Valentino
(2008), Esteban, Morelli, and Rohner (2015), and Uzonyi (2014) find no signifi-
cant effect, while Besançon (2005), Querido (2009), Kathman and Wood (2011),
and Anderton and Carter (2015) find that poverty increases the intensity of one-
sided violence episodes or genocides, or the probability of genocide onset.
Ulfelder and Valentino (2008) find that high infant mortality rates have a closer
relationship to mass killing onset, conditional on instability, than does GDP
per capita. At the microlevel, McDoom (2014) finds that proxies of wealth were
not significantly related to the onset of genocidal violence in Rwanda in 1994,
while Fjelde and Hultman (2014) find that geographic locations with higher
incomes (in Africa) experience less government-and rebel-sponsored violence
against civilians. Wood (2014) makes a similar finding regarding violence against
civilians by the Lord’s Resistance Army in Uganda.
Rummel (1995, 21) briefly examines the association of per capita central gov-
ernment expenditures and energy consumption with the likelihood of democide, but
finds “virtually no relationship.” Somewhat surprisingly, even fewer studies have
found a statistically significant relationship between trade openness and geno-
cide, across different specifications of genocide or mass killing and data struc-
tures (Eck and Hultman 2007; Aydin and Gates 2008, Colaresi and Carey 2008;
Anderton and Carter 2015; Esteban, Morelli, and Rohner 2015). Krain (1997,
2005) finds no relationship between a country’s share of global trade and the
severity of genocidal violence, but he does not adjust for each state’s GDP. Other
economic variables have been explored in the literature, including primary com-
modity exports (Besançon 2005; Montalvo and Reynal-Querol 2008; Esteban,
Morelli, and Rohner 2015), Gini coefficients (Kim 2010), and tax capacity and for-
eign aid (DeMeritt 2012), with few conclusive results.7
Ulfelder and Valentino (2008) present what is probably the most comprehen-
sive assessment of economic conditions. They test sixteen variables that we con-
sider economic in nature against the risk of mass killing in ongoing instability
episodes, but retain in their final model only infant mortality, GATT/W TO member-
ship, and a measure that includes economic discrimination. Ulfelder and Valentino
(2008) also find a significant relationship between nonviolent protest, including
general strikes, and mass killing, which could also be considered an economic
variable. Querido (2009) analyzes one-sided violence episodes in African states
from 1989 to 2005 and finds that secondary (alluvial) diamonds and onshore oil-
significantly increase the odds and intensity of mass killing episodes. Querido
(2009) also finds that opium production increases the intensity of one-sided vio-
lence episodes, but that coca and offshore oil production are unrelated to intensity.
However, Querido (2009) examines a small sample that is susceptible to selec-
tion effects. It is possible that these results are driven by a small number of cases
(the Democratic Republic of Congo, Liberia, and Sierra Leone, for example).
A more recent study by Wood (2014) finds that rebels financed by natural resources
are more likely to victimize civilians, as are rebels with foreign support, a finding
replicated elsewhere (Wood 2013; Salehyan, Siroky, and Wood 2014).
Anderton and Carter (2015) consider a number of additional economic vari-
ables. In addition to income and trade openness, the authors test the effects of
GDP growth, economic and political discrimination, and the number of Internet users
per 100 people. They report a strong positive relationship between economic
discrimination and genocide risk, but find no significant relationships with GDP
growth and the spread of the Internet. Esteban, Morelli, and Rohner (2015) are
mostly concerned with genocide incidence (i.e., years in which a genocide began
or was ongoing). Finding a weak positive relationship between the value of oil
production relative to GDP and genocide onset, they then find stronger relation-
ships between oil production and genocide incidence, along with natural resource
rents as a percentage of GDP, diamond production, trade openness, GDP per
capita, and mineral and energy rents to GDP. With regard to oil and diamond
production they also find significant relationships between these variables and
the presence of military massacres at the group level (i.e., ethnic groups that
live in areas of high mineral production are more likely to be victims of military
massacres). This suggests different processes driving the onset and the duration
of genocide episodes. Kathman and Wood (2011) focus on the effect of third-party
military interventions in increasing the costs of genocide or mass killing, and there-
fore reducing their likelihood (see also Krain 2005). This approach suggests to us
that nonmilitary economic interventions to increase those costs might also be useful
for forecasting, for example, direct financial and material aid to targeted groups
or sanctions against governments to deter genocidal policies.8 Krain (2012)
suggests that “naming and shaming” of perpetrators, raising political costs, can
reduce genocidal violence. Thus, we now turn to the role of economic variables in
forecasting studies.
24.4. Role of Economic Variables in Genocide Forecasting

We survey the economic variables that have been found to be related to the onset
of genocide in forecasting models. It is worth emphasizing at the outset that
economic variables that influence the probability of violent episodes short of
genocide, such as civil wars, ethnic wars, and coups, also drive the probability of
genocide (Stewart 2013). Almost all genocides after World War II have occurred
in the context of an ongoing civil war, a successful coup, or reversion to authori-
tarianism. The variables used include the usual items such as poverty (Hegre and
Sambanis 2006), along with access to lootable natural resources and, more recently,
horizontal inequalities among ethnic groups (Cederman, Weidmann, and Gleditsch
2011). We do not deal extensively with these here. Rather, we take a closer look
at variables found to influence genocide specifically. We define economic factors
somewhat narrowly as those that can be (and commonly are) measured in terms
of monetary value.
Other chapters in this volume demonstrate that there are close theoretical
links between economics and the onset of genocide at macro-, country-year, and
microlevels. We are not saying that economic variables may not be causally related
to genocides; they have, however, played a mixed role in predicting genocides.
Table 24.1 is a summary of variables to date that have proven useful for forecast-
ing, and those that have appeared in the analytical models literature.
An important distinction is that between underlying and proximate factors.
The effect of a temporally slow-moving economic variable that is related to geno-
cide can be swamped statistically by the forecasting power of a related but tem-
porally specific political or social variable. For example, political assassinations
are good predictors of genocide, but likely act as triggers or symptoms of impend-
ing genocide rather than as root causes. Poverty and economic crisis might make
assassinations more likely, but these effects could be masked by the (forecasting)
effectiveness of an assassinations variable, if the latter helps the model to make
better predictions.
Only a handful of forecasting studies of interest exist. The seminal work of
Barbara Harff and the Political Instability Task Force (Harff 2003) aims to pre-
dict the onset of genocide in states that are already experiencing instability or state
failure (political or ethnic civil wars and reversals of democracy). This conditional
Table 24.1 Survey of Variables Retained in Predictive Models and
Significantly Associated to Genocide/Mass Killing in Analytical
Studies1
Variable Predictive Studies Analytical Studies Finding

Retaining Variable in a Significant/Substantive
Forecasting Models Relationship
Income/Economic Harff 2003, Ulfelder Easterly, Gatti, and Kurlat
Development 2013, Rost 2013 2006, Besançon 2005, Querido
2009, Kathman and Wood
2011, Anderton and Carter
2015, Wood 2010, Ulfelder and
Valentino 2008
Trade Openness Harff 2003, Hazlett Esteban, Morelli, and Rohner
2011, Ulfelder 2013 2015
Economic Crisis Ulfelder 2013
Natural Resource Ulfelder 2013 Besançon 2005, Esteban, Morelli,
Dependence and Rohner 2015
Iron and Steel Hazlett 2011
Production
Economic Rost 2013 Anderton and Carter 2015,
Discrimination Ulfelder and Valentino 2008
Energy Production
Male Participation
in the Labor Force
Youth Bulge
Income Inequality Besançon 2005
Tax Capacity
Diamond Querido 2009, Esteban, Morelli,
Production and Rohner, 2015
Onshore Oil Querido 2009
Deposits
Offshore Oil
Cocoa Production
Oil Production to Esteban, Morelli, and Rohner,
GDP 2015
Energy and Mineral Esteban, Morelli, and Rohner,
Rents to GDP 2015
(continued)
Variable Predictive Studies Analytical Studies Finding

Retaining Variable in a Significant/Substantive
Forecasting Models Relationship
Internet Users per
100 people
GATT/W TO Ulfelder and Valentino 2008
Membership
Average democracy
score of trading
partners
Trade Herfindahl
Trade to OECD
Countries (% of all
trade)
Aid (% of GDP)
Energy Depletion
(% of GNI)
Forest Depletion
(% of GNI)
Mineral Depletion
(% of GDP)
Predictive Models Surveyed: Harff 2003; Hazlett 2011; Goldsmith et al. 2013; Ulfelder 2013;
Rost 2013.
Analytical Models Surveyed: Besançon 2005; Montalvo and Reynal Querol 2008; Querido
2009; Kathman and Wood 2011; McDoom 2014; Rummel 1995; Eck and Hultman 2007; Colaresi
and Carey 2008; Aydin and Gates 2008; Esteban, Morelli, and Rohner 2015; Krain 1997, 2005;
Kim 2010; Wood 2010; DeMeritt 2012; Uzonyi 2014; Anderton and Carter 2015; Ulfelder and
Valentino 2008.
1
For a survey of variables tested in analytical studies, see Anderton and Carter (2015).
model aims to distinguish episodes of state failure that result in genocide (such as
in Rwanda, where the genocide occurred in the context of a civil war) from those
that do not (such as the war in Sierra Leone, which did not involve genocide or
politicide). Using cases of genocide and politicide during ongoing state failures
from 1955 to 1997, Harff’s model correctly classifies 74 percent of genocide onsets
and 76 percent of nononsets, in-sample. Two economic variables are retained in
the final model: low economic development (proxied as the infant mortality rate)
increases the probability of genocide, and high trade openness reduces it. These
two findings, however, are the product of a long process of testing variables and
of either retaining or discarding them based on theoretical applicability and
forecasting performance.9
In 2011, Chad Hazlett respecified Harff’s model to resolve ambiguities around
whether the model was producing forecasted probabilities of genocide for an
entire instability episode, or just for the year after an instability onset. He also
introduced variables associated with the strategic approach to understanding
genocide (Valentino 2004; Valentino et al. 2004), and extended the model to pro-
duce forecasts for each year of an instability event. Still working with a conditional
modeling framework, Hazlett (2011) found that the ethnic character of the ruling
elite and the number of prior genocides lost forecasting power in this model, but
the importance of trade openness was retained. Hazlett also found that iron and
steel production, which he considers a general measure of state strength, was effec-
tive in predicting which instability episodes resulted in genocide at any point in
their lifetime.10
Jay Ulfelder (2012, 2013) has been working on a country-year model to predict
mass killings, which explicitly avoids a conditional approach in order to produce
forecasts for all countries in each year. Mass killings are defined differently to
genocide and do not have the same requirements for intent to eradicate a group “in
whole or in part.” As such, the definition of mass killing tends to pick up cases of
lower-level repression over longer periods of time, such as South Africa from 1976
to 1994 and Haiti from 1958 to 1986. Ulfelder uses an ensemble Bayesian model
averaging approach, estimating four different models and averaging the predicted
probabilities of mass killing to produce a single forecast. The four models are as
follows: (1) a model based on Colaresi and Carey (2008) that emphasizes mili-
tarization in combination with the level of constraints upon executive decision-
making power; (2) a model based on a combination of Goldstone et al. (2010)
and Harff (2003); (3) a model based on elite threat that predicts the probability of
coups and civil wars; and (4) a model based on a so-called random forest estima-
tion, which includes all of the variables in models 1 to 3. A number of economic
variables are included in these models. Trade openness is included, as well as mea-
sures of poverty (the infant mortality rate or GDP per capita, often measured cat-
egorically), economic decline (more than 2 percent in a year), and natural resource
wealth as a proportion of GNP. The average probability of mass killing onset
across these different models is then used to produce a final forecast. Using k-fold
cross-validation, this approach yields an AUC score of 0.77 (Ulfelder 2013), and
in a more recent iteration (Ulfelder 2014), 0.837. Ulfelder’s approach is capable of
producing estimates for all countries in a given year.11
Nicholas Rost (2013) produced a number of forecasting models for genocide
and politicide. He uses three different dependent variables: genocide/politi-
cide from Harff (2003), democide from Rummel (1995), and mass killing from
Ulfelder and Valentino (2008). Some forecasts are made for all countries and all
years while others are conditional on existing state failure, and he uses both in-
sample and out-of-sample tests of forecasting effectiveness. Rost’s final models
include GDP per capita as a predictor in the country-year forecasts, although he
also finds that economic discrimination against an ethnic, religious, or communal
group prior to the onset of armed conflict makes it more likely that the group will
be targeted with genocidal policies. Rost also finds that measures of trade depen-
dence help predict democide (“death by government,” as defined by Rummel), but
not genocide or mass killing.
The final model we survey is Goldsmith et al. (2013). The authors adopt a
two-stage setup, using one statistical model to predict the probability of any type
of violent political instability (democratic reversals, ethnic wars, or civil wars)
occurring, and then using that probability as a predictor in a second model to
predict annual probabilities of genocide onset for all countries. Some iterations of
the project also used machine-learning approaches to make five-year forecasts for
2011–2015 (see Butcher et al. 2012; Semenovich, Sowmya, and Goldsmith 2012).
The final list of variables included in the Goldsmith et al. (2013) study was the
product of a long process of testing a large number of predictor variables identi-
fied in the literature as being theoretically associated with instability or genocide
onset. Variables were dropped when they did not contribute to in-sample forecast-
ing accuracy (for the period 1974–1988) as measured by the change in AUC when
they were removed. While the infant mortality rate is retained as a predictor of
violent political instability, and thus an indirect predictor of genocide onset, none
of the variables used to directly forecast genocide is strictly economic. However,
the values on some of these variables can be partly explained by economic factors.
One example is the human defense burden, that is, the number of soldiers under
arms as a proportion of the population (which is interacted with another vari-
able in the model, called “executive constraints”). The extent to which a state can
militarize is partly a function of available economic resources and (probably) the
ability to borrow money on international markets.
In addition to the variables reported in the final model, Goldsmith et al. (2013)
tested energy production and the one-, two-, and three-year changes in energy
production, variations on male participation in the labor force, unemployment rates,
changes in unemployment rates, and the economic costs of natural disasters. These
variables were not retained, as they did not increase in-sample forecasting accu-
racy. It is, we think, not insignificant that many standard economic variables have
not proven to enhance the accuracy of genocide forecasting models.
In summary, forecasting models have found a mixed role for economic vari-
ables, which reflects their ambivalent role in analytical studies. Conditional mod-
els (Harff 2003; Colaresi and Carey 2008; Hazlett 2011) tend to find a role for
trade openness, although Ulfelder (2012, 2013) also uses it in his model. The prem-
ise is that trade-dependent states are less likely to respond to instability with geno-
cide or mass killing because of the economic costs imposed through sanctions or
perceptions of increased investment risk. Trade openness, however, is not a signifi-

cant predictor in Rost (2013), Goldsmith et al. (2013), or the annual predictions in
Hazlett (2011). Even more ambivalent has been the role of aggregate wealth mea-
sures. While a clear predictor of political instability events (Goldstone et al. 2010;
Goldsmith et al. 2013; Ulfelder 2013), GDP per capita is only adopted in Rost’s
(2013) model for directly forecasting genocide. Moreover, while there are a host
of economic variables in Ulfelder’s (2013) model of mass killing, including GDP
decline, natural resources, and infant mortality, these are primarily used to predict
the risk of coups and civil wars, which then increase the risk of genocide.12
In some ways this ambivalent role for economic variables may be surpris-
ing, especially since factors such as economic decline have preceded mass kill-
ing events in prominent cases, or since economic deprivation is sometimes used
to incite or justify mass killing (as in Darfur; see Totten 2009). It is difficult to
imagine the rise of Nazism without the Great Depression. But our survey raises the
possibility that while economic factors might have an underlying effect on the like-
lihood of genocidal violence in a society, as predictors in forecasting models they
are overshadowed by political or demographic factors that are more proximate to
genocide onset. Thus it is important to keep in mind the distinction between mod-
eling to understand causation and modeling to optimize prediction. Of course,
the ambivalent role may be related to the different data structures and forecasting
approaches of the studies discussed in this chapter, with some producing forecasts
only for states already experiencing violent political instability, others produc-
ing forecasts for all countries for all years, and yet others for all countries but for
three-to five-year periods. Moreover, these different studies use different methods
to validate the models, with some using in-sample techniques, and others using
cross-validation and out-of-sample forecasts. Different definitions of genocide and
mass killing are also adopted, making results difficult to compare.
In addition, it may be that economic variables have not been fully exploited
for their forecasting potential, a point that we take up in section 24.5. Finally, it
should be emphasized that we are not saying that economic variables do not pre-
dict genocide in any way—on the contrary, economic variables likely do predict
genocide and mass killing—but in the presence of other institutional, sociologi-
cal, or historical variables they are not optimal predictors.
24.5. Potential of Economic Variables in Future

Forecasting Efforts
There remains much to investigate regarding economic variables and the forecast-
ing of genocide. We point to three ways forward: first, deepening analysis of vari-
ables already tested or included in genocide forecasting models; second, exploring
the contribution of variables that have not yet been tested for predictive power;
third, disaggregating the level of analysis below the state to include rebel groups,
which may enhance the fit between particular economic factors and particular
perpetrators and victims of genocidal violence (see Verwimp 2004; McDoom
2013, 2014).
First, variables that have already been tested in the forecasting literature may
not have been fully exploited for their predictive power. In general, we see three
ways in which future work might examine this: (1) considering longer-term
and dynamic effects of economic change; (2) considering further interactions
between economic variables and economic and other structural variables that
reflect the underlying structure provided by economic factors; and (3) loosening
the assumptions of parametric linearity built into regression-based approaches.
As to the first of these, future work might consider the effects of longer-term
economic decline over five or ten years and the interaction of economic growth
and decline. Persistent poor economic growth may create incentives for gov-
ernments to blame or target out-g roups, which may in turn increase the risk of
genocide. Furthermore, it may be the case that it is periods of economic boom,
followed by periods of sharp economic decline, that create strong incentives
for elites to identify and punish out-g roups acting on motivations of relative
deprivation or rising expectations (Olson 1963; Chandra 2002; Chandra and
Foster 2005).
Regarding the second, there are numerous interactions both between eco-
nomic variables and between economic and other variables that may be impor-
tant for genocide onset. Economic crisis, for example, may only matter in certain
types of regimes, for instance in autocratic regimes that derive their legitimacy
more from nationalism or ideology than good governance. Advances in statis-
tical software (see, e.g., Kenkel and Signorino 2011) may also help forecasters
to identify the most likely interactions and functional forms to include in their
models. Economic inequality has also not been included in forecasting models to
date. Although there is only one quantitative study that we are aware of that tests
income inequality on genocide onset and severity, and with inconclusive results
(Besançon 2005), recent work on horizontal inequalities across different ethnic
groups and the likelihood of civil war points to the possibility that the denial of
economic rights and privileges to specific ethnic groups may be a precursor to
genocide (Cederman, Weidmann, and Gleditsch 2011). Certainly dispossession
and exclusion have preceded some of the worst episodes of mass killing in the
twentieth century, and Besançon (2005) has found that educational inequality
makes severe genocides more likely.
As to the third item within the first set of possible extensions in future research,
exploring nonlinear relationships between economic variables and genocide/
mass killing onset may increase predictive power. Present work has, in general,
assumed that economic variables and genocide onset are related in a linear way
(i.e., a USD 100 change in income per capita has the same effect on genocide onset
for a country with an average income of USD 500 per person than for another
state with an average income of USD 30,000 per person), or at least a monotonic
direction (for an exception see Easterly, Gatti, and Kurlat 2006). However, it
may be the case that middle-income regimes are most at risk because genocide
requires an infrastructure that is costly to build (for preliminary evidence, see
Fein 1995, 190). So, too, with trade openness: it may be the case that states need to
cross a certain threshold of openness, only beyond which their exposure to costs
imposed by the international community is high enough to deter genocidal poli-
cies. Machine-learning approaches to forecasting may help to tease out some of
these nonlinear effects. To date, there has been mixed success using these types of
models. In some instances, machine-learning models have performed marginally
better than standard econometric models (Ulfelder 2012, 2013).13
Second, in addition to expanding the analysis of existing economic variables,
future work might incorporate economic variables that have not yet been con-
sidered in forecasting models. A number of variables used in analytical models
might be candidates, such as measures of natural resource or mineral rent-
dependence, economic discrimination, and income inequality. Economic vari-
ables that have not been tested in either the analytical or predictive literature
on mass atrocity may also be of use. Incorporating information regarding mass
atrocity risk embedded in international credit and bond markets might be a use-
ful step forward, especially for shorter-term forecasts. Markets are generally bad
at predicting internal conflicts (Gelos, Sahay, and Sandleris 2004; DiGiuseppe,
Barry, and Frank 2012), but if governments prepare for genocide years before
its onset through the creation of out-g roups, escalation of xenophobic rhetoric,
and rapid militarization, then these relatively public events may indicate that
a mass killing episode is impending and can be factored into the future costs
of doing business in that country.14 For genocide in particular, there are often
smaller-scale killings or massacres that precede the outbreak of genocide.
Furthermore, the rapid recruitment of soldiers or paramilitary forces that can
precede mass killing episodes is costly, and may be reflected in increased bor-
rowing during times of violent conflict (above the baseline rate that would be
expected in a situation of civil war), for example, by the selling of government
bonds. In short, financial markets may provide information on the likelihood
of mass killing events that might be of use to forecasters. Potentially, condi-
tions of high youth unemployment (perhaps accompanied by high inflation
and society-w ide inequality) may also be useful for forecasting genocide, as it is
this pool of individuals that often provides the recruits for paramilitary groups
(Lemarchand 2009).
Third, while episodes of government-s ponsored mass killing were dead-
liest in the twentieth century, at least since 1989 it has been far more com-
mon for rebel groups to brutalize unarmed civilians. The horrific trail of
destruction left by Boko Haram in Nigeria in recent years is a case in point.
In the future, it may be possible to predict which rebel groups are most
likely to kill civilians. Such an approach has a genuine possibility to deter
such acts, as naming and shaming by international organizations appears to
reduce the severity of these episodes (DeMeritt 2012; Krain 2012). For this,
comprehensive, systematic, and publicly available data on mass killings
by rebel groups would be necessary, and the Uppsala One-S ided Violence
Dataset, which provides information on rebel violence against civilians,
might be a useful starting point to build such a dataset (Eck and Hultman
2007). It may be the case that a handful of easily observable characteris-
tics help sort between benign rebel groups (in terms of the deliberate kill-
ing of civilians) and more dangerous groups (see Cunningham, Gleditsch,
and Salehyan 2013). Recent scholarship (Salehyan et al. 2014; Wood 2013,
2014) suggests that foreign sponsorship and financing of rebel groups, and
the nature of that sponsorship, plays a role in shaping incentives for violence
against civilians. Even if publicly available data cannot be gathered on rebel
groups, we may be able to use structural variables (such as resource depen-
dence in general, or state weakness) to predict which civil wars are more
vulnerable to one-s ided violence.
24.6. Conclusion
In this chapter we have explored the use and untapped potential of economic
variables, defined as those measurable in monetary value, in genocide-forecast-
ing efforts. Our survey of the forecasting literature points to some utility for
per capita income or wealth indicators and national-level trade openness. We
suggest ways these might be further explored to improve their contribution to
forecasting efforts. We also point to a number of factors in the related analyti-
cal model literature that deserve consideration for the purpose of forecasting,
such as inequality or other relative deprivation-based measures, financial mar-
ket activity, and commodity exports. These suggest a move away from broad,
nationally focused indicators toward those focused on the microfoundations of
potential genocidal violence. We believe this is a promising direction. We also
suggest that economic variables’ limited predictive power to date has to do with
their position as underlying rather than proximate factors. But this also points
toward new ways to consider better exploiting their utility for forecasting,
because they can be considered as conditioning factors that can help to better
specify, for example, when ethnic scapegoating or political assassinations are
more likely to be precursors to genocidal violence. We believe there is consider-
able promise for improving the understanding, forecasting, and, ultimately, the
prevention of genocide.
Notes
1. The Atrocity Forecasting Project was developed with the assistance of the Australian Aid
Agency (AusAID). The Political Instability Task force is supported by funding from the
US government.
2. It is becoming more common for studies of genocide to use both statistical significance and
forecasting power for inference. See Colaresi and Carey (2008) and Pilster, Bohmelt, and
Tago (2012).
3. This finding accords with studies of peacekeeping that find that peacekeepers tend to be
deployed in the most violent and intractable cases (Fortna 2004).
4. That is, the percentage of actual positive outcomes (e.g., genocides) that are correctly pre-
dicted. The proportion of all positive forecasts that are true positives is termed “precision.”
5. Out-of-sample pseudo R2 has also been used in economics for probit forecasting models
(Ratcliff 2013).
6. Alternatively, the AUC produced by each predictor individually, excluding all others, can
be assessed.
7. Besançon (2005) does find a positive and significant effect on genocide severity, but the
marginal effects are substantively small.
8. Potentially contradicting this is the finding of Valentino et al. (2004) that guerrilla insur-
gencies receiving support from the civilian population are more likely to be met by geno-
cidal violence from the state. In some situations, mass killing may increase in likelihood as
the costs of defeating an insurgency militarily rise.
9. In a 2013 update of the model, the low development variable was dropped: http://w ww.
gpanet.org/content/r isks-new-onsets-genocide-a nd-politicide-2 013.
10. A measure of youth bulge was also tested for the Harff model but not retained. Hazlett did
not find that trade openness or iron and steel production were useful predictors of geno-
cide when making forecasts one year into the future for ongoing instability events.
11. One point made by Ulfelder (2013), with which we agree, is that, while these forecasts
are likely to be quite noisy, they sometimes identify cases that seem counterintuitive but
have factors that have historically placed them at high risk. These may be cases that, due to
attention bias, we do not usually consider. A recent example would be the Central African
Republic, which, in 2010 and 2011, was generally not considered to be a case at high risk of
mass killing.
12. These variables are also included in the random forest model designed to predict genocide.
13. The models presented in Semenovich et al. (2012), Butcher et al. (2013), and Ulfelder
(2013) have reported better predictions using nonparametric (machine learning–based)
approaches. On machine-learning, also see chapter 23 in this volume.
14. This assumes that genocidal policies actually do influence investment risk.
References
Anderton, C., and J. R. Carter. 2015. “A New Look at Weak State Conditions and Genocide
Risk.” Peace Economics, Peace Science and Public Policy 21, no. 1: 1–36.
Aydin, A., and S. Gates. 2008. “Rulers as Mass Murderers: Political Institutions and Human
Insecurity.” In S. M. Saidman and M. J. Zabar, eds., Intra-State Conflict, Governments and
Security: Dilemmas of Deterrence and Assurance. New York: Routledge, 72–95.
Besançon, M. L. 2005. “Relative Resources: Inequality in Ethnic Wars, Revolutions, and
Butcher, C., B. Goldsmith, D. Semenovich, and A. Sowmya. 2012. “Understanding and
Forecasting Political Instability and Genocide.” In Policy Report for the Atrocity Forecasting
Project, University of Sydney. http://s ydney.edu.au/a rts/research/atrocity_forecasting/

downloads/docs/GenocideForecastingReportGlobal_ 2 012afp.pdf.
Cederman, L., N. B. Weidmann, and K. S. Gleditsch. 2011. “Horizontal Inequalities and
Ethnonationalist Civil War: A Global Comparison.” American Political Science Review 105,
no. 3: 478–95.
Chandra, S. 2002. “Race, Inequality, and Anti-Chinese Violence in the Netherlands Indies.”
Explorations in Economic History 39, no. 1: 88–112.
Chandra, S., and A. W. Foster. 2005. “The ‘Revolution of Rising Expectations,’ Relative
Deprivation, and the Urban Social Disorders of the 1960s: Evidence from State-L evel
Data.” Social Science History 29, no. 2: 299–332.
Colaresi, M., and S. C. Carey. 2008. “To Kill or Protect: Security Forces, Domestic Institutions,
and Genocide.” Journal of Conflict Resolution 52, no. 1: 39–67.
Cunningham, D. E., K. S. Gleditsch, and I. Salehyan. 2013. “Non-State Actors in Civil
Wars: A New Dataset.” Conflict Management and Peace Science 30, no. 5: 516–31.
DiGiuseppe, M. R., C. M. Barry, and R. W. Frank. 2012. “Good for the Money: International
Finance, State Capacity, and Internal Armed Conflict.” Journal of Peace Research 49, no.
3: 391–4 05.
DeMeritt, J. H. R. 2012. “International Organizations and Government Killing: Does Naming
and Shaming Save Lives?” International Interactions 38, no. 5: 597–621.
Economy 123, no. 5: 1087–132.
Fawcett, T. 2006. “An Introduction to ROC Analysis.” Pattern Recognition Letters 27, no.
8: 861–74.
Fein, H. 1995. “More Murder in the Middle: Life-I ntegrity Violation and Democracy in the
World, 1987.” Human Rights Quarterly 17, no. 1: 170–91.
Fjelde, H., and L. Hultman. 2014. “Weakening the Enemy: A Disaggregated Study of Violence
against Civilians in Africa.” Journal of Conflict Resolution 58, no. 7: 1230–57.
Fortna, V. P. 2004. “Does Peacekeeping Keep Peace? International Intervention and the
Duration of Peace after Civil War.” International Studies Quarterly 48, no. 2: 269–92.
Gelos, R. G, R. Sahay, and G. Sandleris. 2004 “Sovereign Borrowing by Developing Countries:
What Determines Access?” IMF Working Paper WP/0 4/221. Washington, DC: IMF.
Goldsmith, B., C. R. Butcher, D. Semenovich, and A. Sowmya. 2013. “Forecasting the Onset
of Genocide and Politicide: Annual Out- of-
Sample Forecasts on a Global Dataset,
1988–2 003.” Journal of Peace Research 50, no. 4: 437–52.
Goldstone, J., R. H. Bates, D. L. Epstein, T. R. Gurr, M. B. Lustik, M. G. Marshall, J. Ulfelder,
and M. Woodward. 2010. “A Global Forecasting Model of Political Instability.” American
Journal of Political Science 54, no. 1: 190–2 08.
Harff, B., and T. R. Gurr. 1988. “Toward Empirical Theory of Genocides and Politicide: Identifi
cation and Measurement of Cases since 1945.” International Studies Quarterly 32, no. 3: 359–71.
Hazlett, C. 2011. “New Lessons Learned? Improving Genocide and Politicide Forecasting.”
http://w ww.ushmm.org/m/pdfs/2 0111102-hazlett-early-_w arning-lessons-learned.pdf.
Washington, DC: United States Holocaust Memorial Museum.
He, H., and E. A. Garcia. 2009. “Learning from Imbalanced Data.” IEEE Transactions on
Knowledge and Data Engineering 21, no. 9: 1263–8 4.
Hegre, H., and N. Sambanis. 2006. “Sensitivity Analysis of Empirical Results on Civil War
Onset.” Journal of Conflict Resolution 50, no. 4: 508–35.
Kathman, J. D., and R. M. Wood. 2011. “Managing Threat, Cost, and Incentive to Kill: The Short-
and Long-Term Effects of Intervention in Mass Killings.” Journal of Conflict Resolution 55,
no. 5: 735– 60.
Kenkel, B., and C. Signorino. 2011. “Data Mining for Theorists.” The Society for Political
Methodology. Working Paper Series No. 1278. http://polmeth.wustl.edu/mediaDetail.
php?docId=1278.
Kim, D. 2010. “What Makes State Leaders Brutal? Examining Grievances and Mass Killing dur-
ing Civil War.” Civil Wars 12, no. 3: 237–6 0.
King, G., and L. Zeng. 2001a. “Explaining Rare Events in International Relations.” International
King, G., and L. Zeng. 2001b. “Improving Forecasts of State Failure.” World Politics 53,
no. 4: 623–58.
Krain, M. 1997. “State-Sponsored Mass-Murder: The Onset and Severity of Genocides and
Politicides.” Journal of Conflict Resolution 41, no. 3: 331–6 0.
Krain, M. 2005. “International Intervention and the Severity of Genocides and Politicides.”
Krain, M. 2012. “J’accuse! Does Naming and Shaming Perpetrators Reduce the Severity of
Genocides or Politicides?” International Studies Quarterly 56, no. 3: 574–89.
Lemarchand, R. 2009. “The 1994 Rwanda Genocide.” In S. Totten and W. S. Parsons, eds.,
Century of Genocide: Critical Essays and Eyewitness Accounts. 3rd ed. New York: Routledge,
483–5 04.
Mason, S. J. 2004. “On Using ‘Climatology’ as a Reference Strategy in the Brier and Ranked
Probability Skill Scores.” Monthly Weather Review 132, no. 7: 1891–95.
McDoom, O. S. 2013. “Antisocial Capital: A Profile of Rwandan Genocide Perpetrator’s Social
Networks.” Journal of Conflict Resolution 58, no. 5: 865–93.
McDoom, O. S. 2014. “Predicting Violence within Genocide: A Model of Elite Competition and
Ethnic Segregation from Rwanda.” Political Geography 42 (September): 34–45.
Montalvo, J. G., and M. Reynal-Querol. 2008. “Discrete Polarisation with an Application to the
Olson, M. 1963. “Rapid Growth as a Destabilizing Force.” Journal of Economic History 23, no.
4: 529–52.
Pilster, U., T. Bohmelt, and A. Tago. 2012. “The Onset of Genocidal Violence and the Structure
of Military Forces.” Paper Presented at the Conference for Advancing the Scientific Study
of Conflict and Cooperation: Alternative Perspectives from the UK and Japan, Colchester,
UK. March 20–22.
Querido, C. M. 2009. “State-Sponsored Mass Killing in African Wars: Greed or Grievance?”
International Advances in Economic Research 15, no. 3: 351–61.
Ratcliff, R. 2013. “The ‘Probability of Recession’: Evaluating Probabilistic and Non-
Probabilistic Forecasts from Probit Models of U.S. Recessions.” Economics Letters 121, no.
2: 311–15.
Rost, N. 2013. “Will It Happen Again? On the Possibility of Forecasting the Risk of Genocide.”
Resolution 39, no. 1 :3–2 6.
Abuse: A Principal-A gent Analysis of Wartime Atrocities.” International Organization 68,
no. 3: 633–61.
Semenovich, D., A. Sowmya, and B. E. Goldsmith. 2012. “Predicting Onsets of Genocide with
Sparse Additive Models.” In Pattern Recognition (ICPR), 2012 21st International Conference:
3549–52.
Schrodt, P. A. 2014. “Seven Deadly Sins of Contemporary Quantitative Political Analysis.”
Sokolova, M., N. Japkowicz, and S. Szpakowicz. 2006. “Beyond Accuracy, F- Score and
ROC: A Family of Discriminant Measures for Performance Evaluation.” In A. Sattar and
B. H. Kang, eds., AI 2006: Advances in Artificial Intelligence, Proceedings, Volume 4304: 1015–21.
Stewart, F. 2013. “The Causes of Civil War and Genocide: A Comparison.” In A. Lupel and E.
Verdeja, eds., Responding to Genocide: The Politics of International Action. Boulder, CO: Lynne
Rienner, 47–8 4.
Straus, S. 2007. “Second-Generation Comparative Research on Genocide.” World Politics 59, no.
3: 476–501.
Straus, S. 2012. “Destroy Them to Save Us: Theories of Genocide and the Logics of Political
Violence.” Terrorism and Political Violence 24, no. 4: 544–6 0.
Totten, S. 2009. “The Darfur Genocide.” In S. Totten and W. S. Parsons, eds., Century of
Genocide: Critical Essays and Eyewitness Accounts. 3rd ed. New York: Routledge, 513–77.
Ulfelder, J. 2012. “Forecasting Onsets of Mass Killing.” Paper Presented at the Annual Northeast
Political Methodology Meeting. New York University, May 4.
Ulfelder, J. 2013. “A Multimodel Ensemble for Forecasting Onsets of State-Sponsored Mass
Killing.” Paper Presented at the American Political Science Association 2013 Annual
Meeting. http://ssrn.com/abstract=2303048.
Ulfelder, J. 2014. “A New Statistical Approach to Assessing Risks of State-L ed Mass Killing.”
http://d artthrowingchimp.wordpress.com/2 014/01/2 2/a-new-s tatistical-approach-to-
assessing-r isks-of-state-led-mass-k illing/ [accessed July 25, 2014].
Ulfelder, J., and B. Valentino. 2008. “Assessing the Risks of State-Sponsored Mass Killing.”
http://ssrn.com/abstract=1703426 or http://d x.doi.org/10.2139/ssrn.1703426.
Uzonyi, G. 2014. “Domestic Unrest, Genocide and Politicide.” Political Studies. Online first
before print. http://onlinelibrary.wiley.com/doi/10.1111/1467-9248.12181/abstract.
Valentino, B. 2004. Final Solutions: Mass Killing and Genocide in the 20th Century. Ithaca,
Guerrilla Warfare.” International Organization 58, no. 2: 375–4 07.
Verwimp, P. 2004. “An Economic Profile of Peasant Perpetrators of Genocide—M icro-L evel
Ward, M. D., B. D. Greenhill, and K. M. Bakke. 2010. “The Perils of Policy by P-Value: Predicting
Civil Conflicts.” Journal of Peace Research 47, no. 4: 363–75.
Research 47, no. 5: 601–14.
Origins of Support and Civilian Victimization in Civil War.” Conflict Management and Peace
Science 31, no. 5: 461–80.
Wood, R. M. 2014. “From Loss to Looting? Battlefield Costs and Rebel Incentives for Violence.”
International Organization 68, no. 4: 979–9 9.
Wooldridge, J. M. 2009. Introductory Econometrics: A Modern Approach. 4th ed. Mason,
OH: Cengage Learning.
25
Business in Genocide
Understanding and Avoiding Complicity
Nor a M . St e l a n d W i m Nau dé
25.1. Introduction
Along with the United Nations (UN) Genocide Convention, we define geno-
cide as mass atrocity “committed with the intent to destroy, in whole or in part,
a national, ethical, racial or religious group” (Kelly 2012, 357; on definitions and
data, see chapters 1 through 3 of this volume). Entrepreneurs and their businesses
are not often associated with genocides and other mass atrocities (GMAs). This is
no doubt due to the fact that, by and large, business, enterprise, and commercial
interactions between people are not zero-sum games, but are based on mutual
benefits. Violent conflict, especially to the extreme embodied in genocides,
destroys markets, infrastructure, assets, and resources; undermines trust; and,
as such, undermines trade and investment (Brück, Naudé, and Verwimp 2013).
Indeed, “where human rights are respected and defended, businesses flourish”
(Mallinson as quoted in Chella 2012, 295). Nevertheless, genocides take place
too often to maintain that there may never be vested interests for businesses to be
complicit in such gross human rights violations (Stokes and Gabriel 2010).
Describing and understanding business complicity in GMAs is the main pur-
pose of this chapter. As the substantial literature on the topic, covered extensively
in this book, has convincingly illustrated, GMAs do not arise spontaneously, but
tend to be meticulously sourced and managed. Entrepreneurs and businesses
(firms, companies) may be particularly effective vehicles through which such
planning, management, and execution, not to mention financing, of GMAs are
facilitated. As such, our interest is not the business of genocide (i.e., the manage-
ment and organization of genocidal processes; see, e.g., c hapter 14 in this volume),
but rather business in genocide (i.e., the role of businesses in these processes),
with a particular focus on the agency and decision making of entrepreneurs and
managers.
591
In the literature, the role of business in genocides is hardly ever discussed

separately from other mass atrocities. But while all genocides are mass atrocities,
not all mass atrocities are genocides. Mass atrocities refer to scale and quantity,
whereas genocide refers to intentionality and quality (Kelly 2012, 357). Indeed,
the wealth of studies on corporate complicity in GMA as a broader category
obscures how little we know about the role of businesses in genocides specifically.
Businesses’ general complicity in military regimes and dictatorships, or deploy-
ment in war zones and conflict areas, is crucially different from their specific com-
plicity in particular mass atrocities (Kaleck and Saage-Maaß 2010, 702–3). This
is so, first, and as elaborated on in c hapters 2 and 3 of this volume, because of the
intent clause in the genocide definition. While mass atrocities can be a means
toward another end or a byproduct of a specific development or policy, genocide
is per definition an end in its own right. This distinction particularly matters, sec-
ond, in light of the companies’ valuable reputations. While it could be argued that
being associated with any mass atrocity is bad for business, it is safe to say that
the public—which consists of customers, clients, voters, and policymakers—is
putatively more concerned about genocide, “the crime of crimes” (Kelly 2012,
339). Unlike the more generic container concept of mass atrocities, which might
easier be sold as unavoidable in certain regions, sectors, or situations, the label
of genocide has a sociopolitical explosiveness, and generates a degree of media
attention, that brings with it an exceptional vulnerability to reputational damage
through “naming and shaming” campaigns (Jacobs 2008, 49).1 For this reason, as
well as for the sake of conceptual and empirical demarcation, this chapter focuses
on corporate complicity in genocide specifically—even if many of our observa-
tions might apply to other mass atrocities as well.
There are many examples in history of business in genocide. One can think of
the role of the Dutch and British East India Companies, the world’s first multi-
nationals, in Asia or in the slave trade (Stephens 2002, 49; Chella 2012, 61; Slim
2012, 907); IBM’s complicity in the Holocaust (Black 2001; Stephens 2002); the
collaboration between Hollywood film studios and the Nazis (Urwand 2013);
and the controversies surrounding multinational corporations in resource-r ich
developing countries (Clapham and Jerbi 2001; Cooper 2002; Rieth and Zimmer
2004; Kaeb 2008; Chella 2012). On the other hand, there are also instances of
business playing peace-promoting and peace-supporting roles (Naudé 2009).
Sweetman (2009) for instance mentions that businesses acted as “early warning
systems” in Burundi during the Rwandan genocide and in Macedonia during the
Yugoslavian conflict and that this reduced violence. However, a holistic analysis
of the involvement of commercial organizations in genocide is so far unavailable.
This chapter aims to start to fill this gap by critically exploring the status quo
of the academic literature on corporate complicity in genocide in light of case-
based evidence in order to contribute to advancing conceptualization of the phe-
nomenon. Ultimately, we seek to generate more insights into what productive
Business in Genocide 593
entrepreneurship and socially responsible businesses can do to lessen the ten-

sions, misunderstandings, exclusions, and marginalization that are among the
complex causes of genocides and other mass atrocities.
We commence in section 25.2 by mapping the scholarly debates about what
manifests, determines, and drives corporate involvement in genocides. In sec-
tion 25.3, we explore the specific role entrepreneurs and businesses played in
three of the most uncontested genocides of recent history: the Jewish, Kurdish,
and Darfurian genocides.2 We integrate insights from these cases with findings
from the literature in section 25.4. In section 25.5 we conclude by offering sev-
eral emerging lessons, venture a tentative conceptualization of the “what,” “how,”
and—particularly—“why” of business in genocide, and provide suggestions for
further research.
25.2. Corporate Complicity in Genocides: How

and Why
25.2.1. How?
The literature offers no consensus on the categorization of the roles businesses
might play in genocide (Sherman 2001, 4). Five core positions (that can over-
lap and coexist) can nevertheless be deduced from scholars’ discussions on cor-
porate complicity: preventer, victim, indirect accomplice, direct accomplice,
and perpetrator (Kaeb 2008, 332; Bernard 2012, 882; Chella 2012, 52–53).
Preventers—here, the emblematic Schindlers and Rusesabaginas of the world
come to mind—m ight have striven to contribute to deescalation, early warn-
ing, or counteracting (Rieth and Zimmer 2004; Oetzel, Getz, and Ladek 2007;
Bray and Crockett 2012, 1070). Businesses considered victims might have lost
personnel or assets or been forced to evacuate, a characterization that applies to
some degree to almost all businesses in genocide-a ffected contexts. Perpetrators
will have instigated and executed genocide as primary actors. While there are
no documented cases of companies acting as the main perpetrator in registered
genocide (Chella 2012, 33; Kelly 2012, 340), it should be kept in mind that in
genocide “the ‘accomplice’ is often the real villain and the ‘principal offender’ a
small cog in the machine” (Schabas 2000, 286). In some cases, the violations are
not perpetrated by the businesses, but on behalf of them, such as can be argued
in the case of the forced relocation of indigenous people for the benefit of extrac-
tive industries (Manby 2000, 6; Clapham and Jerbi 2001, 341; Kelly 2012, 360),
resulting in armchair perpetrators (Kyriakakis 2012, 993) or complicity by
i nstigation (Jacobson 2005, 206).
It is therefore direct and indirect complicity that seem the most prevalent
forms of business in genocide and are hence our main concern. Direct com-
plicity is the provision of finance, material, infrastructure, human resources
and capacity, and intelligence to the perpetrator of the genocide to the extent
that the genocide is facilitated (i.e., could not have been implemented the
way it was without the support of the company in question). In short, when
companies are directly complicit in genocide, they are “providing means with
which wars are fought” (Tripathi 2010, 133); they are, first and foremost, sup-
pliers (Slim 2012, 913). Direct complicity is also often described as aiding
and abetting and can concern the processes of ordering, instigating, solicit-
ing, inducing, inciting, joining, planning, preparing, and conspiring (Chella
2012, 23–2 5, 40). The main issue is supplying, through trade, taxes, royal-
ties, and services or even through nonspecified payments, a genocidal group
or regime with the resources and finances it needs to execute the genocide
(Cooper 2002; Ramasastry and Thompson 2006, 17; Martin-O rtega 2008,
274; Wennmann 2009, 1129; Tripathi 2010, 133; Chella 2012). There are
also many instances where businesses have facilitated human rights viola-
tions by providing the necessary means of transportation or infrastructure,
often through close cooperation dynamics with (para-) military or private
security groups protecting the company’s compounds or assets such as mines
or oil fields (Manby 2000, 6; Khan 2005, 2; Kaeb 2008, 336; Moffatt 2009,
27; Tripathi 2010, 134; Chella 2012). In such instances, providing training
and/or equipment and sharing of intelligence are also forms of complicity
(Martin-O rtega 2008, 275; Kaleck and Saage-M aaß 2010, 709; Bray and
Crockett 2012, 1077).
Indirect complicity can fall in two categories. Beneficial complicity occurs
when the company benefits from the genocide in some way, whether it was aware
of or sympathetic to it or not. Silent complicity is apparent when the company
does not contribute to or benefit from the genocide, but is aware of it and fails to
distance itself from it (Jacobson 2005, 202). Legal scholars have been very keen
to emphasize that “silence is not neutrality” (Chapham and Jerbi 2001, 347) but
“an expression of moral support” (Wettstein 2010, 40). 3
These degrees and forms of business complicity in genocide can be estab-
lished based on the notion of proximity to the violator (perpetrator), the
violated (victims), and the violation (event) (Tripathi 2010, 140). Proximity,
in turn, is closely related to the “knowledge and foreseeability” that compa-
nies had of the genocidal events going on, which can be assumed to depend
on the geographical closeness to the event and the frequency and duration
of the company’s contact with the perpetrator (Huisman and van Sliedregt
2010, 819; Tripathi 2010, 140; Wettstein 2010, 35–37). Several law cases have
stressed that companies can be complicit even if they had not participated in
the event or benefited from it; the mere knowledge that the event was going
on generates complicity (Huisman and van Sliedregt 2010, 823). Indeed, “the
ostrich syndrome cannot be evoked as a legitimate defence” (Bohoslavsky and
Opgenhaffen 2010, 171).
25.2.2. Why?
The main sentiment in the debate on what drives business to become involved in
genocides derives from the businesses’ perceived self-interest. On the one hand,
when it concerns companies’ positive behavior in contexts of genocide, it is often
argued that they do so to avoid damage to their business and reputations (e.g.,
Sherman 2001; Rieth and Zimmer 2004; Oetzel, Getz, and Ladek 2007). Here,
the reputational damage of complicity, and subsequent economic losses, is a
recurrent theme of many authors (Clapham and Jerbi 2001, 340; Sherman 2001,
9; Tripathi 2010, 133; Kelly 2011, 415; Amunwa 2012, 12).
On the other hand, when businesses are negatively complicit, it is also assumed
that their self-interest is at the forefront (Sherman 2001). Guaranteeing profit,
preventing losses, maintaining a competitive advantage, protecting investments,
and resource security are seen as the main motivations for corporations to act the
way they do before, during, and after genocides in a substantial part of the litera-
ture (see, e.g., Jacobson 2005, 2008; Ramasastry and Thompson 2006; Martin-
Ortega 2008; Watts 2008; Moffatt 2009; Huisman and van Sliedregt 2010; Chella
2012; van Baar and Huisman 2012).
Guidolin and La Ferrara (2007, 1978–79) describe how the context of violent
conflict often preceding and enabling genocide can have explicit benefits for com-
panies, predominantly in the extractive industries. It installs entry barriers and
thus limits competition, advantaging those businesses already involved; ensures
that the bargaining power of conflict parties, often including the state, is low
due to their need for immediate revenue to sustain the conflict, which can for
instance undercut licensing costs; and lowers transparency, which makes profit-
able informal deals more likely. Bray and Crockett (2012, 1072) similarly note
that postconflict countries constitute “zones of untapped potential and pent-up
consumer demand.”
Such neutral profit-seeking behavior of companies is often seen, as exemplified
by van der Wilt (2006, 255), as a mitigating circumstance: “After all, it seems
far-fetched—and it may not even be fair—to compare the diabolical minds who
frame the extinction of a whole group, with businessmen whose prime interest
is to make profits and who may have reconciled themselves with the possibil-
ity that their trade may support heinous international crimes.” Jacobson (2005,
205) argues: “An accomplice may not even wish the crime to occur, but he is still
willing to provide the aid to the principal offender for another reason, such as
profit.”
The problem is, however, that since the motive of self-interest is such an
obvious explanation for complicity, it may importantly obfuscate other moti-
vations or concerns leading company decision-making concerning geno-
cide (Tripathi 2010). The debate on what determines businesses’ decision
making in and toward the genocidal process is a rather crude one, painting
genocides as an opportunity for unproductive and destructive entrepreneur-

ship (e.g., Baumol 1990), exemplified by the arms trade or “exploitable” and
“lootable” natural resources, and featuring companies as rational actors who
cynically balance reputational damage on the one hand and the maximization
of profit and minimization of loss on the other. The relative underdevelop-
ment of this scholarly discussion seems to follow from the dearth of empirical
data on companies’ motivations and considerations in this realm. For obvi-
ous reasons—bad public relations—very little is known about the decision-
making processes of companies confronted with genocide processes and the
interests driving them (Sherman 2001, 9). It is indeed very hard to analyze
the concrete roles of business in genocide because both good and bad deeds
are often invisible, kept “away from the public eye,” and poorly documented
(Wennmann 2012, 934). In many situations, moreover, businesses maintain
they did not have any motivations one way or the other because their role was
entirely unintentional. Indeed, Sherman (2001, 2) calls businesses’ concerns in
decision making during genocides a “black box” for outsiders. While it is
beyond the scope of this chapter to veritably open this black box, in the follow-
ing we will endeavor to at least develop some conceptual tools assisting future
researchers in approaching it.
25.3. Three Cases

25.3.1. The Kurdish Genocide in Iraq
In 1987 and 1988 Saddam Hussein attacked the Kurdish community living in
North Iraq with chemical and regular weapons, in what is known as the Anfal
campaign. The attempt to “purify” the area and bring it under Arab influence
resulted in up to 100,000 deaths, 182,000 “disappeared” people, and the displace-
ment of some 1.5 million people (HRW 1993, 5; Kelly 2013, 363). The event has
been recognized by both Iraqi and foreign courts and institutions as genocide
(HRW 1993; Kelly 2013, 360).
Various national and international businesses were directly complicit in this
genocide. In particular, a wide range of European, predominantly German, busi-
nesses supplied Saddam Hussein with the equipment and resources to imple-
ment his genocide (Hippler 1991; HRW 1993, 43; Shenon 2002; Kelly 2013).4
Starcevic (1990) notes that no less than “six Iraqi poison gas plants were built with
German help, according to an independent report commissioned by German
authorities.” Milhollin and Motz (2003) conclude, “[T]‌he data reveals that firms
in Germany and France outstripped all others in selling the most important
thing—specialized chemical-industry equipment that is particularly useful for
producing poison gas. Without this equipment, none of the other imports would
have been of much use.”
Kelly’s (2013) account convincingly demonstrates that these businesses can be

considered direct accomplices in the genocide because, first, they were aware of
the genocidal intent of Hussein vis-à-v is the Kurdish population and the nature
of the resources they sold him; and, second, they nevertheless continued to sell
him these products without which the genocide could not have been executed.
Kelly (2013, 370) argues that the German companies that supplied Saddam
Hussein with the ingredients for his chemical weapons program knew what he was
up to because Iraq’s use of chemical weapons against Iran during the 1980–1988
Iran-I raq War had been widely published. The complicity of the companies, then,
concerns the pregenocidal stage: the businesses helped Hussein prepare the
genocide, but played no role in the execution of the genocide.
While it is credibly shown that these companies are in fact complicit, that they
must have known what was about to happen, the question of why they neverthe-
less continued to participate is somewhat unsatisfyingly answered with the stan-
dard reference to profit maximization. While accounts of corporate complicity
in the Kurdish genocide in Iraq provide welcome insights into how businesses
can, seemingly quite intentionally, facilitate and enable genocide, they do not
tell us much about the concrete decision-making processes driving the compa-
nies toward full-fledged involvement, nor do they shed light on the positioning of
these companies in the postgenocidal phase—for instance, in response to survi-
vors’ calls for prosecution (Dolamari 2014).
25.3.2. The Genocide on the Christian

Population of Darfur, Sudan
In the Darfur region of Sudan, government-sponsored and directed Janjaweed
militias massacred some 80,000 people in the 2003–2005 period and displaced
approximately two million from 1999 to 2006 (Clarke 2012). 5 These militias
explicitly targeted the Christian “African” (as opposed to the Muslim “Arab”)
population of the region, subjugating them to forced displacement, scorched
earth, massacres, and systematic rape (Bannon et al. 2005, 3).
The ethnic targeting of the campaigns as well as their scope have led, if not to
unanimity, then at least to a broad consensus among scholars that the events in
Darfur can be considered a genocide (Bannon et al. 2005, 6; Morse 2005; Jacobs
2008, 45; Kelly 2011, 417–19). Forcese (2011, 42), arguing for a case of correla-
tion, if not causality, between business and genocide in Darfur, describes how
“Islamization of Christian and animist regions of Southern Sudan” went hand-
in-hand with the “development” of the oil fields and quotes a former employee
of an international oil company saying that “when oil profits start flowing into
government hands, Christians in the South of Sudan will be largely eliminated
within two years.”
Considering business complicity, the main attention is on the Canadian

companies Talisman Energy Inc. and Arakis Energy Corporation (Rieth and
Zimmer 2004, 1; Morse 2005; Watts 2005, 390; Jacobs 2008; Huisman and van
Sliedregt 2010; Forcese 2011; Kelly 2011, 428; Chella 2012, 6), the Swedish com-
pany Lundin (Morse 2005; Batruch 2010; Clarke 2012), and the China National
Petroleum Corporation (CNPC) that works through its subsidiary PetroChina
and its Sudanese partner, the Greater Nile Petroleum Operating Company
(GNPOC) (Morse 2005; Jacobs 2008; Kelly 2011).6 These businesses have been
associated with the ethnic cleansing of their oil extraction areas in southern
Sudan as well as the “resultant extrajudicial killings, torture, rape and physical
destruction of civilian homes” (Kaeb 2008, 342).
Most commentators describe this business involvement in terms of direct
complicity that materializes through funding genocide by means of providing
revenue to the Sudanese government (Bannon et al. 2005, 9; Forcese 2011, 41;
Kelly 2011, 417). Direct complicity was also apparent in the provision by compa-
nies of financial, logistical, material, or infrastructural assistance (Forcese 2011,
37; see also Morse 2005; Jacobs 2008, 46). Forcese (2011, 41–42) documents how
Lundin “serves broken military trucks, provides electricity lines to their barracks
and even pipes water to the army camps” and how “airfields and roads built, used
and sometimes operated by the oil company have been employed by the Sudanese
military in attack against civilian population.” The CNPC also had “contracts
with Khartoum to secure their oil operations and allow Sudanese military forces
to use the companies’ air strips, landing pads and mechanical support” (Kelly
2011, 422). Such direct complicity thus is closely related to businesses’ (self-cho-
sen) dependence on state (-sponsored) troops for their security (Morse 2005;
Forcese 2011, 37).
With regard to the involvement of Chinese companies in the genocide in
Darfur, Wettstein (2010, 33) identifies a form of indirect complicity as well.
He argues that the multinational companies that made money from the Beijing
Olympics in 2008 implicitly endorsed and even further enabled the genocide in
Darfur, which is crucially facilitated by CNPC. Bannon et al. (2005, 15–16) make
a similar case for the companies now operating in Sudan and suggest, “the pres-
ence of these major energy companies also arguably legitimizes the government
of Sudan and, in turn, its acts in Darfur” (see also Popper 2007, 27). While an
exploration of the repercussions of such an argument (for instance, regarding the
indirect complicity of European companies active in the United States’ war-on-
terror-related atrocities) is beyond the scope of this chapter, the inevitable nor-
malization and hence social acceptance of atrocities that follows from a lack of
denouncement of associated companies seems intuitively evident.
Business complicity in the Darfur genocide concerns all phases of the geno-
cidal process. First, much of the forced displacement and accompanying atrocities
can be assumed to be implemented on behalf of businesses. Had these businesses
not entered the scene, the genocide would have been much less likely to have
occurred—indeed, before the discovery of oil, violent conflict in the region never
took on a genocidal nature and scale. Forcese (2011, 37) argues that the main rea-
son for the Sudanese government to abuse human rights was to “supply resources
to a company.” Second, as described above, businesses facilitated and enabled
the perpetrators during the genocidal acts. Third, there are indications that busi-
nesses have also been involved in attempts to downplay the gravity of the events
and deny their genocidal nature (Bannon et al. 2005; Kelly 2011; also see Batruch
2010 for an example in the case of Lundin).
As with the Kurdish genocide, commentators on the Darfur genocide seem
relatively uninterested in the motivations for businesses’ complicity. Or, rather,
they assume that the usual motivation of self-interest explains business behavior
in the face of genocide (Forcese 2011, 37). However, the context of the extrac-
tive industry in which the Darfur genocide must be placed encourages analysts
to slightly nuance this pure profit perspective. Jacobs (2008, 44), for instance,
suggests that for CNPC the main driver for involvement in Darfur is not so much
profit, but resource security: securing a particular resource or raw material as
a public good. Here, the distinction between private and public companies and
potential ties between companies and governments might be particularly rele-
vant (van Baar and Huisman 2012, 11). Possible reasons for companies to end
their engagement in the Darfur genocide are predominantly sought in the eco-
nomic disadvantages of a genocide stigma (Kelly 2011, 415). Yet, as Jacobs (2008,
49) warns us, only companies with an embedded human rights consciousness
are susceptible to naming and shaming. State-owned enterprises with a “frontier
human rights” mentality that seek resource security are not. As with the Kurdish
genocide, then, the ways in which businesses are involved in the Darfur geno-
cide is relatively well documented, but the internal considerations and decision-
making concerns that have led companies to this path and have guided them on it
remain underresearched.
25.3.3. The Holocaust

The Holocaust, during which the Nazis exterminated six million European Jews
from 1941 to 1945, does not require any introduction or elucidation here. The pri-
vate sector played a significant role in the implementation of Hitler’s envisioned
extermination of the Jewish people. Strategic corporations, so-called W-Betriebe
(Wehrbetriebe, or defense enterprises) important for the war effort, enjoyed priv-
ileges such as the use of forced labor and the acquisition of confiscated Jewish
businesses (Stallbaumer 1999; Wiesen 1999; van Baar and Huisman 2012, 8). The
post–World War II Nuremberg trials are still a benchmark in today’s jurisdiction
considering business complicity in GMAs (Jacobson 2005, 170; Skinner 2008;
Huisman and van Sliedregt 2010, 80; Tripathi 2010, 138; Kelly 2012, 339, 355).
In particular, four businesses were held accountable for their share in the
Jewish genocide in the trials known as the Farben Case, the Krupp Case, the
Flick Case, and, most infamous, the Zyklon B Case (Jacobson 2005; Baars 2006,
115–16; Skinner 2008). The Farben, Krupp, and Flick cases concerned German
industrialists—the former a chemical and pharmaceutical business, the lat-
ter two steel and coal businesses—t hat facilitated Hitler’s war efforts, which in
turn enabled the genocide (Stallbaumer 1999; Wiesen 1999; Stephens 2002, 34;
Jacobson 2005, 178, 186, 189; Slim 2012, 912). These businesses were charged
with “participating in wars of aggression, in enslavement, in plunder and spolia-
tion of property, and in a common plan or conspiracy to commit crimes against
peace” (Jacobson 2005, 186).
Several authors also highlight the role of businesses in other sectors, such as
those trading the gold stolen from the Nazi’s victims (Hayes 1998, 1; Wiesen
1999; Kyriakakis 2012, 993), and the German banking and insurance sector
(Hayes 1998, 1; van Baar and Huisman 2012, 3). Van Baar and Huisman (2012)
present a case study about the German corporation Topf und Söhne that built
cremation ovens and ventilation systems for the gas chambers in Nazi extermi-
nation camps. Thus, while several cases will be discussed more in-depth here,
ultimately, “by 1943, almost every major private firm in Germany was among
the exploiters” (Hayes 1998, 2). They all had “profited from the use of forced and
slave labor, the ‘Aryanization’ of Jewish property, and the plundering of compa-
nies in Nazi-occupied Europe” (Wiesen 1999; see also Black 1984; Hayes 1998;
and Stallbaumer 1999).
The roles these businesses played in the Holocaust falls into the category of
direct complicity: The businesses were used as “instruments of economic mobili-
zation for war” by the main perpetrator (Jacobson 2005, 171). Where the Farben,
Krupp, and Flick industrialists were held responsible for the broader war effort,
that is, war crimes, Bruno Tesch’s company, the producer of Zyklon B, was con-
sidered complicit in genocide, as it had directly delivered the main resource used
to execute this genocide.
Several studies also document the involvement of the private sector in the pre-
genocidal stage—for instance regarding the “Aryanization” of Jewish companies
(Hayes 1998)—and the postgenocidal stage, particularly the peculiar initial lack
of justification and legitimization and the subsequent denial of criminal behav-
ior (van Baar and Huisman 2012, 7). With reference to their case study of Topf
und Söhne, van Baar and Huisman (2012, 13) note “the need for justification of
the oven deliveries and ventilation systems to the extermination camps did not
occur until the end of the war.” They conclude that, for a long time, “in the under-
standing of those involved, there was possibly no criminal behavior that needed
to be neutralized; the behavior was, presumably, not considered a crime.” Wiesen
(1999), however, stresses that this attitude was reversed from the 1960s onward
when companies started to go to great lengths to generate sympathetic corporate
histories and to exonerate themselves from accusations of involvement in Nazi

crimes.
With regard to the Holocaust, the main interest of businesses to get engaged
in genocide is considered to be, again, to do what businesses do best: make
profit. “Farben acted to enrich itself as part of a general plan to dominate
the industries involved” (Jacobson 2005, 183). According to Stallbaumer
(1999, 3), Flick was moved primarily by business as well, his interest being
based on corporate expansion plans and an urge to protect market share.
Hayes (1998, 18) agrees that business self-i nterest was the overriding motive,
but argues that such business self-i nterest often concerned survival rather than
profit maximization. He states that “the point is not, as is so often assumed,
that German corporations grew rich through participation in the Holocaust,
since in general, they did not. The point is that they took what they could get
and became part of it anyway” (Hayes 1998, 2).
As such, ideological motivations for business complicity in the Holocaust are
often refuted in the literature. Germany’s businesses did not “march together
like infantry men, into the conference room to do battle with the Jews” (Hilberg
1962, as quoted in Stallbaumer 1999, 2) and were not fanatical Nazis (van Baar
and Huisman 2012, 6–7; see also Stallbaumer 1999). Scholars on the Holocaust
have also identified more multifaceted drivers of business complicity in geno-
cide, apart from the profit motive or nationalistic zeal. Van Baar and Huisman’s
(2012) article dedicated to the German business Topf und Söhne, for instance,
pays attention to intracompany dynamics that might generate business com-
plicity, such as competition among subunits, corporate culture, and normaliza-
tion and neutralization. They suggest that “while at I. G. Farben, competition
between corporations was an important explanation for their involvement in the
Holocaust, at Topf, it was mostly competition within the corporation that explains
their involvement” (van Baar and Huisman 2012, 10; also see chapter 14 in this
volume). Van Baar and Huisman (2012, 9) show in great detail how professional
zeal, more than moral or profit considerations, determined the company’s drive
to participate, and participate with dedication, in the genocide perpetrated on
the Jews. “Striving for innovation and technical perfection seems to be the most
prominent motivation for Topf und Söhne,” and it was the company, rather than
the SS, that took the initiative to “improve” the ovens and ventilation systems
(van Baar and Huisman 2012, 9–10). This reading resonates with the broader
literature on business culture and the general nature of bureaucracies where
the “bureaucratic nature of organizations and the emphasis on goal attainment,
along with the division of labour [can] make it unnecessary for employees to
think beyond what they are told to do, causing managers to be solely concerned
with reaching targets, and victimization disappears behind cost and benefit
analysis” (van Baar and Huisman 2012, 2; see also Brants 2007, 315; Kaleck and
Saage-Maaß 2010).
25.4. Discussion and Analysis

Combining the main tenets of the literature on corporate complicity in genocide
with the case-specific observations presented above yields several insights that
can help in further conceptualizing the “what,” the “how,” and the “why” of busi-
nesses’ involvement in genocides.
25.4.1. What?
In asking the question what type of business tends to be involved in genocide,
both the general literature and scholars working on our three cases are especially
concerned with two dimensions: size and sector. Academics have mostly been
interested in big businesses’ complicity in genocide of multinationals (Chella
2012, 22–2 4; Voillat 2012, 1072). In the sources about the three genocides dis-
cussed above, as well, relatively little is said about the role of national businesses
or local entrepreneurs in genocidal processes. This is remarkable because much
of the corporate complicity in human rights violations involves what are essen-
tially small business networks, many very informal and even illegal (Sherman
2001, 6). As van Baar and Huisman (2012, 8) suggest, there are important dif-
ferences between family enterprises and multinational companies when it comes
to motivations to get involved in genocide. With regard to the Holocaust, for
instance, Hayes (1998, 5) is convinced that “[e]‌nvy and greed … found their
home to a much greater extent among … participants in those middle ranges
of economic life where Jews remained conspicuous as competitors and middle-
men, that is to say, among mostly self-employed shopkeepers, artisans, peasant
proprietors, and professionals, especially medical doctors, than in big businesses
and corporations.”
Moreover, a company’s size importantly affects its governance and investment
structure (partnership or subcontractor; sole operator or consortium), includ-
ing its relations with the host government (often the genocide perpetrator) that
shapes the ways companies might be involved in genocides (Sherman 2001, 5).
This last issue is also stressed by Watts (2008, 10), who, with reference to firms
operating in the extractive industries, points out that most African govern-
ments collaborate with multinational resource enterprises through state-owned
enterprises, particularly in the oil industry.
The organizational structures of businesses operating in areas where geno-
cides are likely to occur are often purposefully vague (Stephens 2002, 54; Watts
2005, 387; Guidolin and La Ferrara 2007, 1980). This exacerbates problems of
corporate accountability and the legal challenges to distinguish between individ-
ual directors and managers and the company (Martin-Ortega 2008, 273; Chella
2012, 80–82). Stephens’s (2002, 60) conviction that “whatever it is, it can be held
accountable,” may be too optimistic.
Apart from overlooking the role of smaller and more localized companies, the
literature tends to prioritize certain sectors over others in discussing corporate
complicity in genocide. For instance, as exemplified in the Darfur genocide dis-
cussed, there has been extensive attention paid to oil companies in Africa as well
as Colombia, Ecuador, India, and Indonesia (Huisman and van Sliedregt 2010;
Martin-Ortega 2008, 275; Sherman 2001, 8; Stephens 2002, 51–53; Rieth and
Zimmer 2004, 2; Watts 2005, 390–4 01); for corporations mining for diamonds
and gold in Angola, Sierra Leone, and the Democratic Republic of Congo (DRC)
(Cooper 2002, 936; Guidolin and La Ferrara 2007; Huisman and van Sliedregt
2010, 804; Martin-Ortega 2008, 275; Tripathi 2010, 135); for timber busi-
nesses in Liberia (Cooper 2002, 949; Sherman 2001, 6); and for coltan miners
in the DRC (Cooper 2002, 951). Indeed, due to the specific characteristics of the
extractive industries—asset specificity, long production cycles, valuable conces-
sion agreements—t hey are much less likely to disengage from conflict zones and
hence more probable to become involved in genocide (Sherman 2001, 2).7
For similar reasons, the “lords of war” trading and selling the weapons with
which genocides are executed are also well covered by the literature, for instance
via the emblematic van Kouwenhove (Chella 2012, 1; Huisman and van Sliedregt
2010, 805; Kaleck and Saage-Maaß 2010, 708; Tripathi 2010, 139; Wennmann
2012, 913) and van Anraat (Huisman and van Sliedregt 2010, 805; Tripathi 2010,
132; van der Wilt 2006, 2) cases. The former was accused of trading with a geno-
cidal regime in Liberia and Sierra Leone, and the latter provided Saddam Hussein
with the weapons to execute the genocide on Iraq’s Kurds. The role of mercenar-
ies and private armies and security companies, often operating in the wake of
multinationals active in the extractive industries, is also broadly discussed (Bray
and Crockett 2012, 1072; Guidolin and La Ferrara 2007, 1980; Huisman and van
Sliedregt 2010, 805; Kaeb 2009, 327; Percy 2012; Ramasastry and Thompson
2006, 17; Sherman 2001, 6; Watts 2005, 378).
Remarkably, while in the empirical literature about the Jewish and Kurdish
genocide the focus has been on manufacturing companies, much less attention is
being paid to the productive and service sectors. Considering that the main role
that companies are seen to play in genocides is that of financing, it is also remark-
able that there is not more attention to the role of the banking sector in GMAs
such as displayed in the work of Bohoslavsky and Opgenhaffen (2010), which
touches on banks’ involvement in mass atrocities in Argentina, Colombia, and
the Holocaust; and Weiss and Shamir’s (2012, 167) discussion of the complicity
of Israeli banks in Israel’s mass killings in the Gaza Strip.
Based on the above, we suggest that more attention is due on the role of smaller
firms and family enterprises in genocide to counterbalance the current obses-
sion with multinationals in this context. Perhaps even more important, the roles
of domestic and foreign firms need to be juxtaposed more systematically, so as
to conceptualize the differences that were evident in the cases discussed here
between, for example, German companies in the context of the Holocaust and
German companies operating in Iraq under Saddam Hussein. This might also
provide more handles to study the different, and often purposively ambiguous,
institutional ties that firms have with (host and home) governments.
25.4.2. How?
Business involvement is a feature of all genocides. The cases outlined in this
chapter, however, confirm the literature’s emphasis on complicity (as opposed
to perpetration). We illustrated that in genocides, business complicity is both
direct and indirect (or “silent”). Direct complicity, for instance through provid-
ing weapons (in the Kurdish case) and/or providing facilities and finances (in the
Darfur and Holocaust cases), reflects that businesses were aware of, or could and
hence should reasonably have known about, ongoing genocide. Indirect complic-
ity, such as described in the case of the Darfur genocide, mostly is apparent in the
legitimizing and normalizing of business complicity in the genocide.
The cases also reveal a nuance not regularly noted in the literature regarding
the stages of the genocidal process. Here, we see variation across the cases. In
the Kurdish case, the companies providing the chemicals needed for Saddam
Hussein’s extermination of the Kurds helped prepare and enable the genocide,
but played no role in the actual execution. In Darfur, businesses were complicit
in both the pregenocide phase—providing the incentive for genocidal activi-
ties and enabling them through financing the main perpetrator—and, in some
cases, during the genocide—actively supporting and facilitating genocidal cam-
paigns. In the Holocaust, German businesses normalized genocidal tendencies in
the phase leading up to the Endlösung (the “final solution”) and were active imple-
menters and facilitators during the genocide. Where analyses of the Kurdish and
Darfur genocides pay little attention to the postgenocide phase, moreover, studies
of business complicity in the Holocaust do explore the contribution of businesses
to after-t he-fact normalization and rationalization processes.
25.4.3. Why?
We particularly noted an important difference between the cases regarding the
analysis of the motives for business complicity—a topic that is relatively underap-
preciated in the literature we reviewed. In the literature, and also with regard to
the Kurdish and Darfur genocides, there is ample attention to the “how,” but the
“why” of business complicity is assumed rather than critically explored. The pur-
suit of profit is considered the default driver of all business behavior, no matter the
circumstances. The lack of insights about why businesses are involved in genocide
in the first place and why, subsequently, they opt for specific forms and degrees
of complicity is partly a result of the fact that the perspective of the businesses in
question is not solicited—and where it is, is hardly received. While it might for
various reasons be difficult to get to internal business motivations and reflections,
these nevertheless seem essential to take the next step in understanding how and
why businesses enable genocide—and potentially limiting business complicity in
genocide (Brants 2007).
The literature on corporate complicity in the Holocaust provides a useful
starting point. First of all, the more in-depth discussions of specific compa-
nies’ involvement in that genocide reveals that there is an important difference
between companies actively taking the initiative vis-à-v is genocide perpetra-
tors and those more passively meeting the perpetrators’ requests or orders—a n
insight closely related to the distinction between domestic and foreign compa-
nies’ involvement in genocides. The case of the Holocaust also suggests that the
more historic the genocide in question, the more likely the theme of business
motivation will be picked up—not least because the (legal, socioeconomic, and
hence economic) consequences for businesses will diminish and their willing-
ness to reflect on their previous conduct presumably increases. In addition, what
sets the overall analysis of business complicity in the Holocaust apart from other
cases of business complicity in genocide is the contributions of criminologists to
the substantial legal approach to the topic, which has linked business complicity
in genocide to studies on other corporate crimes. As argued in section 25.3, this
allows for more nuanced institutional analyses of drivers for corporate decision-
making in genocidal contexts (Brants 2007, 309–10; van Baar and Huisman
2012, 4).
The overall emphasis on the conduct of individual persons rather than the oper-
ation of businesses as organizations or institutions might also be the result of this
disciplinary bias toward the legal studies in which the discussion on the liability of
companies seems to overshadow the potential responsibility of businesses beyond
their individual leader or otherwise responsible person (Brants 2007, 321). This
bias, however, might have political roots as well. Wiesen (1999) writes: “During
the trials at Nuremberg, the American prosecutors were careful not to portray
the proceedings as attacks on the market economy, but rather as attempts to pun-
ish individuals who had committed crimes.” This was despite the fact that “while
individuals were nominally on trial, the Krupp Company itself, acting through
its employees, violated international law” (Skinner 2008, 345). Emphasizing the
agency of individuals, according to Wiesen (1999), was for a long time convenient
to keep analyses away from seeing the behavior of businesses as a result of capital-
ist structures, a hotly contested issue during the cold war when “Marxists wanted
to prove and anti-Marxists wanted to refute the claims that capitalism and fas-
cism were linked.” In the post–cold war era, however, a more criminological—or
generally political—perspective would be in order. More attention to the motives
driving business complicity and a reconsideration of the dialectic between struc-
ture and agency in the question of business complicity in genocide might also
allow for a less rigidly legal understanding of the relations between perpetrators
(usually state actors) and accomplices (often companies) in genocide.
This chapter has been explicitly explorative and has been concerned with locat-
ing gaps in the conceptual and empirical literature on corporate complicity in
genocide. We did so by means of three concrete case illustrations. Nevertheless,
we do have the ambition to provide explicit conceptual handles for further
research into this topic. We venture some proposals in the section that follows.

We started this chapter mentioning that it may, on the face of it, not be logical for
businesses to become engaged in genocide because it is bad for their businesses
and their reputations. However, as shown throughout this chapter, reality is more
complex. Businesses, as we have illustrated, hardly ever commit or perpetrate geno-
cide, because their role generally consists of complicity. This also means that their
motivation need not be positive or active, that is, to commit genocide and destroy
a group of people—which would indeed be a rather peculiar businesses objective.
As an accomplice, businesses merely need to have the negative or passive incen-
tive to not intervene or disengage even while knowing what is going on.
To better understand therefore the complicity of business in GMA, first and
foremost, we see a need for a shift from diagnostic attention on how businesses
are engaged in genocide to a more analytical exploration of why businesses have
made the choices they did in the process of their engagement with genocide. This
is also necessary to advance the debate on how to hold businesses accountable for
gross human rights violations and moreover to provide incentives for businesses
not only to avoid doing harm but also to proactively, preventively strive to pro-
tect and extend human rights (see, e.g., Wettstein 2010). To move the explanatory
discussion beyond the platitudes of profit maximization and loss minimization,
however, a more case-specific analysis is required. For instance, as we illustrated,
the size, domestic or foreign background, and sector will shape the institutional
structures governing a company and its relation with genocide-perpetrating
governments.
In light of this, what should and can we ask from businesses? There are ample
guidelines for businesses to avoid complicity in human rights violations and geno-
cide, culminating in the Norms on the Responsibilities of Transnational Corporations
and Other Business Enterprises with Regard to Human Rights (2003)8 and the United
Nations Guiding Principles on Business and Human Rights (2011).9 In terms of the
former, as Oetzel, Getz, and Ladek (2007, 14–15) summarize, multinational
enterprises can (and should) leave, withhold payments that might support vio-
lence, publicly condemn violence, provide support to peacekeeping missions,
lobby the government, and explicitly support the rule of law. In terms of the latter
(the “Ruggie Framework”), the implementation of a set of human rights due-d ili-
gence instruments are promoted. Thus, Econsense (2014) for instance lists more
than 120 tools and resources to support such a human rights due diligence.
Rieth and Zimmer (2004, 9) stress that even without such explicit policies
and tools, businesses are a major contributor to the prevention of genocides if they
strengthen equitable and inclusive economies and build human capital (see also
Bray and Crockett 2012, 1070). Complementing liability with a culture more
geared toward responsibility could further open up the debate on what busi-
nesses can and should do in the face of genocidal events and move away from
only criticizing business for what they should not do. Yusuf (2008, 99) suggests a
fundamental rethinking of the role of the private sector in society and its position
vis-à-v is the public sector—a suggestion that, in fact, goes far beyond the theme
of business complicity in genocide. He argues that
[i]‌t can be posited that the social contract between the individual and
the state dictates that concessions (“sub-contracts” in contract theory)
made by the latter operate to bind the privy (“sub-contractor”) to the
head contract to the extent of the expected impact of such operations. In
that way, licensees and concessionaires (such as MNCs [multinational
corporations]) can be made to take on some of the obligations of their
principal (the state) and become bound to fulfil them based on the doc-
trine of privity of contract. They could then become substantially bound
to perform some of the important “terms” of the contract between state
and society.
In other words, economic contracts would include a default sociopolitical

clause, which would mean that in liaising with governments, companies would be
forced to take on some of these governments’ accountability vis-à-v is a country’s
citizens.
A perhaps more concrete potential of focusing on responsibility instead
of liability or only avoiding doing harm is that it takes seriously the options of
businesses in genocidal settings beyond the usual panacea of disengagement.
Businesses have long argued that withdrawal has human security costs as well
(Manby 2000, 7; Kelly 2011, 433; Weiss and Shamir 2012, 160). Popper (2007,
28) writes that, in the case of an international engineering company, “all stake-
holders advised … against withdrawing from the country. To do so, they said,
would undermine the fragile reconstruction process and deny the country much-
needed infrastructure.” Weiss and Shamir (2012) discuss the case of Israeli utility
service providers in Gaza that were pressured by the Israeli government to stop
their provision as part of the government’s sanction regime. In this case, scholars
argued, disengagement rather than continuation of involvement would constitute
a human rights violation.
Ultimately, however, to be able to help businesses steer away from complicity

in the future and toward promotion of human rights, we first need to better under-
stand what drove them toward complicity in the past. In this chapter, we have
demonstrated that further research on business complicity in genocide would
do well to explore the “why” of such complicity rather than merely the “what” or
“how.” It is an understanding of the motivations and considerations of compa-
nies, more than a technical documentation of their behavior, which will yield the
insights needed to work toward prevention.
Acknowledgments
We are grateful to Charles H. Anderton, Jurgen Brauer, and four anonymous
reviewers for their comments on earlier versions of this chapter. The usual dis-
claimer applies.
Notes
1. Several examples serve to illustrate the apparent stakes involved for businesses in avoid-
ing the specific label of genocide. Oil companies such as Shell and Chevron in Nigeria
do not deny mass atrocities take place, but try to stop such atrocities from being defined
as genocide (Kaleck and Saage-M aaß 2010, 704–5) as has been done by authors, such as
Manby (2000), Watts (2008, 13–15), and Yusuf (2008, 80), who frame the developments
in Nigeria’s Ogoni region as cultural genocide. The involvement of gold mining corpora-
tions such as Goldcorp, Nichromet Extractions, Hudbay Minerals, and BHP Billiton in
Guatemala touches on similar, if less well-k nown contestations (van de Sandt 2009; Nolin
and Stephens 2010, 51; Hurtado 2013). In Palestine, several cases have been made against
companies facilitating Israeli occupation (Baars 2006; Skinner 2008, 321; War on Want
2006). The companies in question have made extensive efforts to neutralize the implicit
accusations of political genocide (politicide) in these cases (Moffatt 2009; War on Want
2006, 3; Baars 2006, 128).
2. It should be noted that these cases feature as illustrations and are not full-fledged case studies
as the literature review underlying them is far from exhaustive.
3. Khan (2005, 7) similarly distinguished between first-order involvement (“actively helped to
design and implement”), second-order involvement (knowing “their products would be used
for repression”), and third-order involvement (“benefited indirectly”).
4. In his article, Kelly (2013, 375–79) provides a full list of companies. The German companies
he registers are: Herberger Bau; Karl Kolb; Ludwig-Hammer; Ceilcote; Klockner Industry; Schott
Glass; Preussag; Reininghaus-Chemie; Tafisa; Weco; Martin Merkel; Lewa Hebert.
5. Figures are contested. Morse (2005) mentions two million dead and four million displaced.
6. Bannon et al. (2005, 18–19) provides a comprehensive list of oil companies implicated in
the Darfur genocide: Al-Th ani Investment (UAE), Lundin Petroleum (Sweden), Marathon
(USA), Nam Fatt (Malaysia), ONGC (India), PECD Berhad (Malaysia), PetroChina
(China), Sinopec (China), Tatneft (Russia), Total Elf Fin (France), Vangold Resources Ltd.
(Canada), Videocon (India), White Nile (UK).
7. Even if disengagement has taken place in some cases, for instance in response to campaigns
against blood diamonds.
8. http://w ww1.umn.edu/humanrts/business/norms-Aug2003.html.
9. http://w ww.ohchr.org/Documents/P ublications/GuidingPrinciplesBusinessHR_ EN.pdf.
References
Amunwa, B. 2012. “Fueling the Violence: Oil Companies and Armed Militancy in Nigeria.”
http:// p latformlondon.org/ w p- c ontent/ u ploads/ 2 012/ 0 8/ F uelling- t he- v iolence-
Oil- C ompanies-a nd-A rmed-M ilitancy-i n-N igeria-A ugust-2 012.pdf [accessed March
29, 2015].
Baars, B. 2006. “Corrie et al v. Caterpillar: Litigating Corporate Complicity in Israeli Violations
of International Law in U.S. Courts.” In M. Lau and F. Nasrallah, eds., Yearbook of Islamic
and Middle Eastern Law. Vol. 12. Leiden, The Netherlands: Brill, 97–134.
Bannon, A., A. Cover, N. Ghoshal, D. Marcuse, N. Robinson, and S. Shivaram. 2005. “An
Analysis of Select Companies’ Operations in Sudan: A Resource for Divestment.” The
Allard K. Lowenstein International Human Rights Project Yale Law School. New Haven,
CT: Yale University. http://acir.yale.edu/pdf/ YaleLowensteinSudanReport.pdf [accessed
May 9, 2015].
Batruch, C. 2010. “Lundin Petroleum AB’s Experience in East Africa: The Role of the Private
Sector in Conflict Prone Countries.” Economics of Peace and Security Journal 5, no. 2: 5–9.
Baumol, W. J. 1990. “Entrepreneurship: Productive, Unproductive and Destructive.” Journal of
Bernard, V. 2012. “Globalisation Will Only Mean Progress If it Is Responsible.” International
Review of the Red Cross 94, no. 887: 881–91.
Black, E. 1984. The Transfer Agreement: The Dramatic Story of the Pact between the Third Reich and
Jewish Palestine. Washington, DC: Dialog Press.
Black, E. 2001. IBM and the Holocaust: The Strategic Alliance between Nazi Germany and America’s
Most Powerful Corporation. New York: Crown Books.
Bohoslavsky, J. P., and V. Opgenhaffen. 2010. “The Past and Present of Corporate Complicity:
Financing the Argentinean Dictatorship.” Harvard Human Rights Journal 23, no. 1:
157–2 03.
Brants, C. 2007. “Gold Collar Crime. The Peculiar Complexities and Ambiguities of War Crimes,
Crimes Against Humanity and Genocide.” In H. N. Pontell and G. Geis, eds., International
Handbook of White-Collar and Corporate Crime. New York: Springer, 309–2 6.
Bray, J., and A. Crockett. 2012. “Responsible Risk-Taking in Conflict-A ffected Countries: The
Need for due Diligence and the Importance of Collective Approaches.” International Review
of the Red Cross 94, no. 887: 1069–89.
Brück, T., W. Naudé, and P. Verwimp. 2013. “Business under Fire: Entrepreneurship and Violent
Conflict in Developing Countries.” Journal of Conflict Resolution 57, no. 1: 3–19.
Chella, J. 2012. “The Complicity of Multinational Corporations in International Crimes: An
Examination of Principles.” PhD dissertation, Bond University, Robina, Queensland,
Australia.
Clapham, A., and S. Jerbi. 2001. “Categories of Corporate Complicity in Human Rights Abuses.”
Hastings International and Comparative Law Review 24: 339–49.
Clarke, P. 2012. “Oil Money Paid for Sudan’s Genocide.” Afton Bladet, April 14. http://w ww.
aftonbladet.se/debatt/a rticle14677302.ab [English version, e.g., at http://w ww.eco-
sonline.org/news/2 012/2 0120414_O il_ Money _ Paid_ f or_ S udan%27s_G enocide/]
Cooper, N. 2002. “State Collapse as Business: The Role of Conflict Trade and the Emerging
Control Agenda.” Development and Change 33, no. 5: 935–55.
Dolamari, M. 2014. “Kurdish People Want German Companies to be Put on Trial.” BasNews,
March 13. http:// w ww.basnews.com/e n/ n ews/ 2 014/ 0 3/ 1 3/ k urdish- p eople- w ant-
german- companies-to-be-put-on-t rial/ [accessed May 9, 2015].
Econsense. 2014. Respecting Human Rights: Tools and Guidance Materials for Business. Berlin: Forum
for Sustainable Development for German Business. www.econsense.de/sites/a ll/fi les/
Respecting_Human_R ights.pdf [accessed May 9, 2015].
Forcese, D. 2011. “‘Militarized Oil’ in Sudan’s Oilfields: Lessons for Canadian Foreign Policy.”
Canadian Foreign Policy Journal 8, no. 3: 37–56.
Guidolin, M., and E. La Ferrara. 2007. “Diamonds Are Forever, Wars Are Not: Is Conflict Bad
for Private Firms?” American Economic Review 97, no. 5: 1978–93.
Hayes, P. 1998. “Profits and Persecution: German Big Business and the Holocaust.” J. B. and
Maurice C. Shapiro Annual Lecture, Washington, DC, February 17. http://w ww.ushmm.
org/m/pdfs/2 0050803-hayes.pdf [accessed May 12, 2015].
Hippler, J. 1991. “Iraq’s Military Power: The German Connection.” Middle East Report 21, no.
168. http://w ww.merip.org/m er/m er168/i raqs-m ilitary-p ower- g erman- c onnection
[HRW] Human Rights Watch. 1993. “Genocide in Iraq: the Anfal Campaign against the Kurds.”
New York: Human Rights Watch.
Huisman, W., and E. van Sliedregt. 2010. “Rogue Traders: Dutch Businessmen, International
Crimes and Corporate Complicity.” Journal of International Criminal Justice 8, no. 3: 803–2 8.
Hurtado, L. 2013. “Despite Historic Conviction, Genocide Continues in Guatemala.” Food
First Website. May 16. http://upsidedownworld.org/main/g uatemala-a rchives-33/4300-
despite-h istoric-conviction-genocide-continues-i n-g uatemala?format=pdf [accessed May
12, 2015].
Jacobs, L. A. 2008. “Differentiated Corporate Legal Consciousness in International Human
Rights Disputes: Security and Transnational Oil Companies in Sudan.” Asia Pacific Program
on Dispute Resolution Research Notes 1, no. 3: 37–49.
Jacobson, K. R. 2005. “Doing Business with the Devil: The Challenges of Prosecuting Corporate
Officials Whose Business Transactions Facilitate War Crimes and Crimes against
Humanity.” Air Force Law Review 56: 167–233.
Kaeb, C. 2008. “Emerging Issues of Human Rights Responsibility in the Extractive and
Manufacturing Industries: Patterns and Liability Risks.” Northwestern Journal of International
Human Rights 6, no. 2: 327–53.
Kaleck, W., and M. Saage-Maaß. 2010. “Setting the Framework: Corporate Accountability for
Human Rights Violations Amounting to International Crimes: The Status Quo and Its
Challenges.” Journal of International Criminal Justice 8, no. 3: 699–724.
Kelly, M. J. 2011. “Ending Corporate Impunity for Genocide: The Case against China’s State
Owned Petroleum Company in Sudan.” Oregon Law Review 90, no. 2: 413–4 8.
Kelly, M. J. 2012. “Prosecuting Corporations for Genocide Under International Law.” Harvard
Law and Policy Review 6, no. 2: 339–67.
Kelly, M. J. 2013. “‘Never Again?’ German Chemical Corporation Complicity in the Kurdish
Genocide.” Berkeley Journal of International Law 31, no. 2: 348–91.
Khan, I. 2005. “Understanding Corporate Complicity: Extending the Notions Beyond Existing
Laws.” Business Human Rights Seminar, London. December 8. https://w ww.amnesty.
org/f r/documents/pol34/0 01/2 006/en/ [accessed May 12, 2015]
Kyriakakis, J. 2012. “Developments in International Criminal Law and the Case of Business
Involvement in International Crimes.” International Review of the Red Cross 94, no.
887: 981–1007.
Manby, B. 2000. “Shell in Nigeria: Corporate Social Responsibility and the Ogoni Crisis.”
New York: Carnegie Council on Ethics and International Affairs.
Martin-Ortega, O. 2008. “Business and Human Rights in Conflict.” Ethics and International
Affairs 22, no. 3: 273–83.
Milhollin, G., and K. Motz. 2003. “The Means to Make the Poisons Came from the West.”
New York Times. April 13. http://w ww.nytimes.com/2 003/0 4/13/weekinreview/t he-
world-t he-means-to-make-t he-poisons-came-f rom-t he-west.html [accessed May 9, 2015].
Moffatt, S. 2009. “Bil’in and Beyond—Prosecuting Corporate Complicity in War Crimes under
Canadian Law.” M.A. thesis, University of Toronto.
Morse, D. 2005. “How Oil Drives the Genocide in Sudan.” Sudan Tribune, August 18. Published
August 23. http://w ww.sudantribune.com/spip.php?article11203 [accessed May 9, 2015].
Naudé, W. 2009. “Entrepreneurship, Post-Conflict.” In T. Addison and T. Brück, eds., Making
Peace Work: The Challenges of Social and Economic Reconstruction. Basingstoke: Palgrave
Macmillan, 251–63.
Nolin, C., and J. Stephens. 2010. “‘We Have to Protect the Investors’: ‘Development’ and
Canadian Mining Companies in Guatemala.” Journal of Rural and Community Development
5, no. 3: 37–70.
Oetzel, J., K. Getz, and S. Ladek. 2007. “The Role of Multinational Enterprises in Responding
to Violent Conflict: A Conceptual Model and Framework for Research.” American Business
Law Journal 44, no. 2: 331–58.
Percy, S. 2012. “Regulating the Private Security Industry: A Story of Regulating the Last War.”
Popper, R. 2007. “ABB and Sudan: The Value of Stakeholder Engagement.” In Embedding Human
Rights in Business Practice II. New York: United Nations Global Compact and Office of the
U.N. Commissioner for Human Rights, 26–30.
Ramasastry, A., and R. C. Thompson. 2006. “Commerce, Crime and Conflict: Legal Remedies for
Private Sector Liability for Grave Breaches of International Law.” FAFO Report 365. Olso,
Norway: FAFO. http://w ww.fafo.no/i ndex.php?option=com_zoo&task=item&item_
id=3676&Itemid=927&lang=en [accessed May 9, 2015].
Rieth, L., and M. Zimmer. 2004. “Transnational Corporations and Conflict Prevention: The
Impact of Norms on Private Actors.” Center for International Relations, Law and Conflict
Studies. Tübingen, Germany: University of Tübingen.
Schabas, W. A. 2000. Genocide in International Law: The Crime Of Crimes. 1st ed. New York:
Shenon, P. 2002. “Declaration Lists Companies That Sold Chemicals to Iraq.” New York Times,
December21. http://w ww.nytimes.com/2 002/12/21/world/t hreats-responses-suppliers-
declaration-l ists-companies-t hat-sold-chemicals-i raq.html [accessed May 9, 2015].
Sherman, J. 2001. “Private Sector Actors in Zones of Conflict: Research Challenges and Policy
Responses.” International Peace Academy Workshop Report. New York: International
Peace Institute. http://w ww.ipinst.org/2 001/0 9/private-sector-actors-i n-zones-of-
conflict-research-challenges-a nd-policy-responses [accessed May 9, 2015].
Skinner, G. 2008. “Nuremberg’s Legacy Continues: The Nuremberg Trials’ Influence on Human
Rights Litigation in U.S. Courts under the Alien Tort Statute.” Albany Law Review 71, no.
1: 321–67.
Slim, H. 2012. “Business Actors in Armed Conflict: Towards a New Humanitarian Agenda.”
Stallbaumer, L. M. 1999. “Big Business and the Persecution of the Jews: The Flick Concern and
the ‘Aryanization’ of Jewish Property before the War.” Holocaust and Genocide Studies 13,
no. 1: 1–27.
Starcevic, N. 1990. “Many German Firms Helped Build Iraqi Arsenal.” AP News, October 29.
http://w ww.apnewsarchive.com/1990/ M any- G erman-F irms-H elped-B uild-I raq- s -
Arsenal/id-92b14b92d9adca36724138a6a8eefbc9 [accessed May 9, 2015].
Stephens, B. 2002. “The Amorality of Profit: Transnational Corporations and Human Rights.”
Berkeley Journal of International Law 20, no. 1: 45–9 0.
Stokes, P., and Y. Gabriel. 2010. “Engaging with Genocide: The Challenge for Organization and
Management Studies.” Organization 17, no. 4: 461–80.
Sweetman, P. 2009. Business, Conflict Resolution and Peace-Building: Contributions from the Private
Sector to Address Violent Conflict. Abingdon: Routledge.
Tripathi, S. 2010. “Business in Armed Conflict Zones: How to Avoid Complicity and Comply
with International Standards.” Politorbis 50, no. 3: 131–42.
Urwand, B. 2013. The Collaboration: Hollywood’s Pact with Hitler. Cambridge, MA: Belknap
Press.
van Baar, A., and W. Huisman. 2012. “The Oven Builders of the Holocaust: A Case Study of
Corporate Complicity in International Crimes.” British Journal of Criminology 52, no.
6: 1033–50.
van de Sandt, J. 2009. “Mining Conflicts and Indigenous People in Guatemala.” Cordaid and the
Amsterdam University Law Faculty.
van der Wilt, H. G. 2006. “Genocide, Complicity in Genocide and International v. Domestic
Jurisdiction: Reflections on the Van Anraad Case.” Journal of International Criminal Justice
4, no. 2: 239–57.
Voillat, C. 2012. “Pushing the Humanitarian Agenda through Engagement with Business
Actors: The ICRC’s Experience.” International Review of the Red Cross 94, no. 887:
1089–1115.
War on Want. 2006. “Profiting from the Occupation. Corporate Complicity in Israel’s Crimes
Against the Palestinian People.” http://waronwant.org/campaigns/w inning-justice-
for-palestine/h ide/i nform/12573-report-profiting-f rom-t he-occupation [accessed May
9, 2015].
Watts, M. J. 2005. “Righteous Oil? Human Rights, the Oil Complex, and Corporate Social
Responsibility.” Annual Review of Environmental Resources 30: 373–4 07.
Watts, M. J. 2008. “Blood Oil. The Anatomy of a Petro-I nsurgency in the Niger Delta, Nigeria.”
Economies of Violence Working Papers No. 22. Institute of International Studies.
University of California, Berkeley, CA.
Weiss, D., and R. Shamir. 2012. “Corporate Accountability to Human Rights: The Case of the
Gaza Strip.” Harvard Human Rights Journal 24, no. 1: 155–83.
Wennmann, A. 2009. “Getting Armed Groups to the Table: Peace Processes, the Political
Economy of Conflict and the Mediated State.” Third World Quarterly 30, no. 6: 1123–38.
Wennmann, A. 2012. “The Role of Business in Armed Violence Reduction and Prevention.”
International Review of the Red Cross 94, no. 887: 919–4 0.
Wettstein, F. 2010. “The Duty to Protect: Corporate Complicity, Political Responsibility and
Human Rights Advocacy.” Journal of Business Ethics 96, no. 1: 33–47.
Wiesen, S. J. 1999. “German Industry and the Third Reich: Fifty Years of Forgetting and
Remembering.” Dimensions: A Journal of Holocaust Studies 13, no. 2. http://a rchive.adl.org/
braun/d im_13_2 _forgetting.html#.VUvwaGZsCyM [accessed May 9, 2015].
Yusuf, H. O. 2008. “Oil on Troubled Waters: Multinational Corporations and Realising Human
Rights in the Developing World, with Specific Reference to Nigeria.” African Human Rights
Law Journal 8, no. 1: 79–107.
26
Valuing Lives You Might Save

Understanding Psychic Numbing in the Face of Genocide
Pau l Sl ov ic, Da n i e l Vä st f jä l l , Robi n Gr egory, a n d K i m be r ly
G. Ol son
26.1. Introduction
The twentieth century is often said to have been one of the bloodiest in recorded
history. In addition to its wars, it witnessed many grave and widespread episodes
of mass atrocities (see chapter 3 in this volume). But what stands out in historical
accounts of those atrocities, perhaps even more than the cruelty of their perpetra-
tion, is the inaction of bystanders. Why do people and their governments repeat-
edly fail to react to genocide and other mass-scale abuses of human beings?
There is no simple answer to this question. It is not because people are insen-
sitive to the suffering of their fellow human beings—w itness the extraordinary
efforts that society will expend to rescue a person in distress. It is not because peo-
ple only care about identifiable victims of similar skin color who live nearby: wit-
ness the outpouring of aid from the north to the victims of the December 2004
tsunami in Southeast Asia. Nor can the blame be apportioned entirely to spe-
cific political leaders. Although Presidents George W. Bush and Barack Obama
have been unresponsive to the murder of hundreds of thousands of people in
Darfur, it was President Bill Clinton who ignored the genocide in Rwanda, and
President Franklin D. Roosevelt who for too long did little to stop the Holocaust.
The American example of inaction has been repeated in other countries as well.
Behind every leader who ignored mass murder were millions of citizens whose
indifference allowed and tacitly supported the inaction.
Every episode of mass murder is different and raises distinct social, economic,
military, and political obstacles to intervention. We therefore recognize that geo-
politics, domestic politics, or failures of individual leadership have been impor-
tant factors in particular episodes. But the repetitiveness of such atrocities, which
too often have been ignored by powerful people and nations and by the general
613
public, calls for explanations that may reflect some fundamental deficiency in our
humanity—a deficiency that, once identified and understood, might possibly be
overcome.
In this chapter, we examine indifference and lack of response to genocide and
other mass atrocities with reference to one of the foundations of the relatively new
field of behavioral economics, prospect theory.
26.2. A Brief Introduction to Behavioral Economics

Economics is, at its base, concerned with human behavior. Economists have
developed sophisticated theoretical models to describe behavior. At least in its
early incarnations, and to some degree even today, the concept of homo economicus
played an important role in these theories, assuming that we can all be represented
by a prototypical “economic man” who is completely informed, infinitely sensi-
tive to economic fluctuations, and rational in the sense of having stable, orderly
preferences that maximize economic utility (for a discussion of the concept of
rationality in economics, see c hapters 1 and 6 in this volume).
For several decades economists and psychologists, along with other behav-
ioral scientists, have questioned this presumption of rationality that, accord-
ing to the influential social scientist, Herbert Simon, permitted economists to
make “strong predictions … about behavior without the painful necessity of
observing people” (Simon and Stedry 1969, 272). Simon, both an economist
and a psychologist, drew upon empirical research on human cognitive limita-
tions to challenge traditional assumptions about the motivation, omniscience,
and computational capacities of “economic man.” He introduced the notion
of bounded rationality, which asserts that cognitive limitations force people to
construct simplified models of how the world works in order to cope with it. To
predict behavior, “we must understand the way in which this simplified model
is constructed, and its construction will certainly be related to man’s psy-
chological properties as a perceiving, thinking, and learning animal” (Simon
1957, 198).
About the same time that Simon was documenting bounded rationality in
the 1960s and 1970s, another psychologist, Ward Edwards, began testing eco-
nomic theories with controlled laboratory experiments, often using choices
among simple gambles to examine how people process information central to
“life’s gambles”—gains and losses and their associated probabilities. Early studies
confirmed that people often violate the assumptions of economic rationality and
are guided in their choices by noneconomic factors. Thus we behave very differ-
ently depending on how information is presented to us, the nature of the deci-
sion-making environment, or what period of our life we are in. In many important
situations we do not really know what we prefer. Instead, we must construct our
Va l u i n g L i v e s Yo u M i g h t S a v e 615
preferences “on the spot” using the available cues and information (Lichtenstein
and Slovic 2006).
Stimulated by creative empirical studies by psychologists such as Amos
Tversky and Daniel Kahneman and economists such as Richard Thaler, what
began as a trickle of studies challenging traditional economic assumptions of
rationality became a torrent. Nobel Prizes in economics awarded to Herbert
Simon in 1978, to George Akerlof in 2001, and to Vernon Smith and Daniel
Kahneman in 2002 for their contributions toward understanding the behav-
ioral dynamics of economic decisions further contributed to what has become a
revolution in thinking called behavioral economics.
We now recognize that a question fundamental to economics—“Are people
rational or irrational?”—is ill-formed. As human beings, we have intuitive and
analytic thinking skills that work beautifully, most of the time, to help us navi-
gate through life and achieve our goals, individually and collectively. But, just as
our exquisite visual system can be misled by certain contextual cues, resulting in
visual illusions, our thinking skills also fail us at times.
The very modes of thought that are usually highly rational can also get
us into trouble when the nature of the environment surrounding us changes.
For example, the affective feelings, including fear, anxiety, love, trust, and
confidence, which help us assess risk and reward, are processed swiftly in our
minds. These feelings form the neural and psychological substrate of what
is important to us and guide many decisions—t hat is, what economists refer
to as utility. In this sense, reliance on feelings enables us to be rational actors
in many important situations. For instance, if you were to see a venomous
snake, you would not likely pause to calculate the mathematical utility of the
possible harmful consequences multiplied by their associated probability in
order to decide what to do. When you see the snake, you act rationally: you
move away fast.
Reliance on our gut feelings works beautifully when our experience enables
us to anticipate accurately how we will like the consequences of our decisions—
that is, when we have a good knowledge of the situation and fully understand
our reactions today and in the future. But it can fail miserably when the set-
ting is novel or when the consequences turn out to be much different in char-
acter than we expected. In this instance of surprise, the rational actor often
becomes, to borrow the words of 1998 Nobel Laureate Amartya Sen (1977), the
rational fool.
To answer our original question—“How rationally might we behave in this
new, uncertain, and more dangerous environment?”—our answer is: “It depends.”
The challenge before us now is to better understand when and how rationality
fails in this modern environment.
In the pages that follow, we examine what can be seen as a failure of rationality
in our intuitive responses to genocides and other, distant mass atrocities.
26.3. Background and Theory: Affect and Analysis

in Situations of Risk
Risk management in the modern world relies upon two forms of thinking (Slovic
et al. 2004; Kahneman 2011). Risk as feelings refers to our instinctive and intuitive
reactions to danger. Risk as analysis brings logic, reason, quantification, and delib-
eration to bear on hazard management. Compared to analysis, reliance on feel-
ings tends to be a quicker, easier, and more efficient way to navigate in a complex,
uncertain, and dangerous world. Hence, it is essential to rational behavior. Yet it
sometimes misleads us. In such circumstances we need to ensure that reason and
analysis also are employed.
Although the visceral emotion of fear certainly plays a role in risk as feelings,
we focus here on the “faint whisper of emotion” called affect and how it contrib-
utes to or, in some cases, undermines effective decision making in the face of
mass atrocities. As used here, affect refers to specific feelings of “goodness” or
“badness” experienced with or without conscious awareness. Positive and nega-
tive feelings occur rapidly and automatically; note how quickly you sense the
feelings associated with the word joy or the word hate. A large research literature
in psychology documents the importance of affect in (1) conveying meaning
upon information and (2) motivating behavior. Without affect, information lacks
meaning and may not be used appropriately in judgment and decision making.
Risk as feelings clearly employs both imagery and affect in remarkably accu-
rate and efficient ways, but this way of responding to risk also has a darker, nonra-
tional side and may misguide us in important ways. Particularly problematic is the
difficulty of comprehending the meaning of catastrophic losses of life when rely-
ing on feelings. Research reviewed below shows that disaster statistics, no matter
how large the numbers, lack emotion or feeling. As a result, they fail to convey the
true meaning of such calamities and they fail to motivate proper action to prevent
them. Our affective system, much like our cognitive abilities, appears to be lim-
ited, leading to nonoptimal behaviors such as insensitivity to magnitude.
The psychological factors underlying insensitivity to large-scale loss of life apply
to catastrophic harm resulting from human malevolence, natural disasters, and
technological accidents. In particular, the psychological account described here can
explain, in part, the failure to respond to the diffuse and seemingly distant threat
posed by global warming or the presence of nuclear weaponry. Similar insensitivity
may also underlie our failure to respond adequately to problems of violence, famine,
poverty, and disease afflicting large numbers of people around the world.
26.3.1. Affect and the Value of Human Lives

This brings us to a crucial question: How should we value the saving of human
lives? An analytic answer would look to basic principles or fundamental values for
(a) (b)
Value of life saving

0 1 2 N 0 1 2 N
Number of lives Number of lives
Figure 26.1 Two normative models for valuing the saving of human lives where
(a) every human life is of equal value and (b) large losses threaten the viability of the
group or society. Source: Slovic (2007).
guidance. For example, Article 1 of the UN Universal Declaration of Human Rights

asserts that “[a]‌ll human beings are born free and equal in dignity and rights.”1 We
might infer from this the conclusion that every human life is of equal value. If so,
then—applying a rational calculation—the value of saving N lives is N times the
value of saving one life, as represented by the linear function in Figure 26.1a.
An argument can also be made for judging large losses of life to be dispropor-
tionately more serious because they threaten the social fabric and viability of a
group or community (see Figure 26.1b). Debate can be had at the margins over
whether one should assign greater value to younger people versus the elderly, or
whether governments have a duty to give more weight to the lives of their own
people, and so on, but a perspective approximating the equality of human lives is
rather uncontroversial.
How do we actually value human lives? Research provides evidence in support
of two descriptive models linked to affect and intuitive thinking that reflect values
for life saving that are profoundly different from those depicted in the norma-
tive (rational) models as shown in Figures 26.1a and 26.1b. Both of these descrip-
tive models demonstrate responses that are insensitive to large losses of human
life, consistent with the frequently observed apathy toward genocide.
26.3.1.1. Model 1: Psychophysical Numbing

There is considerable evidence that our affective responses and the resulting value
we place on saving human lives follow the same sort of psychophysical function that
characterizes our diminished sensitivity to changes in a wide range of perceptual
and cognitive entities—brightness, loudness, heaviness, and wealth—as their
underlying magnitudes increase.
The importance of the psychophysical function is reflected by its role in pros-
pect theory (Kahneman and Tversky 1979). Prospect theory is arguably the
most important descriptive theoretical framework ever developed in the field of
decision making. It has been cited close to 8,000 times by journals in business,
Value
Losses Gains
Figure 26.2 Prospect theory’s value function. Source: Kahneman and Tversky (1979).
Copyright © 1979 The Econometric Society. Reprinted with permission.
economics, law, management, medicine, psychology, and political science. The

heart of the theory is the value function (Figure 26.2), proposing that the carri-
ers of value are positive or negative changes from a reference point. Kahneman
(2011) observed that “if prospect theory had a flag, this image would be drawn on
it” (282). The function is nonlinear, reflecting diminishing sensitivity to magni-
tude. In the positive domain, for example, a gain of two tends to be valued as less
than twice that of a gain of one.
As psychophysical research indicates, constant increases in the magnitude of a stim-
ulus typically evoke smaller and smaller changes in response. Applying this principle
to the valuing of human life suggests that a form of psychophysical numbing may result
from our inability to appreciate losses of life as they become incrementally larger. The
function in Figure 26.3 represents a value structure in which the importance of saving
one life is great when it is the first or only life saved, but diminishes as the total number
of lives at risk increases. Thus, psychologically, the importance of saving one life pales
against the background of a larger threat: We may not “feel” much difference, nor value
the difference, between saving 87 lives, or saving 88.
Fetherstonhaugh et al. (1997) demonstrated this potential for psychophysical
numbing in the context of evaluating people’s willingness to fund various life-
saving interventions. In one study, they found that people were less likely to send
clean water that could save 4,500 lives in a refugee camp when the number of
people in the camp was large (250,000) than when it was much smaller (11,000).
In recent years, vivid images of natural disasters in South Asia and the
American Gulf Coast, and stories of individual victims there—brought to us
through relentless, courageous, and intimate news coverage—unleashed an out-
pouring of compassion and humanitarian aid from all over the world. Perhaps
there is hope here that vivid, personalized media coverage featuring victims could
also motivate intervention to halt the killing.
Perhaps. Research demonstrates that people are much more willing to aid
identified individuals than unidentified or statistical victims. A cautionary note

0 1 2 N
Number of lives
Figure 26.3 A psychophysical model describing how the saving of human lives may
actually be valued. Source: Slovic (2007).
$3.00
$2.00
$1.00
$0.00
Identifiable life Statistical Identifiable life
(Rokia) lives with statistics
Figure 26.4 Mean donations. Source: Reprinted from Organizational Behavior and Human
Decision Processes 102, no. 2: Deborah A. Small, George Loewenstein, and Paul Slovic, “Sympathy and
Callousness: The Impact of Deliberative Thought on Donations to Identifiable and Statistical Victims,”
143–53, Copyright © 2007, with permission from Elsevier.
comes from a study in which Small, Loewenstein, and Slovic (2007) gave people
who had just participated in a paid psychological experiment the opportunity to
contribute up to five dollars of their earnings to the charity Save the Children.
In one condition, respondents were asked to donate money to feed an identified
victim, a seven-year-old African girl named Rokia, of whom they were shown a
picture. They contributed more than twice the amount given by a second group
who were asked to donate to the same organization working to save millions of
Africans (statistical lives) from hunger. Respondents in a third group were asked
to donate to Rokia, but were also shown the larger statistical problem (millions
in need) shown to the second group. Unfortunately, coupling the large-scale sta-
tistical realities with Rokia’s story significantly reduced contributions to Rokia
(see Figure 26.4).
Why did this occur? Perhaps the presence of statistics reduced the attention to
Rokia essential for establishing the emotional connection necessary to motivate
donations. Alternatively, recognition of the millions who would not be helped by
one’s small donation may have produced negative feelings that inhibited dona-
tions. Note the similarity here at the individual level to the failure to help 4,500
people in the larger refugee camp. The rationality of these responses can be ques-
tioned. Why be deterred from helping 1 person or 4,500 people just because there
are many others who cannot be saved? Again, negative affective feelings may
hinder action. It is important to note that these negative feelings are irrelevant for
the decision to help those who can be helped. Clearly, such influences are incon-
sistent with a rational view of human behavior, and they may lead people’s behav-
ior away from their “values.”
In sum, research on psychophysical numbing is important because it demon-
strates that feelings necessary for motivating life-saving actions are not congruent
with the normative and rational models in Figures 26.1a and 26.1b. The nonlin-
earity displayed in Figure 26.3 is consistent with the devaluing of incremental
increases in loss of life in the context of large-scale tragedy. It thus explains why
we do not feel any different upon learning that the death toll in Darfur is closer to
400,000 than to 200,000.
What the psychophysical model does not fully explain, however, is apathy
toward genocide, inasmuch as it implies that the response to initial loss of life
will be strong and maintained, albeit with diminished sensitivity, as the losses
increase. Evidence for a second descriptive model, better suited to explain apathy
toward large losses of lives, follows.
26.3.1.2. Model 2: Compassion Fade

Studies by social psychologists find that a single individual, unlike a group, is
viewed as a psychologically coherent unit. This leads to more extensive process-
ing of information and stronger impressions about individuals than about groups.
Consistent with this, a study in Israel found that people tend to feel more distress
and compassion and to provide more aid when considering a single victim than
when considering a group of eight victims (Kogut and Ritov 2005). A related study
in Sweden found that people felt less positive affect and donated less monetary
aid toward a pair of starving children than to either individual alone (Västfjäll
et al. 2014). The American writer Annie Dillard (1999) reads in her newspaper the
headline “Head Spinning Numbers Cause Mind to Go Slack.” She writes of “com-
passion fatigue” and asks, “At what number do other individuals blur for me?”2
Perhaps the blurring that Annie Dillard asked about begins for groups as small as
two people. The decline in value as losses of life increase has been found in other
studies as well. We have referred to this as compassion fade (Västfjäll et al. 2014).
26.3.2. The Prominence Effect: Confronting the Collapse

of Humanitarian Values in Foreign Policy Decisions
Psychophysical numbing and compassion fade can explain why the public and
some government officials fail to appreciate and take appropriate actions to inter-
vene in genocides and mass atrocities. But many in government are quite aware of
the gravity of these situations, think carefully about them, care deeply about the
suffering, and yet still fail to act. What might account for this?
Again we look to psychological theory for an explanation, specifically theoreti-

cal models of judgment and choice, to explain what we hypothesize is a systematic
bias in decision making that works to inhibit humanitarian intervention.
Decisions to protect civilian lives by intervening in foreign countries are some
of the most difficult and controversial choices facing national decision mak-
ers. Although each situation is unique, these decisions typically involve trade-
offs that pit the value of human lives against other important objectives. There
is often a striking disconnect between the high value placed on saving human
lives expressed by top government officials and the apparent low value revealed
by government decisions not to intervene. Specifically, when multiple objectives
are in play, highly regarded humanitarian values appear to collapse in relation to
national security and economic security objectives.
Although each situation is unique, decisions involving tradeoffs that pit the
value of human lives against other important objectives are quite common. For
example, in 2011 the United States supported military action to protect the lives
of civilians living in Libya and intervened aggressively to protect a threatened
population of Yazidi people in Iraq in 2014. On the other hand, the United States
has done little to intervene in the genocide in Darfur or the mass atrocities in
Syria that have led to hundreds of thousands of deaths and millions of displaced
persons.
One explanation for such inconsistency is that the threat to lives in Darfur or
Syria has not been valued highly enough to compete against other political, eco-
nomic, cultural, or military objectives. What we observe is a discrepancy between
the high value placed on saving human lives expressed by top government offi-
cials and the apparent low value revealed by government inaction when millions
are threatened.
When considering objectives that might influence life-protecting interven-
tions, national security comes foremost to mind. Yet the rhetoric of the two most
recent American presidents leaves no doubt that, in terms of expressed values,
national security and humanitarian life saving are both vital objectives.
In the few situations where the United States has intervened with the stated
objective of saving lives, there were presumed security benefits as well. The United
States justified the attack on Saddam Hussein on the grounds that he possessed
weapons of mass destruction. Only when that proved false did we highlight the
atrocities he had perpetrated. Libyan leader Muammar al-Gaddafi had long been
known as a “loose cannon,” addicted to violence at home and elsewhere. His
menacing visage adorned the cover of Time magazine four times since 1986, when
Ronald Reagan referred to him as “this mad dog of the Middle East.” Security
objectives were also important in the more recent Iraq example. In addition to
protecting the Yazidis, the United States protected American military and diplo-
matic personnel stationed nearby in Erbil. Without that security objective, would
the United States have aided the Yazidis?
In contrast, humanitarian intervention in Darfur may have been blocked by

threats to security. We have long sought to obtain intelligence regarding terrorist
operations from the Sudanese government (Albright 2004). Sudanese President
Omar al-Bashir, who takes a back seat to no one as a murderer, has at times been
providing the American government with information about terrorist activities.
These examples suggest a hypothesis that the United States only seems to launch
humanitarian interventions when security interests are also served by such
actions. The threat to human rights posed by security objectives is well recognized
in the political world (e.g., Office of the United Nations High Commissioner for
Human Rights 2008).
Turning to theory, rational choice models typically assume that choices are
consistent with expressed values. However, a great deal of empirical research
has shown that the values indicated by these two modes of assessment often
differ. One explanation for such inconsistency has centered on evidence for
systematic discrepancies in weighting associated with expressed and revealed
preferences.
A study by Slovic (1975) found that difficult choices were systematically
decided in favor of the alternative that was superior on the most important attri-
bute. Tversky, Sattath, and Slovic (1988) used this finding as a springboard to
a general theory of choice called the contingent weighting model. At the heart of
this model was the prominence effect, which recognized that the values revealed
through choices or decisions tend to differ systematically from directly expressed
or stated values. Specifically, in certain contexts, the more prominent attributes
are weighted more heavily in choices than in judgments reflecting expressed
preferences or values. The presumed explanation for this effect is that, unlike
expressed values, chosen actions need to be justified, and decisions congruent
with prominent attributes are inherently more defensible.
We argue that the prominence effect may underlie the apparent discrepancy
between expressed and revealed values regarding whether or not to act to protect
large numbers of civilian lives under attack in foreign countries. Specifically, we
hypothesize that national security is the prominent dimension in the context we
are studying here. As chosen actions need to be justified, deciding in favor of secu-
rity likely makes a stronger argument than deciding in favor of protecting name-
less, faceless foreign lives, no matter how many thousands or millions of lives are
at stake.
But if the prominence effect is indeed infiltrating top-level policy decisions
and causing decision makers to systematically devalue humanitarian actions,
we doubt that the decision makers are consciously aware of this. The promi-
nence mechanism we assume to be driving the decision-making process is not
a consciously expressed devaluation of distant lives; this would be abhorrent to
leaders who truly do value those lives. Rather, we believe that prominent objec-
tives, in particular those offering enhanced security, draw attention away from
less-prominent goals. All eyes are on options that protect the homeland, and deci-
sion makers fixated on security objectives likely fail to consider as seriously the
numbers of people under siege and left to die. Therefore, compensatory weighing
of costs and benefits associated with seeking security and protecting distant lives
is not carefully addressed.
Thus, meaningful action to prevent genocides and other mass atrocities faces
two psychological obstacles. The prominence effect may lead to decisions that
favor inaction, even when this contravenes deeply held values. And decision
makers can get away with this because the public is psychologically numbed.
As Samantha Power (2003, XXI) observed: “No U.S. president has ever made
genocide prevention a priority, and no U.S. president has ever suffered politi-
cally for his indifference to its occurrence. It is thus no coincidence that genocide
rages on.”
26.4. Implications of the Psychological Account

At a theoretical level, these results suggest that important descriptive decision-
making accounts, such as prospect theory, may not always adequately describe
how people value human lives (Kahneman and Tversky 1979; Dickert et al.
2012). We propose that, for life-saving decisions, both the gain and loss domains
of the value function may not only be characterized by a decreased sensitivity
as magnitude increases (the psychophysical functions in Figures 26.2 and 26.3),
but may sometimes even show a decline in value resulting from compassion fade.
Accordingly, Figure 26.5 hypothesizes a decreased sensitivity and at some point
decreasing value in both the loss and gain domains (an inverted U-shape in the
gain domain and a U-shaped one for losses). There is considerable evidence out-
side the field of judgment and decision making for a value function that follows
such an inverted U-shape function in the gain domain. While it is difficult to find
evidence of negative utility resulting from increases in monetary wealth, other
forms of basic economic behavior sometimes follow a declining function even to
the point of collapse. For example, food consumption often follows this trajec-
tory where the value of initial food intake is very high. After attaining some level
of satiation (a psychophysical function consistent with prospect theory), further
food intake may no longer be attractive (i.e., the value is changing). Importantly,
at some point (a threshold that may vary with individuals and over time and
contexts), the value of further food intake is going to start to decline, perhaps
precipitously (Smith 1983; Blundell et al. 2009). We believe that such a model
can describe how we value magnitudes more generally in at least some domains,
including valuation of lives. While the diminishing sensitivity to magnitude in
prospect theory may be explained (at least in part) by diminishing sensitivity
to abstract numbers (Schley and Peters 2014), the further collapse in domains
Value
Losses Gains
Figure 26.5 A modified value function. Value increases at a diminishing rate and, at
some point, begins to decline. Source: Västfjäll et al. (2014).
such as valuation of lives may be determined by decreased positive affect to those

additional lives.
The insensitivity to life saving portrayed by the psychophysical numb-
ing and compassion fade depicted in Figure 26.5 is unsettling. Whereas
Robert J. Lifton (1967) coined the term psychic numbing to describe the “turn-
ing off” of feeling that enabled rescue workers to function during the horrific
aftermath of the Hiroshima bombing, Figure 26.5 depicts forms of numbing
that are not beneficial. Rather they lead to apathy and inaction, consistent with
what is seen repeatedly in response to genocides and other mass atrocities.
Implications of the prominence effect may be even more unsettling, as it sug-
gests that undisciplined decision processes may lead top-level officials to act,
or fail to act, in ways that contradict strongly held values in support of protect-
ing foreign lives.
26.5. Implications for International Law and Policy

Behavioral research, supported by common observation and the record of
repeated failures to prevent or halt catastrophic losses of life, sends a strong and
important message: Our moral intuitions often fail us. They seduce us into calmly
turning away from mass atrocities, rather than attempting to intervene. 3
Fortunately, we can address these crises with slow thinking, such as
thoughtful deliberation, and use our deliberative skills to design more
responsive international legal and political structures. The United Nations
was created in part to deal with such issues, but structural problems built into
its charter have made it an ineffective mechanism to deal with mass atroci-
ties. The failure of our collective moral intuition and failure of the institu-
tions created to address mass atrocities combine to produce a cycle of willful
inaction. Genocide prevention urgently requires new laws and institutional

arrangements that will force us to doggedly pursue the hard measures needed
to combat mass atrocities when our attention strays and our feelings lull us
into complacency.
Clearly there are many serious obstacles to consistent, meaningful interven-
tion to prevent genocide and similarly grave abuses. In addition to the more obvi-
ous political, material, and logistical impediments, the international community
must overcome the psychological constraints described here. Indeed, the cogni-
tive and affective limitations we identify make it much more difficult to mobilize
global public sentiment to overcome the more obvious material and logistical
constraints. The question is whether and how international law and institutions
might be reformed to account for these cognitive limitations. In this section, we
briefly examine several implications of this research for the law and policy of
atrocity prevention.
Several of the following proposals are ambitious—especially those involving
change to the use-of-force regime—and that ambition raises questions about
their political viability. But there are several factors that may increase their via-
bility. First, to the extent that psychic numbing exists, and is masking a prefer-
ence for antigenocide action, unmasking that preference may produce powerful
political will. Second, political actors may be more willing to embrace these vari-
ous reforms if the changes are not intended to overcome political interests, but
to overcome cognitive failures. The psychological research shows a collapse of
rational calculation and evaluation that causes us to artificially devalue human
life. Therefore, the need for reform should be grounded in an understanding that
cognitive deficiencies can prevent actors from realizing a preference for stopping
mass human rights violations—even when doing so would serve their overall values
and interests.
Appreciation of the failure of moral intuition should inform the develop-
ment of new legal rules and institutional arrangements concerned with atroc-
ity prevention and human rights more generally. We next introduce several
strategies: (1) to insulate institutions from the effects of psychic numbing;
(2) to remove or restrict institutional features that foster psychic numbing;
(3) to promote careful analysis and deliberation directly; and (4) to employ
fast intuitive thinking to channel actors toward more adequate, deliberative
processes.
26.5.1. Insulate Institutions from Effects of Psychic Numbing

One approach is to insulate decision-making processes from the adverse psycho-
logical effects that we have identified. For example, policymakers might design
institutions to be less susceptible to psychic numbing or to operate despite the
psychological effects on actors within the institution.
26.5.1.1. Construct Default Rules and Precommitment Devices

The international regime could construct precommitment enforcement strat-
egies to deal with genocides and other human rights atrocities of similar scale.
Consider a few options: the UN Security Council could preauthorize, subject
perhaps to an ex post council override, the use of force in any situation in which
atrocities reach a certain scale. Another possibility is that the Security Council
could order (rather than authorize) all member states to take coercive action once
the commission of atrocities reached a certain level. Alternatively, states could
conclude a treaty in which state parties would preinvite foreign intervention and/
or UN peacekeepers in the event that genocide occurs on their own territory.
Similarly, the psychological evidence provides a powerful reason for support-
ing the Responsibility to Protect notion, an emerging doctrine that shifts from a
right of states to a duty of states to intervene in another country to stop an atrocity
(Wheeler 2005). That is, the novelty of the Responsibility to Protect idea is that
states are under an affirmative obligation—not just a license—to intervene once
the Security Council has authorized such action. The psychological findings pro-
vide an independent and unique reason to place pressure on states in the form of
this legal responsibility. The starting point should favor intervention (at the very
least when the Security Council has determined force is appropriate).
Other precommitment strategies could be implemented to insulate institutions
from the effects of psychic numbing with respect to human rights more generally.
Aside from the use-of-force regime, multilateral organizations could preautho-
rize economic sanctions on the part of their member states. Nations could pass
domestic legislation that triggers such sanctions or that automatically increases
foreign aid in the event of a humanitarian catastrophe (and could perhaps require
repeal of such aid by a supermajority). States could preauthorize UN Special
Rapporteurs to visit their country in the event of mass human rights violations. In
all these instances, multilateral bodies, foreign countries, and the affected nation
might be ill-equipped—w ithout the assistance of a precommitment device—to
confront a situation after deaths and deprivations begin to mount.
Questions about whether and how to intervene in ongoing conflicts—m ilitar-
ily, economically, and so on—tend to occupy the field of the genocide-response
debate, and one appealing feature of the psychic-numbing literature is that it may
offer a simple metric for determining when to intervene. Say, for example, that
valuations of life begin to drop off significantly after 100 deaths. At 100 deaths, a
preauthorized UN investigation would automatically be triggered (implementing
new reporting methods, as discussed below); at 1,000 deaths, that investigatory
body would immediately acquire certain authorities. These lockstep provisions
can be justified on the grounds that any more subjective metric raises the risk
of psychic numbing. If such a system could be implemented, it could limit the
opportunity for genocidal states to stall international intervention under the
guise of diplomatic debate.
26.5.1.2. Emphasize Early Warning and Preventive Action

Another approach is to act before psychic numbing sets in. Apart from the fact
that prevention is in many ways easier, less costly, and less difficult than inter-
vention (Hamburg 2008), reaction strategies must necessarily overcome the psy-
chic numbing generated by the instant crisis. This insight recommends a range
of law and policy options, including more vigorous international monitoring or
intervention in situations likely to generate wide-scale atrocities (e.g., civil wars,
military coups, etc.) or even “anticipatory” humanitarian intervention (Richter
and Stanton n.d.). It recommends establishing a general, preventive disclosure
mechanism to preclude trafficking in resources that are at risk for funding human
rights abuses, as a recent US law attempts for “conflict minerals” in Congo. It also
calls for greater financial and political support for criminal trials—i f that instru-
ment can be expected to deter future violations or to help halt cycles of violence.
26.5.1.3. Empower Institutions and Actors Less Likely

to Succumb to Psychic Numbing
Psychological research also provides good reason to support a form of subsidiarity
within the humanitarian rights and use-of-force regime. Regional and local actors
who are closer to the situation are more likely to appreciate the reality and the
gravity of the atrocities being committed. Accordingly, international law might
provide regional organizations (e.g., the Economic Community of West African
States, the African Union) greater leeway to use force to stop genocides before or
even without Security Council action. The objective here is to create a one-way
ratchet, providing more proximate and local actors an option to intervene with-
out complete international backing. The design would not work the other way
around—to provide regional actors authority to bar outside intervention by the
international community.
Regional actors could also be empowered in intergovernmental settings involv-
ing enforcement measures not involving the use of force. Examples of such enforce-
ment measures include formal resolutions condemning a state for extremely
poor human rights conditions, the creation of a special rapporteur to monitor
the country, the ousting of a state from an intergovernmental organization, and
the imposition of economic sanctions. Voting rules could be fashioned whereby
such measures would be adopted either if a majority of state parties approves or if a
supermajority of states from the relevant region approve. For example, the imposi-
tion of sanctions against Zimbabwe could be approved either (1) by a majority of
all state parties to an international organization or (2) by approval of three-fourths
of African states, even if a majority of the whole does not agree. Once again, these
devices are intended to function as a one-way ratchet. Such a design principle
would be important due to the fact that, for political and psychological reasons,
regional actors may protect their neighbors from enforcement actions.
Outside monitoring and independent international review are key compo-

nents of the international regime. The foregoing discussion suggests that, in fact,
outside reviewers may be more susceptible to numbing effects. Responding by
empowering local actors to conduct investigations may not solve the numbing
problem, but rather replace it with a neutrality problem if local actors are less
likely than their international counterparts to be impartial observers. One poten-
tial solution would be to turn to intermediate actors—regional bodies or hybrid
local/international bodies. Another would be to train, to the extent possible, the
relevant rapporteurs to recognize and counter the risks of psychic numbing. But
however the problem is addressed, institutional capacities must be assessed and,
although it is not currently recognized as such, psychic numbing is a relevant
factor to consider in making this assessment.
26.5.2. Address Institutional Features That Foster

Psychic Numbing
26.5.2.1. Change the Method and Content of Human Rights Reporting
By challenging the assumption that information makes positive change more
likely, the research presented in this chapter calls into question one of the strategic
pillars of human rights advocacy. Documentation, including the presentation of
data showing mass and systematic violations, is often thought to raise awareness.
Efforts by international organizations to document mass human rights violations
typically focus on the widespread nature of violations rather than on narratives
or other information about the individuals who have been harmed. Statistics pre-
vail over stories. A good example of this is the Darfur Atrocities Documentation
Project (Totten 2006), which compiled a database of over 10,000 eye-w itnessed
incidents but reported mostly the percentages of different types of abuses.
International legal procedures amplify the problem. First, consider strict
page limitations on reports to the UN Human Rights Council. These page con-
straints apply to reports by nongovernmental organizations as well as by UN
human rights officials. As a result, the authors of the reports condense informa-
tion into compact pieces of data, and are unable to delve deeply into descrip-
tions of individuals’ lives. Under these pressures, statistics are also considered
an efficient method for conveying information. Second, in official settings little
opportunity exists for conveying information in the form of visual media. Third,
important international legal forums impose an express or implicit requirement
that violations meet a quantitative threshold (UN Human Rights Council’s
1503 Complaints Procedure), which incentivizes advocates to frame their
appeal through the representation of large numbers of cases. It is not difficult
to conceive of innovations to repair these problems. Procedural and substantive
requirements could be softened or exceptions could be made to expand the

forms of information conveyance.
26.5.2.2. Reconsider Human Rights Indicators

Many now call for the use of quantitative indicators in global governance (e.g.,
measures of good governance by the World Bank; see, e.g., Davis, Kingsbury,
and Merry 2010). Some psychological research suggests that significant perverse
effects may result from the collection and circulation of quantitative human
rights indicators, because actors involved in these processes may become desensi-
tized to human rights violations. With this caveat in mind, however, in many situ-
ations indicators will prove invaluable for responding to psychic numbing. First,
indicators can provide a valuable tool for tracking the likelihood of numbing
effects—t he larger the numbers involved, the greater the risks. Second, indica-
tors can prove to be a valuable tool for overcoming vague expressions of concern
and instead encourage precise thinking and aid effective communication. Third,
many of the problems that might be associated with indicators can be overcome
if we are mindful of the difference between the collection of data and its final,
presented form. For example, data collection and data reporting could be done by
different agencies, and those persons charged with the collection of data should
also be trained to look for other signs of human rights abuse and, in particular, for
stories that serve to illustrate the significance of a given atrocity.
26.5.2.3. Reconsider Substantive Elements of Human Rights Law

Even the substantive law of genocide might be considered problematic as it
conceptualizes genocide as a collective or group injury, rather than as harm to
individuals. As a result of the legal definition, the discourse surrounding the pre-
sentation of grievances may focus too extensively on the group-based harms. In
this light, it is instructive to reflect on the characterization by Holocaust survivor
Abel Hertzberg: “There were not six million Jews murdered: there was one mur-
der, six million times” (United States Holocaust Memorial Museum 2005).
The definition of crimes against humanity raises a similar concern. Generally
defined as a “widespread and systematic” attack against a civilian population,
the elements of the crime might also emphasize the representation of aggregate
numbers rather than individual cases. The particular definition of crimes against
humanity in the UN Statute for the Rwanda Tribunal includes an unusual require-
ment that the attack be directed against a “civilian population on national, politi-
cal, ethnic, racial or religious grounds.” That definition (which was altered in the
treaty for the International Criminal Court) shares some of the same concerns as
the group-based focus of genocide.
26.5.3. Employ Intuitive Thinking to Activate

and Support Deliberative Processes
Despite the limitations of fast, intuitive responses noted above, we should never-
theless attempt to bolster the feelings they generate, to motivate support for ratio-
nal deliberation. Such attempts should capitalize on the findings described earlier
demonstrating that we care greatly about aiding individual people in need, even
more so when we can attach a name and a face to them.
26.5.3.1. Affective Imagery

The data in this chapter present a striking irony: in an effort to emphasize
objective facts, human rights proponents risk losing their ability to connect
with sympathizers on a human level. To be sure, we do not advocate wholesale
abandonment of current reporting mechanisms or the exclusive adoption of
emotion-laden stories. After all, the goal of overcoming psychic numbing is to
better calibrate our interventions to the scale of the atrocities that we face. But
there is ample room for the future of human rights reporting to exhibit mixed
methodologies.
The increasing availability of mixed media may help in this regard. As people
post visceral digital content depicting human rights abuses, audiences may exhibit
responses which otherwise had been masked by numbing effects. In April 2010,
the website WikiLeaks posted a video of US soldiers firing indiscriminately upon
civilians in Iraq, creating a media and political uproar. Dozens of news reports
had already reported on the problem of indiscriminate targeting, none of which
garnered the same attention as the online video. The same phenomenon can be
said of the Abu Ghraib prisoner abuse scandal—during the entire US occupation
of Iraq, nothing created the same backlash as the release of photos of prisoner
mistreatment, despite several reports that suggested much more violent and more
widespread practices.
Thus, one possibility is to infuse human rights reporting with powerful affec-
tive imagery such as that associated with Hurricane Katrina, the South Asian tsu-
nami, and the earthquake in Haiti. This would require pressure on the media to
report the slaughter of innocent people aggressively and vividly. Another way to
engage our experiential system would be to bring people from abused populations
into our communities and our homes to tell their stories.
Above we discuss the disadvantages of reports that focus on numbers of viola-
tions. While it is obviously necessary to document the scope of such atrocities,
neglecting the stories of individuals certainly contributes to numbing. Human
rights advocates should reorient documentation and reporting of abuses to
prompt fast, emotion-laden thinking. In some cases, in-depth narratives and
visual personal stories describing the predicament of individual victims should
be emphasized instead of more abstract descriptions of the scale of abuses, that is,
stories over statistics.
At the same time, scale and systematicity presumably remain important for cali-
brating the appropriate response to any human rights problem. As a consequence,
human rights documentation should not abandon the reporting of scale and system-
level effects. The central challenge of applying the psychological research to human
rights advocacy is identifying when or how much “statistics” and when or how much
“storytelling” should be employed in the documentation and reporting of abuses.
Arresting visual displays and photographs of victims and atrocities should be
included in the reporting and publicly distributed information presented by human
rights advocates. Indeed, the future success of the human rights movement requires
training not only advocates skilled in documenting large numbers of cases, and pro-
fessionals skilled in quantitative methods, but also professionals skilled in compos-
ing and representing narratives about the lives of individual victims.
On this last point, Paul Farmer (2005) has written eloquently about the
power of images, narratives, and first-person testimony to overcome our “fail-
ure of imagination” in contemplating the fate of distant, suffering people. Such
documentation can, he asserts, render abstract struggles personal and help make
human rights violations “real” to those unlikely to suffer them.
26.5.3.2. Victim Empowerment

Another domain is victim empowerment. Where deliberative processes are
systematically lacking, victims could be empowered to trigger a range of insti-
tutional responses such as initiating international court proceedings, placing an
issue on the agenda of an international political body, or making a presentation as
part of the deliberative process. Human rights organizations, including the UN
Office of the High Commissioner for Human Rights, could personally involve
victims in making such presentations or reading their organization’s statement
before such bodies. In the abstract, such measures risk biasing decision makers
toward purely emotional responses, which would be inappropriate in certain
decision-making forums. Regime designers would need to consider the condi-
tions for crafting such interventions primarily to prod deliberative mechanisms
into action when they are otherwise deficient.
26.5.4. Directly Promote Deliberation

Even when moral intuitions are distorted, human cognition can rely on the ratio-
nal, deliberative mode of thinking. Where emotion and affect let us down, we
still can be spurred into action if we can employ deliberative processes capable of
weighing the costs and benefits of possible intervention options.
Rather than solely focusing on obligations to act, international and domestic

law could require policymakers and governmental actors to reason about actions
to take in order to overcome psychic numbing. Even better would be an approach
that constructs a policy response through facilitating effective deliberation, with the
participants faced with development of a defensible policy able to make use of a
common structure and language for expressing and communicating their feel-
ings and thoughts. The obligation to deliberate should apply to omissions (e.g.,
the failure to respond meaningfully to genocide) as well as actions. Experience
from other decision contexts that require policymakers to balance a range of costs
and benefits (Gregory and Keeney 1994) indicates that this simple act of engaging
in structured dialogue may help overcome both affective and cognitive obstacles
to intervention and, in many cases, could help to generate new actions that might
not otherwise be considered.
In particular, a deliberative approach commonly known as decision analysis
(Keeney 1982) has been used widely to address a variety of public policy choices
characterized by the type of emotional, health-based, and ethics-based tradeoffs
that lead feelings to override more thoughtful analysis. Decision analysis, par-
ticularly if combined with the insights of psychology and economics, works best
in organizing or structuring difficult choices that involve multiple dimensions of
value and different outcomes of possible actions that are subject to substantial
uncertainty (Gregory et al. 2012).
At first glance, the human suffering involved in mass atrocities is seemingly
at odds with a more dispassionate, quantitative approach to problem solving. Yet
any dialogue among government leaders concerning the choice to intervene or
stand by in the face of reports concerning mass atrocities necessarily involves
dimensions other than those directly concerned with the victims’ suffering: the
costs of providing assistance, the potential risks to those providing military or
humanitarian aid, the domestic and international implications of action or inac-
tion, and so forth. Decision makers who face tough tradeoffs across this array of
impacts can easily feel confused or overwhelmed and, as a result, decision quality
suffers.
As noted earlier, when faced with multidimensional choices that include dispa-
rate values (e.g., national security versus life saving), people may focus only on a
single, prominent dimension of value to the neglect of other important consider-
ations (Lichtenstein, Gregory, and Irwin 2007). Thus, for policymakers, national
security may overwhelm life-saving objectives. Structured decision-a iding exer-
cises can make explicit the conflict between humanitarian and security objectives
and ensure that the weight given to the former is not in violation of held values.
Systematic frameworks for addressing tough multidimensional choices exist
and are widely employed. A common set of decision-analytic methods, based in
multi-attribute decision theory (Keeney and Raiffa 1993), seek first to structure
difficult policy choices in terms of a small set of important concerns and then
generate and evaluate alternatives in terms of how their consequences are likely to
affect these key objectives. These concerns can include both commonly discussed
values (e.g., cost or time) and others that often are omitted from official dialogue
(e.g., protecting democratic institutions or enhancing political relations). A criti-
cal element of this decision-a iding process is coming up with good indicators or
measures to track the performance of different policies. Experience shows that
it is relatively easy to identify broad terms that might help to evaluate contem-
plated actions—protecting civilians, reducing famine, or establishing regional
stability—but, so long as these concerns remain abstractions, different people may
disagree about the extent to which a specific policy alternative addresses them. As
a result, decision-analytic methods highlight the role of performance measures
(or attributes) that seek to operationalize key considerations and develop concise,
agreed-upon measures to aid in their communication and the implementation of
actions that help to address them (Keeney and Gregory 2005).
As noted, one of the difficulties in addressing human rights decisions is that
they typically involve a range of seemingly incommensurable value dimensions.
Choices of this type are often made on the basis of intuition or “gut feelings,” in
the absence of a defensible framework or guidelines. One implication is that deci-
sion makers are likely to evaluate the pros and cons of actions in each new situa-
tion or crisis on an ad hoc and inconsistent basis, without sufficient deliberation
or peer review. Another result is that at least some key considerations are likely
to remain poorly defined or, perhaps, entirely omitted from deliberations at the
same time that other concerns are given undue weight and influence. A third com-
mon result is that each decision is likely to be viewed as a one-off dilemma, thus
making it more difficult to develop consistent standards that might encourage
learning about genocide-prevention strategies by incorporating lessons gained
from one experience to inform later evaluations. Hillary Clinton expressed this
difficulty well in a March 2011 interview with Ryan Lizza (2011, 55) in Tunis:
I get up every morning and I look around the world. People are being
killed in Cote d’Ivoire, they’re being killed in the Eastern Congo, they’re
being oppressed and abused all over the world by dictators and really
unsavory characters. So we could be intervening all over the place. But
this is not a—what is the standard? Is the standard, you know, a leader
who won’t leave office in Ivory Coast and is killing his own people? Gee,
that sounds familiar. So part of it is having to make tough choices and
wanting to help the international community accept responsibility.
(p. 55)
The lack of a clear analytic framework is a serious shortcoming, particularly

when the decision contexts are as significant as those brought to mind by recent
examples of genocide. Fortunately, there exist a variety of practical tools, drawn
Objectives Performance Desired Alt 1 Alt 2 Alt 3

measures direction
maximize intervention Constructed more
effectiveness scale (1–5)
maximize domestic Constructed more
support scale (1–5)
maximize international Constructed more
support scale (1–5)
minimize intervention Dollars less
costs (millions)
minimize fatalities Number less
among military
minimize fatalities in Number less
country
advance post- Dollars more
intervention economic (millions)
opportunities
promote reform Constructed more
scale (1–5)
Figure 26.6 Consequence matrix for clarifying intervention tradeoffs.

Source: Paul Slovic, Daniel Västfjäll, and Robin Gregory, “Informing Decisions to Prevent
Genocide,” SAIS Review of International Affairs 32, no. 1 (2012): 44, Figure 6. Copyright ©
2012 The Johns Hopkins University Press. Reprinted with permission of The Johns Hopkins
University Press.
from the insights and practices of psychologists and decision scientists, which
can help to inform deliberations about the development of a defensible decision-
making framework concerning decisions to prevent mass atrocities. Over time,
these methods could assist policymakers to adopt a more responsible approach
to dealing quickly and effectively with the tough issues and tradeoffs raised by
emerging problems of genocide.
Figure 26.6 illustrates a simple approach that has been widely used to help
decision makers organize their thinking and deliberate more effectively about
complex, multi-issue, and multistakeholder problems. This deceptively simple
visual tool is called a consequence matrix (Keeney 1992) or, in some circles, a facts
box. The rows report a set of critical concerns, defined in terms of specified per-
formance measures, and with a preferred direction (i.e., either more or less is
better). The columns to the right represent alternative policies or actions, which
are ranked or rated by filling in each of the cells of the matrix in terms of how well
the various consequences of the alternative are expected to achieve progress on
each of the specified objectives. In a typical case, there is considerable variation.
One action or intervention policy might be best in terms of anticipated domestic
support but rate poorly in terms of financial costs, whereas a second alternative
might maximize the expected effectiveness of the intervention but run the risk of
causing high numbers of military casualties.
The objectives can also be used to help generate novel alternatives (Alternatives
4, 5, etc.) that seek higher levels of achievement across a number of the different
concerns. This general approach can be used informally, as a tool to stimulate dis-
cussions, or it can be expressed more formally, through development of a value
model that first specifies the key objectives and then weights them in terms of
their contribution to the specific decision context (Keeney and von Winterfeldt
2011). This weighting capability allows tradeoffs to be addressed explicitly across
different intervention contexts, so that country-by-country strategies can be
developed in light of the relevant opportunities and constraints.
26.6. Conclusion
In the course of the past century, national decision makers often have been
informed of imminent or ongoing mass murders and genocides but have chosen
not to intervene. As Samantha Power (2003) sadly notes, America’s record is one
of strong abstract support for principles and ideals opposed to genocide but of
little or no action when a real-world situation arises that calls for immediate, effec-
tive intervention in order to prevent or halt a possible genocide.
There are many reasons for this consistent record of neglect. In this chapter we
focus on two possible explanations. First, we emphasize the role of psychology
and, in particular, affective responses in shaping our values and reactions to mass
atrocities. Drawing upon behavioral research, we argue that we cannot depend
only upon our moral intuitions to motivate us to take proper action against geno-
cide and mass abuse of human rights. This places the burden of response squarely
upon moral argument and international law. Second, we emphasize the need for
an explicit decision framework that can incorporate the multiple dimensions of
value that influence choices about genocide and can provide insights about trade-
offs, the various pros and cons, that characterize alternative responses.
It is time to reexamine repeated failures to act in the face of mass atrocities with
an acknowledgment of the psychological challenges described in this chapter,
along with recognition of the possibilities offered by methods that communicate
the individuality and emotional reality underlying the statistics and encour-
age thoughtful deliberation. The latter is a remedy that is quite modest in that it
requires only the acknowledgment of the potential for structured deliberations to
help bridge strong emotions with the discipline of a reasoned approach. With this
new model as a guide, it may be possible to design legal and institutional mecha-
nisms that will enable us to respond to genocide and other mass harms with a
degree of intensity that is commensurate with the high expressed value we place
on protecting human lives.
Acknowledgments
This chapter is based on research supported by the National Science Foundation
under Grants SES-1024808 and SES-1227729. Any opinions, findings, and con-
clusions or recommendations are those of the authors and do not necessarily
reflect the views of the National Science Foundation.
Notes
1. See http://w ww.un.org/en/documents/udhr/.
2. Dillard (1999) struggles to think straight about the great losses that the world ignores: “More
than two million children die a year from diarrhea and eight hundred thousand from mea-
sles. Do we blink? Stalin starved seven million Ukrainians in one year, Pol Pot killed two
million Cambodians” (130–31).
3. This section draws heavily on the contributions of David Zionts, Andrew Woods, Ryan
Goodman, and Derek Jinks as presented in Slovic et al. (2013). Excerpts from Slovic et al.
(2013) are reprinted with permission of Princeton University Press; permission conveyed
through Copyright Clearance Center, Inc.
References
Albright, M. K. 2004. “Statement of Madeleine K. Albright: National Commission of Terrorist
Attacks upon the United States, March 23, 2004.” http://govinfo.library.unt.edu/911/
hearings/hearing8/a lbright_ statement.pdf.
Blundell, J. E., K. de Graaf, G. Finlayson, J. C. G. Halford, M. Hetherington, N. A. King, and R.
J. Stubbs. 2009. “Measuring Food Intake, Hunger, Satiety and Satiation in the Laboratory.”
In D. B. Allison and M. L. Baskin, eds., Handbook Of Assessment Methods for Eating Behaviours
and Weight-Related Problems: Measures, Theory and Research. 2nd ed. Newbury Park, CA: Sage,
283–325.
Davis, K. E., B. Kingsbury, and S. E. Merry. 2010. Indicators as a Technology of Global Governance
(Report No. 2010/2). Institute for International Law and Justice. http://w ww.iilj.org/pub-
lications/2 010-2 .Davis-K ingsbury-Merry.asp.
Dickert, S., D. Västfjäll, J. Kleber, and P. Slovic. 2012. “Valuations of Human Lives: Normative
Expectations and Psychological Mechanisms of (Ir)rationality.” Synthese 189, no. 1
(supp.): 95–105.
Dillard, A. 1999. For the Time Being. New York: Alfred A. Knopf.
Farmer, P. 2005. “Never Again? Reflections on Human Values and Human Rights.” Paper pre-
sented at the Tanner Lectures on Human Values, Salt Lake City, UT (March). http://w ww.
tannerlectures.utah.edu/lectures/documents/Farmer_ 2 006.pdf.
Fetherstonhaugh, D., P. Slovic, S. M. Johnson, and J. Friedrich. 1997. “Insensitivity to the Value
of Human Life: A Study of Psychophysical Numbing.” Journal of Risk and Uncertainty 14, no.
3: 283–300.
Gregory, R., L. Failing, M. Harstone, G. Long, T. McDaniels, and D. Ohlson. 2012. Structured
Decision Making: A Practical Guide to Environmental Management Choices. Chichester, West
Sussex, UK: Wiley-Blackwell.
Gregory, R., and R. L. Keeney. 1994. “Creating Policy Alternatives Using Stakeholder Values.”
Management Science 40, no. 8: 1035–4 8.
Hamburg, D. A. 2008. Preventing Genocide: Practical Steps toward Early Detection and Effective
Action. Boulder, CO: Paradigm.
Kahneman, D. 2011. Thinking, Fast and Slow. New York: Farrar, Straus & Giroux.
Kahneman, D., and A. Tversky. 1979. “Prospect Theory: An Analysis of Decision under Risk.”
Econometrica 47, no. 2: 263–91.
Keeney, R. L. 1982. “Decision Analysis: An Overview.” Operations Research 30, no. 5: 803–38.
Keeney, R. L. 1992. Value-Focused Thinking: A Path to Creative Decisionmaking. Cambridge,
MA: Harvard University Press.
Keeney, R. L,. and R. Gregory. 2005. “Selecting Attributes to Measure the Achievement of
Objectives.” Operations Research 53, no. 1: 1–11.
Keeney, R. L., and H. Raiffa. 1993. Decisions with Multiple Objectives: Preferences and Value Tradeoffs.
Keeney, R. L., and D. von Winterfeldt. 2011. “A Value Model for Evaluating Homeland Security
Decisions.” Risk Analysis 31, no. 9: 1470–87.
Kogut, T., and I. Ritov. 2005. “The Identified Victim Effect: An Identified Group, or Just a Single
Individual?” Journal of Behavioral Decision Making 18, no. 3: 57–167.
Lichtenstein, S., R. Gregory, and J. Irwin. 2007. “What’s Bad Is Easy: Taboo Values, Affect, and
Cognition.” Judgment and Decision Making 2, no. 3: 169–88.
Lichtenstein, S., and P. Slovic, eds. 2006. The Construction of Preference. New York: Cambridge
University Press.
Lifton, R. J. 1967. Death in Life: Survivors of Hiroshima. New York: Random House.
Lizza, R. 2011. “The Consequentialist.” The New Yorker, May 2, 2011, 44–55.
Office of the United Nations High Commissioner for Human Rights. 2008. Human Rights,
Terrorism, and Counter-Terrorism (Fact Sheet No. 32). http://w ww.ohchr.org/Documents/
Publications/Factsheet32EN.pdf.
Power, S. 2003. A Problem from Hell: America and the Age of Genocide. New York: Harper Perennial.
Richter, E., and G. Stanton. n.d. “The Precautionary Principle: A Brief for the Genocide
Prevention Task Force.” http://w ww.genocidewatch.org/aboutus/bydrgregorystanton.
html.
Schley, D. R., and E. Peters. 2014. “Assessing ‘Economic Value’: Symbolic Number Mappings
Predict Risky and Riskless Valuations.” Psychological Science 25, no. 3: 753–61.
Sen, A. K. 1977. “Rational Fools: A Critique of the Behavioral Foundations of Economic Theory.”
Philosophy and Public Affairs 6, no. 4: 317–4 4.
Simon, H. A. 1957. Models of Man: Social and Rational. New York: Wiley.
Simon, H. A., and A. C. Stedry. 1969. “Psychology and Economics.” In G. Lindzey and
E. Aaronson, eds., The Handbook of Social Psychology. Vol. 5. Reading, MA: Addison Wesley,
269–314.
Slovic, P. 1975. “Choice between Equally Valued Alternatives.” Journal of Experimental
Psychology: Human Perception and Performance 1, no. 3: 280–87.
Slovic, P. 2007. “If I Look at the Mass I Will Never Act: Psychic Numbing and Genocide.”
Judgment and Decision Making 2, no. 1: 79–95.
Slovic, P., M. Finucane, E. Peters, and D. G. MacGregor 2004. “Risk as Analysis and Risk as
Feeling: Some Thoughts about Affect, Reason, Risk, and Rationality.” Risk Analysis 24, no.
2: 311–22.
Slovic, P., D. Västfjäll, and R. Gregory. 2012. “Informing Decisions to Prevent Genocide.” SAIS
Review 32, no. 1: 33–47.
Slovic, P., D. Zionts, A. K. Woods, R. Goodman, and D. Jinks. 2013. “Psychic Numbing and
Mass Atrocity.” In E. Shafir, ed., The Behavioral Foundations of Public Policy. Princeton,
NJ: Princeton University Press, 126–42.
Small, D. A., G. Loewenstein, and P. Slovic 2007. “Sympathy and Callousness: The Impact of
Deliberative Thought on Donations to Identifiable and Statistical Victims.” Organizational
Behavior and Human Decision Processes 102, no. 2: 143–53.
Smith, B. D. 1983. “Extraversion and Electrodermal Activity: Arousability and the Inverted-U.”
Personality and Individual Differences 4, no. 4: 411–19.
Totten, S. 2006. Genocide in Darfur: Investigating the Atrocities in the Sudan. New York: Routledge.
Tversky, A., S. Sattath, and P. Slovic. 1988. “Contingent Weighting in Judgment and Choice.”
Psychological Review 95, no. 3: 371–8 4.
United States Holocaust Memorial Museum. 2005. Life after the Holocaust: Thomas Buergenthal—
Personal History. http://w ww.ushmm.org/wlc/en/media_oi.php?ModuleId=10007192&
MediaId=5603.
Västfjäll, D., P. Slovic, M. Mayorga, and E. Peters. 2014. “Compassion Fade: Affect and Charity
Are Greatest for a Single Child in Need.” PLOS ONE 9, no. 6: e100115.
Wheeler, N. J. 2005. “A Victory for Common Humanity? The Responsibility to Protect and
the 2005 World Summit.” Journal of International Law and International Relations 2, no.
1: 95–105.
27
Genocides and Other Mass Atrocities

A Law and Economics Approach
J u rge n Br au e r , C h a r l e s H . A n de rton, a n d Dav i d Sc h a p
27.1. Introduction
Summarizing large-sample datasets on atrocities involving civilians, chapter 3 in
this volume identifies 201 distinct cases of state-perpetrated genocides and mass
killings from 1900 to 2013. Its cautious estimate of the sum total of intention-
ally caused civilian fatalities in these cases is 84 million people. Other estimates,
which appear less cautious, run to double or even triple that number (Rummel
1998, vii). Even the lower-bound estimate of the death toll is a staggering num-
ber, and other chapters in this book spell out causes, consequences, and poten-
tial remedies in great detail. Perhaps surprisingly, little has been written on the
economic analysis of domestic and international law as it pertains to atrocity
crimes such as genocides. In this chapter, we therefore focus on the role of law
in the prevention of atrocity crimes and how the economic analysis of law may
help us to understand law’s failures and successes as well as future pitfalls and
opportunities.
Section 27.2 recounts definitions of various types of atrocity crimes and
briefly discusses some domestic and international legal instruments and institu-
tions to deal with such crimes. Section 27.3 is a synopsis of basic concepts, ideas,
and illustrations from the field of law and economics that includes a subsection
on the economics of international treaty law, such as the United Nations (UN)
Genocide Convention. As will be seen, the economic analysis of law begins to
explain the persistent presence of the “too little, too late” intervention syndrome
that afflicts virtually all instances of atrocity crimes. Section 27.4 places the law
and economics discussion within the even broader realm of global public goods
(GPGs) and asks exactly what sort of goods are treaties such as the Convention,
what one may expect from them, and who best should provide them. Section 27.5
concludes.
639
27.2. Atrocity Crimes and Legal Instruments

and Institutions
Quite apart from coining the very word “genocide,” and from single-m indedly
instigating the 1948 United Nations Convention on the Prevention and
Punishment of the Crime of Genocide, Raphael Lemkin’s (1944) conception of
genocide also was novel because it included criminal acts committed in peace-
time, a view the Convention adopted in Article 1. Prior to the Convention, at
the Nuremberg trials of 1945–1946, for example, litigation had been limited to
atrocities committed during wartime (Schabas 2010, 126–27). Article 2 of the
Convention defines genocide as “any of the following acts committed with intent
to destroy, in whole or in part, a national, ethnical, racial, or religious group, as
such: (a) Killing members of the group; (b) Causing serious bodily or mental
harm to members of the group; (c) Deliberately inflicting on the group condi-
tions of life calculated to bring about its physical destruction in whole or in part;
(d) Imposing measures intended to prevent births within the group; (e) Forcibly
transferring children of the group to another group” (United Nations 1951).
Note that the crime of genocide need not involve direct killing as only the first
and the third items in the five-item list in Article 2 define genocide with refer-
ence to the physical destruction—t he killing—of people. Note also that intent is
not restricted to “bad faith” (male fides). Thus, even though the latter part of the
genocide against Aborigines and Torres Strait Islander peoples in Australia was
carried out bona fides—to civilize backward peoples—the process nonetheless
aimed at the destruction of their unique identities.1 The Convention’s definition
has not expanded since 1948, but Schabas (2010, 141) argues that one need not
be overly concerned, in principle, about crimes going unpunished as the continu-
ously evolving concept of crimes against humanity has “emerged as the best legal
tool to address atrocities.” Nevertheless, the UN’s definition has been criticized
by scholars, for instance in regard to (1) groups left out (e.g., political groups);
(2) how to identify, document, and prove intent; (3) the apparent inability of the
Convention to prevent genocide; (4) the relationship of genocide to other atroci-
ties; and (5) various misuses of the term (e.g., Curthoys and Docker 2008).
In nongenocidal mass killing, perpetrators kill but do not seek to physically
destroy a group as such (Waller 2007, 14). Such killings come under the rubrics of
war crimes and crimes against humanity. Along with genocide, they are defined
in the 1998 Rome Statute of the International Criminal Court (ICC). As defined
there, crimes against humanity encompass widespread or systematic attacks
against civilians involving inhumane means such as extermination, forcible pop-
ulation transfer, torture, rape, and disappearances. War crimes are grave breaches
of the Geneva Conventions and include the willful killing, torture, causing of
great suffering or serious injury, and extensive destruction and appropriation
Genocides and Other Mass Atrocities 641
of property.2 (See chapters 1, 2, and 3 in this volume for context, definitions,

and data.)
Table 27.1 shows a selection of domestic and international legal instruments
and institutions designed to prevent, punish, and/or foster restitution for atroc-
ity crimes, as well as sources that provide further information. Adjudication of
atrocity crimes began in earnest following World War I with the establishment of
the Turkish Military Tribunal (TMT) (1919–1920), which prosecuted organiz-
ers of the Armenian genocide. Genocide was not then defined, of course, hence
the Turkish government’s insistence not to brand the crimes as genocide. The tri-
als, characterized as “a milestone in Turkish legal history” (Dadrian 1997, 30),
revealed the systematic planning behind the atrocities, enrichment of perpetra-
tors through looting of victims’ assets, and lack of military necessity for the forced
relocation of Armenians. However, the TMT convicted only fifteen men among
the hundreds who orchestrated the crimes (Dadrian 1997).
Following World War II, the International Military Tribunal (IMT) at
Nuremberg tried leading officials for crimes against peace, war crimes, and crimes
against humanity (1945–1946). 3 Twelve Nazi leaders received the death sentence
and many others were given long prison terms. The trials had an important influ-
ence on the growth of international criminal law including the UN Genocide
Convention, the International Criminal Tribunal for the Former Yugoslavia
(ICTY), the International Criminal Tribunal for Rwanda (ICTR), and the ICC.
As of December 2014, the ICTY and ICTR have indicted, respectively, 161 and
95 people. Importantly, the ICTR established, for the first time, mass rape in war-
time as an act of genocide. (Unlike common-law precedent rules in the United
States for instance, the ICTR’s findings do not bind other courts.) Also as of
December 2014, the ICC has indicted thirty-six individuals for atrocity crimes
including three current or former heads of state: Omar al-Bashir (Sudan), Uhuru
Kenyatta (Kenya), and Laurent Gbagbo (Côte d’Ivoire).
Parties to international treaties incur the added step of ratification, which
refers to the passing of domestic legislation compatible with their obligations
under international law. Following the Convention, dozens of nations thus devel-
oped domestic laws to put on trial suspected Nazi war criminals and/or perpetra-
tors of more recent atrocities (Schabas 2003). In 2000, Chile’s Court of Appeals
lifted former President Augusto Pinochet’s immunity from prosecution, paving
the way to try him for his role in civilian atrocities that occurred during his presi-
dency. Although he died prior to any conviction, the case is notable for two rea-
sons. First, it involved a state’s prosecution of its own former leader; and second,
Pinochet was initially arrested in London, based on an application of the prin-
ciple of universal jurisdiction by European judges, a principle by which a state
(or states, in Pinochet’s case) asserts a right to prosecute a person for an alleged
crime regardless of the crime’s location or the accused’s residence or nationality
(Lunga 1992). In contrast, however, a number of African states have refused to
Table 27.1 Selection of Legal Institutions, Jurisprudence, and International
Norms Related to Genocide Prevention and Postgenocide Justice
Selection of Legal Institutions Selection of Sources for Further

(or Norms) Information
International
International Military Tribunal at US Holocaust Memorial Museum
Nuremberg (IMT) (1945–1946) (http://w ww.ushmm.org)
Convention on the Prevention United Nations (https://t reaties.un.org),
and Punishment of the Crime of Schabas (2010), US Holocaust Memorial
Genocide (1948, 1951) Museum (http://w ww.ushmm.org)
International Criminal Tribunal United Nations (http://w ww.icty.org)
for the Former Yugoslavia (ICTY)
(1993)
International Criminal Tribunal United Nations (http://w ww.unictr.org)
for Rwanda (ICTR) (1994)
International Criminal Court International Criminal Court (http://
(ICC) (2002) www.icc-c pi.int)
Norms on the Responsibilities of Hillemanns (2003)
Transnational Corporations and
Other Business Enterprises with
regard to Human Rights (2003)
Extraordinary Chambers in the Extraordinary Chambers in the Courts of
Courts of Cambodia (ECCC) Cambodia (http://w ww.eccc.gov.kh/en)
(2003)
Responsibility to Protect (R2P) United Nations (http://daccess-dds-
(2005) ny.un.org/doc/U NDOC/GEN/
N05/487/60/PDF/N0548760.
pdf?OpenElement)
Domestic
Turkish Military Tribunal Dadrian (1997)
(1919–1920)
US Alien Tort Claims Act (1789, Michalowski (2013)
1980)
Prosecution of civilian atrocities Schabas (2003), Prevent Genocide
(not necessarily genocide) International (http://w ww.
in domestic courts (includes preventgenocide.org)
Nuremberg and others)
enforce an international arrest warrant and detain Omar al-Bashir (Sudan) when
he was on their soil after having been indicted before the ICC.
Another important international development occurred at the 2005 UN
General Assembly. Member states unanimously adopted a norm known as the
Responsibility to Protect (R2P). Even though without legal force (UN Doc.
A/R ES/60/1, paras. 138, 139), it nevertheless was part of the impetus for UN
Security Council Resolution 1973, passed on March 17, 2011, which authorized
member states to take actions, including enforcement of a no-fly zone, to protect
civilians from attacks by the Libyan military.
Although individual agents of corporations can be tried, corporations alleged
to have been complicit in atrocity crimes have not usually faced prosecution
(Kelly 2012). This is because the ICC followed the ICTY and ICTR in assuming
jurisdiction only over “natural persons,” not “legal persons” (Cernic 2010, 141).
Nevertheless, efforts have been made under domestic law, for example under the
US Alien Tort Claims Act, to bring litigation against corporations for alleged
complicity in atrocities and other human rights abuses. Such litigation has led
companies to develop their own norms to avoid such complicity (Michalowski
2013; also see c hapter 25 in this volume).
27.3. Law and Economics

Law and economics, broadly considered, is a field that may be thought to include
industrial organization, business regulation, and antitrust public policy, and thus
might be traced as far back as Adam Smith’s observation, in 1776, that people in
the same trade seldom get together without the discussion turning to price fix-
ing. More narrowly considered, as in the remainder of this chapter, the phrase
“law and economics” has come to be understood as the “new” law and economics,
which began around 1960 when economists and legal scholars initiated research
examining the economic underpinnings of legal doctrines and norms, especially
those that developed as part of English common law. Common law is judge-made
law arising from court decisions, as opposed to statutory law emanating from leg-
islatures. As judges rendered decisions and similar cases later arose, subsequent
jurists would look to earlier decisions for guidance rather than treating each new
case as a “case of first impression.” A rich history of deep-rooted precedents con-
stituted the doctrines and norms of the legal system, many of which were later
codified into statutes in the United States.4
The economic logic contained in common law norms became the subject mat-
ter of the new law and economics. The field developed apace as scholarly journals
devoted to the field came into existence, and a number of joint JD/PhD programs
emerged, linking university law schools with the economics departments in their
arts and sciences divisions. In 1991, the Nobel Memorial Prize in Economic
Sciences was awarded to Ronald Coase, bringing heightened recognition to his

work in general and to the field of law and economics in particular. Successive
waves of researchers in the field took the approach first applied to common law
doctrines and extended it to many areas of law. Insights from sundry applications
in the field are germane to the topic of atrocity crimes. This section provides a syn-
opsis of selected concepts, ideas, and illustrations in law and economics. The last
subsection steers toward the economics of international agreements, to be taken
up and expanded upon in more detail in section 27.4.
27.3.1. The New Law and Economics

The approach used in the new law and economics is applied microeconomics, typ-
ically with reliance on the assumption of purposeful actors in either an implicit
model or more formally in an explicit, often game-t heoretic model. The assump-
tion of goal-d irected behavior makes theory, prediction, and statistical testing
possible. As a general rule, microeconomics is positive and descriptive, address-
ing what is—and how it came to be—as opposed to normative and prescriptive,
addressing what ought to be. It tends to examine not beliefs (“preferences”) but acts
based on beliefs and how these acts are facilitated or constrained by factors such
as resource availability.
An exception to the general rule of eschewing normative or prescriptive pro-
nouncements in microeconomics is economists’ notion of efficiency (defined
momentarily), a shared norm by which rules, systems, events, or behaviors may be
condemned when deemed to be inefficient. In Posnerian law and economics—t hat
mimicking the canonical approach of Richard A. Posner (2011)—institutional
arrangements are evaluated relative to the Kaldor-H icks welfare norm of eco-
nomic efficiency. Under Kaldor-H icks efficiency, an outcome is deemed more
efficient if those who are made better off retain a net gain even after they fully
compensate, in principle, those who are made worse off. 5 Roughly speaking, the
Kaldor-H icks criterion searches for the largest economic pie, independent of how
the pie is divided. Reliance on the Kaldor-H icks criterion is not without its critics,
particularly either those who emphasize a norm of distributive justice (ignored
under Kaldor-H icks) or welfare theorists concerned with formal properties of
abstract models.6 Regardless, the Kaldor-H icks criterion lies at the heart of cost-
benefit analysis wherein all calculations involve monetized values.
Another consideration of importance to the foundations of the economic
analysis of law is the notion of transaction costs, such as in regard to contract
formation and enforcement or when purposeful actors seek to rearrange owner-
ship of rights in a system that protects such rights as are granted. Indeed, Coase
(1992, 717) himself indicated that abstraction from transaction costs can only be
a “stepping stone” on the way toward the analysis of real-world situations where
such costs are always greater than zero, sometimes substantially so. Of particular
concern are two special kinds of transaction costs. First, there is the cost associ-
ated with the ubiquitous problem of incompatible incentives. Whereas two par-
ties may have an incentive to reach an agreement resulting in gains from trade,
each has an incentive to attempt to capture the lion’s share of the gains. Tenacious
bargaining over the apportionment of these gains, however, can frustrate a deal
entirely. Second, in multiparty bargaining, there can be costs associated with free-
rider problems on the one hand and holdout problems on the other hand, as self-
interested, strategic individuals resist group assimilation, coalescence, or even
just mere agreement in an instant case, again frustrating deal accomplishment.
Areas of common law having efficiency and transaction cost aspects relevant
to the study of atrocity crimes include property, contract, tort, and criminal law.
The following subsections briefly address and illustrate each of these.7
27.3.2. Property Law

In the realm of property law, among the questions that are addressed by law
and economics theorists are three major ones: Who should be awarded rights?
How should rights be awarded? And in what manner should rights be protected?
Related to the first question—W ho should be awarded rights?—is the so-called
Coase Theorem, the gist of which is that if parties can bargain at low cost to a
mutually preferred position, then the initial assignment of rights to one party or
the other is immaterial to economic efficiency. While it privately matters who sells
and who buys the right to a resource, it does not socially matter who is initially
awarded the right to that resource as it ultimately flows to the user who values it
the highest (pays the most for it).8 By way of example consider the question “Who
should be awarded rights to Jerusalem?” If Jerusalem were a relatively insignifi-
cant piece of real estate, the initial assignment to a Jewish or a Palestinian person
would not matter socially: whoever values the land most will either keep it or pur-
chase it from the other party. An immediate corollary is that if transaction costs
are prohibitively high, then the initial assignment of rights does matter indeed.
High transaction costs due to cultural enmity, ethnic hostility, or religious fac-
tionalism based on self-i mage, identity, and attachment may outright preclude the
kind of mutually advantageous bargaining needed to advance diverse peoples to
greater economic well-being.9
The second prominent question—How should rights be awarded?—concerns
the method of assigning rights. The prospect of being awarded a special right to
some resource tempts prospective owners to compete with one another. Some
modes of competition, like violent contests, are themselves very costly, thus dis-
sipating resources in the process of obtaining the award. Some systems thought
to be fair allocation schemes—like that of first come, first served—a re seen as
needlessly expensive once one observes the time and other resources wasted by
the allocation mechanism. Political favoritism also is decried due to the waste of
resources involved in the process. Termed “rent-seeking” in the law and econom-
ics and political economy literatures and known, in expanded form, as “directly
unproductive, profit-seeking activities” in the development literature, favoritism
brings strategic, game-theoretic aspects into play as agents move to influence
future resource distribution.10 In contrast, awards by auction to the highest bid-
der generally involve only a modicum of resources exhausted in the process of
making the award.
For example, imagine various states in a costly first-come, first-serve race to
capture an item at the bottom of the sea in international waters. The item might
be a rare ore or a sunken treasure. As competitors race to obtain the item, the
hasty dispatch of means of recovery would include costly infrastructure need-
lessly duplicated across states, namely, their respective large oceangoing vessels,
and the higher costs of rapid deployment due to premium payments for goods and
services acquired to assure quickly going to sea. Yet the item sought has rested
at sea-bottom perhaps for centuries (sunken treasure) or thousands of millennia
(rare ore), so the race is arbitrary and unnecessary except as a degenerate means
to solve the allocation problem of who will get the item first to the exclusion of the
others. An international auction coupled with an accord that specifies a singular,
mutually agreed-to harvester with apportioned proceeds would result in dimin-
ished costs of recovery and thus be more efficient, economically. In the absence of
such an agreement, losses abound.
In an atrocity crimes context, the method of assigning rights and the nature of
the rights so assigned are often based on combinations of political favoritism and
discrimination against groups that are ill-favored along ethnic, religious, cultural,
or political lines. “Rights” so conveyed can become sources of tension, protest, vio-
lence, repressive crackdowns, and even mass atrocity. In the late nineteenth and
early twentieth centuries, for example, Armenians living in the Ottoman Empire
faced various forms of political, economic, and religious discrimination, which
led to formation of political organizations and protests, rising national conscious-
ness, and efforts to gain relief from their plight from European nations. In turn,
the Ottoman Empire’s leaders perceived the Armenians’ behavior as threatening,
which led them to carry out massacres to deal with the “Armenian Question” (see
Adalian 2013, 121–22).
The third question—I n what manner should rights be protected?—concerns
the protection of entitlements.11 For example, imagine a world court with author-
ity to enforce an international agreement governing the geographic extent of
radio broadcasts. When a transmitter in a foreign country overwhelms a domes-
tic broadcast in another country by exceeding broadcast limitations previously
agreed to, a violation of an entitlement occurs. The form of the violation may
be deemed a nuisance or a trespass. The domestic broadcaster may be entitled
to ex post compensation (liability rule protection, as occurs with a nuisance),
that is, a grant, for example, of money damages to be paid by the violator of the
entitlement. The idea is that a credible threat of being held liable ex post inhibits
the harmful act from being committed ex ante. Alternatively, the domestic broad-
caster may seek an ex ante injunctive remedy from the court by which the foreign
broadcaster, here deemed a trespasser, is forced to partially or fully cut back its
planned broadcast stream (property rule protection). With a property rule protec-
tion against trespass, someone wishing to use another’s property is encouraged to
bargain with the owner ex ante to actual use if the transaction costs in so doing
are low relative to the high cost of a court attempting to assess ex post what may
be a highly subjective loss due to trespass. In this set of circumstances, the prop-
erty rule would be regarded as more efficient than the liability rule. If, instead, ex
ante bargaining costs are high and a court ex post has the competence to assess
money damages at low cost to itself, then the circumstances favor use of liability
rule protection. Broadcasting rights might be at issue, for example, if a state jams
hate-message broadcast signals in an atrocity-committing state (on media and
genocide, see chapter 12 in this volume).
A third manner of entitlement protection is inalienability, applied to one’s right
to oneself. For example, international agreements recognizing inalienable rights
include accords against human trafficking. Thus, one may not legally promise
one’s body parts or sell oneself into slavery no matter how desperate one is for
present funds—a contract based on such terms would be unenforceable today.12
Prohibitions against human trafficking are among the enumerated items in the
definition of crimes against humanity (e.g., Art. 7(2)c of the Rome Statute). The
rationale behind having certain inalienable rights rests with a paternalistic argu-
ment: Despite being in the mutual interest of a borrower (who obtains funds
under collateral of one’s person) and a lender (who earns interest on a reasonably
well-assured loan repayment), a third party would be offended by observing the
adverse consequences (to another or others) of a deal gone bad or of a slave trade.
Property law also brings up the issue of asset seizure legislation based on the
holding of stolen or illegally obtained funds through criminal activity such as may
be committed during atrocity crimes. When the United States freezes such assets,
it is because it is alleged that they have been obtained through illegal activity, and
whatever contract the alleged perpetrator may have had with a financial institu-
tion holding those funds is null and void, pending conviction. At the very least,
access to the funds can be “frozen” during criminal proceedings.13
27.3.3. Contract Law

Contract law exists to enforce a mutually agreed-to exchange of promises in the
face of intervening circumstances that may make subsequent breach of a prom-
ise advantageous to one of the parties. Whether or not the intervening event was
contemplated a priori by either party, contract law handles unforeseen events in
a manner similar to how the parties themselves would have handled them had
they been foreseen. Knowing this aspect, the parties need not attempt the impos-
sible task of including every possible contingency in their agreement. Reliance
on contract law lowers their transaction cost and encourages commerce. There
is of course a risk associated with each possible contingency that neither party
wishes to bear, but if one of the two parties can bear the risk at lower cost, the par-
ties ex ante—in crafting the contract—w ill assign the risk to the correct party.
In the absence of explicit recognition of the risk in the formal contract, the court
will mimic ex post what the parties would have done ex ante had the matter been
explicitly addressed. For example, two parties stipulate in a contract for the ship-
ment of goods such that, if an act of war precludes shipment, then the agreement
is voided and money damages are not owed. In the absence of any other explicit
clause, a court ex post in deciding what to do about a shipment failure due to a
tidal wave might reckon that the parties similarly would have wished to have the
agreement voided in the unanticipated (or at least not explicitly addressed) cir-
cumstance. Inasmuch as Posner (2011) regards international agreements among
sovereign states as akin to contracts, questions arise, in the absence of a global
enforcement mechanism and enforcement institution, when one party to an
agreement appears not to live up to its obligations. For example, South Africa
did not arrest and render Sudan’s Omar al-Bashir, indicted by the International
Criminal Court, when the opportunity to do so was present in June 2015.
An important finding from the economics of contract law is that not all con-
tracts should be fulfilled as contemplated because the occurrence of some event
(anticipated or not) may make completion of the contract suboptimal, that is, eco-
nomically inefficient, from the standpoint of the joint-wealth maximization of the
parties. Hence, and perhaps surprisingly, the law provides for breach of contract.
To maintain confidence in the contracting process, the sufferer of the breach is
awarded, for example, money damages ex post that preserve the benefit of the bar-
gain, leaving the party suffering the breach in as good a position as if the contract
had been fulfilled.14
Much in the contracting process may be regarded as existing outside of or along-
side the legal sphere, inasmuch as some agreements may be self-policing or self-
enforcing. For example, if two parties reach an agreement and each has a reputation
to preserve, then a party’s failure to complete the terms of the agreement involves
loss of reputation and, presumably, loss of future opportunities. Thus, it is in the
interest of both parties not to break their agreement in the first place, but to work
things out between themselves, even in the absence of a formal legal structure.
27.3.4. Tort Law

Tort law is civil law, allowing private parties to access government-provided
mechanisms and procedures without involving government as one of the con-
tending parties. Tort is an intentional or unintentional personal injury or other
harm stemming from an event other than breach of contract. A transoceanic oil
tanker, for example, that accidentally crashes and spills massive amounts of its
contents on foreign soil, would constitute an unintentional tort in an interna-
tional context. In contrast, one country blockading another country’s coastline in
an effort, for example, to extract bounty or to commit an atrocity crime like delib-
erately inducing mass starvation, represents an international intentional tort. In
either case, harm is caused. Tort law governs such harm and serves two primary
purposes, namely, to deter harm and to provide insurance. (Retributive justice
may be a tertiary purpose, but this is not especially connected to law and econom-
ics analysis.) As under the liability rule in property law, the law here strives to
deter future injury by making injurers pay money damages ex post. Ex post com-
pensation attempts to put victims in a position as if the injury had not occurred,
thus providing implicit insurance that can make victims “whole.”
The US Alien Tort Claims Act of 1789 states that “[t]‌he district courts shall
have original jurisdiction of any civil action by an alien for a tort only, commit-
ted in violation of the law of nations or a treaty of the United States.” In recent
decades, courts have interpreted the Act to permit foreign citizens to seek rem-
edies in US courts for human rights violations for conduct committed outside
the United States. While plaintiffs thus won, in February 2010, a case brought
in the Southern District of Florida against Charles Taylor Jr.—t he son of former
Liberian President Charles Taylor—and awarded them damages of over USD
22 million on April 17, 2013, in a case of alleged corporate liability under the Act,
the US Supreme Court held that the Act did not create jurisdiction for a claim
regarding conduct occurring in the territory of a foreign sovereign.
Both intentional and unintentional torts are of interest to genocide schol-
ars inasmuch as deliberate acts of causing harm are frequently accompanied by
unintended collateral damage. As with contract law, risk and attitudes/behaviors
toward risk therefore loom large in tort law. Because accidents or collateral harm
are stochastically (randomly) determined and thus defy prediction down to each
particular incident or individual, natural and legal persons take out insurance to
guard against the risk of being held liable. But anytime insurance is involved, one
must be concerned with moral hazard, the phenomenon by which a rational actor,
once insured, is led at the margin to act more recklessly than otherwise. In the eco-
nomic approach, injuries do not simply flow from injurer to victim, but the behav-
ior of both actors—potential injurer and potential victim—is thought to affect the
likelihood and severity of the injury. In the atrocity crimes context, this raises the
highly contentious issue of possibly blaming the victims for misfortunes they suffer.
(Might they have been able to protect themselves or did they negligently contribute
to their suffering? Is there, or should there be, a market for “genocide insurance,”
much as there has been a debate regarding a market for “terrorism insurance”?)
Even though collateral risk is randomly distributed, whether related to atroc-
ity crimes or otherwise, the average or typical attitude and behavior toward risk
of a “representative,” rational actor in an economic model is purposeful and thus

lends itself to prediction. When tort includes the loss of human life, two distinct
literatures address this aspect. First, compensation is owed in the death of a par-
ticular, identifiable individual. Compensation is paid by the tortfeasor (injurer),
depending on the rules of the particular jurisdiction, to either the estate of the vic-
tim or to the specifically designated statutory beneficiaries, typically immediate
family members and sometimes more distant next of kin. The amount of compen-
sation is based on the expected future earnings of the decedent absent the death
event, reduced to take account of the portion of those earnings that would have
been self-consumed. In the subfield of forensic economics, two journals (Journal
of Forensic Economics and Journal of Legal Economics) exist in part to provide guid-
ance on how best to estimate such losses.
A second literature delves into loss of human life but has nothing to do with
compensation ex post. Instead, it attempts to put a dollar value on the loss of a
statistical life (i.e., a random draw from the population) for the purpose of subse-
quently applying cost-benefit analysis to deciding how safe to make our roads,
buildings, and cars ex ante. No person, in ordinary day-to-day living, behaves
as if his or her own life is infinitely valuable. A risk of death, however slight, for
example, is taken on whenever a person travels from point A to point B to achieve
some related finite benefit, a risk that would be shunned as unacceptable if con-
tinuation of life were indeed infinitely valuable to the person. By exploiting the
fact that, everything else equal, riskier jobs pay more than safe jobs, or by look-
ing at people’s willingness to pay for added safety features in automobiles, the
value of a randomly selected person’s life has been estimated in recent years in
the range of high single-d igit millions of US dollars.15
By way of application of these distinct literatures, war reparations might
include compensation to surviving family members methodically calculated
based on expected net earnings of decedents absent their death events (or per-
haps by a formulaic process akin to that used for victims of the September 11,
2001 attacks on the United States). In contrast, a decision to enter a multinational
agreement to drain marshland to prevent malaria might be tipped by the implicit
value attached to each statistical life to be saved. Such thinking can be extended
to the creation or modification of atrocity crime prevention law and treaties. How
to value lives lost in atrocity crimes, and thus compensate surviving relatives, and
how to value statistical lives saved to subsequently apply cost-benefit analysis to
designing atrocity prevention mechanisms ex ante, can be advanced using well-
established methods from law and economics.
27.3.5. Criminal Law

Criminal law has at its heart a deterrence motive designed to inhibit criminal
acts. In the economic approach, crime is regarded as a rational (purposeful)
choice, given the opportunity set confronting the individual. Becker (1995, 9, 15)
observes that the economic approach to crime is “amazingly simple,” with crimes
carried out (or not) by people who base their decision on the benefits and costs of
undertaking crime; thus “crime is not inevitable,” and is instead the consequence
of public policies “not only about police and prisons, but about education and
a number of other things.” Punishment may be meted out to accomplish either
specific deterrence (preventing recidivism) or general deterrence (making an
example of one to inhibit similar action by another). Criminal law enforcement,
including punishment, is costly, however, and policymakers need to weigh the net
benefit to society. If resources spent on enforcement far outpace the value of the
last crime committed, the law overdeters. Thus, an optimal level of law enforce-
ment exists in societies, yielding a residual level of crime that economists would
also deem optimal.
A major difference between criminal law and the areas of property, contract,
and tort law is that a criminal court action is brought by the state, as opposed
to a civil action brought by a harmed individual. The harm done is perceived to
threaten society at large, hence the perceived need for the sovereign to step in and
prosecute an alleged wrongdoer on behalf of all its citizens.16 Another important
difference is that the consequence of a criminal proceeding may involve penal
incarceration apart from any ordered monetary payment. Thus, prior to the afore-
mentioned civil suit brought against Charles Taylor Jr., he had already been tried,
convicted, and imprisoned under US criminal law.
One area of domestic criminal law that has parallel application to genocide in
an international context is domestic law governing so-called hate crimes, namely,
crimes directed at individuals of particular classes of people and motivated by
sheer malice toward the specific group. As with genocide, the criminal action is
motivated not by what a person has done but by who this person is and represents,
namely, a class or group of people (OSCE 2009). The state has a higher burden of
proof for a hate crime as opposed to crime motivated by greed or motivated by
animus toward a particular person (rather than toward a group). Domestic hate
crimes have been singled out for special criminal sanctions in part due to the pos-
sibility or likelihood of group retaliation and resultant civil unrest (Posner 2011,
298, citing Wisconsin v. Mitchell, 508 U.S. 476).
27.3.6. International Agreements

Just as ordinary contracts involve natural or legal persons, treaty agreements
involve states, but they differ from ordinary contracts in that there is no world gov-
ernment to oversee and enforce such agreements.17 To a large extent, therefore,
international agreements need to be designed to be self-policing or self-enforcing.
In fact, even under national law, many ordinary contracts are self-enforcing,
owing to “concern with reputation and because of the availability of self-help
remedies such as repossession and other forms of forfeiture and the withhold-
ing of reciprocal performance (or payment)” (Posner 2011, 175). To illustrate the
point, Posner mentions illegal cartel agreements, stabilized by threat of retalia-
tion by any one member on any other member who would violate the pact. While
reputation may lead to self-enforcement, the former does not guarantee the latter.
From an economic perspective, compliance with an agreement is determined by
the benefits and costs at hand, which may differ both among contemporaneous
members and from those in a previous or altogether different context in which
reputation as an adherent (or violator) was established (Posner 2011, 178). This
observation explains why a particular country may have a reputation for adhering
to some agreements while violating others (Posner 2011, 177 [citing Downs and
Jones 2002]).
International agreements may be bilateral, as are most contracts, or multi-
lateral. International conventions, like those governing war, the treatment of
prisoners, and the prevention of the use of chemical weapons, tend to be mul-
tilateral. Posner (2011, 175) attributes this to two reasons. First, a multilateral
agreement functions like tort protection (a right of protection against the world)
as opposed to contract protection (a right to money damages under conditions
of breach by a specifically named other party), and tort protection is particu-
larly prized when one is uncertain which party will become one’s future enemy.
Second, although the same protection as in a multilateral agreement could in
principle be achieved by a series of bilateral agreements, the multilateral agree-
ment economizes on the transaction costs involved. The costs of formulating a
multilateral agreement are not insubstantial, however, which is why according
to Posner agreements tend to cover a single topic (2011, 175), and do so recipro-
cally (2011, 176), as opposed to covering a range of different topics and doing
so by nonreciprocal means, such as requiring side payments. Of course, one can
find exceptions to these generalizations (e.g., the United States providing aid to
North Korea [in a failed attempt] to forestall development of nuclear weapons).
Posner (2011, 179) also observes that, in distinction to bilateral agreements like
contracts, which tend to have a short duration coincident with the agreed-to
term of performance, multilateral agreements tend to have longer or even open-
ended duration to avoid exceedingly costly renegotiations. Given the longer
terms of multilateral agreements, greater built-i n flexibility is desirable, such as
use of an adjudicating body that takes account of changing circumstances over
time. For example, the Treaty on the Nonproliferation of Nuclear Weapons, or
NPT for short, is subject to a review conference among parties every five years.
With over 190 treaty members, the review conferences are costly, costs that are,
presumably, outweighed by more than commensurate benefits. If such provi-
sion is not possible, then use of a variety of escape clauses coupled with only
mild sanctions for agreement violation can achieve permanence. High transac-
tion costs require correspondingly high benefits, and escape clauses are a form
of negative outcomes for potentially affected populations (i.e., they offer a lower
degree of protection).
Posner also observes, discouragingly, an unintended consequence of, for exam-
ple, a convention governing the humane treatment of prisoners of war. Humane
treatment lowers recruitment costs, and hence the overall cost of war (Posner
2011, 177). Thus, such conventions may unintentionally provoke an increase in
the likelihood of war. Conversely, Posner observes, as have others before him,
that so-called mutually assured destruction in the nuclear era between the former
Soviet Union and the United States may have preserved the peace by increasing
the costs of unlimited war.
Posner (2011, 177) also offers an unsettling prediction of how defection from
arms control agreements may occur in a general war: “In a limited war, one
expects oneself and one’s opponent to survive, and this reduces both the benefits
of winning and the costs of losing, and so [reduces] the benefits of obtaining an
advantage by violating the conventions. In a general war expected to lead to the
annihilation of one of the combatant regimes, the benefits of such an advantage are
greater, and so less compliance should be anticipated.” However, at least in regard
to the NPT, the UN writes that “[m]‌ore countries have ratified the NPT than any
other arms limitation and disarmament agreement, a testament to the Treaty’s
significance.”18
Finally, in a genocide or other mass atrocity context, it has been noted that
conventions that would require third-party intervention against an atrocity-
perpetrating state could provide an incentive for a rebel group to strategically
maneuver the state into committing atrocities because the intervention that
would follow could increase the relative power of the rebel group. Clearly, the
design of laws and institutions should take such moral hazards into account
(also see c hapter 7 in this volume).
27.4. Economics of International Law

Despite some cases of genocide and other atrocity crimes having been brought
to trial in national and international courts or tribunals—the Armenian tri-
als in Turkey, the Nuremberg trials, the Pinochet case, and tribunals regard-
ing Cambodia, Rwanda, and the Balkan wars of the 1990s—t he overall record
of reducing the risk of atrocity crimes appears only mildly encouraging. There
are several reasons for this. First, even assuming away issues of ignorance and
apathy, as a matter of economics, unilateral action runs into the problem of suf-
ficient scale; and multilateral, collective action runs into issues related to stra-
tegic behavior, free-r iding, coordination, agency, benefit appropriation, and cost
shifting. Even assuming that none of these pose a problem, all options rely on the
existence of well-codified and well-functioning norms and regimes of national
and international laws, or agreements, and their enforcement.19 Second, as a

matter of law, then, reducing atrocity crime risk is difficult because state sover-
eigns are cautious to accede to any international treaty that may later commit
them to undertake or refrain from actions (the expected cost must not exceed
the expected benefit) and because sovereigns generally do not cede jurisdiction
over nonstate atrocity crime actors to international bodies (e.g., Nigeria main-
tains jurisdictional prerogative over Boko Haram). (And if a nonstate actor were
to prevail in an internal conflict the actor may not be brought to justice at all.) And
third, as a matter of institutional design—our focus in this section—questions arise
as to what kind of bads atrocity crimes are in the first place and, correspondingly,
what kind of goods atrocity crime–related domestic and international laws are,
and how to best supply them (Williamson 1999).
Start with the “what kind” question. When classifying types of goods, econo-
mists consider two specific characteristics of the benefits that goods can yield to
users (consumers), exclusion and rivalry. Benefits are excludable when agents can
be shut out from accessing the good. For example, a bicycle owner can prohibit
(exclude) others from using the bicycle. Benefits are rivalrous when the good can
effectively be enjoyed by only one person at a time. For example, when one per-
son rides a bicycle, another cannot ride it at the same time. A good meeting these
characteristics is called a private good. In contrast, a radio station’s FM broadcasts
are nonexcludable (no one can feasibly be prevented from tuning into the broad-
cast) and nonrivalrous (many people can listen to the broadcast at the same time).
The benefit is “lumpy,” and cannot be divided into distinct “chunks” for sale to
individual listeners. The station’s broadcasts are called public goods (not meaning
publicly provided, but meaning jointly consumable goods). A third class consists
of the somewhat ill-named common-pool resource goods (nonexcludable and rival-
rous). An emergency door at a dance club is an example. In principle, no one is
excluded from accessing the door when the need arises, but crowding at the door
effectively limits its benefits and only a few may escape in case of fire. A fourth
class is club goods (excludable and nonrivalrous), like a neighborhood swimming
pool. Several swimmers can enjoy the water simultaneously, but an access fee
excludes nonpayers. Goods can be blended to varying degrees along the exclud-
ability and rivalry dimensions. An impure public good thus exhibits degrees of
nonexcludability and nonrivalry, but not perfectly so.
The thinking can be applied to atrocity crimes, in both regular (goods) and
inverted (bads) form. To illustrate, consider the example of indiscriminate chem-
ical weapons gassing. If it is infeasible to exclude oneself from being gassed (the
gassing is nonexclusionary), and if the amount of gas inhaled by one person does
not appreciably reduce the amount of gas inhaled by another (inhalation is nonri-
valrous), one would conceptualize gassing as a public bad. Those who possess gas
masks enjoy them as private goods (exclusionary and rivalrous). If there is sealed
shelter available, those who manage to get in enjoy it as a club good (exclusionary
and, once inside, nonrivalrous). In contrast, genocide would be a club bad precisely
because its architects differentiate and select a specific group of victims: the bad
is nonrivalrous (all in the selected group partake in its imposed “consumption”)
and exclusionary (members of unselected groups do not become victims). The
rounding up of 500 villagers and randomly shooting 50 of them is an example of
a common-pool resource bad. No one is excluded as a potential victim (nonexclud-
ability), but only some are in fact shot “as if ” they had jostled to be killed (rivalry).
Examples of a private bad suffered in violent conflict include isolated instances of
unorchestrated rape in war or the death of a soldier in the performance of his or
her duties (the “expected” bad in war, but not a war crime).
In regard to the good that atrocity crime–related law may provide, interna-
tional law of war is generally intended as a public good, indeed, a global public good.
In the case of the Rwandan genocide of 1994, the UN Genocide Convention was
in force, but if any one party had intervened at its own cost to prevent or mitigate
the genocide, all other UN members would have benefited at zero cost of contri-
bution and with zero risk of exclusion from the receipt of these benefits (nonrival-
rous and nonexclusionary). This gives rise to the classic free-r ider problem: every
member waits for any other member to intervene. Moreover, no one UN member
had enough private interests at stake so as to bear the cost alone. In contrast, in
the case of Chad and western Sudan (i.e., Darfur), the government of Chad had a
clear private interest to prevent atrocity crimes in Sudan in order to forestall the
possibility of large numbers of refugees crossing the border and burdening Chad.
(Alas, Chad did not have the resources to act on its private interest.)20 The case of
European Union participation in the intervention in the Balkan wars of the 1990s
is one of an impure public good. Some benefits were private to the intervening states
(namely, to reduce costly refugee flows to France, Germany, Italy, etc.), but the
benefits of reduced refugee flows also accrued to states that made no or few con-
tributions to ending the wars (nonexcludable benefits).
Even this cursory “walk around goods space” (Brauer and van Tuyll 2008) sug-
gests that the good or bad in question may change its particular form depending
on the circumstances, geographic space, and time. An atrocity can morph from
a public bad at one place and time to a club bad at another, and a common-pool
resource bad at yet another. The point of the exercise is to suggest that neither
atrocity crimes nor interventions to deal with them necessarily share unitary
exclusion and rivalry characteristics and may therefore need to be dealt with dif-
ferently in policy and law. To conceive of atrocity crime simply as “the” global
public bad requiring “the” global public good response is inadequate. Moreover,
as Shaffer (2012) points out, global public goods such as international laws can
be rivalrous to each other and their construction is designed, in part, to trade off
against multiple national laws (legal pluralism).
Now turn to the “how best to supply the goods” question, that is, the supply
side. Are atrocity crime–prevention goods best provided by private or by public
actors, or by some changing combination of the two? What is the technology

of their production (e.g., best-shot, weakest-link, aggregate effort, or variants
thereof)? What sort of issues in agency, transaction costs, and institutional design
arise? While a considerable global public goods (GPG) literature has sprung up in
economics (e.g., Kaul 2006 and literature cited therein), application to the design
of international agreements as an instance of GPGs is thin in general and almost
entirely absent in regard to atrocity crime law specifically (see, e.g., a recent sym-
posium of papers in the European Journal of International Law 23, no. 3 [2012]).
For example, the economically efficient provision of atrocity crime law in
response to such crimes may depend on the summation technology of its produc-
tion and enforcement. Applying Hirshleifer’s (1983) insight—t hat some GPGs
are best provided as best-shot products (the single-best effort suffices; no need
for anyone else to contribute to its provision), weakest-link products (the weak-
est provider limits the good’s effectiveness), or aggregate effort products (the
more is provided by all, the better for all)—Shaffer argues that best-shot GPGs
are best dealt with in global administrative law, weakest-link GPGs by fostering
legal pluralism, and that only aggregate effort GPGs may require a global consti-
tutionalist approach (Shaffer 2012, esp. 690, Table 2). To illustrate, when a single
state has effectively become the world’s only superpower to intervene in other
states’ actual or alleged atrocity crimes, it may be tempted to overreach or under-
reach according to its own cost-benefit calculation, regardless of the wishes of
all other UN members. Superpower intervention or nonintervention solely at its
own discretion, however, challenges global legitimacy. (The United States is often
accused in this regard; France, in regional interventions, less so.) Such situations,
Shaffer (2012) argues, are best dealt with by global administrative law, which
might hold the incumbent of the superpower office responsible for its actions. We
imagine (since Shaffer does not directly address atrocity crimes) that, instead of
a UN Genocide Convention with its free-r ider problem, there might exist a UN-
approved automatic trigger obligating the superpower (or a standing UN force)
to intervene in cases of atrocities, subject to global administrative law. As of this
writing, little has been theorized in this regard.
An additional issue pertains to transgenerational global public goods, also
insufficiently theorized but probably of great importance in cases of atrocity
crimes since each event carries significant generational implications (see c hapters
4 and 11 in this volume).21 For public goods provision, Sandler (1999) speaks for
four levels of awareness rules. First, a myopic view considers making a marginal
cost (MC) contribution to the provision of a GPG only up to the sum of the mar-
ginal benefits (MB) a state expects to receive for its own current generation, so that
MC = ΣMB. Second, although still selfish, a forward-looking view is to include
one’s own offspring generations, i, such that MC = ΣMBi. Since the expected ben-
efits are larger when more generations are considered, this translates into greater
willingness to make a larger MC contribution. Third, a more generous view of
benefits summation includes other states’ populations, j, but only for the current
world generation (MC = ΣMBj). The most enlightened view of all—we call this
the Buddha rule—sums the expected benefits across all generations across all
populations, MC = ΣMBij. Since the sum total of such benefits is likely to be large,
it justifies correspondingly large outlays to fund the provision of the good. Design
criteria for GPGs that would take account of goods-(or bads-) spaces, summation
technologies, and transboundary and transgenerational aspects have been dis-
cussed in the literature (e.g., Sandler 1997) but rarely in regard to atrocity crime–
related national and international law (chapter 28 in this volume is an exception).
All this opens up the possibility of a great international legal experiment
involving the codependency of certain treaties, conventions, and agreements in
national and international law.22 This might involve stating outright in the law that
a state party to a statute would gain access to particular economic arrangements,
trade agreements for example, that it would not have otherwise. This would be a
privately appropriable benefit for the ratifying state, a design in furtherance of
the purpose of the law: To realize benefit capture in one context requires a state’s
commitment to renounce and forgo undesired behavior in another. Since this
benefit is alienable (revocable), it also serves as a credible threat in case of non-
compliance. For example, linking the Rome Statute to participation in specific
international trade agreements would be advantageous to any state party. This
presupposes that the state is not already party to such agreements, but is meant
here as an example of how the Statute might have been designed. Yet note that this
is what the European Union (EU) already does with new member applicants—
allowing access to the political, economic, and cultural benefits of becoming part
of the EU but requiring a certain level of adherence to human rights norms and
laws. Despite imperfect implementation, dangling the carrot of EU membership
is an example of a benefit available for adhering to obligations concerning human
rights. Similarly, the design of the African Union mandates automatic member-
ship suspension, and suspension of the benefits of membership, in case of any
member experiencing a nondemocratic regime change. While implementation
and enforcement of these ideas are another matter, they seem to us to point in the
right direction, infusing law and treaty design with an economic analysis of incen-
tive structures. (On benefits linking and benefits capture, see, e.g., Sandler 1997.)
27.5. Conclusion
The chapter reviewed some concepts of the economic analysis of law and applied
them to unconventional examples related to atrocity crimes. It also reviewed con-
cepts related to global public goods, club goods, common-pool resource goods,
and private goods (and bads) and suggested that viewing atrocity crimes uni-
formly as global public bads, and international treaties aimed at their prevention
uniformly as global public goods, may be misplaced and may lead to unwarranted,
ineffective, and inefficient policy prescriptions. A more differentiated view of the
“bads-ness” of the crimes and the “goods-ness” of the proposed remedies in law,
as well more discerning thinking regarding the provision of such goods, may well
be warranted.
Perhaps the primary takeaway from this chapter is that there exists a capable
cohort of law and economics and other scholars who, if they extended and applied
their learning specifically to issues of atrocity crimes, might well generate cre-
ative and feasible ideas regarding the effective design of legal instruments and
institutions to prevent (or at least to mitigate) any such crimes.
Acknowledgments
With kind permission of Springer Science+Business Media, selected portions of
sections 27.1 and 27.2 of this chapter are paraphrased from the entry “Genocide”
by Charles H. Anderton and Jurgen Brauer, Encyclopedia of Law and Economics,
2015 (February 1, 2015; latest version), © Springer Science+Business Media,
New York. DOI 10.1007/978-1-4 614-7883-6 _581-1. Online ISBN 978-1-4 614-
7883-6. Section 27.3 is entirely new, and section 27.4 has been significantly modi-
fied and expanded. In addition, we are grateful to Samantha A. Capicotto, Esq.,
Director of Policy and Planning and Program Director, Raphael Lemkin Seminar
for Genocide Prevention, Auschwitz Institute for Peace and Reconciliation,
New York, for taking the time to comment on an earlier draft of this chapter. All
errors are, of course, our own.
Notes
1. This has been referred to as cultural genocide (Tatz 2013). Also see the May 2015 Executive
Summary of a Report by the Truth and Reconciliation Commission of Canada, whose first
sentence reads: “For over a century, the central goals of Canada’s Aboriginal policy were to
eliminate Aboriginal governments; ignore Aboriginal rights; terminate the Treaties; and,
through a process of assimilation, cause Aboriginal peoples to cease to exist as distinct
legal, social, cultural, religious, and racial entities in Canada. The establishment and opera-
tion of residential schools were a central element of this policy, which can best be described
as ‘cultural genocide.’ ” See http://w ww.trc.ca/websites/t rcinstitution/i ndex.php?p=890
[accessed June 18, 2015].
2. The Rome Statute of 1998 provides for a fourth crime, the crime of aggression, but notes
that the Court does not have jurisdiction until the crime is defined and the conditions of
jurisdiction are set out. In 2010, the first-ever Rome Statute review conference was held
(in Kampala, Uganda) at which “ICC States Parties agreed upon a jurisdictional regime
for the crime of aggression” but also “determined that the activation of jurisdiction” will
have to wait at least until 2017 (see http://w ww.iccnow.org/?mod=review; accessed June
18, 2015). The removal of people of a particular group from a state or region using means
such as forced migration is ethnic cleansing (Pégorier 2013); in contrast to genocide the ele-
ment of intent to destroy is missing. “Ethnic cleansing” has become a popular term but it
is not a legally defined atrocity crime under the Rome Statute. An additional category that
some researchers employ is violence against civilians (or VAC), which includes “smaller” mass
atrocities, specifically, those with fewer than 1,000 afflicted civilians per case or per year.
Although also not legally defined, this can, and does, include cases of genocides and mass
killings. The Uppsala Conflict Data Program (UCDP) defines one-sided violence against
civilians as “the use of armed force by the government of a state or by a formally organized
group against civilians which results in at least 25 deaths. Extrajudicial killings in custody
are excluded” (Pettersson 2012, 2; also see Eck and Hultman 2007).
3. See Article 6 of the IMT founding charter, http://avalon.law.yale.edu/i mt/i mtconst.
asp#art6 [accessed June 18, 2015].
4. Posner (2011, 33–3 4) indicates that the economic approach to law is not exclusively an
Anglo-A merican exercise, but can be applied to the Continental (European) tradition that
emphasizes a legislative code as the source of law, yielding far less of a role for judges as
policymakers. See also, e.g., Mackaay (2014).
5. Under the more well-k nown Pareto efficiency criterion, an outcome is more efficient if at
least one person is made better off, and strictly no one is made worse off. In practice, how-
ever, it is almost impossible to take any social action, such as a change in law or economic
policy, without making at least one person worse off. The Kaldor-H icks criterion is less
demanding in that it requires only an “in principle” compensation, not an actual compensa-
tion. Thus, under Kaldor–H icks more efficient outcomes can leave some people worse off.
6. For example, under certain circumstances, Kaldor-H icks can be shown to violate transitiv-
ity. For more on efficiency in law and economics, see, e.g., Zerbe (2014).
7. Other examples of areas of law and economics that may be connected to the topic of
atrocity crimes include constitutional law, administrative law, and international treaties.
Subsection 27.3.6 and section 27.4 deal with aspects of treaty and administrative law. On
the importance of constitutional law, see, e.g., chapter 28 in this volume.
8. The pioneering work was Coase (1960), with the term Coase Theorem evolving subsequently.
9. On identity, see c hapters 12–14, 17, 21, and 22 in this volume.
10. The rent-seeking concept was introduced by Tullock (1967); the specific term was intro-
duced by Krueger (1974). The concept and label of “directly unproductive, profit-seeking
activities” is due to Bhagwati (1982). On resource distribution, see, e.g., c hapter 9 in this
volume. On strategy and game theory, see c hapters 7, 19, 20, and 22, all in this volume.
11. The analysis may be found, in decreasing order of complexity, in Calabresi and Melamed
(1971), Krauss (1999), and Solum (2014).
12. Recently, there has been some relaxation in laws prohibiting the sale of blood or vital
organs, as some economists have argued in favor of greater reliance on markets rather than
donation and nonmarket allocation schemes (citing, for example, greater quality assur-
ance for blood, and shorter wait times and better matches in organ transplant surgery).
For more on quasi-markets for vital organs, see Barnett and Kaserman (1995) and Beard,
Kaserman, and Saba (2006).
13. The practice of asset seizure domestically in the United States, however, has been criticized
as having been applied too readily in many cases of questionable guilt, as a means of financ-
ing police authorities, since the burden for retrieval of seized assets falls on the accused
(a perverse inversion of the ordinary constitutional protection of private property from
government seizure).
14. Preserving the benefit of the bargain is but one of several ways in which the amount of
money damages may be formulated in cases of contract breach.
15. For example, the US Transportation Department has it at USD 9.4 million in 2015 (http://
www.dot.gov/officepolicy/t ransportation-policy/g uidancetreatment- e conomic-v alue-
statistical-l ife). In 2010, the US Environmental Protection Agency cited a figure of USD
9.1 million and the U.S. Food and Drug Administration proffered a figure of USD 7.9 mil-
lion (http://w ww.nytimes.com/2 011/0 2/17/ business/e conomy/17regulation.html?_
r=1&pagewanted=all).
16. In contrast to the perception that sovereign action is necessary, Benson (1994, 250) explains
that in England private policing predated police services provided by the Crown: “In the
case of policing, for example, before English kings began to concentrate and centralize
power, individuals had rights to a very important private benefit arising from successful
pursuit and prosecution: victims received restitution. Effective collection of restitution
required the cooperation of witnesses and of neighbors to aid in pursuit; but anyone who
did not cooperate with victims could not obtain similar support when victimized, and
therefore could be excluded from this very important benefit of law enforcement.”
17. This section borrows heavily from Posner (2011, 174–79).
18. See http://w ww.un.org/disarmament/W MD/Nuclear/NPT.shtml [accessed June 12, 2015].
19. In this section, we use international agreements, laws, and treaties as synonyms.
20. Benson (1994) considers the public goods argument for police and highways to be an “ex
post justification for claiming that the only efficient policy is public provision of these
services at zero marginal prices” (249), arguing essentially that what was once privately
provided was subsequently co-opted by the state. A related argument has been made by
Brauer (1999) and Brauer and van Tuyll (2008) in regard to the employment of private mil-
itary companies to intervene in cases of atrocity crimes. A government that contracts for
its protection with a private military company acquires a private good, but international
disapproval of “mercenary” firms has made these contracts odious. By credible threats of
sanction, other states in the international system effectively compel vulnerable states to
rely on the provision of a global public security good that may not arrive in time and suf-
ficient force. Examples include the civil wars in Angola and Sierra Leone in the 1990s, both
of which were tamped down when private military companies were engaged, and both of
which flared up viciously after their withdrawal.
21. The idea of transgenerational goods probably goes beyond Posner’s intent when he dif-
ferentiates (2011, 176) between short-and long-duration international agreements, but
his general point carries: The costlier negotiation of long-duration agreements, let alone
transgenerational ones, requires commensurate higher benefits. A “negative” form of high
benefits, from the point of view of potentially threatened populations of a ratifying state,
would be an escape hatch large enough to drive genocide through!
22. We are grateful to Samantha A. Capicotto, Esq., for suggesting this paragraph.
References
Adalian, R. P. 2013. “The Armenian Genocide.” In S. Totten and W. S. Parsons, eds., Centuries of
Genocide: Essays and Eyewitness Accounts. 4th ed. London: Routledge, 117–55.
Barnett, A. H., and D. L. Kaserman. 1995. “The ‘Rush to Transplant’ and Organ Shortages.”
Economic Inquiry 33, no. 3: 506–15.
Beard, T. R., D. L. Kaserman, and R. P. Saba. 2006. “Inefficiency in Cadaveric Organ
Procurement.” Southern Economic Journal 73, no. 1: 13–2 6.
Becker, G. 1995. “The Economics of Crime.” Cross Sections (Federal Reserve Bank of Richmond)
(Fall): 8–15.
Benson, B. 1994 “Are Public Goods Really Common Pools? Considerations of the Evolution of
Policing and Highways in England.” Economic Inquiry 32, no. 2: 249–69.
Bhagwati, J. N. 1982. “Directly Unproductive, Profit-Seeking (DUP) Activities.” Journal of
Brauer, J. 1999. “An Economic Perspective on Mercenaries, Military Companies, and the
Privatisation of Force.” Cambridge Review of International Affairs 13, no. 1: 130–4 6.
Brauer, J., and H. van Tuyll. 2008. Castles, Battles, and Bombs. Chicago: University of Chicago
Press.
Calabresi, G., and A. D. Melamed. 1971. “Property Rules, Liability Rules, and Inalienability: One
View of the Cathedral.” Harvard Law Review 85, no. 6: 1089–1128.
Cernic, J. L. 2010. Human Rights Law and Business: Corporate Responsibility for Fundamental Human
Rights. Groningen, The Netherlands: Europa Law Publishing.
Coase, R. H. 1960. “The Problem of Social Cost.” Journal of Law and Economics 3 (October): 1–4 4.
Coase, R. H. 1992. “The Institutional Structure of Production.” American Economic Review 82,
no. 4: 713–19.
Curthoys, A., and J. Docker. 2008. “Defining Genocide.” In C. Stone, ed., The Historiography of
Genocide. New York: Palgrave Macmillan, 9–41.
Dadrian, V. N. 1997. “The Turkish Military Tribunal’s Prosecution of the Authors of the
Armenian Genocide: Four Major Court-Martial Series.” Holocaust and Genocide Studies 11,
no. 1: 28–59.
Downs, G. W., and M. A. Jones. 2002. “Reputation, Compliance and International Law.” Journal
of Legal Studies 31: S95–S114.
Hillemanns, C. F. 2003. “UN Norms on the Responsibilities of Transnational Corporations
and Other Business Enterprises with Regard to Human Rights.” German Law Journal 4,
no. 10: 1065–80.
Hirshleifer, J. 1983. “From Weakest-Link to Best-Shot: The Voluntary Provision of Public
Goods.” Public Choice 41, no. 3: 371–86.
Kaul, I., and P. Conceição, eds. 2006. The New Public Finance: Responding to Global Challenges.
Kelly, M. J. 2012. “Prosecuting Corporations for Genocide under International Law.” Harvard
Law and Policy Review 6, no. 2: 339–67.
Krauss, M. I. 1999. “Property Rules vs. Liability Rules.” In B. Bouckaert and G. De Geest, eds.,
Encyclopedia of Law and Economics. Northampton, MA: Edward Elgar. http://encyclo.find-
law.com/3800book.pdf.
Krueger, A. O. 1974. “The Political Economy of the Rent-Seeking Society.” American Economic
Review 64, no. 3: 291–303.
Lunga, L. 1992. Individual Responsibility in International Law for Serious Human Rights Violations.
New York: Springer.
Mackaay, E. 2014. Law and Economics for Civil Law Systems. Northhampton, MA: Edward Elgar.
Michalowski, S., ed. 2013. Corporate Accountability in the Context of Transitional Justice. New York:
Routledge.
[OSCE] Organisation for Security and Cooperation in Europe. 2009. Hate Crime Laws: A Practical
Guide. OSCE Office for Democratic Institutions and Human Rights. Warsaw: OSCE.
www.osce.org/odihr/36426?download=true [accessed June 24, 2015].
Pégorier, C. 2013. Ethnic Cleansing: A Legal Qualification. New York: Routledge.
Pettersson, T. 2012. “UCDP One-Sided Violence Codebook Version 1.4.” Department of Peace
and Conflict Research, Uppsala University.
Posner, R. A. 2011 [1973]. Economic Analysis of Law. 8th ed. New York: Aspen.
Rummel, R. J. 1998. Statistics of Genocide: Genocide and Mass Murder since 1900. Piscataway,
NJ: Transactions.
Sandler, T. 1997. Global Challenges. New York: Cambridge University Press.
Sandler, T. 1999. “Intergenerational Public Goods: Strategies, Efficiency and Institutions.” In

I. Kaul, I. Grunberg, and M. A. Stern, eds., Global Public Goods: International Cooperation in
the 21st Century. New York: Oxford University Press, 20–50.
Schabas, W. A. 2003. “National Courts Finally Begin to Prosecute Genocide, the ‘Crime of
Crimes.’” Journal of International Criminal Justice 1, no. 1: 39–63.
Schabas, W. A. 2010. “The Law and Genocide.” In D. Bloxham and A. D. Moses, eds., The Oxford
Handbook of Genocide Studies. New York: Oxford University Press, 123–41.
Shaffer, G. 2012. “International Law and Global Public Goods in a Legal Pluralist World.”
European Journal of International Law 23, no. 3: 669–93.
Solum, L. 2014. “Legal Theory Lexicon: Property Rules and Liability Rules.” Legal Theory Blog.
February 16. http://lsolum.typepad.com/legaltheory/2 014/02/legal-t heory-lexicon-
property-r ules-l iability-r ules.html.
Tatz, C. 2013. “Genocide in Australia.” In S. Totten and W. S. Parsons, eds., Centuries of
Genocide: Essays and Eyewitness Accounts. 4th ed. New York: Routledge, 54–87.
Tullock, G. “The Welfare Costs of Tariffs, Monopolies and Theft.” Western Economic Journal 5,
no. 3: 224–32.
Waller, J. 2007. Becoming Evil. 2nd ed. New York: Oxford University Press.
Williamson, O. 1999. “Public and Private Bureaucracies: A Transaction Cost Economics
Perspective.” Journal of Law, Economics, and Organization 15, no. 1: 306–42.
Zerbe, R. O., ed. 2014. Efficiency in Law and Economics. Northhampton, MA: Edward Elgar.
28
Local and National Democracy

in Political Reconstruction
Roge r B. M y e r son
28.1. Introduction
Too often, bright hopes for new democracies have faded, and we need to under-
stand why. In Egypt, for example, public demands for democratically accountable
government in 2011 were followed by elections to choose a national assembly and
a president in 2012, but then the 2012 constitution offered only a vague promise
to introduce elected local governments sometime in the next decade. The down-
fall of the elected president in 2013 has led to questions about what went wrong
in the process of building a new democracy in Egypt. Many have asked whether
the nation might have moved too quickly into a presidential election, but few have
asked whether the move to introduce democratic local government was too slow.
The purpose of this chapter is to examine such questions from a basic theoretical
perspective.
When a nation is beginning a transition to democracy after a conflict or break-
down of the state, others in established democratic countries may naturally want to
offer their help and support, but we need to think deeply and carefully about how to
help. Effective assistance in postconflict political reconstruction must depend on
our fundamental understanding of how successful democratic societies are devel-
oped. We must try to understand the foundations of a democratic state. As in other
forms of construction, the chances of success in national political reconstruction
can be improved only with some understanding of what makes a stable political
architecture and in what order should its structural elements be introduced.
When we live in a successful democratic society, we are surrounded by a com-
plex system of political, legal, economic, and social institutions, each of which
seems to depend on many of the others. When these institutions do not exist or
are not functioning, which institutions must be established first to begin mov-
ing from anarchy toward prosperity? I approach this question as a theorist in
663
economics and political science, with an understanding that there is no greater

question in social theory. From this perspective, I would argue that the founda-
tions of democratic development depend on both local and national politics, and
that elections for local councils may be as important in political reconstruction as
elections for leadership of the national government.
A common tendency to focus only on development of national democracy may
have several causes. News media regularly focus on national elections as the most
important political events in established democracies, and so they may naturally
focus also on national elections in a newly democratic state. Coordinators of
international assistance programs are accustomed to working with national gov-
ernments, and so they may put priority on establishing a national government to
be their partner. Most importantly, the major leaders who hope for national power
in the newly reconstructed state may acquire a vested interest in the centraliza-
tion of power, and so they may resist any devolution of power to independently
elected local governments.
There are, however, strong reasons to suggest that those who would truly sup-
port a cause of democratic political reconstruction should encourage a balanced
development of democratic government at several levels, from local to national.
Indeed, one could even argue for the priority of introducing democratic govern-
ment at the municipal and provincial levels before national elections, as actually
occurred in American history. This chapter offers a summary overview of such
arguments (Myerson 2014, 2015a). (For other general perspectives on state-
building see Dobbins et al. 2007 and Ghani and Lockhart 2008.)
Of course, every nation’s political development must depend on its own tradi-
tions and culture. In this regard, the most important aspect of culture is what
people expect of their leaders. In a nation where history has led people to expect
that any political leader would maintain a corrupt patronage network that may
violate laws with impunity, this expectation is likely to be fulfilled. For a success-
ful transition to democracy, people must somehow come to expect better public
service from their leaders. But an incumbent national leader has no incentive to
raise voters’ expectations, and voters have no reason to believe promises of bet-
ter government from a candidate who has never exercised power. An incentive to
raise expectations may be combined with the potential to do so only when vot-
ers allocate power at two or more levels of government, so that officials at lower
levels can hope for election to higher office if they demonstrate an ability to serve
the voters better. This is our basic argument for democratic decentralization.
28.2. Leaders’ Reputations and the Foundations

of the Constitutional State
The foundations of the state are not just an abstract topic of study for political
philosophers or organizers of state-building assistance missions. The practical
L o c a l a n d N a t i o n a l D e m o c ra c y i n Po l i t i c a l R e c o n s t r u c t i o n 665
problem of establishing the state must be solved anew in every generation by

political leaders who rise to the summit of political power, whether by election
or conquest. Such leaders know the vital importance of maintaining good reputa-
tions with their supporters and with the broader public.
To compete for power in any political system, a leader needs to build a base of
active supporters, and the essential key to motivating this base is the leader’s repu-
tation for reliably distributing patronage benefits to loyal supporters. Any leader
needs supporters to trust his or her promises that their service will be rewarded.
We cannot expect a leader to do anything that would cause his supporters to lose
this basic confidence.
Thus, to maintain this essential trust of their supporters, leaders at all levels
are fundamentally constrained by cultural norms that define what their factional
supporters should expect of them. Constitutional constraints on powerful leaders
can be enforced by such norms when a leader who violated the constitution would
shock his supporters and so would risk losing their trust (Myerson 2008).
In power, a successful leader must oversee a government that is composed of
people. The state is a network of agents who manage public resources and who
enforce the laws that can sustain property rights and help maintain trust in con-
tractual relations for people in other organizations of society. Agents of the state
could profit from abuse of their powers, and so they must be motivated by the
expectation of greater long-term rewards for good service. But promised rewards
for good service become a debt of the state that its leaders might subsequently
prefer to deny. The ultimate social function of political leadership is to maintain
trust of long-term incentives in the government itself (Myerson 2011).
Doctrine for state-building missions often emphasizes training for profes-
sional development of security forces and administrative agencies (US Army
and Marine Corps 2007). But incentives in such units and agencies ultimately
depend on political leadership. If political leaders do not support the standards for
evaluating and rewarding the service of professionals in public service, then these
standards cannot be maintained. Security forces, no matter how well trained,
cannot be expected to protect a community in the long run without a clear line of
accountability to the community’s political leaders. Thus, in answer to our basic
question of what comes first in building a successful democratic society, political
development should be seen as the essential first priority that is fundamental to
everything else in national reconstruction.
We have noted that any political leader must reliably protect the rights of
political supporters and government agents to enjoy their promised rewards.
The critical question of political economy, then, is whether property rights are
to be securely protected only for a small elite who actively support the national
ruler, or whether the protected circle of trust will extend more broadly to
include people throughout the nation. Members in the securely protected
group require some legal and political power that could be used against a
government official who failed to protect their rights. A broad distribution of
such power to threaten the privileged status of government officials may natu-
rally seem inconvenient to established national leaders, but people who have
been admitted into this circle of political trust can invest securely in the state,
increasing economic growth. A fundamental fact of modern economic growth
is that it requires decentralized economic investments by many individuals
who must feel secure in the protection of their right to profit from their invest-
ments. Thus, modern economic growth requires a wide distribution of political
voice and power throughout the nation.
Political systems can differ on at least two major dimensions that fundamen-
tally affect the distribution of power in a society: democracy and decentralization.
Democratic political systems distribute political voice more broadly in a nation
by making leadership of government dependent on free expressions of popular
approval from a large fraction of the nation’s citizens. Decentralized political sys-
tems distribute power more widely to autonomous units of local government at
the provincial or municipal levels.
Relationships between local and national political leaders are vital elements in
the structure of any political system. National leaders can wield their power only
with trust and support of local officials throughout the nation, and local leaders
in turn rely on national leaders to affirm their privileged positions of local power.
But under different constitutional systems, the primary leaders of local govern-
ment may be agents appointed by the national leadership, or they may earn their
positions by autonomous local politics. This distinction between centralized and
decentralized states should be seen as one of the primary dimensions on which
states vary, potentially as important as the distinction between democratic and
authoritarian states. Decentralized federal democracy and centralized unitary
democracy may have significantly different implications for economic and politi-
cal development. In particular, I will argue, political decentralization can sig-
nificantly increase the chances of s uccess for a new democracy (Myerson 2006).
28.3. Effective Democratic Competition Requires

More than Just Elections
The basic argument for democracy can be expressed by analogy with competi-
tion in economic markets. As profits motivate economic production in markets,
so privileges of power can motivate political efforts in government. But even a
benevolent autocrat would find it difficult to resist his courtiers’ urge for greater
privileges if further exploitation of the public would entail no risk of losing power.
Thus, one may argue, as competition in markets can limit producers’ profits and
yield better values for consumers, so democratic competition should limit elite
privileges and yield better government services for the public.
Even with free elections, however, a corrupt political faction could win reelec-
tion from the voters and maintain its grip on power if the voters believed that
other candidates would not be any better. Thus, a successful democracy requires
more than just elections. It requires alternative candidates who have good demo-
cratic reputations for using power responsibly to benefit the public at large, not
merely to reward a small circle of supporters. For democracy to be effective, vot-
ers must have a choice among qualified candidates with proven records of public
service who have developed good reputations for exercising power responsibly
in elected office.
However, a nation that has just emerged from autocratic rule and violent con-
flict is unlikely to have many widely trusted political leaders with such reputa-
tions for good public service. When such trusted alternative leadership is lacking,
national elections can become simple exercises in ratifying the authority of the
incumbent national leadership, with little effect on their incentives to serve the
public better.
The essential supply of trusted democratic leadership can develop best in
responsible institutions of local government, where successful local leaders can
prove their qualifications to become strong competitive candidates for higher
office. When locally elected leaders have some real responsibility for both the
successes and failures of their local administration, then those who succeed will
enlarge the nation’s vital supply of popularly trusted leaders. Thus, democratic
decentralization can be an effective way to ensure that national elections are truly
competitive and that their winners must act to earn the voters’ trust. The chances
for a successful transition to democracy should be greater if the first transitional
government includes locally elected councils that have substantial autonomous
responsibility for local public services.
As an application of this point, consider a situation where a new democratic
state has been established by a foreign state-building intervention and the foreign
interveners have selected the initial national leadership for the new state. If the
first national leader is the only one in the new state (since the expulsion of the
leaders of the old regime) who has had any opportunity to oversee public services
and develop a patronage network, then his victory in the first presidential election
will be very likely. After such an election, however, people would still understand
that the national leader has achieved supreme power, not by earning broad popu-
lar trust, but by foreign influence. Thus, in such a situation, a national presidential
election alone cannot prove that a state-building mission has established a truly
sovereign democratic state. To avoid such a conclusion, foreign interveners who
have pledged to rebuild a nation as an independent democratic state must develop
the nation’s supply of trusted democratic leadership, and they should do so by giv-
ing substantial responsibilities to elected local governments as soon as possible in
the transitional regime.
28.4. Advantages of Democratic Decentralization

We have argued that local democracy can help to make national democracy more
competitive, as a record of using public resources responsibly in local government
can qualify a local leader to become a competitive candidate for power at higher
levels of government. In effect, local democracy can reduce barriers against entry
into national democratic competition. (Economists understand that barriers
against new entrants can be important determinants of the level of profit-taking
by suppliers in an imperfectly competitive market.) From this perspective, an
ideal system of federal democracy should have several levels of subnational gov-
ernments, so that elected offices at different levels could form a ladder of demo-
cratic political advancement that effective leaders can climb from local politics to
provincial and national politics.
Conversely, the threat of small unrepresentative cliques or warlords dominat-
ing local governments can be countered by the participation of national political
parties in local politics. Local political bosses should know that, if they lose popu-
lar support, they could face serious challengers supported by a rival national party.
Thus, the introduction of democracy in different levels of government, from
local to national, can strengthen democratic competition at all levels. A multi-
party national assembly and elected local councils can together provide the insti-
tutional pillars on which a strong democratic system can be built.
Democratic decentralization can also help economic development by pro-
viding better mechanisms for local public investment. A poor community can
mobilize its resources for public investments that are essential for its economic
development only when members of the community are coordinated by local
leaders whom they can trust to appropriately reward contributors and discipline
free-r iders. Such trust can be expected only from leaders whose authority is based
in local politics. Local officials whose positions depend on national political
patronage are inevitably less concerned about developing trust among the resi-
dents of a small, poor community. Thus, integrated efforts to achieve economic
development throughout a nation may depend on a political system that admits
autonomously elected local leaders into the national network of power (Fortmann
1983; Myerson 2015b).
At any point in time, in any society, there are formal or informal structures
of local social leadership in all communities. When a state has failed, such local
leadership can become even more important to people as a source of basic pro-
tection. The long-term successful establishment of a political regime will depend
on its general recognition and acceptance by such local leaders in all parts of the
nation. If a new regime is endorsed by an overwhelming majority of local leaders
throughout the nation, then the others will feel compelled to acquiesce. But if
there are communities where the regime lacks any local supporters, then these
communities can become a fertile ground for insurgents to begin building a rival
system of power with encouragement from disaffected local leaders.
The regime’s constitutional distribution of power can determine how many
local leaders will find a comfortable place for themselves in the regime as well as
how many local leaders will feel excluded from power in it. Everyone understands
that in the long run, once a state is firmly established, it will be able to redefine and
redistribute positions of local leadership in the nation. When a state constitution-
ally devolves a share of power to locally elected officials, it gives these local leaders
a stake in the political system that they should be willing to defend. Thus, political
decentralization can actually strengthen a state against external challenges and
insurgency by ensuring that, throughout the nation, there are local leaders who
have a substantial interest in defending the state and who (by the fact of their
election) have a proven ability to mobilize local residents for political action.
28.5. Forces Against Decentralization

In spite of the aforementioned advantages, there can be powerful forces against
the introduction of democratic local government in nations where it has not previ-
ously existed. The potential of autonomous subnational governments to become
sources of new competition for national power is one important reason why estab-
lished national leaders might not want to share power with locally elected gover-
nors and mayors. A national leader who appoints and dismisses local governors
can use this power to prevent these positions from being used as proving grounds
for new competitors against him. In a centralized unitary state, a governor who
begins to develop an independent reputation for serving the public better than
the leader who appointed him should expect that the leader may soon replace him
with a loyal supporter who is less politically threatening.
The most important national leaders in a newly democratic nation will have
a vested interest in maximizing the power of the national government and so
may resist any devolution of power to independently elected local governments.
A national leader is likely to find particularly strong opposition to political decen-
tralization among his most valued supporters, those who may hope someday to be
appointed to an office of mayor or governor as a reward for their political support.
In a centralized state where governors are appointed by the national leader, these
powerful local offices can be among the most prized positions that the national
leader can use to reward loyal supporters. Then a move to let local voters select
their own mayors and governors would, from the perspective of these key sup-
porters, be a disappointing diminution of their potential rewards for good politi-
cal service. It is very risky for any political leader to do anything that would so
disappoint his most important supporters.
Thus, we have fundamental reasons to expect that political decentralization

may often be undersupplied, relative to what would be best for the general pop-
ulation, because it runs against the vested interests of those who hold power at
the national level. Evidence can be found to support this prediction. In Pakistan,
elected politicians of national and provincial governments have three times dis-
solved institutions of local democracy that had been created by military rulers
(Cheema, Khan, and Myerson 2015). In Egypt, three constitutions in a row
(those introduced in 1971, 2012, and 2014) promised an eventual devolution
of power to locally elected councils but then allowed current national leaders
to postpone such decentralization and continue the centralized appointment of
local governors.
28.6. Problems of Separatism and Ethnic Violence

Ethnic rivalries can complicate the problems of democratization in important
ways. When people feel divided by different ethnic or tribal identities, a local
leader who has governed well in one region may nonetheless be viewed with sus-
picion by people of other regions who have different ethnic identities. Thus, the
potential of democratic local governments to make national politics more com-
petitive may be seriously reduced when ethnic rivalries make it harder for a suc-
cessful local leader to present himself as a candidate whom voters throughout the
nation can trust. For example, Nigeria has had imperfectly competitive elections
at both the national and provincial levels since 1999. A few governors there seem
to have earned reputations for providing better local government, but their ability
to offer themselves as candidates for the national presidency has apparently been
limited by interregional suspicions in Nigeria.
Varshney (2002) has emphasized that ethnic violence may be dependent on
structures and traditions of local politics. We must acknowledge a risk that the
introduction of local democracy could itself exacerbate ethnic tensions. There
are many parts of the world where traditional autocratic regimes have long relied
on ethnic or tribal leaders to provide basic justice and security within their local
communities. When such local authority is transferred to the office of a locally
elected mayor whose administration covers all citizens in a district, at most one
ethnic leader can win election to mayor. Then there can be a serious danger that
the traditional leaders of other ethnic groups may react against local democracy
if they feel that it threatens them with a loss of power to serve their traditional
constituents.
This risk of inciting ethnic violence could be reduced by vesting local power
more broadly in a council where leaders of all traditional groups could hope to
win seats. That is, where local ethnic tensions are a problem, it may be better to let
an elected local council choose the mayor or head of local government by a local
version of the normal parliamentary system, rather than to allocate such a power-
ful local office by a winner-take-a ll popular election.
There may also be concerns about decentralization exacerbating regional
separatism. In a region that has a strong popular separatist movement, its can-
didates would be likely to win local elections, but local democracy would not
then be causing the separatist movement. In fact, separatist movements are
often caused by a history of oppressive centralized rule that leaves no place for
local leadership. Election to local offices can actually give local leaders more
interest in preserving the political status quo due to concerns that the next suc-
cessor state might reduce or redistribute their local powers. In a province that
is large enough to stand alone against the rest of the nation, however, the top
provincial leaders could perceive some chance of gaining sovereign national
power by cultivating a separatist movement. Thus, where separatism is a con-
cern, political decentralization may be better limited to local councils for small
districts.
Ultimately, ethnic divisions in national politics cannot be bridged unless there
are some political leaders who can be trusted by people of all major ethnic groups.
Responsible local governments can provide more opportunities for such leader-
ship to develop. In a nation where such broadly trusted leadership is lacking, a
local leader who began to develop a reputation for working reliably and justly with
members of all ethnic groups could hope to become a strong candidate with broad
support for national leadership.
28.7. Establishing a Stable Federal Division of Powers

We have noted that national leaders and their most important supporters may
feel a powerful vested interest against any decentralization of power in a tradi-
tionally centralized unitary state. Political decentralization can seem undesirable
or burdensome to national leaders because it entails more difficult negotiations
with local leaders, some of whom may have the potential to become new rivals for
national power. But a national leader who accepts this cost may find, in the long
run, that a reputation for working effectively with local leaders within an accepted
constitutional system can become an essential asset for building strong, broad-
based political coalitions.
In this way, a constitutional system with democratic local government can
become politically stable once it is established. When governors and mayors are
locally elected, they become local power-brokers from whom national politicians
must regularly seek support in their competition for national power, and then it
would be very costly for any national leader to threaten the constitutional powers
of these elected local officials. Thus, a transition to a decentralized democratic
system, once achieved, can be self-sustaining.
It may be argued that, to demonstrate an appropriate respect for national sov-

ereignty, foreign supporters of a new state should try not to influence its constitu-
tional structure. However, when foreign military or economic support is needed
to maintain a national leader’s authority, the offer of foreign support itself may
affect the state’s constitutional development. If there were no foreign support, the
national leader could hope to gain effective national authority only by negotiat-
ing more political deals with local leaders. Thus, a greater centralization of power
may itself be a result of foreign support. In this case, the constitutional impact of
foreign support could actually be reduced when foreign supporters press national
leaders to accept more political decentralization, even as such decentralization
could reduce the state’s costly dependence on its foreign supporters.
28.8. Conclusions
Questions of how to help a nation develop a strong democratic political system
call for a deeper understanding of political systems in general and of democracy
in particular. Under any political system, power is held by leaders who organize
political networks or parties by promising their supporters that loyal service will
be well rewarded. In a dictatorship, national power is exercised by one leader’s
political network, which tolerates no rival. In democracy, different leaders with
rival political networks must compete for voters’ approval as the key to power.
But democratic competition can effectively provide political incentives for bet-
ter public service only when voters can identify two or more qualified candidates
with good reputations for each elective office.
Thus, the key to successful democratic development in a nation is to increase its
supply of leaders who have reputations for using public funds responsibly to pro-
vide public services, and not just to give patronage jobs to their supporters. This
essential supply of trusted democratic leadership can develop in responsible insti-
tutions of democratic local government, where successful local leaders can prove
their qualifications to compete for higher office. A presidential election by itself
can give prestige to its winner, but it does nothing to develop the broader supply of
trusted alternative candidates on which the success of democracy will ultimately
depend. Indeed, one might find more opportunities for independent political
development of reputations for responsible public service in a decentralized fed-
eral system without multiparty democracy, as in China today, where autonomous
local governments have provided vital leadership for economic growth.
We have argued that interactions between local politics and national poli-
tics can strengthen democracy at both levels. Local democracy can strengthen
national democratic competition when elected offices in municipal and provin-
cial governments provide a ladder of democratic advancement that effective lead-
ers can climb from local politics into national politics. But conversely, national
democracy can strengthen local democratic competition when national politi-

cal parties support competitive challengers to established local bosses. For such
mutually reinforcing interactions between local politics and national politics, the
institutional pillars for a strong democratic system should include both a multi-
party national assembly and elected local councils with clear autonomous bud-
gets and responsibilities.
Those who would encourage and support the development of democracy in
a nation should try to support a balanced development of democracy across dif-
ferent levels, from municipal and provincial governments to the national gov-
ernment. Urging national elections before local elections, or providing foreign
support only for political leaders at the national level, could upset this balance.
International donors might better support balanced democratic development
by insisting that some share of development-assistance funds should be allo-
cated directly to autonomously elected institutions of provincial and municipal
governments.
We may ask, for example, how the fate of Egypt’s first democratic system might
have been different if global advocates of democracy in 2011 had urged a transi-
tion to democracy that included local elections with or before national elections.
If Egypt’s transition to democracy had started with local elections, many factions
would have gotten opportunities to start building reputations for responsible
democratic leadership in different areas. But in a presidential election, only one
candidate can win the prize of centralized national power. In fact, the winners of
the national elections in 2012 chose to delay the introduction of local democracy
and retained centralized control of local government. Such centralization might
have seemed convenient for the short-term interests of those who had won con-
trol of the national government, but it left Egypt’s new democracy perilously vul-
nerable to fears of another autocracy. Empowerment of trusted local leadership
throughout the country could have done much to reduce such fears.
For an example of a better transitional regime for democratic state-building,
we may consider America’s Articles of Confederation (1776–1788), which dis-
tributed power widely among thirteen locally elected provincial assemblies. This
decentralization of power admittedly created difficulties in financing the war
effort, but it guaranteed that every community had local leaders with a vested
interest in the new regime, and this broadly distributed political strength made
the American Revolution unbeatable. The wide supply of local leaders with estab-
lished reputations for public service in elected office was the best guarantee that
strong competitive democracy would endure in America after the Revolution. It is
well understood that the Articles’ weak national government was not suitable for
the long run, but it was ideal as a transitional regime for the initial establishment
of democratic national government. The contrast is stark between the decentral-
ized political structures of American history and the centralized regimes that
America has often supported abroad.
References
Cheema, A., A. Q. Khan, and R. Myerson. 2015. “Breaking the Countercyclical Pattern of
Local Democracy in Pakistan.” In J. Faguet and C. Pöschl, eds., Is Decentralization Good for
Development: Perspectives from Academics and Policy Makers. New York: Oxford University
Press, 68–79.
Dobbins, J., S. G. Jones, K. Crane, and B. C. DeGrasse. 2007. The Beginner’s Guide to Nation-
Building. Santa Monica, CA: RAND.
Fortmann, L. 1983. “The Role of Local Institutions in Communal Area Development.”
Gaborone, Botswana: Applied Research Unit, Ministry of Local Government and Lands.
http://pdf.usaid.gov/pdf_docs/PNAAT392.pdf [accessed February 2013].
Ghani, A., and C. Lockhart. 2008. Fixing Failed States. New York: Oxford University Press.
Myerson, R. 2006. “Federalism and Incentives for Success of Democracy.” Quarterly Journal of
Political Science 1, no. 1: 3–23.
Myerson, R. 2008. “The Autocrat’s Credibility Problem and Foundations of the Constitutional
State.” American Political Science Review 102, no. 1: 125–39.
Myerson, R. 2011. “Toward a Theory of Leadership and State-Building.” Proceedings of the
National Academy of Sciences U.S.A. 108 (supp. 4): 21297–21301.
Myerson, R. 2014. “Standards for State-Building Organizations.” In R. Solow and J. Murray,
eds., Economics for the Curious. New York: Palgrave Macmillan, 174–88.
Myerson, R. 2015a. “Democratic Decentralization and Economic Development.” In C. Monga
and J. Y. Lin, eds., Oxford Handbook of Africa and Economics, Volume 1. New York: Oxford
University Press, 756–69.
Myerson, R. 2015b. “Local Agency Costs of Political Centralization.” University of Chicago
Working Paper. http://home.uchicago.edu/~rmyerson/research/localagency.pdf.
U.S. Army and Marine Corps. 2007. Counterinsurgency Field Manual FM 3-24. Chicago: University
of Chicago Press.
Varshney, A. 2002. Ethnic Conflict and Civic Life. New Haven, CT: Yale University Press.
NA M E IN DEX
Figures, notes, and tables are indicated by f, n, and t respectively. Names starting with "al-" are alphabetized by the
subsequent part of the surname.
Abadie, A., 125, 127, 128 Apenszlak, J., 383, 385

Acemoglu, D., 106, 218, 221, 222, 284, 329 Aranguren, M., 438, 444
Adalian, R. P., 195, 196, 199, 295, 298, 304, 646 Archibald, R. C., 460
Adena, M., 281, 282, 283, 284 Arendt, H., 46
Adhikari, P., 254, 255, 257 Arias, M., 262
Agüero, J. M., 264 Arjona, A., 220
Aguirre, C., 562n1 Armstrong, D., 98n5
Aguirre, J., 462 Arnado, G., 39
Aivazian, V. A., 463 Aronson, J. D., 104, 563n2
Ajzenman, N., 476n2 Astorga, L., 455
Akerlof, G. A., 309, 318, 320, 321, 322, 463, 487, Aust, B., 367, 368
488, 489, 501, 502, 532n5, 615 Autesserre, S., 211
Akresh, R., 224n1, 263, 264, 352 Aydin, A., 238, 239, 576, 580t
Albright, M. K., 4, 173, 622 Azam, J.-P., 242, 252, 253, 254, 426
Alderman, H., 224n1, 263, 264
Alesina, A., 238, 532n1
Alhinawi, H., 447 Baars, B., 600, 608n1
Allison, P. D., 522 Baccara, M., 200
Aly, G., 319, 325, 326, 327, 335n7, 335n9, 387 Baddeley, J. F., 406
Amegashie, J. A., 203 Bae, S., 175, 199
Amesberger, H., 386 Baker, S. F., 399
Amunwa, B., 595 Bakke, K. M., 402, 569
Anderson, M. B., 242 Balakian, P., 30
Anderton, C. H., 5, 23, 62, 67, 69, 71, 99n8, 146, Balcells, L., 252, 253, 254, 255
147, 152f, 168, 169nn6–9, 186n1, 186n4, Balch-L indsay, D., 173, 174, 213, 220, 235t, 238,
190, 200, 207, 217, 244, 290, 291, 294, 296, 239, 241, 242, 402, 405, 571
298, 306, 308, 310–315t, 318, 476, 481, Baliga, S., 194, 207
482, 513, 531, 532n1, 576, 577, 579–580t, Ball, P., 536, 563n2
608, 658 Ballesteros, A., 459
André, C., 233, 254, 255, 340, 341, 344, Bandura, A., 323
348, 349 Banerjee, A. V., 400, 515
Andreas, P., 406 Banken, R., 336n15
Andreev, K., 107 Bannon, A., 597, 598, 599, 608n6
Angrist, J., 263, 442 Bar-I saac, H., 200
Annan, J., 214 Barnard, A., 406
Annan, K., 42 Barnett, A. H., 659n12
675
676 Name Index
Barnett, R., 35 Bongaarts, J., 107

Barrett, C., 260 Boucher, J. P., 523
Barrett, J., 33, 34 Boudreaux, K., 341, 344
Barry, C. M., 585 Boulding, K., 299, 309, 393–394n1, 394n2, 403,
Bartels, S., 365 487, 501
Bartov, O., 4 Bowles, S., 511, 514
Bartrop, P. R., 370 Boylan, R., 562
Bashford, A., 121n5, 122n18 Bozzoli, C., 255, 261
al-Bashir, O., 622, 641, 643, 648 Brakman, S., 258
Bates, R., 447 Brants, C., 601, 605
Batruch, C., 598, 599 Brauer, J., 146, 168, 169n9, 290, 291, 298,
Bauer, M., 129, 265, 266 308, 310–315t, 394n1, 476, 608, 655,
Baum, S., 232 658, 660n20
Baumol, W. J., 332, 596 Bray, J., 593, 594, 595, 603, 607
Beard, T. R., 659n12 Breiman, L., 545
Beardsley, K., 402 Breitman, R., 29
Becker, G. S., 23, 194, 299, 651 Breton, A., 324, 325
Bel, G., 326 Bridgman, J., 316n4
Bellamy, A. J., 176 Brockett, C. D., 216
Bellows, J., 258, 265, 399, 402 Browning, C. R., 4, 335n7
Bemba, J. P., 368 Brück, T., 211, 216, 217, 255, 261, 264, 591
Benabou, R., 515 Brustein, W., 319, 326
Benhabib, J., 517 Bryukhanov, A. I., 415, 418
Ben-Sefer, E., 386 Buccheim, C., 326
Benson, B., 660n16, 660n20 Bueno de Mesquita, B., 199
Berelson, B., 275, 280, 285n1 Buggeln, M., 336n12
Bergan, D., 275 Buhaug, H., 244, 399, 402, 403
Berkeley, B., 425 Bundervoet, T., 224n1, 262, 264, 352
Berlage, L., 264, 352 Burge, R., 366
Berman, E., 532n1, 536 Burnet, J., 387
Bernard, V., 593 Burnett, P. H., 292
Bernheim, B. D., 515 Bush, G. W., 276, 613
Bernholz, P., 323, 324 Bush, K. D., 222
Berry, A., 260 Bussmann, M., 61t, 173
Besançon, M. L., 175, 236t, 244, 247, 576, 577, Butcher, C., 582, 587n13
579–580t, 584, 587n7 Butcher, T., 33
Bhagwati, J. N., 659n10
Bhatia, P., 563n14
Bhavnani, R., 218, 219 Cabral, L., 195
Biddle, S., 536 Calabresi, G., 659n11
Biernacki, C., 543 Calderón, F., 15, 453, 454, 455–456, 457f,
Bijak, J., 366, 374n4 461–4 62, 475
Bismarck, O. von, 303 Calderón, G., 476n2
Bjørnlund, M., 387 Caldwell, J. C., 105
Black, E., 592, 600 Callen, J. L., 463
Black, P., 330 Callen, M., 129, 265, 266
Blair, G., 169n4 Camacho, A., 224n1, 264
Blattman, C., 99n9, 214, 240, 265, 268n2 Cameron, A. C., 521, 523
Bloom, D. E., 106, 117 Canning, D., 106, 117
Bloxham, D., 25n2 Capicotto, S. A., 658, 660n22
Blundell, J. E., 623 Carey, S. C., 174, 576, 580t, 581, 582, 587n2
Boeke, J. H., 491 Carment, D., 187n17
Bohmelt, T., 587n2 Carr-Saunders, A. M., 121n6
Bohoslavsky, J. P., 594, 603 Carter, J. R., 62, 147, 169nn7–8, 190, 244, 576,
Bohra-M ishra, P., 252, 255, 256, 257 577, 579t, 580t
Boix, C., 218 Carter, M., 260, 265, 266, 269n4
Name Index 677
Cartwright, E., 164 Curthoys, A., 640

Caruso, R., 328 Cuvelier, J., 358
Casper, B. A., 186n12 Czaika, M., 252, 253, 255, 256
Cassar, A., 265 Czukas, K., 269n4
Castaño, C., 438, 449n4
Castillo, J. C., 455, 462
Cates, W., Jr., 104 Dadrian, V. N., 641, 642t
Caves, R. E., 200, 202 Dal Bó, E., 447
Cederman, L.-E ., 244, 403, 420n5, 578, 584 Das Gupta, M., 116
Celeux, G., 543 Dauter, L., 191
Cenarro, A., 449n5 Davenport, C., 98n5
Cernic, J. L., 643 Davidson, J. S., 486
Cerra, V., 126, 128, 131, 259 Davis, D., 258, 462
Chacón, M., 449n11 Davis, K. E., 629
Chalk, F., 512 Day, R. H., 517
Chamarbagwala, R., 224n1 de Groot, O. J., 125, 127, 463
Chandra, S., 584 Deininger, K., 255
Charny, I. W., 57t, 213, 371 Dell, F., 326
Chatterji, M., 510, 511 Dell, M., 462
Cheema, A., 670 DellaVigna, S., 275, 283, 285n2
Chella, J., 591, 592, 593, 594, 595, 598, 602, 603 DeMeritt, J. H. R., 569, 577, 580t, 586
Chen, S., 128 den Boer, A., 122n16
Chicoine, L., 462 Deng, F., 380
Chirot, D., 173, 174 Denov, M., 392
Chowdhury, S., 203 Denuit, M., 523
Church, J. R., 193 Deolalikar, A., 264
Churchill, W., 28, 29, 42 Derderian, K., 387
Cincotta, R. P., 106 Des Forges, A., 341, 344, 346, 347, 348
Clapham, A., 592, 593, 595 DeSoysa, I., 242
Clark, J. F., 358 de St. Jorre, J., 406
Clarke, P., 597, 598 de Waal, A., 185
Clegg, S., 335n6 de Walque, D., 212, 222, 224n1, 263, 264, 348, 352
Clinton, B., 613 Diamond, J., 340
Clinton, H., 633 Diamond, P. A., 515
Coase, R. H., 644, 659n8 Dickert, S., 623
Cockayne, J., 476n5 DiGiuseppe, M. R., 585
Coghlan, B., 365, 374n3 Dihigo, E., 35
Cohen, B., 104 Dillard, A., 620, 636n2
Cohen, W. S., 4, 173 Dillon, M., 369
Colaresi, M., 174, 576, 580t, 581, 582, 587n2 Dixit, A., 203, 328
Colby, W., 537 Dixon, J., 99n9
Collier, P., 99n9, 125, 127, 128, 214, 217, 218, Dobbins, J., 664
219, 237, 242, 500, 532n1 Dobby, E. H. G., 411
Comber, L., 536 Doces, J., 106
Confino, A., 382 Docker, J., 640
Conley, M., 564n21 Dolamari, M., 597
Cooper, J., 49nn1–2 Dolan, C., 389
Cooper, N., 476n5, 592, 594, 603 Donaldson, D., 400
Cooper, R., 515 Dowd, C., 61t
Cornes, R., 505 Downs, G. W., 652
Cornwell, J., 335n4 Doyle, K., 562n1
Cramer, C., 221, 348 Dranove, D., 200
Cribb, R., 483 Dube, A., 462
Crockett, A., 593, 594, 595, 603, 607 Dube, O., 439, 462
Croicu, M., 62 Dubost, C., 34
Cunningham, D. E., 121n7, 402, 586 Duch-Brown, N., 190
678 Name Index
Duffield, M., 369, 476n5 Ferguson, T., 326

Duflo, E., 400 Fergusson, L., 285n3
Dunnigan, J. F., 403, 404 Fernald, J., 128
Dunning, T., 482, 495, 499 Fernández, M., 263
Duque, V., 263 Ferreira, O. R., 452
Duran-M artinez, A., 462 Ferrero, M., 143
Durante, R., 278 Festinger, L., 322
Durlauf, S., 511, 514 Fetherstonhaugh, D., 618
Dyson, T., 105 Finnegan, W., 452, 453
Finnoff, K., 352
Fischhoff, B., 104, 563n2
Earl, H., 37 Fisher, M., 390
Easterly, W., 53, 54t, 57t, 66t, 98n1, 98n5d–e, Fjelde, H., 99n6, 576
129, 174, 186n6, 213, 219, 221, 231, 236t, Fleischhacker, J., 329
237, 238, 239, 240, 241, 248n2, 283, 483, Fligstein, N., 191
532n1, 571, 576, 579t, 585 Fonfría, A., 190
Eastwood, R., 104 Forcese, D., 597, 598, 599
Eccles, H. E., 399 Fortmann, L., 668
Eck, K., 57t, 62t, 66t, 220, 234, 236t, 237, 243, Fortna, V. P., 587n3
425, 571, 576, 580t, 586, 659n2 Foster, A. W., 584
Edwards, W., 614 Fox, V., 455
Ehrenreich, E., 335n4 Foxton, P. D., 399
Eisensee, T., 278 Frank, R. F., 23
Ekmekcioglu, L., 387 Frank, R. W., 585
Elder, T., 37 Frey, B. S., 169n7
Enders, W., 99n9, 147, 169n7 Friedman, J., 536
Engel, S., 252, 254, 255, 256, 257, 259 Friedman, W., 212, 214, 222, 345, 346, 349, 351
Englebert, P., 364 Friend, T., 485
Enikolopov, R., 277, 278, 284 Frieze, D., 30, 31, 32, 34, 35, 36, 37, 39, 40, 41,
Enterline, A. J., 402 42, 49n1, 49n3
Escobar, V. M., 455–456 Frumkin, G., 117
Esteban, J., 71, 174, 175, 176, 177, 178, 179, Fudenberg, D., 202, 203
182, 184, 186n8, 186n11, 219, 236t, 237, Fujii, L. A., 217, 219, 351
239, 241, 243, 245, 248n2, 513, 576, 577,
579t, 580t
Gabriel, Y., 200, 591
Gaddafi, M., 391, 621
Fafchamps, M., 269n4 Galiani, S., 476n2
Fagan, J., 536 Gallie, W., 43, 49n6
Fajnzylber, P., 247 Galor, O., 120n2
Fargues, P., 122n18 Galula, D., 402, 406, 411
Farmer, P., 631 Gangopadhyay, P., 511, 513, 532n1, 532n2
Favretto, K., 187n16 Garcia, E. A., 574
Fawcett, T., 573 García-Ponce, O., 462
Fealy, G., 486 García Tinoco, M., 460
Fearon, J. D., 218, 402, 429, 512 Gardeazabal, J., 125, 127, 128
Feenstra, R. C., 126, 130 Garfinkel, M. R., 147, 328
Feierstein, D., 370, 371 Garretsen, H., 258
Feigenbaum, M. J., 517 Gartzke, E., 562
Fein, H., 4, 217, 218, 370, 452, 476n1, 585 Gates, S., 219, 238, 239, 399, 402, 576, 580t
Feldman, G. D., 327, 335n9 Gatti, R., 53, 54t, 57t, 66t, 98n1, 98n5d–e, 174,
Felter, J., 536 186n6, 213, 221, 231, 236t, 237, 238, 239,
Feng, Y., 104 240, 241, 248n2, 283, 483, 571, 576, 579t
Fenrick, W., 44 Gaudet, H., 275, 280, 285n1
Ferencz, B. B., 36–37, 329, 330 Gbagbo, L., 641
Ferguson, N. T., 453, 463 Gellately, R., 318
Name Index 679
Geller, D. S., 99n9 Guiso, L., 129

Gelman, A., 536 Gupta, R., 513, 532n2, 532n3
Gelos, R. G., 585 Gurr, T. R., 56t, 66t, 126, 131, 136t, 137, 214,
Gentzkow, M., 275, 285n2 219, 239, 244, 569
Gerber, A. S., 275 Gürtler, O., 203
Getz, K., 593, 595, 606
Ghani, A., 664
Ghobarah, H. A., 125, 129 Habyarimana, 279, 346
Gibney, M., 57t Hall, B. H., 522
Gibson, J., 492 Haltiwanger, J., 515
Gilbert, R. J., 203 Hamburg, D. A., 627
Gilmore, E., 242 Hamman, J. R., 335n5
Giraldo, F., 440, 440t, 444, 445t Hand, J. L., 196
Girumuhatse, J., 385 Harff, B., 4, 56t, 66t, 98n5a, 98n5g, 126, 131,
Glaeser, E. L., 192, 319, 322, 490, 502 136t, 147, 173, 174, 175, 212, 218, 219, 220,
Glass, J. M., 318 221, 235t, 237, 238, 239, 240, 245, 247, 569,
Gleditsch, K. S., 402, 578, 584, 586 576, 578, 579t, 580, 580t, 581, 582, 587n10
Gleditsch, N. P., 66t, 68, 242, 244, 448n3 Harsanyi, J., 513
Glick, L. B., 4 Hartley, K., 192, 205
Goebbels, J., 275 Hassan, M., 404
Goenjian, A., 265 Hassan, T., 222, 284
Goeth, A., 34 Hatzfeld, J., 349
Goldenberg, M., 385 Hausman, J. A., 522
Goldhagen, D. J., 232, 330, 335n2, 336n12 Hayes, P., 600, 601, 602
Goldsmith, B. E., 247, 573, 574f, 580t, 582, 583 Hazlett, C., 579t, 580t, 581, 582, 583, 587n10
Goldstone, J. A., 121n5, 581, 583 He, H., 574
González, N. M., 452 Healy, A., 265
Goodman, R., 636n3 Hearn, P. P., 420n5
Goodsell, L., 562 Hedgepeth, S. M., 385
Goodwin, J., 214 Heger, L., 175
Gorbachev, M., 316n9 Hegre, H., 399, 578
Göring, H., 328 Heim, S., 335n7
Gottschang, T. R., 255 Heinle, K., 452
Gould, W. T. S., 104, 105 Hendrix, C., 62t
Gourevitch, P., 348, 385 Herbert, U., 329, 330, 331
Govaert, G., 543, 563n14 Herf, J., 335n4
Grabowski, J., 335n4 Hernandez, D., 460
Granovetter, M., 206 Hernández-Salazar, D., 562n1
Graunt, J., 121n5 Herrera, H. A., 462
Grayson, G. W., 455, 456, 456f, 460 Hertzberg, A., 629
Green, D. P., 283 Hesketh, T., 116
Greenhill, B. D., 569 Heuveline, P., 117
Greenhill, K. M., 121n14 Heydrich, R., 325, 330
Gregory, R., 632, 633, 634f Heyes, K., 366
Greif, A., 532n1 Hicks, H., 99n6
Greiser, A., 34 Hilberg, R., 328, 335n8, 335n10, 601
Grigoryan, A., 187n16 Hill, H., 498
Griliches, Z., 522 Hill, K., 104, 107, 120n1
Gross, E., 39 Hillemanns, C. F., 642t
Grossman, H. I., 199, 214, 463 Hillman, A. L., 323, 335n2
Grun, R. E., 255 Himmler, H., 324, 330, 387
Guariso, A., 351 Hinde, A., 366, 374n4
Guerlac, H., 399 Hintjens, H. M., 351
Guidolin, M., 595, 602, 603 Hippler, J., 596
Guillen, M., 523 Hirshleifer, J., 199, 214, 328, 433,
Guilmoto, C. Z., 115 449n10, 656
680 Name Index
Hitler, A., 31, 32, 34, 199, 281–2 82, 318, 319, Jenkins, D., 485
322, 324, 325, 330, 335n11, 419, 599, 600 Jerbi, S., 592, 593, 594, 595
Hobbes, T., 513, 532n4 Jinks, D., 636n3
Hoddinott, J., 224n1, 263, 264 John, A., 515
Hoeffler, A., 71, 99n9, 125, 127, 129, 214, 217, Johnson, D. H., 425
218, 219, 240, 241, 242, 252, 253, 254, Johnson, S., 106
426, 500 Johnson, S. D., 399
Hoess, R., 34 Jomini, A.-H ., 399, 400
Hoex, L., 218, 220 Jonas, S., 195
Hoffman, S., 176 Jonassohn, K., 512
Hogg, N., 387 Jones, A., 366, 370, 379, 391
Holck, S. E., 104 Jones, M. A., 652
Hollenbach, F. M., 284 Jonker, L. B., 518
Holzgrefe, J. L., 176 Josse, J., 550
Homer-Dixon, T., 340, 343 Justino, P., 211, 212, 214, 215, 216, 217, 218, 220,
Horowitz, D., 218 221, 222, 223, 224nn1–2 , 254, 259, 260,
Horton, L., 449n12 261, 264, 352
Horvath, S., 545
Hudson, V., 122n16
Huisman, W., 594, 595, 598, 599, 600, 601, 602, Kabila, J., 78t, 362, 368
603, 605 Kaeb, C., 592, 593, 594, 598, 603
Hulten, C. M., 127, 128 Kagame, P., 280
Hultman, L., 57t, 62t, 66t, 99n6, 173, 176, 220, Kahneman, D., 164, 165, 194, 615, 616, 617, 618,
234, 236t, 237, 243, 425, 571, 576, 580t, 618f, 623
586, 659n2 Kaleck, W., 592, 594, 601, 603, 608n1
Humphreys, M., 169n3, 214, 217 Kalyvas, S., 99n5, 169n3, 214, 215, 220, 233,
Hurtado, L., 608n1 251, 252, 253, 254, 255, 257, 404, 426, 429,
Hussein, S., 596, 597, 603, 604, 621 431, 448n1, 449n4, 464, 537, 544, 562
Husson, F., 550 Kammen, D., 485, 486
Huth, P., 125, 129, 173, 174, 213, 220, 235t, 238, Kantorová, V., 107
239, 241, 242, 405, 571 Kapend, R., 366, 374n4
Kaplan, E., 276
Karlan, D., 275
Iannaccone, L., 487 Kaserman, D. L., 659n12
Ibáñez, A. M., 252, 254, 255, 256, 257, 259, 260, Kathman, J. D., 576, 577, 579–580t
262, 263, 268n1 Katz, E., 280, 285n1
Ichino, A., 216 Kaufmann, E. P., 121n5, 122n18
Ignatieff, M., 42, 43 Kaul, I., 656
Imai, K., 169n4 Kayitesi, C., 349
Inklaar, R., 125, 126, 130 Kazianga, H., 269n4
Iovleff, S., 563n14 Keen, D., 214, 216, 221
Irvin-Erickson, D., 33 Keeney, R. L., 632, 633, 634, 635
Irwin, J., 632 Kelly, M. J., 591, 592, 593, 595, 596, 597, 598,
Isham, J., 243 599, 607, 608n4, 643
Ishwaran, H., 545, 564n17 Kenkel, B., 584
Kenyatta, U., 641
Keohane, R., 176
Jackson, R., 33–3 4, 45 Kessler, C., 265
Jackson, S., 358, 359, 359t, 360, 361, 368 Kessler, R., 265
Jacob, J., 496t Khan, A. Q., 670
Jacobs, L. A., 592, 597, 598, 599 Khan, I., 594, 608n3
Jacobson, K. R., 593, 594, 595, 599, 600, 601 Kibbe, W. C., 292, 301
Jaggers, K., 137 Kiernan, B., 4, 386
Jahan, R., 197 Kim, D., 219, 220, 238, 577, 580t
Japkowicz, N., 573 King, C., 521
Jaspers, W., 367, 368 King, E. W., 325
Name Index 681
King, G., 574, 575 Laudati, A., 363, 368

Kingsbury, B., 629 Lavelle, J., 265
Kinsey, B., 224n1, 263, 264 Lawrence, T. E., 406
Kirk, D., 105, 121n4 Lazarsfeld, P. F., 275, 280, 285n1
Kirk, R., 449n4 Lê, S., 550
Kis-K atos, K., 252, 253, 255, 256 le Carré, J., 364
Kiss, A., 536 Lederman, D., 247
Kleinschmidt, H., 302 Lee, R., 104, 107, 121n9
Klingner, J., 536 Leeson, P. T., 200
Kocher, M., 214, 215, 252, 253, 254, 255, 537, Leites, N., 401
544, 562 Lemarchand, R., 585
Kodi, M., 371 Lemke, D., 121n7
Kogalur, U., 564n17 Lemkin, R., 4, 12, 13, 28, 29–37, 39, 40–43,
Kogut, T., 620 49n1, 49n4, 69, 70, 71, 143, 144, 145, 157,
Komer, R. W., 406 159, 169n10, 211, 289, 290, 296, 316n1,
Kondylis, F., 212, 222, 260, 263 318, 327, 332, 379–380, 383, 384, 392,
Konrad, K., 203 511, 640
Kony, J., 389–391 Lenin, V., 296, 297
Kopf, D., 452, 476n1 Leone, M., 216
Korey, W., 34, 35, 49n1 Lessing, B., 462
Korf, B., 252, 254, 255 Levene, M., 70, 71
Koubi, V., 125, 127 Levine, R., 219
Krain, M., 125, 173–174, 175, 186, 235t, 237, Ley, R., 380
245, 248n2, 569, 576, 577, 578, 580t, 586 Li, L., 116
Kranton, R. E., 309, 318, 320, 321, 322, 487, 488, Liaras, E., 426
489, 501, 502 Lichbach, M., 214
Krauss, M. I., 659n11 Lichtenstein, S., 615, 632
Kress, M., 399, 401 Liddle, W., 494
Krueger, A. O., 659n10 Lieberman, B., 371
Kugler, J., 104, 120n3, 125, 263 Lifton, R. J., 624
Kugler, T., 117, 118t, 120f, 120n3, 122n17 Limongi, F., 501
Kuper, L., 4, 217, 218, 379 Lincoln, A., 316n9
Kuperman, A. J., 156, 176, 187n17 Lindau, J. D., 455, 456
Kuran, T., 23 Lindbeck, A., 449n11
Kurlat, S., 53, 54t, 57t, 66t, 98n1, 98n5d–e, 174, Lindley, A., 252, 255, 256, 257
186n6, 213, 221, 231, 236t, 237, 238, 239, Lindsay, J., 562
240, 241, 248n2, 283, 483, 571, 576, 579t Lipset, S., 500
Kydd, A. H., 176, 177, 178, 179, 180, 182, 183, Lipton, M., 104
185, 186n7, 186n9, 186n11, 186n13 Lizza, R., 633
Kyriakakis, J., 593, 600 Loayza, N. V., 128, 247
Lockhart, C., 664
Loewenstein, G., 335n5, 619, 619f
Lacina, B. A., 448n3 Lohlé-Tart, L., 366, 374n4
Ladek, S., 593, 595, 606 Longerich, H. P., 330
La Ferrara, E., 532n1, 595, 602, 603 Lopez, H., 126, 127, 137
Lahuerta, Y., 431 Loveland, T. R., 420n5
Laitin, D., 218, 402, 532n1 Loveridge, S., 224n2
Lakwena, A., 390 Lower, W., 387
La Mattina, G., 352 Lozada, R., 440, 440t, 444, 445t
Lambert, A., 366, 374n4 Lozano-Gracia, N., 252, 255, 256, 257
Lambert, D., 522 Luechinger, S., 169n7
Lancaster, K. J., 159 Lujala, P., 242
Langer, A., 219 Lunga, L., 641
Langer, L. L., 306 Luxemburg, R., 296
Lanoszka, A., 562 Lyall, J., 169n4, 536
La Porta, R., 532n1 Lybbert, T., 269n4
682 Name Index
Mace, J., 296, 305, 306, 316n10 Merrouche, O., 216

MacGaffey, J., 361 Merry, S. E., 629
Mack, P., 4 Meyer, D., 562
Mackaay, E., 659n4 Mialon, H. M., 201
MacQueen, M., 336n15 Michaelsen, M. M., 476n2
Madden, D., 562 Michalowski, S., 642t, 643
Maddison, A., 105, 107, 109t, 118t, 120f Midlarsky, M., 169n1
Madefu Yahisule, J. O., 364 Miguel, E., 99n9, 218, 240, 258, 259, 265,
Madley, B., 292, 299, 300, 301 268n2, 399, 402
Magaloni, B., 455, 476n2 Milanovic, B., 507n4
Makarin, A., 284 Milgram, S., 323
Malthus, T. R., 121n5, 340, 342–3 43 Milhollin, G., 596
Mamdani, M., 347, 348 Miller, D. E., 387
Mampilly, Z., 220 Miller, L. T., 387
Manby, B., 593, 594, 607, 608n1 Minani Bihuzo, R., 364
Mann, M., 173, 175 Minow, M., 48, 49
Manne, R., 388 Mitchell, F., 35
Mansfield, E., 175 Mixon, F. G., 325
Mao Zedong, 15, 401 Mobutu Sese Seko, 361, 363, 434, 486, 495, 496, 500
Marks, D., 491, 493t, 495t, 497 Moffatt, S., 594, 595, 608n1
Markusen, E., 452, 476n1 Mollica, R., 265
Marriage, Z., 360, 362, 368, 369 Moloeznik, M. P., 456, 460
Marshall, A., 12 Mombo, L. M., 364
Marshall, M. G., 56t, 66t, 126, 131, 136t, 137 Montalvo, J. G., 175, 218, 219, 231, 236t, 238,
Martinez, C., 460 239, 241, 247, 576, 577, 580t
Martinez, P., 460 Monteiro, J., 264
Martin-Ortega, O., 594, 595, 602, 603 Montgomery, M. R., 104
Marwaha, J., 536 Morán, H. E., 224n1
Mason, S. J., 573 Morelli, M., 71, 174, 175, 176, 177, 178, 179, 182,
Massey, D. S., 252, 255, 256, 257 184, 185, 186n8, 186n11, 219, 236t, 237,
Matijasevic, T. M., 262 239, 241, 243, 248n2, 513, 576, 577, 579t,
Matti, S., 372 580t, 585
Maxwell-F yfe, D., 34 Morrison, A. R., 252, 253, 255, 256, 257
May, R. A., 252, 253, 255, 256, 257 Morse, D., 597, 598, 608n5
May, R. M., 517 Moses, A. D., 25n2, 43, 388
Mayersen, D., 147 Moshonas, S., 364
Mbembe, A., 357, 358, 360, 361, 369, 370, 373, 374 Motz, K., 596
Mbi, E., 366 Moya, A., 260, 263, 265, 266
McAfee, R. P., 201 Moyar, M., 536, 539
McCargar, J., 537 Mueller, H., 125, 127, 131, 233
McCauley, C., 173, 174 Mueller, K. P., 399
McDoom, O. S., 98n2, 217, 218, 219, 221, 223, Mugabe, R., 492
342, 345, 347, 350, 351, 576, 580t, 584 Mullahy, J., 522
McDougal, T. L., 462, 463 Muller, E., 218
McFate, M., 399 Mulligan, C. B., 199
McGee, J., 201 Muñoz, J., 262
McGregor, K., 486 Muñoz, P., 440, 440t, 444, 445t
McKay, A., 224n2 Murphy, K. M., 194
McPhee, W. N., 275 Murshed, S. M., 99n10, 219, 399, 487
McQuinn, B., 402 Museveni, Y., 389, 425
Meierhenrich, J., 25n2 Myerson, R., 664, 665, 666, 668, 670
Mejia, D., 455, 462 Myrdal, G., 482, 492, 494
Melamed, A. D., 659n11
Mercille, J., 462
Meredith, J., 35 Nadif, M., 543
Meron, T., 41 Naidoo, S., 476n5
Name Index 683
Naimark, N., 39 Petrovic, D., 48

Natapoff, A., 536 Pettersson, T., 57t, 62t, 659n2
Nath, S., 532n1 Pham, P. N., 389
Naudé, W., 591, 592 Pierskalla, J. H., 284
Neiman, B., 128 Pilster, U., 587n2
Nero, 30 Pinchotti, S., 214
Nersessian, D., 34 Pinker, S., 232
Neumayer, E., 242 Pinochet, A., 486, 641, 653
Neville, A. O., 387–388 Pinto, M. E., 431
Newman, P. C., 326 Pischke, J., 442
Ngoma-Binda, P., 364 Piven, B., 402, 406
Nietschke, H., 476n5 Platteau, J. P., 233, 254, 255, 340, 341, 344,
Niewyk, D. L., 197, 199 348, 349
Nkunda, L., 362 Pobol’, N. L., 412
Noh, S. J., 199 Poch, B., 117
Nolin, C., 608n1 Polatel, M., 387
North, D. C., 129 Polk, W. R., 430, 449n7
Nunn, N., 129, 242 Pol Pot, 636n2
Polyan, P. M., 412
Popper, R., 598, 607
Obama, B., 613 Porfirio Diaz, J., 431
Oetzel, J., 593, 595, 606 Porter, M., 201, 202
Okello, T., 389 Posner, R. A., 644, 648, 651, 652–653, 659n4,
Olken, B., 277 660n17, 660n21
O’Loughlin, J., 402 Powell, C., 49n6
Olson, J. M., 340, 342 Power, S., 31, 32, 36, 41, 623, 635
Olson, M., 214, 489, 502, 584 Prat, A., 285n2
Omran, A. R., 105 Pratt, A., 532n1
Opgenhaffen, V., 594, 603 Pratto, F., 197
Ordover, J. A., 203 Preston, S. H., 104
Organski, A. F. K., 125 Price, M., 563n1, 563n2
Ortiz, K., 264 Prunier, G., 341, 344–3 46, 347, 348, 349–350
Østby, G., 219, 244 Przeworski, A., 501
Ott, A., 175, 199 Pugh, M., 476n5
Overy, R. J., 329 Purdekova, A., 252, 255
Owen, R. C., 399 Putin, V., 277–278
Putnam, R. D., 129, 511, 514
Paige, J., 214, 218

Paluck, E. L., 283 Qian, N., 242, 400
Parsons, W. S., 204, 211, 289, 290, 292–306, Querido, C., 175, 243
316nn3–4 Querido, C. M., 238, 576, 577, 579t, 580t
Pasha, E., 304 Quigley, J. B., 370
Pasha, T., 30
Patterson, D., 382
Pégorier, C., 25n1, 659n2 Raeymaekers, T., 367
Peña, X., 263 Raiffa, H., 632
Pench, M., 203 Raleigh, C., 234, 242, 243, 399, 407
Percival, V., 340, 343 Raleigh, R., 61t
Percy, S., 603 Ramasastry, A., 594, 595, 603
Perrot, S., 358 Ramsay, K., 562
Persson, T., 449n11 Ratcliff, R., 572, 573, 587n5
Peteraf, M., 200, 201 Ratcliffe, J. H., 399
Peters, E., 623–624 Ratner, S., 44
Petersen, R., 214, 217, 426, 449n11 Rauchhaus, R., 187n17
Petrova, M., 277, 278, 284 Ray, D., 186n8, 219, 245
684 Name Index
Reagan, R., 316n9, 621 Saage-M aaß, M., 592, 594, 601, 603, 608n1
Redmond, H., 462 Saba, R. P., 659n12
Regan, P., 244 Safire, W., 47
Reid, J., 369 Sahay, R., 585
Reid, T., 359, 360 Saidel, R. G., 385
Reno, W., 221 Saideman, S. M., 402
Restrepo, H. E., 104 Salardi, P., 216, 476n2
Restrepo, J., 435, 438, 439 Salehyan, I., 62t, 175, 244, 402, 577, 586
Restrepo, P., 455, 462 Saloner, G., 203
Reynaert, J., 163, 164 Salop, S. C., 203
Reynal-Querol, M., 125, 127, 128, 129, 175, 218, Saltarelli, D., 222
219, 231, 236t, 238, 239, 241, 247, 576, Sambanis, N., 70, 71, 99n5, 99n9, 125, 578
577, 580t Samuelson, P. A., 335n11
Riad, A. M. B., 35 Sanchez, F., 224n1, 263
Richards, P., 214, 216, 218, 391 Sandler, T., 99n9, 147, 169n7, 505, 656, 657
Richter, E., 627 Sandleris, G., 585
Rieth, L., 592, 593, 595, 598, 603, 607 Sankoh, F., 391
Rios, V., 453, 462, 467, 470, 474, 476n2 Sapienza, P., 129
Ritov, I., 620 Sarkees, M. R., 66t, 67, 68
Robinson, G., 486 Sattath, S., 622
Robinson, J., 218, 221, 222, 447, 449n11 Satyanath, S., 218
Robles, G., 476n2 Sawyer, W. C., 325
Rocha, R., 264 Saxena, S. C., 126, 128, 131, 259
Rockmore, M., 262 Schabas, W. A., 4, 34, 36, 45, 46, 47, 640,
Rod, J. K., 399, 403, 420n5 641, 642t
Rodríguez, C., 224n1, 263 Schaller, D., 293, 298, 303
Rogall, T., 126, 233, 234, 237, 351, 352, 400 Scheffman, D. T., 203
Rogers, D. B., 200 Scheinkman, J. A., 194
Rohner, D., 71, 129, 174, 175, 176, 177, 178, 179, Scherer, T., 562
182, 185, 186n3, 186n8, 186n11, 219, 236t, Scherner, J., 326
237, 239, 241, 243, 248n2, 513, 576, 577, Scheweitzer, A., 326
579t, 580t, 585 Schindler, K., 264
Rohrbach, P., 303 Schley, D. R., 623–624
Roland, G., 258 Schmitz, D. F., 486
Romero, M., 439 Schneider, G., 61t, 173, 219
Roosa, J., 484, 485 Schock, K., 218
Roosevelt, F. D., 28, 32, 613 Schramm, M., 258
Rosefielde, S., 306 Schrodt, P. A., 571
Rosenbaum, D. I., 199 Schumpeter, J. A., 199
Rosenthal, A. M., 37 Schutte, S., 403
Ross, M., 435 Scott, J., 214, 218
Rost, N., 237, 244, 247, 579t, 580t, 581, 582, 583 Scully, G., 175
Rotberg, R., 176 Seelke, C. R., 455
Rotfeld, A., 42 Seira, E., 476n2
Roth, J. K., 336n13 Seligson, M., 218
Rouanet, H., 550 Semelin, J., 379, 392
Rousseau, J. J., 532n4 Semenovich, D., 582, 587n13
Roux, B., 550 Sen, A. K., 244, 488, 501, 615
Rowlands, D., 187n17 Sergenti, E., 218
Rozental, M., 104 Serneels, P., 126, 127, 129
Ruhe, C., 173 Sevilla, J., 106, 117
Rummel, R. J., 53, 54t, 57t, 98n5d–e, 173, 174, Seybolt, T. B., 104, 563n2
221, 235t, 237, 238–2 39, 240, 404, 576, Shaffer, G., 655, 656
580t, 581, 582, 639 Shamil, I., 405
Russell, D. E. H., 427 Shamir, R., 603, 607
Russett, B., 125, 129 Shanley, M., 200
Name Index 685
Shapiro, J., 536, 562 Stanton, G., 291, 627

Shapiro, J. M., 275 Starcevic, N., 596
Sharlach, L., 365 Stargardt, A. W., 327
Shaw, M., 70, 71, 98n1, 169n3, 204, 232, 246 Starr, H., 403
Shawcross, H., 34 Staub, E., 4, 232
Shemyakina, O., 222, 224n1, 263, 264 Stearns, J., 363, 364
Shenon, P., 596 Stedry, A. C., 614
Sherman, J., 593, 595, 596, 602, 603 Steele, A., 251, 252, 253, 254, 404
Sherman, W. T., 406 Stefanelli, J., 44, 47
Shi, T., 545 Stein, S., 33
Shirk, D. A., 452, 455 Stephens, B., 592, 600, 602, 603
Shneider, H., 36 Stephens, J., 608n1
Shortland, A., 463, 476n5 Stewart, F., 125, 126, 132, 138, 192, 219, 244,
Sidanius, J., 197 343, 578
Sienkiewicz, H., 30 Stigler, G. J., 23, 299
Signorino, C., 584 Stokes, P., 200, 591
Sikkink, K., 41 Stover, E., 389
Silva, R., 536 Straus, S., 176, 177, 178, 179, 180, 182, 183, 185,
Simon, H. A., 196, 614, 615 186n7, 186n9, 186n11, 186n13, 213, 217,
Simpson, B., 486 219, 348, 351, 353n1, 452, 476n1, 569, 570
Singer, J. D., 68, 99n9 Strom, A., 31
Siroky, D., 244, 402, 577 Strömberg, D., 277, 278, 285n2
Sjöström, T., 194, 207 Suharto, 481, 482, 483, 484, 485, 486, 489, 492,
Skaperdas, S., 147, 328, 463, 512 494, 495, 497, 499, 500
Skarbek, D., 200 Sukarno, 481, 483, 484, 485, 486, 491, 492, 494, 507n1
Skinner, G., 599, 600, 605, 608n1 Summers, H., 536, 539
Skully, G. W., 240 Sun Tzu, 44
Slim, H., 592, 594, 600 Sundberg, R., 62
Slovic, P., 615, 616, 617f, 619, 619f, 622, Swee, E. L., 224n1, 264
634f, 636n3 Sweetman, P., 592
Small, D. A., 619, 619f Syropoulos, C., 463
Small, M., 68 Szpakowicz, S., 573
Smeulers, A., 218, 220
Smith, A., 643
Smith, B. D., 623 Tabeau, E., 107, 121n8, 121n12, 536
Smith, V., 615 Tabellini, G., 449n11
Snyder, J., 175, 278 Tadjoeddin, M. Z., 497t, 498, 498t, 499
Snyder, R., 462 Tago, A., 239, 587n2
Snyder, T., 415 Tahoun, A., 284
Sobel, E., 200 Tan, T. K., 491
Sokolova, M., 573 Tatz, C., 294, 295, 297, 301, 302, 658n1
Sollenberg, M., 68 Taylor, C., 391, 649
Solow, R. M., 127, 128 Taylor, C., Jr., 649, 651
Solum, L., 659n11 Taylor, P. D., 518
Sowmya, A., 582 Tehlirian, S., 30
Spagat, M., 435, 438, 439 Teitelbaum, M. S., 105, 120n3
Speer, A., 330 Tesch, B., 600
Spence, A. M., 203 Teson, F. R., 176
Spoerer, M., 329 Thaler, R., 615
Spolaore, E., 129 Thayer, T., 537, 560
Sprout, H., 403 Themnér, L., 66t
Sprout, M., 403 Thoenig, M., 129, 186n3
Stalin, J., 7, 14, 15, 39, 204, 289, 296, 297, 305, Thom, K., 462
306, 307, 407, 411–415, 419, 636n2 Thomas, H., 201
Stallbaumer, L. M., 599, 600, 601 Thompson, R. C., 594, 595, 603
Stanley, W. D., 253, 255 Thucydides, 172
686 Name Index
Tilly, C., 121n6, 218, 367 Vélez, C., 252, 254, 255, 256, 257
Timmer, M., 126, 130 Vergara, A., 431
Tirole, J., 202, 203 Verpoorten, M., 98n2, 126, 127, 129, 254, 255,
Tisdell, C., 192, 205 261, 264, 341, 352
Titeca, K., 390, 391 Verwimp, P., 98n2, 211, 212, 213, 214, 216, 217,
Toft, M. D., 121n5, 399, 402 220, 222, 224nn1–2 , 254, 255, 257, 259,
Torvik, R., 449n11 260, 261, 262, 264, 341, 342, 344, 345,
Totten, S., 204, 207n1, 211, 289, 290, 292–306, 348, 351, 352, 536, 584, 591
316n3, 370, 583, 628 Vilalta, C., 462
Tough, P., 269n5 Vinck, P., 265, 389
Townsend, R. M., 515 Vlassenroot, K., 358, 362
Townsley, M., 399 Voigtländer, N., 322, 482
Trapido, J., 364 Voillat, C., 602
Trevino, L. J., 325 von Bülow (Chancellor), 303
Tripathi, S., 594, 595, 599, 603 von Hagenbach, P., 44
Trivedi, P. K., 521, 523 von Joeden-Forgey, E., 361, 366, 381
Tsui, K., 199 von Krosigk, L., 325
Tull, D. M., 364 von Neurath, K., 34
Tullock, G., 199, 659n10 von Trotta, L., 293, 303, 336n14
Tversky, A., 164, 165, 615, 617, 618f, 622, 623 von Winterfeldt, D., 635
Tyson, S. A., 186n12 Voors, M. J., 128, 129, 265, 267
Voth, H. J., 322, 326, 482
Udry, C., 269n4

Ulfelder, J., 53, 54t, 56t, 62, 66t, 97tn1, 98tnn1– Wacziarg, R., 129
2, 98tn5h, 98n1, 453, 576, 577, 579–580t, Wald, P., 46, 47
581, 582, 583, 585, 587n11, 587n13 Wallensteen, P., 66t, 68
Üngör, U. U., 387 Waller, J., 3, 4, 53, 217, 218, 219, 232,
Urdal, H., 106 512, 640
Uribe, A., 439 Wallis, J., 126, 129
Urlanis, B., 117 Walter, B., 214
Urwand, B., 592 Ward, J., 546f, 554, 555t, 560t
Uvin, P., 342, 344 Ward, M. D., 569, 571
Uzonyi, G., 253, 576, 580t Ware, R., 193
Warr, P., 496, 507n3
Watts, M. J., 595, 598, 602, 603, 608n1
Vahabi, M., 207 Waxman, M. C., 406
Valentino, B. A., 4, 53, 54t, 56t, 62, 66t, 97tn1, Wayman, F. W., 66t, 67, 68, 239
98tnn1–2 , 98tn5f, 98tn5h, 98n1, 99n5, Weber, R. A., 335n5
169n3, 173, 174, 213, 220, 232, 235t, 238, Weibull, J., 449n11
239, 241, 242, 405, 453, 512, 571, 575, 576, Weidmann, N. B., 420n5, 578, 584
577, 579–580t, 581, 587n8 Weinstein, D., 258
van Baar, A., 595, 599, 600, 601, 602, 605 Weinstein, J. M., 169n3, 200, 214, 217,
van Creveld, M., 399, 401 233, 402
van der Wilt, H. G., 595, 603 Weiss, D., 603, 607
van de Sandt, J., 608n1 Weiss, T. G., 176
van Schaack, B., 39, 40 Weitz, E., 382
Vansina, J., 347 Welch, D., 335n4
van Sliedregt, E., 594, 595, 598, 599, 603 Weller, J. B., 292
van Tuyll, H., 655, 660n20 Wennmann, A., 594, 596, 603
van Zanden, J. L., 491, 493t, 495t, 497 Wettstein, F., 594, 598, 606
Vargas, J. F., 219, 285n3, 435, 439 Wheeler, N. J., 176, 626
Varian, H., 25n2 Wibowo, Sarwo Edhie, 485
Varshney, A., 670 Wickham-Crowley, T., 214
Västfjäll, D., 620, 624f, 634f Wiesen, S. J., 326, 599, 600, 605
Vela, M. A., 285n3 Wilkinson, T., 462
Name Index 687
Williams, M. A., 201 Yanagizawa-Drott, D., 126, 234, 237, 279, 280,
Williams, N. E., 252, 253, 255, 256, 257 281, 282, 283, 284, 342, 351, 352
Williams, P. D., 176 Yehouda, S., 335n6
Williamson, O., 654 Yehuda, R., 265
Winter, J., 36 Yellen, J. L., 463
Winter-Ebmer, R., 216 Yusuf, H. O., 607, 608n1
Wintrobe, R., 324, 325, 487, 489, 502
Witmer, F. D., 402
Wodon, Q., 126, 127, 137 Zak, P., 104
Wolff, C., Jr., 401 Zakic, M., 335n9
Wolitzky, A., 329 Zambrano, 262
Wood, E. J., 214, 216, 217 Zeng, L., 574, 575
Wood, R. M., 57t, 99n6, 220, 237, 238, 244, Zerbe, R. O., 659n6
251, 252, 253, 254, 402, 576, 577, Zetter, R., 252, 255
579–580t, 586 Zhu, W. X., 116
Woods, A., 636n3 Zhukov, Y. M., 399, 400, 402, 405, 411, 415,
Wooldridge, J. M., 571 418, 421n12
Worley, L., 316n4 Zhuravskaya, E., 277, 278
Wrong, M., 363 Zilibotti, F., 129, 186n3
Wyshak, G., 265 Zimmer, M., 592, 593, 595, 598, 603, 607
Zingales, L., 129
Zionts, D., 636n3
Xiem, T. V., 541 Zwierzchowski, J., 107, 121n8, 121n12
SU BJECT I N DE X
Figures, notes, and tables are indicated by f, n, and t respectively. Names of people are found in the separate
Name Index.
Abkhazia, 95t Age distribution disruption: demography of

Aborigines (Australia), 14, 74t, 93t, 289, 292, genocide, 102, 112–115
306, 307, 309, 640; child removals, 302, Agglomeration, 10, 309, 532n6
388; constrained optimization theory Aid: appropriation, 242; development aid,
(COT) and genocide of, 300t, 301–302; 242, 495, 577, 578. See also Disaster aid;
economic interdependencies and genocide Humanitarian aid
of, 294–295, 297–298 Albania, 73t
Abu Ghraib prisoner abuse scandal, 630 Algeria: civilian deaths in massacres, 64,
Accountability, 41, 221, 224, 600, 602, 606, 607, 425, 448n1; civilians switching sides
663, 665 in civil war, 431; civil war overlapping
ACLED. See Armed Conflict Location and with genocide, 220; datasets and trends
Event Data of genocides, 73t; external support,
“Acts Constituting a General (Transnational) disruption of, 406; population losses and
Danger Considered as Offences Against the demographical rates, 108t; population
Law of Nations” (Lemkin’s proposal), 31 recovery, 118t
Advanced Research Projects Agency, 539 Alien Tort Claims Act (US), 642t, 643, 649
Affect: defined, 616; value of human lives and, ALiR (Army for the Liberation of Rwanda),
616–620, 617f 95t, 358
Affective imagery, 630–631 Al-Jazeera, 275
Afghanistan: age distribution disruption, 112, Alliance of Democratic Forces for the Liberation
113f, 114; behavioral responses to violence, of Congo-K inshasa, 95t
266; datasets and trends of genocides, Al Qaida, 62, 95t
72–73t; gender-i mbalance ratios, 115, 115f, Analytical models: correlation and causation
116t; harm inflicted by ISAF led to reduced treated together for purposes of, 572;
support, 169n4; migration rates, 112, 112f; differences from predictive models, 570–
population losses and demographical rates, 576; economic risk factors, role of, 576–578
108t, 113f; population recovery, 20, 114, Angola: civil war overlapping with genocide,
118t, 119, 120f; road networks and rebel 136t, 660n20; corporations mining for
violence, 402, 420n3 diamonds, 603; datasets and trends of
Africa: cell phone coverage linked with violence genocides, 73t, 96–9 7t; migration rates,
in, 284; civilian violence in contemporary 112f; natural resources used to finance
civil wars, 238, 242, 243, 407–411, 408f, insurgents, 435; population losses and
409–410t, 411f; terrorizing civilians in, demographical rates, 109t, 110; population
426. See also specific countries recovery, 118t, 119, 121n10, 122n15
African Union, 369, 627, 657 Anti-A mericanism, 275
689
690 Subject Index
Anti-Semitism, 281–2 82, 319, 321, 322, 323, Behavioral poverty traps, 19, 267
335n2. See also Holocaust Behavioral responses to displacement, 265–2 67
Anti-Serbian sentiment, 283. See also Belarus, Soviet partisan railroad sabotage in
Serbo-Croatian war World War II in, 415–418, 416f, 417–419t
Appropriation economy, 11, 121n14, 297, 329, Bernoulli Latent Block Model, 543
393n1. See also Natural resources; Wealth Biafra, 85t, 406
appropriation Bleakness theorem, 17, 158–163; defined, 162
Arab Revolt of 1916‒1918, 406 Blood and human organs, sale of, 659n12
Arab Spring, 58, 284, 663 Boko Haram (Nigeria), 391, 585, 654
Arakis Energy Corporation, 598 Bosnia: age distribution disruption, 20, 113f,
Argentina: banks’ involvement in mass 114; civil war overlapping with genocide,
atrocities, 603; civil war overlapping with 77t, 94t, 136t; early death estimates, 121n8;
genocide, 136t; datasets and trends of gender-i mbalance ratios, 20, 115, 115f,
genocides, 73–74t 116t; government using thugs, instead of
Armed Conflict Location and Event Data armed forces, to commit ethnic cleansing,
(ACLED), 58, 59t, 61t, 225n3, 407, 408 233; microeconomic effects of genocide,
Armed Islamic Group of Algeria (GIA), 431 212, 222; migration rates, 112, 121n8;
Armenian genocide (1915‒1923), 30, 45, 92t, population losses and demographical
108t, 196, 199, 289, 292, 307–309, 493, rates, 107, 108t, 110, 121n12; population
641, 653; causes of, 646; constrained recovery, 118, 118t, 120f
optimization theory (COT) and Bounded rationality, 8, 614
genocide of, 300t, 304–305; economic Brazil, 38, 39, 74t
interdependencies and genocide of, British All-Party Parliamentary Group
295–296, 298, 306; enrichment/ (APPG), 361
impoverishment dynamic of, 382, 387; Brutality: advancing aims of perpetrators, 17;
forced labor and sex slaves, 387 economics of, 11–12. See also specific events
Army for the Liberation of Rwanda (ALiR), of genocide
95t, 358 Bulgaria, 74t
Articles of Confederation (US), 673 Burundi: civil war overlapping with genocide,
Assassinations, 30, 82t, 252, 483, 537, 540, 551, 136t, 175; datasets and trends of genocides,
578, 586 74t, 89t, 95–9 7t; democratization and
Asset seizure, 6, 306, 647, 659n13 ethnic cleansing, 175; migration rates,
Atrocity Forecasting Project, 587n1 112f; population losses and demographical
Auschwitz Institute for Peace and rates, 108t, 264; population recovery, 118,
Reconciliation, 42 118t, 119, 120f
Australia: Aboriginals Protection and Business complicity, 6, 16, 21, 22, 591–612;
Restriction of the Sale of Opium Act of banking sector, 603; contract terms
1897, 301; child removals, 301, 302, 388; setting out sociopolitical clause to prohibit
datasets and trends of genocides, 74t; genocide, 607; Darfur and, 597–599,
Genocide Convention ratification and, 603; direct vs. indirect complicity, 604;
41, 302; independence of, 316n5. See also genocide prevention, business role in, 607;
Aborigines Holocaust and, 599–6 01; importance
Authoritarian regimes, 175, 199, 200, 221, 434, of understanding of how and why, 606;
492, 494, 495, 500, 578, 666, 672 Kurdish genocide from chemical weapons,
Autocracy, 186n5, 239, 284, 500, 584, 667, 596–597, 603; level of involvement, 608n3;
670, 673 Nuremberg trials and, 599–6 00, 605; types
Azerbaijan, 74t, 420n7 and sizes of businesses, 602–6 04
“Butter” and “guns,” 328, 335n11
Backfire condition, 21–22, 25, 154, 156, 169n8

Bangladesh, 4, 67, 74t, 197, 512 Cambodia: age distribution disruption, 112,
Behavioral economics, 9, 16, 22, 23, 145, 164– 113f, 114; civil war overlapping with
167, 194, 309, 614–615; policy implications genocide, 135, 136t, 220; datasets and
for civilian atrocity prevention, 166–167; trends of genocides, 4, 67, 75t; early death
reference point dependence and loss estimates, 107–110, 121n8; external
aversion, 164–165, 166f, 168. See also support, disruption of, 406; Extraordinary
Intuitive thinking; Psychic numbing Chambers in the Courts of, 642t, 653;
Subject Index 691
gender-i mbalance ratios, 20, 113f, CNDP (National Congress for the Defense of
114–116, 115f, 116t; microeconomic the People, Congo), 163, 362–363
effects of genocide, 212, 222; migration CNN, 275
rates, 112f, 121n8; population losses and Coase Theorem, 645, 659n8
demographical rates, 108t; population Cognitive dissonance, 322–323
recovery, 118t, 119, 120f; school enrollment Cold war, 30, 176, 363, 411, 481, 486, 500;
of displaced children, 263 interventions, effect of, 434; linkage
Canada, 75t, 658n1 alleged between capitalism and
Capital effects. See Macroeconomic effects of fascism, 605
genocide Collective political violence datasets, 53
Carnegie Endowment for International Colombia, 15, 17, 285n3, 289, 437–4 47; banks’
Peace, 32 involvement in mass atrocities, 603;
Caucasus Wars of 1816‒1864, 405–4 06 children’s health, 264; civilian casualties,
Cell phone coverage linked with violence in 425, 439–4 44, 440–4 41t, 443t, 449n14;
Africa, 284 civilians switching sides in civil war, 431,
Central African Republic, 75t, 97t, 587n11 449n5; datasets and trends of genocides,
CERAC (Conflict Analysis Resources Center), 76–77t, 96t; displacement correlated
439–4 40, 440–4 41t, 448n2 with violence, 255; displacement of
Chad, 75t, 655 population, 254, 259–2 60; human rights
Chemical weapons, use of, 45, 596–597, 600, nongovernmental organizations’ list of
604, 654 conflict events, 439; IDPs in, 251; militia
Child removals, 301, 302 empowerment, 435, 437–4 47, 445–4 46t;
Child soldiers, 17, 18, 361, 389, 391–392 paramilitary forces, 438; population
Chile: civil war overlapping with genocide, losses and demographical rates, 109t;
136t; datasets and trends of genocides, population recovery, 118t, 119; post-war
75t; Pinochet’s liability for civilian recovery, 259; psychological issues of
atrocities, 641 displaced households, 265; risk assessment
China: African mineral resources and, 362, and livelihoods, 262, 266; school
372, ; civil war overlapping with genocide, enrollment of displaced children, 263;
136t; datasets and trends of genocides, 76t; third-party interventions, effects of, 17,
displacement correlated with violence, 434; unemployment rate among displaced
255; external support, disruption of, 406; persons, 263; US military aid to, 434;
as most murderous regime, 239; population Venezuela military aid to rebels in, 434
losses and demographical rates, 108–109t; Colonization, 79t, 157, 248n2, 293, 297–298,
population recovery, 118t 303, 315t, 316n4, 340, 347, 353, 491, 492;
China National Petroleum Corporation, 598 faced with impossibility of defeating
Choice, 5, 6f, 143, 191, 290; defined, 7; genocidal insurgency supported by majority of
conflict as deliberate choice, 291–292, 298, people, 430, 449n7. See also Decolonization
482, 487; lexicographic ordering and, 324; Communist Party of India‒Maoist, 95t
optimal choice, 7–8; violence compelling, Communist regimes, 238–2 39, 481, 500
18. See also Constrained optimization Comparative advantage and the exchange
theory; Displacement of people; Rational economy, 10–11
choice models Compassion fade, 16, 620, 623, 624
Cingranelli-R ichards (CIRI) Human Rights Conflict Analysis Resources Center. See CERAC
Dataset, 53–56, 55t, 57t Conflict economics, 190, 328
Civilian atrocities, datasets on, 53; state- Conflict trap, 237–2 38
nonstate group civilian atrocity datasets, Congo. See Democratic Republic of
53–56; trends and comparative measures Congo (DRC)
of seriousness, 64–67, 65f, 66t Congolese Rally for Democracy, 95–96t
Civilian violence. See Violence against civilians Consequence matrix for clarifying intervention
Civil war: inequality as cause of, 244–2 45; tradeoffs, 633– 634, 634t
overlapping with genocide, effects of, 135, Constrained optimization theory (COT), 7,
136t, 174, 186n7, 220, 232, 237, 246, 578; 8, 13, 17, 145–146, 290–291; baseline,
as predicter of mass atrocities, 247. See also 147–148, 152f; definition of, 144; high
Violence against civilians; specific countries input substitution possibilities in, 157, 158,
Civil war literature, 237, 240, 241, 242 159f; pre-Holocaust genocides, 298–306;
Class as impetus for genocide, 14–15, 204, 353 rebel-helping price policies and, 154–156;
692 Subject Index
usefulness of, 15, 167–168. See also GMA 72–95; Easterly, Gatti, and Kurlat
risk and prevention (EGK) mass killing dataset, 53, 54t, 57t,
Contest success function (CSF), 428, 63, 64; Event Data on Armed Conflict
436, 449n10 and Security (EDACS), 59t, 61t; events
Contingent weighting model, 622 datasets, 58, 439; Konstanz One-Sided
Contract law, 647–6 48, 649; sociopolitical Violence Event Dataset (KOSVED), 58,
clause to prohibit genocide, 607 59t, 61t; Minorities at Risk (MAR), 55t,
Contras (Nicaragua), 449n11 56, 58t; nonstate groups as perpetrators
Convention on the Prevention and Punishment (1989‒2013), 95–98; Political Instability
of the Crime of Genocide. See Genocide Task Force’s geno-politicides (PITF-G),
Convention 53, 54t, 56t, 63, 64, 67; Political Instability
COPDAB (Conflict and Peace Databank), 52 Task Force Worldwide Atrocities Dataset
Corner solution, 149, 150f, 151, 153f, 187n13, (PITF-A), 58, 59t, 61t, 63, 64, 65f, 69;
430, 449n7, 506 Political Terror Scale (PTS), 53, 55t, 57t;
Correlates of War (COW) Project, 4, 52, 63, Rummel Democides dataset, 53, 54t,
67–68, 69, 99n12 57t; "Shia Genocide Database: Killings
Corruption, 193, 277, 285n3, 364, 367; in Pakistan from 1963 to 31 May 2013,"
democracy and political development in 524; state-and nonstate-perpetrated mass
corrupt cultural tradition, 664; Mexico atrocities, 63; terrorism datasets, 59–62;
drug wars and, 15, 455, 463–4 66, 470, two-t rack development of conflict datasets,
473, 475 67–71; Ulfelder and Valentino (UV)
Costs. See Logistics of violence; mass killing dataset, 53, 54t, 56t, 63, 64;
Opportunity costs Uppsala Conflict Data Program One-Sided
COT. See Constrained optimization theory Violence Dataset (UCDP-V), 53, 54t, 57t,
Côte d’Ivoire, 77t, 633, 641 63, 64; Virtual Research Associates (VRA),
COW. See Correlates of War (COW) Project 58. See also Two-t rack development of
Crime of aggression, 658n2 conflict datasets
Crimes against humanity, 28, 38, 45–47, 48, Decentralization, 454, 499, 666, 672, 673;
647; associated with genocide, 3, 43– democratic decentralization, advantages of,
44, 347, 569, 640; defined, 25n1, 629, 668–6 69; forces against, 669–671
640; distinguished from genocide, 640; Decision analysis, 632–634; consequence
encompassing war crimes and genocide, 47, matrix for clarifying intervention tradeoffs,
641; history of, 45–4 6; Nuremberg trials 633– 634, 634t
and, 45–4 6, 641 Decision to stay or leave. See Displacement
Criminal law, 31, 641, 645, 650–651 of people
Croatia, 77t, 94t, 283. See also Decolonization, 237, 315t, 358, 391, 486,
Serbo-Croatian war 491–492
CSF. See Contest success function Degenerate war, 70
Cuba, 35, 77t Dehumanization, 280, 291, 319, 323, 331, 334
Cultural ideologies, 197 Democide, 53, 54t, 57t, 235t, 238–2 39, 404, 576,
Customary law, 6, 44, 47, 255 581, 582
Czechoslovakia, 77t Democracy, 43, 174–175, 238–2 40, 241, 247,
368, 372, 449n1, 484-4 85, 491-492, 507n1,
578; democratic competition, requirements
Darfur, 91t, 185, 207n1, 369, 583, 597–599, of, 666–6 67; democratic decentralization,
603, 604, 608n6, 613, 620; global public advantages of, 668–6 69; distribution of
good and Chad’s rights, 655; US failure to power in society and, 666; forces against
provide humanitarian aid to, 621–622 decentralization, 669–670; human
Darfur Atrocities Documentation Project, 628 rights and, 244; leaders’ reputations and
Dashnak Party (Turkey), 30 foundations of constitutional state, 664–
Datasets, use of, 12–13, 24, 52–62; Armed 666; local level equal in importance to
Conflict Location and Event Data national level, 664; modernization theory
(ACLED), 58, 59t, 61t; Cingranelli- of endogenous democracy, 500–501; in
Richards (CIRI) Human Rights Dataset, political reconstruction, 663–674; stable
53–56, 55t, 57t; civilian atrocity trends federal division of powers, 671–672
and comparative measures of seriousness, Democratic Liberation Forces of Rwanda
64–67, 65f, 66t; country list (1900‒2013), (FDLR), 96t, 358, 368, 369, 373, 406
Subject Index 693
Democratic Republic of the Congo (DRC), 111–112, 112f, 121n13; natural growth
8, 19, 356–377; betrayal mechanisms, rates, 107; population recovery, 117–120,
366–367; civil war overlapping with 118t; postgenocide population “catch-
genocide, 136t; collaboration of aid up,” 102, 104; total population and
groups and political leaders shutting out demographic growth rates, 107–111;
majority of population, 21, 372–373; willingness of polity to welcome back
collusion between northern donors and returning migrants, 103, 111, 251
domestic elites, 372–373; Commission to Deprivation: as factor in genocide, 57t, 75t, 215,
Accompany the Transition (CIAT), 363; 219, 244, 246, 583, 584, 586, 626. See also
continuing pillage and taxation, 362–363; Scarcity; Starvation
datasets and trends of genocides, 77–78t; Detention, costs of operations, 329, 403, 405,
disinterest demonstrated by aid donors 485, 542f, 546f
during fighting, 372; economy and GDP Developing countries, 19, 25, 362, 419;
rise during genocide, 365f, 366; excess mortality rates in, 121n11; polarized,
mortality rates, 365, 366, 374n3; extractive poor, and resource-abundant countries,
taxation, 363; framing peace as economic 175, 242–2 44, 592; population losses and
transaction, 371–373; Global and All- demographical rates in, 110
Inclusive Agreement on the Transition Development aid, 242, 315t, 360-370, 373-374
of the Democratic Republic of Congo Development and risk of mass atrocities, 15,
(2002), 356, 362, 371; Goma crisis, 163; 20–21, 25, 230–250, 293, 360-370, 373-
high-profile politicians, 367; high-value 374, 383, 481-4 82, 494-500, 521, 525-
goods, 367–368; ineffectiveness of civilian 530, 576, 579t, 580, 597, 668; activist vs.
protection in, 163–164; involvement of opportunistic rebellions, 233; civil war
Congolese in mass atrocities, 360–361; overlapping with genocide, 232; economic
Lutundula Report, 364; Mai Mai and, gap in study of genocide, 231; economy
362; mortality causes during and after and, 240–2 46; empirical literature, 230;
war, 373; natural resources used to finance ethnicity and, 238; higher standard
insurgents, 435; nature of genocide, of living and, 231, 240; ideology and
370–371; northern markets and aid, 360; democracy and, 238–2 39; large-n studies
outcomes, 365–366; Peace, Security and of mass atrocities, 13, 234–2 37, 235–2 36t,
Cooperation Framework (2013), 373; 246–2 47; measurement of economic
peace-making initiatives, 361–366; pillage development, 231; natural resources as
of aid, 363–364; Rwanda, Uganda, and factor, 242–2 44; outside sponsorship
their militias, 358–359, 434; unregulated increasing likelihood of violence, 243; past
liberalization, 364, 373; victims, atrocities, likelihood of recurrence, 237–
difficulty in describing, 371, 374; violence 238; propaganda, effect of, 234; war and
overlooked, 368–369; war machine, political upheaval, 23
357–361; willful blindness and reckless Development research, demography in, 121n5
disregard, 370, 373. See also Kiwanja Diamond, Jared: Collapse, 340
massacre Diaspora support to rebels, 308, 402, 434–435
Demography of genocide, 13, 20, 102–124, Dictatorships. See Authoritarian regimes
215; age distribution disruption, 102, Direct persuasion, 14, 207n2, 253, 274–276,
112–114; birth rates, 103, 107, 120, 279–2 84
121n10; birth rates, increase post-v iolence, Disaster aid, effects of specific media content on,
122n18; data quality issues, 107; decline 278, 618
in death rate due to medical advances and Displacement of people, 5, 14, 47, 251–273,
improved sanitation, 121n6; demographic 356, 361, 373, 389, 429, 510, 596, 597-598;
transition theory, 104–106, 120n3, 121n4; armed groups’ strategies, 253–254; assets
development research’s relationship with, and livelihoods, 259–2 61; behavior and,
121n5; differentiating normal fluctuations 265–2 67; business complicity in Darfur,
and mere growth, 103–104; gender, age, 598–599; causes and consequences of,
and education status of Mexico drug war 252, 256; civilian strategies, 254–255;
victims, 461, 461f; gender-i mbalance ratios, compared to weather shocks, 268n4;
103, 114–116, 122n16, 215; likelihood of decision to leave, 18, 252–253, 255–256,
political instability and violent conflict 261; decision to stay, 14, 18, 252–255, 257,
associated with demographic transition 426, 449n12; economic factors, 256–258;
phases, 106; migration effects, 107, future research needs, 267–2 68, 268n3;
694 Subject Index
health issues and, 264; human capital structure of income, 242–2 44; wealth
effects, 262–2 65; logistics of resettlement, appropriation, 290, 296–297. See also
405, 411–415, 412f, 413–414t, 420– Macroeconomic effects of genocide;
421nn7–10; psychological issues and, 265, Microeconomic causes and consequences
269n5; risk assessment and livelihoods, of genocide; specific countries and events of
261–2 62; school enrollment and, 263; genocide
significant asset losses due to decision to Economic discrimination, 24, 244; as predictive
stay, 260; socioeconomic consequences, factor for genocide, 577, 579t, 582, 585
258–2 67; stayers vs. displaced persons, Economic marginalization, 235t, 245–2 46
254–255, 261. See also Refugees Economic protection, 214–216
Dispossessed engineered migrations, 121– Economics of density or agglomeration, 10, 12,
122n14, 298, 308, 326–327, 330 308, 532n6
Dispossessive mass killing, 512 Economics of information, 11, 14, 25. See
Disruption of economic activities, 5, 6f, 127, also Media, role in mass persuasion;
128, 203, 205, 261, 373, 438 Propaganda, effect in genocide
Distributional advantages sought, 222–223; Economies of learning, 10
future research agenda, 223 Economies of scale, 10, 308, 311t, 342, 404, 435
Distrust. See Trust issues Economists, lack of study of mass atrocities
Diversion of resources, 5, 6f, 127 by, 4, 69
Dominance, 192, 194, 195, 196, 199, 204, EDACS (Event Data on Armed Conflict and
205, 326 Security), 59t, 61t
Dominican Republic, 38, 39, 41, 78t Education, 6f, 19, 20-21, 159, 196, 205, 215, 222,
DRC. See Democratic Republic of Congo 231, 235t, 262, 269n5, 311t, 342, 346, 349,
Drug cartels. See Mexico drug wars 351–352, 443t, 461, 489, 490, 496, 498,
Drug crops as income source for insurgents, 438 502, 503, 515, 651; effect of displacement
Drug trafficking organizations (DTOs). See on, 263–2 64; reducing GMA risk, 6, 20,
Mexico drug wars 245, 281, 584. See also Human capital
Dutch disease, 482, 494, 496, 507n3 Egypt, 38, 58, 284, 402, 663, 670, 673
Einsatzgruppen, 29, 320, 335n1
Electoral coercion, 285n3
Early warning systems, 58, 72, 172, 186n2, 247, Elites, 21, 23, 56t, 131, 137, 173, 197, 212, 238,
308, 316n2, 592, 593, 627 239, 242, 314t, 344, 353, 438, 486, 487,
East Asian miracle, 494, 498 496, 499, 581, 584, 665, 666; in Congo,
Easterly, Gatti, and Kurlat (EGK) mass killing 369, 371–373; mass persuasion, role in,
dataset, 53, 54t, 57t, 63, 64, 66t, 72–98t 274, 283
Eastern Europe, civilians in resistance ELN, 76t, 96t, 437, 440t, 444, 445t, 449n5
movements in, 449n11 El Salvador: armed groups’ strategies, 253;
East German Stasi files, 562n1 civil war overlapping with genocide, 136t,
East Pakistan, 197 220; datasets and trends of genocides, 78t;
Economic concepts and theories applicable displacement correlated with violence, 255;
to GMAs, 7–11, 310–315t; comparative noncombatants who did not collaborate
advantage and the exchange economy, 10– with armed forces, likelihood of death of,
11; constrained optimal choice, 7–8, 144; 254–255
opportunity cost, 9; production function, Empowerment: altering balance of power, 425,
9–10; rationality, 8–9; scarcity, 7. See also 434–437; of institutions and actors less
Constrained optimization theory likely to succumb to psychic numbing,
Economic dimensions of genocides, 5–7, 6f, 627–628; militia empowerment in
190–191, 240–2 46, 482; business model, Colombia, 435, 437–4 47, 445–4 46t; of
290, 295–296; correlation of GDP growth victims, 631
and economic and political discrimination Energy production, 579t, 582
with genocide, 577, 579t, 582, 583; factors England. See United Kingdom
in displacement decisions, 256–258; Equatorial Guinea, 78t, 136t
income inequality, 21, 244–2 45; income Equilibrium, 506, 507t; defined, 510; levels
levels, 240–2 42, 259–2 61; international for mass killings, 149, 150t, 199-2 00, 510,
trade, 245–2 46; pre-Holocaust genocides 511, 514; with one-sided mass killings
illustrating interdependencies between risk, 182–183; replicator dynamics, 518–
economics and genocide, 290, 291–298; 520; with two-sided mass killings risk,
Subject Index 695
183–185; violence against civilians, 427, Forced displacement of population. See

431–436, 448 Displacement of people
Eritrean People’s Liberation Front, 435 Forced labor, 6, 290, 303, 312t, 316n4, 386, 388;
Ethiopia: civil war overlapping with genocide, Holocaust, 329–332, 333–334, 336nn12–
135, 136t, 220; datasets and trends of 14, 387, 599, 600
genocides, 78–79t; Genocide Convention Forces démocratiques de libération du Rwanda.
ratification and, 41; population losses and See Democratic Liberation Forces of
demographical rates, 109t; population Rwanda (FDLR)
recovery, 118t, 119 Forecasting models. See Predictive modeling
Ethnic cleansing, 47–4 8, 55t, 58t, 111; of GMAs
classification of, 3, 44; Darfur, 597–599; Foreign nongovernmental organizations
defined, 25n1, 47–4 8, 658–659n2; (NGOs). See Nongovernmental
dispossessed engineered migrations and, organizations
122n14; escalation to as unintended Former Yugoslavia genocide, 47, 94–95t;
consequence of third-party efforts, 156 Genocide Convention and, 40; Kosovo
Ethnic hatred. See Hatred war, NATO bombing in, 156. See also
Ethnicity, 192, 197, 222, 238, 247, 347; ethnic International Criminal Tribunal for the
fragmentation, 219; ethnic rivalries and former Yugoslavia (ICTY)
violence, 670–671; ethnic sectarianism, Fox News, 276
173; ethnolinguistic fractionalization France, 79t, 655, 656. See also Algeria
index (ELF) to measure, 238; interethnic Frente Sandinista (Nicaragua), 449n11
inequality, 339, 346 Front de Libération Nationale (FLN), 406
European Union: membership requirements Front for National Integration, 96t
and benefits, 657; role in intervention in Frozen assets, 151, 647, 659n13
Balkan wars, 655. See also specific countries Future research needs, 22–25, 102, 122n18, 167-
Event Data on Armed Conflict and Security 168, 186, 190, 206-2 07, 224, 261, 267, 490,
(EDACS), 59t, 61t 584, 596
Events datasets, 24, 58, 439; Armed Conf lict
Location and Event Data (ACLED),
58, 59t, 61t; Event Data on Armed Game theory, 8, 9, 13, 15, 16, 17, 71; choice
Conf lict and Security (EDACS), 59t, models and, 145; civilians under crossfire,
61t; Konstanz One-S ided Violence approach to, 426–427; equilibrium win
Event Dataset (KOSVED), 58, 59t, probabilities in, 186n8; evolutionary, 23,
61t; Political Instability Task Force 167, 309; favoritism and, 646, 659n10;
Worldwide Atrocities Dataset (PITF-A), incentives and constraints for mass killings,
58, 59t, 61t approach to, 172–189, 427–437; new law
Exchange economy, 10–11 and economics using, 644
Exclusion. See Political exclusion; Social Gang violence, 463. See also Mexico
exclusion drug wars
Explorative engineered migrations, 122n14 GDP. See Economic dimensions of genocides;
Ex post compensation, 646, 649–650 specific countries
Extraconstitutional changes, 235t, 237 Gender and genocidal economy, 264, 365,
378–395; enrichment/i mpoverishment
dynamic and, 382, 383–386; genocidal
Family bonds as target of life force atrocities, 17, economy, described as zero sum, 378–
381–382 379, 392; genocidal economy of atrocity,
Famine. See Starvation 381–383, 392–393; genocide prevention
Farben Case, 600–6 01 and, 393; life force atrocities and, 381,
FARC, 76t, 96t, 431, 434, 437–438, 440, 440t, 383; resource-poor genocides, 388–392;
444, 445t, 449n5 resource-r ich genocides, 386–388; root-
FDLR. See Democratic Liberation Forces and-branch genocide and, 379
of Rwanda Gender-i mbalance ratios, 102–103, 114–116,
Federal Emergency Relief Administration 122n16, 215, 352; Mexico drug wars,
(FERA), 277 civilian casualties in, 461, 461f; school
Flick Case, 600–6 01 enrollment of displaced children, 264
Food aid, 242, 306, 308. See also Geneva Conventions (1949), 25n1, 44, 640;
Humanitarian aid Additional Protocols (1977), 44
696 Subject Index
Genocide and mass atrocities (GMAs)/ forecasting power for inference, 587n2; on
genocides and mass killings unbalanced social structure, 197
(GMKs): characterization of, 4, 43–4 4, Georgia, 79t; behavioral responses to
203, 205–2 06, 291–292, 371, 379, 629; violence, 266
classification of, 3, 48; coercive mass Germany: datasets and trends of genocides,
killing, 512; country list of (1900‒2013), 79t; population losses and demographical
72–95; datasets on, 53; defined, 3, 33, 53, rates, 108t; population recovery, 118, 118t;
192, 212–213, 246, 365, 507n2, 591, 640; post-war recovery, 259. See also Herero and
economic theory of mass killing, 512–513; Nama peoples; Holocaust; Nazis
exogenous social interactions and, 514– GIA (Armed Islamic Group of Algeria), 431
517; genocide as form of mass atrocity, 48, Global Terrorism Database (GTD), 58, 63,
98, 592; independent of war, 232; literature 99n7, 99n8
survey of features and causes, 511–512; GMA risk and prevention, 20, 63–6 4, 71,
macro-vs. microlevel analysis, 520–521; 143–171, 172, 185, 285, 292, 307, 315t,
mass atrocity distinguished from genocide, 316n2, 332, 333, 569–590, 608, 623, 655,
592; mass killing, defined, 453; mass 657–658; affective imagery of human
killing distinguished from genocide, 3, 581; rights violations as means to prevent
nonrational factors for, 5, 9; pre-Holocaust, genocide, 631; assessment of benefits and
289–317; rational factors for, 5, 8–9, 200; costs, 23–2 4, 143; backfire condition,
record of, since 1900, 4, 52, 53; resource- 156, 169n8; baseline constrained
poor genocides, 388–392; resource-r ich optimization model, 147–148, 152f;
genocides, 386–388; root-a nd-branch baseline of pre-G MK model, 146–147;
genocide, 390; speed of killing, 308–309; behavioral economics on, 164–167; civilian
stylized facts about, 172–175; symbioses killing, risk factors for, 149–151, 150f,
between war and mass atrocities, 71; types 152f, 169n6; constrained optimization
of, 512; war crimes distinguished from, 45. theory in study of, 145–146, 167–168;
See also Genocide Convention; GMA risk courts serving to discourage atrocities,
and prevention; Lemkin, Raphael 653–654; cross price effects on civilian
Genocide Convention, 70, 294, 314t, 365, 642t, killing, 154–155, 155f; decision-a nalysis
655, 656; Article I, 38; Article II, 3, 38– approach to prevent genocide, 632–633;
40, 640; Article VI, 40; Article XIII, 41; deliberate thinking processes, as means to
criticism of, 640; definition of genocide, prevent genocide, 631–635; economic risk
186n1, 212, 289, 302, 370, 507n2, 511, 569, factors and predictive modeling of, 569–
591, 640; description of articles, 37–4 0; 590; economic variables and conditions,
domestic laws adopted after passage of, effect of, 20–21, 23–2 4; empowerment
641, 642t; drafting of, 36–37, 46; General of victims as means to prevent genocide,
Assembly adoption of, 37; passage and 631; future research needs on, 24;
ratification of, 40–41 gender influences on genocide and, 393;
Genocide gap in field of defense and peace ineffectiveness of civilian protection in
economics, 4, 22–2 3, 68–69, 72, 99n10 DR Congo, 163–164; ineffectiveness
Genocide studies, field of: on conditions in of piecemeal civilian protection
which authority group believes it has policies, 158–163, 161f; new laws and
tactical and strategic benefits for attacking institutions for, 22; policy implications
civilians, 169n3; datasets, use of, 24, for civilian atrocity prevention, 166–
52–62; on decisiveness of a conflict, 199; 167; post-G MA reconstruction efforts
economics gap in, 5, 22–2 3, 231, 244; and, 21; precommitment strategies to
forward-looking forecasting analysis, 24; prevent genocide, 626; preconditions
future research needs, 22–25; industrial for appearance of violent conflict and
organization, need to study, 25; large-n genocide, 193; rebel-helping price policies,
studies of mass atrocities, 13, 234–2 37, 154–155; reference point dependence
235–2 36t, 246–2 47; Lemkin Book Award and loss aversion, 164–165, 166f; solution
for best scholarly book, 42; organizational of the baseline model, 149; substitution
studies, 200; origins of, 4; pre-Holocaust possibilities, 157–164; tools of genocide
genocides, 307–308; short-and long-term, and mass killing prevention, 151–156,
micro-and macroeconomic consequences, 153–154f, 309, 320, 393, 419, 607, 625, 627,
24, 264; statistical significance and 639, 642t, 650; tradeoffs, 633–635, 634t.
Subject Index 697
See also Business complicity; Early warning economic life), 327–329, 332, 335nn9–
systems 10, 600; Atonement Tax, 327; banks’
Goods: club bad, 655; club goods, 11, 25, 501, involvement, 603; business complicity,
504, 505, 654; common-pool resource bad, 599–6 01, 603; Churchill learning of
655; common-pool resource goods, 654; and referring to, 29; conflict economics
creation of “bads,” 12, 655; global public in, 328; crime without a name (1941),
goods, 655; high-value goods, 367–368; 29; development of term “genocide”
impure public goods of prevention, 655; (Lemkin), 29–33; economic benefits to
low-value goods, 368; private bad, 655; Nazi family from, 327; economics of scale
private goods, 654, 660n21; public goods, and, 10; enrichment/i mpoverishment
11, 25, 654, 656, 660n20; types of, 654 dynamic of, 382; flight tax, 327; forced
Government logistics, 401–4 02, 420n4 impoverishment, 327–329, 382; forced
Grants economy, 11, 393–394n1 labor and extermination camps, 329–332,
Great Depression, 21, 296, 583 336nn12–14, 600; German knowledge
Greece, 233, 254–255 of, 318; heavy taxation on Jews, 327–328;
Guatemala, 154–155, 608n1; armed groups’ identity and incentives, 320–327; material
strategies, 253; civil war overlapping incentives in public sector, 324–325;
with genocide, 136t; datasets and trends microeconomic effects of genocide, 212,
of genocides, 79–8 0t; displacement 222; rational solution to implement
correlated with violence, 255; intergroup superior productivity, 335n7; supremacy
social interactions of Mayans and non- thesis vs. economic model for, 332; Topf und
Mayan population, 195 Söhne case study, 600; ultra-Orthodox Jews
Guatemalan National Police Archive, 562n1 rebuilding populations after, 122n18
Guerrilla wars, 77t, 220, 235t, 242, 575, 587n8 Homicide, analogy to genocide, 33, 36, 40, 452,
Guinea, 80t 456- 4 62
Hopefulness theorem, 17
Horizontal inequality, 219, 244, 245, 339, 343,
Hague Conventions (1899 & 1907), 44, 45 345–350, 353; rhetoric of, 346–3 47
Haiti, 80t, 581, 630 Household economics and capital, 13, 14, 18, 19,
Hate crimes, 651 103, 126, 211–217, 221–224, 252, 254–257,
Hate speech, 234 259–2 68, 268–2 69n4, 326–327, 340–3 42,
Hatred, 83t, 172–173, 192, 218, 233, 274–2 86, 345, 352, 361, 387, 412, 426
319, 322, 323, 347, 426, 429, 488–490, 500, Human capital, 117, 125, 129, 175, 215–216, 221,
501–507, 512. See also Anti-Semitism 236t, 245, 311t, 346, 352, 369–370, 476n2,
Health issues, 6, 33, 104, 106, 205, 231, 262, 265, 489, 502; businesses’ role in building, part
267, 297, 331, 333, 388, 443t, 496, 498, of genocide prevention, 607; destruction
632; Colombia, children’s health in, 264; of, 20; displacement of people, 262–2 65;
displacement of people and, 264; relationship investment in, 498
to GMAs, 6. See also Human capital Human Development Index (HDI), 231, 248n1
Healthy economic opportunities, 19–21 Humanitarian aid: effects of specific media
Hegemony, 198 content on, 278, 618; psychological theory
Herero and Nama peoples (South-West Africa), to explain inhibiting decision making
79t, 289, 306, 308, 316n13, 336n14; on, 621
constrained optimization theory (COT) Human rights, 29–30, 37, 41, 42, 53–57, 244,
and genocide of, 300t, 302–304, 308; 245, 251, 283, 315t, 438–439, 591, 594,
economic interdependencies and genocide 602, 606–6 08, 617, 622, 625–627, 633,
of, 293–294, 298, 316n4 635, 642t, 643, 649, 657; affective imagery
Heterogeneous societies, 238, 245, 510, of violations, 630–631; Darfur abuses, 599;
513–514, 532n1 democracies and, 244; indicators, 459, 460,
High-value goods, 367–368 476n1, 629; reporting methods, changes
History of genocides, pre-Holocaust, 289–317. to, 628
See also Pre-Holocaust genocides Human rights law, 629
Hitler’s Willing Executioners, 18 Human Rights Watch, 163–164, 388,
Hobbesian scenarios, 232, 494, 513, 532n4 460, 476n4
Holocaust, 28–36, 318–338; Aryanization of Human Security Report (HSR) (2009-2 010),
Jewish property (expulsion of Jews from 366, 374n4
698 Subject Index
Hungary, 80t, 87t 493t, 499f; GOLKAR party, 494; identity

Hutu Rebels, 96t. See also Rwanda and motivation to participate in genocide,
487–490; mass killing, orchestration
of, 485–4 86; Nahdatul Ulama (NU)
ICTR. See International Criminal Tribunal role, 485–4 86; political background of
for Rwanda parliamentary democracy (1949‒1957),
Identity: choice of, 321, 487–4 88; 491; political background of Sukarno’s
discontinuities, 323–324; economics of, 6, Guided Democracy (1957‒1965), 491–492,
14, 15, 19, 23, 195, 196, 238, 295, 298, 303, 507n1; postpoliticide “New Order” regime,
309, 311t, 320–327, 374, 510, 645, 659n9; 482, 494–499, 495t, 497–499f; principal
future research needs on, 23; Indonesia, identity and resistance to genocide
482–4 83, 487–490; investing in, 322; Nazi, (model), 490, 501–506, 507f; RPKAD
319, 321; persistence of, 322–323; principal role, 485; Suharto’s socioeconomic
identity and resistance to genocide, 490, performance, 494–499, 495t, 497–498f;
501–506, 507f; “us vs. them” identity, 218, transition from Sukarno to Suharto,
219. See also Holocaust; Nazis 481, 483
Inalienability, 647 Indonesian Communist Party (PKI), 483–4 86
Incentives and constraints for mass killings, Indonesian National Party (PNI), 485, 486
146–167, 172–189, 624–635, 640–657; Industrial organization theory, 11, 13, 22, 25,
equilibrium with one-sided mass killings 194, 200, 203, 206, 290, 643
risk, 182–183; equilibrium with two- Inequality, 5, 24, 175, 218–219, 222, 230,
sided mass killings risk, 183–185; identity 231, 236t, 244–2 45, 247, 258, 282, 494,
and incentives, 320–327; incentive 498f, 507n4, 570, 579t, 584–585. 586;
amplification factor, 186; integrated interethnic inequality, 339, 346; rhetoric
model, 178–184; level of fairness making of, 346–3 47. See also Horizontal inequality;
opposition indifferent between peace Vertical inequality
and conflict, 181–182; likelihood of mass Infant mortality rates, as predictive factor for
killings when revenue share is produced genocide, 114, 235t, 576, 577, 581, 582
rather than rent-derived, 181; material Information, economics of, 11, 14, 25
incentives in private sector, 325–327; Institute for the Study of Genocide, 42
material incentives in public sector, Institutional change and effects, 220–221;
324–325; nonproduced rents as necessary future research agenda, 216, 224
condition, 181; stylized facts, 172–175; Institutional Revolutionary Party (PRI,
third-party intervention, 175–178 Mexico), 455
Income. See Economic dimensions of genocides; Interahamwe (Rwanda), 344, 347, 349, 358
Macroeconomic effects of genocide; Internally displaced persons, 5, 121n13,
Microeconomic causes and consequences 251–2 68, 429; Lemkin as, 32; number
of genocide of, 251
Interdependencies (economics and genocide) International agreements and treaties, 641,
schema, 290, 291–298. See also Economic 644, 647, 648, 651–653, 657–658, 659n7,
dimensions of genocides 660n19, 660n21,
India, 35, 80t, 95–9 7t, 492, 494, 603 International Coffee Organization, 344
Indirect persuasion, 14, 274, 280, 281, 285n1 International Criminal Court (ICC), 153, 629,
Indonesia, 15, 19, 481–509; 30 September 641, 642t, 643. See also Rome Statute
Movement, 483–4 84; aftermath of International Criminal Tribunal for Rwanda
mass atrocity, 490–499; armed groups’ (ICTR), 44, 46, 279, 370, 629, 641,
strategies, 253; civil war overlapping 642t, 643
with genocide, 136t; communist purge International Criminal Tribunal for the former
(politicide) of 1965‒1966, 483–4 87, 492– Yugoslavia (ICTY), 46, 48, 370, 641,
493; compared to Zaire, 495–496, 500; 642t, 643
datasets and trends of genocides, 80–81t; International Crisis Behavior (ICB) Project, 52
displacement correlated with violence, International law and policy, 15, 22, 28, 31–35,
255; displacement decision for civilians, 41, 44, 47, 48, 289, 370, 379, 511, 605,
256; diversification and redistribution in 624–635, 639, 641; default rules and
post-mass atrocity economy, 495–499, precommitment devices, 626; economics
496t, 497f; economy and GDP, 492–493, of, 653–657; empowerment of institutions
Subject Index 699
and actors less likely to succumb to psychic Jews, 29, 34, 35, 79t, 318–335, 336n15, 382, 385,
numbing, 627–628; human rights law, 629 387, 599, 601, 602, 629; anti-Semitism,
International Military Tribunal (IMT) at 281–2 82, 335n2, 335n3; occupations in
Nuremberg, 33–3 4, 44, 641, 642t, 659n3. pre-Nazi Germany, 197; threat in terms of
See also Nuremberg trials social privilege and political opposition in
International Monetary Fund (IMF), data Nazi Germany, 204, 382; ultra-Orthodox
collection by, 576 Jews rebuilding populations after the
International Rescue Committee, 365 Holocaust, 122n18
International Terrorism: Attributes of Terrorist Jordan, 82–83t, 96t
Events (ITERATE), 52 Justice, interest in, 307, 642t, 654, 670
International trade. See Trade Justice cascade, 41
Intervention. See Third-party intervention
Intolerance in heterogeneous societies, 512,
513–518, 532n2 Kaldor-H icks efficiency, 644, 659nn5–6
Intrastate conflict, 4, 52, 63, 66t, 67–70, 99n12, Kashmir, 80t, 96t
169n2, 237, 279, 400, 513; defined, 420n1 Kenya, 83t, 93t, 430
Intuitive thinking: affective imagery and, Khmer Rouge, 75t, 107, 386, 406
617, 630–631; genocide prevention by Kiwanja massacre (2008), 8, 163–164
employing, 630–631, 633; moral intuition, Knights Templars, 476
failure of, 624–625, 635; risk management Konstanz One-Sided Violence Event Dataset
and, 615, 616 (KOSVED), 58, 59t, 61t
Inversion rituals, 381 Kosovo Liberation Army (KLA), 156, 435
Iran, 38, 39, 524, 597; civil war overlapping Kosovo war, NATO bombing in, 156
with genocide, 136t; datasets and trends Krupp Case, 600, 605
of genocides, 81t; population losses and Kurds. See Iraq
demographical rates, 109t; population
recovery, 118t, 119
Iran-I raq War, 597 La Familia Michoacana (Mexico), 452
Iraq, 21, 305, 429, 630; Abu Ghraib prisoner Land mine explosions, 264
abuse scandal, 630; civil war overlapping Land pressure. See Malthusian crisis
with genocide, 136t; datasets and trends Laos, 83t
of genocides, 82t, 96t; Kurdish genocide, Law and economics, 11, 22, 25, 639–6 62;
596–597, 603, 604; multilateral conflict, atrocity crimes and legal instruments and
185; population losses and demographical institutions, 640–6 43, 642t; background
rates, 109t; population recovery, 118t, 119; of field of, 643–6 44; contract law, 647–
reconstitution and reconstruction efforts 648; criminal law, 650–651; economics of
in, 21; US intervention to protect Yazidi international law, 653–657; future research
people, 621 needs on, 25; international agreements,
Irish Republican Army (IRA), 435, 449n4 651–653; new version of, 644–6 45;
Iron and steel production, as predictive factors property law, 645–6 47; tort law, 648–650
for genocide, 579t, 581, 587n10 League of Nations, 31, 315t
ISIS [Islamic State], 96t Lebanon, 38; datasets and trends of genocides,
Israel, 82t, 122n18, 278, 402, 406, 603, 83t, 92t, 97t; migration rates, 112f;
607, 608n1 population losses and demographical
Israeli-Palestinian conflict, 278, 402, 406, 603, rates, 108t; population recovery, 118, 118t,
607, 608n1 119, 120f
Italy, 82t, 122n18, 655 Legitimacy: of autocratic regimes, 584; business
complicity and, 604; Congo government,
372, 373; damage to state’s postconflict
Jama’atu Ahlis Sunna Lidda’awati legitimacy, 129; indigenization and,
wal-Jihad, 96t 297; Indonesia dictatorship, 500; of
Japan, 122n18; datasets and trends of genocides, justice cascade, 41; legitimizing myths
82t; population losses and demographical for intergroup behaviors, 197, 198, 204;
rates, 108t; population recovery, 118, 118t, Mexican government, 453, 455, 462, 465–
119, 120f; post-war recovery, 258 472, 477n12; of racial supremacy, 335n4;
Jerusalem, 645 of Sudan government’s acts in Darfur, 598;
700 Subject Index
of superpower intervention, 656; symbiotic 413–414t, 420–421nn7–10; transport

relationship between rulers and capitalists costs, increase in, 127
creating, 367; weak, 221 London Charter of the International Military
Lemkin, Raphael, 12, 28–33, 318; “Acts Tribunal at Nuremberg (1945), 44, 45
Constituting a General (Transnational) Lord’s Resistance Army (LRA, Central and East
Danger Considered as Offences Against the Africa), 93t, 96t, 388–392, 425, 576
Law of Nations” (proposal), 31; archival Love and affection, outlawing of, 386
papers of, 49n1; Axis Rule in Occupied Lower-level civilian atrocities, 12–13, 53, 54–
Europe, 4, 32–33, 35, 145, 157, 211, 383; on 55t, 57t, 59–6 0t, 64, 235t
biological aspects of genocidal techniques, Low-value goods, 19, 368
380, 383–384, 392; creation of crime of Lundin (Swedish energy company), 598,
genocide, 33–36, 318, 379, 511, 640; death 599, 608n6
of, 42; development of term “genocide” by, Lutundula Report, 364
33, 70, 289, 379; as internally displaced
person, 32; legacy, 41–43; Nuremberg
trials and, 33–35; power of place and, 30– M23 insurgents (Rwanda), 363, 369, 374n2
31; on techniques of genocide, 144, 145, Macroeconomic effects of genocide, 13, 18–2 0,
157, 159, 169n10, 290, 296, 316n1, 327; 125–139; capital flight, 5; capital reduction,
Totally Unofficial, 49n1; United Nations and 127, 215; caveats on conclusions, 137–138;
drafting of Genocide Convention, 35–37; civil war overlapping with genocide, effects
on wars of extermination, 69–71, 332; of, 135, 136t; consequences to country’s
World War II role of, 32 infrastructure, 125–126; future research
Leverage, 196, 204, 205, 206 agenda, 223–224; future research needs on,
Lexicographic ordering, 323–325 24; high-value goods, 367–368; impulse
Liability rule, 646, 647, 649 response functions (IRFs) and, 126–127;
Liberation Tigers of Tamil Eelam (LTTE), income inequality, 244–2 45; income
90t, 96t levels, 240–2 42, 259–2 61; international
Liberia, 391, 435, 577, 603; datasets and trends trade and, 245–2 46; literature review,
of genocides, 83–8 4t, 96–9 7t; migration 127; longer-term effects, 24, 128–129,
rates, 112, 112f; natural resources used to 215–216, 264; loss of trust and, 125–126;
finance insurgents, 435, 603; population methodology of study, 126, 130–132;
losses and demographical rates, 108t; productivity loss, 127–128, 132–137, 132t,
population recovery, 118t, 119, 120f 133–134f, 135t; recovery phase, 128–129,
Liberia Peace Council, 96t 130f; resources diverted to militarization
Libya, 175–180, 391, 643; datasets and trends of and security, 127; roots of genocide in
genocides, 84t; multilateral conflict, 185; economic circumstances, 138, 293–294;
US intervention in, 621 structure of income, 242–2 44; transaction
Life force atrocities, 19–2 0, 361, 381, 383, and transport costs, increase in, 127
390, 391 Magnification theorems, 23
Local and national leaders, relationships Malawi, 84t
between, 666 Malaya, 406
Logistics of violence, 399–424; civilian Mali, 84t
violence in contemporary civil wars, Malthusian crisis, 335n7, 339, 340–3 43, 352,
407–411, 408f, 409–410t, 411f; definition 353; skepticism about Malthusian claims,
of logistics, 400–4 01; disruptions to 342–3 43
logistics, 404–4 06; empirical evidence, Manu, code of, 44
406–418; external support, disruptions of, Marshall, Alfred: Principles of Economics, 12
405–4 06; government logistics, 401–4 02, Massacres, defined, 450n15
420n4; large-scale violence’s need for Mass atrocities and killings. See Genocide and
external resources, 404; local and external mass atrocities (GMAs)/genocides and mass
resources, 401; local support, disruptions killings (GMKs); GMA risk and prevention
of, 404–4 05; railroad sabotage in World Mass persuasion, 274. See also Media, role in
War II, 400, 415–418, 416f, 417–419t; mass persuasion; Propaganda, effect in
rebel logistics, 402; scale of violence and, genocide
403–4 04; sources of logistical costs, 403; Media, role in mass persuasion, 274–2 86;
Stalin’s mass resettlements, 411–415, 412f, in conflict environments, 279–2 83;
Subject Index 701
differences-i n-d ifferences (DiD) approach, migrations and, 122n14; militarized

276; effects of specific media content, 278– engineered migrations and, 122n14. See
279; empirical challenges and strategies, also Displacement of people
275–276; field experimentation to study, Militarization, 127, 128, 295, 361, 585; as
276; instrumental variables approaches, predictive of genocide, 581; state-building
276–277; Irregular Terrain Model (ITM), mission focusing on, 665
277, 281; postconflict persuasion, 283; Militarized engineered migrations, 122n14
predisposition to media messages, 18, Millennium Development Goals, 369
282, 284 Minorities at Risk (MAR), 55t, 56, 58t, 236t
Men. See Gender and genocidal economy; Mobility barriers, 200–2 02
Gender-i mbalance ratios Mobutu Sese Seko, 78t, 361, 363, 434, 486,
Mexico drug wars, 14, 15, 17, 452–4 80; 495–496, 500
Calderón’s administration and increased Modernization theory of endogenous
violence, 15, 21, 461–4 62, 475; Calderón’s democracy, 500–501
strategy and security spending, 455–456, Mongolia, 84t
456f; cartels’ actions in model, 464–4 65; Monopoly power, 13, 192, 199–2 00, 359,
cartels’ income from narcotics trafficking, 420n4, 462
476; civilian risk, increase of, 21; Moral hazard: in humanitarian interventions,
corruption and, 15, 463–4 64, 475; current 26n5, 156, 187n11, 315t, 454, 474, 475; in
levels of violence, 456–4 60, 457–459f; torts law, 649, 653
evidence of civilian casualties, 460–4 61, Moral intuition, failure of, 624–625, 635
461f; gender and age of civilian casualties, Movement for the Liberation of Congo
461, 461f; government’s willingness to (MLC), 362
tolerate violence, 476n6; grand coalition’s Moving Out of Poverty (World Bank study), 262
actions in model, 467–470, 469f; history Mozambican National Resistance, 96t
of drug trafficking, 455; literature on Mozambique: datasets and trends of genocides,
determinants of increased homicide 84t, 96t; economic growth post-conflict,
rates, 461–4 62; massacres, 459–4 60, 258–259, 261; population losses and
459f; model of incentives for cooperation demographical rates, 109t; population
and violence, 463–475; models of gang recovery, 118t
violence, 463; National/Subnational Muslims in Pakistan. See Pakistan
(NS) crackdown in model, 465, 470–471; Myanmar (Burma), 84t
number of civilian casualties, 452; pax Myths, legitimatizing of, 173, 197, 198, 204
narcótica, 453–454, 455, 462–4 64, 467,
473, 475; private economic incentives,
464; subnational government-c artel NAFTA (North American Free Trade
(SC) couplet, role in model, 471–473; Agreement), 462
unintended consequences, 17 Nama people. See Herero and Nama peoples
Microeconomic causes and consequences of (South-West Africa)
genocide, 13, 18, 211–229; consequences, "Naming and shaming," 360, 569, 578, 586,
19–2 0, 221–223, 259; disaggregation 592, 599
of violence-related data, need for, 224; National Action Party (PAN, Mexico), 455, 462
distributional advantages sought, 222; National Congress for the Defense of the People
future research agenda, 223–224; future (CNDP, Congo), 163, 362–363
research needs on, 24; incentives, 18– National Consortium for the Study of Terrorism
19, 217–221; inequality and, 218–219; and Responses to Terrorism (University of
literature review, 213–216; pre-Holocaust Maryland), 4, 66t
genocides, 294–295; structures and National groups, definition of, 33
institutions that support or hinder Nationalism, 29, 173, 283, 295, 305, 492, 584
violence, 219–221 National Liberation Army (Colombia). See ELN
Middle-i ncome countries, 105, 186n6, 585 National Liberation Front (NLF, Vietnam), 93t,
Migration, 13, 25n1, 102, 103, 107, 108t, 110, 539, 548t, 549, 554, 555t, 563n13
111–112, 112f, 114, 115, 117–119, 121n13, National nihilism, 296
186, 215, 252–2 68, 327, 342, 658n2; National Patriotic Front of Liberia, 97t
dispossessed engineered migrations, National Resistance Army (NRA, Uganda), 93t,
121–122n14; explorative engineered 254, 425
702 Subject Index
National Socialist German Workers’ Party 85t; national and local elections, 670;
(NSDAP), 281–2 82, 318, 322, 326, 392. See population losses and demographical
also Nazis rates, 108t
National Union for the Total Independence of Nongovernmental organizations (NGOs), 314t,
Angola, 97t 346–3 47, 439; complicity of, 21; displaced
Native Americans. See Yana people persons as focus of, 261; on ethnic
NATO bombing in Kosovo war, 156 cleansing, 47. See also specific NGO by name
Natural disasters: affective imagery in media Nonstate groups as perpetrators, 4, 52–53, 54t,
reporting of, 630; disaster aid, effects of 58, 63, 67, 95–98, 99n6, 231, 234, 235–
specific media content on, 278; economic 236t, 243, 420n1, 425, 654. See also Mexico
costs of, predictive factor for genocide, drug wars
582; willingness to fund life-saving Nonviolent protests, 577
interventions, 616, 618–619 Norm establishment and norm shifting, 14, 18,
Natural resources, 7, 10, 178; business 19, 23, 29–30, 103–104, 167, 168–169, 194,
complicity and, 598, 603, 608n1; economic 206, 211, 214, 216, 218, 221, 223, 224, 233,
diversification and, 495–496; as factor in 274, 283, 360, 385, 511, 514, 532n5, 642t,
rebellion and/or genocide, 19, 175, 181, 643, 653–654, 657, 665
185, 187n14, 233, 242–2 44, 290, 292, Norms on the Responsibilities of Transnational
310t, 356, 358–359, 359t, 391, 402, 435, Corporations and Other Business Enterprises
500, 596; funding genocide in Sudan, 598; with Regard to Human Rights, 606, 642t
linked to intensity of mass killings, 577, North American Free Trade Agreement
587n7; as predictive factor for genocide, (NAFTA), 462
578, 579t, 581, 583 Northern Ireland, 222
Nazis, 15, 21, 28–29, 32, 69, 94t, 318–326, North Korea, 63, 83t, 165, 652
328–330, 332–333, 335n6, 335nn10–11, NPT (Treaty on the Nonproliferation of
583, 592, 596, 600–6 01, 641; Cartel Law, Nuclear Weapons), 652–653
326; charges against and convictions at NRA (National Resistance Army, Uganda), 93t,
Nuremberg, 33–3 4; dividing conquered 254, 425
people into races that could be Aryanized, Nuremberg trials, 33–37, 44–4 6, 386, 640–
387; economic policy of, 9, 10, 326, 380– 641, 642t, 653; business complicity cases,
381, 384–387; eliminating threats to “pure” 599–6 00, 605. See also London Charter
German people and values, 204, 382–383; of the International Military Tribunal at
historical antecedent of pogroms at time Nuremberg
of Black Death in 1300s, 322, 335n3;
media persuasion used by, 275, 281–2 82;
as most murderous regime, 239; racism to Oil companies, business complicity of, 608n1,
strengthen Third Reich, 387; rise to power, 608n6. See also Natural resources
199; in wake of Great Depression, 21, 583. Operation Barbarossa, 28
See also Holocaust; World War II Opportunity costs, 9, 12, 23, 143, 167, 187n14,
Necropolitics, 369, 373 252, 257, 316n7, 339, 343, 462, 513, 521,
Neoliberalism, 369 524–525, 532n3
Nepal: armed groups’ strategies, 253; datasets Optimal choice, 7–8, 145, 146–156, 179–180.
and trends of genocides, 84t; displacement See also Constrained optimization theory
correlated with violence, 254, 255 Optimization, defined, 144
Netherlands, 84t, 360 Organizational studies, 200
NGOs. See Nongovernmental organizations Ottoman Empire. See Turkey/O ttoman Empire
Nicaragua: Contra repression causing civilians Oxfam, 363, 373
to switch sides, 449n12; datasets and
trends of genocides, 84–85t; population
losses and demographical rates, 109t; Pakistan, 4, 16, 67, 289, 510–533; agriculture
population recovery, 118t, 119 and Shiite killings, 525–530, 526–528t,
Niger, 85t 529t; civil war overlapping with genocide,
Nigeria, 391, 499, 499f, 585, 654; blockade of 136t; datasets and trends of genocides, 85t,
Biafra, 406; business complicity, 608n1; 97t, 524; democratic decentralization, 670;
civil war overlapping with genocide, economic development and killings, 525;
136t; datasets and trends of genocides, Green Revolution, 525; methodology and
Subject Index 703
model for study, 521–530; Sunni and Shia Postconflict reconstruction, 6, 16, 20–21, 104,
branches of Islam, 532-533n7 268, 607, 663–673. See also specific countries
Papua New Guinea, 86t, 304 Postgenocide reconciliation, role of media
Paraguay, 86t in, 283
Pareto efficiency criterion, 328, 659n5 Post-t raumatic stress disorder, 265
Patani Insurgents, 97t Poverty: cycle of, as by-product of genocide,
Path-dependence, 170n11 222; as driver of genocide, 576, 578, 579t,
Peacekeeping studies, 587n3 581; as factor for participating in armed
Penn World Table, 13, 126 conflict, 215; forced impoverishment,
People’s Revolutionary Party (PRP, 327–329, 382; Rwanda genocide and, 339,
Vietnam), 549 343–3 45, 576; vulnerability to, 215, 260
Persuasion. See Direct persuasion; Indirect Poverty traps, 117, 260, 267, 268, 268n3. See also
persuasion; Media, role in mass persuasion Behavioral poverty traps
Peru, 38, 86t, 425, 431 Power: analysis of, 204; defined, 196;
Philippines, 38, 86t, 136t dominance, 196, 204; as factor in genocide,
Phoenix Program (South Vietnam) database, 173–174, 238–2 39; within rebel army, 233;
537–539, 540, 559 social structure as source of, 196
Poisson model, 408, 409t, 414, 414t, 416, Predictive modeling of GMAs, 569–590;
442–4 43, 450n21, 521–523, 527t, 529; calibration of the probabilities and,
negative binomial equation and, 523, 530t; 573; differences between predictive and
zero-i nflated poisson (ZIP) equation, analytical models, 570–576; economic
522–523, 529t variables, role of, 578–583, 579–580t; false
Poland, 30, 32, 42, 46, 86t, 251, 320, 383 positive rate as fall-out, 573; forecasting
Polarization, 23, 175, 219, 236t, 238, 247, 291, efforts, 583–586; hypothesis testing vs.,
426, 484, 486, 500 571; inferences, 572; potential of economic
Policy lessons, 20–22; behavioral economics, for variables, 583–586; precision, 587n4;
civilian atrocity prevention, 166–167. See Receiver Operating Characteristic (ROC)
also Law and economics; Reconstruction analysis, 573–574, 574f; true positive rate
Polish Supreme National Tribunal, conviction of as sensitivity or recall, 573
Nazis by, 34 Predisposition to hatred, 18, 275, 281–2 82, 284.
Political conduct, 198–2 05; aggressive behavior, See also Norm establishment and norm
203; defined, 191, 192; hegemony, 198; shifting
interplay with social structure, 193; Pre-Holocaust genocides, 17, 289–317;
making genocide more likely, 190; mobility constrained optimization theory, 298–306;
barriers, 201; political competitiveness, genocidal conflict as deliberate choice,
199; sabotage, 203 291–292; inderdependencies schema and,
Political exclusion, 13, 19, 38–39, 58t, 173, 193, 291–298
197, 203, 218, 222, 239, 298, 310t, 328, Prevention. See GMA risk and prevention
373, 584 Principal-agent relationship, 244, 315t, 329,
Political Instability Task Force, 578 330, 335n5
Political Instability Task Force (PITF), 52, 61t, Production function, 9–10
174, 186n1 Productive efficiency, 10
Political Instability Task Force’s geno-politicides Productivity. See Macroeconomic effects of
(PITF-G), 53, 54t, 56t, 63, 64, 67 genocide
Political Instability Task Force Worldwide Program on Forced Migration and Health at
Atrocities Dataset (PITF-A), 58, 59t, 61t, the Mailman School of Public Health
63, 64, 65f, 69 (Columbia University), 120n1
Political reconstruction, democracy in, 663–674 Prominence effect, 620–623
Political Terror Scale (PTS), 53, 55t, 57t Propaganda, effect in genocide, 234, 274–2 86;
Political upheaval, 235t, 237. See also Civil war Holocaust and, 319, 322; Rwanda and,
Politicide. See Indonesia 234, 279–2 83, 284, 347, 351–352. See also
Population pressure leading to Rwanda Media, role in mass persuasion
genocide, 339, 340–3 43 Property law, 645–6 47
Portugal, 86t Prospect theory, 23, 617–618, 618f, 623
Posner, Richard A., 644, 648, 652–653, Psychic numbing, 22, 23, 613–638; affect
659n4, 660n21 and analysis in risk situations, 616–623;
704 Subject Index
behavioral economics, overview of, civilians to join, 254–255; rebel-helping

614–615; compassion fade, 620, 624; price policies, 154–155; short-term rewards
empowerment of institutions and actors for, 233; third-party support, effects of, 434
less likely to succumb to, 627–628; Receiver Operating Characteristic (ROC)
implications for international law and analysis, 573–574, 574f
policy, 624–635; implications of the Reference point dependence and loss aversion,
psychological account, 623–624, 624f; 145, 164–167, 166f, 618
institutional features that foster, need to Refugees, 5, 111, 121n13, 234, 242, 251–273,
address, 628–629; insulating institutions 294, 314t, 342, 429, 460, ; as focus of
from effects of, 626–628; prominence attention, 21, 102; global public good and
effect, 620–623; psychophysical function, Chad’s rights, 655
617–618, 624; rapporteurs to recognize and Regional organizations, empowerment to stop
counter the risks of, 628 genocides, 627
Psychological costs of GMAs, 24, 306; forced Reinforcement mechanisms, 19
displacement and, 265, 269n5 Religious groups, 97t, 222; LRA as, 390–391;
Psychophysical numbing, 617–618, 624 religious mass killings in Pakistan
Public economics, 25. See also Goods (1978‒2012), 520–533. See also Holocaust;
Punishment: as deterrent, 651; by eradication, Jews; Pakistan
174; credible threat of, 329; UN Genocide Rent-seeking, 221, 464, 646, 659n10
Convention, 3, 37, 212, 370, 640, 642t, Reprisal killings, 58t, 415, 421n12
Purification by genocide, 204, 310t, 379 Resettlement. See Displacement of people
Resistance, 12, 41, 68, 167, 169n2, 297, 302,
304, 306, 312–313t, 329, 364, 430, 449n11,
Quigley’s The Genocide Convention: An 465, 490
International Law Analysis, 370 Resource-abundant countries, 175, 187n14,
242–2 43
Resource-poor genocides, 388–392
Radio Télévision Libre des Mille Collines Resource-r ich genocides, 386–388
(RTLM), 279–2 81, 282, 347 Responsibility to Protect (R2P), 63, 315t, 626,
Railroad sabotage. See Logistics of violence 642t, 643
Rally for Congolese Democracy (RCD), Restitution, 641, 660n16
359, 362 Retributive justice, 649
Rape and sexual violence, 6, 17, 20, 30, 46, Revolutionary Armed Forces of Colombia.
640–6 41; in Armenian genocide, 387; See FARC
Congo and, 163, 365; genocide and, 102; Revolutionary United Front (RUF, Sierra
Holocaust and, 385; as life force atrocities, Leone), 388, 391–392
381, 393; by LRA and Uganda, 389–390; Risk. See GMA risk and prevention
by peacekeeping troops, 26n5; by RUF in Risk management, 268n3, 616
Sierra Leone, 391; Rwanda and, 350, 352; Ritual desecrations, 381
Sudan and, 597–598 Rivalry, 192, 200–2 02, 205, 513, 654; for
Raphael Lemkin Human Rights Award, 42 collaboration of locals, 426
Raphael Lemkin Seminars for Genocide ROC (Receiver Operating Characteristic)
Prevention (Auschwitz Institute for Peace analysis, 573–574, 574f
and Reconciliation), 42 Romania, 86–87t
Rapid economic decline, 24, 344, 584 Rome Statute of the International Criminal
Rapporteurs to recognize and counter the risks Court, 25n1, 40, 657, 658–659n2; crimes
of psychic numbing, 626–628 against humanity and, 46–47, 640, 647;
Rational choice models, 8–9, 146, 481– war crimes and, 44, 640
482, 614, 622. See also Constrained Root-a nd-branch genocide, 379, 390
optimization theory RTLM. See Radio Télévision Libre des Mille
RCD (Rally for Congolese Democracy), Collines
359, 362 Rummel Democides dataset, 53, 54t, 57t
Rebels, 169n2; in activist rebellions, 233; Russia/USSR, 46, 239, 295, 304, 308; Caucasus
diaspora support to, 434–435; killing Wars of 1816‒1864, 405–406; cold war
civilians, 242, 243; logistics, 402; in interventions, effect of, 434; datasets
opportunistic rebellions, 233; reasons for and trends of genocides, 87–88t; forced
Subject Index 705
collectivization of agriculture, 296–297, 305; Sabotage, 202–2 03, 206; Soviet partisan
Lenin’s New Economic Policy (NEP) of 1921, railroad sabotage in World War II, 415–
297; media persuasion, 277; social media 418, 416f, 417–419t, 421n12
linked to political participation, 277, 284; Saddam Hussein, 82t, 596–597, 603–6 04, 621
Soviet partisan railroad sabotage in World War Save the Children, 619
II, 415–418, 416f, 417–419t, 421n12. See also Scarcity, 7, 9, 193, 255, 340–3 42, 583. See also
Stalinism; Ukrainians (genocide under Stalin) Deprivation
Rwanda, 98n2, 339–355; age distribution Security of property and person, 6, 290,
disruption, 112, 113f, 114; aid through 297–298, 632
Congo war, 360; civil war overlapping Seleka Rebels, 75t, 97t
with genocide, 136t, 220, 580; Congo war Self-help remedies, 651– 652
and, 358–359, 434; datasets and trends Self-i mage. See Identity
of genocides, 88–89t; democratization Separatism, 670– 671
and ethnic cleansing, 175; difficult- Serbo-Croatian war, 283, 406
to-reach villages, less violence in, 235; Sherman’s March to the Sea, 406
displacement of large-scale landholders Shiite in Pakistan. See Pakistan
and cattle owners, 254; distributional Shining Path (Peru), 86t, 425, 431
effects of genocide, 222; enrichment/ Siberia, mass exile to. See Stalinism
impoverishment dynamic of genocide, Sierra Leone: behavioral responses to violence,
382, 387; exports and production of 266; civil war without genocide, 580;
natural resources, 358–359, 359t; gender- datasets and trends of genocides, 89t, 97t;
imbalance ratios, 115–116, 115f, 116t, economic growth post-conflict, 258–
352; Genocide Convention and, 40, 259; natural resources used to finance
655; government using thugs, instead of insurgents, 435, 577, 603; population
armed forces, to commit ethnic cleansing, shifts, 266. See also Revolutionary United
233; horizontal inequality and, 339, Front (RUF)
345–350; as intermediate conflict, 68; Sikh Insurgents, 97t
kill or be killed dynamic, 219–220; Slavery. See Forced labor
logistical constraints during Rwandan Social capital, 19, 20, 24, 129, 136–137, 196,
genocide, 400; looting and attack on 204–2 06, 265, 277, 311t, 313t, 511, 514–515
richest, 347–3 49; M23 insurgents and, Social Conflict in Africa Database (SCAD),
363, 369, 374n2; macroeconomic effects 60t, 62t
of genocide in, 126, 137; Malthusian crisis Social contract, 25, 192, 500, 513–514, 516,
(population pressure) and, 339, 340–3 43, 532n4, 607
352; microeconomic effects of genocide, Social exclusion, 222, 297–298
212, 222; migration rates, 111, 112, 112f; Social media: affective imagery of, 630; linked
number killed, 4, 67, 68, 99n8, 353n1; to political protests, 284
poor prospects for advancement as factor Social niche, defined, 195
in genocide, 339, 343–3 44; population Social structure, 6, 193–197; defined, 191,
losses and demographical rates, 107, 108t, 192; dominance, 196, 204; exogenous
110–111; population recovery, 118t, 120f; vs. endogenous, 196, 241–2 42, 514–
postgenocide aftermath, 352; postgenocide 517; group-based inequalities, 197;
reconciliation, role of media in, 283; homogeneous, 195, 205; intergroup
poverty as factor in genocide, 339, 343– competition and social interactions, 195–
345, 576; propaganda’s role in, 234, 279– 197; interplay with political conduct, 193;
283, 284, 347, 351–352; recruitment from leverage, 196, 204, 206; making genocide
those with fewest economic opportunities, more likely, 190; new forms of, 222;
344–3 45; regulated limited mobility, relationships changed after mass atrocities,
341–3 42; rhetoric of inequality, 346–3 47; 222–223; rivalry, 192; Rwanda genocide
school enrollment of displaced children, and, 339, 350–352; social classification as
263; social structure, 339, 350–352; Tutsi mechanism of genocide, 204; as source of
defined as adversaries, 204, 280–2 81; power, 196. See also Class as impetus
victims identified as troublemakers, 233, Societies, defined, 194
349–350. See also International Criminal Somalia: datasets and trends of genocides, 89t;
Tribunal for Rwanda (ICTR) displacement decision for civilians, 256;
Rwandan Patriotic Front, 97t population losses and demographical rates,
706 Subject Index
109t; population recovery, 118t, 119; UN Suharto. See Indonesia

authorization of Unified Task Force in, 176 Sukarno. See Indonesia
South Africa, 39, 89t, 222, 302, 373, 581, 648 Sunk cost, defined, 198
South Korea, 122n18, 493t, 501, 560, 560t; Sunni in Pakistan. See Pakistan
datasets and trends of genocides, 83t; Sun Tzu, 44
population losses and demographical rates, Superpower intervention/nonintervention, 656.
108t; population recovery, 118t See also Third-party intervention
South Vietnam. See Vietnam Supreme values, 323–324
Sovereignty, 31, 38–39, 156, 282–2 83, 314-315t, Symbioses between war and mass atrocities, 71
370, 420n1, 660n16, 672 Symbols of existential strength of groups, 6,
Spain: Basque region, economic effects of long- 381–382, 385–386
running conflict in, 128; datasets and Syria: civil war overlapping with genocide, 136t;
trends of genocides, 89t; displacement datasets and trends of genocides, 91t, 96t;
correlated with violence, 254–255 external support, disruption of, 406; forced
Spanish civil war, 89t, 449n5 deportations to, 295, 304–305; IDPs in,
Speed of killing, 32, 308–309, 335, 339, 350 251; multilateral conflict, 185; US lack of
Srebrenica, 26n5, 48 response to mass atrocities, 621
Sri Lanka: civil war overlapping with genocide,
136t; datasets and trends of genocides, 90t,
96t; post-war recovery, 259 Tajikistan, 91t, 224n1, 263, 421n7
Stalinism: freedom of movement, limitations on, Taliban, 72t, 97t, 169n4
297; "Marxism and the National Question," Talisman Energy Inc., 598
296; mass resettlements, logistical costs of, Tamil Tigers, 434
411–415, 412f, 413–414t, 420–421nn7–10; Targeting of certain populations intentionally,
purges of kulaks, 39, 204, 289, 297; scarcity 186n1, 212; lack of prior intent, 232
and, 7; in wake of Great Depression, 21 Taxation on Jews during Holocaust, 327
Starvation, 15, 32, 85t, 145, 157–162, 294, 296, Terra nullius, 294
301, 305–306, 406, 649 Terrorism datasets, 59–62, 99n9; Global
State-and nonstate-perpetrated mass Terrorism Database (GTD), 58–62
atrocities, 63 Theorem, defined, 26n4
State-building missions, 664–6 65. See also Thick market externalities, 515, 532n6
Third-party intervention. Third-party intervention, 7, 13, 22–2 3, 151,
State failure, 235–2 36t, 237, 471, 578, 580 154, 156–157, 162, 175–178, 185, 199,
State-owned enterprises, 492, 599, 602 299, 490, 538, 653; automatic triggers for,
Stay or leave decision. See Displacement 26n5; balance of power, effect on, 434;
of people increasing costs of genocide, 577; state-
Strategic atrocities, 425–451. See also Violence building intervention and selection of
against civilians leadership, 667; superpower intervention/
Strategic complementarities, 194 nonintervention, 656; willingness to fund
Strategic substitutes, 194 life-saving interventions, 618–619. See also
Subnational analysis in correlation of income Unintended consequences; United Nations
level with mass atrocities, 241 Thucydides: The War of the Peloponnesians and the
Subsistence crops, 262 Athenians, 172
Substitution possibilities in GMK, 17, 157–164; TMT (Turkish Military Tribunal) (1919‒1920),
high input substitution possibilities in 641, 642t
constrained optimization model, 158, 159f; Topf und Söhne case study, 600–6 01
synchronized attack against out-g roup’s Tort law, 648–650, 651, 652
existence, 157–158 Total genocides, 382–383, 388
Sudan: business complicity in genocide on Totalitarian regimes, 205, 238–2 39
Christian population, 597–599; civilians Total war, 69, 384
targeted by People’s Liberation Army, 425; Trade: African mineral trade, 364; arms trade,
civil war overlapping with genocide, 136t; 596; comparative advantage and, 10; cost-
datasets and trends of genocides, 90–91t, benefit analyses and, 256; disruption of,
96t; duration of mass atrocity, 63; IDPs in, 19, 205, 290, 314t, 591; drug trade, 15, 455,
429; population losses and demographical 459, 461, 476; funding genocidal regimes,
rates, 109t; population recovery, 118t, 119. 594, 595; Herero in German South-West
See also Darfur Africa as traders, 293, 294; high-value
Subject Index 707
goods, 368; in Indonesia GDP growth, Ulfelder and Valentino (UV) mass killing
493t, 495t; material incentives in public dataset, 53, 54t, 56t, 62, 63, 64
sector and, 324; new law and economics Unemployment: during Holocaust, 197; in
and, 645; openness as driver of genocide, Pakistan, 525; in Rwanda, 339, 345; of
5, 24, 152f, 175, 230, 235–2 36t, 240, 245– displaced persons, 263; predictive factor
246, 576, 579t, 581, 582–583, 585, 586, for genocide, 582
587n10; primary commodity exports, 577; Unilateral intervention, 22, 314t
Rwanda and, 358–361; slave trade, 592 Unintended consequences, 8, 13, 21–22, 26n5;
Tradeoffs, 148, 280, 311t, 313–314t, 324, 330, third-party efforts to protect oppressed
336n14, 449n11, 465, 472–473, 621, 632– groups producing, 17, 22, 156, 169n4, 653.
635, 634f, 655 See also Backfire condition
Transaction costs, 126, 128, 206, 644–6 45, Union of Congolese Patriots, 97t
647–6 48, 652, 656 United Kingdom, 93t, 660n16; aid to
Transport. See Logistics of violence Rwanda, 360
Treaty on the Nonproliferation of Nuclear United Liberation Movement for Democracy in
Weapons (NPT), 652–653 Liberia, 97t
Trends of genocides. See Datasets, use of United Nations: Congo and UN forces, 368;
Trespass, 647 Darfur peacekeeping mission, 369; data
Tribal clashes, 83t, 97t collection by, 576; General Assembly,
T’ruah: The Rabbinic Call for Human Rights, 42 Resolution 96 (I), 36; Guiding Principles
Trust issues, 6, 125–126, 136, 173, 216, 221, on Business and Human Rights, 606; High
224, 313t, 324‒325, 489‒490, 513, 591, 615, Commissioner for Human Rights,
665‒667 622, 631; High Commissioner for
Turkey/O ttoman Empire, 195; Arab Revolt of Refugees (UNHCR), 251; humanitarian
1916‒1918, 406; Armenians and education, intervention by, 621; Human Rights
196; datasets and trends of genocides, 92t; Council reports, 628; Lemkin and, 35;
Young Turks, 195, 199. See also Armenian "Methods for Testing Adult Mortality,"
genocide 121n11; Mission in the DRC (MONUC,
Turkish Military Tribunal (TMT) (1919‒1920), later UN Stabilization Mission in the DRC
641, 642t or MONUSCO), 163; Panel of Experts,
Twitter, 284 360; Population Division database, 107;
Two-t rack development of conflict datasets, purpose of, 624; Rwanda, role in, 282–2 83;
67–71; genocide gap in defense and peace Security Council, Resolution 794, 176;
economics and, 68–69; symbioses between Security Council, Resolution 1973, 175,
war and mass atrocities, 71; war/mass 643; Security Council authorization of use
atrocity nexus and, 69–71 of force for atrocities reaching certain level,
626. See also specific conventions
United Self-Defense of Colombia (AUC), 438–
UCDP. See Uppsala Conflict Data Program 439, 441, 449n14, 575
Uganda: Acholi national group, 389; aid through United States: Alien Tort Claims Act, 642t, 643,
Congo war, 360; armed groups’ strategies, 649; co-fi nancing Yana extermination,
254; civilian deaths in massacres, 425; 298; cold war interventions, effect of, 434,
civil war overlapping with genocide, 500; datasets and trends of genocides,
136t, 220; Congo war and, 358–359, 370, 93t; genocide definition, 39; firearms
434; datasets and trends of genocides, smuggling, 462; ignoring genocides, 613,
92–93t; economic growth post-conflict, 623; Mexico drug trafficking and, 455;
258; exports and production of natural role of AM radio in New Deal, 277; state-
resources, 358, 359t; GDP per capita, 365f; building and Articles of Confederation,
population losses and demographical 673; superpower intervention/
rates, 109t; population recovery, 118t, nonintervention by, 621, 656; Vietnam
119; risk perception, 262. See also Lord’s War, 539
Resistance Army Unity party (Russia), 277
Ukrainians (genocide under Stalin), 289, 309; Universal jurisdiction, 40, 48; principle of, 641
constrained optimization theory (COT) Unwilling executioners, 18
and genocide of, 300t, 305–306, 307; Uppsala Conflict Data Program (UCDP),
economic interdependencies and genocide 52, 60t, 62t, 63, 98n5, 448n3, 453, 576,
of, 296–297 586, 659n2
708 Subject Index
Uppsala Conflict Data Program One-Sided 559–561, 559f, 560t; population losses
Violence Dataset (UCDP-V), 53, 54t, 57t, and demographical rates, 109t; population
62, 63, 64 recovery, 118t, 119; post-war recovery, 258;
Uppsala Conflict Data Program/Peace Research Provincial Reconnaissance Units (PRU),
Institute Oslo (UCDP/PRIO) armed 560; real GDP per capita, 499f; taxonomy
conflict dataset, 4, 68, 69, 99n11 of suspects, 550–552, 552f; taxonomy of
Urdu language, 197 targeting operations, 557–558, 557t, 558f;
US Genocide Prevention Task Force, 172–173 Tet Offensive (1968), 406, 538; To Van
US National Research Council, 120n1 Xiem as example narrative and coding, 541;
USSR. See Russia/USSR; Stalinism variable selection in database, 544–5 47,
Utility payoffs from behavior, 321, 333–335, 615 546f; victims of targeting, 547–550, 548t,
551t; war in South Vietnam, 538–539
Villagerization, 260
VAC. See Violence against civilians Violence: benefits to certain groups, 216;
Value of human lives, 616–620, 617f, 619f; causes of organized violence, 214–215;
damages in tort actions, 650; US action and compelling choice, 18; consequences of,
inaction in certain humanitarian situations 215–216; ethnic, 670–671; incentives,
related to, 621 214–215; inequality and, 219; intrastate
VCI Neutralization and Identification conflict, defined, 420n1; logistics of, 399–
Information System (VCINIIS), 540 424; nonparticipation risks in rebellion,
Vertical inequality, 244–2 45, 343–3 47, 353 214–215; participation risks in rebellion,
Victims: difficulty in describing (Congo), 371, 214; preconditions for appearance of, 193;
374; empowerment of, 631; minimizing recurring outbreaks post-genocide, 222;
victimization by understudy of, 24; redressing imbalances, 213; structures
opportunity cost for, 9; passivity of, and institutions that support or hinder,
307, 316n11; troublemakers as label for 219–221; vulnerability to, 215; welfare
(Rwanda), 233, 349–350; undermining outcomes and, 214, 216. See also Civil war;
as essential element to perpetrator Ethnic cleansing; Logistics of violence;
sustainability, 380–381. See also Rape and sexual violence; Violence against
Demography of genocide; Displacement of civilians
people; Resistance; Value of human lives; Violence against civilians (VAC), 13, 15, 17,
Violence against civilians 19, 221, 236t, 425–451, 659n2; civilians
Viet Cong, 16, 93t, 402, 406, 549, 562 switching sides in civil wars, 431, 449n9,
Viet Cong Infrastructure (VCI), 549 449n11; collaboration of locals and, 18,
Vietnam, 16, 536–565; Advanced Research 426, 449n4; in contemporary civil wars,
Projects Agency, 539; attributes, 541–5 42; 407–411, 408f, 409–410t, 411f; data
civilian targeting program (Phoenix regarding, 52–53, 58, 61–62t, 71, 85t;
Program), 537–539, 540, 559, 561; in death toll, 431, 449n14; in DR Congo,
coffee trade, 344; Combined Military 163; economic determinants of, 234;
Interrogation Center (CMIC), 541; empowerment and altering balance of
datasets and trends of genocides, 93t, power, 434–437; equilibrium in model,
536, 563n3; external support, disruption 432–434, 448; ethnolinguistic groups
of, 406; Greenbook coding for jobs, 553, and, 238; fear and terror, 426; literature
555t; grouping attributes and records, review of, 230–2 47; logistics and, 399;
542–5 44; Ho Chi Minh Trail for model to study pattern of killings, 427–
transport, 402; impossibility of defeating 431; rebels commit, 220, 233; timing
insurgency supported by majority of of actions, 431–432. See also Civilian
people, 430; Intelligence Coordination and atrocities, datasets on; Colombia; Mexico
Exploitation, 563n4; jobs held by suspects, drug wars
553–555, 553f; methods of targeting, Virtual Research Associates (VRA), 57t, 58,
556–557, 556t; National Police Command 60t, 62t
Data Management Center, 539; National Voting and voter turnout, 279, 285n3, 298, 670
Police Infrastructure Analysis Subsystem Vulnerability to poverty, 18, 19, 215, 259, 260,
II (NPIASS-I I), 536, 539, 540t, 541, 542t, 265, 267
543, 543f; overview of targeting database, Vulnerability to reputational damage, 592
539–5 47; perpetrators of targeting, Vulnerability to violence, 18, 215
Subject Index 709
Vulnerable groups, 17; persistent exclusion of, Yana people (California), 14, 289, 309;
19, 173, 218 constrained optimization theory (COT)
and genocide of, 299–301, 300t; economic
interdependencies and genocide of,
Wannsee Conference (1942), 325, 330 292, 298
War crimes, 28, 34, 44–45, 600, 641; associated Yemen: datasets and trends of genocides, 94t;
with genocide, 3, 43–4 4; defined, 25n1, population losses and demographical
44, 640; distinguished from genocide, 45, rates, 109t; population recovery,
47–4 8, 640; history of, 44 118t, 119
War on drugs. See Mexico drug wars Young Turks, 195, 199, 387
Wealth appropriation, 6–7, 159, 290, 296–297. Youth bulge, 106, 579t; as predictive factors for
See also Natural resources genocide, 587n10
WikiLeaks video of US soldiers indiscriminately
killing Iraq civilians, 630
Women. See Gender and genocidal economy; Zaire, 174, 434; compared to Indonesia,
Gender-i mbalance ratios 486, 495–496, 500. See also Democratic
World Bank: Congo economy and, 360, 364, Republic of the Congo (DRC)
365, 365f, 366, 368; data collection by, 576; Zanzibar, 95t
good governance measures, 629; Moving Zapatista revolution (Mexico), 431
Out of Poverty study, 262 Zero-sum dynamic, 379–380, 382, 391–392,
World Events Interaction Survey (WEIS), 52 393n1, 426, 591
World food price crisis (2007), 364 Zimbabwe, 492; Congo war and, 434;
World War II, 4, 15, 28–29, 70, 119, 147, 184, datasets and trends of genocides,
220, 258, 283, 289, 365, 486, 492, 578, 599, 95t; school enrollment of displaced
641; Soviet partisan railroad sabotage in, children, 263; stunting, 264;
400, 407, 415–418, 416f, 417–419t, 421n12. sanctions, 627
See also Holocaust; Nazis Zyklon B Case, 600

(Oxford Scholarship Online) Anderton, Charles H. - Brauer, Jurgen - Economic Aspects of Genocides, Other Mass Atrocities, and Their Prevention-Oxford University Press (2016)

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Oxford Scholarship Online) Anderton, Charles H. - Brauer, Jurgen - Economic Aspects of Genocides, Other Mass Atrocities, and Their Prevention-Oxford University Press (2016)

Uploaded by

Copyright:

Available Formats

Economic Aspects of Genocides,

Other Mass Atrocities,

Published in the United States of America by Oxford University Press

© Oxford University Press 2016

All rights reserved. No part of this publication may be reproduced, stored in

You must not circulate this work in any other form

Library of Congress Cataloging-​i n-​P ublication Data

Dedicated to my children, Mahlobo, Lwandle, and Nokuthula,

PART ONE ECONOM IC S A N D M A SS AT ROCI T I E S: OV ERV I E W

1. On the Economics of Genocides, Other Mass Atrocities, and Their

2. “A Crime without a Name”: Defining Genocide and Mass Atrocity 28

3. Datasets and Trends of Genocides, Mass Killings, and Other Civilian

4. The Demography of Genocide 102

5. The Macroeconomic Toll of Genocide and the Sources of Economic

PART T WO T H EOR ET IC A L A PPROACH E S A N D R E V I E WS OF

6. Genocide and Mass Killing Risk and Prevention: Perspectives from

7. Incentives and Constraints for Mass Killings: A Game-​Theoretic

8. Genocide: From Social Structure to Political Conduct 190

9. The Microeconomic Causes and Consequences of Genocides and Mass

10. Development and the Risk of Mass Atrocities: An Assessment of the

12. Media Persuasion, Ethnic Hatred, and Mass Violence: A Brief

13. “For Being Aboriginal”—​Economic Perspectives on Pre-​Holocaust

14. Identity and Incentives: An Economic Interpretation of the

15. The Economics of Genocide in Rwanda 339

17. Gender and the Genocidal Economy 378

18. On the Logistics of Violence: Evidence from Stalin’s Great Terror,

19. Strategic Atrocities: Civilians under Crossfire—​Theory and Evidence

20. From Pax Narcótica to Guerra Pública: Explaining Civilian Violence in

21. Long-​Term Economic Development in the Presence of an Episode of

22. Economic Foundations of Religious Killings and Genocide with

23. Understanding Civil War Violence through Military

PART FIVE TOWA R D PR EDICT ION A N D PR E V EN T ION

24. Economic Risk Factors and Predictive Modeling of Genocides and

25. Business in Genocide: Understanding and Avoiding Complicity 591

26. Valuing Lives You Might Save: Understanding Psychic Numbing in

27. Genocides and Other Mass Atrocities: A Law and Economics

28. Local and National Democracy in Political Reconstruction 663

Name Index 675

In compiling this book, we accumulated great intellectual and personal debt to

Charles H. Anderton is W. Arthur Garrity Sr. Professor in Human Nature,

Willa Friedman is Assistant Professor, Department of Economics, University of

Peacebuilding, Graduate Institute for International and Development Studies,

Maren M. Michaelsen is Assistant Professor of Microeconomics, Faculty of

Massimo Morelli is Professor of Economics, Department of Policy Analysis

Andrés Moya is Assistant Professor of Economics, Department of Economics,

S. Mansoob Murshed is Professor, International Institute of Social Studies,

Roger B. Myerson is Glen A. Lloyd Distinguished Service Professor of

Wim Naudé is Dean-​ Director of the Maastricht School of Management,

Kimberly G. Olson is an independent researcher and policy analyst, Oregon

Maria Petrova is Research Professor, Department of Economics and Business,

Dominic Rohner is Assistant Professor, Department of Econometrics and

David Schap is Professor of Economics, Department of Economics and

Paul Slovic is President, Decision Science Research Institute, and Professor of

Dimitrios Soudis is Lecturer, Faculty of Economics and Business, Department of

Nora M. Stel is Research Fellow, Maastricht School of Management, Maastricht,

ECONOMICS AND MASS

On the Economics of Genocides, Other

1.2. What Has Economics Got To Do With It?

4. GMA as a 6. GMA Distorts

Choice Types Diagnostics 5 Ds Lootable Assets Facets 1. Health

Figure 1.1 Economic aspects of genocides and other mass atrocities.

of post-​GMA political, economic, and cultural reconstitution and reconstruc-

Library of Congress Cataloging-i n-P ublication Data

7. Incentives and Constraints for Mass Killings: A Game-Theoretic

13. “For Being Aboriginal”—Economic Perspectives on Pre-Holocaust

19. Strategic Atrocities: Civilians under Crossfire—Theory and Evidence

21. Long-Term Economic Development in the Presence of an Episode of

Wim Naudé is Dean- Director of the Maastricht School of Management,

of post-GMA political, economic, and cultural reconstitution and reconstruc-

ill-considered expectations. Such phenomena are nonrational (but not necessar-

nineteenth-century economist Alfred Marshall put it, economics is “a

reviews the literature on GMA-related population displacement. It explains how

1.3.2.4. Policy Lessons for GMA Prevention and Post-G MA

and—in subsumed form—ethnic cleansing. These categories of mass atrocity

Petrovic, D. 1994. “Ethnic Cleansing—A n Attempt at Methodology.” European Journal of

3.2.2. State/Nonstate Group Civilian Atrocity Datasets

Table 3.2 Definitions and Sources for State/Nonstate Group Civilian

3.3. State-and Nonstate-Perpetrated Mass

3.5. Two-Track Development of Conflict Datasets

3.5.1. Examples of the Two-Track Development

In the post-Holocaust world in which quantitative research on peace and war

3.5.2. The Two-Track Development and the Genocide Gap

3.5.3. The War/Mass Atrocity Nexus