Professional Documents
Culture Documents
To cite this article: Charles H. Anderton & John R. Carter (2011): CONFLICT DATASETS: A PRIMER
FOR ACADEMICS, POLICYMAKERS, AND PRACTITIONERS, Defence and Peace Economics, 22:1, 21-42
This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden.
The publisher does not give any warranty express or implied or make any representation
that the contents will be complete or accurate or up to date. The accuracy of any
instructions, formulae, and drug doses should be independently verified with primary
sources. The publisher shall not be liable for any loss, actions, claims, proceedings,
demand, or costs or damages whatsoever or howsoever caused arising directly or
indirectly in connection with or arising out of the use of this material.
Defence and Peace Economics, 2011
Vol. 22(1), February, pp. 21–42
Prominent conflict datasets used in the social scientific study of war and peace are summarized. These include
Defence
10.1080/10242694.2010.491677
1024-2694
Original
0Taylor
00
canderton@holycross.edu
CHARLESANDERTON
2010
000
00 &andArticle
Francis
(print)/1476-8267
Peace Economics(online)
datasets for armed conflicts (e.g., COW and UCDP/PRIO wars and sub-war conflicts), terrorism (e.g., GTD and
ITERATE terrorist incidents), and events (e.g., WEIS, KEDS, and VRA cooperative and conflict actions). Topics
explored include the diversity of conflict types, alternative definitions of war and sub-war conflict, historical patterns
of violence in the international system, and degree of overlap across datasets.
INTRODUCTION
The empirical study of war has exploded in recent decades, owing in large part to extraor-
dinary advances in computer power and the development of numerous datasets on violent
conflict. This article examines selected conflict datasets that are prominent in the social
scientific study of war and peace. As the title suggests, the article is designed to be a
primer for those who are relatively unfamiliar with conflict datasets. As such, we do not
cover the historical development, relative empirical performance, or intricate coding proce-
dures of the datasets. Rather, we organize, summarize, and compare the various datasets
and thereby provide what we hope will be a useful gateway to the empirical studies of war
and peace.
We begin with a typology of conflicts that focuses on the level of hostility and the nature
of the actors. We then use the conflict typology to categorize and summarize 30 well-
established datasets, including those for armed conflicts (e.g., COW and UCDP/PRIO wars
and sub-war conflicts), terrorism (e.g., ITERATE and GTD terrorist incidents), and events
(e.g., WEIS, KEDS, and VRA cooperative and conflict actions). Among the topics explored
ISSN 1024-2694 print: ISSN 1476-8267 online © 2011 Taylor & Francis
DOI: 10.1080/10242694.2010.491677
22 C.H. ANDERTON AND J.R. CARTER
are the diversity of conflict types, operational definitions of war and sub-war conflict,
selected empirical patterns of violence in the international system, and degree of overlap
across related datasets.1
A TYPOLOGY OF CONFLICTS
in Figure 1, replicated here from Anderton and Carter (2009). The horizontal axis measures
time, while the vertical axis gauges the degree of friendliness or hostility between parties. The
bell-shaped curve is a simplified representation of the full cycle of a conflict, showing how
relations between parties might range from durable and stable peace to unstable peace, crisis,
and war. In some cases, the parties might continue for years in just one or two of the five
phases shown along the vertical axis; for example, the USA and the USSR operated in the
unstable peace and crisis phases during the Cold War. Around the outside of the curve are vari-
ous third-party interventions that can affect the course of a conflict. As Lund (1996: 385)
notes, the ‘P series’ (preventive diplomacy, peacemaking, peace enforcement, etc) is common
in documents and discussions associated with the United Nations, whereas the ‘C series’
(conflict prevention, crisis management, conflict management, etc.) is widespread in the
academic literature. Figure 1 also depicts arrows emanating from the curve, which show that
conflicts can be prevented, escalate, or recur.
We focus on the middle and upper phases of the conflict cycle corresponding to unstable
FIGURE 1 Lund’s (1996) conflict cycle (adapted from Lund, 1996: 386).
peace, crisis, and war. Under unstable peace, the parties perceive themselves as rivals and
competitively structure their armaments vis-à-vis one another, but outbreaks of violence are
relatively rare and limited in severity. Once the crisis stage of the cycle is reached, relations
are overtly hostile, armed forces are on high alert, low-level violence may occur, and the risk
of high-level violence is palpable. War is a high level of violence involving sustained fight-
ing between the rivals and substantial casualties. In what follows, we deviate slightly from
Lund’s terminology. First, we note that in a similar conflict cycle model, Pfetsch and Rohloff
(2000: 385) describe ‘latent conflict’ in terms very similar to those used in Lund’s descrip-
tion of unstable peace. In our categorization of conflict datasets we follow Pfetsch and
Rohloff and adopt the ‘latent conflict’ language. Second, in order to avoid confusion with the
International Crisis Behavior Project’s formal definition of ‘crisis,’ we designate the crisis
level of hostility in the conflict cycle as ‘sub-war conflict.’ Hence, in our categorization of
conflict datasets we distinguish three levels of hostility: (1) latent conflict, (2) sub-war
conflict, and (3) war.
1
Some of our coverage of conflict datasets adapts and extends similar work found in Anderton and Carter (2007,
2009). For advanced coverage of the difficulties associated with the development and use of conflict datasets, see the
edited volume of Kauffmann (2008). For valuable summaries of more than 60 conflict datasets, see Eck (2005). The
International Peace Research Institute in Oslo (PRIO) provides dozens of replication conflict datasets for articles
published in the Journal of Peace Research and in its Centre for the Study of Civil War (see www.prio.no/CSCW/
Datasets). Among the data areas we do not cover are arms trade and arms production (Brauer, 2007; Stockholm
International Peace Research Institute, 2008), military alliances (Leeds et al., 2002; Gibler and Sarkees, 2004), mili-
tary spending (Brzoska, 1995; Stockholm International Peace Research Institute, 2008), strategic or enduring inter-
state rivalries (Colaresi and Thompson, 2002; Klein et al., 2006), enduring intrastate rivalries (DeRouen and
Bercovitch, 2008), third-party intervention (Regan et al., 2009; see also UCDP’s data on managing low intensity
conflict at www.pcr.uu.se/research/UCDP/data_and_publications/datasets.htm), peace missions (Stockholm Interna-
tional Peace Research Institute at www.sipri.org/contents/webmaster/databases), peace agreements (UCDP peace
agreements at www.pcr.uu.se/research/UCDP/data_and_publications/datasets.htm), and conflict geography and
resources (PRIO data at www.prio.no/CSCW/Datasets/Geographical-and-Resource).
CONFLICT DATASETS 23
Downloaded by [FNSP Fondation National des Sciences Politiques] at 12:34 16 September 2012
FIGURE 1 Lund’s (1996) conflict cycle (adapted from Lund, 1996: 386)
We distinguish datasets for armed conflicts, terrorism, and events. Armed conflict datasets
generally code the presence or absence of a certain level of hostility, be it latent, sub-war, or
war. Although terrorist incidents appear in some armed conflict and events datasets, terrorism
datasets have generally developed along their own track. Hence, we summarize terrorism
datasets separately from those on armed conflicts and events. In contrast to armed conflict and
terrorism datasets, events datasets generally code a larger number and wider range of cooper-
ative and hostile actions between or within states so as to provide a much more detailed picture
of cooperation and conflict.
Conflict characteristics
Armed conflict datasets Latent Conflict Sub-War War Fatality Criterion Interstate Intrastate Extra-State Non-State Time period
across all four actor types. The next three datasets are provided by the Political Instability Task
Force (PITF), covering revolutionary and ethnic intrastate wars, genocides/politicides, and
adverse regime changes. The 12th is the International Crisis Behavior (ICB) Project’s dataset
on interstate crises, while the 13th is the Minorities at Risk (MAR) Project’s dataset on civil,
inter-communal, and factional conflicts. The final two datasets cover civil wars and are
provided by political scientists James Fearon, David Laitin and Nicholas Sambanis.
As shown in Table I, there is variation among the datasets in terms of the hostility levels
gauged, whether a fatality criterion is used to determine hostility levels, the types of actors
involved, and the time periods covered. Further points of contrast among the datasets are
Downloaded by [FNSP Fondation National des Sciences Politiques] at 12:34 16 September 2012
revealed in Table II, where we summarize the specific conflict definitions used by the vari-
ous organizations. For datasets that use a fatality criterion, there are differences in whether
fatalities are counted per year or in total and whether fatalities include both military and
civilian deaths. Table II includes the Web addresses for the conflict datasets where interested
readers can find additional details in coding manuals, downloadable data files, and additional
literature.
Differences in definitions and other coding practices can and do lead to differences in the
classification of specific conflicts, even among datasets that cover the same hostility level,
actor types, and time period. For example, the Correlates of War (COW) Project classifies the
conflict between the Sri Lankan government and the rebel group, Janatha Vimukthi Peramuna,
as an intrastate war for the period 1987–1989. COW presents no data on intrastate conflict in
Sri Lanka during the 1990s, presumably because the seriousness of conflict was below its war
threshold. In contrast, the UCDP/PRIO dataset classifies conflict between the Sri Lankan
government and the rebel group, Liberation Tigers of Tamil Eelam, at a war level of intensity
for the years 1987, 1990–2001, and 2006–2007.
Terrorism Datasets
In Tables III and IV we summarize five well-known terrorism datasets. Along the top of Table
III we distinguish various characteristics of the datasets including the range or types of terror-
ist incidents, whether international or domestic incidents (or both) are covered, and the time
period. The first dataset is International Terrorism: Attributes of Terrorist Events (ITERATE),
which provides information on terrorist incidents that transcend state boundaries. The second
through fourth datasets cover both international and domestic terrorism incidents and are
provided by the Global Terrorism Database (GTD), the National Counterterrorism Center’s
Worldwide Incidents Tracking System (WITS), and RAND’s Worldwide Terrorism Incident
Database (RWTID). The fifth dataset, Terrorism in Western Europe: Events Data (TWEED),
provides information on domestic terrorist incidents for various Western European countries.
Table IV provides additional information about each dataset’s terrorism definition and Web
address. A previously well-known and popular online terrorism data source was the Terrorism
Knowledge Base (TKB). Unfortunately, TKB is no longer freely available to the public,
although it appears that RWTID incorporates TKB data. For a summary of TKB’s demise, see
Houghton (2008).
Events Datasets
At any given phase of a conflict cycle, the parties under consideration will typically direct a
variety of actions toward one another. For example, a nation in the crisis phase of the conflict
cycle might threaten a rival with military action, recall diplomats, call on international medi-
ators to help dampen the crisis, request assistance from an ally, offer new carrots to the rival,
and/or call on the rival to accept the latest peace offer. Each of these events is a component of
CONFLICT DATASETS 27
‘[E]xplicit contention between two or more nation-states’ over a piece of territory, a river, or a maritime zone
(Hensel and Mitchell, 2009).
UCDP/PRIO Armed Conflicts (www.prio.no/CSCW/Datasets/Armed-Conflict/UCDP-PRIO/)
War is substantial combat between parties leading to a minimum of 1,000 battle-related deaths (military and civil-
ian) per year among the parties involved. Sub-war conflict applies the definition for war, but with battle-related
deaths between 25 and 999 (UCDP/PRIO Codebook 4-2007).
UCDP One-Sided Violence Against Civilians (www.pcr.uu.se/research/UCDP/data_and_publications/)
Use of armed force by a state, or by a formally organized group, against civilians which results in at least 25 civilian
deaths per year (UCDP One-Sided Violence Codebook 1.3).
UCDP Non-State Conflicts (www.pcr.uu.se/research/UCDP/data_and_publications/datasets.htm)
Violence between two organized groups, neither of which is a state, leading to at least 25 battle-related deaths (mil-
itary and civilian) per year among the parties involved (UCDP Non-State Conflict Codebook 1.0).
CSP Major Episodes of Political Violence (www.systemicpeace.org/warlist.htm)
Violent episodes involve at least 100 directly-related (military and civilian) deaths per year and at least 500 directly-
related deaths in total (information accessed from CSP Website on 24 November 2009).
CSP Coups d’Etat (www.systemicpeace.org/inscr/inscr.htm)
The ‘forceful seizure of executive authority and office by a dissident/opposition faction within the country’s ruling
or political elites that results in a substantial change in the executive leadership and the policies of the prior regime’
(Marshall and Marshall, 2007: 1).
PITF Revolutionary/Ethnic Wars (globalpolicy.gmu.edu/pitf/pitfpset.htm)
Conflict between a government and a politically organized group or communal minority leading to a minimum of
1000 directly-related deaths (military and civilian) for the whole war among the parties involved. In addition, each
party mobilizes at least 1000 people and there must be at least one year where the annual conflict-related deaths
exceed 100 (Marshall et al., 2009: 5).
PITF Genocides/Politicides (globalpolicy.gmu.edu/pitf/pitfpset.htm)
‘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 et al., 2009: 14).
PITF Adverse Regime Change (globalpolicy.gmu.edu/pitf/pitfpset.htm)
‘[A]dverse shifts in … governance, including … revolutionary changes in political elites and the mode of gover-
nance; contested dissolution of federated states or secession of a substantial area of a state by extrajudicial means;
and complete or near-total collapse of central state authority’ (Marshall et al., 2009: 10).
ICB Crises (www.cidcm.umd.edu/icb/)
A state’s foreign policy leaders ‘perceive a threat to basic values, time pressure for response and heightened prob-
ability of involvement in military hostilities’ (Brecher and Wilkenfeld, 2007: 12).
MAR Conflicts (www.cidcm.umd.edu/mar/)
Hostility ranges from harassment to sporadic violence to protracted warfare (Minorities at Risk Project, 2005).
Fearon and Laitin Civil Wars (www.stanford.edu/group/ethnic/)
Conflict between a state and non-state groups with at least 1,000 fatalities for the whole war, a yearly average of at
least 100 fatalities, and at least 100 killed on both sides (Fearon and Laitin, 2003: 76).
Sambanis Civil Wars (pantheon.yale.edu/∼ns237/index/research.html#Data)
Conflict between a state and one or more internal groups characterized by at least 500 deaths caused by conflict in
the first year or 1000 cumulative deaths in the initial three-year period, no three-year period with fewer than 500
deaths caused by conflict, and substantial resistance by the weaker party (Sambanis, 2004: 829–830).
the larger crisis and can potentially provide valuable information about the evolution of the
crisis through time, how the crisis is resolved, or why the crisis escalates to war. Events
datasets use news sources or other information to code a large number of cooperative and
28 C.H. ANDERTON AND J.R. CARTER
Dataset characteristics
nappings
3. National Counterterrorism 14 categories including arson, bomb- X X 2004–2009
Center’s (NCTC) Worldwide ings, hijackings, and kidnappings
Incidents Tracking System
(WITS)
4. RAND Worldwide Terrorism 10 categories including arson, assas- X X 1968–present
Incident Database (RWTID) sinations, bombings, hijackings, and
kidnappings
5. Terrorism in Western Europe: 10 categories including arson, vari- X 1950–2004
Events Data (TWEED) ous types of bombings (e.g., letter,
car, fire), rocket/grenade attacks, and
kidnappings
conflictive actions between and within states. In so doing, events datasets can provide numer-
ous details about relations between and within the parties through the various phases of a
conflict cycle.
In Table V we summarize ten well-known events datasets. Along the top row of the table
we distinguish various characteristics of the datasets including the range of cooperative and
conflictive events, the nature of the parties (interstate, intrastate, extra-state, or non-state), and
the time period. The first dataset is the seminal World Events Interaction Survey (WEIS),
which documents cooperative and conflictive actions for all four actor types across 22 broad
categories and 61 sub-categories of events. The second is the Conflict and Peace Databank
(COPDAB), which reports interstate and intrastate cooperative and conflictive events along a
15 point scale. The third dataset is Behavioral Correlates of War (BCOW), which codes data
across more than 100 sub-categories of events in the context of interstate crises. The next
dataset, the Intranational Political Interactions (IPI) Project, focuses on cooperation and
conflict events between groups residing within the state, one of which can be the state itself.
The fifth dataset is Armed Conflict Location and Event Data (ACLED), which provides daily
information about battlefield events for selected states affected with civil war. ACLED notes
in its codebook that battlefield events can occur between a government and an internal or
external non-state group or between two non-state groups. The sixth dataset is the Center for
Systemic Peace’s (CSP) data for conflict events within India. The seventh is FAST Interna-
tional, developed by the Swiss Peace Foundation (SPF) for the early warning of conflicts. The
eighth is the dataset provided by political scientists Gary King and Will Lowe, which encom-
passes more than 150 sub-categories of conflict and cooperative events. The final two datasets
are from the Virtual Research Associates (VRA) and the Kansas Events Data System (KEDS).
VRA and KEDS can be thought of as ‘events data factories’ because they create numerous
datasets across a large number (100 or more) of sub-categories of conflict and cooperative
events.
Further points of contrast between events datasets are shown in Table VI where we provide
summary information about data sources, coding methods, geographic scope, and number of
event records. Notice that the first six datasets use manual methods (human readers) to code
events data. The seventh dataset, FAST International, uses local staff in various countries to
CONFLICT DATASETS 29
gather raw data, which are passed on to country coordinators for coding using standardized
methods (Hämmerli et al., 2006: 161–163). The final three datasets, King/Lowe, VRA, and
KEDS, use specialized software programs that can read and code an extraordinary amount of
information from news sources.
In our experience, the armed conflict and terrorism datasets summarized in Tables I–IV are
relatively user-friendly, whereas the events datasets summarized in Tables V and VI are less
so. There are a number of reasons for this. First, armed conflict and terrorism datasets typically
contain a few thousand (or tens of thousands of) rows of data in a familiar and accessible
spreadsheet format. In contrast, events datasets can contain hundreds of thousands or even
millions of records, which utilize zip files and less familiar formats such as SPSS or Stata.
Second, the move to computer-coding of information sources by some data organizations
requires that the user have some familiarity with the datasets’ specialized methodologies.
Third, event data in raw form often need to be translated into indices of conflict or cooperation
before they can be used in statistical studies of conflict.
Downloaded by [FNSP Fondation National des Sciences Politiques] at 12:34 16 September 2012
30
Dataset characteristics
Events datasets Range of events Interstate Intrastate Extra-State Non-State Time period
1. World Events Interaction Survey (WEIS) 22 broad categories and more than 61 sub-categories X X X X 1966–1978
of conflict and cooperation events
2. Conflict and Peace Databank (COPDAB) 15 point scale of conflict and cooperation events X X 1948–1978
3. Behavioral Correlates of War (BCOW) 7 broad categories of physical and verbal events and X 1816–1980
more than 100 sub-categories of conflict and cooper-
ation events
4. Intranational Political Interactions (IPI) 20 broad categories and more than 100 sub-categories X X 1979–1992
Project of conflict and cooperation events
5. Armed Conflict Location and Event Data Locations, dates, and other aspects of battlefield X X X 1960–2004
(ACLED) events in states affected with civil war
6. Center for Systemic Peace (CSP) India Locations, dates, and other aspects of violent events X X 1960–2004
Sub-National Problem Set within India
7. Swiss Peace Foundation (SPF) FAST Locations, dates, and other aspects of conflict and co- X X X X 1998–2008
International operation events
8. King/Lowe Events Data 34 broad categories and more than 150 sub-categories X X X X 1990–2004
of conflict and cooperation events
9. Virtual Research Associates (VRA) 38 broad categories and more than 200 sub-categories X X X X Mid 1990s-present
of conflict and cooperation events and real time basis
C.H. ANDERTON AND J.R. CARTER
WEIS (www.icpsr.umich.edu/cocoon/ICPSR/STUDY/05211.xml)
Data source: New York Times
Coding method: manual
Geographic scope: global
No. of event records: 98,043
COPDAB (www.icpsr.umich.edu/cocoon/ICPSR/STUDY/07767.xml)
Data sources: newspapers, chronologies, and other historical sources
Coding method: manual
Geographic scope: global
Downloaded by [FNSP Fondation National des Sciences Politiques] at 12:34 16 September 2012
In an influential article, Goldstein (1992) developed a scale for translating 61 WEIS sub-
category codes into numerical scores of conflict and cooperation. At the high end of conflict
is a military attack, clash, or assault (WEIS sub-category 223), which receives a score of –10
on the Goldstein scale. The high end of cooperation is the extension of military assistance
(WEIS sub-category 072), which receives a Goldstein score of +8.3. Intermediate actions
receive Goldstein scores close to zero; for example, when an actor asks for information (WEIS
sub-category 091), the Goldstein score is +0.1. Figure 3 shows the Goldstein translation of the
32 C.H. ANDERTON AND J.R. CARTER
Downloaded by [FNSP Fondation National des Sciences Politiques] at 12:34 16 September 2012
FIGURE 3 WEIS event data record in raw form translated according to the Goldstein scale
first event record in the WEIS dataset, which contains close to 100,000 such records. The
record indicates that the United States visited the United Kingdom on January 1, 1966 to
consult about the Vietnam conflict. The visit (WEIS sub-category 032) receives a Goldstein
score of +1.9. As events datasets have developed and evolved, WEIS event categories and sub-
categories have been rearranged and expanded. For example, the King/Lowe, VRA, and
KEDS datasets cited earlier in Table V contain 100 or more event sub-categories. These event
sub-categories are translated into indices of conflict and cooperation using scales that are
modifications of, or roughly equivalent to, the Goldstein scale.2
Figure 4 shows an example of an events data trend for India based on materials provided
FIGURE 3 WEIS event data record in raw form translated according to the Goldstein scale.
by the Virtual Research Associates (VRA). On a daily basis, VRA receives numerous elec-
tronic stories from various news agencies, including Thomson Reuters and Agence France
Presse. The stories are processed by VRA software, which assigns parts of speech and
semantic tags to each clause in the news stories. VRA further processes the codes to
develop several dozen diagnostics measures of conflict and cooperation including indices
for country stability, interstate interactions, violence, and contentiousness. Figure 4 shows
the 5-week simple moving average for India’s contentiousness in the weeks and months
prior to the November 2008 Mumbai attacks. VRA operationalizes contentiousness by first
totaling the number of contentious actions for India for a given week as defined by WEIS
category codes 11–22. The number of contentious actions is then divided by the number of
all actions, giving rise to contentious actions as a proportion of all actions, which is
measured on the vertical axis. Note the generally higher moving averages for India’s
contentiousness in September and October, prior to the November 2008 Mumbai attacks.
Figure 4 is an example of how events data can potentially serve as an early warning indica-
tor of conflict. The example of course is highly stylized because it is based on a simple
moving average. To predict changes in conflict risk in a more sophisticated way, VRA
applies various market analysis techniques to measures of conflict and cooperation derived
from events data (e.g., moving average convergence/divergence (MACD), Bollinger Bands,
and Ultimate Oscillator).
FIGUREAdapted
Source: 4 Simple
from5-week
data provided
moving by
average
Virtual
forResearch
India’s contentiousness,
Associates, Inc. January-October 2008.
2
In the last row of Table V, KEDS develops events data using WEIS codes. Hence, these KEDS data can be trans-
lated into indices of conflict and cooperation using the Goldstein scale. KEDS also offers CAMEO-coded events,
which are translated into numerical scores of conflict and cooperation in a way that is roughly equivalent to the
Goldstein scale (see http://web.ku.edu/keds/cameo.dir/CAMEO.SCALE.txt). Integrated Data for Events Analysis
(IDEA) is an events data categorization scheme that rearranges and extends the WEIS categories and sub-categories
of conflict and cooperative events (see Bond et al., 2003; King and Lowe, 2003). The IDEA categories and sub-cate-
gories are prominent in the events data of FAST International, King and Lowe, and VRA. Goldstein scores of
cooperation and conflict have been adapted to the IDEA framework (see King and Lowe, 2003: 622–623).
CONFLICT DATASETS 33
Downloaded by [FNSP Fondation National des Sciences Politiques] at 12:34 16 September 2012
FIGURE 4 Simple 5-week moving average for India’s contentiousness, January-October 2008
Source: Adapted from data provided by Virtual Research Associates, Inc.
We now turn our attention to historical patterns of violent conflict in the international system
as revealed in selected conflict datasets. We focus in particular on wars, sub-war conflicts,
genocides, and terrorism, drawing on information from prominent conflict data organizations
such as COW, PRIO, UCDP, PITF, ITERATE, and GTD. The data reveal not only the
frequency of violent conflicts over time, but also the diverse nature of conflict types that
pervade global society. We also report results of a detailed analysis of overlaps between
conflict datasets based on methods similar to those used by Hewitt (2003) to compare COW
militarized interstate dispute dyads and ICB interstate crisis dyads. For the common time
frame of 1918–1992, Hewitt found that 23% of the 2155 dispute dyads could be matched with
corresponding crisis dyads, while 75% of the 756 crisis dyads could be matched with corre-
sponding dispute dyads. In our own analysis, we compare the degrees of overlap across COW
and UCDP/PRIO for interstate and extra-state wars; across COW, UCDP/PRIO, and PITF for
intrastate wars; and across COW and UCDP/PRIO for sub-war interstate conflicts. The
Appendix to this article describes our comparison methodology.
Interstate Wars
Figure 5 shows the stock of interstate wars by year since 1946 based on data from COW
(1946–1997) and UCDP/PRIO (1946–2008). For a majority of years shown in the figure, there
is at least one interstate war present. In recent years, however, there have been a relatively low
number of interstate wars. Figure 6 also reveals that COW and UCDP/PRIO have non-trivial
differences in their counts of interstate wars. Of the 52 years in which the two datasets overlap
(1946–1997), 16 (31%) have different counts on the number of interstate wars. Our more
34 C.H. ANDERTON AND J.R. CARTER
Downloaded by [FNSP Fondation National des Sciences Politiques] at 12:34 16 September 2012
detailed analysis indicated that of the 23 interstate wars codified by COW from 1946–1997,
16 (70%) are also classified as interstate war by PRIO. Going the other way, of the 25 inter-
state wars tracked by PRIO over the same period 1946–1997, 19 (76%) are also classified as
interstate war by COW.
FIGURE Uppsala
Sources: 56 Number
Conflict
of interstate
intrastate
Data Program
wars in(UCDP),
the international
International
system,
Peace
1946–2008.
Research Institute, Oslo (PRIO), Gleditsch et al. (2002), and Correlates of War (COW) Project Intrastate
Interstate War Version 3.0.
Intrastate Wars
In Figure 6 we plot the stock of intrastate wars by year since 1946 based on data from COW
(1946–1997), UCDP/PRIO (1946–2008), and PITF (1948–2007). The figure shows that there
Downloaded by [FNSP Fondation National des Sciences Politiques] at 12:34 16 September 2012
has been at least one intrastate war for each year shown and, in some years, there have been a
dozen or more intrastate wars. Comparing Figures 5 and 6 we see that intrastate wars have
been far more prevalent than interstate wars in recent decades. Figure 6 also shows that COW,
UCDP/PRIO, and PITF have substantial differences in their counts of intrastate wars. In none
of the years do all three data organizations report the same number of intrastate wars, and in
some years PITF reports more than double the number of wars counted by the other datasets.
Inspection of the definitions of war in Table II shows that PITF has a more inclusive war
threshold compared to COW and UCDP/PRIO. In particular, PITF classifies a case as war if
there are 1000 battle-related (military and civilian) deaths for the whole war. COW, however,
only considers military fatalities in its otherwise equivalent fatality threshold. UCDP/PRIO
considers military and civilian deaths, but its 1000 fatality threshold is on a per year basis
rather than on the per war basis followed by PITF and COW. Hence, it is not surprising that
PITF has larger counts of intrastate wars.
Our more detailed analysis revealed that of the 104 intrastate wars in the COW dataset from
1946–1997, 71 (68%) are also classified as intrastate war by UCDP/PRIO. Reversing direc-
tion, of the 74 intrastate wars tracked by UCDP/PRIO over the period 1946–1997, 57 (77%)
are also classified as intrastate war by COW. We also found that of the 94 intrastate wars codi-
fied by COW from 1955–1997, 89 (95%) are also classified as intrastate wars by PITF. Of the
116 intrastate wars tracked by PITF from 1955–1997, 77 (66%) of them were also classified
as intrastate wars by COW. We also compared intrastate war cases between UCDP/PRIO and
PITF. Of the 120 intrastate wars tracked by PITF over the period 1955–2007, 68 (57%) were
also classified as intrastate wars by UCDP/PRIO. Going the other way, of the 69 UCDP/PRIO
intrastate wars from 1955–2007, 63 (91%) were also classified as intrastate wars by PITF.
Extra-state Wars
Figure 7 shows the stock of extra-state wars by year since 1946 based on data from COW
(1946–1997) and UCDP/PRIO (1946–2008). A striking feature of the figure is the lack of
extra-state wars over the past two decades. The figure also reveals substantial differences in
the counts of extra-state wars by year across COW and UCDP/PRIO. Of the 52 years in which
the two datasets overlap (1946–1997), only 12 (23%) have equal counts on the number of
extra-state wars. More detailed analysis revealed that of the 17 extra-state wars codified by
COW from 1946–1997, seven (41%) are also classified as extra-state war by PRIO. Going the
other way, of the seven extra-state wars tracked by PRIO over the period 1946–1997, all seven
are also classified as extra-state war by COW.
FIGURE Uppsala
Sources: 7 Number
Conflict
of extra-state
Data Program
wars in
(UCDP),
the international
International
system,
Peace
1946–2008.
Research Institute, Oslo (PRIO), Gleditsch et al. (2002), and Correlates of War (COW) Project Extra-state War Version 3.0.
Non-state Wars
To date there has been a lack of systematic data available to social scientists on non-state wars.
COW, however, may soon introduce data on non-state wars, and UCDP has introduced a new
dataset on non-state conflict (Sundberg, 2009). The UCDP data currently cover the period
36 C.H. ANDERTON AND J.R. CARTER
Downloaded by [FNSP Fondation National des Sciences Politiques] at 12:34 16 September 2012
2002–2007, and two non-state conflicts in the dataset report yearly fatalities exceeding 1000,
which is UCDP’s threshold for war.
Sub-war Conflicts
In Figure 8 we summarize the stock of sub-war conflicts in the international system since 1946
based on data from UCDP/PRIO and COW. Note that since the late 1950s UCDP/PRIO reports
a preponderance of intrastate sub-war conflicts relative to interstate and extra-state sub-war
conflicts. The figure also indicates a relatively large number of non-state sub-war conflicts
reported by UCDP, albeit for the more limited period 2002–2007 for which data are available.
The Fatal MIDs series is provided by COW and represents sub-war interstate conflicts involv-
ing at least one fatality. Since the UCDP/PRIO and Fatal MIDs series track sub-war interstate
conflicts, there is a basis to compare their degree of overlap. We found that of the 32 interstate
sub-war conflicts coded by UCDP/PRIO from 1946–1997, 30 (94%) are also classified as sub-
war fatal MIDs by COW. Going the other way, of the 282 sub-war fatal MIDs tracked by COW
over the period 1946–1997, 40 (14%) are also classified as sub-war interstate conflicts by
PRIO. The asymmetry in these counts is not surprising given that COW needs only a single
fatality for a sub-war interstate conflict to qualify as a fatal MID, whereas UCDP/PRIO
requires a minimum of 25 battle-related fatalities per year for inclusion in its dataset.
FIGURE Uppsala
Sources: 8 Number
Conflict
of sub-war
Data Program
conflicts(UCDP),
in the international
International
system,
Peace 1946–2008.
Research Institute, Oslo (PRIO), and Gleditsch et al. (2002) for interstate, intrastate, and extra-state data; Uppsala Conflict Data Program (UCDP) and Kreutz and Eck (2005) for non-state data; Correlates of War (COW) Project Militarized Interstate Disputes v3.1 for fatal MIDs.
Genocides/Politicides
Like war, genocide can create substantial casualties, but unlike most wars, genocide is large-
scale violence or repression directed against non-combatants that is designed to kill a substantial
CONFLICT DATASETS 37
Downloaded by [FNSP Fondation National des Sciences Politiques] at 12:34 16 September 2012
portion of people from a particular group (Harff, 2003). In Figure 9 we present annual data on
the stock of genocides/politicides (left axis) and estimated fatalities from genocides/politicides
(right axis) for the period 1956–2007 based on data from PITF. The figure shows that for the
last half century there has been at least one genocide/politicide ongoing per year in the inter-
national system, and in some years there have been ten or more. Estimated annual fatalities
from genocides/politicides are above 50,000 for most years shown in Figure 9, and in many
years the fatality estimates are in the hundreds of thousands.
FIGURE Political
Sources: 9 NumberInstability
and estimated
Task Force
fatalities
(PITF)of Dataset
genocides/politicides,
on Genocides and
1956–2007.
Politicides at http://globalpolicy.gmu.edu/pitf/pitfpset.htm, Marshall et al. (2009), and Harff (2003).
Terrorism
Figure 10 shows the number of terrorist incidents worldwide over time, while Figure 11 does
the same for casualties (deaths plus injured) per incident. Each figure plots two time series,
one drawn from the Global Terrorism Database (GTD) and the other from International
Terrorism: Attributes of Terrorist Events (ITERATE). Three summary observations are noted.
First, the huge differences in the frequency of incidents shown by the two sources in Figure
10 almost surely are due primarily to the fact that GTD includes both international and domes-
tic incidents, while ITERATE excludes domestic incidents. Thus, the gap between the two
series suggests that domestic terrorism incidents worldwide may have been 8–10 times more
frequent than international incidents in recent decades (Anderton and Carter, 2009: 130).
Second, perhaps surprisingly, neither frequency series shows any obvious trend, upward or
downward, over the past 30 years. The surprise is muted, however, when attention shifts to the
intensity of terrorist incidents, which leads to the third observation. As shown in Figure 11, the
levels of casualties per incident are noticeably higher over the past 15 years, suggesting that
the severity of terrorist attacks has increased markedly.
38 C.H. ANDERTON AND J.R. CARTER
Downloaded by [FNSP Fondation National des Sciences Politiques] at 12:34 16 September 2012
CONCLUSION
FIGURE LaFree
Sources: 11
10 Casualties
Terrorism
and Duganincidents
per(2007),
incident
worldwide,
Global
from terrorism,
Terrorism
1968–2007.
1968–2007.
Database-START (with
accessed
1993
onmissing)
12/31/09,accessed
and International
on 31 December
Terrorism:
2009,
Attributes
and International
of Terrorist
Terrorism:
Events. Attributes of Terrorist Events.
In this age of information we have witnessed a remarkable and immensely valuable prolifera-
tion of conflict datasets. As more is learned about the nature and scope of conflict, established
datasets have been refined and new datasets have been created. We are deeply indebted to the
authors of these datasets. They have wrestled with definitions, typologies, and coding rules;
they have toiled in the collection of hard facts and figures; and they have shared the fruits of
their work widely. Without their datasets a truly quantitative and cumulative social scientific
study of war and peace would not be possible.
ACKNOWLEDGEMENTS
We are grateful to Doug and Joe Bond of Virtual Research Associates (VRA) for their assis-
tance with VRA events data and to Daniel Arce and an anonymous referee for valuable input.
Conflict data information presented herein is as of December 2009. Recently, COW updated
its war data through 2007.
References
Anderton, C.H. and Carter, J.R. (2007) A survey of peace economics. In Handbook of Defense Economics, Volume
2, edited by T. Sandler and K. Hartley. New York: Elsevier, 1211–1258.
Anderton, C.H. and Carter, J.R. (2009) Principles of Conflict Economics: A Primer for Social Scientists. New York:
Cambridge University Press.
Bond, D., Bond, J., Oh, C., Jenkins, J.C. and Taylor, C.L. (2003) Integrated data for events analysis (IDEA): an event
typology for automated events data development. Journal of Peace Research 40(6) 733–745.
Brauer, J. (2007) Arms industries, arms trade, and developing countries. In Handbook of Defense Economics, Volume
2, edited by T. Sandler and K. Hartley. New York: Elsevier, 973–1015.
40 C.H. ANDERTON AND J.R. CARTER
Brecher, M. and Wilkenfeld, J. (2007) Codebook for ICB1 – System Level. International Crisis Behavior Project.
Brzoska, M. (1995) World military expenditures. In Handbook of Defense Economics, Volume 1, edited by K. Hartley
and T. Sandler. New York: Elsevier, 45–67.
Colaresi, M. and Thompson, W.R. (2002) Strategic rivalries, protracted conflict, and crisis escalation. Journal of
Peace Research 39(3) 263–287.
DeRouen, K.R. and Bercovitch, J. (2008) Enduring internal rivalries: a new framework for the study of civil war.
Journal of Peace Research 45(1) 55–74.
Eck, K. (2005) A Beginner’s Guide to Conflict Data: Finding and Using the Right Dataset. Uppsala, Sweden: Uppsala
Conflict Data Program.
Engene, J.O. (2007) Five decades of terrorism in Europe: the TWEED dataset. Journal of Peace Research 44(1)
109–121.
Fearon, J.D. and Laitin, D.D. (2003) Ethnicity, insurgency, and civil war. American Political Science Review 97(1)
Downloaded by [FNSP Fondation National des Sciences Politiques] at 12:34 16 September 2012
75–90.
Gibler, D.M. and Sarkees, M.R. (2004) Measuring alliances: the Correlates of War formal interstate alliance dataset,
1816–2000. Journal of Peace Research 41(2) 211–222.
Gleditsch, N.P., Wallensteen, P., Eriksson, M., Sollenberg, M. and Strand, H. (2002) Armed conflict 1946–2001: A
new dataset. Journal of Peace Research 39(5) 615–637.
Goldstein, J.S. (1992) A conflict-cooperation scale for WEIS events data. Journal of Conflict Resolution 36(2)
369–385.
Hämmerli, A., Gattiker, R. and Weyermann, R. (2006) Conflict and cooperation in an actors’ network of Chechnya
based on event data. Journal of Conflict Resolution 50(2) 159–175.
Harff, B. (2003) No lessons learned from the Holocaust? Assessing risks of genocide and political mass murder since
1955. American Political Science Review 97(1) 57–73.
Hensel, P.R. and Mitchell, S.M. (2009) The Issue Correlates of War Project homepage, downloaded 11/24/2009 from
www.paulhensel.org/icow.html.
Hewitt, J.J. (2003) Dyadic processes and international crises. Journal of Conflict Resolution 47(5) 669–692.
Houghton, B.K. (2008) Terrorism Knowledge Base: a eulogy. Perspectives on Terrorism 2(7) 18–19.
Jones, D.M., Bremer, S.A. and Singer, J.D. (1996) Militarized interstate disputes, 1816–1992: rationale, coding rules,
and empirical patterns. Conflict Management and Peace Science 15(2) 163–212.
Kauffmann, M. (Ed.) (2008) Building and Using Datasets on Armed Conflicts. Amsterdam: IOS Press.
King, G. and Lowe, W. (2003) An automated information extraction tool for international conflict data with
performance as good as human coders: a rare events evaluation design. International Organization 57(3)
617–642.
Klein, J.P., Goertz, G. and Diehl, P.F. (2006) The new rivalry dataset: procedures and patterns. Journal of Peace
Research 43(3) 331–348.
Kreutz, J. and Eck, K. (2005) UCDP non-state conflict codebook, version 1.0. Uppsala Conflict Data Program.
LaFree, G. and Dugan, L. (2007) Global Terrorism Database II, 1998--2004: Codebook and Data Documentation
[Computer file]. ICPSR22600-v1. College Park, MD: University of Maryland [producer], 2007. Ann Arbor, MI:
Inter-university Consortium for Political and Social Research [distributor], 2008–06–16.
Leeds, B.A., Ritter, J.M., Mitchell, S.M. and Long, A.G. (2002) Alliance treaty obligations and provisions, 1815–
1944. International Interactions 28(3) 237–260.
Lund, M.S. (1996) Early warning and preventive diplomacy. In Managing Global Chaos: Sources of and Responses
to International Conflict, edited by C.A. Crocker, F.O. Hampson and P. Aall. Washington: US Institute of Peace,
379–402.
Marshall, M.G. and Marshall, D.R. (2007) Coup d’etat events, 1960–2006 codebook. Center for Systemic Peace.
Marshall, M.G., Gurr, T.R. and Harff, B. (2009) PITF problem set codebook. Political Instability Task Force, avail-
able online at globalpolicy.gmu.edu/pitf/PITFProbSetCodebook2008.pdf.
Mickolus, E.F., Sandler, T., Murdock, J.M. and Flemming, P.A. (2003) International terrorism: attributes of terrorist
events (Iterate) 1968–2002 data codebook. Mimeo.
Minorities at Risk Project (2005) College Park, MD: Center for International Development and Conflict Management.
Retrieved from http://www.cidcm.umd.edu/mar/ on 2/12/07.
National Counterterrorism Center (2008) 2007 report on terrorism, 30 April 2008. Downloaded on 6/6/08 from
www.nctc.gov/.
Pfetsch, F.R. and Rohloff, C. (2000) KOSIMO: a databank on political conflict. Journal of Peace Research 37(3)
379–389.
Regan, P.M., Frank, R. and Aydin, A. (2009) Diplomatic interventions and civil wars: a new dataset. Journal of Peace
Research 46(1) 135–146.
Sambanis, N. (2004) What is civil war? Conceptual and empirical complexities of an operational definition. Journal
of Conflict Resolution 48(6) 814–858.
Sarkees, M.R. (2000) The Correlates of War data on war: an update to 1997. Conflict Management and Peace Science
18(1) 123–144.
Stockholm International Peace Research Institute (2008) SIPRI Yearbook 2008. New York: Oxford University Press.
Sundberg, R. (2009) UCDP non-state conflict codebook version 2.1–2009, available at http://www.pcr.uu.se/
research/UCDP/data_and_publications/datasets.htm (accessed 3 December 2009).
Wayman, F.W., Sarkees, M.R., and Singer, J.D. (2005) Inter-state, intra-state, extra-state, and non-state wars, 1816–
2004. Paper presented at the International Studies Association meeting, Honolulu, March 1–5, 2005.
CONFLICT DATASETS 41
Discrepancies across conflict datasets that track the same type of conflict arise from variations
in definitions, coding practices, assessments of historical materials, and the like. Numerous
types of discrepancies are possible including nature of the actors (e.g., the Indo-Hyderabad
1948 war is extra-state according to COW and interstate according to UCDP/PRIO), level of
violence (e.g., the 1982 Falklands/Malvinas conflict is a war in COW but a sub-war in UCDP/
PRIO), time period (First Kashmir war is 1948–49 in COW and 1948 in UCDP/PRIO), and
number of conflicts (e.g., UCDP/PRIO has three conflict IDs for the 1967 Middle East war,
Downloaded by [FNSP Fondation National des Sciences Politiques] at 12:34 16 September 2012
while COW has one ID for this war). Given such variations, assessing the degree of overlap
across conflict datasets requires that certain assumptions be made. The conflict data overlap
percentages presented in this article are based on the following assumptions.
1. Only overlapping time periods are used to assess the degree of overlap across datasets.
For example, UCDP/PRIO war data cover 1946–2008, while COW war data cover 1816–
1997. Comparisons across these data organizations were based on the common time
period 1946–1997.
2. Conflicts between the same actors that are temporally disconnected (i.e., have at least a
one calendar year break in the conflict) are treated as distinct conflicts even if a data orga-
nization applies the same conflict ID to the episodes of conflict. For example, UCDP/
PRIO applies the same conflict ID to the India/Pakistan wars of 1948, 1965, and 1971.
Our temporal disconnectedness assumption leads us to treat these as three wars.
3. A conflict case classified as a war by one data organization, but as a sub-war by another
data organization does not represent an overlapping case owing to the discrepant hostility
levels. For example, the 1969–70 conflict between Egypt and Israel is classified as a war
by COW but a sub-war by UCDP/PRIO.
4. A conflict case with a certain level of hostility that is classified as one actor type (e.g.,
interstate) by one data organization, but as another actor type (e.g., extra-state) by another
data organization does not represent an overlapping case owing to the discrepant actor
types. For example, the Indo-Hyderabad war of 1948 noted earlier is extra-state according
to COW and interstate according to UCDP/PRIO.
5. A conflict case with a certain actor type and level of hostility but with different time peri-
ods across data organizations represents an overlap as long as their time frames are not
separated by more than one calendar year. For example, COW’s Cambodia-Khmer Rouge
intrastate war of 1993–97 is classified as not matching with UCDP/PRIO because the
latter has a war level of violence only for 1989, which is greater than the one year gap.
6. A conflict case with a certain actor type and level of hostility that is treated as multiple
conflicts by one data organization but a single conflict by another data organization repre-
sents an overlap. For example, UCDP/PRIO’s three conflict IDs for the 1967 Middle East
war are each treated as matched by COW’s one conflict ID for this war; moreover,
COW’s 1967 Middle East war is treated as matched by UCDP/PRIO.
7. For UCDP/PRIO, the intensity index (as distinct from the cumulative intensity index) was
used to determine whether war or sub-war pertained in a given year.
8. For PITF, the revolutionary and ethnic war datasets were concatenated with redundant
cases culled.
9. The Fatal MIDs series constructed for this article deleted all cases where the hostility level
for at least one player was 5 (indicating war) or all players had a fatality index of 0 (no
fatalities), –9 (missing fatality data), or 6 (war level of fatalities).
Table A1 shows how we compared interstate war cases between COW and UCDP/PRIO.
COW interstate wars in rows 1–16 of the first column of the table have matches in the UCDP/
42 C.H. ANDERTON AND J.R. CARTER
PRIO interstate wars column. Note that the war counts differ between the two columns even
when matches occur because the Six Day War (COW row 9) is counted as three wars in
UCDP/PRIO and the Yom Kippur War (COW row 12) is treated as two wars in UCDP/PRIO.
The wars shown in bold at the bottom of the table represent cases from each column’s dataset
that do not have a match in the other dataset. For example, the Israel–Egypt 1969–70 war in
COW row 17 does not have a match in the UCDP/PRIO column. Likewise, the China–Taiwan
wars in UCDP/PRIO (cases 21–23) do not have matches in COW. At the bottom of the table
are the summaries of the degrees of overlap between COW and UCDP/PRIO interstate war
cases. As shown, 16 of 23 (70%) COW interstate wars are matched in UCDP/PRIO, and 19 of
Downloaded by [FNSP Fondation National des Sciences Politiques] at 12:34 16 September 2012
25 (76%) of UCDP/PRIO interstate wars are matched in COW. These are the overlap summa-
ries presented earlier in the article for COW and UCDP/PRIO interstate wars. Similar methods
were used to arrive at the overlap counts for the other types of conflicts and data organizations
indicated in the article.