Professional Documents
Culture Documents
in a
RISING
Africa
Poverty
in a
RISING
Africa
K at h l e e n B e e g l e
Lu c C h r i s t i a e n s e n
A n d r e w Da b a l e n
Isis Gaddis
© 2016 International Bank for Reconstruction and Development / The World Bank
1818 H Street NW, Washington DC 20433
Telephone: 202-473-1000; Internet: www.worldbank.org
1 2 3 4 19 18 17 16
This work is a product of the staff of The World Bank with external contributions. The findings, interpretations,
and conclusions expressed in this work do not necessarily reflect the views of The World Bank, its Board of Execu-
tive Directors, or the governments they represent. The World Bank does not guarantee the accuracy of the data
included in this work. The boundaries, colors, denominations, and other information shown on any map in this
work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the
endorsement or acceptance of such boundaries.
Nothing herein shall constitute or be considered to be a limitation upon or waiver of the privileges and immuni-
ties of The World Bank, all of which are specifically reserved.
This work is available under the Creative Commons Attribution 3.0 IGO license (CC BY 3.0 IGO)
http://creativecommons.org/licenses/by/3.0/igo. Under the Creative Commons Attribution license, you are free to
copy, distribute, transmit, and adapt this work, including for commercial purposes, under the following conditions:
Attribution—Please cite the work as follows: Beegle, Kathleen, Luc Christiaensen, Andrew Dabalen, and Isis Gad-
dis. 2016. Poverty in a Rising Africa. Washington, DC: World Bank. doi:10.1596/978-1-4648-0723-7. License:
Creative Commons Attribution CC BY 3.0 IGO
Translations—If you create a translation of this work, please add the following disclaimer along with the attribu-
tion: This translation was not created by The World Bank and should not be considered an official World Bank
translation. The World Bank shall not be liable for any content or error in this translation.
Adaptations—If you create an adaptation of this work, please add the following disclaimer along with the attribu-
tion: This is an adaptation of an original work by The World Bank. Views and opinions expressed in the adap-
tation are the sole responsibility of the author or authors of the adaptation and are not endorsed by The World
Bank.
Third-party content—The World Bank does not necessarily own each component of the content contained within
the work. The World Bank therefore does not warrant that the use of any third-party-owned individual component
or part contained in the work will not infringe on the rights of those third parties. The risk of claims resulting
from such infringement rests solely with you. If you wish to re-use a component of the work, it is your responsibil-
ity to determine whether permission is needed for that re-use and to obtain permission from the copyright owner.
Examples of components can include, but are not limited to, tables, figures, or images.
A ll queries on rig hts a nd licenses shou ld be add ressed to t he P ublish i ng a nd K nowledge Divi-
sion, The World Bank, 1818 H Street N W, Washington, DC 20433, USA; fax: 202-522-2625; e-mail:
pubrights@worldbank.org.
ISBN: 978-1-4648-0723-7
eISBN: 978-1-4648-0724-4
DOI: 10.1596/978-1-4648-0723-7
Foreword . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
Acknowledgments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvii
Key Messages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Overview ........................................................... 3
Assessing the Data Landscape . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Improving Data on Poverty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Revisiting Poverty Trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Profiling the Poor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Taking a Nonmonetary Perspective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Measuring Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1. The State of Data for Measuring Poverty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Types of Data for Measuring Monetary Poverty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
The Political Economy of Data Production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
Reappraising the Information Base on Poverty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Concluding Remarks and Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
v
vi Contents
Boxes
1.1 Sources outside the national statistical system provide valuable information on
well-being . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.2 How did poverty change in Guinea and Mali? Lack of comparable data makes
it difficult to know . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.3 Many kinds of data in Africa are unreliable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
1.4 Can donors improve the capacity of national statistics offices? Lessons learned
from MECOVI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
1.5 What is the threshold for being poor? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.1 Adjusting the data for Nigeria has a huge effect on estimates of poverty reduction . . . 59
2.2 How do spikes in food prices affect the measurement of poverty? . . . . . . . . . . . . . . . . 64
2.3 Can wealth indexes be used to measure changes in poverty? . . . . . . . . . . . . . . . . . . . 67
3.1 How useful are subjective data in monitoring poverty? . . . . . . . . . . . . . . . . . . . . . . . 85
3.2 Tracking adult literacy with data remains challenging . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.3 What happens to Africans who flee their homes?. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
3.4 Demographic and Health Surveys make it possible to measure multidimensional
poverty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
3.5 What is the multidimensional poverty index (MPI)? . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.1 A primer on the Gini index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
4.2 Can the Gini index be estimated without a survey? . . . . . . . . . . . . . . . . . . . . . . . . . . 123
4.3 Are resources within households shared equally? Evidence from Senegal . . . . . . . . . 131
Contents vii
Figures
O.1 Good governance and statistical capacity go together . . . . . . . . . . . . . . . . . . . . . . . . . . 7
O.2 Adjusting for comparability and quality changes the level of and trends in poverty . . . 8
O.3 Other estimates also suggest that poverty in Africa declined slightly faster and is
slightly lower . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
O.4 Fragility is associated with significantly slower poverty reduction . . . . . . . . . . . . . . . . 10
O.5 Acceptance of domestic violence is twice as high in Africa as in other developing
regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
O.6 Residents in resource-rich countries suffer a penalty in their human development . . . 14
O.7 Declining inequality is often associated with declining poverty . . . . . . . . . . . . . . . . . . 16
I.1 Poverty reduction in Africa lags other regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.1 All regions have increased the number of household surveys they conduct . . . . . . . . . 27
1.2 Africa conducts more nonconsumption surveys than consumption surveys . . . . . . . . 29
1.3 Many African countries lack surveys with which to gauge changes in poverty . . . . . . 29
1.4 Comparability of consumption surveys has improved, but it remains a major
problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.5 Different survey designs can result in very different consumption estimates . . . . . . . . 35
1.6 Data errors may account for some of the reported change in consumption . . . . . . . . . 36
1.7 The weights used to construct consumer price indexes in Africa are outdated . . . . . . 37
1.8 Both the prices and weights used to construct consumer price indexes in Africa
reflect a strong urban bias . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
1.9 Adoption of the 2011 purchasing power parity values increased GDP per capita
figures across Africa . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
1.10 Rebasing increased GDP values in many African countries . . . . . . . . . . . . . . . . . . . . 42
1.11 Good governance and statistical capacity go together . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.1 Adjusting for comparability and quality changes the level, depth, and severity of
poverty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.2 Analysis based only on comparable surveys suggests that poverty reduction in
Africa was faster than previously thought . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.3 Survey-to-survey imputation and evidence from comparable surveys provide
similar estimates of poverty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.4 Survey-to-survey imputations suggest that poverty in Africa is lower than household
survey data indicate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
B2.2.1 Food inflation does not always exceed overall inflation . . . . . . . . . . . . . . . . . . . . . . . . 64
2.5 Correcting for CPI bias suggests that poverty reduction is underestimated . . . . . . . . . 65
2.6 Fragility is associated with significantly slower poverty reduction . . . . . . . . . . . . . . . . 69
2.7 Urban poverty in Southern and West Africa fell by almost half between 1996
and 2012 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
2.8 Across Africa, more and more households are headed by women . . . . . . . . . . . . . . . . 71
2.9 Estimates of movements into and out of poverty vary widely across Africa . . . . . . . . . 73
2.10 The share of poor people in Africa who fall into poverty is about the same as the
share of poor people who move out of poverty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
2.11 Africa’s poor are clustered around the poverty line . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.1 Africa’s literacy rate is the lowest in the world . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.2 Literacy rates are lowest in West Africa . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.3 The gender gap in literacy varies widely across Africa . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.4 Illiteracy is higher among poorer people, older people, rural dwellers, and people
in resource-rich and landlocked countries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
viii Contents
Maps
O.1 Lack of comparable surveys in Africa makes it difficult to measure poverty trends . . . . 5
O.2 The number of violent events against civilians is increasing, especially in Central
Africa and the Horn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
O.3 Inequality in Africa shows a geographical pattern . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.1 More than half of African countries completed a consumption survey between
2011 and early 2015. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.2 Lack of comparable surveys in Africa makes it difficult to measure poverty trends . . . 33
3.1 HIV prevalence remains very high in Southern Africa . . . . . . . . . . . . . . . . . . . . . . . . 93
3.2 The number of violent events against civilians is increasing, especially in
Central Africa and the Horn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
3.3 Multiple deprivation is substantial in the Western Sahel and Africa’s populous
countries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.1 Inequality in Africa shows a geographical pattern . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Tables
I.1 Classification of countries in Africa . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.1 Africa lags in the number of comparable surveys per country, conducted between
1990 and 2012 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.2 Only a few country characteristics are correlated with the number and share of
comparable and open consumption surveys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.1 Addressing quality and comparability reduces the surveys available for poverty
monitoring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.2 Many country-level factors affect asset ownership of the near-poor . . . . . . . . . . . . . . 68
4.1 Inequality in Africa, 1993–2008 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
4.2 Likelihood of remaining in one’s father’s sector in selected African countries . . . . . . 135
4.3 Gross and net occupational intergenerational mobility out of farming in selected
African countries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
Foreword
A
fter two decades of unprecedented evidence on core measures of poverty and
e c o no m i c g row t h , how mu c h inequality, along both monetary and non-
have the lives of African families monetary dimensions. The findings are both
improved? The latest estimates from the encouraging and sobering.
World Bank suggest that the share of the Considerable progress has been made in
African population in extreme poverty did terms of data for measuring the well-being
decline—from 57 percent in 1990 to 43 per- of the population. The availability and qual-
cent in 2012. At the same time, however, ity of household survey data in Africa has
Africa’s population continued to expand improved. At the same time, not all coun-
rapidly. As a result, the number of people tries have multiple and comparable surveys to
living in extreme poverty still increased by track poverty trends. Reevaluating the trends
more than 100 million. These are stagger- in poverty, taking into account these data
ing numbers. Further, it is projected that the concerns, suggests that poverty in Africa may
world’s extreme poor will be increasingly be lower than what current estimates suggest.
concentrated in Africa. In addition, Africa’s population saw progress
With the adoption of the Sustainable in nonmonetary dimensions of well-being,
Development Goals, including the eradica- particularly in terms of health indicators
tion of extreme poverty by 2030, successful and freedom from violence. While the avail-
implementation of the post-2015 develop- able data do not suggest a systematic increase
ment agenda will require a solid understand- in inequality within countries in Africa, the
ing of poverty and inequality in the region, number of extremely wealthy Africans is
across countries and population groups, and increasing. Overall, notwithstanding these
in different dimensions. broad trends, caution remains as data chal-
Poverty in a Rising Africa is the first of lenges multiply when attempting to measure
two sequential reports aimed at better under- inequality.
standing progress in poverty reduction in While these findings on progress are
Africa and articulating a policy agenda to encouraging, major poverty challenges
accelerate it. This first report has a modest, remain, especially in light of the region’s
but important, objective: to document the rapid population growth. Consider this: even
data challenges and systematically review the under the most optimistic scenario, there
xi
xii Foreword
are still many more Africans living in pov- To maintain and accelerate the momen-
erty (more than 330 million in 2012) than in tum of progress of the past two decades, con-
1990 (about 280 million). Despite improve- certed and collective efforts are also needed
ments in primary school enrollment rates, to further improve the quality and timeliness
the poor quality of learning outcomes, as of poverty statistics in the region. Domestic
evidenced by the fact that two in five adults political support for statistics can be the most
are illiterate, highlights the urgency of poli- important factor in the quest for better data.
cies to improve educational outcomes, par- Development partners and the international
ticularly for girls. Perpetuation of inequality, community also have an important role to
in the absence of intergenerational mobility play in terms of promoting regional coop-
in education, further highlights the long-run eration, new financing models, open access
consequences of failure to do so. Not surpris- policies, and clearer international standards.
ingly, poverty reduction has been slowest in This volume is intended to contribute toward
fragile states. This trend is compounded by improving the scope, quality, and relevance
the fact that violence against civilians is once of poverty statistics. Because, in the fight
again on the rise, after a decade of relative against poverty in Africa, (good) data will
peace. There is also the paradoxical fact that make a difference. Better data will make for
citizens in resource-rich countries are expe- better decisions and better lives.
riencing systematically lower outcomes in
all human welfare indicators controlling for Makhtar Diop
their income level. Clearly, policies matter Vice President, Africa Region
beyond resource availability. World Bank
Acknowledgments
T
his volume is part of the African Isabel Almeida, Prospere Backiny-Yetna,
Regional Studies Program, an ini- Yele Batana, Abdoullahi Beidou, Paolo Brun-
tiative of the Africa Region Vice ori, Hai-Anh Dang, Johannes Hoogeveen,
Presidency at the World Bank. This series La-Bhus Jirasavetakul, Christoph Lakner,
of studies aims to combine high levels of Jean-François Maystadt, Annamaria Mi -
analytical rigor and policy relevance, and lazzo, Flaviana Palmisano, Vito Peragine,
to apply them to various topics important Dominique van de Walle, Philip Verwimp,
for the social and economic development and Eleni Yitbarek.
of Sub-Saharan Africa. Quality control and The team benefited from the valuable
oversight are provided by the Office of the advice and feedback of Carlos Batarda,
Chief Economist for the Africa Region. Haroon Bhorat, Laurence Chandy, Pablo
This report was prepared by a core team Fajnzylber, Jed Friedman, John Gibson, Jéré-
led by Kathleen Beegle, Luc Christiaensen, mie Gignoux, Ruth Hill, José Antonio Mejía-
Andrew Dabalen, and Isis Gaddis. It would Guerra, Berk Ozler, Martin Ravallion,
not have been possible without the relentless Raul Santaeulalia-Llopis, and Frederick Solt.
efforts and inputs of Nga Thi Viet Nguyen Valentina Stoevska and colleagues from the
and Shinya Takamatsu (chapters 1 and 2), ILO provided valuable data.
Umberto Cattaneo and Agnes Said (chapter Stephan Klasen, Peter Lanjouw, Jacques
3), and Camila Galindo-Pardo (chapters 3 Morisset, and one anonymous reviewer pro-
and 4). Rose Mungai coordinated the vided detailed and careful peer review
massive effort to harmonize data files; Wei comments.
Guo, Yunsun Li, and Ayago Esmubancha The World Bank’s Publishing and Knowl-
Wambile provided valuable research assis- edge team coordinated the design, type-
tance. Administrative support by Keneth setting, printing, and dissemination of the
Omondi and Joyce Rompas is most grate- report. Special thanks to Janice Tuten,
fully acknowledged. Stephen McGroarty, Nancy Lammers, Abdia
Francisco H. G. Ferreira provided gen- Mohamed, and Deborah Appel-Barker.
eral direction and guidance to the team. Robert Zimmermann and Barbara Karni
Additional contributions were made by edited the report.
xiii
About the
Authors and Contributors
Kathleen Beegle is a lead economist in the Luc Christiaensen is a lead agriculture econ-
World Bank’s Africa Region. Based in Accra, omist in the World Bank’s Jobs Group and
she coordinates country programs in Ghana, an honorary research fellow at the Maas-
Liberia, and Sierra Leone in the areas of edu- tricht School of Management. He has writ-
cation, health, poverty, social protection, ten extensively on poverty, secondary towns,
gender, and jobs. Her broader area of work and structural transformation in Africa and
includes poverty, labor, economic shocks, East Asia. He is also leading the “Agriculture
and methodological studies on household in Africa: Telling Facts from Myths” project.
survey data collection. She was deputy direc- He was a core member of the team that pro-
tor of the World Development Report 2013: duced the World Development Report 2008:
Jobs. She holds a PhD in economics from Agriculture for Development. He holds a
Michigan State University. PhD in agricultural economics from Cornell
University.
Umberto Cattaneo is a research assistant at
the World Bank and a doctoral fellow at the Andrew Dabalen is a lead economist in the
European Center for Advanced Research in World Bank’s Poverty and Equity Global
Economics and Statistics at the Université Practice. His work focuses on policy analysis
Libre de Bruxelles. His research interests and research in development issues, such as
include development economics, civil war, poverty and social impact analysis, inequal-
poverty analysis, applied microeconomet- ity of opportunity, program evaluation, risk
rics, and agricultural and environmental and vulnerability, labor markets, conflict,
economics. He recently completed a study and welfare outcomes. He has worked in the
on the impact of civil war on subjective and World Bank’s Africa and Europe and Cen-
objective poverty in rural Burundi. He holds tral Asia Regions on poverty analysis, social
a master’s degree in development econom- safety nets, labor markets, and education
ics from the School of Oriental and African reforms. He has coauthored regional reports
Studies of the University of London and a on equality of opportunity for children in
master’s degree in economics and finance Africa and vulnerability and resilience in the
from the University of Genova. Sahel and led poverty assessments for several
xv
xvi About the Authors And Contributors
countries, including Albania, Burkina Faso, Nga Thi Viet Nguyen is an economist in the
Côte d’Ivoire, Kosovo, Niger, Nigeria, and World Bank’s Poverty and Equity Global
Serbia. He has published scholarly articles Practice, where her work involves poverty
and working papers on poverty measure- measurement and analysis, policy evaluation,
ment, conflict and welfare outcomes, and and the study of labor markets and human
wage inequality. He holds a PhD in agricul- development. She was part of the team that
tural and resource economics from the Uni- produced the 2013 report Opening Doors:
versity of California-Berkeley. Gender Equality and Development in the
Middle East and North Africa. In Africa she
Isis Gaddis is an economist in the World investigated the impact of Nigeria’s import
Bank’s Gender Group. She previously served bans on poverty, the role of social safety net
as a poverty economist for Tanzania based programs in rural poverty in Malawi, and
in Dar es Salaam. Her main research inter- the contribution of labor income to poverty
est is empirical microeconomics, with a focus reduction in five African countries and con-
on the measurement and analysis of poverty tributed to various poverty assessments. She
and inequality, gender, labor economics, and holds a master’s degree in public policy from
public service delivery. She holds a PhD in Harvard University.
economics from the University of Göttingen,
where she was a member of the development Agnes Said is a lawyer who has been work-
economics research group from 2006 to ing with the World Bank since 2009. Her
2012. work focuses on public sector governance and
social protection. She is part of the manage-
Camila Galindo-Pardo worked as a research ment team of a multidonor trust fund for the
analyst in the Chief Economist’s Office of the Middle East and North Africa Region that is
Africa Region of the World Bank, where she striving to strengthen governance and increase
studied the link between sectoral economic social and economic inclusion in the region.
growth and poverty, income inequality and Her work on justice and fundamental rights
extreme wealth, gender based-violence, and has been published by the European Commis-
the prevalence of net buyers of staple foods sion and the European Parliament. She holds
among African households. She is a PhD a master of laws degree from the University
student in economics at the University of of Gothenburg and a master’s degree in inter-
Maryland-College Park. national relations and international econom-
ics from the School of Advanced International
Rose Mu nga i is a senior econom ist / Studies of the Johns Hopkins University.
statistician with the Africa Region of the
World Bank and the region’s focal point on Shinya Takamatsu is a consultant in the
poverty data. She has more than 15 years of Poverty and Equity Global Practice of the
experience designing household surveys and Africa Region of the World Bank, where he
measuring and analyzing poverty. For sev- is a core member of the region’s statistical
eral years she led production of the Bank’s development team. He has published several
annual Africa Development Indicators working papers on poverty imputations and
report. Before joining the World Bank, she survey methodology and conducted research
worked as a senior economist/statistician on the educational spillover effect of a condi-
at the Kenya National Bureau of Statistics, tional cash transfer program and the poverty
where her core role was measuring pov- impacts of food price crises. He holds a PhD
erty. She holds a master’s degree in devel- in agricultural and resource economics, with
opment economics from the University of a minor in statistics, from the University of
Manchester. Minnesota.
Abbreviations
xvii
Key Messages
1
Overview
P
erceptions of Africa changed dramati- and projections that the world’s poor will
cally over the past 20 years. Viewed be increasingly concentrated in Africa even
as a continent of wars, famines, and if the average 1995–2014 growth rates are
entrenched poverty in the late 1990s, there maintained suggest the need to focus the
is now a focus on “Africa rising” and an global poverty agenda on Africa.
“African 21st century.”1 At 4.5 percent a This report is the first of a two-part vol-
year, average economic growth was remark- ume on poverty in Africa. This study docu-
ably robust, especially when contrasted with ments the data challenges and revisits the
the continuous decline during the 1970s and core broad facts about poverty in Africa; the
1980s. second report will explore ways to accelerate
Substantial improvements in well-being its reduction.
should have accompanied this expansion. The report takes a broad, multidimen-
Whether or not they did remains unclear sional view of poverty, assessing progress
given the poor quality of the data (Devara- over the past two decades along both mon-
jan 2013; Jerven 2013), the nature of the etary and nonmonetary dimensions. The
growth process (especially the role of natural dearth of comparable, good-quality house-
resources) (de la Briere and others 2015), the hold consumption surveys makes assessing
emergence of extreme wealth (Oxfam 2015), monetary poverty especially challenging.
the heterogeneity of the region, and persis- The report scrutinizes the data used to
tent population growth of 2.7 percent a year assess monetary poverty in the region and
(Canning, Raja, and Yazbeck 2015). explores how adjustments for data issues
Expectations are also rising. All develop- affect poverty trends. 2
ing regions except Africa have reached the At the same time, the remarkable expan-
Millennium Development Goal (MDG) of sion of standardized household surveys on
halving poverty between 1990 and 2015 nonmonetary dimensions of well-being,
(UN 2015). Attention will now shift to the including opinions and perceptions, opens
set of new global development goals (the Sus- up new opportunities. The report examines
tainable Development Goals [SDGs]), which progress in education and health, the extent
include the ambitious target of eradicating to which people are free from violence and
poverty worldwide by 2030. The poten- able to shape their lives, and the joint occur-
tial for a slowdown in economic growth rence of various types of deprivation. It also
3
4 POVERTY IN A RISING AFRICA
reviews the distributional aspects of poverty, ranks second to South Asia in terms of the
by studying various dimensions of inequality. number of national household surveys per
To shed light on Africa’s diversity, the country, according to the International
report examines differences in performance Household Survey Network catalog. The
across countries, by location, and by gen- region has an average of 24 surveys per coun-
der. Countries are characterized along four try conducted between 1990 and 2012—
dimensions that have been shown to affect more than the developing world average
growth and poverty: resource richness, fra- of about 22. This expansion was confined
gility, landlockedness (to capture geographic almost entirely to surveys that do not collect
openness and potential for trade), and income consumption data, however.
status (low, lower-middle, upper-middle, and The increase in household consump-
high income). tion surveys, which are the building blocks
for measuring poverty and inequality, was
sluggish, though coverage increased. Since
Assessing the Data Landscape 2009 only 2 countries did not conduct a
According to World Bank estimates from single consumption survey over the past
household surveys, the share of people liv- decade (down from 10 in 1990–99). The
ing on less than $1.90 a day (in 2011 inter- number of countries that either did not
national purchasing power parity [PPP]) conduct a consumption survey or do not
fell from 57 percent in 1990 to 43 percent allow access to the microdata declined
in 2012, while the number of poor still from 18 in 1990–99 to 4 in 2003–12; and
increased by more than 100 million (from the number of countries with at least two
288 to 389 million). consumption surveys increased, from 13
These estimates are based on consumption in 1990–99 to 25 in 2003–12. Many frag-
surveys in a subsample of countries cover- ile states—namely, Chad, the Democratic
ing between one-half and two-thirds of the Republic of Congo, Sierra Leone, and
region’s population. Poverty rates for the rest Togo—were part of this new wave of sur-
of the countries are imputed from surveys veys. Nonetheless, fragile states still tend to
that are often several years old using gross be the most data deprived.
domestic product (GDP) trends, raising ques- The lack of consumption surveys and
tions about the accuracy of the estimates. On accessibility to the underlying data are obvi-
average only 3.8 consumption surveys per ous impediments to monitoring poverty. But
country were conducted in Africa between the problems do not end there. Even when
1990 and 2012, or one every 6.1 years. In available, surveys are often not comparable
the rest of the world, one consumption sur- with other surveys within the country or are
vey was conducted every 2.8 years. The aver- of poor quality (including as a result of misre-
age also masks quite uneven coverage across porting and deficiencies in data processing).
countries. For five countries that together rep- Consequently, countries that appear to be
resent 5 percent of the African population, no data rich (or have multiple surveys) can still
data to measure poverty are available (either be unable to track poverty over time (exam-
because no household surveys were con- ples include Guinea and Mali, with four sur-
ducted or because the data that were collected veys each that are not comparable).
are not accessible, or, as in the case of one At a country level, lack of comparability
survey for Zimbabwe, were collected during between survey rounds and questions about
a period of hyperinflation and unsuitable for quality issues often prompt intense technical
poverty measurement). As of 2012, only 27 of debates about methodological choices and,
48 countries had conducted at least two com- national poverty estimates within countries
parable surveys since 1990 to track poverty. (see World Bank 2012 for Niger; World Bank
To be sure, the number of household sur- 2013 for Burkina Faso; World Bank 2015b
veys in Africa has been rising. Africa now for Tanzania). But much regional work in
OVERVIEW 5
MAP O.1 Lack of comparable surveys in Africa makes it difficult to measure poverty trends
Cabo Mauritania
Verde
Mali Niger
Sudan Eritrea
Senegal Chad
The Gambia
Guinea-Bissau Burkina Faso
Guinea Benin
Nigeria
Côte Ethiopia
Sierra Leone d’Ivoire Ghana Central African South Sudan
Republic
Liberia Cameroon Somalia
Togo
Equatorial Guinea Uganda
São Tomé and Príncipe Rep. of Kenya
Gabon Congo
Rwanda
Dem. Rep. of Burundi
Congo Tanzania
Seychelles
Comoros
Number of comparable surveys conducted, Angola
1990–2012 Malawi
0 or 1 survey (9 countries) Zambia
No comparable surveys (12 countries)
2 comparable surveys (17 countries) Zimbabwe Mauritius
Mozambique
More than 2 comparable surveys (10 countries) Namibia
Botswana Madagascar
Swaziland
South Lesotho
Africa
IBRD 41865
SEPTEMBER 2015
Source: World Bank data.
Africa and elsewhere disregards these impor- conducted only 1.6 comparable surveys in the
tant differences, relying on databases such 23 years between 1990 and 2012.
as the World Bank’s PovcalNet that has not The challenge of maintaining compara-
consistently vetted surveys on the basis of bility across surveys is not unique to Africa
comparability or quality. or to tracking poverty (see, for example,
If surveys that are not nationally repre- UNESCO 2015 for data challenges in track-
sentative (covering only urban or rural areas, ing adult literacy). However, in Africa lack
for example), that were not conducted at of comparability exacerbates the constraints
similar times of the year (in order to control imposed by the already limited availability of
for seasonality in consumption patterns), consumption surveys. It becomes especially
and that collected consumption data using problematic when the challenges concern
different instruments or reporting periods populous countries, such as Nigeria. Only 27
are dropped, the typical African country countries (out of 48) conducted two or more
6 POVERTY IN A RISING AFRICA
90
Mauritius
80 Rwanda
Malawi Mozambique
Nigeria Tanzania Senegal South Africa
São Tomé and Príncipe Burkina Faso Lesotho
70 The Gambia Niger
Cabo Verde
Statistical capacity indicator
Guinea-Bissau Gabon
40 Comoros
Equatorial Guinea
30 Eritrea
20 Somalia
0 10 20 30 40 50 60 70 80
Safety and rule of law score
political arrangements thus favor less (or less by failure to adhere to methodological and
autonomous) funding for statistics because it operational standards. While this problem
represents one way to exercise influence over partly reflects the lack of broader political
statistical agencies. In some countries donor support domestically, regional cooperation
financing has replaced domestic financing, and peer learning, as well as clear interna-
but the interests of donors are not always tional standards, could help improve techni-
aligned with the interests of governments. cal quality and consistency. The Program for
This problem highlights the need for alterna- the Improvement of Surveys and the Measure-
tive financing models, including cofinancing ment of Living Conditions in Latin America
arrangements, preferably under a coordi- and the Caribbean (known by its acronym in
nated regional umbrella and with adequate Spanish, MECOVI) provides a compelling
incentives for quality improvements. model for achieving better poverty data.
Politics and funding are not the only rea-
sons statistics are inadequate. The evidence
presented here suggests that better outcomes Revisiting Poverty Trends
were possible even with the set of surveys that
were conducted. African countries collected Various technical approaches can be applied
on average 3.8 consumption surveys in the to address some of the data shortcomings
past two decades, but many of them could in tracking regional poverty trends. They
not be used to track poverty reliably because include limiting the sample to comparable
of comparability and quality concerns caused surveys of good quality, using trends in other
8 POVERTY IN A RISING AFRICA
nonconsumption data rather than GDP to (37 percent instead of 43 percent). The series
impute missing poverty estimates, and gaug- of comparable and good-quality surveys only
ing inflation using alternative econometric excludes some of the surveys from Burkina
techniques. Faso, Mozambique, Tanzania, and Zambia
Taking these steps affects the view of how and replaces the poverty estimates of the
poverty has evolved in Africa. The estimate two comparable but poorer-quality surveys
from PovcalNet in figure O.2 shows the of Nigeria (Nigeria Living Standards Sur-
now-familiar trend in poverty from surveys veys 2003/04 and 2009/10) with the estimate
in the World Bank PovcalNet database. It from the General Household Survey Panel
provides the benchmark. These estimates are 2010/11, which has been deemed of good
population-weighted poverty rates for the 48 quality. Poverty gap and severity measures
countries, of which 43 countries have one or follow similar trajectories, after correction
more surveys.4 For years for which there were for comparability and quality.
no surveys, poverty was estimated by impu- In the series depicted based on the subset
tation using GDP growth rates. of comparable and good-quality surveys, the
The estimate based on only comparable information base for Nigeria, which encom-
surveys shows the trends when only com- passes almost 20 percent of the population
parable surveys are used and the same GDP of Africa, shifts. The 2003/04 and 2009/10
imputation method is applied. It largely mir- surveys showed no change in poverty in Nige-
rors the PovcalNet estimate. In contrast, ria. The poverty rate indicated by the alter-
when in addition to controlling for compara- native survey for 2010/11 (26 percent) is half
bility, quality is taken into account, the 2012 the estimate obtained from the lower-quality
estimate of poverty in Africa is 6 percentage survey (53 percent) in 2009/10. Given that
points lower than the PovcalNet estimate only one survey is retained, the estimated
poverty trend for Nigeria also relies more
FIGURE O.2 Adjusting for comparability and quality changes the on the GDP growth pattern (which was high
level of and trends in poverty during the 2000s) as well as a lower poverty
rate for 2010/11. Reesti mating the poverty
65 rate with only comparable surveys of good
quality but without Nigeria indicates that
Nigeria accounted for a large fraction of the
60
additional decline observed using the cor-
rected series (the red line). Without Nigeria,
Poverty rate (percent)
55
the corrected series declines from 55 percent
to 40 percent (a 15 percentage point drop),
50 compared with 57 percent to 43 percent (a 14
percentage point drop) in PovcalNet. Confi-
45 dence in the revised regional series depends
significantly on how reliable the trends in
40 Nigeria’s poverty obtained using the good-
quality survey and greater dependence on
GDP imputation are considered.
35
Consumption data gaps can also be filled
1990 1993 1996 1999 2002 2005 2008 2010 2012
by applying survey-to-survey (S2S) imputa-
PovcalNet tion techniques to nonconsumption survey
Comparable surveys only data. In this method, at least one survey with
Comparable and good-quality surveys only
Comparable and good-quality surveys only without Nigeria consumption and basic household character-
istics is combined with nonconsumption sur-
Sources: World Bank Africa Poverty database and PovcalNet. veys with the same basic characteristics for
Note: Poverty is defined as living on less than $1.90 a day (2011 international purchasing power parity). different years. Consumption for the years
OVERVIEW 9
with no survey is then estimated based on FIGURE O.3 Other estimates also suggest that poverty in Africa
the evolution of the nonconsumption house- declined slightly faster and is slightly lower
hold characteristics as well as the relation
between those characteristics and consump- 65
tion, as estimated from the consumption
survey. Where they have been tested, these
prediction techniques perform mostly well
This exercise also underscores the need gap in performance is 12 percentage points
for more reliable and comparable consump- in favor of nonfragile countries. Conditional
tion data to help benchmark and track prog- on the three other country traits, the differ-
ress toward eradicating poverty by 2030, ence in poverty reduction between fragile
as envisioned under the SDGs. More gener- and nonfragile countries rises to 15 percent-
ally, it counsels against overinterpreting the age points (figure O.4). Middle-income coun-
accuracy conveyed by point estimates of tries as a group did not achieve faster poverty
poverty—or other region- or countrywide reduction than low-income countries, and
statistics of well-being. These estimates pro- being resource rich was associated with pov-
vide only an order of magnitude of levels and erty reduction that was 13 percentage points
changes, albeit one that becomes more pre- greater than in non-resource-rich countries
cise the more comparable and reliable is the after controlling for other traits. The main
underlying database. driver for the difference in poverty reduction
in resource-rich and resource-poor countries,
however, is corrections to the Nigeria data.
Profiling the Poor More surprisingly, once resource richness,
What distinguishes countries that have suc- fragility, and income status are controlled for,
ceeded in reducing poverty from those that landlocked countries did not reduce poverty
have failed? What are the effects of income less than coastal economies (the effect is not
status, resource richness, landlockedness, statistically significant and the point estimate
and fragility? is even negative). This finding contradicts the
Not surprisingly, fragility is most detri- common notion that landlocked countries
mental to poverty reduction. Between 1996 perform worse than coastal countries because
and 2012, poverty decreased in fragile states transport costs impede trade and lower com-
(from 65 percent to 53 percent), but the decline petitiveness (Bloom and Sachs 1998).
was much smaller than in nonfragile econo- Although Africa is urbanizing rapidly, in
mies (from 56 percent to 32 percent). The the majority of countries, 65–70 percent of
the population resides in rural areas (Can-
ning, Raja, and Yazbeck 2015). Across coun-
FIGURE O.4 Fragility is associated with significantly slower tries rural residents have higher poverty rates
poverty reduction
(46 percent in rural areas in 2012 versus
18 percent in urban areas, using corrected
data for all countries). But the gap between
–1.1 Middle income the poverty rate in rural and urban areas
declined (from 35 percentage points in 1996
to 28 percentage points in 2012). Among the
–7.1 Landlocked four geographic regions, only urban areas in
West Africa halved poverty. Poverty among
rural populations in West and Southern
–12.6*** Resource rich Africa declined about 40 percent.
Africa is distinguished by a large and
rising share of female-headed households.
Fragile 15.1***
Such households represent 26 percent of all
households and 20 percent of all people in
–15 –10 –5 0 5 10 15 20 Africa. Southern Africa has the highest rate
Change in poverty rate (percentage points) of female-headed households (43 percent).
compared to alternative category West Africa exhibits the lowest incidence
(20 percent), partly reflecting the continu-
Source: World Bank Africa Poverty database. ing practice of polygamy, together with high
Note: Figure shows results of a regression on the change in the poverty rate for 43 countries from
1996 to 2012 based on estimated poverty rates using comparable and good-quality surveys.
remarriage rates among widows. The poverty
*** Statistically significant at the 1% level. rates among people living in male-headed
OVERVIEW 11
of deaths from politically motivated vio- At the other end of the spectrum, obesity is
lence declined by 75 percent, and both the emerging as a new health concern.
incidence and tolerance of gender-based Africans enjoyed considerably more peace
domestic violence dropped. Scores on voice in the 2000s than they did in earlier decades,
and accountability indicators rose slightly, but the number of violent events has been on
and there was a trend toward greater par- the rise since 2010, reaching four times the
ticipation of women in household decision- level of the mid-1990s (map O.2). Violence is
making processes. increasingly experienced in terms of political
These improvements notwithstanding, unrest and terrorism rather than large-scale
the levels of achievement remain low in all civil conflicts.
domains, and the rate of progress is leveling Africa also remains among the bottom
off.6 Despite the increase in school enroll- performers in terms of voice and account-
ment, today still more than two out of five ability, albeit with slightly higher scores than
adults are unable to read or write. About the Middle East and North Africa and East
three-quarters of sixth graders in Malawi Asia and the Pacific. Tolerance of domestic
and Zambia cannot read for meaning—just violence (at 30 percent of the population) is
one example of the challenge of providing still twice as high as in the rest of the devel-
good-quality schooling. The need to rein- oping world (figure O.5), and the incidence
vigorate efforts to tackle Africa’s basic educa- of domestic violence is more than 50 percent
tional challenge is urgent. higher. Higher tolerance of domestic violence
Health outcomes mirror the results for lit- and less empowered decision making among
eracy: progress is happening, but outcomes younger (compared with older) women sug-
remain the worst in the world. Increases in gest that a generational shift in mindset is
immunization and bednet coverage are slow- still to come.
ing. Nearly two in five children are malnour- Around these region-wide trends there is
ished, and one in eight women is underweight. also remarkable variation across countries
MAP O.2 The number of violent events against civilians is increasing, especially in Central Africa and the Horn
a. 1997–99 b. 2009–11 c. 2014
IBRD 41867
SEPTEMBER 2015
Sources: Armed Conflict Location and Events Dataset (ACLED); Raleigh and others 2010.
Note: Maps indicate annual number of violent events against civilians; number in parentheses indicates the number of countries. For the following countries there are no data: Cabo
Verde, Comoros, Mauritius, São Tomé and Príncipe, and the Seychelles.
OVERVIEW 13
FIGURE O.5 Acceptance of domestic violence displaced persons—have traits that may
is twice as high in Africa as in other developing make them particularly vulnerable. In 2012,
regions 3.5 million children in Africa were two-
parent orphans (had lost both parents), and
50 another 28.6 million children were single-
parent orphans, bringing the total number
41
of orphans to 32.1 million. The prevalence
Tolerance of domestic violence
40
of orphanhood is particularly high in coun-
(percent of population)
30
30 tries in or emerging from major conflict and
in countries severely affected by HIV/AIDS.
22
Because it can be correlated with wealth and
20
14
urban status, orphanhood does not always
confer a disadvantage on children in terms of
10 schooling. Data on school enrollment among
10- to 14-year-olds in the most recent Demo-
0 graphic and Health Surveys show that in half
2000–06 2007–13 of the countries surveyed, orphans were less
Developing countries in other regions likely to be enrolled than nonorphans.
Sub-Saharan Africa In a sample of seven African countries for
which comparable data are available, almost
Source: Data from Demographic and Health Surveys 2000–13.
Note: Figures are population-weighted averages of 32 African and 28 1 working-age adult in 10 faces severe dif-
non-African developing countries. ficulties in moving about, concentrating,
remembering, seeing or recognizing people
across the road (while wearing glasses), or
and population groups. Literacy is especially taking care of him- or herself. People with
low in West Africa, where gender dispari- disabilities are more likely to be in the poor-
ties are large. High HIV prevalence rates are est 40 percent of the population, largely
holding life expectancy back in Southern because of their lower educational attainment
Africa. Conflict events are more concentrated (Filmer 2008). They score 7.2 percent higher
in the Greater Horn of Africa and the Demo- on the multidimensional poverty index than
cratic Republic of Congo. people without disabilities (Mitra, Posärac,
Rural populations and the income poor are and Vick 2013). Not unexpectedly, disability
worse off in all domains, although other fac- rates show a statistically significant correla-
tors, such as gender as well as the education tion with HIV/AIDS and conflict.
of women and girls, often matter as much or Africa had an estimated 3.7 million refu-
more (at times in unexpected ways). Women, gees in 2013, down from 6.7 million in 1994
for example, can expect to live in good health but up from 2.8 million in 2008. In addition,
1.6 years longer than men; and, among chil- there were 12.5 million internally displaced
dren under 5, boys, not girls, are more likely people, bringing the number of people dis-
to be malnourished (by 5 percentage points).7 placed by conflict to 16.2 million in 2013, or
At the same time, illiteracy remains substan- about 2 percent of Africa’s population (May-
tially higher among women, women suffer stadt and Verwimp 2015). The main source
more from violence (especially domestic vio- of refugees is the Greater Horn of Africa,
lence), and they are more curtailed in their although the number of refugees from Cen-
access to information and decision making. tral Africa is still about 1 million, about half
Multiple deprivation characterizes life for a of them from the Democratic Republic of
sizable share of African women (data on men Congo.
are not available). Although the suffering associated with dis-
Several groups—including orphans, placement is tremendous, the displaced are
the disabled, and refugees and internally not necessarily the poorest; and fleeing often
14 POVERTY IN A RISING AFRICA
helps them mitigate the detrimental effects points), suffer more from domestic violence
of conflict (Etang-Ndip, Hoogeveen, and (by 9 percentage points), and live in countries
Lendorfer 2015). Refugee status is also not that rank low in voice and accountability
always associated with weaker socioeconomic measures (figure O.6).
outcomes. Finally, local economies often also Third, better-educated women (secondary
benefit from the influx of refugees (Maystadt schooling and above) and children in house-
and Verwimp 2015) through increased holds with better-educated women score
demand for local goods (including food) and decisively better across dimensions (health,
services, improved connectivity (as new roads violence, and freedom in decision). More
are built and other transport services pro- rapid improvement in female education and
vided to refugee camps), and entrepreneur- women’s socioeconomic opportunities will be
ship by refugees themselves. game changing in increasing Africa’s capabil-
Three overarching aspects stand out from ity achievement.
a review of the nonmonetary dimensions of
poverty in Africa. First, fragile countries tend
to perform worse and middle-income coun-
Measuring Inequality
tries better. This unsurprising finding con- Although not all aspects of inequality are
firms the pernicious effects of conflict and is necessarily bad (rewarding effort and risk
consistent with the widely observed associa- taking can promote growth), high levels of
tions with overall economic development. inequality can impose heavy socioeconomic
Second, controlling for these factors, costs on society. Mechanically, higher initial
there is a worrisome penalty to residing in inequality results in less poverty reduction
a resource-rich country: people in resource- for a given level of growth. Tentative evidence
rich countries tend to be less literate (by 3.1 also suggests that inequality leads to lower
percentage points), have shorter life expec- and less sustainable growth and thus less
tancy (by 4.5 years) and higher rates of poverty reduction (Berg, Ostry, and Zettel-
malnutrition among women (by 3.7 percent- meyer 2012) (if, for example, wealth is used
age points) and children (by 2.1 percentage to engage in rent-seeking or other distortion-
ary economic behaviors [Stiglitz 2012]). The
pathway by which inequality evolves thus
FIGURE O.6 Residents in resource-rich countries suffer a
matters for poverty reduction and growth.
penalty in their human development
The report measures inequality using the
Gini index, which ranges from 0 (perfect
equality) to 1 (perfect inequality). It shows
Incidence of domestic violence (% points) 9
that inequality is especially high in Southern
Africa (Botswana, Lesotho, Namibia, South
Children’s malnutrition (% points) 2.1 Africa, Swaziland, and Zambia), where Gini
indices are well above 0.5 (map O.3).
Women’s malnutrition (% points) 3.7 Of the 10 most unequal countries in the
world today, 7 are in Africa. Excluding these
Life expectancy (years) –4.5
countries (five of which have populations
of less than 5 million and most of which
are in Southern Africa) and controlling for
Literacy (% points) –3.1
country-level income, Africa has inequality
levels comparable to developing countries
–6 –4 –2 0 2 4 6 8 10
in other parts of the world. Inequality levels
do not differ significantly between coastal
Source: Staff calculations based on World Health Organization and multiple Demographic and
Health Surveys. and landlocked, fragile and nonfragile, or
Note: Figure shows the gap between resource-rich and other countries in Africa. Results control resource-rich and resource-poor countries,
for demographic factors, education, poverty, and other country characteristics (income, fragility,
landlockedness). controlling for subregion.
OVERVIEW 15
Cabo Mauritania
Verde
Mali Niger
Senegal Sudan Eritrea
The Gambia Chad
Guinea-Bissau Burkina Faso
Guinea Benin
Nigeria
Côte Ethiopia
Sierra Leone d’Ivoire Ghana Central African South Sudan
Republic
Liberia Cameroon Somalia
Togo
Equatorial Guinea
Uganda
São Tomé and Princípe Rep. of Kenya
Gabon Congo
Rwanda
Dem. Rep. of Burundi
Congo
Tanzania
Seychelles
Comoros
Gini index Angola
Malawi
0.60–0.63 Zambia
0.50–0.59
0.46–0.49 Mozambique
Zimbabwe Madagascar Mauritius
0.41–0.45
Namibia
0.36–0.40 Botswana
0.31–0.35
No data Swaziland
South Lesotho
Africa
IBRD 41869
SEPTEMBER 2015
For the subset of 23 countries for which Although declines in inequality are associ-
comparable surveys are available with which ated with declines in poverty, poverty fell,
to assess trends in inequality, half the coun- despite increasing inequality, in many coun-
tries experienced a decline in inequality and tries (figure O.7, quadrant 1).
the other half saw an increase. No clear pat- For Africa as a whole, ignoring national
terns are observed by countries’ resource boundaries, inequality has widened. The
status, income status, or initial level of Africa-wide Gini index increased from
inequality. While one might have expected a 0.52 in 1993 to 0.56 in 2008. A greater
more systematic increase in inequality given share of African inequality is explained by
Africa’s double decade of growth and the role gaps across countries, even though within-
the exploitation of natural resources played country inequality continues to dominate.
in that growth, the results presented here do These results stand in contrast to changes
not provide strong evidence for such a trend. in global inequality (Lakner and Milanovic
16 POVERTY IN A RISING AFRICA
FIGURE O.7 Declining inequality is often associated with declining between households. In Rwanda, South
poverty Africa, and Zambia, educational attain-
ment of the household head explains about
Annualized percentage change in Gini index
Berg, Andrew, Jonathan D. Ostry, and Jeromin People.” Policy Research Working Paper 7253,
Zettelmeyer. 2012. “What Makes Growth World Bank, Washington, DC.
Sustained?” Journal of Development Econom- Ferreira, Francisco H. G., and María Ana Lugo.
ics 98 (2): 149–66. 2013. “Multidimensional Poverty Analysis:
Bloom, David, and Jeffrey Sachs. 1998. “Geog- Looking for a Middle Ground.” World Bank
raphy, Demography, and Economic Growth Research Observer 28 (2): 220–35.
in Africa.” Brookings Papers on Economic Filmer, Deon. 2008. “Disability, Poverty, and
Activity 2: 207–95. Schooling in Developing Countries: Results
Canning, David, Sangeeta Raja, and Abdo from 14 Household Surveys.” World Bank
Yazbeck, eds. 2015. Africa’s Demographic Economic Review 22 (1): 141–63.
Transition: Dividend or Disaster? Africa Florian, Krätke, and Bruce Byiers. 2014. “The
Development Forum Series. Washington, DC: Political Economy of Official Statistics:
World Bank. Implications for the Data Revolution in Sub-
Carletto, Calogero, Dean Jolliffe, and Raka Saharan Africa.” PARIS21 Discussion Paper 5.
Banerjee. 2015. “From Tragedy to Renais- Hamilton, Bruce W. 2001. “Using Engel’s Law
sance: Improving Agriculture Data for Better to Estimate CPI Bias.” American Economic
Policies.” Journal of Development Studies 51 Review 91 (3): 619–30.
(2):133–48. Hoogeveen, Johannes, and Nga Thi Viet Nguyen.
CGD (Center for Global Development). 2014. 2015. “Statistics Reform in Africa: Align-
ing Incentives with Results.” Working Paper,
Delivering on the Data Revolution in Sub-
World Bank, Poverty and Equity Global Prac-
Saharan Africa. Final Report of the Data for
tice, Washington, DC.
African Development Working Group. Center
IMF (International Monetary Fund). 2003.
for Global Development and African Popula-
Ghana: First Review under the Three-Year
tion and Health Research Center, Washington,
Arrangement under the Poverty Reduction
DC.
and Growth Facility. IMF Country Report
Christiaensen, Luc, Peter Lanjouw, Jill Luoto, and
03/395, Washington, DC.
David Stifel. 2012. “Small Area Estimation-
———. 2007. Ghana: Article IV Consultation:
Based Prediction Methods to Track Poverty: Staff Report. IMF Country Report 07/210,
Validation and Applications.” Journal of Eco- Washington, DC.
nomic Inequality 10 (2): 267–97. Jerven, Morten. 2013. “Comparability of GDP
Costa, Dora L. 2001. “Estimating Real Income in Estimates in Sub-Saharan Africa: The Effect of
the United States from 1888 to 1994: Correct- Revisions in Sources and Methods since Struc-
ing CPI Bias Using Engel Curves.” Journal of tural Adjustment.” Review of Income and
Political Economy 109 (6): 1288–310. Wealth 59 (S1): S16–S36.
Dabalen, Andrew, and Hai-Anh Dang. 2015. Lakner, Christoph, and Branko Milanovic. 2015.
“The Transition of Welfare over Time for “Global Income Distribution: From the Fall of
Africa: Evidence from Synthetic Panel Analy- the Berlin Wall to the Great Recession.” World
sis.” Background paper prepared for this Bank Economic Review. Advance access pub-
report, World Bank, Washington, DC. lished September 26, 2015.
de la Briere, Benedicte, Deon Filmer, Dena Lanjouw, Peter, and Martin Ravallion. 1995.
Ringold, Dominic Rohner, Karelle Samuda, “Poverty and Household Size.” Economic
and Anastasiya Denisova. 2015. From Mines Journal 105 (433): 1415–34.
to Minds: Turning Sub-Saharan’s Mineral Loayza, Norman V., and Claudio Raddatz. 2010.
Wealth into Human Capital. Washington, “The Composition of Growth Matters for
DC: World Bank. Poverty Alleviation.” Journal of Development
Demographic and Health Surveys. 2015. Calver- Economics 93 (1): 137–51.
ton, MD: ICF International. Maystadt, Jean-François, and Philip Verwimp.
Devarajan, Shantayanan. 2013. “Africa’s Statisti- 2015. “Forced Displacement and Refugees in
cal Tragedy.” Review of Income and Wealth Sub-Saharan Africa: An Economic Inquiry.”
59 (S1): S9–S15. Policy Research Working Paper 7517, World
Etang-Ndip, Alvin, Johannes Hoogeveen, and Bank, Washington, DC.
Julia Lendorfer. 2015. “Socioeconomic Impact Milazzo, Annamaria, and Dominique van de
of the Crisis in North Mali on Displaced Walle. 2015. “Women Left Behind? Poverty
OVERVIEW 19
and Headship in Africa.” Policy Research UN (United Nations). 2015. The Millennium
Working Paper 7331, World Bank, Washing- Development Goals Report 2015. New York:
ton, DC. United Nations.
Mitra, Sophie, Aleksandra Posärac, and Brandon UNESCO (United Nations Educational, Scientific
Vick. 2013. “Disability and Poverty in Devel- and Cultural Organization). 2015. Education
oping Countries: A Multidimensional Study.” for All 2000–2015: Achievements and Chal-
World Development 41: 1–18. lenges. EFA Global Monitoring Report. Paris:
New World Wealth. 2014. Wealth in Kenya: The UNESCO.
Future of Kenyan HNWIs. Johannesburg. van de Walle, Dominique, and Annamaria Mila-
Newhouse, David, S. Shivakumaran, Shinya zzo. 2015. “Are Female-Headed Households
Takamatsu, and Nobuo Yoshida. 2014. “How Poorer? New Evidence for Africa.” mimeo,
Survey-to-Survey Imputation Can Fail.” Policy DECRG. World Bank, Washington, DC.
Research Working Paper 6961, World Bank, World Bank. 2007. “Underreporting of Consumer
Washington, DC. Price Inflation in Tanzania 2002–2006.”
Raleigh, Clionadh, Andrew Linke, Håvard Hegre, World Bank Policy Note, Washington, DC.
and Joakim Karlsen. 2010. “Introducing ———. 2011. World Development Report 2012:
ACLED-Armed Conflict Location and Event Gender and Development. Washington, DC:
Data.” Journal of Peace Research 47 (5). World Bank.
Oxfam. 2015. “Wealth: Having It All and Want- ———. 2012. Niger: Investing for Prosperity: A
ing More.” Oxfam Issue Briefing, January. Poverty Assessment. Washington, DC: World
Oxford. Bank.
Sandel, Michael J. 2012. What Money Can’t Buy: ———. 2013. “Burkina Faso: A Policy Note:
The Moral Limits of Markets. New York: Poverty Trends and Profile for 2003–2009.”
Farrar, Straus and Giroux. World Bank, Washington, DC.
Sen, Amartya. 1985. Commodities and Capabili- ———. 2015a. A Measured Approach to End-
ties. Amsterdam: North-Holland. ing Poverty and Boosting Shared Prosperity:
———. 2002. “Why Health Equity?” Health Concept, Data, and the Twin Goals. Policy
Economics 11 (8): 659–66. Research Report. Washington, DC: World
Stiglitz, Joseph E. 2012. The Price of Inequality: Bank.
How Today’s Divided Society Endangers Our ———. 2015b. “Tanzania Mainland Poverty
Future. New York: W. W. Norton. Assessment.” World Bank, Washington, DC.
Introduction
A
frica has experienced a dramatic 2015), Africa remains the only developing
turnaround since the mid-1990s. region where the MDG 1 target of halving
Following 20 years of economic extreme poverty by 2015 will not be attained.
decline in the 1970s and 1980s, it grew at Understanding and addressing poverty is
a robust pace of 4.5 percent a year, a more complicated by the fact that poverty statistics
rapid pace than in the rest of the developing in the region are often limited and sometimes
world, excluding China. Thanks to a sharp of poor quality. Poverty estimates are based
decline in large-scale conflicts during the on data from a patchwork of household sur-
1990s, better macroeconomic fundamentals veys that are conducted at irregular intervals,
and governance, a commodity supercycle, and that are sometimes incomparable and
and discoveries of new natural resources, the of questionable quality. Concerns about the
narrative of Africa as a “growth tragedy” has availability, comparability, and quality of
shifted to one of Africa rising. poverty data are not unique to Africa, but the
Despite this growth, a large share of the challenges in Africa are perceived as much
African population continues to live below greater than in other regions.
the international poverty line of $1.90 a day. Some researchers have used alternative
Africa’s poverty rate declined from 57 per- data and methods to estimate poverty. They
cent in 1990 to 43 percent in 2012, according find that substantially more people have been
to the latest estimates from the World Bank’s lifted out of poverty than the traditional esti-
PovcalNet database. Because of population mates suggest (Pinkovskiy and Sala-i-Martín
growth, however, the number of poor people 2014; Young 2012). Others are more cau-
implied by these estimates increased, from tious and question such optimism (Chen and
288 million in 1990 to 389 million in 2012. Ravallion 2010; Harttgen, Klasen, and
Poverty reduction in Africa significantly Vollmer 2013).
lags other developing regions. East Asia and The lack of reliable and timely statistics
South Asia started out with poverty rates that in Africa across a range of areas, including
were about as high as Africa’s in the 1990s; poverty, is increasingly recognized as a mat-
their poverty rates are much lower today ter demanding greater international attention
(figure I.1). According to the latest Millen- (Devarajan 2013; Garcia-Verdu 2013; Jer-
nium Development Goal (MDG) report (UN ven 2013). The United Nations’ post-MDG
21
22 POVERTY IN A RISING AFRICA
FIGURE I.1 Poverty reduction in Africa lags other regions it examines five classifications of countries
(table I.1). The literature has identified these
groupings as capturing deep currents that
Percent of population living in poverty
70
determine Africa’s performance in poverty
60 reduction and growth.
50 The report consists of four chapters.
40
Chapter 1 maps out the availability, com-
parability, and quality of the data needed
30 to track monetary poverty (consumption,
20 price, gross domestic product, and census
10
data); reflects on the governance and politi-
cal processes that underpin the current situ-
0 ation with respect to data production; and
describes some approaches to addressing the
90
92
94
96
98
00
02
04
06
08
10
12
19
19
19
19
19
20
20
20
20
20
20
20
data gaps. It is unique in that studies of pov-
Sub-Saharan Africa erty in Africa typically overlook the impor-
South Asia
Latin America and the Caribbean tant yet mundane details of the data on hand.
East Asia and Pacific
Europe and Central Asia
Developing world
TABLE I.1 Classification of countries in Africa
Source: World Bank 2016. Classification Number of countries
Resource-richa 17
Fragileb 17, including 6 that are
frameworks calls for a “data revolution” also resource rich
(UN 2014) to provide timely and reliable Incomec
household surveys and other statistics (such Low 26
Lower-middle 14
as indicators from national accounts). If it
Upper-middle- and high 8
occurs, such a revolution will surely change Landlocked 16
the terms of the debate about living standards Subregion
in Africa, which is now often dominated by Central Africa 9
data and methodological aspects. Data qual- East Africa 18
ity considerations remain very much at the Southern Africa 5
forefront of any assessment of poverty in West Africa 16
Africa.
Note: Countries are classified into subregions according to the UN DESA
Given the state of the data, what is the best classification, with the exception of Sudan, which is classified in that sys-
way to study poverty in Africa and put forth tem as North Africa. Central Africa includes Angola, Cameroon, the Central
African Republic, Chad, the Democratic Republic of Congo, the Republic
an agenda to accelerate poverty reduction? of Congo, Equatorial Guinea, Gabon, and São Tomé and Príncipe. East
This report is the first of two reports that Africa includes Burundi, Comoros, Eritrea, Ethiopia, Kenya, Madagascar,
Malawi, Mauritius, Mozambique, Rwanda, Seychelles, Somalia, South
seek to improve the understanding of poverty Sudan, Sudan, Tanzania, Uganda, Zambia, and Zimbabwe. Southern Africa
reduction in Africa (report 1) and articulate includes Botswana, Lesotho, Namibia, South Africa, and Swaziland.
West Africa includes Benin, Burkina Faso, Cabo Verde, Côte d’Ivoire,
policies to accelerate it (report 2). It reassesses The Gambia, Ghana, Guinea, Guinea-Bissau, Liberia, Mali, Mauritania, Niger,
trends in poverty and inequality in Africa Nigeria, Senegal, Sierra Leone, and Togo.
a. Resource-rich countries include countries that had average rents
by examining the primary data sources and from natural resources (excluding forests) that exceeded 10 percent of
identifying potential biases in them. Careful GDP in 2006–11; countries with diamonds (Botswana, Liberia, Namibia,
and Sierra Leone); and Niger (which has uranium). The group does
evaluation of the data for monitoring poverty not include Somalia, for which inadequate data are available for
in Africa will help sharpen the focus on data classification.
b. Fragile countries are countries that appear on the World Bank’s 2015
issues in Africa in general and on consump- harmonized list of fragile situations, which classifies countries as fragile
tion data in particular. if they (a) had an average Country Policy and Institutional Assessment
(CPIA) rating of 3.2 or less or (b) hosted a UN or regional peace-keeping
A regional report like this cannot provide or peace-building mission in the previous three years.
in-depth analysis for each country. Instead, c. Country income categories are from World Development Indicators.
INTRODUCTION 23
Chapter 2 evaluates the robustness of the Devarajan, Shantayanan. 2013. “Africa’s Statisti-
estimates of poverty in Africa. It concludes cal Tragedy.” Review of Income and Wealth
that poverty reduction in Africa has not been 59 (S1): S9–S15.
overestimated and in fact may be slightly Garcia-Verdu, Rodrigo. 2013. “The Evolution of
Poverty and Inequality in Sub-Saharan Africa
greater than traditional estimates suggest,
over the Period 1980–2008: What Do We (and
although even the most optimistic estimates
Can We) Know Given the Data Available?”
of poverty reduction imply that more than International Monetary Fund, Washington,
330 million people were living in poverty in DC.
2012. The chapter also presents a very broad- Harttgen, Kenneth, Stephan Klasen, and Sebas-
stroke profile of poverty and trends in pov- tian Vollmer. 2013. “An African Growth Mir-
erty in the region. acle? Or: What do Asset Indices Tell Us about
Chapter 3 broadens the view of poverty Trends in Economic Performance?” Review of
by considering nonmonetary dimensions of Income and Wealth 59 (S1): S37–S61.
well-being, such as education, health, and Jerven, Morten. 2013. “Comparability of GDP
freedom, using Sen’s (1985) capabilities and Estimates in Sub-Saharan Africa: The Effect of
Revisions in Sources and Methods since Struc-
functionings approach. In contrast to the
tural Adjustment.” Review of Income and
dearth of good-quality and comparable sur-
Wealth 59 (S1): S16–S36.
veys on household expenditures, there has Pinkovskiy, Maxim L., and Xavier Sala-i-Martín.
been a surge in survey-based information on 2014. “Africa Is on Time.” Journal of Eco-
these and related nonmonetary dimensions of nomic Growth 19 (3): 311–38.
poverty. Sen, Amartya. 1985. Commodities and Capabili-
Chapter 4 reviews the evidence on ties. Amsterdam: North-Holland.
inequality in Africa. In addition to patterns UN (United Nations). 2014. A World that Counts:
of monetary inequality, it examines other Mobilising the Data Revolution for Sustain-
dimensions, including inequality of opportu- able Development. Independent Expert Advi-
nity and intergenerational mobility in occu- sory Group on a Data Revolution for Sustain-
able Development, New York.
pation and education. Viewing inequality
———. 2015. The Millennium Development
from beyond the realm of household surveys,
Goals Report 2015. New York: UN.
this work also explores extreme wealth (bil- World Bank. 2016. Global Monitoring Report
lionaires and millionaires) in Africa. 2015/2016: Development Goals in an Era
of Demographic Change. Overview booklet.
Washington, DC: World Bank.
References Young, Alwyn. 2012. “The African Growth Mir-
Chen, Shaohua, and Martin Ravallion. 2010. acle.” Journal of Political Economy 120 (4):
“The Developing World Is Poorer Than We 696–739.
Thought, but No Less Successful in the Fight
against Poverty.” Quarterly Journal of Eco-
nomics 125 (4): 1577–625.
The State of Data for
Measuring Poverty 1
A
frica has grown robustly for two number of household surveys, particularly
decades—performance that lies surveys that collect data on the nonmonetary
in stark contrast to the “growth dimensions of poverty, has increased, thanks
tragedy” of the 1980s (Easterly and Levine to donor-funded programs such as the Demo-
1997). The statistics suggest that Africa’s graphic and Health Surveys (DHS) and the
people are faring better and that poverty has Multiple Indicator Cluster Surveys (MICS).
come down. But scrutiny of these statistics The frequency and coverage of data on citizen
has raised doubts about the quality of the opinions on a wide range of topics, including
underlying data and the exact magnitude of governance, political leadership, democracy,
Africa’s progress. The World Bank’s Bulletin and corruption, have increased, and data
Board on Statistical Capacity indicator gave tracking salient events, such as conflict and
Africa a regional score of 59 in 2014, well weather events, are now widely available. In
below the world average of 66 and low even addition to national statistical offices, the
relative to the average for the low-income cat- actors in data collection now include non-
egory of countries. The lack of good-quality governmental organizations (NGOs), polling
and accessible data to assess socioeconomic firms, and universities.
changes now regularly features in discus- These improvements notwithstanding,
sions of the development agenda for Africa major concerns remain. Problems with the
(Devarajan 2013; Jerven 2013). availability, comparability, and quality of the
There is no doubt that Africa needs bet- data, combined with different approaches
ter data to monitor the evolution of both the and methods to correct for these shortcom-
monetary and nonmonetary dimensions of ings, are at the center of the divergent views
living conditions. Progress on this front will regarding the direction and magnitude of
also be crucial to monitor the post-2015 Sus- poverty reduction in Africa over the past two
tainable Development Goals (SDGs). To be decades (Chen and Ravallion 2010; Hartt-
sure, there have been improvements in data gen, Klasen, and Vollmer 2013; Pinkovskiy
availability in Africa in recent years. The and Sala-i-Martín 2014; Young 2012).
Consider the measurement of monetary
This chapter was written with Rose Mungai, Nga Thi poverty, for example. The share of Africa’s
Viet Nguyen, and Shinya Takamatsu. population consuming less than $1.90 a day
25
26 POVERTY IN A RISING AFRICA
(in 2011 international purchasing power par- estimate poverty for small areas in a country.
ity [PPP] dollars) declined, according to the Gross domestic product (GDP) from national
World Bank’s PovcalNet, falling from 57 per- income accounts is used to fill gaps between
cent in 1990 to 43 percent in 2012.1 How- surveys to provide annual poverty estimates.
ever, this estimate is based on surveys in a This chapter reviews the state of these
subsample of countries that cover only one- data in Africa. It reflects on the governance
half to two-thirds of the population. For the and political incentives that influence data
remaining population, the poverty rate was production, in order to help understand why
imputed from surveys that were often sev- multiple challenges beset the data for poverty
eral years old. For five countries (Equatorial measurement, and discusses some approaches
Guinea, Eritrea, Somalia, South Sudan and for addressing data shortfalls.
Zimbabwe), which together represent 5 per-
cent of the African population, no data were
available with which to measure poverty.
Types of Data for Measuring
Equally if not even more important are Monetary Poverty
concerns about the comparability and quality Estimating poverty requires consumption
of the underlying household survey and price or income data from household surveys, but
data. Guinea and Mali, for example, each other data are also needed. This includes price
fielded four surveys since the mid-1990s, but data to adjust nominal consumption values
no two of these surveys is considered compa- for changes in price levels over time, census
rable for measuring poverty. data to estimate the population, and national
Against this background and as a start- accounts data to impute poverty in years in
ing point in revisiting estimates of poverty which no household survey was conducted.
in Africa, this chapter takes stock of the
data available to measure the evolution of
monetary poverty in the region. It focuses Household Survey Data
on household-level consumption and price Household surveys are essential for obtaining
data but also briefly reviews auxiliary data the socioeconomic data necessary to under-
sources needed to estimate poverty. stand the welfare of populations across the
The cornerstone of poverty estimates in world. Some 50 years ago, regular household
Africa (and most other developing regions) surveys were virtually nonexistent in devel-
are consumption data from household sur- oping countries. Although both the number
veys that are representative of the popula- of surveys conducted in Africa and their com-
tion. 2 By themselves, consumption data parability and quality have improved, sub-
are not sufficient to analyze changes in liv- stantial gaps remain.
ing standards. Monitoring changes in real
terms requires data on inflation at the coun- Frequency and scope of data collection
try level—such as a consumer price index Only a handful of household surveys were
(CPI)—to adjust nominal consumption into collected in Africa in the 1980s. The num-
real values. Estimating global or regional pov- ber grew modestly for almost a decade,
erty levels requires setting a common poverty expanding rapidly in the mid-1990s, partly
line, such as the international poverty line as a result of growing interest among gov-
of $1.90 per capita per day, and converting ernments and the international community
local currency units to a common reference in monitoring the Millennium Development
currency. Auxiliary data sources also have a Goals (MDGs). The first decade of the 2000s
bearing on Africa’s poverty estimates. Popu- was one of the most productive for household
lation censuses are needed to derive popula- data collection in Africa. By 2010 the number
tion statistics from sample surveys and, when of national household surveys in Africa was
used jointly with a consumption survey, the second highest in the developing world,
THE STATE OF DATA FOR MEASURING POVER T Y 27
after South Asia (Demombynes and Sandefur FIGURE 1.1 All regions have increased the number of household
2014; Garcia-Verdu 2013) (figure 1.1).3 surveys they conduct
The breadth of the socioeconomic data
that surveys cover has also increased. A 2.5
majority of African countries collect data
on welfare and key MDG indicators from
BOX 1.1 Sources outside the national statistical system provide valuable information on
well-being
Impressive large-scale household survey efforts are Each wave has covered a wider range of topics, some
being conducted outside the national statistical sys- of which are harmonized across countries. Eleven
tem. They elicit data on nonconsumption aspects of African countries have been included, some with
well-being and perceptions.a multiple rounds.
three countries over the period 2000–09, 23 FIGURE 1.2 Africa conducts more nonconsumption surveys than
countries conducted one survey and another consumption surveys
21 had at least two surveys.
A wave of consumption surveys was con- 160
140 139 138
ducted in the region between 2011 and 2015.
FIGURE 1.3 Many African countries lack surveys with which to gauge changes in poverty
50 47 47 47 47 47 47 47 47 47 47 47 47 48 48
45 3 3 3 4 4
7 6 6 5 5 5
8 8 8
40 10 9
3 8 6 13 10
5 6 7 12 16 15
Number of countries
35 2 14
2 3 5
2 4 4
30 5 3
4 4
25 16 4 6
16 15 7
14 20
20 15 21 23
15 22 25
23 21 19
17
10 18 16 16 15
5 13 12
9 7 5 4 3 3 4 4
0
9
2
–9
00
00
00
00
00
00
00
00
00
–0
–1
–1
–1
90
00
01
02
03
–2
–2
–2
–2
–2
–2
–2
–2
–2
19
91
92
93
94
95
96
97
98
99
20
20
20
20
19
19
19
19
19
19
19
19
19
MAP 1.1 More than half of African countries completed a consumption survey between 2011 and early
2015
Cabo Mauritania
Verde
Mali Niger
Senegal Sudan Eritrea
The Gambia Chad
Guinea-Bissau Burkina Faso
Guinea Benin
Nigeria
Côte Ethiopia
Sierra Leone d’Ivoire Ghana Central African South Sudan
Republic
Liberia Cameroon Somalia
Togo
Equatorial Guinea
Uganda
São Tomé and Príncipe Rep. of Kenya
Gabon Congo
Rwanda
Dem. Rep. of Burundi
Congo
Tanzania
Seychelles
2007 and earlier
2008 to 2010 Comoros
Angola
2011 and later Malawi
Zambia
No data
Mozambique
Zimbabwe Madagascar Mauritius
Namibia
Botswana
Swaziland
South Lesotho
Africa
IBRD 41864
OCTOBER 2015
conducted. 5 Eritrea and Somalia have not comparable with one another (or with those
fielded national consumption surveys over of other countries). Tracking poverty trends
the past 20 years. These five countries repre- is difficult when changes in measured con-
sent 5 percent of the region’s population. sumption partly reflect changes in survey
design or implementation.
Comparability of consumption data The survey design literature documents
The lack of consumption surveys is an obvi- multiple ways in which two surveys can be
ous impediment to monitoring poverty, but rendered noncomparable. For this report,
problems with consumption data do not household consumption surveys are consid-
end there. Even where multiple surveys are ered comparable if the following features are
available for a country, they are often not consistent across surveys:6
THE STATE OF DATA FOR MEASURING POVER T Y 31
FIGURE 1.4 Comparability of consumption surveys has improved, but it remains a major problem
1990 1995 2000 2005 2010
Angola ◆ ◆ ◆
Benin ◆ ◆ ◆ ◆ ◆ ◆ ◆ ◆
Botswana ◆ • •
Burkina Faso ◆ • • ◆ ◆ ◆
Burundi • •
Cameroon ◆ • •
Cabo Verde ◆ ◆ ◆
Central African Republic ◆ ◆ ◆ ◆
Chad ◆ • •
Comoros ◆ ◆
Congo, Dem. Rep. • •
Congo, Rep. ◆ ◆
Côte d’Ivoire • ◆ • • •
Equatorial Guinea ◆
Eritrea ◆
Ethiopia • • • •
Gabon ◆ ◆
Gambia, The • ◆ • ◆ ◆ ◆
Ghana • • • •
Guinea ◆ ◆ ◆ ◆ ◆
Guinea-Bissau ◆ ◆ ◆ ◆
Kenya ◆ ◆ ◆ ◆
Lesotho ◆ ◆ ◆
Liberia ◆
Madagascar ◆ • • • • •
Malawi ◆ ◆ • •
Mali ◆ ◆ ◆ ◆
Mauritania • • ◆ • ◆ •
Mauritius ◆ ◆ ◆ • •
Mozambique • • •
Namibia ◆ • •
Niger ◆ ◆ ◆ ◆ ◆
Nigeria ◆ ◆ • •
Rwanda • • •
São Tomé and Príncipe ◆ ◆
Senegal ◆ ◆ ◆ • •
Seychelles ◆ • •
Sierra Leone • •
Somalia
South Africa ◆ ◆ • • • ◆ •
South Sudan ◆
Sudan ◆
Swaziland ◆ • •
Tanzania ◆ • • ◆
Togo • •
Uganda • • • • • • • • • •
Zambia ◆ ◆ ◆ • ◆ • • ◆
Zimbabwe* ◆ ◆ ◆ ◆ ◆ ◆ ◆
1990 1995 2000 2005 2010
MAP 1.2 Lack of comparable surveys in Africa makes it difficult to measure poverty trends
Cabo Mauritania
Verde
Mali Niger
Sudan Eritrea
Senegal Chad
The Gambia
Guinea-Bissau Burkina Faso
Guinea Benin
Nigeria
Côte Ethiopia
Sierra Leone d’Ivoire Ghana Central African South Sudan
Republic
Liberia Cameroon Somalia
Togo
Equatorial Guinea Uganda
São Tomé and Príncipe Rep. of Kenya
Gabon Congo
Rwanda
Dem. Rep. of Burundi
Congo Tanzania
Seychelles
Comoros
Number of comparable surveys conducted, Angola
1990–2012 Malawi
0 or 1 survey (9 countries) Zambia
No comparable surveys (12 countries)
2 comparable surveys (17 countries) Zimbabwe Mauritius
Mozambique
More than 2 comparable surveys (10 countries) Namibia
Botswana Madagascar
Swaziland
South Lesotho
Africa
IBRD 41865
SEPTEMBER 2015
Source: Data from the World Bank microdata library.
years, and the average Latin American coun- than the benchmark method of personal
try conducts at least one survey every two diary with daily visits yielded poverty rates
years. If comparability is taken into account, that were 7–19 percentage points higher.
the picture is even worse, with African coun- Most instruments, including household-level
tries producing just 1.6 comparable poverty diaries or recall questionnaires of different
estimates per country between 1990 and granularity, thus underreport consumption
2012. compared with the supervision-intensive per-
Does noncomparability matter? Survey sonal diary. Backiny-Yetna, Steele, and Djima
experiments show that changes in question- (2014) show that poverty estimates in Niger
naire design can matter a lot. According to are sensitive to the reporting period, with
Beegle and others (2012), use of diary ver- estimates of 51 percent, 47 percent, and 43
sus recall, shorter versus longer reporting percent depending on the approach. Results
periods, and changes in the number of con- from the 2005/06 survey in Kenya also point
sumption items drastically affect poverty and to significant differences in poverty calcu-
inequality measures. Using methods other lations depending on whether the recall or
34 POVERTY IN A RISING AFRICA
BOX 1.2 How did poverty change in Guinea and Mali? Lack of comparable data makes
it difficult to know
Guinea conducted four household surveys between nonfood items, the 2002/03 survey included 240
1994 and 2012. The 1994/95 and the 2002/03 sur- food and 425 nonfood items, and the 2007 and 2012
veys were conducted over 12 months, the 2007 sur- surveys included 110 food and 130 nonfood items.
vey was conducted in July–October 2007, the 2012 Mali implemented four surveys between 1994
survey was conducted in February–March 2012. In and 2012; the surveys vary in a number of ways.
1994/95 each household was visited 11 times, one The 1994/95 survey included 10 food and 34 non-
visit every three days for a month. Food consump- food items, the fewest among the surveys, and a
tion data were collected from visit 2 to visit 11, using 15-day food recall period. In 2001/02 every house-
a three-day recall period. A 12th of the sample was hold was interviewed every quarter. Food consump-
visited each month. In 2002/03 each household was tion data were collected through a seven-day diary;
visited three times, or once every four months (the in theory each household was visited 7 times a
survey is thus a panel of three observations). Dur- quarter, for a total of 28 visits during the year. The
ing each visit, food consumption data were collected 2006 and 2010 surveys were Core Welfare Indica-
using a three-day recall period in urban areas and tors Questionnaire (CWIQ)–type surveys fielded in
two-day recall in rural areas. In the 2007 and 2012 July–November 2006 and December 2009–August
surveys, each household was visited once. Food con- 2010. Food consumption data were collected using
sumption data were collected by asking about typical the usual-month approach. The number of items on
monthly consumption (not actual consumption, such the questionnaires was similar, although some types
as consumption the previous week). The 2007 and of expenditures (food eaten away from home, bev-
2012 surveys were conducted in different seasons. erages, cigarettes) were reported by each individual
The number of consumption items also differed: the household member using an open list.
1994/95 questionnaire included 116 food and 110
TABLE 1.1 Africa lags in the number of comparable surveys per country, conducted between 1990 and 2012
Developing countries that conducted at least Average
one consumption survey Average number of
number of comparable
Country Population Median year surveys per surveys per
Number of coverage coverage of most recent developing developing
Region countries (percent) (percent) survey country country
East Asia and Pacific 15 63 96 2010 3.9 2.8
Europe and Central Asia 21 100 100 2011 10.0 6.4
Latin America and the Caribbean 22 85 98 2011 11.1 6.3
Middle East and North Africa 12 92 98 2007 3.2 1.8
South Asia 8 100 100 2010 4.1 2.8
Africa 47 98 99 2010 3.8 1.6
World 125 89 98 2010 6.0 3.5
Sources: Data from the World Bank microdata library, PovcalNet, and World Development Indicators.
Note: The table includes low-income, lower-middle-income, and upper-middle-income countries, with the exception of Equatorial Guinea, which is a high-income country.
diary approach to consumption was used the General Household Survey-Panel (GHS-
(Dabalen and others 2015).9 Panel), which was launched in the last quar-
In Nigeria two household surveys were ter of 2010. The NLSS, which relied on the
conducted the same year. The Nigeria Liv- diary approach, reported much lower con-
ing Standards Survey (NLSS) was fielded in sumption than the GHS-Panel, which used
2009/10. It overlapped with the first wave of the recall approach (figure 1.5). The surveys
THE STATE OF DATA FOR MEASURING POVER T Y 35
FIGURE 1.5 Different survey designs can result in (Biemer and Lyberg 2003; Gryna and Juran
very different consumption estimates 1980). At the core of data quality problems
is often a process failure.10 Interviewers
may fail to make contact with respondents
and subsequently report fake data, perhaps
Food
because supervision was lax or insufficient (as
Finn and Ranchhod [forthcoming] document
in a survey in South Africa). Enumerators
may not have been given sufficient training
to probe for the responses intended by the
Nonfood questions. Respondents may refuse to par-
ticipate, or they may provide false informa-
tion. Modes of data collection—computers,
phones, paper—could also be compromised
0 20 40 60 because the infrastructure needed was not
Thousands of naira planned appropriately. Errors may be intro-
Nigeria Living Standard Survey (NLSS) duced in entering (or keying) data. Poor data
General Household Survey (GHS)–Panel
quality can undermine comparability over
Sources: Data from the NLSS and GHS-Panel for overlapping months in 2010.
time because process failures that occur one
year may not be repeated in another.
Misreported data are clearly the most seri-
were also different in other salient ways, in ous way data quality can be compromised.
particular with respect to field supervision There is little value in all the other dimensions
and field team composition, both of which of data (such as timeliness, richness of detail,
may affect quality. relevance, availability, and even comparabil-
At the country level, noncomparability ity), if the data are erroneous and hence can-
between survey rounds is often a concern; not be used for the purposes for which they
country-level poverty reports are replete were designed (Biemer and Lyberg 2003).
with discussions of survey comparability (see The systematic detection of poor quality
World Bank 2013 for Burkina Faso; World is challenging. Judge and Schechter (2009)
Bank 2012 for Niger; World Bank 2015b for apply Benford’s law—a statistical method
Tanzania). These differences are often over- for reviewing the digits in reported statistics
looked at the regional level, partly because for abnormal patterns as a sign of fraudu-
databases such as PovcalNet do not vet sur- lence—to surveys in Bangladesh, Ghana,
veys on the basis of comparability. Mexico, Pakistan, Paraguay, Peru, South
Lack of survey comparability within coun- Africa, the United States, and Vietnam. They
tries across time is not unique to consump- find widespread evidence of fake crop and
tion measures. It has been reported in the livestock production data. Among the sur-
measurement of literacy, for example (see box veys reviewed, data quality was far worse in
3.2 in chapter 3). Although more systematic surveys in developing countries. Consump-
documentation of these differences in a meta- tion data for almost 40 percent of households
database would not resolve these issues, it surveyed in the Malawi 1997/98 household
would be helpful to analysts. survey were incomplete or inaccurate, and
the data were unusable in poverty analysis
Quality of consumption data (Benson, Machinjili, and Kachikopa 2004).
The closest approximation of a broad defini- One commonly observed manifestation of
tion of good-quality data involves fitness for poor quality is deterioration in reporting over
use: data should be accurate, rich in detail, the survey period that cannot be explained by
relevant, timely, and likely to achieve the seasonality. In Tanzania average household
purposes for which the survey was intended size fell significantly over the course of surveys
36 POVERTY IN A RISING AFRICA
FIGURE 1.6 Data errors may account for some of standards between countries calls for the use
the reported change in consumption of PPP exchange rates to achieve parity in the
purchasing power of people’s incomes. The
35 same principle applies within countries, where
consumers in rural and urban areas often face
30
different prices, but the evidence for Africa is
Million Leones/number of items
Combining the price data with weights to FIGURE 1.7 The weights used to construct
construct the CPI is a complex process that consumer price indexes in Africa are outdated
often differs significantly across countries.
Partly because of these variations and partly Missing
because the CPI is not designed specifically 11% 2010 or later
to apply to the measurement of poverty, CPIs 2%
Before 1995
may not always accurately depict changes in 2%
the cost of living experienced by the average
household or (particularly) the poor. 1995–99
CPIs suffer from several potential sources 11%
of bias. Commodity substitution bias relates 2005–09
to the use of an imperfect indexing formula 51%
and outdated weights. The most common 2000–04
23%
index for CPIs is the Laspeyres index, which
uses weights from a base (reference) period.
This index disregards substitution behavior
that may stem from inflation itself—that is,
it ignores the fact that when the prices of Source: ILO 2013.
Note: Figures indicate the share of Africa’s population in 2013 living in
some goods rise more quickly than the prices countries in which the weights used to calculate the consumer price index
of others, households shift consumption to (CPI) in July 2012 came from each time period.
similar but cheaper items. It therefore overes-
timates inflation and underestimates poverty and brands increases living standards. Econo-
reduction. metric techniques seek to estimate the gains
Updating weights can address this prob- that occur as a result. Hausman (1996, 1999)
lem, but CPI weights are often many years measures the consumer gains resulting from
old. As of July 2012, for example, 13 per- the introduction of new breakfast cereals and
cent of the African population was living in mobile phone services by estimating virtual
countries in which the CPI basket was based (reservation) prices. Whether such techniques
on data from the 1990s (or earlier), and data should find their way into the estimation of
on 11 percent of the population were missing the CPI remains controversial.12 New prod-
altogether (figure 1.7). uct bias is by definition positive. It leads to
Outlet substitution bias is related to an overestimation of inflation in the CPI
changes in the retail landscape. Price data and therefore an underestimation of poverty
for the CPI are often collected from a fixed reduction.
set of stores or markets. With the advent of Plutocratic bias arises because CPI weights
discount retail stores in some countries in are computed in a way that implicitly weights
Africa, failure to adjust where the price data households in proportion to their total con-
are collected is expected to lead to an overes- sumption (so-called plutocratic weights) and
timation of inflation and underestimation of are hence more representative of wealth-
poverty reduction. ier households (Deaton 1998; Ley 2005;
Quality change bias reflects the fact that Oosthuizen 2007). Plutocratic weights are the
the quality of a product can change (typically, natural choice in the deflation of economic
improve) while the price remains unchanged. aggregates, such as national accounts, but
Evidence from the developed world suggests generally not the first choice for measuring
that quality change bias generally leads to an poverty and welfare. The alternative would be
overestimation of inflation (Hausman 2003). weighting all households equally (Prais 1959).
Overestimating inflation thus understates If consumption patterns and rates of inflation
poverty reduction. differ among poor, average, and better-off
New products bias is similar to quality households, the CPI will not accurately track
change bias. The introduction of new products the changes in prices experienced by the poor.
38 POVERTY IN A RISING AFRICA
In Africa and other developing regions, lower inflation than the better off between
there is empirical evidence that inflation 1998 and 2003.
inequality can be important—that is, the Urban bias arises because many CPIs
poor and the nonpoor may experience differ- in Africa are based on prices collected only
ent inflation rates. Whether these differences in urban areas. Some countries also base
result in over- or underestimation of the infla- weights only on urban consumption patterns.
tion faced by the poor is less clear. In Burkina Urban-based prices and weights are signifi-
Faso in 1994–98, food crop prices increased cantly more prevalent in Africa than else-
much more quickly than the prices of other where (figure 1.8). There is reason to believe
consumer items (Günther and Grimm 2007). that the urban bias in prices and weights is
Because the poor spend a larger share of even more common than suggested by the
their budgets on food, they experienced data of the International Labour Organiza-
higher inflation than other consumers. Infla- tion (ILO). For instance, Kenya, which is
tion inequality has also been documented listed as having nationwide coverage in the
in Brazil, Colombia, Indonesia, Mexico, ILO database, reports, in its CPI publica-
Peru, South Africa, Tanzania, and Uganda tion, that, outside of Nairobi, urban centers
(Goñi, López, and Servén 2006; McCull- were selected to represent each province
och, Weisbrod, and Timmer 2007; Mkenda (KNBS 2010). Whether urban bias matters
and Ngasamiaku 2009; Okidi and Nsubuga in measuring poverty depends on whether
2010; Oosthuizen 2007). While some stud- rural inflation does or does not track urban
ies find that the inflation poor households inflation.
experience is higher, in some countries it is Bias from the treatment of own consump-
better-off households that face higher rates tion stems from the practice of including
of inflation. Even within the same country, only market purchases in the CPI weights,
the direction of bias can change. In Burkina excluding consumption from food grown
Faso, for example, the poor encountered by the household. One-quarter of Africa’s
FIGURE 1.8 Both the prices and weights used to construct consumer price indexes in Africa reflect a
strong urban bias
a. Prices b. Weights
100 100
90 90
80 80
70 70
60 60
Percent
Percent
50 50
40 40
30 30
20 20
10 10
0 0
Africa Low-income and Africa Low-income and
lower-middle-income lower-middle-income
countries in countries in
other regions other regions
Missing Urban areas Main cities Main city Nationwide
revolved around whether the world has then rank countries on the basis of these pov-
become more or less equal and whether it erty rates and compare these ranks to ranks
has become less poor relative to the United obtained using 2005 PPP and 2011 PPP. For
States, whose currency is taken as the bench- a sample of five African countries, the 2011
mark when calculating these exchange rates. PPP ranking followed the ranking from this
Such debates have become routine with every imputation approach more closely than the
round of ICP PPP releases (see the discussion 2005 PPP did. In contrast, there was no
in Almås 2012; Ciccone and Jarociński 2010; major difference in the rankings of the 2011
Deaton 2010), partly because in each round and 2005 PPP on the one hand and the rank-
major revisions have been made to methods, ing based on the imputation approach for a
the number of countries participating, and sample of countries in Europe and Central
coverage (rural and urban) within countries, Asia and Latin America and the Caribbean.
so that some reranking becomes inevitable. What do the latest PPPs say about the
In the latest release, the consumption and change in national income levels (GDP per
income of the average developing country capita) in Africa? The region remains the
rose by 25 percent (Inklaar and Rao 2014). world’s poorest, even though its share of
The new PPPs project large declines in pov- global income inched higher, from 3.3 per-
erty and a shift in the geography of the poor cent in 2005 to 4.5 percent in 2011. All 10
from Asia to Africa (Dykstra, Kenny, and of the world’s poorest economies were in
Sandefur 2014; Jolliffe and Prydz 2015). Africa. Country rankings within Africa
Experts are divided over whether the 2005 remained fairly stable, but there were some
or the 2011 PPP better describes the world. changes in rank, such as Botswana and
Supporters of the 2011 round (Deaton and Gabon at one end and Ghana and Zambia in
Aten 2014) argue that the methodological the middle (figure 1.9).
changes introduced in 2011, in particular
the use of a core global list of goods rather
Population Census and GDP Data
than 18 ring countries in 2005, undid some
of the mistakes made in the 2005 PPP, which Surveys and price data are not the only data
inflated the price ratios for Africa, Asia needed for estimating poverty. Census data
(without Japan), and western Asia by 20–30 are needed both to select the sample for a
percent. On the other side of the debate, survey and to estimate the size of the popula-
Ravallion (2014) finds that the 2011 PPP tion. GDP data from the system of national
places more weight on strongly internation- accounts are used to estimate poverty in
ally traded goods than do past ICP rounds, years with no survey.
seen through a convergence of price levels and
exchange rates, especially in Asia. He argues Census data
that these results are inconsistent with expec- A census is essential for measuring and moni-
tations from the methodological changes toring monetary and nonmonetary poverty,
introduced in the 2011 ICP round. for several reasons. First, it is the basis for
Lanjouw, Massari, and van der Weide the sample frame for surveys and the selec-
(2015) use a multiple imputations approach tion of the primary sampling units (commu-
that avoids the use of PPPs entirely to rank nities) from which households are sampled.
poverty rates of countries. Their method gen- At the back end of surveys, censuses—spe-
erates multiple imputed consumption and cifically the population projections from the
poverty rates for each country (so for a sam- past census to the survey year—are needed to
ple of five countries, there are five estimates obtain the population statistics from the sur-
per country), each corresponding to the esti- vey estimates. The absence of an up-to-date
mate obtained when a particular country census introduces significant uncertainty into
is used as the reference in the model. They population-level statistics on living standards
THE STATE OF DATA FOR MEASURING POVER T Y 41
FIGURE 1.9 Adoption of the 2011 purchasing power parity values increased GDP per capita figures across Africa
18,000
16,000
14,000
GDP per capita (PPP)
12,000
10,000
8,000
6,000
4,000
2,000
0
-B e
Bo uin s
Gu Zim ibe .
in ba ria
hi u
n N pia
L e
an e
T da
ad u i o
a a
r
nz a
ia
rk o li
a s
m o
Gh bia
p er
M mb ala c
oz ia wi
Si amb The
enna
é L Ben a
an e in
Cô Prí tho
d’ ipe
M ene re
rit al
C ia
Ni had
Ca jibo ia
er i
V n
e
Sw , R o
yp N zila .
t, am nd
ab bia
ut un .
ua M Af a
ria ur a
Ga na
n
Ca Su on
ng roc la
ts ea
L ep
a ep
So T ep
m ut
Ug sca
l G itiu
in ro
i
Bu Com Ma
to a ric
Ta and
h isi
y
ag ne
e a bw
Rweon
Anerd
ra qu
c
M G og
bo
Et issa
Za Fas
bo da
Ga M ubl
au g
an
an
D ger
Co o go
Re i g
S oi
a
a
o
.R
R
o
Ar i
te nc
d so
w
er i
Iv
,
K
m
o
M
De
ca
o,
ng
fri
Eg
m
lA
Co
Eq
To
ra
o
nt
Sã
Ce
2005 2011
(or any measures from household surveys) correct count of the poor there is critical for
(World Bank 2015a). Second, census data regional estimates.
have been used to estimate poverty rates and Only a handful of countries make their
poverty counts at the smallest possible juris- census data sets available to the public.
diction, through poverty mapping techniques The Integrated Public Use Microdata Series
(Elbers, Lanjouw, and Lanjouw 2003). Third, (IPUMS)—the world’s largest collection of
census data are useful for understanding a public use census microdata files—currently
number of nonmonetary dimensions of living includes 19 African countries.15
standards, such as housing conditions and
educational attainment. National accounts data
Because of the enormous financial, per- National accounts are the comprehensive
sonnel, and managerial demands of cen- economic statistics that measure economic
suses, they are ideally conducted once activity in a country. They are also impor-
every 10 years. The coverage of population tant for estimating poverty in years in which
censuses in Africa improved significantly no survey has been conducted. Rather than
in the last two rounds. In the 2000 round assume a steady rate of change in poverty
(1995–2004), 33 of 47 countries partici- between survey rounds, researchers apply
pated; only 8 countries had no census in the per capita growth rates of GDP or private
2010 round (2005–14).14 The eight countries consumption (referred to as household final
represent about 13 percent of Africa’s popu- consumption expenditure in the World Devel-
lation. The Democratic Republic of Congo opment Indicators) to the household survey
has not conducted a census since 1984. means to interpolate the pattern of poverty
Because it is estimated to be the third most between two surveys or extrapolate it beyond
populous country in Africa, obtaining the the survey range (when no other survey is
42 POVERTY IN A RISING AFRICA
available).16 For a country with only one sur- grows), the base year becomes less and less
vey, the survey mean is adjusted forward and representative of the economy and therefore
backward using the real growth rate of GDP requires updating. The international recom-
per capita to give poverty estimates in other mendation is to update the base year at least
years (see World Bank 2015a). These calcula- every five years. This process of replacing the
tions assume that GDP per capita or private base year is known as rebasing.
consumption per capita grows at the same Thanks to rebasing, a national economy
rate for everyone. can grow statistically overnight (figure 1.10).
W hen used to interpolate, national The GDP rebasing exercise carried out by
accounts imputation is preferred over assum- Ghana in 2010, for example, caused such a
ing a steady rate of poverty change between large increase in GDP that Ghana jumped
survey rounds. This approach helps capture from low-income to low-middle-income
possible downturns and upswings between country status. Rebasing in Nigeria in 2014
surveys. The assumption that each house- propelled it to surpass South Africa as the
hold’s consumption expands uniformly at the biggest economy in Africa. The announce-
rate of the overall economy becomes more ment drew much attention from the media,
tenuous when extrapolating beyond the sur- business community, economists, and inter-
veys, especially farther into the future (or the national organizations (BBC 2014; Econo-
past). mist 2014; Magnowski 2014).
One reason why the reliability of GDP- Only 22 countries in Africa (less than half
imputed poverty estimates declines the far- of all countries) use base years that are more
ther away the estimate is from the actual recent than 2004. Growing sectors may thus
survey is that the structure of the economy be undercounted, leading to underestimation
changes over time. Every year statistical agen- of GDP, GDP growth, and poverty reduction.
cies collect proxy information on the level of Given that rebasing typically gives greater
production in various sectors. They aggregate weight to nonagricultural sectors, which are
these values assuming the structure of the not as powerful at reducing extreme poverty
economy in the base year. As the structure of as agricultural growth, underestimation of
the economy changes (for example, the agri- poverty reduction is likely to be smaller than
cultural sector shrinks and the service sector underestimation of GDP (Christiaensen,
Demery, and Kuhl 2011; Loayza and Raddatz
FIGURE 1.10 Rebasing increased GDP values in many African 2010).
countries Of the 14 countries that rebased their
GDP in the last 10 years, only 3 reported a
decline in GDP. Some of the upward revisions
Uganda
Tanzania were large, partly because the base year had
Sierra Leone not been changed in many years.
Nigeria Interpolation and extrapolation are neces-
Niger sary to estimate poverty in years in which no
Lesotho
Kenya
survey data are available. Should the imputa-
Ghana tions be based on GDP or private consump-
Ethiopia tion data from national accounts? Private
Congo, Dem. Rep. consumption is preferred, because it captures
Cameroon
a set of goods and services that more closely
Cabo Verde
Burundi mirrors consumption from household surveys
Botswana (see Deaton 2005 for a critique of private con-
–20 –10 0 10 20 30 40 50 60 70 80
sumption as a proxy for household survey
consumption). In practice, however, consider-
Percent change in GDP after rebasing
Number of years between base years ations such as the availability and quality of
GDP and private consumption data and the
Source: Data from national statistical agencies for each country. strength of correlations between data from
THE STATE OF DATA FOR MEASURING POVER T Y 43
national accounts and household surveys In Kenya, for example, where the last
typically influence the choice. PovcalNet uses household survey was conducted in 2005, the
private consumption per capita for interpola- poverty rate associated with the $1.90 pov-
tions, except in Africa, where it uses GDP per erty line was 34 percent. Extrapolating from
capita. the 2005 survey using a real average GDP
For 1991–2012 the average ratio of aver- per capita growth rate of 2.3 percent yields
age consumption per capita from household a poverty estimate of 26 percent for 2012.
surveys to average private consumption per Reducing the growth rate by 0.5 percent-
capita from national accounts (based on 83 age point a year increases the estimate to 28
household surveys in Africa) was 0.86. This percent. The larger the measurement error in
figure is similar to the global average but less GDP growth rates and the older the survey
than the ratio of 1.0 for Africa estimated in data the projections rely on, the larger the
Deaton (2005). The ratio of average con- difference between the “true” and the esti-
sumption per capita from household surveys mated poverty rate using projections.
to GDP per capita for the same sample of sur-
veys was 0.61. This figure is two-thirds of the
global average (0.9) and 60 percent of the 1.0 The Political Economy of
ratio reported in Deaton (2005). The lower Data Production
ratio when using GDP is expected, because
After years of investment in statistics by
GDP includes more than private household
African governments and the international
consumption.
development community, a feeling of disap-
What about growth rates? For a subset of
pointment is noticeable in recent discussions
countries for which two comparable surveys
about the absence of adequate data for pov-
are available, annual per capita growth rates
erty measurement, let alone high-quality
from the household consumption surveys can
data. The issues are not unique to consump-
be compared with the corresponding annual
tion data (box 1.3). Explanations for the
per capita growth of GDP and private con-
delays in the availability of data and quality
sumption from national accounts. Annual
improvements point to inadequate funding,
growth rates are 0.41 percentage points
the limited capacity of national statistical
higher for private consumption per capita
offices, the lack of strategic planning, and
and 1.2 percentage points higher for GDP
administrative cultures. The response of some
per capita than estimates of consumption per
supporters of statistics in the region has been
capita growth from household surveys (based
to ask for more money and more capacity
on a simple country average for each period
building. But there is increasing recognition
for which comparable pairs of survey data
that the problem may be more deeply seated
are available). For Africa overall, without
than lack of money or technical expertise.
restricting to years with comparable surveys,
GDP and private consumption per capita
growth rates from national accounts are Country-Level Factors Associated with
very close, with the GDP per capita growth the Availability, Comparability, and
rate higher by only 0.02 percentage points Openness of Data
on average. This finding suggests that the
performance of GDP in tracking consump- Do richer countries in Africa tend to have
tion from surveys is worse in the subset of more surveys and more surveys that are com-
countries for which comparable surveys are parable? Are countries that receive more aid
available. Overall, using private consump- doing a better job of collecting data, perhaps
tion from national accounts rather than GDP because donors have an interest in show-
to impute poverty when surveys are lacking ing results? Which countries collect more
does not appear to make a significant differ- frequent and comparable consumption sur-
ence. Both sources lead to overestimation of vey data and make the data available to the
the decline in poverty. public?17
44 POVERTY IN A RISING AFRICA
Poor quality and lack of comparability affect many example, Gaddis and Hoogeveen 2015). Although
kinds of data in Africa, not just consumption data. political incentives to show positive results may
One telling sign is the wide variance in indicators drive some of the differences between surveys and
such as health care use, educational enrollment, administrative data (Sandefur and Glassman 2015),
adult literacy, child mortality, and access to water data quality problems also play a role. Estimates of
and sanitation for the same country from different maize yields for Malawi for 2006/07, for example,
surveys (box 3.2, in chapter 3, shows the challenge range from 1,700 kilograms per hectare to more
of tracking adult literacy). Another is the divergence than 2,500 (a difference of almost 50 percent)
between survey and administrative data (see, for (Carletto, Jolliffe, and Banerjee 2015).
This section groups countries in four comparable and open to the public is higher in
ways—by income level, natural resource fragile than in nonfragile countries in Africa.
endowment, geographical location (land- Countries receiving more development
locked versus coastal), and fragility—to iden- aid (as a share of the government budget)
tify patterns. Besides these broad groupings, might be expected to have more and higher-
the analysis draws attention to the role of quality poverty data (defined narrowly as
governance and development aid in data pro- having consumption surveys that are compa-
duction. The upper panel in table 1.2 reports rable), in part because donors are presumably
results for Africa, whereas the bottom panel interested in collecting data with which to
shows results for developing countries in assess whether their aid is having an impact.
other regions. There is no strong evidence that they do. In
Lack of financial resources is generally the non-African sample, there is a negative
considered as a major constraint to statistics correlation between aid and the number of
in Africa. Surprisingly, this is not supported consumption surveys. In the African sample,
by the results. In Africa, middle-income coun- there is no statistically significant relation-
tries neither collect more consumption surveys ship between aid and the number of con-
than low-income countries, nor are the surveys sumption surveys or the share of comparable
they collect more likely to be comparable or surveys. In fact, the more aid a country in
open to the public. Outside of Africa, middle- Africa receives, the less likely it is to open its
income countries collect more consumption surveys to the public.
surveys than low-income countries, but the The lack of positive correlation between
relationship turns insignificant after control- aid and data production in Africa is puzzling.
ling for the share of aid in the budget, political It may be that donors do not explicitly or
freedoms, and government effectiveness. implicitly demand more or better data. Alter-
African countries that are rich in natural natively, the incentives of donors and govern-
resources conduct fewer consumption sur- ments could be misaligned. An example of
veys than non-resource-rich countries in the such misaligned interests is the case in which
region. Both in Africa and in other regions, donors ask and are willing to pay for data
fragile countries collect fewer consumption that are high in quality (small sample, multi-
surveys than nonfragile countries, although topic surveys) though less frequently collected,
in Africa, the statistical significance disap- whereas governments prefer larger samples
pears after controlling for the share of aid in that are representative at lower administrative
the budget, political freedoms, and govern- levels (CGD 2014). National statistical agen-
ment effectiveness. Unexpectedly, in some cies can be caught between the preferences of
specifications, the share of surveys that are donors and those of their governments.
THE STATE OF DATA FOR MEASURING POVER T Y 45
TABLE 1.2 Only a few country characteristics are correlated with the number and share of comparable and
open consumption surveys
Number of Share of consumption Share of consumption
consumption surveys that are surveys that are
Country characteristic surveys comparable open
Africa
Middle-income –0.781 –0.343 –0.072 –0.141 0.068 0.069
Resource-rich –0.869* –1.115* –0.096 0.075 0.016 –0.079
Landlocked 0.794 1.093 0.047 –0.268* –0.056 0.015
Fragile –1.963*** –0.823 –0.084 0.396* 0.169*** –0.010
Log of aid share of government budget –0.146 0.031 –0.076*
Worldwide Governance Indicators
— government effectiveness index 0.363 0.581*** –0.280**
Political rights freedom index –0.165 0.101* –0.022
Outside Africa
Middle–income 4.107** 2.360 0.090 0.146 0.094 0.160
Resource-rich –0.954 –2.755 0.233** 0.166 0.050 –0.067
Landlocked 1.349 3.675** 0.189** 0.122 0.046 –0.001
Fragile –6.236*** –4.766*** 0.025 0.156 0.020 –0.094
Log of aid share of government budget –1.707*** 0.020 –0.003
Worldwide Governance Indicators
— government effectiveness index –1.371 –0.018 –0.073
Political rights freedom index –0.895 0.026 0.009
Unlike aid, good governance is positively a country’s score on the statistical capac-
correlated with higher-quality data in Africa. ity indicator (which measures a country’s
The government effectiveness indicator—one data collection, data availability, and data
of six dimensions of governance tracked in practices) and a country’s safety and rule of
the Worldwide Governance Indicators (WGI) law score (one of the governance indicators
database—is highly correlated with greater tracked in the Ibrahim Index of African gov-
comparability of surveys. However, the indi- ernance) (figure 1.11). Countries with better
cator is negatively correlated with the share scores on safety and rule of law also have
of household surveys that are open. Political higher statistical capacity scores.
openness is (measured by the freedom index)
also positively correlated with a greater share
Political Aspects of the Lack of
of comparable surveys.
Good-Quality Data
Alternative indicators of statistical capac-
ity and governance yield stronger results. The production of statistics is a technically
There is a strong positive correlation between complex task. It involves mobilizing financial
46 POVERTY IN A RISING AFRICA
90
Mauritius
80 Rwanda
Malawi Mozambique
Nigeria Tanzania Senegal South Africa
São Tomé and Príncipe Burkina Faso Lesotho
70 The Gambia Niger
Cabo Verde
Statistical capacity indicator
Guinea-Bissau Gabon
40 Comoros
Equatorial Guinea
30 Eritrea
20 Somalia
0 10 20 30 40 50 60 70 80
Safety and rule of law score
and human resources on a large scale and these studies, autonomous statistical agencies
establishing robust quality control mecha- fail to emerge even where legislation man-
nisms. Pervasive asymmetries of information, dates them because the norms and proce-
which create difficulties if users or buyers dures for making decisions remain informal
seek to verify the quality of the product, ren- (personalized), centralized, and even ad hoc
der the task even more complicated. (Krätke and Byiers 2014). As a consequence,
These challenges partly explain the lack of statistical agencies are unable to produce
high-quality consumption surveys. But gov- timely, good-quality data that are free of
ernments in Africa have been able to meet bias. This failure leaves the agencies vulner-
their capacity needs in performing other able to pressure from local political and well-
activities that are more or equally complex organized advocacy groups (CGD 2014). In
technically, such as delivering antiretroviral addition, where outside financiers tie funding
drugs to people with AIDS and conducting to specific indicators (such as school enroll-
national elections (Hoogeveen 2015). Why ment), both statistical offices and local poli-
have they failed to produce more and better ticians may have incentives to exaggerate
data on living standards? achievements—and produce unreliable data
Several recent reports and papers advance to support them.
the proposition that data are weak because The political environment in many Afri-
of the political preferences of elites (Car- can countries is characterized by ethnic
letto, Jolliffe, and Banerjee 2015; CGD 2014; divisions, fractious alliances, high degrees
Devarajan 2013; Krätke and Byiers 2014; of competition for political leadership and
Hoogeveen 2015; Jerven 2013). According to economic resources, and vague rules of the
THE STATE OF DATA FOR MEASURING POVER T Y 47
game. Many elites in such contexts may take (Hoogeveen 2015). Second, because sup-
a hostile attitude toward reliable and timely porting the patronage network is costly, the
data collection, which they consider a par- opportunity costs of funding high-quality sta-
tisan audit of their performance. This ten- tistics are high in terms of political survival.
dency creates strong incentives to establish Third, poor-quality statistics allow elites
competing and politicized statistical units, to escape accountability, because they can
which in turn leads to fragmentation, dupli- contest bad outcomes. This lack of demand
cation, wastage, and, ultimately, ineffective by and support from the top of the political
agencies. hierarchy may be the most important con-
Political elites may not favor good- straint to changing the poverty data land-
quality statistics for other reasons as well. scape in Africa. However, experiences from
First, where clientelism exists and opportuni- other regions (notably Latin America and the
ties to engage politically are limited, as is the Caribbean) suggest that regional cooperation
case in most African countries, a record of and peer learning, together with international
achievement that can be supported by good- standards and technical guidelines, can still
quality data is unnecessary, because support go a long way in improving the quality and
from a small group of power brokers suffices consistency of existing data (box 1.4).
BOX 1.4 Can donors improve the capacity of national statistics offices? Lessons learned
from MECOVI
The Program for the Improvement of Surveys and • Commitment and ownership were key. The
the Measurement of Living Conditions in Latin national statistical office in each country clearly
America and the Caribbean (known as MECOVI, defined its resources, activities, and work plans.
its acronym in Spanish) was a coordinated effort • Defining the governance structures of the three
led by the Inter-American Development Bank, the sponsoring institutions was important.
United Nations Economic Commission for Latin • Regional training and experience-sharing activi-
America and the Caribbean, and the World Bank ties focused on South-South exchanges were
to provide technical assistance to national statisti- critical.
cal offices to increase their capacity to produce high-
quality household surveys in a sustainable manner. The focused nature of MECOVI’s support for
Launched in 1996, the program was active until household surveys created “islands of efficiency” in
2005. The concept and framework it developed still some of the least-developed statistical offices. Survey
influence household surveys in the region. departments became the “favorite child”—with the
The program has been widely recognized as most funding and the best resources—but the tech-
successful in building the capacity of participating nical nature of the support allowed for significant
countries’ statistical agencies, encouraging regional spillovers to other departments, which benefited
cooperation and peer learning, and establishing the from improvements in areas such as data quality
foundations for the sustainability of household sur- control, questionnaire design, sampling, and data
vey programs. Several lessons emerge from the pro- entry.
gram’s success: Is MECOVI replicable? Some factors that con-
tributed to its success (such as significant interest in
• Planning for the medium term was crucial. The household surveys to measure poverty) cannot be
minimum timeframe for all activities was four reproduced. Others, however, can be. They include
years. close coordination among donors, cooperation
• Concentrating on a limited and focused set of between countries, a long-term view, clearly defined
activities related specifically to household surveys and limited goals, heavy involvement of national sta-
helped obtain objectives. Clearly allocating local tistical offices, well-focused objectives, and secure
funds to surveys and outside resources to techni- funding.
cal assistance rather than data collection led to
sustainability. Contribution by Jose Antonio Mejia-Guerra.
48 POVERTY IN A RISING AFRICA
Reappraising the Information that this method does not pose major issues,
at least when there are no dramatic turn-
Base on Poverty
arounds in the economy or the predictions
The ability to track poverty accurately in are not too far in the future (Christiaensen
Africa hinges on overcoming the many data and others 2012; Douidich and others 2013;
challenges identified in this chapter. Among Kijima and Lanjouw 2003).
these challenges, one set of issues concerns
the availability, comparability, and quality
of consumption data. A second involves the
Using the Engel Curve Approach to
quality and possible biases in the most com-
Avoid the Biases Inherent in the CPI
monly used price data (the CPI) used to mon- Engel’s Law is based on the observation
itor real standards of living. that the share of food in households’ con-
sumption declines as income increases. The
Engel curve method exploits this empiri-
Filling in Years with No Consumption cal regularity to estimate changes in real
Survey incomes based on changes in food budget
One major data challenge is that consump- shares over time, controlling for other fac-
tion surveys are not conducted every year. tors that affect the household’s allocation of
Global or regional poverty estimates fill in its budget between food and nonfood items
gaps between surveys by relying on GDP or (for example, the demographic composi-
private consumption data as an approxima- tion of the household and the relative prices
tion of consumption growth. Additionally, of food and nonfood items) (Costa 2001;
some consumption surveys may be noncom- Hamilton 2001). Inconsistencies between
parable or of dubious quality. If comparabil- changes in real incomes estimated by the
ity and quality concerns result in excluding Engel curve method and measured changes
some surveys, greater reliance will need to be in real incomes (for instance, CPI-deflated
placed on GDP-based imputations. nominal incomes) are regarded as evidence
The alternative to using GDP imputations of measurement bias in the CPI. A drift of
to fill in missing data is to use survey-to- Engel curves to the left, so that over time a
survey (S2S) imputations. This approach given food budget share is associated with
relies on at least one survey with consump- a smaller level of real income, is an indica-
tion (the reference survey), which is used tion that the CPI overstates increases in the
to build a model that can be used to esti- “true” cost of living and that real incomes
mate consumption in other surveys based are increasingly underestimated (Hamilton
on other household traits. The fact that this 2001).
approach can make use of many types of The key identifying assumption of this
nonconsumption surveys, such as the DHS approach is that no unobserved factors affect
and the MICS, is one of its main attractions. the share of the budget spent on food (that is,
The approach can be used to address multi- there are no changes in preferences or price
ple data problems, including low frequency, changes beyond the broad factors for which
lack of comparability, and poor quality. If the model controls). This assumption is not
the model eschews regressors that require trivial and can be violated (because of shifts
adjustments in the cost of living, concerns in preferences toward specific consumer
about the CPI bias can be addressed simulta- durables, such as mobile phones, for exam-
neously (because the imputation is effectively ple). For this reason, although the method
in real terms). The model’s success depends can provide useful indications of CPI bias,
on the stability of the estimated relationship especially when applied to a large number of
between consumption and the household countries, the results should not be overinter-
traits tracked. The evidence mostly suggests preted for any specific country.
THE STATE OF DATA FOR MEASURING POVER T Y 49
Measuring poverty requires setting a level of con- historically been defined as a line that is representa-
sumption below which people are defined as poor. tive of national poverty lines in the poorest coun-
Most developing countries define a national pov- tries, after conversion into a common currency using
erty line based on the cost of a “basic needs” food PPP exchange rates (World Bank 1990; Ravallion,
basket, with some allowance for fundamental non- Datt, and van de Walle 1991; Chen and Ravallion
food requirements (such as clothing and housing). 2010). In 2008 this international line was estimated
Although these national lines have the advantage at $1.25 per capita per day at 2005 prices. In 2015
of measuring poverty according to country-specific the line was updated to $1.90 at 2011 prices based
standards and circumstances, they are not compara- on results from the 2011 PPP round, the value used
ble across countries. For instance, Uganda’s national in this report.
poverty lines are based on a minimum daily calorie Several researchers have proposed alternative pov-
intake of 3,000 kcal per adult, which is much higher erty lines. Ravallion and Chen (2011) and Chen and
than the norms used in neighboring Kenya (2,250 Ravallion (2013) propose “weakly relative” poverty
kcal) and Tanzania (2,200 kcal). Many other salient lines, which combine features of an absolute poverty
differences also undermine cross-country compari- line for the poorest countries with the notion that
sons of national poverty lines. once a country has passed a certain income threshold
To measure poverty at the global or regional level the poverty line should increase with rising per capita
and to compare poverty across countries, it is com- income. Klasen and others (forthcoming) propose
mon practice to apply the same absolute standard an international poverty line of about $1.70 in 2011
in each country to estimate the number of poor. prices, derived using a method that is similar to the
The World Bank’s international poverty line has one used by Jolliffe and Prydz (2015).
50 POVERTY IN A RISING AFRICA
tions associated with the introduction of new on Poverty and Inequality Measures in Niger.”
goods in the U.S. CPI. Policy Research Working Paper 7090, World
13. Unlike national CPIs, PPPs are not intended Bank, Washington, DC.
for assessing changes in country-level prices Barrionuevo, Alexei. 2011. “Inflation, an Old
over time (Feenstra, Inklaar, and Timmer Scourge, Plagues Argentina Again.” New York
2015). Times, February 5.
14. The 14 countries that failed to participate BBC. 2014. “How Nigeria Will Become Africa’s
in the 2000 round were Angola, Burundi, Biggest Economy.” April 4. http://www.bbc
Cameroon, Chad, the Democratic Republic .com/news/world-africa-26873233.
of Congo, Eritrea, Ethiopia, Guinea Bissau, Beegle, Kathleen, Joachim De Weerdt, Jed Fried-
Liberia, Madagascar, Nigeria, Somalia, Sudan, man, and John Gibson. 2012. “Methods
and Togo. The 8 countries that did not par- of Household Consumption Measurement
ticipate in the 2010 round were the Central through Surveys: Experimental Results from
African Republic, Comoros, the Democratic Tanzania.” Journal of Development Econom-
Republic of Congo, Equatorial Guinea, Eritrea, ics 98 (1): 3–18.
Madagascar, Sierra Leone, and Somalia. Sierra Benson, Todd, Charles Machinjili, and Lawrence
Leone conducted a census in late 2015. Kachikopa. 2004. “Poverty in Malawi, 1998.”
15. See https://international.ipums.org Development Southern Africa 21 (3): 419–41.
/international/. Berumen, Edmundo, and Victor A. Beker. 2011.
16. To calculate the poverty rate for years “Recent Developments in Price and Related
between two surveys, one can take the first Statistics in Argentina.” Statistical Journal of
survey and apply the GDP growth rate for- the IAOS 27 (1–2): 7–11.
ward to the interim year, take the second sur- Biemer, Paul, and Lars E. Lyberg. 2003. Introduc-
vey and apply the GDP growth backward to tion to Survey Quality. Wiley Series in Survey
the interim year, and take the average of the Methodology. Hoboken, NJ: John Wiley &
two poverty estimates, weighted by the num- Sons.
ber of years to the first and second survey. Carletto, Calogero, Dean Jolliffe, and Raka
This weighting gives a survey closer to the Banerjee. 2015. “From Tragedy to Renais-
interim year more weight. sance: Improving Agricultural Data for Better
17. Openness in this section is defined as access Policies.” Journal of Development Studies 51
to the public and hence differs from the con- (2):133–48.
cept of availability in the previous discussion, CGD (Center for Global Development). 2014.
which considers only whether data are acces- Delivering on the Data Revolution in Sub-
sible to the report team. Saharan Africa. Final Report of the Data for
18. These challenges feature prominently not African Development Working Group, Center
only in cross-country poverty measurement for Global Development and African Popula-
but also in poverty measurement for a single tion and Health Research Center, Washington,
country using national poverty lines. DC.
Chen, Shaohua, and Martin Ravallion. 2010.
“The Developing World Is Poorer Than We
References Thought, but No Less Successful in the Fight
Against Poverty.” Quarterly Journal of Eco-
Adam, Christopher, David Kwimbere, Wilfred nomics 125 (4): 1577–625.
Mbowe, and Stephen O’Connell. 2012. “Food ———. 2013. “More Relatively-Poor People in
Prices and Inflation in Tanzania.” Working a Less Absolutely-Poor World.” Review of
Paper, International Growth Centre, London. Income and Wealth 59 (1): 1–28.
Almås, Ingvild. 2012. “International Income Christiaensen, Luc, Lionel Demery, and Jesper
Inequality: Measuring PPP Bias by Estimating Kuhl. 2011. “The (Evolving) Role of Agricul-
Engel Curves for Food.” American Economic ture in Poverty Reduction: An Empirical Per-
Review 102 (2): 1093–117. spective.” Journal of Development Economics
Backiny-Yetna, Diane Steele, and Ismael Yacou- 96 (2): 239–54.
bou Djima. 2014. “The Impact of Household Christiaensen, Luc, Peter Lanjouw, Jill Luoto,
Food Consumption Data Collection Methods a nd Dav id S t i fel. 2012 . “S m a l l A re a
THE STATE OF DATA FOR MEASURING POVER T Y 53
Estimation-Based Prediction Methods to Track Africa, Little Change in Poverty at the Grass-
Poverty: Validation and Applications.” Journal roots.” Afrobarometer Policy Brief 1. http://
of Economic Inequality 10 (2): 267–97. www.afrobarometer.org /publications/pp1
Ciccone, Antonio, and Marek Jarociński. 2010. -after-decade-growth-africa-little-change
“Determinants of Economic Growth: Will -poverty-grassroots.
Data Tell?” American Economic Journal: Dykstra, Sarah, Charles Kenny, and Justin Sand-
Macroeconomics 2 (4): 222–46. efur. 2014. “Global Absolute Poverty Fell by
Costa, Dora L. 2001. “Estimating Real Income in Almost Half on Tuesday.” Center for Global
the United States from 1888 to 1994: Correct- Development, Washington, DC. http://www
ing CPI Bias Using Engel Curves.” Journal of .cgdev.org/blog/global-absolute-poverty-fell
Political Economy 109 (6): 1288–310. -almost-half-tuesday.
Dabalen, Andrew, Paul Gubbins, Johan Misti- Easterly, William, and Ross Levine. 1997. “Afri-
aen, and Ayago Wambile. 2015. “Diary ver- ca’s Growth Tragedy: Policies and Ethnic Divi-
sus Recall in Food Consumption: Example sion.” Quarterly Journal of Economics 112
from Kenya.” World Bank, Poverty and Equity (4): 1203–50.
Global Practice, Washington, DC. The Economist. 2014. “Nigeria’s GDP. Step
Deaton, Angus. 1998. Getting Prices Right: What Change: Revised Figures Show that Nige-
Should Be Done? Journal of Economic Per- ria Is Africa’s Largest Economy.” April 12.
spectives 12 (1): 37–46. http://www.economist.com/news/finance-and
———. 2005. “Measuring Poverty in a Grow- -economics/21600734-revised-figures-show
ing World (or Measuring Growth in a Poor -nigeria-africas-largest-economy-step-change.
World).” Review of Economics and Statistics Elbers, Chris, Jean O. Lanjouw, and Peter Lan-
87 (1): 1–19.
jouw. 2003. “Micro-Level Estimation of Pov-
———. 2010. “Price Indexes, Inequality, and the
erty and Inequality.” Econometrica 71 (1):
Measurement of World Poverty.” American
355–64.
Economic Review 100 (1): 5–34.
Elvidge, Christopher D., Paul C. Sutton, Tilot-
Deaton, Angus, and Bettina Aten. 2014. “Try-
tama Ghosh, Benjamin Tuttle, Kimberly E.
ing to Understand the PPPs in ICP2011: Why
Baugh, Budhendra Bhaduri, and Edward
Are the Results So Different?” NBER Working
Bright. 2009. “A Global Poverty Map Derived
Paper 20244, National Bureau of Economic
from Satellite Data.” Computers & Geosci-
Research, Cambridge, MA.
ences 35 (8): 1652–60.
Deaton, Angus, and Olivier Dupriez. 2011. “Spa-
tial Price Differences within Large Countries.” Etang-Ndip, Alvin, Johannes Hoogeveen, and
Working Paper 1321, Woodrow Wilson School Julia Lendorfer. 2015. “Socioeconomic Impact
of Public and International Affairs, Research of the Crisis in North Mali on Displaced Peo-
Program in Development Studies, Princeton ple.” Policy Research Working Paper 7253,
University, Princeton, NJ, and World Bank, World Bank, Washington, DC.
Washington, DC. Feenstra, Robert, Robert Inklaar, and Marcel
Demombynes, Gabriel, and Justin Sandefur. Timmer. 2015. “The Next Generation of the
2014. “Costing a Data Revolution.” Working Penn World Table.” American Economic
Paper 383, Center for Global Development, Review 105 (10): 3150–82.
Washington, DC. Finn, Arden, and Vimal Ranchhod. Forthcom-
Devarajan, Shantayanan. 2013. “Africa’s Statisti- ing. “Genuine Fakes: The Prevalence and
cal Tragedy.” Review of Income and Wealth Implications of Data Fabrication in a Large
59 (S1): S9–S15. South African Survey.” World Bank Economic
Douidich, Mohamed, Abdeljaouad Ezzrari, Roy Review.
van der Weide, and Paolo Verme. 2013. “Esti- Gaddis, Isis, and Johannes Hoogeveen. 2015.
mating Quarterly Poverty Rates Using Labor “Primary Education in Mainland Tanzania:
Force Surveys: A Primer.” Policy Research What Do the Data Tell Us?” In Preparing the
Working Paper 6466, World Bank, Washing- Next Generation in Tanzania: Challenges and
ton, DC. Opportunities in Education, edited by Arun
Dulani, Boniface, Robert Mattes, and Carolyn Joshi and Isis Gaddis. Washington, DC: World
Logan. 2013. “After a Decade of Growth in Bank.
54 POVERTY IN A RISING AFRICA
Garcia-Verdu, Rodrigo. 2013. “The Evolution of IMF (International Monetary Fund). 2003.
Poverty and Inequality in Sub-Saharan Africa Ghana: First Review under the Three-Year
over the Period 1980–2008: What Do We (and Arrangement under the Poverty Reduction
Can We) Know Given the Data Available?” and Growth Facility: Staff Report. IMF Coun-
International Monetary Fund, Washington, try Report 03/395. Washington, DC: IMF.
DC. ———. 2007. Ghana: Article IV Consultation:
Goñi, Edwin, Humberto López, and Luis Servén. Staff Report. IMF Country Report 07/210.
2006. “Getting Real about Inequality: Evi- Washington, DC: IMF.
dence from Brazil, Colombia, Mexico, and Inklaar, Robert, and D. S. Prasada Rao. 2014.
Peru.” Policy Research Working Paper 3815, “Cross-Country Income Levels over Time: Did
World Bank, Washington, DC. the Developing World Suddenly Become Much
Gryna, Frank, and Joseph Juran. 1980. Quality Richer?” Groningen Growth and Development
Planning and Analysis, 2nd ed. New York: Centre, University of Groningen, Netherlands.
McGraw-Hill. Jerven, Morten. 2013. “Comparability of GDP
Günther, Isabel, and Michael Grimm. 2007. Estimates in Sub-Saharan Africa: The Effect of
“Measuring Pro-Poor Growth When Relative Revisions in Sources and Methods since Struc-
Prices Shift.” Journal of Development Eco- tural Adjustment.” Review of Income and
nomics 82 (1): 245–56. Wealth 59 (S1): S16–S36.
Hamilton, Bruce W. 2001. “Using Engel’s Law Jolliffe, Dean Mitchell, and Espen Beer Prydz.
to Estimate CPI Bias.” American Economic 2015. “Global Poverty Goals and Prices: How
Review 91 (3): 619–30. Purchasing Power Parity Matters.” Policy
Harttgen, Kenneth, Stephan Klasen, and Sebas- Research Working Paper 7256, World Bank,
tian Vollmer. 2013. “An African Growth Mir- Washington, DC.
acle? Or: What Do Asset Indices Tell Us about Judge, George, and Laura Schechter. 2009.
Trends in Economic Performance?” Review of “Detecting Problems in Survey Data Using
Income and Wealth 59 (S1): S37–S61. Benford’s Law.” Journal of Human Resources
Hausman, Jerry. 1996. “Valuation of New Goods 44 (1): 1–24.
under Perfect and Imperfect Competition.” Kaminski, Jonathan, Luc Christiaensen, and
In The Economics of New Goods, edited by Christopher L. Gilbert. 2014. “The End of
Timothy F. Bresnahan and Robert J. Gordon, Seasonality? New Insights from Sub-Saharan
209–48. Chicago: University of Chicago Press. Africa.” Policy Research Working Paper 6907,
———. 1999. “Cellular Telephone, New Prod- World Bank, Washington, DC.
ucts, and the CPI.” Journal of Business and Kijima, Yoko, and Peter Lanjouw. 2003. “Poverty
Economic Statistics 17 (2): 188–92. in India during the 1990s: A Regional Perspec-
———. 2003. “Sources of Bias and Solutions to tive.” Policy Research Working Paper 3141,
Bias in the Consumer Price Index.” Journal of World Bank, Washington, DC.
Economic Perspectives 17 (1): 23–44. K la s en , S teph a n , Tat ya n a K r ivobokova ,
Hoogeveen, Johannes, and Nga Thi Viet Nguyen. Friederike Greb, Rahul Lahoti, Syamsul
2015. “Statistics Reform in Africa: Align- Pasaribu, and Manuel Wiesenfarth. Forthcom-
ing Incentives with Results.” Working Paper, ing. “International Income Poverty Measure-
World Bank, Poverty and Equity Global Prac- ment: Which Way Now?” Journal of Eco-
tice, Washington, DC. nomic Inequality.
Hoogeveen, Johannes, Kevin Croke, Andrew KNBS (Kenya National Bureau of Statistics).
Dabalen, Gabriel Demombynes, and Marcelo 2010. The New Consumer Price Index (CPI)
Giugale. 2014. “Collecting High-Frequency Users’ Guide.
Panel Data in Africa Using Mobile Phone Krätke, Florian, and Bruce Byiers. 2014. “The
Interviews.” Canadian Journal of Develop- Political Economy of Official Statistics:
ment Studies 35 (1): 186–207. Implications for the Data Revolution in Sub-
ILO (International Labour Organization). 2013. Saharan Africa.” PARIS21 Discussion Paper 5.
All Countries CPI Descriptions: Method- Lanjouw, Peter, Renzo Massari, and Roy van der
ologies of Compiling Consumer Price Indi- Weide. 2015. “International Poverty Compari-
ces. 2012 ILO Survey of Country Practices. sons: An Imputation-Based Approach.” World
Geneva: ILO. Bank, Washington, DC.
THE STATE OF DATA FOR MEASURING POVER T Y 55
Ley, Eduardo. 2005. “Whose Inflation? A Char- Unit Working Paper 07–129, University of
acterization of the CPI Plutocratic Gap.” Cape Town, Rondebosch, Cape Town, South
Oxford Economic Papers 57 (4): 634–46. Africa.
Loayza, Norman, and Claudio Raddatz. 2010. Phelps, Glenn, and Steve Crabtree. 2013. “More
“The Composition of Growth Matters for Than One in Five Worldwide Living in
Poverty Alleviation.” Journal of Development Extreme Poverty.” Gallup. http://www.gallup
Economics 93 (1): 137–51. .com/poll/166565/one-five-worldwide-living
Magnowski, Daniel. 2014. “Nigerian Economy -extreme-poverty.aspx.
Overtakes South Africa’s on Rebased GDP.” Pinkovskiy, Maxim, and Xavier Sala-i-Martín.
April, Bloomberg. 2014. “Africa Is on Time.” Journal of Eco-
Majumder, Amita, Ranjan Ray, and Kompal nomic Growth 19 (3): 311–38.
Sinha. 2012. “Calculating Rural-Urban Food ———. 2015. “Lights, Camera…, Income! Esti-
Price Differentials from Unit Values in House- mating Poverty Using National Accounts,
hold Expenditure Surveys: A Comparison with Survey Means, and Lights.” Staff Report 669,
Existing Methods and A New Procedure.” Federal Reserve Bank of New York.
American Journal of Agricultural Economics PovcalNet. Database. http://iresearch.worldbank
94 (5): 1218–35. .org/PovcalNet/.
McCulloch, Neil, Julian Weisbrod, and Peter Prais, Sigbert. 1959. “Whose Cost of Living?”
Timmer. 2007. “Pathways out of Poverty dur- Review of Economic Studies 26 (2): 126–34.
ing an Economic Crisis: An Empirical Assess- Ravallion, Martin. 2014. “An Exploration of the
ment of Rural Indonesia.” Policy Research International Comparison Program’s New
Working Paper 4173, World Bank, Washing- Global Economic Landscape.” NBER Working
ton, DC.
Paper 20338, National Bureau of Economic
Mkenda, Adolf F., and Wilhelm Ngasamiaku.
Research, Cambridge, MA.
2009. “An Analysis of Alternative Weighting
Ravallion, Martin, and Shaohua Chen. 2011.
System on the National Price Index in Tan-
“Weakly Relative Poverty.” Review of Eco-
zania: The Implication to Poverty Analysis.”
nomics and Statistics 93 (4): 1251–61.
Botswana Journal of Economics 6 (10): 50–70.
Ravallion, Martin, Gaurav Datt, and Dominique
Muller, Christophe. 2008. “The Measurement of
van de Walle. 1991. “Quantifying Absolute
Poverty with Geographical and Intertemporal
Poverty in the Developing World.” Review of
Price Dispersion: Evidence from Rwanda.”
Income and Wealth 37 (4): 345–61.
Review of Income and Wealth 54 (1): 27–49.
Sandefur, Justin, and Amanda Glassman. 2015.
Mveyange, Anthony. 2015. “Night Lights and
Regional Income Inequality in Africa.” Depart- “The Political Economy of Bad Data: Evidence
ment of Business and Economics, University of from African Survey and Administrative Sta-
Southern Denmark, Odense, Denmark. tistics.” Journal of Development Studies 51
NBS (National Bureau of Statistics). 2009. House- (2): 116–32.
hold Budget Survey 2007: Final Report. Dar Schultze, Charles, and Christopher Mackie. 2002.
es Salaam, Tanzania. At What Price? Conceptualizing and Measur-
Noor, Abdisalan M., Victor A. Alegana, Peter ing Cost-of-Living and Price Indexes. Wash-
W. Gething, Andrew J. Tatem, and Robert W. ington, DC: National Academy Press.
Snow. 2008. “Using Remotely Sensed Night- UN (United Nations). 2009. Practical Guide
Time Light as a Proxy for Poverty in Africa.” to Producing Consumer Price Indices. New
Population Health Metrics 6: 5. http://www York: United Nations. http://www.unece.org
.biomedcentral.com/content/pdf/1478-7954 /fileadmin/DAM/stats/publications/Practical
-6-5.pdf. _Guide_to_Producing_CPI.pdf.
Okidi, John, and Vincent Nsubuga. 2010. Infla- World Bank. 1990. World Development Report
tion Differentials among Ugandan House- 1990: Poverty. Washington, DC: World Bank.
holds: 1997–2007. Research Series 72, Eco- ———. 2007. “Underreporting of Consumer Price
nomic Policy Research Centre, Kampala, Inflation in Tanzania 2002–2006.” World
Uganda. Bank Policy Note, Washington, DC.
Oosthuizen, Morné. 2007. “Consumer Price ———. 2012. Niger: Investing for Prosperity, a
Inflation across the Income Distribution in Poverty Assessment. Washington, DC: World
South Africa.” Development Policy Research Bank.
56 POVERTY IN A RISING AFRICA
———. 2013. “Burkina Faso: A Policy Note: Pov- Research Report. Washington, DC: World
erty Trends and Profile for 2003–2009.” World Bank.
Bank, Washington, DC. ———. 2015b. “Tanzania Mainland Poverty
———. 2014. Purchasing Power Parities and the Assessment.” World Bank, Washington, DC.
Real Size of World Economies: A Comprehen- ———. 2015c. The Socio-Economic Impacts of
sive Report of the 2011 International Com- Ebola in Sierra Leone: Results from a High-
parison Program. Washington, DC: World Frequency Cell Phone Survey: Rounds 1–3.
Bank. Washington, DC: World Bank.
———. 2015a. A Measured Approach to End- Young, Alwyn. 2012. “The African Growth
ing Poverty and Boosting Shared Prosperity: Miracle.” Journal of Political Economy 120
Concepts, Data, and the Twin Goals. Policy (4): 696–739.
Revisiting Poverty Trends
2
T
his chapter examines trends in pov- The third section provides a brief profile of
erty in Africa using household con- the poor, based on country typology, location
sumption, generally the variable of within a country (urban/rural), and gender.
choice for tracking poverty there.1 In many The fourth section examines the dynamics of
African countries, such data are collected poverty—the movement of people into and
infrequently, are of poor quality, or are not out of poverty. The last section summarizes
comparable across surveys. How these data the chapter’s main findings.
challenges are dealt with often underlies dif-
fering views about Africa’s progress toward
reducing poverty, including the Millennium Trends Using Comparable
Development Goal (MDG) target of halving and Better-Quality Data
poverty by 2015.2 According to the latest estimates in Povcal-
The chapter is divided into five sections. Net, the share of Africa’s population living
The first section looks at whether correct- below the international poverty line of $1.90
ing for the comparability and quality of data declined from 57 percent in 1990 to 43 per-
changes the view of how poverty has evolved cent in 2012. This rate of poverty reduction
in Africa. It focuses on region-wide trends, was the slowest among the major regions of
with specific countries featured only for illus- the world.
trative purposes. The results are benchmarked Consensus about the accuracy of these
against the World Bank’s PovcalNet, the most figures is lacking, because of debate over the
comprehensive repository for poverty data quality of the data (Pinkovskiy and Sala-i-
for calculating regional and global trends. Martín 2014; Young 2012). What does the
Scrutiny of data quality and comparability trend in poverty look like if known data com-
entails excluding some data, which leads to parability issues across surveys within coun-
reliance on imputations to obtain long-term tries and quality problems are addressed?3
trends. The second section checks whether Figure 2 .1 shows four trends. T he
these imputations drive the alternative trends PovcalNet line shows changes in poverty
reported here, by reporting poverty trends based on all surveys in its database. These
using alternative methods and assumptions. estimates are population-weighted poverty
This chapter was written with Nga Thi Viet Nguyen
rates from 47 of Africa’s 48 countries. Of
and Shinya Takamatsu. the 47 countries for which poverty estimates
57
58 POVERTY IN A RISING AFRICA
FIGURE 2.1 Adjusting for comparability and quality changes the have been computed, one or more surveys are
level, depth, and severity of poverty available for 43.4 For each of these countries,
a poverty rate is estimated from actual survey
a. Poverty rate data (regardless of comparability or quality).
70 For years without surveys, per capita growth
60 in gross domestic product (GDP) is used to
simulate consumption growth between sur-
50
vey years (see World Bank 2015b for a dis-
40 cussion of the method).
Percent
30
Additional estimates are based on only
comparable surveys, comparable and good-
20
quality surveys (as described in chapter
10 1, and henceforth referred to as corrected
data), and comparable and good-quality
0
1990 1993 1996 1999 2002 2005 2008 2010 2011 2012 surveys without Nigeria. 5 For the subset of
comparable surveys identified in each coun-
b. Depth
try, the imputation methodology used in
30 PovcalNet, which relies on growth in GDP
per capita, was applied to fill gaps between
25 surveys. By design, this method relies on
fewer surveys and more imputed estimates
20
of poverty.
Another set of estimates goes a step farther
Percent
15
by taking quality as well as comparability
10 into account. Starting from the subset of sur-
5 veys deemed comparable, this estimate drops
surveys of poor quality. This step affected five
0 countries (Burkina Faso, Mozambique, Nige-
1990 1993 1996 1999 2002 2005 2008 2010 2011 2012 ria, Tanzania, and Zambia), which together
represent 30 percent of Africa’s population.
c. Severity Detailed descriptions of the quality of the sur-
18 veys were used to determine which to exclude
16 (Alfani and others 2012; World Bank 2012,
14 2013, 2014b, 2015c). For Nigeria, home to
18 percent of the population of Africa, this
12
implied dropping the two comparable surveys
Percent
10
(both of poor quality), and replacing them by
8
one deemed of good quality (at the expense of
6
greater reliance on imputation). The last set of
4 estimates is based on a sample that corrects
2 for comparability and quality and excludes
0 Nigeria.
1990 1993 1996 1999 2002 2005 2008 2010 2011 2012
Correcting only for comparability shows
slightly higher regional poverty rates between
PovcalNet
1990 and 1999 but little change in trends
Comparable surveys
compared with the PovcalNet estimates.
Comparable and good-quality data
Comparable and good-quality data
Correcting for quality and comparability
(without Nigeria) leads to a change in level after about 2002.
Using these surveys only, the estimate of pov-
Source: World Bank Africa Poverty database. erty in Africa is 6 percentage points lower
REVISITING POVERTY TRENDS 59
(37 percent instead of 43 percent) than the Measures of the depth and severity of pov-
PovcalNet estimate in 2012. Nigeria accounts erty follow trajectories similar to the pov-
for a large fraction of this change. The fourth erty rate (see panels b and c of figure 2.1).
estimate, based on surveys that were both In 1990 the depth of poverty was 25 per-
comparable and of good quality and excludes cent using PovcalNet (compared to 23 per-
Nigeria, shows that poverty declined from cent using corrected data), indicating that
about 55 percent to 40 percent (15 percent- resources equivalent to 25 percent of the
age points), compared to the 14 percentage- value of the poverty line per person would
point decline (from 57 percent to 43 percent) have been needed to eliminate the shortfall in
revealed by the PovcalNet data. consumption among the poor. By 2012 this
The headcount poverty rate is a simple share had fallen to 14–17 percent, depending
measure of the share of the population liv- on the sample used. The severity of poverty
ing below the poverty line; it does not dis- also declined, falling from about 12 per-
tinguish among the poor. Depth of poverty cent in 1990 (compared to 14 percent using
captures the amount of shortfall in consump- PovcalNet) to 7–8 percent using corrected
tion among the poor as a share of the poverty data (9 percent with PovcalNet).6
line. Severity of poverty adds more weight to The trends based on corrected data raise
the shortfall of the poorest and thus captures two major concerns, both of which poten-
inequality among the poor. tially bias the results in a way that may
BOX 2.1 Adjusting the data for Nigeria has a huge effect on estimates of poverty
reduction
Nigeria is home to 18 percent of Africa’s population. difference also affects changes in poverty. The com-
As a result, it has a major effect on regional levels bination of using imputations and the GHS-Panel
and trends in poverty. instead of the NLSS leads to significant changes in
Nigeria has conducted household budget sur- Nigeria and regional poverty trends.
veys since the early 1990s, but design changes made The confidence one can attach to the revised
them noncomparable. Since 2003 it has measured regional series depends crucially on the acceptance
poverty by conducting two Nigeria Living Standard of the trends in poverty in Nigeria that are obtained
(NLSS) and two General Household Survey Panel based on the GHS-Panel and GDP growth projec-
(GHS-Panel) surveys. Official national poverty tions. The recent exercise in rebasing the GDP lends
measures and PovcalNet use the NLSS 2003/04 support to the use of the GHS-Panel data, which
and 2009/10. better describe the link between growth and pov-
The NLSS and GHS-Panel are not comparable, erty, urban and rural gaps, the spatial distribution of
and they differ in the quality of implementation poverty (World Bank 2014b), and Nigeria’s perfor-
(World Bank 2014c). The poverty estimates and mance relative to its peers. The implied poverty rates
trends from the two sources also differ sharply. At in the GHS-Panel suggest that Nigeria is no longer
the $1.90 poverty line (2011 PPP), poverty rates the poorest country in West Africa (as implied by
from the NLSS 2009/10 (53 percent) are twice as the NLSS).
high as rates obtained from the GHS-Panel 2010/11 Additional evidence in support of the corrected
(26 percent). The NLSS shows no change in pov- data comes from the use of survey-to-survey (S2S)
erty between 2003/04 and 2009/10, whereas the imputations (discussed later in the chapter) rather
GHS-Panel suggests a decline from 26 percent in than GDP projections to look at trends. The impu-
2009/10 to 23 percent in 2012/13. Using the GHS- tations using GDP growth suggest that the poverty
Panel instead of the NLSS changes poverty levels in rate in Nigeria dropped by 12 percentage points
Nigeria—and therefore the region. between 2004 and 2012. The S2S imputations using
Nigeria’s GDP growth rates were higher in the GHS-Panel consumption suggest a 10 percentage
2000s than in the 1990s. Because GDP is used to fill point decline for the same period (Corral, Molini,
in data gaps for years when there are no surveys, this and Oseni 2015).
60 POVERTY IN A RISING AFRICA
TABLE 2.1 Addressing quality and comparability reduces the surveys available for poverty monitoring
(percent of total data points available from surveys)
FIGURE 2.2 Analysis based only on comparable surveys suggests that poverty reduction in Africa was faster than
previously thought
4.9
5
Annual percentage point
4
3.1
reduction in poverty
3 2.4
1.8 1.8 1.8 1.9 2.0
2 1.5 1.6 1.7 1.5 1.5 1.6
1.2 1.2 1.3 1.2
1 0.7 0.8
0.3 0.5
0
–0.1 0 0
–1
–1.0 –1.0
–2 –1.5
–2.3
un s
es
in ad
M an el)
ag ia
Cô me ar
d’ n
Se oire
Ni au gal
(N s
M go
Si az i
M Le d
rit e
So G ia
h na
Na frica
ia
Ni ngo Rw na
ria em da
-P p.
am nia
Et que
Ug pia
Bu C da
r 2 ric u ca
ric ou s
co rie
To )
Sw alaw
S
ria tiu
Af n c trie
au on
tri
as
te roo
ra n
ad mb
an
Bo ib
HS e
c
LS
fo Af co fri
T n
rk h
ut ha
a
ge , D an
an nt
er ila
Ca as
oz za
ne
(G . R
aF
a
ge ri
m
w
a
Iv
bi
A
ta all ing l A
a n
hi
Za
t s
da or op Al
M
M
Co
ed a f el
7
ct at ev
rre d d D
Co cte
rre
Co
Mean poverty reduction among countries that reduce poverty
Mean poverty reduction of all countries
Source: Data for individual African countries are from World Bank Africa Poverty Database. Developing country data are from PovcalNet.
Note: Positive values denote a reduction in poverty, while negative values denote an increase. The survey years are as follows: Botswana (2002 and 2009), Burkina Faso (1998 and
2003), Cameroon (2001 and 2007), Chad (2003 and 2011), Democratic Republic of Congo (2004 and 2012), Côte d’Ivoire (2002 and 2008), Ethiopia (1999 and 2010), Ghana (1998 and
2005), Madagascar (2001 and 2010), Malawi (2004 and 2010), Mauritania (2000 and 2008), Mauritius (2006 and 2012), Mozambique (2002 and 2008), Namibia (2003 and 2009), Nigeria
(2003 and 2009 [Nigeria Living Standards Survey] and 2010 and 2012 [GHS-Panel]), Rwanda (2000 and 2010), Senegal (2005 and 2011), Sierra Leone (2003 and 2011), South Africa
(2005 and 2010), Swaziland (2000 and 2009), Tanzania (2000 and 2007), Togo (2006 and 2011), Uganda (1999 and 2012), and Zambia (1998 and 2006). Nigeria GHS-Panel data are
shown but were not used to estimate averages. Data on all Africa and developing countries are for 1999–2012. “Corrected data for 27 African countries” reports poverty estimates
based on comparable and good-quality data for countries with data from at least two comparable surveys, excluding Nigeria. “Corrected data for all African countries” shows aver-
age based on comparable and good-quality data for all of Africa.
reduced poverty by more than 1 percentage reduction from actual data remains about
point a year. On average these 24 countries 1 percentage point a year. By contrast, the
achieved an annual rate of poverty reduction rate of poverty reduction based on only com-
of 0.92 percentage points. In contrast, the parable and good-quality surveys and GDP
corrected data suggest an average annual pov- imputations to fill the data gaps for both
erty reduction rate of 0.8 percentage points all countries and the 27 countries for which
between 1990 and 2012 for Africa. Annual comparable data are available is about 1.6
poverty reduction for the developing world as percentage point a year in the 2000s—a
a whole, using uncorrected data, is 1.5 per- much higher rate of poverty reduction than
centage points. the actual data imply. The corrected data are
Except for a few countries (Ethiopia, heavily influenced by the data on Nigeria.
Ghana, and Uganda, where the earliest sur- Excluding Nigeria reduces the implied pov-
veys started in the first half of the 1990s), erty reduction obtained from corrected data
most of these comparable surveys were con- to 1.2 percentage point a year, which is closer
ducted during the 2000s. Limiting the analy- to the poverty reduction rate of 1 percentage
sis to surveys in the 2000s does not change point a year based on actual data if Nigeria is
the results: the implied average poverty excluded for the entire period.
62 POVERTY IN A RISING AFRICA
These 24 countries represent 75 percent economics and statistics. They have been
of the total population of Africa and 83 per- used to recover missing values of one or
cent of its poor. The list includes large and more variables because respondents did not
small countries, some that fell into conflict provide the needed information, the data
between surveys, coastal and landlocked were corrupted, or errors that cannot be
countries, and countries with different levels ignored arose during the measurement of
of resource endowments. The experience of variables. S2S imputations are attractive in
these countries arguably captures the experi- Africa because they can address the chal-
ences of countries in the region. The poverty lenges posed by the noncomparability of
estimates suggest that the average annual surveys, the poor quality of consumption
poverty reduction from these surveys is rea- data, the low frequency of consumption
sonably close to the rate obtained from an surveys and the paucity of poverty data
appropriate comparison of poverty estimates points, and missing or poor-quality price
based on GDP imputations. data. Validation of S2S imputation against
actual poverty trends based on reliable data
suggests that the method can track poverty
Survey-to-Survey Imputation as a
well, provided there are no major economic
Robustness Check
turnarounds and the periods are not too
Instead of using GDP growth rates to fill far apart (Christiaensen and others 2012;
gaps in consumption survey data, the S2S Douidich and others 2013).
imputation takes advantage of nonconsump- Figure 2.3 illustrates how an S2S imputa-
tion household surveys. Survey-based impu- tion can be used to estimate a poverty trend
tation techniques have a long tradition in and why accounting for comparability is
FIGURE 2.3 Survey-to-survey imputation and evidence from comparable surveys provide similar
estimates of poverty
2
Change in poverty (percentage points)
–1
–2
–3
–4
Comparable surveys Noncomparable surveys
–5
2
10
05
10
01
12
05
01
04
06
10
12
10
–1
01
–0
5–
20
20
20
2–
20
20
20
1–
6–
20
1–
02
03
–2
00
00
00
1–
7–
1–
20
20
98
97
93
20
,2
99
99
,2
99
l, 2
99
i, 2
9
19
19
19
a,
o,
a,
ca
ea
,1
), 1
), 1
i, 1
,1
al
ga
an
bi
as
n,
a,
r,
ri
M
in
da
ia
aw
ne
m
se
se
ca
aF
Af
ny
Gh
oo
an
Gu
an
Za
ba
ba
as
Se
al
Ke
th
in
er
nz
Ug
ag
M
rk
u
10
Ta
99
So
Bu
ad
Ca
20
(1
M
(
ia
ia
op
op
hi
hi
Et
Et
Actual S2S
important. It presents poverty estimates for FIGURE 2.4 Survey-to-survey imputations suggest that poverty in
10 countries in which surveys are not com- Africa is lower than household survey data indicate
parable and 4 in which the surveys are com-
parable. The results lead to two conclusions. 65
First, for comparable surveys, imputed and
actual changes are in the same direction, and
the estimates are similar in magnitude (in 3
data because they rely too heavily on GDP associated with the food and fuel crises that
imputations. occurred during the period under study
(1990–2012) (box 2.2).14 CPI basket weights
typically reflect the expenditure patterns of
The Role of Price Adjustments in
wealthier households, which spend a much
Measuring Poverty
smaller share of their budgets on food than
Consumer price indexes (CPIs), which are the average poor family does. If food prices
used to estimate real consumption in 2011 increase much more quickly than general
(the base year of the poverty line) may not consumer prices, CPIs may underestimate
have taken full account of the inflation the true inflation experienced by the poor. In
BOX 2.2 How do spikes in food prices affect the measurement of poverty?
Poverty estimates indicate that poverty reduction For the longer period (2002–12, panel b), food
accelerated beginning around 2002. One concern CPIs increased more quickly than the general CPIs in
with this finding is that the CPIs in the 2000s may Burkina Faso, Ethiopia, Mozambique, and Uganda
have understated the sharp rise in food prices, espe- and less quickly in Ghana, Malawi, and Zambia.
cially for major staples such as maize, wheat, and In Nigeria—which, because of its large population,
rice, observed in 2007/08 and 2011. has a substantial effect on the regional trend—the
A comparison of trends in food prices and the two inflation rates almost coincide. It is possible that
overall CPI in African countries with long-run CPI these patterns would look different if price deflators
series highlights the effect of the 2007/08 food price that are more tailored to the consumption patterns
crisis. Most countries experienced significantly of the poor than the food CPIs had been used. But
higher food price inflation over this period than over the evidence suggests that the broad increase in pov-
the 2000s as a whole. Between 2007 and 2009, food erty reduction after 2002 is not merely a reflection
CPIs increased more quickly than general CPIs in of a failure to account for rapidly rising food prices.
seven of nine countries (figure B2.2.1, panel a).
FIGURE B2.2.1 Food inflation does not always exceed overall inflation
a. Overall and Food CPI, 2007–09 b. Overall and Food CPI, 2002–12
30 30
Annual inflation rate (percent)
25 25
20 20
15 15
10 10
5 5
0 0
ea
aa
Za a
Ug a
Ni e
Za a
Ca aso
ia
a
Et n
am i
Ug a
Ca aso
d
ia
oz alaw
Et n
am i
ri
a
d
qu
qu
oz aw
bi
bi
oo
ri
an
op
an
oo
an
ge
op
an
ge
aF
m
m
bi
F
bi
Gh
er
l
hi
Gh
er
Ni
hi
M
a
in
m
in
rk
rk
Bu
Bu
Sources: Databases of the International Labour Organization (http://laborsta.ilo.org/STP/guest) and the Food and Agriculture Organization (http://faostat3.fao.org
/download/P/CP/E).
a. Series for Mozambique and Zambia run only through 2011.
REVISITING POVERTY TRENDS 65
this case, the rate of poverty reduction will to urban areas, only urban households are
be overstated. There are other reasons, out- used for these estimations, except for Ethio-
lined in chapter 1, why CPIs may not accu- pia, Mauritius, Nigeria, and Rwanda.17 The
rately depict the inflation experience of the estimation of the CPI bias in Africa fol-
poor. If CPIs do not adjust correctly for price lows the methodology outlined in Gibson,
increases, the measurement of poverty will be Stillman, and Le (2008) (for a review of
flawed. An underestimated (overestimated) these methods, see also Gaddis 2015). This
CPI will overstate (understate) poverty reduc- approach is based on the assumptions that
tion. In terms of the level of poverty, when (a) food and nonfood consumption are mea-
adjusting a survey from before 2011 for- sured consistently across surveys and without
ward to 2011, if the CPI is overestimated, serious measurement error and (b) prefer-
the poverty rate in the year before 2011 will ences remain stable over time.
be underestimated. When adjusting a survey Unobserved time-varying factors that are
from after 2011 back to 2011, an overesti- correlated with the food budget share could
mated CPI will cause the poverty rate in the also potentially bias the estimates. They may
year after 2011 to be overestimated. also explain why the Engel curve method has
How does correcting for these biases been shown to perform poorly when compar-
affect poverty rates (and trends)? There are ing cost of living differences across space—
two broad approaches to investigating biases such as regions or provinces (Gibson, Le, and
in the CPI and reassessing poverty rates and Kim 2014).
trends. One approach uses item-level price
data (for example, unit record data from
the consumer price collection) to check for FIGURE 2.5 Correcting for CPI bias suggests that poverty
aggregation errors, experiment with different reduction is underestimated
weights, and perform more detailed demand
estimations to approximate the relative con-
Tanzania 2008–12 –5.6
tribution of various sources of CPI bias (see,
South Africa 2005–10 –3.4
for instance, Boskin and others 1996; Diew-
ert 1998; Hausman 2003). Nigeria 2003–09 –2.9
An alternative approach exploits the Ethiopia 2004–10 –2.0
empirical regularity that food budget shares Madagascar 2001–10 –2.0
decline as consumption increases—that is, Congo, Dem. Rep. 2004–12 –1.7
they act according to Engel’s law.15 Accord- Togo 2006–11 –1.5
ingly, provided that nominal consumption Rwanda 2005–10 –1.3
has been measured consistently over time, Tanzania 2000–07 –1.1
differences in the food budget share among Senegal 2005–11 –1.0
demographically similar households with the Côte d’Ivoire 2002–08 –0.9
same level of consumption at different points Mozambique 2002–08 –0.8
in time indicate the CPI’s mismeasurement Mauritius 2006–12 0.0
of the true change in cost of living (Costa Nigeria 2011–13 0.0
2001; Hamilton 2001). Any wedge between Cameroon 2001–07 0.0
the estimated changes in real consumption Burkina Faso 1998–2003 0.1
derived from demand functions for food (that Uganda 2009–12 2.0
is, Engel curves) and measured changes in Ghana 2005–12 5.3
real consumption (that is, CPI-deflated nomi-
nal consumption) is attributed to CPI mea- –8.0 –6.0 –4.0 –2.0 0 2.0 4.0 6.0
surement bias in this approach. Difference in poverty reduction resulting from correction of CPI bias
The Engel approach is applied to compa- (percentage points per year)
rable surveys from 16 countries to estimate
Source: World Bank Africa Poverty database.
the direction and magnitude of CPI bias.16 Note: A negative value indicates that the CPI underestimates the reduction in poverty (or in few
Because CPI data collection is often restricted cases, overestimates the increase in poverty). A positive value denotes the opposite.
66 POVERTY IN A RISING AFRICA
Figure 2.5 displays estimates of the extent contrast, the Engel curve captures inflation
of CPI bias for pairs of surveys. The later rates of a household whose position in the dis-
year in each pair was used as the reference; tribution is unknown. Some of the measured
the implied poverty rate that corrects for the difference in poverty reduction between the
size of the bias during the period of the sur- two methods may reflect differential growth
veys was computed for the other year. Esti- in the inflation of the households represented
mates of poverty reduction from the Engel by these deflators.20 The large differences in
curve and the CPI are then compared. poverty rates the two methods yield in some
The Engel curve estimates suggest that countries suggest that more work is needed
CPIs in Africa tend to overstate increases in to corroborate the Engel curve estimation
the cost of living.18 In 11 countries, the cost of results. Ideally, such work would extend
living for an average urban household rose by these overall bias estimates by examining the
less than what is suggested by the official CPI. CPI product list using the method suggested
(For a detailed discussion of the estimation by Hausman (2003).
see Dabalen, Gaddis, and Nguyen 2015). The
difference in annualized poverty reduction
Asset Ownership as a Measure of
between the Engel method and CPI updates
Poverty Trends
ranges from about 5 percentage points in
Ghana to almost –6 percentage points in Given the low frequency and measurement
Tanzania. Burkina Faso, Ghana, and Uganda problems common to consumption surveys
are the three countries whose estimated dif- (discussed in chapter 1), might other sources
ferences are positive, although for Burkina of data offer a substitute for consumption?
Faso the difference is not statistically signifi- Some efforts have focused on using asset
cant. This means that CPI updates in these ownership as an alternative measure of
countries understate the increase in the cost consumption change and a means to track
of living (and therefore overstate poverty poverty.
reduction) for the period studied. The size of Assets as a proxy for consumption or
the divergence in Nigeria depends on which income have several advantages that have
survey is used. The poorer-quality survey made them popular since the 1990s.21 First,
(Nigeria Living Standard Surveys 2003/04 nonconsumption household surveys contain-
and 2009/10) yields a 3 percentage point dif- ing asset information covering many coun-
ference in poverty reduction between the CPI tries and years, such as the Demographic and
and Engel methods, implying that the CPI Health Survey, have become available. Data
overstates cost of living and therefore under- on assets are easier to collect than data on
states poverty reduction; the higher-quality consumption, which require detailed ques-
survey (Nigeria GHS-Panel 2011 and 2013) tionnaires. Second, the asset approach avoids
yields no difference between the two methods. the need to monetize values, which requires
The 16 countries in figure 2.5 represent 70 price data.
percent of the African population. The results Although they find that assets have a
imply that on average, CPI updates understate robust correlation with nonincome dimen-
poverty reduction by 1 percentage point a sions of poverty (including nutrition, health
year.19 They also provide prima facie evidence care use, educational enrollment, fertility,
that poverty in many African countries may and child mortality), Filmer and Scott (2012)
have declined more quickly than indicated by show that the correlation between consump-
trends in international poverty rates. tion and asset indexes is weak. Assets and
These estimates come with an important asset indexes are more strongly correlated
caveat, however. The Engel curve estimates with consumption in urban areas and in set-
do not necessarily imply that CPIs provide tings in which transitory shocks are mild,
biased estimates of general inflation. CPIs, measurement error in consumption is lim-
by design, capture inflation faced by house- ited, and the share of privately consumed
holds in the 70th or even 80th percentiles. By goods, such as food, in consumption is small.
REVISITING POVERTY TRENDS 67
These factors are likely to lead to a weaker equally poor may accumulate different levels
correlation between assets and consumption of the same asset because of various factors,
in Africa than in other settings. Howe and including conflict, trade restrictions on the
others (2009) assess the correlation between asset, or poor provision of a public good that
asset indexes and expenditure in 36 data- is highly complementary to the asset (unreli-
sets and conclude that the indexes are a poor able electricity would reduce the acquisition
proxy for consumption data. of refrigerators, for example). Third, because
Assets have been frequently used to rank assets are stocks, having more assets reflects
households in country-level analysis and then both current and past consumption or
differentiate households in the poorest and income. Fourth, the extent to which house-
richest quintiles. Can assets also be used to holds opt to accumulate assets may be a func-
assess poverty levels and trends? There are tion of alternative means of saving or storing
several methodological concerns about using wealth, which varies across countries.
assets to monitor poverty. First, households A fifth concern is the challenge of setting a
may increase their assets in the absence of poverty line based on asset indexes. For con-
consumption growth (“asset drift”) (Hartt- sumption measures, there is a cost-of-basic
gen, Klasen, and Vollmer 2013). Second, the needs anchor. In contrast, there is no consen-
ability to accumulate assets varies substan- sus on the minimum set of assets needed to
tially across countries for reasons that may meet basic needs. Moreover, there is no con-
have little to do with the ability to purchase sensus on how to aggregate assets (box 2.3).
them. Populations in two countries that are The choice of which assets to include in the
Three indexes measure asset ownership. wealth score is rescaled to range between 0 and 100.
If a new asset or a new country is introduced, the
The DHS Wealth Index index needs to be recalibrated. Although not iden-
The Demographic and Health Survey (DHS) wealth tical, this index is highly correlated with the DHS
index is the most commonly used asset index. It is wealth index. Its correlation with consumption is
constructed from a large set of household assets and low (0.5) for the two countries for which it was eval-
utility services in the DHS and includes country- uated (Malawi and Niger) for this report.
specific items (Rutstein and Johnson 2004). This
index is a standardized score with a mean of 0 and a The Comparative Wealth Index
standard deviation of 1. Principal component anal- The comparative wealth index aims to make existing
ysis is used to assign the indicator weights to each country-specific DHS wealth indexes comparable
asset or service. Because the number of assets or with one another, to enable trend analysis within
utility services and the weights change over time and and across countries (Rutstein and Staveteig 2014).
across countries, this index is not comparable across The approach adjusts households’ country-specific
surveys within a country, over time, or between DHS wealth index based on the country-level rela-
countries. tionship between some “unsatisfied basic needs” and
ownership of four assets (car, refrigerator, fixed tele-
The International Wealth Index phone, and television) relative to a reference coun-
To circumvent concerns about varying the assets try. For each survey, thresholds for ownership of the
included in an asset index across countries and assets are determined using a logistic regression, and
years, the international wealth index is constructed unsatisfied basic needs are estimated based on the
from a small set of common assets. Principal compo- cumulative distribution of unsatisfied needs. These
nent analysis is used to determine the asset weights thresholds are regressed against the thresholds for
(Smits and Steendijk 2015). Countries are weighted the reference country and the coefficients used to
by the square root of population size; the weighted reweight the national wealth index for each survey.
68 POVERTY IN A RISING AFRICA
index, how to weight them, and what weights pattern with respect to asset ownership con-
to choose matters, because survey-specific ditional on household consumption level;
asset indexes are tailored to the asset patterns but the statistically significant correlations
in a particular country for a specific year. indicate that, conditional on consumption,
The most common index, the national wealth context partly drives asset ownership. This
index (NWI), relies on statistical procedures finding speaks to the concern about identify-
(for example, principal component analysis) ing a set of assets across countries that is con-
to determine weights. Even within countries, sistently associated with monetary poverty.
such an approach to weighting is sensitive to As indicated by the coefficient on the time
the choice of assets for the index calculation. indicator, asset ownership of each of the five
The result is a lack of comparability over assets increased from the earlier to later sur-
time and across countries (Abreu and John- vey, conditional on household consumption,
son 2013; Gwatkin and others 2007; McKen- suggesting asset drift.
zie 2005). Weights matter because different For this set of countries as a whole, there
countries often hold assets that are different is evidence of asset drift, but there is varia-
in type or quality. They have a strong bearing tion across countries. The share of countries
on whether the index shows a close correla- displaying asset drift is about 50 percent for
tion with consumption. television ownership, 36 percent for motor-
We explore some of these issues and bikes, 33 percent for computers, 20 percent
examine the patterns of accumulation for for refrigerators, and 10 percent for cars.
five privately held assets (television, refrigera- This evidence is consistent with the size and
tor, computer, motorbike, and car), without significance of the time indicator in pooled
indexing them into an aggregate indicator. country results in table 2.2.
Following the approach of Harttgen, Klasen, Data on assets may be useful in specific
and Vollmer (2013), we restrict the focus to ways as a proxy for consumption, such as
near-poor households (households with con- ranking households within a survey. But
sumption within 5 percent above or below given the methodological concerns and the
the poverty line). limited empirical evidence, these data do
Table 2.2 shows the results of regressions not seem to offer a robust alternative to con-
of asset ownership on consumption, the time- sumption data for measuring poverty and its
fixed effect, and the country typology using trends.
32 household surveys for 16 countries with
two comparable surveys. As consumption
rises among the set of near-poor households,
Profiling the Poor
they are more likely to own each asset. The This section provides a brief description
country typologies do not indicate a clear of the profiles of the poor. It begins with
TABLE 2.2 Many country-level factors affect asset ownership of the near-poor
Item Television Refrigerator Computer Motorbike Car
Consumption 0.378*** 0.335*** 0.004 0.164 –0.062
Middle income 0.202*** 0.123*** 0.003 0.082*** 0.011**
Resource rich –0.015** –0.081*** –0.003* 0.070*** 0.027***
Landlocked –0.014 –0.067*** –0.007 0.001*** –0.008**
Fragile 0.108*** –0.048*** –0.008** –0.019*** –0.012***
Later survey 0.113*** 0.014*** 0.007*** 0.068*** 0.019***
Number of observations 16,884 16,847 12,269 15,678 11,859
Source: World Bank Africa Poverty database for recent surveys of Botswana, Cameroon, the Democratic Republic of Congo, Côte d’Ivoire, Ethiopia, Ghana,
Madagascar, Malawi, Mozambique, Nigeria, Rwanda, Senegal, Sierra Leone, South Africa, Tanzania, and Uganda.
Note: Sample is households with consumption within 5 percent above or below the poverty line. Consumption is the log of consumption per capita
(PPP 2011). Other variables are indicators taking a value of 0 or 1.
Statistical significance: * = 10 percent, ** = 5 percent, *** = 1 percent.
REVISITING POVERTY TRENDS 69
identification of the location of the poor FIGURE 2.6 Fragility is associated with significantly slower
using broad country classifications. Then poverty reduction
it looks at urban and rural patterns within
countries, and concludes with a discussion
of poverty of female-headed households. –1.1 Middle income
female-headed households are more common associated with lower rates of female head-
in urban areas. Their prevalence is positively ship, presumably partly explained by lower
correlated with country income status but work-related migration by men associated
exhibits no relationship with state fragility or with growing local economies, there was
resource wealth. an Africa-wide annual trend increase of 0.4
Both the share of the population liv- percent in the share of the population liv-
ing in female-headed households and the ing in female-headed households (evaluated
share of households headed by women have at the mean share over the entire sample)
been rising, across regions and with age during the period of growth from the 1990s
(figure 2.8). According to Milazzo and van to 2013. Second, this seeming paradox is
de Walle (2015), two recent developments resolved by the fact that other things such
across Africa explain this finding. 22 First, as demographic and population characteris-
although economic growth is found to be tics, social norms, education, and the nature
FIGURE 2.8 Across Africa, more and more households are headed by women
by women (perccent)
40 40
20 20
0 0
15 25 35 45 55 65 75 15 25 35 45 55 65 75
Woman’s age Woman’s age
by women (perccent)
40 40
20 20
0 0
15 25 35 45 55 65 75 15 25 35 45 55 65 75
Woman’s age Woman’s age
Earliest survey Latest survey
of the family are changing across Africa and in East Africa poverty rates are similar in
encouraging female headship. female- and male-headed households.
Should this steady rise in the incidence The smaller household size of female-
of female-headed households cause con- headed households means that using per
cern? Do female-headed households tend capita household consumption as the wel-
to be poorer and more vulnerable than oth- fare indicator will tend to overestimate the
ers? Female heads are a diverse group that poverty of male-headed households relative
includes widows, divorced women, separated to female-headed households if larger house-
women, abandoned women, married women holds enjoy economies of scale (Lanjouw
with nonresident husbands (polygamous or and Ravallion 1995). Differences in poverty
migrant), and single women. Households according to the gender of the head thus
headed by certain categories of women— depend on the consumption indicator used
widows, divorced or separated women, and to measure poverty. As the share of female
single women—frequently appear to be dis- heads continues to grow, this sensitivity to
advantaged. Widow-headed households are per capita or alternate adjustments for demo-
significantly poorer than other households in graphic composition may grow with it.
Madagascar, Mali, Uganda, and Zimbabwe
(Appleton 1996; Horrell and Krishnan 2007;
van de Walle 2013; World Bank 2014a). But The Movement of People into
female-headed households that receive trans- and out of Poverty
fers from male members have consistently
higher consumption or income than male- To this point, this chapter provides a snap-
headed households and are substantially bet- shot of poverty at different points in time.
ter off than other female-headed households. It does not describe dynamics—movements
Female- and male-headed households dif- into and out of poverty. Many investiga-
fer in terms of demographics in ways that tions of poverty dynamics rely on panel
potentially disadvantage female-headed data, which track households and individu-
households. On average, female heads are als over time. This analysis is complicated
older (reflecting the many widowed heads) by a host of issues, such as the impact of
and have fewer years of education (4.1 ver- attrition, measurement error, and sample
sus 5.1 years). Their households tend to be selection bias (Christiaensen and Shorrocks
smaller (3.9 people compared with 5.1 people 2012). In addition, few of the earlier and
in households headed by men) but have higher long-running panels in Africa are nationally
dependency ratios (1.2 compared with 1.0). representative.23
Female heads are many times more likely to Two main messages emerge from the esti-
be living in households in which they are the mation of poverty dynamics from panel data
only adult. Three-quarters of male-headed in Africa. First, perhaps unsurprisingly, there
households, compared to just 44 percent of is huge variation in estimates of both chronic
female-headed households, are composed and transient poverty (figure 2.9). Chronic
of two adults and children. Female-headed poverty estimates range from 6 percent to
households are also more likely to be sin- 70 percent. Chronic poverty estimates for
gle-adult households (16 percent versus 10 the same country—and in some cases using
percent). the same datasets—can also vary widely,
Poverty rates based on household per cap- depending on the method and the number of
ita consumption are higher among people liv- spells used.
ing in male-headed households (48 percent) Second, movement into and out of poverty
than female-headed households (40 percent). is substantial: in 20 of 26 studies, transient
But there are differences across region. By poverty rates are higher than chronic poverty
this metric, poverty in Southern Africa is rates. The median transient poverty rate is
higher among female-headed households; about 32 percent while the median chronic
REVISITING POVERTY TRENDS 73
FIGURE 2.9 Estimates of movements into and out of poverty vary widely across Africa
Uganda 2005/6–2011/12 15 29 56
Nationally Representative
Uganda 1992–99 13 30 57
Tanzania 2010/11–2012/13 8 24 68
Nigeria 2010/11–2012/13 27 20 53
Malawi 2010/11–2013/14 23 32 44
Lesotho 1993–2002 26 42 32
0 10 20 30 40 50 60 70 80 90 100
Percent
Chronic poor (always poor) (%) Transient poor (sometimes poor) (%) Never poor (%)
Sources: Baulch 2011; Duponchel, McKay, and Ssewanyana 2014 (Uganda 2005/06–2011/12); Finn and Leibbrandt 2013 (South Africa, National Income Dynamics Study); World Bank
poverty assessments.
Note: Estimates for South Africa are based on Finn and Leibbrandt transition matrixes and a poverty rate of 45 percent using a national poverty line of R 620 a month in 2011.
74 POVERTY IN A RISING AFRICA
poverty rate is 21 percent, implying that a surveys in the region, an alternative approach
household or individual is more likely to be to obtaining evidence on the movement into
sometimes poor than always poor (compare and out of poverty is to use statistical meth-
the median of chronic poverty [blue bars] to ods to construct synthetic panels from avail-
the median of transient poverty [orange bars] able cross-sections (Dang and Lanjouw 2013,
in figure 2.9). Health, labor market, conflict, 2014; Dang and others 2014). In addition
and weather shocks have been identified as to generating more data on dynamics, the
major drivers of these transitions. synthetic panel approach applies the same
How much of transitory poverty is real methodology and uses the same standard and
and how much reflects measurement error welfare measure for all countries, which is
is a matter of debate. According to some not the case in most panel studies. Synthetic
researchers, measurement error of income panel data may also be more representative of
or consumption may explain as much as half the population than panel data, which suffer
of transitory poverty (Dercon and Krishnan from attrition.
2000; Glewwe 2012). In constructing synthetic panels, we
Revisiting the same household or indi- selected countries with comparable surveys.
vidual over several years has its advantages, Figure 2.10 decomposed each country’s pov-
but doing so is costly—the main reason why erty over time into components: chronic
large, nationally representative panels over poverty (households that were poor in both
long periods are rare. Given the paucity of periods), downwardly mobile (households
nationally representative household panel that fell into poverty in the second period),
FIGURE 2.10 The share of poor people in Africa who fall into poverty is about the same as the share of
poor people who move out of poverty
Mauritania
Nigeria
Ghana
Cameroon
Côte d'Ivoire
Botswana
Senegal
Swaziland
Ethiopia
Tanzania
Sierra Leone
Chad
Uganda
Togo
Rwanda
Zambia
Burkina Faso
Mozambique
Malawi
Madagascar
Congo, Dem. Rep.
All countries
0 10 20 30 40 50 60 70 80 90 100
Percentage (%)
Chronic poor Downwardly mobile Upwardly mobile Never poor
and nonpoor. Rates of chronic poverty vary FIGURE 2.11 Africa’s poor are clustered around the
across countries and do not appear to be poverty line
linked to overall poverty rates. 24 The non-
70
poor are further decomposed into two com-
90
93
96
99
02
05
20 8
2010
2011
12
26 percent of the nonpoor population
0
19
19
19
19
20
20
20
emerged from poverty (that is, were poor Poverty line (level of daily consumption)
in the first period but not the second). 24 $1.90 $2.20 $2.40
This group could be considered vulnerable
Source: World Bank Africa Poverty database. The estimates use data
to falling back into poverty. Second, coun- corrected for comparability and quality.
tries that are similar in terms of poverty
rates may be dissimilar in terms of poverty
dynamics. For instance, Ethiopia and Sene- rate by 12 percentage points. Poverty rates
gal both show similar average poverty rates, have declined, but the level of vulnerability
but the share of chronically poor people is remains very high.
larger in Ethiopia. Third, in some countries
with low poverty rates, a large share of the
Concluding Remarks
poor are chronically poor. Botswana, for
example, has poverty rates that are among How much poverty reduction has been
the lowest in the sample, but almost all achieved since Africa’s economic recovery
of its poor are chronically so (Dang and began 15 years ago? The answer has been
Dabalen 2015). contentious, partly because the poverty data
The review of the literature on poverty have not been properly scrutinized for com-
dynamics and the synthetic panel results parability and quality.
depict a situation in which vulnerability is Assessment of the data leads to three
high, as evident from the prevalence of tran- important conclusions. First, once known
sient poverty. Because Africa’s poor appear to data problems are corrected, current poverty
be clustered around the poverty line, a small rates are lower and poverty reduction at least
positive shock to incomes could lift many out as large as international poverty estimates
of poverty, but a small negative shock could suggest. The most comprehensive source of
drive as many into poverty. household consumption survey data that pro-
How large is the clustering around the vides country and regional estimates of pov-
poverty line? Raising the poverty line by erty is the World Bank PovcalNet database.
$0.30–$0.50 (equivalent to a 16–26 percent According to the surveys available on the
negative shock to incomes) increases the pov- database, Africa’s poverty rate—defined in
erty rate by 5 to 12 percentage points (figure this report as people living on less than $1.90
2.11). Raising the poverty line by $0.30 in per person per day (PPP 2011)—was 43 per-
1990 increases the poverty rate from 55 per- cent in 2012, a 14 percentage point decline
cent to 60 percent. Raising the poverty line since 1990. Accounting for the comparability
from $1.90 to $2.40 (that is by $0.50—or and quality of data suggests that the decline
26 percent) in 2012 increases the poverty may have been larger. The adjusted data
76 POVERTY IN A RISING AFRICA
imply that the poverty rate could be as much lack of panel surveys with national coverage
as 6 percentage points lower (37 percent over long periods makes it difficult to estab-
instead of 43 percent) in 2012. Important lish this fact with certainty. The share of the
drivers of the larger decline are corrections transient poor (the sum of the upwardly and
to the Nigeria data (which account for a downwardly mobile), at roughly 25 percent
large fraction of the difference between the of the population, also suggests a significant
estimates of the adjusted data and the Pov- share of vulnerable population.
calNet data) and greater reliance on GDP
simulations.
A number of robustness checks support Notes
the notion that poverty reduction may have 1. The term poverty is used here to refer to
been larger than assumed. Based on spells of people with consumption levels below the
comparable surveys only and excluding Nige- international poverty line. The MDGs use
ria, the implied annual change in poverty the term extreme poverty to describe these
using GDP imputation is similar to the one people.
recorded in the data correcting for compara- 2. Some scholars argue, for example, that the
bility and quality. The results derived from African poverty rate has been falling much
survey-to-survey imputation methods suggest more quickly than internationally accepted
conventional wisdom suggests (Pinkovskiy
that the decline was larger than previously
and Sala-i-Martín 2014; Young 2012).
thought. This also applies to the S2S results 3. This report does not address the problem of
for Nigeria, which supports the notion that comparability across countries.
poverty in Nigeria declined faster than cur- 4. South Sudan—for which there are no pur-
rent official estimates suggest. In addition, chasing power parity (PPP) exchange rates
results from Engel curve estimation imply and, until recently, no consumer price index
that CPIs may overestimate changes in the (CPI) data—was not included in the regional
cost of living and hence underestimate pov- poverty estimate. No survey data were avail-
erty reduction. able for four countries (Equatorial Guinea,
Second, although this is good news, the Eritrea, Somalia, and Zimbabwe). For these
challenge remains substantial; the region did countries, the average regional poverty rate
not meet the MDG target of halving poverty was assigned. Together these countries are
home to about 5 percent of the population of
by 2015 and many more people are poor in
Africa.
2012 than in 1990 (even under the most opti- 5. Where there are multiple surveys that are not
mistic scenario of poverty reduction). If the comparable, only the survey that included
pace of poverty reduction does not pick up, it the most comprehensive consumption data
will take the region another decade to reach was used.
this target. 6. These poverty trends are robust to changes
A major drag on reaching the goal is fra- in country composition. The same imputa-
gility. Among the four types of countries tion methods were applied to two subsam-
assessed—fragile, resource rich, landlocked, ples: the 23 most populous countries and
and low-income—fragile countries had the the 27 countries with at least two compa-
slowest rate of poverty reduction. Between rable surveys. For the 23 largest countries,
1996 and 2012, this group of countries which account for more than 88 percent of
the total and the poor population, poverty
reduced poverty by 12 percentage points—13
declined from 55 percent to 36 percent (19
percentage points less than nonfragile coun-
percentage points) based on the comparable
tries. Controlling for other characteristics and good-quality data and from 57 percent
(resource richness, landlockedness, and low- to 43 percent (14 percentage points) based
income status) increases the difference in on the full sample of surveys (PovcalNet).
poverty reduction to 15 percentage points. Among the 27 countries for which there are
Third, about 58 percent of the poor in at least two comparable surveys, which rep-
Africa may be chronically poor, although the resent about 76 percent of the population
REVISITING POVERTY TRENDS 77
and almost 80 percent of the poor, poverty for the Democratic Republic of Congo at the
dropped from 57 percent to 38 percent (19 2005–09 rate.
percentage points) based on the comparable 12. For the period 1990–94, there was no sur-
and good-quality data. As with the pat- vey coverage or surveys in the immediately
tern among all countries, poverty measures following period for 4 of the 23 countries,
peaked in the mid-1990s and declined more so regional averages computed from the rest
sharply after 2002 when comparable and of the 19 countries were used. Similarly,
good-quality data are used. regional averages were used for 3 countries
7. For Burundi, Gambia, and Seychelles, only for 1995–99, 2 countries for 2000 – 04,
one of the comparable consumption aggre- 1 country for 2005–09, and 1 country for
gates is available for use at the time of this 2010–12.
report. 13. In general, only data that were subject to
8. One of these countries is Zambia, where the rigorous vetting (in terms of completeness
finding is based on poor-quality data. of the sample and consumption aggregate,
9. One of these countries is Nigeria, where the proper documentation, and consistency with
finding is based on poor-quality data. consumption measures used by countries
10. Because the richness of survey data within in their monitoring and analysis) are used
and across countries varied widely over in PovcalNet estimates. What is referred
time, attempts were made to maintain the to as PovcalNet results here are estimates
same model across time within but not obtained by applying the methods used in
across countries. Overall, for each model PovcalNet (described in World Bank 2015b)
four clusters of variables were analyzed: to the vetted data for these 23 countries.
demographics, education of the household We were able to closely replicate the official
head, housing and assets, and rural and PovcalNet estimates for the period 1990–
urban location. 2012, in some cases differing only by a deci-
11. More specifically, if a survey and an estimate mal point.
for a country were available in the period 14. This discussion focuses on the role of the
immediately before or after the period with- CPI in adjusting consumption in a given sur-
out a survey, the nearest available estimate vey year to the benchmark year. Prices also
was used for the period without a survey. matter for the profile of poverty within a
For example, Ethiopia conducted a sur- country. For instance, urban-rural poverty
vey in 1994/1995. Assigning the poverty gaps may be overestimated if price differ-
rate from 1994/95 to 1995–99 leaves the ences between urban and rural areas are
1990–94 period without a poverty estimate underestimated. Cross-country compari-
for Ethiopia, as there were no surveys dur- sons—and therefore regional poverty levels
ing this period. Therefore, we used the esti- and trends—will also be sensitive to changes
mate from 1994/95 for both 1990–94 and and adjustments to PPP exchange rates. This
1995–99, keeping Ethiopia’s poverty rate for section does not address these issues.
that period unchanged. The main goal of the 15. Engel Law is the observation that, as income
exercise is to avoid using GDP imputations rises, the share of income devoted to food
to fill in missing data points and to avoid falls, even if actual expenditure on food may
creating a series that would seem implau- be rising.
sible. For instance, there are no surveys for 16. Where there are more than two comparable
the Democratic Republic of Congo before surveys per country, the CPI bias is esti-
2005. In 2005 the extreme poverty rate mated separately for each subperiod. The
estimated from survey data was 91 percent. estimation is further restricted to countries
If we assign a regional poverty rate for the for which monthly CPI data (food, nonfood,
period without surveys, the poverty rate in and all-item CPIs) from the national statis-
the Democratic Republic of Congo would be tical agency are available, as these data are
half what the actual survey says and would needed to control for relative price changes.
make the country one of the least poor coun- The method only partially accounts for the
tries in Africa before 2005. To avoid such a quality change bias and does not capture the
series break, for all periods before 2005–09, consumer surplus arising from the introduc-
we were compelled to hold the poverty rate tion of new commodities (Gibson, Stillman,
78 POVERTY IN A RISING AFRICA
and Le 2008). Plutocratic bias (whereby the 19. Dropping outliers (differences of more than
CPI gives more weight to the consumption 3 percentage points in absolute value) does
of richer households) is addressed because not change this result substantially (–1.1
the results are democratically weighted esti- becomes –0.8).
mates (that is, use household sample weights 20. Nakamura, Steinsson, and Liu (2014) show
that are more representative of their share that if inflation rates at different points in
in the population) among the subsample of the income distribution are similar, the
urban households and do not weight house- fact that the Engel curve deflator is for
holds according to their total expenditures. one unknown household and the CPI is for
Studies on the Russian Federation (Gibson, another household should not matter: one
Stillman, and Le 2008) and Brazil and Mex- can attribute most of the gap between the
ico (de Carvalho Filho and Chamon 2012) two to genuine CPI bias. And in a recent
use income as an instrumental variable for analysis Hobijn and Lagakos (2005) sug-
consumption to address endogeneity arising gest that, over long periods of time, the CPI
from the fact that total consumption enters inflation rate accurately represents changes
both sides of the regression equation (that in the cost of living for households at differ-
is, when computing budget shares and when ent parts of the income distribution.
controlling for consumption levels). The 21. See Ainsworth and Filmer (2006); Bicego,
results suggest that ordinary least square Rutstein, and Johnson (2003); Bollen, Glan-
estimates, such as the ones presented here, ville, and Stecklov (2002); Case, Paxon, and
may suffer from some degree of bias because Ableidinger (2004); Filmer and Pritchett
of correlated measurement error but are (1999, 2001); Gwatkin and others (2000);
unlikely to show a different direction of bias McKenzie (2005); Rao and Ibanez (2005);
than the instrumental variable estimates. Sahn and Stifel (2000); Schellenberg and
Because many of the household surveys used others (2003); and Stifel and Christiaensen
in this report do not contain income aggre- (2007).
gates, endogeneity concerns could not be 22. This trend is estimated from a regression
addressed in the same manner. (the log of the odds ratio) of the share of the
17. For Nigeria and Rwanda, urban and rural population living in female-headed house-
CPI series were used. For Ethiopia, regional holds using 98 country-year DHS surveys
CPI (but collected from urban areas) were covering the last 25 years. Milazzo and
used. Finally in Mauritius, urban CPI was van de Walle (2015) report that the trend is
applied to rural and urban households dur- explained largely by rising age at marriage
ing the Engel curve estimation because the and higher education levels.
household survey does not have urban and 23. Since the introduction of the Living Stan-
rural identifiers. dards Measurement Study–Integrated Sur-
18. This finding contrasts with the view of veys on Agriculture, surveys have been
Sandefur (2013), who argues that CPI infla- nationally representative.
tion understates true inflation and hence 24. In principle, poverty mobility is likely to
provides too optimistic a view of poverty be greater over longer intervals (see, for
reduction in Africa. His analysis is based example, Dang and Lanjouw 2014). For
on a database of national poverty lines that these data, however, the Pearson correlation
tend to increase (in nominal terms) at a more between chronic poverty and the length of
rapid rate than official CPI inflation. Under time between the two cross-sections is weak
certain conditions (related to how these (0.35 and not statistically significant).
national poverty lines are constructed), 25. Notice that the average upward mobility
the poverty lines he proposes can reveal in these countries is about 14 percent and
changes in the cost of living among the poor. the nonpoor population is around 54 per-
However, the vast majority of the poverty cent (40 percent never poor plus 14 percent
lines Sandefur uses do not meet the neces- upwardly mobile). Therefore, the fraction
sary conditions (see Gaddis 2015) and are of the upwardly mobile among the non-
therefore inappropriate for inferring price poor is 14/54, which is roughly 26 percent.
changes between surveys. Similarly, on average about 35 percent of
REVISITING POVERTY TRENDS 79
the poor were poor in both periods. The Case, Anne, Christina Paxson, and Joseph Able-
fraction of the population that was poor idinger. 2004. “Orphans in Africa: Paren-
at least once in both periods includes the tal Death, Poverty, and School Enrollment.”
chronic poor (35 percent), the downwardly Demography 41 (3): 483–508.
mobile (11 percent) and the upwardly mobile Christiaensen, Luc, Peter Lanjouw, Jill Luoto, and
(14 percent). Therefore, the fraction of the David Stifel. 2012. “Small Area Estimation-
chronic poor among the poor is about 58 Based Prediction Methods to Track Poverty:
percent (35/60). Validation and Applications.” Journal of Eco-
nomic Inequality 10 (2): 267–97.
Christiaensen, Luc, and Anthony Shorrocks.
References 2012. “Measuring Poverty over Time.” Jour-
nal of Economic Inequality 10 (2): 137–143.
Abreu, A., and D. Johnston. 2013. “Asset Indices
Corral, Paul, Vasco Molini, and Gbemisola
as a Proxy for Poverty Measurement in African
Oseni. 2015. “No Condition Is Permanent:
Countries: Reassessment.” Paper presented at
Middle Class in Nigeria in the Last Decade.”
the Conference African Economic Develop-
Policy Research Working Paper 7214, World
ment: Measuring Success and Failure. Vancou-
Bank, Washington, DC.
ver: British Columbia, Canada. April 18-20,
Costa, Dora L. 2001. “Estimating Real Income in
2013.
the United States from 1888 to 1994: Correct-
Ainsworth, Martha, and Deon Filmer. 2006.
ing CPI Bias Using Engel Curves.” Journal of
“Inequality in Children’s Schooling: AIDS,
Orphanhood, Poverty, and Gender.” World Political Economy 109 (6): 1288–310.
Development 34 (6): 1099–128. Dabalen, Andrew, Isis Gaddis, and Nga Thi Viet
Alfani, Fedrica, Carlo Azzarri, Marco d’Errico, Nguyen. 2015. “CPI Bias and Its Implication
and Vasco Molini. 2012. “Poverty in Mozam- for Poverty in Sub-Saharan African Coun-
bique: New Evidence from Recent Household tries.” Background paper prepared for the
Surveys.” Policy Research Working Paper Poverty in a Rising Africa report, World Bank,
6217, World Bank, Washington, DC. Washington, DC.
Appleton, S. 1996. “Women-Headed Households Dang, Hai-Anh, and Andrew Dabalen. 2015.
and Household Welfare: An Empirical Decon- “How Large Is Chronic Poverty in Africa:
struction for Uganda.” World Development 24 Poverty Dynamics using Synthetic Panels.“
(12): 1811–27. Background paper prepared for the Poverty in
Baulch, Bob. 2011. Why Poverty Persists: Poverty a Rising Africa report, World Bank, Washing-
Dynamics in Asia and Africa. Northampton, ton, DC.
MA: Edward Elgar Publishing. Dang, Hai-Anh, and Peter Lanjouw. 2013. “Mea-
Bicego, George, Shea Rutstein, and Kiersten suring Poverty Dynamics with Synthetic Pan-
Johnson. 2003. “Dimensions of the Emerging els Based on Cross-Sections.” Policy Research
Orphan Crisis in Sub-Saharan Africa.” Social Working Paper 6504, World Bank, Washing-
Science & Medicine 56 (6): 1235–47. ton, DC.
Bloom, David, and Jeffrey Sachs. 1998. “Geog- ———. 2014. “Welfare Dynamic Measurement:
raphy, Demography, and Economic Growth Two Definitions of a Vulnerability Line.”
in Africa.” Brookings Papers on Economic Policy Research Working Paper 6944, World
Activity 2: 207–95. Bank, Washington, DC.
Bollen, Kenneth, Jennifer Glanville, and Guy Dang, Hai-Anh, Peter Lanjouw, Jill Luoto, and
Stecklov. 2002. “Economic Status Proxies in David McKenzie. 2014. “Using Repeated
Studies of Fertility.” Population Studies 56 (1): Cross-Sections to Explore Movements in and
81–96. out of Poverty.” Journal of Development Eco-
Boskin, Michael J., Ellen R. Dulberger, Robert J. nomics 107: 112–28.
Gordon, Zvi Griliches, and Dale W. Jorgenson. Dang, Hai-Anh, and Minh Nguyen. 2014.
1996. “Toward a More Accurate Measure of “povimp: Stata Module to Impute Poverty in
the Cost of Living: Final Report to the Sen- the Absence of Consumption Data.” World
ate Finance Committee from the Advisory Bank, Development Research Group, Pov-
Commission to Study the Consumer Price erty and Inequality Unit and Global Poverty
Index.” Practice.
80 POVERTY IN A RISING AFRICA
Luke, Gallup, Jeffrey D. Sachs, and Andrew Mshinda, and Jennifer Bryce. 2003. “Inequali-
Mellinger. 1999. “Geography and Economic ties among the Very Poor: Health Care for
Development.” In Annual Conference on Children in Rural Southern Tanzania.” Lancet
Development Economics 1998, edited by 361 (9357): 561–66.
Boris Pleskovic and Joseph E. Stiglitz, 127–78. Smits Jeroen, and Roel Steendijk. 2015. “The
Washington, DC: World Bank. International Wealth Index (IWI).” Social
McKenzie, David. 2005. “Measuring Inequality Indicators Research 122 (1): 65–85.
with Asset Indicators.” Journal of Population Stifel, David, and Luc Christiaensen. 2007.
Economics 18 (1): 229–60. “Tracking Poverty over Time in the Absence of
Milazzo, Annamaria, and Dominique van de Comparable Consumption Data.” World Bank
Walle. 2015. “Women Left Behind? Poverty Economic Review 21 (2): 317–41.
and Headship in Africa.” Policy Research van de Walle, Dominique. 2013. “Lasting Welfare
Working Paper 7331, World Bank, Washing- Effects of Widowhood in Mali.” World Devel-
ton, DC. opment 51: 1–19.
Nakamura, Emi, Jon Steinsson, and Miao Liu. World Bank. 2012. Zambia Poverty Assessment:
2014. “Are Chinese Growth and Inflation Too Stagnant Poverty and Inequality in a Natural
Smooth? Evidence from Engel Curves.” NBER Resource-Based Economy. Washington, DC:
Working Paper No. 19893. Issued in February World Bank.
2014. ———. 2013. Burkina Faso. A Policy Note: Pov-
Pinkovskiy, Maxim, and Xavier Sala-i-Martín. erty Trends and Profile for 2003–2009. Wash-
2014. “Africa Is on Time.” Journal of Eco- ington, DC: World Bank.
nomic Growth 19 (3): 311–38. ———. 2014a. Face of Poverty in Madagascar:
Rao, Vijayendra, and Ana Maria Ibanez. 2005. Poverty, Gender, and Inequality Assessment.
“The Social Impact of Social Funds in Jamaica: Report 78131-MG, Washington, DC: World
A ‘Participatory Econometric’ Analysis of Tar- Bank.
geting, Collective Action, and Participation in ———. 2014b. Nigeria Economic Report, No. 2,
Community-Driven Development.” Journal of July. Washington, DC: World Bank.
Development Studies 41 (5): 788–838. ———. 2014c. “Nigeria: Where Has All the
Rutstein, Shea, and Kiersten Johnson. 2004. The Growth Gone?” Policy Note 78908, World
DHS Wealth Index. DHS Methodological Bank, Washington, DC.
Report 6, ICF International, Rockville, MD. ———. 2015a. Africa’s Demographic Transition:
Rutstein, Shea, and Sarah Staveteig. 2014. Making Dividend or Disaster? edited by David Can-
the Demographic and Health Surveys Wealth ning, Sangeeta Raja, and Abdo S. Yazbeck.
Index Comparable. DHS Methodological Washington, DC: World Bank.
Report 9, ICF International, Rockville, MD. ———. 2015b. A Measured Approach to Ending
Sahn, David, and David Stifel. 2000. “Poverty Poverty and Boosting Shared Prosperity: Con-
Comparisons over Time and across Coun- cept, Data and the Twin Goals. Policy Research
tries in Africa.” World Development 28 (12): Report. Washington, DC: World Bank.
2123–55. ———. 2015c. United Republic of Tanzania:
Sandefur, Justin. 2013. “Africa Rising? Using Tanzania Mainland Poverty Assessment.
Micro Surveys to Correct Macro Time Series.” Report AUS6819. Washington, DC: World
Center for Global Development, Washington, Bank.
DC. Young, Alwyn. 2012. “The African Growth Mir-
Schellenberg, Joanna, Cesar Victora, Adiel Mushi, acle.” Journal of Political Economy 120 (4):
Don de Savigny, David Schellenberg, Hassan 696–739.
Poverty from a
Nonmonetary Perspective 3
C
hapter 2 considers poverty in mone- Second, the benefit of income can be lim-
tary terms. Using income or the mon- ited when it interacts with other conditions.
etary value of consumption as a basis The benefits of a bicycle as a means of trans-
for defining the poor is appealing on several port, for example, are quite different for an
grounds. It allows for different preferences in able-bodied and a handicapped person. Such
purchases, the definition of a poverty thresh- resources are instrumentally valuable; they
old in “objective” ways (such as the cost of have no intrinsic value. Relying on their mon-
a minimum-calorie diet), and aggregation etary valuation as a measure of well-being
across domains (the value of food and non- can therefore be misleading.
food consumed). Third, income measures are at the house-
Yet, income fails to provide a complete hold level, which assumes the equal distribu-
picture of well-being for several reasons. tion of income across household members.
First, many aspects of well-being are not Yet intrahousehold inequalities in the distri-
just difficult to price but valuable in ways bution of household income across house-
that cannot be monetized (Sandel 2012; hold members can be substantial (Chiappori
Sen 1985). For example, commoditizing the and Meghir 2015; see box 4.3 in chapter 4).
right to vote by allowing people to sell it Direct information on individuals avoids this
would yield a market value of voting rights strong assumption.
that would not capture the full meaning This chapter briefly reviews how the capa-
and value of the right as an expression of bility approach motivates a nonmonetary
citizenship and political participation. The multidimensional perspective on poverty. It
list of difficult-to-monetize aspects of well- then assesses Africa’s progress in literacy and
being is long, including the ability to read education, life expectancy and health, free-
and write, longevity and good health, secu- dom from violence, and self-determination
rity, political freedoms, social acceptance (freedom to decide). It devotes special atten-
and status, and the ability to move and tion to displaced and disabled people, two
connect. vulnerable groups that are rarely covered in
standard poverty reports (because of data
This chapter was written with Umberto Cattaneo, limitations). Finally, it considers the four
Camila Galindo-Pardo, and Agnes Said. dimensions of well-being jointly, in order to
83
84 POVERTY IN A RISING AFRICA
identify countries and individuals that are and functionings also underlies the rich and
deprived in multiple dimensions. vibrant literature on multidimensional pov-
erty (Alkire 2008; Bourguignon and Chakra-
varty 2003; Robeyns 2005; Sen 1999).
The Capability Approach Using the capability approach to measure
Sen (1980, 1985, 1999) proposes the capabil- well-being is challenging. There are some
ity approach as an alternative to monetary common approaches to measuring certain
measures of poverty. This approach focuses basic functionings, such as the ability to read
on what people effectively do and are (their and write, adequate nourishment, and good
functionings) and on the capacity of people health, even though measurement issues
to freely choose and achieve these function- remain here as well (de Walque and Filmer
ings (that is, their capability) rather than on 2012; UNESCO 2015). There is much less
the commodities bought or consumed. There experience in measuring other functionings
is broad consensus that functionings such as (including mobility, social integration, and
the ability to read and write and being well even the capacity to aspire) and in measur-
nourished, healthy, and free from violence ing capability. An added complexity is deter-
and oppression are vital for human devel- mining thresholds below which a person is
opment. They are ontological needs (stem- considered poor, as these cutoffs arguably
ming from the condition of being human) depend on the individual’s choices and pref-
that apply to every person regardless of geo- erences. Finally, there is the challenge of
graphic location or time (Max-Neef, Elizalde, aggregation. For example, how much poorer
and Hopenhayn 1991). Focusing on individ- should a person be considered when deprived
ual achievements in these areas thus provides in several functionings compared to when
a good basis to begin assessing poverty from deprived in only one?
a nonmonetary perspective. The human development index (HDI) and
Education, health, and security also the multidimensional poverty index (MPI)
expand the choices people can make and the (Alkire and Santos 2014) are applications of
range of things they can do and be (that is, the capability approach to assessing progress
their capabilities). But they are not sufficient. in societies.1 Both indexes focus on achieve-
Social and political institutions often impede ments in education, longevity/health, and liv-
self-realization. Basic personal and political ing standards (through income and assets).
freedoms are equally essential. To appreci- The approach pursued in this chapter is in
ate the importance of opportunity and choice this tradition, though with a different empha-
for assessing well-being, consider two people, sis in three areas.
both teachers. One chooses teaching from First, to provide a more comprehensive
among a range of occupational options. The view of people’s basic capabilities, the chap-
other becomes a teacher because other, pre- ter considers two additional dimensions:
ferred options are excluded because of cul- freedom from violence and the opportunity
tural constraints (engineering is closed to for self-determination (freedom to decide).
women) or location (engineering jobs are not Poverty analyses have largely ignored these
available in remote villages) or because some- dimensions.
one else chose the profession for her (Foster Second, the degree of joint deprivation is
2011). Can they be considered equally well explored by estimating the share of people
off? Clearly, personal autonomy and self- deprived in one, two, and more dimensions.
determination matter for well-being. The This approach achieves a middle ground
study of outcomes should not be indifferent between a single index of poverty (which
to the process by which the choice was made. requires weighting achievements in the vari-
The capability approach provides the ous dimensions) and a dashboard approach
philosophical underpinnings for the non- (which simply lists achievements dimension
monetary perspective on poverty examined by dimension, ignoring jointness in depriva-
in this chapter. Sen’s vision of capabilities tion) (Ferreira and Lugo 2013).
POVERT Y FROM A NONMONE TARY PERSPEC TIVE 85
Third, the focus is on outcomes measured the World Value Surveys. These data enable
at the individual (not the household) level. the much wider dimensional and more indi-
Where data on outcomes are not available, vidualistic scope of this chapter. Some of the
information on inputs (such as use of bednets concerns regarding data availability, compa-
and vaccination rates in lieu of disease prev- rability, and quality highlighted in the context
alence measures) and proximate measures of the expenditure surveys apply here as well,
(such as governance indicators for freedom to however. Their implications are discussed
decide) are used. throughout the chapter where relevant.
Data on nonmonetary dimensions of pov- There has also been an upsurge in the
erty are now much more widely and more availability and use of subjective measures
regularly available than they once were, of well-being and poverty, such as measures
including at the individual level. This follows based on ordinal questions about happiness
the rapid expansion and public availability of or life satisfaction (box 3.1). Given the lack of
the Demographic and Health Surveys (DHS), a common frame of reference, which makes it
as well as the Africa-wide and globally com- difficult to compare across people and time,
parable national opinion surveys, such as the these measures are not used here to assess
Afrobarometer, the Gallup World Poll, and poverty.
Subjective measures of well-being reflect utility as interpersonal comparisons become difficult. Adapta-
a mental state (happiness) or as a cognitive reflec- tion of happiness standards and aspirations—lower-
tion of the condition of one’s life. Unlike income ing them when conditions go awry and raising them
measures of poverty, they do not rely on prices or when conditions improve—are pervasive worldwide.
monetary valuations, although they include both Countries with higher rates of HIV prevalence, for
monetary and nonmonetary dimensions of well- example, do not systematically report lower life
being. These measures are based on the personal satisfaction (Deaton 2008); people who lose limbs
evaluation of individuals themselves, reflecting the still record good well-being scores (Loewenstein and
value attached to individual sovereignty. Being one- Ubel 2008; Oswald and Powdthavee 2008).
dimensional, they facilitate complete orderings. The subjective well-being approach also does
Answers to subjective well-being questions, such not adjust for individual tastes or aspirations. This
as questions based on ordinal questions about hap- could lead to paradoxical policy actions, such as
piness, economic welfare, or life satisfaction, are redistribution from poor happy subsistence farmers
intuitive and not time consuming to collect. They to unhappy millionaires. Subjective well-being data
confirm that many dimensions beyond income and may therefore not yet be appropriate to monitor liv-
material consumption—health, job market status, ing standards. They do, however, contain important
the quality of relations and social interactions, and complementary information about people’s prefer-
even political rights and freedom of speech (Frey ences that could help inform policy makers about
and Stutzer 2002)—matter and that happiness and how to value public goods or weight nonmonetary
life satisfaction increase with income at a declining dimensions of well-being (Decanq, Fleurbaey, and
pace (or not at all beyond a certain level of income, Schokkaert 2015) or set the poverty line (Ravallion
according to the Easterlin paradox (Easterlin 1974), 2012). As the capability approach emphasizes, per-
though the existence of this paradox remains sonal preferences and choices cannot be ignored in
debated [Stevenson and Wolfers 2008]). assessing an individual’s level of poverty and well-
One challenge with subjective well-being is the being. How to use questions about subjective well-
lack of a common frame of reference. As individuals being to learn about those aspects of people’s prefer-
adapt their tastes and aspirations to their circum- ences that policymakers want to take on board is an
stances, intrapersonal comparisons over time and important research agenda.
86 POVERTY IN A RISING AFRICA
100
93
90
83
80 79
70 67
Adult literacy rate (%)
62
60 54
58
50
50
40
30
20
10
0
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
East Asia and Pacific Middle East and North Africa Sub-Saharan Africa
Latin America and the Caribbean South Asia
Following the launch of the Education for All initia- literacy rates, the trends reported here are thus rea-
tive in 2000, much effort has been devoted to moni- sonably well supported by actual data, despite the
toring adult literacy. Nonetheless, data on adult lit- small number of observations.
eracy are still collected relatively infrequently, and How comparable are these data? Until the mid-
both the definitions and methods of measurement 2000s, estimates of African literacy were based
keep changing, raising validity and comparability on self- or proxy declarations on whether a per-
issues similar to the ones encountered in compiling son could read or write, with countries sometimes
expenditure data for poverty tracking (see chapter assuming literacy among people who had completed
1). On the upside, the UNESCO Institute for Statis- primary school. Estimates were obtained from
tics provides detailed and publicly accessible meta- census or survey data. Since 2006, where self- or
data of the data sources, definitions, and actual proxy declaration–based estimates are not avail-
measures used (UNESCO 2015). able, literacy scores have increasingly been derived
For the period 1995–2012, 109 annual literacy from direct assessments, in which respondents are
estimates are available—13 percent of the total pos- asked to read a sentence from a card (this technique
sible number of 828 (18 years * 46 countries [no is used in the DHS and Multiple Indicator Cluster
literacy data are available for Somalia and South Surveys [MICS]). Twenty of the 56 literacy estimates
Sudan]). Although the figure seems low, annual recorded in Africa during 2006–12 were test based.
changes in literacy rates are small, meaning that the Literacy rates obtained through direct assess-
small number of estimates is less important than it ments were 8 percentage points lower on average
would otherwise be. A more relevant metric is the across a sample of 20 countries (UNESCO 2015).
number of countries with two or more surveys or Their increased use may partly explain why Africa’s
censuses to estimate literacy rates and the proximity recorded progress on adult literacy has not been
of the data sources to the beginning and end points more rapid.
of the study period (1995 and 2012). The measure of literacy as the self- or proxy-
On this count the picture is better. Only four declared ability to read and write only a short simple
countries, which together represent 6.4 percent of statement about everyday life is rudimentary. Liter-
the 2013 African population, have only one esti- acy today is seen as a “continuum of skills, such as
mate; the remaining 42 have two or more records. the ability to identify, understand, interpret, create,
For these countries, linear interpolation and extrap- communicate, and compute using printed and writ-
olation are used to fill in the missing years. For ten materials associated with varied contexts, that
countries with only one observation, the average enables individuals to achieve their goals in work
African trend was applied to extrapolate. The aver- and life and participate fully in society” (UNESCO
age population-weighted literacy estimate in each 2015, 137). This shift toward a more demanding
country is 4.5 years removed from 1995 and 3.3 notion of literacy mirrors the notion of rising pov-
years from 2012. Given the small annual change in erty lines as countries develop.
This gap partly accounts for the low levels of higher-income countries (by about 32 per-
overall adult literacy there. Gender parity is centage points in upper-middle-income and
much higher in Southern Africa. The ratio high-income countries and about 14 percent-
of literate women to literate men is only 0.32 age points in low-middle-income countries)
in Guinea and 0.38 in Niger. In contrast, (figure 3.4). In resource-rich countries, how-
women are more likely to be literate than ever, illiteracy rates are about 3 percentage
men in Lesotho (1.34) and Namibia (1.08). points higher than in resource-poor countries
What traits of households and countries (irrespective of the country’s income level,
explain the gender gap in literacy? Over- landlockedness, or fragility), indicating that
all, female illiteracy rates are substantially governance factors matter. Women in poor
higher in low-income countries than in rural households are 36 percent more likely
88 POVERTY IN A RISING AFRICA
FIGURE 3.2 Literacy rates are lowest in West Africa countries could not read for meaning (figure
3.5). Even in Kenya 20 percent of sixth grad-
ers could not read for meaning. Among fran-
Equatorial Guinea cophone countries in the region, 55 percent
South Africa
Seychelles
of fifth graders did not reach the minimum
Mauritius performance threshold, and half of them per-
Botswana formed at or below the level expected from
Cabo Verde random guessing. Scores for numeracy skills
Swaziland and mathematics are equally poor.
Zimbabwe
Congo, Rep.
Comoros Life Expectancy, Health, and Nutrition
Uganda
Ghana A widely used measure of the ability to live
Cameroon a long and healthy life is life expectancy at
Angola birth. It provides a comprehensive reflec-
Eritrea
Tanzania
tion of the various factors that affect health
Rwanda and mortality. A more refined measure is
Malawi healthy life expectancy, the number of years
Zambia a newborn can expect to live in full health.
Togo Life expectancy and mortality indicators are
Africa
estimated for a population (usually at the
Guinea-Bissau
Senegal country level). In contrast, nutrition (and dis-
The Gambia ability) indicators provide individual views of
Sierra Leone health status.
Côte d’Ivoire Life expectancy. Over the past decade,
Chad Africa experienced a massive increase in life
Central African Republic
Mali
expectancy: babies born in 2013 are expected
Guinea to live 6.2 years longer than babies born in
Niger 2000 (figure 3.6). The change makes the
0 20 40 60 80 100 region one of the strongest recent perform-
Adult literacy rate in 2012 (%)
ers in the world, above South Asia, where
life expectancy increased by 6.0 years since
Source: EDSTAT data. 1995. This progress follows directly from the
Note: Figures cover only countries for which a survey was conducted in 2010–12. Missing years were rapid decline in under-five mortality rates in
inter- or extrapolated. Africa average is population-weighted.
the region.
Even so, at 57 years, life expectancy in
to be illiterate than their urban counterparts the region remains well below the average
in richer households. Literacy is positively rate for the world (70.9 years). At the cur-
correlated with being divorced, widowed, rent annual rate of increase, it will take about
or single (20 percent less likely to be illiter- two decades to reach the levels in South Asia
ate). Illiteracy is much lower among younger (almost 67 in 2013), which lags other regions
people, holding hope for gender parity and by several years.
overall literacy levels. Healthy life expectancy in Africa was
Progress has been slow despite the rapid 49 years in 2012, 8 years less than total
increase in gross primary school enrollment, life expectancy (WHO 2015). The gender
which rose from 75 percent in 1995 to 106 gap favors women: in 2012 African women
percent in 2012. 2, 3 Despite gross primary could expect to live 1.6 years longer in good
school enrollment rates of 124 percent in health than men.4 As with literacy, the dif-
Malawi and 119 in Zambia in 2007, a stag- ferences in healthy life expectancy across
gering 73 percent of sixth graders in the two countries are significant, ranging from 39 to
POVERT Y FROM A NONMONE TARY PERSPEC TIVE 89
FIGURE 3.3 The gender gap in literacy varies widely across Africa
80
1.2
70
60 1.0
50 0.8
40 0.6
30
0.4
20
10 0.2
0 0
S e am o
ua Mh A nd
G a ia
N b ia
h i ia
ia
B o he a
Sw swa es
So a na
ria ur a
u s
M G inea
Zi aga on
Ca ba scar
e
m de
an s
Co G d a
Ca o , a
er .
Ta en n
nz ya
an a
Su da
d Eri an
ín a
A e
g i
Th a ola
in a F o
-B o
er ne u
Le a l
te Ch e
d ’ ad
n g za M e
fri M a . R e
n rita .
pu i a
Li nin
N ia
in r
ea
De biq li
Be lic
M frica
m Rep
ca u ep
An law
Gu ige
Rw oro
l G itiu
o, m a
t o a f r ic
Ug ani
Pr tre
bo bw
n g h an
yc ibi
cip
on
m u
Co Mo voir
e a as
Si Se ssa
N oth
Gu rkin Tog
K o
Et iger
r
op
Re n
e mb
ra g
t ll
be
d
Co Ver
ad ab
u t zila
m
a
i
s
I
Le
Z
m
Cô
an
Bu
é
m
lA
Eq
To
ra
o
nt
Sã
Ce
Male adult literacy ratio, 2012 Ratio of female to male literacy rate
FIGURE 3.4 Illiteracy is higher among poorer people, older people, rural dwellers, and people in
resource-rich and landlocked countries
45–49 18.8
40–44 12.8
35–39 11.5
30–34 9.2
25–29 6.8
20–24 2.2
FIGURE 3.5 Many sixth graders in Africa lack basic reading skills
a. SACMEQ reading test scores in selected countries, 2007 b. PASEC reading test scores in selected countries, 2004–09
Swaziland Gabon
Tanzania Cameroon
Kenya
Mauritius Burundi
Zanzibar Senegal
Seychelles Burkina Faso
Botswana
Zimbabwe Madagascar
Namibia Congo, Dem. Rep
Mozambique Côte d’Ivoire
Uganda
South Africa Chad
Lesotho Benin
Zambia Comoros
Malawi
SAQMEC PASEC
100 50 0 50 100 100 50 0 50 100
Percent failing Percent passing Percent failing Percent passing
Pre, emergent, and basic reading (1, 2, 3) Level 1: Students perform at or below the level
Reading for meaning (level 4) expected for random guessing (score of less than 25%)
Interpretive and above (levels 5, 6, 7, 8) Level 2: Students score between 25% and 40%
Level 3: Students perform at or above a level
determined to represent “basic knowledge”
Sources: Hungi and others 2010; World Bank estimates based on PASEC data.
Note: SACMEQ = Southern Africa Consortium for Measuring Educational Quality. PASEC = Programme d’Analyse des Systèmes Educatifs de la CONFEMEN. SAQMEC and PASEC statis-
tic are the country averages.
FIGURE 3.6 Life expectancy in Africa is rising, but it remains the 67 years (figure 3.7). Many of the countries
lowest in the world in which healthy life expectancy is shortest
are fragile or conflict-affected states. Healthy
80 life expectancy is also low in some of Afri-
ca’s oil giants, such as Angola and Nigeria.
75 Among the top performers in 2012 are the
island economies (Cabo Verde, Mauritius,
70 and the Seychelles), which recorded healthy
life expectancies of more than 60 years. Some
Age (years)
65
countries saw very little change in healthy life
60
expectancy between 2000 and 2012 (South
Africa saw no change at all). Other coun-
55 tries—including some that were in conflict in
the 1990s, such as Eritrea and Rwanda (15
50 years) and Ethiopia (11 years)—recorded sig-
nificant improvements.
45 Healthy life expectancy is related to
1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 four key variables: country income, natu-
East Asia and Pacific North America ral resources, fragility, and landlockedness.
Europe and Central Asia Sub-Saharan Africa There are clear signs of a resource curse in
Latin America and the Caribbean South Asia
Middle East and North Africa World terms of longevity (literacy is also inversely
correlated with natural resource endow-
Source: World Development Indicators. ment) (figure 3.8): on average people born
POVERT Y FROM A NONMONE TARY PERSPEC TIVE 91
70
60
50
40
Years
30
20
10
N so
M iger
A wi
i a
Zi Uga nea
ba da
C Za we
ut ero ia
S n
ru n
to a- Ma i
ria Bi li
te ui u
Iv a
SwNig ire
Co M S ilan a
ng oza om d
De bi a
m que
Ch p.
ng ad
n so la
Si Rep tho
m Ca aur lles
d Ve s
ín e
ad m e
ia
Et neg r
Rw opi l
a a
Ga nda
Gh on
Er ana
M mo ea
ri s
e ud ia
G a an
Bo Ken ia
ts y a
Libana
ut za a
Af a
T ica
Co B ogo
rk , R n
aF .
Le ic
e
in ep
nd
hi a
Se asca
an o iu
au ro
az eri
o, m ali
So Tan eri
h ni
Gufric
d’ ne
P r rd
M Na cip
on
h o
Cô l G sa
B u g o ni
B u uda
ra bl
So am mb
ca Le o
ag ib
Th S tan
e
a
a
Co itr
m n
o
b
r
é b it
.R
m
e
M he
s
al
er u
w
A
yc
n
Se
ua ine
Eq Gu
fri
lA
To
ra
o
nt
Sã
Ce
Healthy life expectancy at birth, 2012
Change in healthy life expectancy, 2000–12
in resource-rich countries have life spans FIGURE 3.8 Healthy life expectancy is lower in
that are 4.5 years shorter than people born resource-rich countries
in non-resource-rich countries (a difference
of about 10 percent), after controlling for 8
income level, fragility, and landlockedness. 6.5
Di erence in healthy life expectancy (years)
FIGURE 3.9 Vaccination rates rose and child mortality from malaria fell
Percent of population
Thousands of deaths
600
Thousands of deaths
350 70
300 60 500
250 50 400
200 40 300
150 30
200
100 20
50 10 100
0 0 0
20 0
20 1
20 2
20 3
20 4
20 5
20 6
20 7
20 8
20 9
20 0
20 1
20 2
13
13
20 0
20 1
20 2
20 3
20 4
20 5
20 6
20 7
20 8
20 9
20 0
20 1
20 2
0
0
0
0
0
0
0
0
0
0
1
1
1
0
0
0
0
0
0
0
0
0
0
1
1
1
20
Sources: Panel a: Data from Health Nutrition and Population Statistics and WHO 2014a. Panel b: Data from WHO 2014b.
Note: Immunization measles age group is 12–23 months. Immunization DTP3 age group is 12–23 months. Measles age group is 1–59 months. Pertussis age group is 0–4 years.
Tetanus age group is 0–4 years. DTP = diphtheria, tetanus, and pertussis.
19.1 in South Africa, 21.9 percent in by assessing height and weight. For adults the
Botswana, 22.9 percent in Lesotho, and 27.4 body mass index (BMI)—the ratio of weight
percent in Swaziland). Prevalence rates of 5–7 to height—is often used. Very low BMIs are
percent are observed in East Africa (Kenya, indicative of undernourishment; high BMI is
Tanzania, and Uganda) (map 3.1). Despite how obesity is often defined. Systematic BMI
substantial progress and the increased avail- measures are not available for men. Among
ability of better treatment options, HIV/AIDs women in Africa, 13 percent are underweight
will continue to hold back life expectancy in and 5 percent are obese (population-weighted
a number of countries, especially in Southern averages from Demographic and Health Sur-
but also in East Africa. veys 2006–12).
Nutrition. A healthy life is also reflected in Underweight is less common in middle-
good nutritional status, commonly measured income countries and more prevalent in
Cabo Mauritania
Verde
Mali Niger
Senegal Sudan Eritrea
The Gambia Chad
Guinea-Bissau Burkina Faso
Guinea Benin
Nigeria
Côte Ethiopia
Sierra Leone d’Ivoire Ghana Central African South Sudan
Republic
Liberia Cameroon Somalia
Togo
Equatorial Guinea
Uganda
São Tomé and Príncipe Rep. of Kenya
Gabon Congo
Rwanda
Dem. Rep. of Burundi
Congo
Tanzania
Seychelles
Percent of population
that is HIV+ Comoros
0.4 Angola
Malawi
0.5 to 5.5 Zambia
5.6 to 10.6
10.7 to 15.7 Mozambique
Zimbabwe Madagascar Mauritius
15.7 and higher
Namibia
No data Botswana
Swaziland
South Lesotho
Africa
IBRD 41866
SEPTEMBER 2015
Source: DHS Statcompiler from latest Demographic and Health Surveys.
94 POVERTY IN A RISING AFRICA
fragile states and, especially, resource-rich more malnourished than girls (39.5 percent
countries (where it is 3.7 percentage points compared with 35.2 percent). This differ-
higher than for non-resource-rich coun- ence largely reflects biological differences in
tries) (figure 3.10). This finding holds even health and survival between boys and girls
after controlling for other country and (Kraemer 2000; Waldron 1983). If this bio-
household features, suggesting that policy logical disadvantage is not offset by cultural
choices underpin this poor health outcome preferences for boys (as in Asia), higher mal-
in resource-rich countries. Malnutrition is nutrition rates among boys result (Wamani
more prevalent among poor households (by and others 2007).
3.2 percentage points) and in rural areas (by The prevalence of stunting is high in
1.6 percentage points). It declines with educa- Burundi (57 percent), Madagascar (50 per-
tion. Widows, divorcées, and single women cent), and Africa’s most populous countries—
are at significantly greater risk than married Nigeria (37 percent), Ethiopia (44 percent),
women of being undernourished (by 2.7 per- and the Democratic Republic of Congo (42
centage points). The role of marital status percent). Only two countries (Gabon and
in women’s health capabilities is an under- Senegal) register rates under 20 percent. The
appreciated aspect of well-being in Africa overall level of development of a country mat-
and highlights the importance of indicators ters for child nutrition, though other factors
of individual well-being (van de Walle and are likely even more important (Harttgen,
Milazzo 2015). Klasen, and Vollmer 2013). Children born
Trends in obesity suggest that poor to educated women enjoy chances of proper
nutritional habits are accompanying rising growth development that are 9.9 percent-
incomes. The condition is most prevalent age points higher (for secondary education)
among highly-educated women, women in and 19.8 percentage points higher (for higher
urban settings, and women in middle-income education). Children in poor, rural house-
countries. Based on an extrapolation of the holds with undernourished mothers are 20
data shown in figure 3.10, the total num- percent more likely to be stunted. Everything
ber of obese adults in Africa (both men and else equal, being born in a fragile or resource-
women) is estimated at 26.7 million. The fig- rich country also reduces one’s chances of
ure is likely to reach epidemic proportions proper early childhood growth. A continued
in the near future, presenting Africa with a focus on increasing education among women
new health challenge (Popkin 2001; Ziraba, will have dramatic and long-lasting effects on
Fotso, and Ochako 2009). Africa’s human capabilities.
The long-run nutritional status of young Physical impairment and disability also
children, often measured by low height-for- deprive people of opportunities (capabili-
age (stunting) provides an important addi- ties) and the ability to do and be what they
tional indicator of a population’s capability value (their functionings) (Mitra 2006). As a
of living a long and healthy life as well as group, the disabled are typically either under-
an outlook on the future.7 Chronically mal- sampled or poorly identified in representative
nourished children face a higher risk of mor- surveys and as a result often understudied.
tality and disease. Early growth retardation From a sample of seven countries across
also impedes cognitive development and Africa on which comparable data are avail-
schooling achievements (Dercon and Port- able, it appears that almost 1 working-age
ner 2014). adult in 10 in Africa suffers from a disabil-
The prevalence of stunting has been ity, defined as reporting severe difficulties
declining across Africa. It fell from 44.6 per- in moving about, concentrating, remember-
cent in 1995 to 38.6 percent in 2012 (DHS ing, seeing or recognizing people across the
2015). Unlike in Asia, where there is a strong road (while wearing glasses), or taking care
cultural preference for boys, who are there- of themselves (figure 3.11). The prevalence of
fore better fed, in Africa boys under five are disability ranges from 5.3 percent in Kenya
POVERT Y FROM A NONMONE TARY PERSPEC TIVE 95
FIGURE 3.10 Many factors contribute to underweight and obesity in African women
–5.7
Low-middle income compared Low-middle income compared
with low income with low income 4.8
Upper-middle and high income –1.5 Upper-middle and high income 8.3
compared with low income compared with low income
Urban compared with rural –1.6 Urban compared with rural 4.1
Poor compared with nonpoor 3.2 Poor compared with nonpoor –4.6
–1.9 3.6
Attended primary education Attended primary education
–8 –6 –4 –2 0 2 4 6 8 10 –8 –6 –4 –2 0 2 4 6 8 10
Percentage point difference Percentage point difference
FIGURE 3.11 About 1 in 10 Africans suffers from the past year; and 33 percent report that they
a disability or a family member had feared crime in their
home at least once in the past year.
14
Freedom from political violence. After
12 years of multiple large-scale conflicts in the
Prevalernce of disability (%)
10
1990s, Africa enjoyed a period of relative
peace during the first decade of the 21st cen-
8 tury (map 3.2). Between 1997 and 2014, the
6 number of violent events against civilians
more than quadrupled, reaching more than
4
4,000 in 2014. The number of victims per
2 event declined (from 20 during the late 1990s
0 to 4 in 2014), however, reflecting the chang-
ing nature of the events. The more conven-
us
a
i
e
aw
ny
an
bi
bw
as
m
aF
al
Ke
Gh
ba
au
Za
M
in
rk
Zi
Bu
MAP 3.2 The number of violent events against civilians is increasing, especially in Central Africa and the Horn
Source: Armed Conflict Location and Events Dataset (ACLED); Raleigh and others 2010.
Note: Maps indicate annual number of violent events against civilians; numbers in parentheses indicate number of countries. No data are available for Cabo Verde, Comoros,
Mauritius, São Tomé and Príncipe, and Seychelles.
(Institute of Statistics and Economic Studies domestic violence may also reflect broader
2009). Conflict has also held back progress social norms toward violence and gender
toward reducing under-five mortality and roles.
increasing life expectancy (figure 3.12). Domestic violence affects more than 700
Freedom from domestic violence. Physi- million women across the world. Africa and
cal and sexual violence (and the threat of South Asia have the largest shares of women
such violence) at home are negatively associ- in partnerships who have been victims of
ated with health outcomes, empowerment, domestic violence—an astounding 40 per-
employment trajectories, and the ability to cent in Africa and 43 percent in South Asia
engage in productive activities (Campbell (World Bank 2014). North America has the
2002; Coker, Smith, and Fadden 2005; Duflo lowest share (21 percent).
2012; MacQuarrie, Winter, and Kishor Acceptance of domestic violence is mea-
2013; Nyamayemombe and others 2010; sured by attitudes reported by women toward
Stöckl, Heise, and Watts 2012; Vyas 2013; domestic violence. Women are considered
Wayack, Gnoumou, and Kaboré 2013). The accepting of domestic violence if they respond
effects also extend well beyond the direct that husbands are justified in beating their
victims. Children’s health and educational wives if the wives do any of the following:
achievements are impeded, and social norms go out without telling the husband, neglect
that condone violence perpetuate it (Rico the children, argue with the husband, refuse
and others 2011). A child whose mother to have sex, or overcook food. Between
experienced domestic violence is more likely 2000–06 and 2007–13, acceptance of domes-
to become a victim or a perpetrator of such tic violence by women in Africa declined by
violence later in life (Kishor and Johnson almost 10 percentage points (figure 3.13);
2004). The incidence of and attitudes toward the incidence of domestic violence, which is
98 POVERTY IN A RISING AFRICA
Africa’s refugee population peaked at about 6.5 mil- Fourteen percent of IDPs, 4 percent of returnees,
lion people in 1994 following the Rwandan geno- and 1 percent of refugees reported that they had
cide. It declined to 3.5 million in the late 1990s experienced death or physical violence within their
and 2.8 million in 2008, following the end of the household. Overall, better-educated and wealthier
genocide and the decline in large-scale conflicts in households managed to flee the conflict area; poorer
Southern and West Africa. The number of refugees people had to stay behind. Among people who
increased again in 2010–13, to 3.7 million. Add- returned by 2014, mainly IDPs, escape seemed to
ing the estimated 12.5 million internally displaced have helped them mitigate the effects of violence.
persons (IDPs) brings the total number of African They suffered less than the average population of
people displaced by conflict to about 16.2 million northern Mali. But many people also responded to
at the end of 2013, or about 2 percent of the total the crisis by leaving the country, and refugee situa-
population. (Estimates of the number of refugees are tions often become protracted, extending the suffer-
from the United Nations High Commissioner for ing (Kreibaum 2014).
Refugees; estimates of the number of IDPs are from Over the past decade there has been an expan-
the International Displacement Monitoring Centre sion of microhousehold studies examining the evo-
[see Maystadt and Verwimp 2015 for a discussion].) lution of well-being among refugees, host commu-
The Greater Horn of Africa and Central Africa nities, and returnees. These studies show refugees
(especially the Democratic Republic of Congo) have also as economically active people who often engage
been major sources of refugees. In some countries in entrepreneurship; they are not always worse off
(Somalia, South Sudan, and Sudan), refugees have than nonrefugees or the hosting communities, partly
fled in the wake of extreme weather shocks and not because of the support received. Singh and others
only due to conflict (Calderone, Headey, and May- (2005) find, for example, no difference in under-
stadt 2014; Gambino 2011; Maystadt and Ecker five mortality rates between refugee and nonrefugee
2014; O’Loughlin and others 2012). households in western Uganda and South Sudan. In
Most African refugees remain in Africa. Since contrast, Verwimp and Van Bavel (2005) find higher
2005 the region has also been receiving a large under-five mortality rates and fertility among (for-
inflow of refugees from North Africa and, since mer) Rwandan refugees in the Democratic Repub-
2013, Iraq, Syria, and Yemen, bringing the total lic of Congo. Verwimp and Van Bavel (2013) report
number of refugees in Africa to 5.6 million. a reduction in schooling among Burundi children
Socioeconomic data on refugees and IDPs dur- associated with displacement that is distinct from
ing or immediately after conflicts are scant. A recent the effects of exposure to violence.
study tracking the welfare of people displaced dur- Insights from three country case studies (of
ing the 2012 crisis in northern Mali sheds some light Kenya, Tanzania, and Uganda) suggest that the
on the consequences (Etang-Ndip, Hoogeveen, and local economy often benefits from the influx of refu-
Lendorfer 2015). Welfare losses were substantial: gees, through increased demand for local goods and
the value of durable assets fell by 20–60 percent, services and better connectivity following invest-
and the value of livestock declined by 75–90 per- ment in new roads and transport services to reach
cent. But loss of welfare and wealth is only part of the camps (Maystadt and Verwimp 2015). But not
the story. In June 2014, 52 percent of the IDPs in everyone benefits. The landless and agricultural
Bamako felt insecure on the street at night, and 30 workers, whom refugees may compete with on the
percent felt insecure during the day. The share rose labor market, and net food buyers suffer, at least in
to 85 percent among returnees in Gao and Kidal. the short run.
correlated with acceptance, also fell. Accep- Both the levels of and changes in accep-
tance of domestic violence in the region is tance of violence vary widely across countries.
still exceptionally high, however (30 percent), Women’s acceptance of domestic violence is
more than twice the average in the rest of the deeply engrained in some countries (77 per-
developing world (14 percent) (figure 3.14). cent acceptance rates in Mali and Uganda); in
POVERT Y FROM A NONMONE TARY PERSPEC TIVE 99
FIGURE 3.12 Conflict slows progress in reducing under-five mortality and increasing life expectancy in Africa
600
1,000
800
400
600
400
200
200
0 0
Source: Data from the Armed Conflict Location and Events Dataset (ACLED) and World Development Indicators.
Note: Results are population weighted (the size of each dot represents the population). Fatalities are measured for 2000–10. Under-five mortality is taken from the last Demographic
and Health Survey (DHS) in the 20th century in a country (up to 2004 if no earlier survey available) and the last DHS survey in the first decade of the 21st century (up to 2013 if no
earlier survey available).
FIGURE 3.13 The incidence and acceptance of FIGURE 3.14 Acceptance of domestic violence is twice
domestic violence in Africa has declined as high in Africa as in other developing countries
62% 41%
51%
30%
41%
32% 22%
14%
Sources: Demographic and Health Surveys 2000–13; World Development Sources: Demographic and Health Surveys 2000–13; World Development
Indicators. Indicators.
Note: Figures are population-weighted averages of ever-partnered Note: Figures are population-weighted averages of 32 African and 28 non-
women in 20 African countries. African developing countries.
others, only small minorities accept domestic point. In Mali incidence increased 8 percent,
violence (13 percent in Malawi, 16 percent but there was no change in acceptance rates.
in Benin) (figure 3.15). Declining acceptance Acceptance of domestic violence is much
does not always translate into declining inci- greater among women in resource-rich (16
dence, however. In Malawi, for example, percent) and fragile (9 percent) countries
while acceptance decreased 13 percentage (controlling for other country traits) (fig-
points, incidence rose almost 1 percentage ure 3.16). Surprisingly, tolerance of violence
100 POVERTY IN A RISING AFRICA
FIGURE 3.15 Women’s acceptance of domestic violence varies widely across countries in Africa
80
accept domestic violence
Percent of women who
60
40
20
0
nd B wi
oz ín n
Sw biq e
M az ue
ag nd
g r
m a
Ghibia
Zi so a
b o
m e
rk ibe s
C a ria
t e e ro o
d’ on
G a i re
K n
nz a
a a
an a
Se da
ng Nig l
o, er
er m .
L ia
hi e
o, Bu opia
. i
an .
da
i
Si Za Rep
Ug Rep
al
Co ega
m nd
Ni sca
Bu L oro
Le an
Ug nd
Na eri
Co abw
Et eon
Rw ani
am cip
Ta eny
m th
M Pr eni
Cô am Fas
bo
ra b
M
a
o
ad ila
De ru
al
Iv
a
n
M
in
éa
ng
m
Co
To
o
Sã
is also greater among younger women; it tolerant of domestic violence. The incidence
declines with age, possibly because its inci- of domestic violence is just 3.9 percent lower,
dence rises (domestic violence is more com- however. Africa’s upper-middle-income and
mon in the 20–35 age group than among the high-income countries have higher rates of
15–19 age group). Tolerance of violence fell domestic violence (despite lower acceptance
by 1.7 percent a year between 2000 and 2013 rates) than poorer countries. After control-
and the incidence of violence fell by 0.6 per- ling for age, educational attainment, and
cent, but no broad generational shift in mind- income, there is no discernable difference
set has yet occurred. between rural and urban areas.
A main distinguishing factor in accep-
tance is education. Better-educated women
Freedom to Decide
are 31 percent less likely to be tolerant of
domestic violence than women with no edu- The second critical dimension of the capa-
cation, and women with secondary educa- bility approach is the ability to shape one’s
tion are 16 percent less likely to be tolerant. life—that is, to determine what one values.
Education is not associated with a lower inci- This dimension concerns opportunities. A
dence of domestic violence, however. In fact, woman who cannot leave her house without
women with primary and secondary educa- her husband’s permission or who has no say
tion are more than 10 percent more likely about her own health is not free to deter-
to have experienced domestic violence than mine her choices in life. Homosexuals who
uneducated women, among whom incidence are afraid of revealing their sexual orienta-
rates are similar to rates among women with tion for fear of persecution are constrained in
higher education. their life choices.
Income reduces tolerance of domestic People gain access to a broader set of
violence, especially in upper-middle-income opportunities if they can participate in the
and high-income countries and the richest processes that affect their lives and are
segments within countries. Women in the allowed to make their own choices. These
richest quintile are 7.1 percent less likely choices are often politically and socially
than women in the poorest quintile to be constrained.
POVERT Y FROM A NONMONE TARY PERSPEC TIVE 101
FIGURE 3.16 Acceptance and incidence of domestic violence are greater among younger women and women in resource-
rich and fragile states; acceptance is also higher among uneducated women, but not incidence
30–34 –9.1
30–34 2.6
25–29 –7
25–29 4.6
20–24 –3.3
20–24 4.5
This dimension is not so much about WGI scores range from −2.5 to 2.5 units in a
democracy per se but about the degree to normal standard distribution.
which political systems give people voice and The WGI data indicate that perceptions of
participation in the processes that affect their political constraints have not changed much
lives at all levels of society. It is about not worldwide in the past few years, although
only political freedom and participation but there was a slight improvement in Africa,
also social norms and the freedom to decide albeit from low levels (figure 3.17). The
about routine matters in life, including within region is doing better than the Middle East
the household. Constraints can be based on and North Africa and East Asia and Pacific.
gender, religion, ethnicity, sexual orientation, Improvements have been especially note-
or other reasons. worthy in West Africa (Burkina Faso, Ghana,
Indicators that measure freedom to decide Liberia, Niger, and Nigeria) (figure 3.18).
are often not available, particularly at the Countries that experienced a large decline in
individual level. We draw on three measures: their voice and accountability scores include
a country-level measure of voice and account- the Central African Republic, Eritrea, Gabon,
ability, as a broad indicator of enabling the and Madagascar. The results for country
expression of voice; exposure to mass media, groupings are consistent with the findings
as an indicator of access to information to about education, health, and violence. Coun-
inform decisions; and the extent to which tries that are resource rich or fragile are less
women have control over decision making in well off than countries that are not (by −0.5
various domains of living. units each), and upper-middle-income and
The Worldwide Governance Indicators high-income countries score 0.6 points higher
(WGI) project scores countries in terms of than low-income countries, controlling for
voice and accountability. It captures percep- other country traits. The WGI findings are
tions of the extent to which a country’s pop- highly correlated with the findings of the
ulation is able to participate in selecting the Afrobarometer.8 There is no systematic dif-
government and enjoy freedom of expression, ference in perceptions of political freedoms by
freedom of association, and a free media. gender and area of residence (rural or urban).
The second measure of freedom to decide
reflects the ability to make informed deci-
FIGURE 3.17 Voice and accountability levels remain low in Africa sions. Access to media provides an impor-
tant source of information, and educational
2.5
attainment helps people digest the informa-
2.0
tion and act on it.
Voice and accountability indicator
FIGURE 3.18 Voice and accountability are stronger in middle-income countries, and often lower in resource-rich
economies
2.5
2.0
Voice and accountability indicator
1.5
1.0
0.5
0
–0.5
–1.0
–1.5
–2.0
–2.5
te ige a
ria Su lic
ui n
So au rde
M Af s
ri a
w ha s
h al
m s
M S ritr a
oz en ea
as e
Sw o, R la
.
Th Rw land
hi ia
n .R u
E r ea
í n
an a
aF e
o
er N ali
Le er
Co ibe e
m ria
an s
zil na
d Be d
s e
Se en ho
rk biq al
Af a
ad Iv a
M Ga car
ru a
Ca T d i
er o
Gu oon
Co A inea
a da
ia
Ce C G uth ab d
nt on uin S we
pu p.
rit n
ea
fri De is n
az ep
Ug oro
h iu
G tiu
Za elle
Cô N ric
d
M d ’ ri
Bu ani
E d
R w bi
au ric
in u
L on
ag oir
Le ncip
as
l G da
m og
So imb Cha
ca m sa
au bo
Pr ni
an
l A o, - B a
o
yc eg
op
Bu am eg
Et mb
Re e
n
ra ig
b
M
n
itr
e G an
ra g ea ud
S ot
ut rit
ng ng
M Ve
i
a
b o
Ca
Z
an
to
Si
ua
é
m
Eq
To
o
Sã
FIGURE 3.19 Less than half of Africa’s population has regular access to mass media
100
Percent of population with regular
90
80
exposure to mass media
70
60
50
40
30
20
10
0
Cô dag roo i
Sw am pe
í n
Be ire
Mn
Bu am fr i
Zi na Fque
ba so
Ni we
M u ria
rit ea
T ia
Co err ib o
ng a L eria
De N ne
. r
hi .
Ch ia
ad
az ibia
G h nd
K na
pi n a
w al
ld
so a
n wa o
pu da
m c
ng za a
Coo, R ia
or .
M os
Ca r u i
te a s n
Iv r
r k b ica
m ep
M m nd
Et Rep
Bu law
oz A al
m ige
d’ ca
Za bli
Le and
Co Tan bi
lo Se eny
c a R th
S i L og
Pr bo
ni
an
op
ng eg
n
Ug or
m a
au in
a
N nci
Re n
ge
o
eo
ila
b
i i
nd Ga
a e
o,
éa
M
ve
fri
m
lA
De
To
ra
o
nt
Sã
Ce
Women Men
average media access is 15 percentage points in fragile states. Increased use of cellphones
lower among women than men (54 percent can partly substitute for traditional media
versus 69 percent). Poverty, rural residence, (Aker and Mbiti 2010).
and lack of education are key differentia- The third set of indicators of freedom to
tors. Media access is also 6 percent lower in decide are measures of decision making in
resource-rich countries and 5 percent lower the lives of women from household surveys.
104 POVERTY IN A RISING AFRICA
FIGURE 3.20 Women’s participation in their own health care decisions is lower among younger women,
women in poor and rural households, and women in resource-rich and landlocked countries
45–49 –20.3
40–44 –20.0
35–39 –18.2
30–34 –15.6
25–29 –12.5
20–24 –7.6
The share of husbands who have the final in their own health care decisions tends to
say in decisions regarding their wives’ health be lower among younger women, women in
care is 21 percent in the Middle East and poor and rural households, and women in
North Africa, 39 percent in South Asia, and resource-rich and landlocked countries (fig-
46 percent in Africa. Women’s participation ure 3.20). It is greater in fragile states. That
POVERT Y FROM A NONMONE TARY PERSPEC TIVE 105
such participation increases with age is con- In Country A, 10 people are illiterate and the
sistent with the trends in women’s attitudes other 10 are in poor health. In Country B, 10
toward domestic violence. people are both illiterate and in poor health
The final decision on whether a married and the other 10 are literate and healthy.
woman can visit friends or family lies with Under the dashboard approach, which con-
the husband alone in 40 percent of African siders poverty dimension by dimension, both
households, compared with 33 percent in the countries are equally poor (10 people are
rest of the developing world. Control over a deprived in each dimension). But because
women’s earnings lies fully with someone else the deprivation associated with simultaneous
in only 10 percent of households. Overall, the deprivation in two dimensions may be worse
general trend in Africa is toward greater par- than the sum of the deprivations associated
ticipation of women in household decision- with each of them, the case could be made
making processes. that B is poorer. The dashboard approach
ignores jointness in deprivation.
Important insights regarding the degree of
Multiple Deprivation interdependency can be obtained by count-
ing the number of dimensions in which an
Thus far this chapter has examined the individual is deprived and calculating the
region’s well-being by assessing progress on shares of the population deprived in a given
each functioning and capability separately. number of dimensions (Ferreira and Lugo
Using a dashboard approach (listing achieve- 2013). This counting approach does not
ments by dimension) instead of aggregating require that weights be imposed on dimen-
the measures into an index avoids having to sions or that the degree to which deprivations
assign weights to different dimensions.9 It are substitutable be determined (Atkinson
also allows researchers to draw on different 2003). This approach is akin to the MPI
datasets. It does not require a measure of sev- proposed by Alkire and Foster (2011), but it
eral dimensions of poverty (for the same indi- does not impose a number of deprivations to
vidual or household) simultaneously. From a qualify as poor. By capturing the essence of
practical perspective, policies typically aim to the interest in multidimensional poverty, it
address shortcomings in a particular dimen- provides a middle ground between the dash-
sion (education, health care, the incidence of board approach (Ravallion 2011), which
violence). The gains from combining scores ignores jointness in deprivation, and the sca-
across dimensions to obtain a complete rank- lar MPI, which assigns a minimum number
ing may be limited. of deprivations for a person to qualify as
These advantages come at the expense of poor (Alkire and Foster 2011; Decancq and
being able to assess the extent to which people Lugo 2013).
suffer multiple deprivations. People suffering Measuring multidimensional deprivation
in different dimensions are arguably worse requires information on each dimension for
off than people suffering in one dimension. the same individual. To look at Africa wide
Omission of valuable dimensions underesti- patterns, such information is available only
mates their poverty, especially when dimen- for women of reproductive age from 25
sions are poorly correlated (that is, when they countries covered in the DHS. Proxy indica-
are poor substitutes or poor complements).10 tors for the four dimensions are used (box
In addition, the deprivation associated with 3.4). Considering each dimension separately
simultaneous deprivation in two dimensions (as in the dashboard approach), about one
may well be worse than twice the deprivation adult woman in two is illiterate (56 percent),
associated with each of them. As a result, exposed to violence (54 percent), or insuffi-
country rankings may differ when simulta- ciently empowered (51 percent), and about
neity in deprivations is considered. Consider, one in seven (14 percent) is undernourished.
for example, two countries with 20 people. For the four dimensions considered here, the
106 POVERTY IN A RISING AFRICA
BOX 3.4 Demographic and Health Surveys make it possible to measure multidimensional
poverty
average woman suffers 1.75 deprivations (56 Multiple deprivations and the concentra-
+ 54 + 51 + 14 = 175/100). tion of deprivation are more common among
Does everyone suffer equally, or is depriva- women with less wealth: 42 percent of asset-
tion concentrated among a subset of the pop- poor women versus 18 percent of non-asset-
ulation? Under a perfectly equal distribution poor women are deprived in at least three
of deprivations, everyone would be deprived
in 1.75 dimensions. Under perfect concen-
tration (or full inequality), all deprivations FIGURE 3.21 A large share of African women
would be concentrated within a single group: suffers from multiple deprivations
43.7 percent (175/4) of the population would
suffer in each of these four dimensions, while
Cumulative share of adult females (%)
100
97
the remaining 56.3 percent would be depriva- 90
tion free. The larger the share of people suf- 80 82 ↑
95
fering in three or more dimensions, the more 70 92
60 69
concentrated is the deprivation. 58
Deprivation among African women is 50
58
40
widespread: more than four women in five
30 26 38
(86 percent) are deprived in at least one
20
dimension; only 14 percent are free of depri- 10 14 22
vation (figure 3.21). Multiple deprivation 0
4
characterizes a sizable group of women: 0 1 2 3 4
almost one woman in three is poor in three or Number of dimensions deprived
four dimensions; 55 percent suffer in one or
Women who are not asset poor
two dimensions. Deprivation is widespread, Women who are asset poor
but for a sizable group it is also highly con- All women
centrated: about one-third of women realize
only one functioning or none at all. Source: Data from Demographic and Health Surveys 2005–13.
POVERT Y FROM A NONMONE TARY PERSPEC TIVE 107
dimensions. But three out of four non-asset- experience on average half the deprivation
poor women suffer at least one deprivation, of women 15–19 (figure 3.22). After con-
confirming that income poverty provides trolling for education and illiteracy, toler-
only a partial picture of a population’s ance for domestic violence and social control
well-being. over one’s actions tend to decline with age.
Multiple deprivation is more prevalent This evidence suggests that there is a posi-
among younger women: women 35– 49 tive dynamic as life progresses, but it is also
FIGURE 3.22 Multidimensional poverty is more prevalent among young women, divorced women, poor
women, rural women, and women living in low-income, fragile, and resource-rich countries
45–49 –0.1
40–44 –0.2
35–39 –0.2
30–34 –0.2
25–29 –0.2
20–24 –0.1
MAP 3.3 Multiple deprivation is substantial in the Western Sahel and Africa’s populous countries
Cabo Mauritania
Verde
Mali Niger
Senegal Sudan Eritrea
The Gambia Chad
Guinea-Bissau Burkina Faso
Guinea Benin
Nigeria
Côte Ethiopia
Sierra Leone d’Ivoire Ghana Central African South Sudan
Republic
Liberia Cameroon Somalia
Togo
Equatorial Guinea
Uganda
São Tomé and Príncipe Rep. of Kenya
Gabon Congo
Rwanda
Dem. Rep. of Burundi
Congo
Tanzania
Percent of women deprived in at
Seychelles
least three dimensions
Comoros
0%–10% (4 countries) Angola
Malawi
10%–20% (7 countries) Zambia
20%–30% (7 countries)
30%–40% (1 countries) Mozambique
Zimbabwe Madagascar Mauritius
40%–100% (6 countries)
Namibia
No data Botswana
Swaziland
South Lesotho
Africa
IBRD 41869
SEPTEMBER 2015
indicative of the strong persistence of cultural in West Africa and the Sahel (Guinea, Mali,
habits across generations. Niger) display high levels of multiple depri-
Poor women experience 0.6 more depri- vation, as do Africa’s most populous coun-
vations than rich women, and rural women tries (map 3.3): the share of women suffering
suffer 0.5 more deprivations than urban three or more deprivations is 68 percent
women, holding other factors constant. As in Ethiopia, 40 percent in the Democratic
these women also have lower levels of educa- Republic of Congo, and 22 percent in Nige-
tion and poverty is more prevalent in rural ria. High rates of multiple deprivation in
areas, the unadjusted gaps are much larger. these populous countries partly explain the
Multiple deprivations are also more com- large share of multidimensional poverty in
mon in low-income, fragile, and resource- Africa, where 31 percent of women in the
rich states. Multiple deprivation is 10 percent 25 countries studied are deprived in three
higher in resource-rich countries. Countries dimensions or more.
POVERT Y FROM A NONMONE TARY PERSPEC TIVE 109
Considering the share of women deprived in one, The multidimensional poverty rate (H) is the share of
two, three, … k dimensions (with k the total num- the population that is poor in at least k dimensions.
ber of dimensions considered) is similar to one of Alkire and Foster also consider the intensity of
the family of multidimensional poverty measures deprivations (A), the average number of dimensions
proposed by Alkire and Foster (2011). They use two in which the multidimensionally poor are deprived.
thresholds to determine whether a person is multi- Adjusting the multidimensional poverty rate (H) for
dimensionally poor: a dimension-specific cutoff deprivation intensity (A) helps differentiate countries
to determine whether he or she is deprived in each with an equal share of multidimensionally poor. A
dimension and a dimension threshold (k) that is the country in which 30 percent of women have three
number of dimensions in which a person needs to be deprivations and none has four would rank higher
deprived to be considered multidimensionally poor. than a country in which 30 percent of women are
Relative rather than equal weighting of the dimen- multidimensionally poor but half of them suffer four
sions can be applied. The second cutoff is then a pro- deprivations. The MPI can then be written as M =
portion (not the number) of weighted deprivations. H × A.
The approach adopted here is similar to In Alkire/Foster notation, figure 3.23 ranks
the MPI approach proposed by Alkire and countries based on a multidimensional pov-
Foster (2011) (box 3.5). To illustrate this sim- erty rate based on k of 3, with no adjust-
ilarity, figure 3.23 displays the share of the ment for intensity of deprivation (A). Using
population in each country that is deprived in the MPI, that is adjusting the results in figure
one, two, three, and four dimensions. Coun- 3.23 for A, does not change the ranking.
tries are ranked by the share of the popula- Mitra, Posärac, and Vick (2013) use this
tion deprived in three or more dimensions. approach to compare poverty among abled
FIGURE 3.23 Country ranking changes only slightly when the dimension threshold changes
100
90
Percent of adult women
80
70
60
50
40
30
20
10
0
az e
Ga d
Le n
Na ho
ia
a
am ya
Co que
Lib s
T a ri a
m a
U g on
Ni a
Za ia
a
i
ng in re
m o
ra p.
E t ea
ia
Gu i
te nd
al
ge
o
an
Ca ani
bi
d
Sw cip
on
bo
ni
De as
n
op
ib
er R e
Co urk Ivoi
or
o z K en
in
ge
M
an
e
t
ila
m
Be
Cô uru
Ni
so
o, a F
Gh
m
Le
bi
er
ín
nz
m
hi
.
B d’
Pr
B
nd
Si
éa
M
m
To
o
Sã
and disabled populations. They include 10 the rate of progress has leveled off. Despite
dimensions capturing both monetary and substantial increases in school enrollments,
nonmonetary aspects of poverty among more than two out of five adults in Africa
individuals (primary school completion, cannot read or write, and the quality of
employment) and households (nonhealth schooling is poor. Improving Africa’s pri-
expenditures, the ratio of health expendi- mary educational outcomes is urgent. Health
tures to total expenditures, and six indi- outcomes mirror the results for literacy. Prog-
cators of assets, amenities, and housing ress is happening, but outcomes are still the
conditions). People are considered poor if the worst in the world. Increases in immuniza-
weighted sum of their deprivations in each tion and bednet coverage are slowing. Nearly
of these dimensions exceeds 40 percent. In two in five African children is malnourished,
the seven countries in their sample, the MPI one in eight adult women is underweight, and
is on average 7.2 percent larger for people obesity is emerging as a new health concern.
with disabilities. The difference is largest in Africans enjoyed considerably more peace
Kenya (12 percent) and smallest in Malawi in the 2000s than before, but since 2010 the
(5 percent). number of violent events has been four times
what it was in the mid-1990s. Violence in
Africa is experienced not only in terms of
Concluding Remarks political unrest and large-scale civil conflicts
This chapter reviews Africa’s progress since but also in the form of domestic violence. At
the mid-1990s in a number of nonmone- 30 percent, tolerance of domestic violence is
tary dimensions of poverty. The dimensions twice as high as in the rest of the developing
include education and health, two focus areas world and the incidence of domestic violence
of the Millennium Development Goals, as is more than 50 percent higher. Higher toler-
well as freedom from violence and freedom ance of domestic violence and less empowered
to decide. Wider data availability makes this decision making among younger (compared
possible, though some measurement issues with older) women suggest that a generational
remain, even when tracking traditional indi- shift in mindset is still to come. On voice and
cators, such as adult literacy. Progress has accountability, Africa remains among the
been achieved in all four domains, albeit with bottom performers, albeit with slightly higher
wide variation across countries and popula- scores than countries in the Middle East and
tion groups. North Africa and East Asia and Pacific.
Between 1996 and 2012, Africa’s adult Around these region-wide trends is
literacy rates rose 4 percentage points, the remarkable variation across countries and
gender gap shrank, and gross primary enroll- population groups. Rural populations and
ment rates increased dramatically. Life the income poor are worse off in all domains,
expectancy at birth rose 6.2 years, and the though other factors, such as gender and
prevalence of chronic malnutrition among female education, often matter as much
under-five-year-olds fell 6 percentage points or more and at times in unexpected ways.
(to 38.6 percent). The number of deaths from Women in Africa can, for example, expect
politically motivated violence declined, and to live in good health 1.6 years longer than
tolerance and the incidence of gender-based men, and boys under five years are 5 per-
domestic violence dropped 10 percentage centage points more likely to be malnour-
points each. Scores on the voice and account- ished than girls. At the same time, the gender
ability indicators rose slightly, and women’s gap in literacy remains substantial, women
participation in household decision-making suffer more than men from violence (espe-
processes increased. cially domestic violence), and they are more
This progress notwithstanding, levels of curtailed in their access to information and
deprivation remain high in all domains and decision making. Literacy is especially low
POVERT Y FROM A NONMONE TARY PERSPEC TIVE 111
in West Africa, where gender disparities are choices are not available, many other oppor-
large. High HIV prevalence rates are hold- tunities remain inaccessible.”
ing life expectancy back in Southern Africa. 2. UNESCO (2015) discusses reasons for lim-
Conflict events are concentrated in the ited progress in global adult literacy since
the 2000s, including the underperformance
greater Horn of Africa and the Democratic
of adult literacy programs. All progress has
Republic of Congo. The low levels of Africa’s come from better literacy among younger
capability achievements are driven partly by cohorts.
below-average performance in its three most 3. The gross enrollment ratio can exceed 100
populous countries (Nigeria, the Democratic percent because of the inclusion of over-age
Republic of Congo, and Ethiopia). Multiple and under-age students following early or
deprivations characterize life for a sizable late school entrance and grade repetition.
share of African women (data on men are 4. Higher life expectancy for women is possible
unavailable). even in an environment that is disadvanta-
Two important findings stand out. First, geous to them, given that women are geneti-
cally predisposed to live longer (Sen 2002;
fragile and resource-rich countries tend to
World Bank 2011).
perform worse and middle-income countries 5. The results are based on a country fixed-
better than other countries. This finding con- effect regression analysis of life expectancy
firms the pernicious effects of conflict and in 2000–12 in 39 countries on the under-five
is consistent with the widely observed asso- mortality rate, the HIV prevalence rate, an
ciation with overall economic development. indicator variable taking the value of 1 if
People in resource-rich countries experience the average annual number of deaths from
a resource penalty in their human develop- conflict in the five years preceding the year
ment outcomes. They are less literate (by of recorded life expectancy exceeded 100,
and GDP (in constant 2005 U.S. dollars
3.1 percentage points), have shorter average
per capita) and its square. De Walque and
life spans (by 4.5 years) and higher rates of
Filmer (2013) also find no effect of GDP on
malnutrition among women (by 3.7 percent- adult mortality in Africa and relatively little
age points) and children (by 2.1 percentage effect of recent conflict, unless the conflicts
points), suffer more from domestic violence escalated, as in the Rwandese genocide. Else-
(by 9 percentage points), and have less voice where in the world GDP is negatively corre-
and accountability than people in non- lated with adult mortality.
resource-rich countries.11 6. The increase in funding has slowed in recent
Women’s education (secondary schooling years, causing both the increase in the use of
and above) makes a decisive difference across treated bednets and the decline in child mor-
tality from malaria to level off (WHO 2013,
dimensions (health, violence, and freedom in
2014b).
decision), among both adults and children. 7. Children are considered stunted if their
Improving women’s education and socioeco- height-for-age is more than two standard
nomic opportunities can be game changing deviations from the median height-for-age of
for Africa’s capability achievement. the reference population.
8. There is a high correlation between the
WGI’s voice and accountability score and
Notes the responses from 35 African countries to
1. UNDP (1990, page 10) describes the HDI as the Afrobarometer’s “freedom to say what
follows: “Human development is a process you think” (0.67) and “freedom to join any
of enlarging people’s choices. In principle, political organization” (0.65) questions; the
these choices can be infinite and change over correlation with “the extent of democracy”
time. But at all levels of development, the is 0.58. Because the Afrobarometer does not
three essential ones are for people to lead a measure free media but only exposure to
long and healthy life, to acquire knowledge mass media, the correlation with the WGI’s
and to have access to resources needed for a voice and accountability score is slightly
decent standard of living. If these essential lower.
112 POVERTY IN A RISING AFRICA
9. The debate about defining weights is lively Alkire, Sabina. 2008. “Choosing Dimensions:
(see Alkire and Foster 2011 and critiques The Capability Approach and Multidimen-
by Ravallion 2011). Some of it concerns sional Poverty.” MPRA Paper 8862, Munich
whether deprivations should be treated Personal RePEc Archive.
as substitutes or complements (Bourgui- Alkire, Sabina, and James Foster. 2011. “Under-
gnon and Chakravarty 2003). Appropriate standings and Misunderstandings of Multi-
weights should reflect ethically or empiri- dimensional Poverty Measurement.” Journal
cally grounded trade-offs among the compo- of Economic Inequality 9 (2): 289–314.
nents of deprivation (see Decancq and Lugo Alkire, Sabina, and Maria Emma Santos. 2014.
2013; Ferreira and Lugo 2013); they should “Measuring Acute Poverty in the Developing
not be set for the sake of convenience. World: Robustness and Scope of the Multi-
10. At the country level, there is limited cor- dimensional Poverty Index.” World Develop-
relation in the population shares of people ment 59: 251–74.
deprived in the four dimensions. The cor- Atkinson, Anthony B. 2003. “Multidimensional
relation coefficient is 0.22 on average (in Deprivation: Contrasting Social Welfare and
absolute value); it ranges from 0.12 (for the Counting Approaches.” Journal of Economic
correlation between the voice and account- Inequality 1 (1): 51–65.
ability indicator and the illiteracy indicator) Bourguignon, François, and Satya R. Chakra-
to 0.39 (for the correlation between the voice varty. 2003. “The Measurement of Multi-
and accountability indicator and the indica- dimensional Poverty.” Journal of Economic
tor of the number of fatalities from violence). Inequality 1 (1): 25–49.
This low correlation is consistent with lack Calderone, Margherita, Derek Headey, and Jean-
of interchangeability across functionings François Maystadt. 2014. Enhancing Resil-
and capabilities (as emphasized by the capa- ience to Climate-Induced Conflict in the Horn
bility approach). The overlap is greatest in of Africa, vol. 12. International Food Policy
the prevalence of $1.25 income poverty (33 Research Institute, Washington, DC.
percent) for asset-poverty and each of the Campbell, Jacquelyn C. 2002. “Health Conse-
other four dimensions, which could be seen quences of Intimate Partner Violence.” Lancet
as providing support for the welfarist (mone- 359 (9314): 1331–36.
tary poverty) approach to measuring poverty Chiappori, Pierre-André, and Costas Meghir.
(that asset poverty is an indicator of multiple 2015. “Intrahousehold Inequality.” In Hand-
deprivation). Yet, even though the overlap is book of Income Distribution, vol. 2, edited by
highest, the correlation remains nonetheless Anthony B. Atkinson and François Bourgui-
rather low, underscoring that income pov- gnon, 1369–418, Amsterdam: Elsevier.
erty remains a rather incomplete proxy for Christiaensen, Luc, and David Stifel. 2007.
well-being and that good scores on income “Tracking Poverty over Time in the Absence of
poverty hide deprivation in many basic func- Comparable Consumption Data.” World Bank
tionings and capabilities. Economic Review 21 (2): 317–41.
11. De la Brière and others (2015) discuss how Coker, Ann L., Paige H. Smith, and Mary K. Fad-
resource-rich countries could harness their den. 2005. “Intimate Partner Violence and
mineral wealth to build more human capital. Disabilities among Women Attending Family
Practice Clinics.” Journal of Women’s Health
14 (9): 829–38.
References Deaton, Angus. 2008. “Income, Health and Well-
ACLED (Armed Conflict Location and Event Being around the World: Evidence from the
Data) Project. n.d. http://www.acleddata.com Gallup World Poll.” Journal of Economic Per-
/about-acled/. spective 22 (2): 53–72.
Afrobarometer. 2010 –12. http://www.afro Decanq, Koen, Marc Fleurbaey, and Erik Schok-
barometer.org. kaert. 2015. “Inequality, Income and Well-
Aker, Jenny C., and Isaac Mbiti M. 2010. Being.” In Handbook of Income Distribution,
“Mobile Phones and Economic Development vol. 2, edited by Anthony B. Atkinson and
in Africa.” Journal of Economic Perspectives François Bourguignon, 67–140, Amsterdam:
24 (3): 207–32. Elsevier.
POVERT Y FROM A NONMONE TARY PERSPEC TIVE 113
Decancq, Koen, and María Ana Lugo. 2013. J. Arrow, Amartya Sen, and Kotaro Suzumura,
“Weights in Multidimensional Indices of Well- 687–728. Amsterdam: Elsevier.
Being: An Overview.” Econometric Reviews Frey, Bruno S., and Alois Stutzer. 2002. “What
32 (1): 7–34. Can Economists Learn from Happiness
de la Brière, Benedicte, Deon Filmer, Dena Rin- Research?” Journal of Economic Literature
gold, Dominic Rohner, Karelle Samuda, 40 (2): 402–35.
and Anastasiya Denisova. 2015. From Mines Gambino, Tony. 2011. “Democratic Republic of
to Minds: Turning Sub-Saharan’s Mineral the Congo.” Background paper for the World
Wealth into Human Capital. Washington, Development Report 2011. World Bank,
DC: World Bank. Washington, DC.
Dercon, Stefan, and Catherine Portner. 2014. Harttgen, Kenneth, Stephan Klasen, and Sebas-
“Live Aid Revisited: Long-Term Impacts of the tian Vollmer. 2013. “Economic Growth and
1984 Ethiopian Famine on Children.” Journal Child Undernutrition in Sub-Saharan Africa.”
of the European Economic Association 12 (4): Population and Development Review 39 (3):
927–48. 397–412.
de Walque, Damien, and Deon Filmer. 2012. Hungi, Njora, Demus Makuwa, Kenneth Ross,
“The Socioeconomic Distribution of Adult Mioko Saito, Stephanie Dolata, Frank van
Mortality during Conflicts in Africa.” Peace Cappelle, Laura Paviot, and Jocelyne Vellein.
Economics, Peace Science, and Public Policy 2010. “SACMEQ III Project Results: Pupil
18 (3): 1–12. Achievement Levels in Reading and Math-
———. 2013. “Trends and Socioeconomic Gradi- ematics.” Working Document 1, Southern
ents in Adult Mortality around the Developing Africa Consortium for Measuring Educational
World.” Population and Development Review Quality (SACMEQ), Harare, Zimbabwe.
39 (1): 1–29. http://www.sacmeq.org/sites/default/files/sac-
DHS (Demographic and Health Surveys). Various meq/reports/sacmeq-iii/working-documents
years. ICF International, Calverton, MD. /wd01_sacmeq_iii_results_pupil_achievement
Duflo, Esther. 2012. “Women’s Empowerment .pdf.
and Development.” Journal of Economic Lit- Institute of Statistic and Economic Studies. 2009.
erature 50 (4): 1051–79. Welfare and Poverty in Rural Burundi: Results
Easterlin, Richard. 1974. “Does Economic of the Priority Survey, Panel 2007. Bujum-
Growth Improve the Human Lot? Some bura, Burundi.
Empirical Evidence.” In Nations and House- Kishor, Sunita, and Kiersten Johnson. 2004. Pro-
holds in Economic Growth: Essays in Honor filing Domestic Violence: A Multi-Country
of Moses Abramovitz, edited by Paul A. David Study. Calverton, MD: ORC Macro.
and Melvin W. Reder. New York: Academic Kraemer, Sebastian. 2000. “The Fragile Male.”
Press. British Medical Journal 321 (7276): 1609–12.
EDSTAT. Education Statistics. World Bank, Kreibaum, Merle. 2014 “Their Suffering, Our
Washington, DC. http://datatopics.worldbank Burden? How Congolese Refugees Affect the
.org/education. Ugandan Population.” Households in Con-
Etang-Ndip, Alvin, Johannes G. M. Hoogeveen, flict Network Working Paper 181, Institute
and Julia Lendorfer. 2015. “Socioeconomic of Development Studies, University of Sussex,
Impact of the Crisis in North Mali on Dis- Brighton, UK.
placed People.” Policy Research Working Loewenstein, George, and Peter A. Ubel. 2008.
Paper 7253, World Bank, Washington, DC. “Hedonic Adaptation and the Role of Decision
Ferreira, Francisco H. G., and María Ana Lugo. and Experienced Utility in Public Policy.” Jour-
2013. “Multidimensional Poverty Analysis: nal of Public Economics 92 (8–9): 1795–810.
Looking for a Middle Ground.” World Bank MacQuarrie, Kerry L., Rebecca Winter, and
Research Observer 28 (2): 220–35. Sunita Kishor. 2013. Spousal Violence and
Filmer, Deon, and Kinnon Scott. 2012 “Assessing HIV: Exploring the Linkages in Five Sub-
Asset Indices.” Demography 49 (1): 359–92. Saharan African Countries. DHS Analytical
Foster, James E. 2011. “Freedom, Opportunity Study 3. Calverton, MD: ICF International.
and Well-Being.” In Handbook of Social Max-Neef, Manfred A., Antonio Elizalde, and
Choice and Welfare, vol. 2, edited by Kenneth Martin Hopenhayn. 1991. Human Scale
114 POVERTY IN A RISING AFRICA
Development: Conception, Application and ———. 2012. “Poor, or Just Feeling Poor? On
Further Reflections. New York: Apex Press. Using Subjective Data in Measuring Poverty.”
Maystadt, Jean-François, and Olivier Ecker. Policy Research Working Paper 5968, World
2014. “Extreme Weather and Civil War: Does Bank, Washington, DC.
Drought Fuel Conflict in Somalia through Rico, Emily, Bridget Fenn, Tanya Abramsky, and
Livestock Price Shocks?” American Journal of Charlotte Watts. 2011. “Associations between
Agricultural Economics 96 (4): 1157–82. Maternal Experiences of Intimate Partner
Maystadt, Jean-François, and Philip Verwimp. Violence and Child Nutrition and Mortality:
2015. “Forced Displacement and Refugees in Findings from Demographic and Health Sur-
Sub-Saharan Africa: An Economic Inquiry.” veys in Egypt, Honduras, Kenya, Malawi and
Background paper prepared for this study. Rwanda.” Journal of Epidemiology and Com-
World Bank, Washington, DC. munity Health 65 (4): 360–67.
Milazzo, Annamaria, and Dominique van de Robey ns , I ng rid. 20 05. “T he Capabilit y
Walle. 2015. “Women Left Behind? Poverty Approach: a Theoretical Survey.” Journal of
and Headship in Africa.” Policy Research Human Development 6 (1): 93–117.
Working Paper 7331, World Bank, Washing- SACMEQ (Southern and Eastern Africa Consor-
ton, DC. tium for Monitoring Educational Quality).
Mitra, Sophie. 2006. “The Capability Approach 2007. United Nations Educational, Scientific
and Disability.” Journal of Disability Policy and Cultural Organization (UNESCO), Inter-
Studies 16 (4): 236–47. national Institute for Educational Planning,
Mitra, Sophie, Aleksandra Posärac, and Brandon Paris.
Vick. 2013. “Disability and Poverty in Devel- Sahn, David, and David Stifel. 2000. “Poverty
oping Countries: A Multidimensional Study.” Comparisons over Time and across Coun-
World Development 41: 1–18. tries in Africa.” World Development 28 (12):
Nya mayemomb e , C a rol i ne , C . B ened ic t , 2123–55.
V. Mishra, M. Gwazane, S. Rusakaniko, and Sandel, Michael J. 2012. What Money Can’t Buy:
P. Mukweza. 2010. “The Association between The Moral Limits of Markets. New York: Far-
Violence against Women and HIV: Evidence rar, Straus and Giroux.
from a National Population-Based Survey in Sen, Amartya. 1980. “Equality of What?” In The
Zimbabwe.” Zimbabwe Working Paper 4, ICF Tanner Lectures on Human Values, vol. 1,
Macro, Calverton, MD. edited by MacMurrin, Sterling M., 195–220,
O’Loughlin, John, Frank D. W. Witmer, Andrew Cambridge, UK: Cambridge University Press.
M. Linke, Arlene Laing, Andrew Gettelman, ———. 1985. Commodities and Capabilities.
and Jimy Dudhia. 2012. “Climate Variability Amsterdam: North-Holland.
and Conflict Risk in East Africa, 1990–2009.” ———. 1999. Development as Freedom. New
Proceedings of the National Academy of Sci- York: Anchor Books.
ences 109 (45): 18344–49. ———. 2002. “Why Health Equity?” Health
Oswald, Andrew J., and Nattavudh Powdthavee. Economics 11 (8): 659–66.
2008. “Death, Happiness, and the Calculation Singh, Kavita, Unni Karunakara, Gilbert Burn-
of Compensatory Damages.” Journal of Legal ham, and Kenneth Hill. 2005. “Forced
Studies 37: S217–S251. Migration and Under-Five Mortality: A Com-
PASEC (Programme d’Analyse des Systèmes parison of Refugees and Hosts in North-
Educatifs de la CONFEMEN). http://www western Uganda and Southern Sudan.”
.confemen.org/le-pasec/. European Journal of Population/Revue Euro-
Popkin, Barry M. 2001. “The Nutrition Transi- péenne de Démographie 21 (2–3): 247–70.
tion and Obesity in the Developing World.” Stevenson, Betsey, and Justin Wolfers. 2008.
Journal of Nutrition 131 (3): 871S–873S. “Economic Growth and Subjective Well-Being:
Raleigh, Clionadh, Andrew Linke, Håvard Reassessing the Easterlin Paradox.” Brookings
Hegre, and Joakim Karlsen. 2010. “Introduc- Papers on Economic Activity, 1–87.
ing ACLED: Armed Conflict Location and Stöckl, Heidi, Lori Heise, and Charlotte Watts.
Event Data.” Journal of Peace Research 47 (5): 2012. “Moving beyond Single Issue Priority
651–60. Setting: Associations between Gender Inequal-
Ravallion, Martin. 2011. “On Multidimensional ity and Violence and Both HIV Infection and
Indices of Poverty.” Journal of Economic Poor Maternal Health in Malawi.” UNAIDS,
Inequality 9 (2): 235–48. Geneva.
POVERT Y FROM A NONMONE TARY PERSPEC TIVE 115
U N DP (United Nations Development Pro- Tylleskär. 2007. “Boys Are More Stunted than
gramme). 1990. Human Development Report Girls in Sub-Saharan Africa: A Meta-Analysis
1990. Oxford, UK: Oxford University Press. of 16 Demographic and Health Surveys.” BMC
UNESCO (United Nations Educational, Scientific Pedriatics 7 (17): 1–10.
and Cultural Organization). 2015. Education Wayack Pambè, Madeleine, Bilampoa Gnou-
for All 2000–2015: Achievements and Chal- mou, and Idrissa Kaboré. 2013. “Relationship
lenges. EFA Global Monitoring Report 2015. between Women’s Socioeconomic Status and
Paris. Empowerment in Burkina Faso: A Focus on
UNHCR (United Nations High Commissioner Participation in Decision-Making and Expe-
for Refugees). 2014. UNHCR Global Trends rience of Domestic Violence.” DHS Working
2013. Geneva. Paper 99, Demographic and Health Surveys
van de Walle, Dominique, and Annamaria Mila- and USAID.
zzo. 2015. “Are Female Headed Households WHO (World Health Organization). 2013. World
Poorer? New Evidence for Africa.” mimeo, Malaria Report 2013. Geneva: WHO.
World Bank, Washington, DC. ———. 2014a. Global Health Observatory 2015.
Verwimp, Philip, and Jan Van Bavel. 2005. “Child Geneva: WHO.
Survival and Fertility of Refugees in Rwanda.” ———. 2014b. World Malaria Report 2014.
European Journal of Population/Revue Euro- Geneva: WHO.
péenne de Démographie 21 (2–3): 271–90. ———. 2015. Global Health Observatory 2015.
———. 2013 “Schooling, Violent Conflict, and Geneva: WHO.
Gender in Burundi.” World Bank Economic World Bank. 2011. World Development Report
Review 28 (2): 384–411. 2012: Gender and Development. Washington,
Vyas, Seema. 2013. Estimating the Association DC: World Bank.
between Women’s Earnings and Partner Vio- ———. 2014. Voice and Agency Report: Empow-
lence: Evidence from the 2008–2009 Tanza- ering Women and Girls for Shared Prosperity.
nia National Panel Survey. Women’s Voice, Washington, DC: World Bank.
Agency, & Participation Research Series 2, World Development Indicators (database).
World Bank, Washington, DC. World Ban k, Wash ing ton, DC . ht t p: //
Waldron, Ingrid. 1983. “Sex Differences in d a t a .w o r l d b a n k . o r g /d a t a - c a t a l o g
Human Mortality: The Role of Genetic Fac- /world-development-indicators.
tors.” Social Science & Medicine 17 (6): Ziraba, Abdhalah K., Jean C. Fotso, and Rhoune
321–33. Ochako. 2009. “Overweight and Obesity in
Wamani, Henry, A nne N. Åstrøm, Stefan Urban Africa: A Problem of the Rich or the
Peterson, James K. Tumwine, and Thorkild Poor?” BMC Public Health 9 (1): 465.
Inequality in Africa
4
I
nequality in Africa is complex. Of the 10 entrenched. The expectation of having no
most unequal countries in the world, 7 chance of obtaining wealth or the feeling that
are in Africa. But African countries other the cards are stacked against one can yield
than these seven do not have higher inequal- precisely these outcomes, for example. This
ity than developing countries elsewhere in the lack of a level playing field—the structural or
world. For the region as a whole, however, ex ante component of inequality—is usually
inequality is high, because of the wide varia- perceived to be unfair. Cultures around the
tion in income across countries. As a comple- world value fairness—so much that in some
ment to the description of poverty, freedoms, cases people make seemingly irrational deci-
and capabilities in the previous two chapters, sions (that is, decisions that do not serve their
this chapter profiles inequality in Africa, self-interest) to punish others who behave
describing it in terms of consumption inequal- unfairly (World Bank 2005).
ity (including from the perspective of extreme Inequality in outcomes—the gap between
wealth) as well as inequality of opportunity. the poorest and the richest—depends not
An important distinction is between only on opportunities but also on effort and
inequality of outcomes (such as income, the degree to which individuals take risks.
consumption, and wealth) and inequality of Rewarding people for effort or risk taking
opportunity. In the case of the latter, in many can incentivize and motivate them. From
settings, circumstances over which a per- this perspective, not all aspects of inequal-
son has little control—mother’s education, ity are necessarily bad, although high levels
father’s occupation, birth in a rural area or of inequality can impose large socioeconomic
into a particular ethnic group—may largely costs on society.
dictate one’s future. Being born poor often Inequality can influence the ability of
means being the beneficiary of less invest- communities to coordinate and provide social
ment in human development, which deter- services and public goods.1 It can also induce
mines future living standards. conflict, although the empirical evidence that
Being born poor can also influence one’s substantial inequality leads to conflict or is
aspirations. Hoff (2012) describes how the source of most conflict is mixed (Cramer
aspirations can be affected if inequality is 2005; Lichbach 1989).2
Inequality influences how economic
This chapter was written with Camila Galindo-Pardo. growth translates into poverty reduction,
117
118 POVERTY IN A RISING AFRICA
a. Ghana b. Nigeria
Percent of respondents
Percent of respondents
25 25
20 20
15 15
10 10
5 5
0 0
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Need Need Need Need
more less more less
equality equality equality equality
c. Rwanda d. Zimbabwe
Percent of respondents
25 25
Percent of respondents
20 20
15 15
10 10
5 5
0 0
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Need Need Need Need
more less more less
equality equality equality equality
Source: World Values Surveys of Ghana (2012), Nigeria (2011), Rwanda (2012), and Zimbabwe (2012).
Note: 1 = “Incomes should be made more equal,” 10 = “We need larger income differences as incentives for individual effort.”
FIGURE 4.2 Survey respondents’ perceptions of the adequacy of their government’s efforts to narrow
the income gap differ across countries
1.0 100
0.9 90
0.8 80
Percent of respondents
0.7 70
0.6 60
Gini index
0.5 50
0.4 40
0.3 30
0.2 20
0.1 10
0 0
Z a rd e
M T an n d a
au i a
w i
Li oire
oz Le o
Ca m b n e
V e ue
rk h s
So na a n a
az rica
Ca Gui d
Zi er a
ba n
ne r
M l
m in
ru a
te To i
d’ g o
a a
as a
N e
Na en i
Ke ar
Ni nya
ri a
Si Le ana
Sw Af o
ts w
B al
nd
ga
Se ige
Bu G itiu
Bu ibi
ag n i
Ug eri
m ne
bw
m oo
h s
M ra th
n
M mb
c
ut Fa
Bo a l a
bo i q
ge
a o
ila
ad za
er so
b
Iv
r
M
Cô
Sources: Survey responses: Afrobarometer, Round 5 (2011–13). Gini index: World Bank Africa Poverty database.
Note: Blue bars show the share of the population that perceives that the government is not doing enough to narrow the income gap (right axis). Orange
diamonds are Gini indexes (left axis).
120 POVERTY IN A RISING AFRICA
The Gini index can be explained using the Lorenz FIGURE B4.1.1 The Lorenz curve illustrates the Gini
curve, which plots the cumulative share of total con- measure of inequality
sumption on the vertical axis against the cumulative
proportion of the population on the horizontal axis, Y
100%
starting with the poorest individual or household
(figure B4.1.1). If there is perfect equality, the bot-
consumption (or income)
centage difference between the consumption of a was expressed in terms of gross national income. It
randomly selected individual and the population’s is an intuitive measure of inequality that highlights
average consumption. One attractive feature of the large gaps in consumption often found between
the MLD is that it is sensitive to inequality among the rich and the poor.
the poor. Another is that, unlike the Gini index, the Each of these measures has different properties
MLD is decomposable: the contribution of inequal- and can produce different results. But cross-country
ity across different groups and the contribution of rankings of inequality in Africa are not strongly
the inequality within these groups can be calculated. affected by the measure of inequality used. Figure
Doing so helps unpack the nature of inequality, as B4.1.2 plots country inequality rankings according
done later in this chapter. to the MLD (panel a) and Palma (panel b) against
A third, more recent inequality measure is the the ranking based on the Gini index. In most cases,
Palma ratio, the ratio of the consumption share of countries line up on the diagonal, which means that
the richest 10 percent of the population to the share their rank position is unaffected by the measure
of the poorest 40 percent of the distribution (Palma used. These findings are similar to the finding by
2006, 2011). In its original formulation, the index Cobham and Sumner (2013).
40 40
Gini index rank
Gini index rank
30 30
20 20
10 10
0 0
0 10 20 30 40 50 0 10 20 30 40 50
Mean log derivation (MLD) rank Palma ratio rank
is almost twice as large as consumption on the Middle East in Alvaredo and Piketty
inequality. In rural areas, the wealth Gini 2015). One study that attempts to assess the
is 0.60, compared with 0.54 for income extent of the underestimation (the study of
and 0.39 for consumption. In urban areas, Egypt by Hlasny and Verme 2013) shows,
these estimates are 0.84 for wealth, 0.71 for perhaps surprisingly, that it is not large.
income, and 0.44 for consumption. A similar Another approach for gauging underes-
picture emerges in Tanzania and Uganda. timation at the top of the distribution is to
A second concern with consumption compare consumption from household sur-
inequality is that, in practice, consump- veys with private consumption in national
tion inequality measures will be biased accounts. Although there are conceptual
downward if the set of goods in the con- differences between these two measures
sumption measure does not include items of consumption, the growing gap between
consumed by the rich (luxury goods such as national accounts and survey consumption
vacations as well as irregularly purchased in countries such as China and India is often
consumer durable purchases such as cars). interpreted as an indicator that surveys miss
These goods are sometimes not included in out on a growing share of private expendi-
surveys or are excluded from the measure tures (Deaton 2005). This problem appears
of consumption if they are. 6 Consump- to be less important in Africa, where house-
tion surveys also struggle to include hard- hold surveys and national accounts have not
to-survey populations, including both the been observed to be diverging, as discussed
extreme poor (who may live in remote areas in chapter 1.
or informal settlements) and the extreme To study inequality in the distribution of
rich (who may refuse to participate in sur- consumption, the Gini index across coun-
veys). Applying imputation methods for tries is compared. The Gini index is a widely
mismeasured income data and accounting used measure of inequality (box 4.1). It
for expatriates not included in surveys in ranges between 0 (every individual enjoys
Côte d’Ivoire and Madagascar significantly the same level of consumption per capita,
increase measured inequality, according to perfect equality) and 1 (a single individual
Guénard and Mesplé-Somps (2010). The accounts for all consumption). A Gini index
net effect of missing these households is of 0.4 means that the expected difference in
ambiguous in terms of the bias in inequal- consumption between any two people cho-
ity, contingent on which household groups sen from the population at random will be
are excluded from the survey. However, if 80 percent (two times the Gini). This chap-
top income earners or the very poor are sys- ter focuses on Gini indexes as derived from
tematically excluded, inequality measures household surveys, rather than efforts to
will be understimated. impute a Gini from other sources (box 4.2)
Methods have been proposed to address
some of these problems (see Korinek, Misti-
aen, and Ravallion 2006). One approach is
Inequality Patterns and Trends
to compare top incomes in household surveys This section explores both national and
with tax records (Atkinson, Piketty, and Saez regional aspects of inequality and then
2011; Banerjee and Piketty 2005). Studies describes core household traits that explain
adopting this approach typically conclude inequality across groups in countries.
that surveys underestimate top incomes.
The evidence on South Africa is ambiguous,
Inequality across African Countries
because most surveys provide estimates of
top income shares that are close to the tax Gini indexes from the most recent household
data (Morival 2011). Many developing coun- surveys in Africa range from 0.31 (Niger
tries lack administrative tax data with which and São Tomé and Príncipe) to 0.63 (South
to assess the level of underreporting in house- Africa). Comparing these estimates with
hold surveys (see, for example, the discussion estimates from other countries (based on the
INEQUALITY IN AFRICA 123
Issues of comparability and data availability hamper SWIID imputations show substantial variability
studies of inequality in Africa. For the Gini results in the region, as Solt (forthcoming a) notes (figure
in this chapter, only nine countries have more than B4.2.1). Most of the estimates computed directly
three data points, and seven countries have just a from the surveys are within the SWIID confidence
single data point. interval, but that interval is wide.
Can this dearth of data be circumvented by esti- The two sources are highly correlated (with a
mating the Gini? The Standardized World Income correlation of 0.83 between the survey estimate and
Inequality Database (SWIID) takes this approach, the average SWIID estimate from 100 imputations).
seeking to maximize the comparability and coverage The correlation is higher (0.91) if the comparison is
of Gini estimates worldwide (Solt forthcoming a). This limited to surveys deemed comparable within the
effort works best in countries with better and more country. The correlation is low (only 0.15) among
data, but it is still subject to critique (see Jenkins 2014 the nine countries in Central Africa.
and the response to his critique in Solt forthcoming b). The direction of the changes in the Gini in the
Using a missing-data algorithm and drawing on SWIID does not match well with the trends revealed
information from proximate years within a country by the surveys (as in figure 4.4). In only 11 of 20
and various data collection efforts (such as the World countries with a trend in both sources does the
Bank’s PovcalNet, the UNU-WIDER database, and direction of change match. There is a high degree of
country statistical reports), SWIID produces Gini uncertainty in the SWIID estimates. In only 1 of the
estimates for 45 countries in Africa. For 1991–2012, 20 countries studied is the change in the Gini statis-
SWIID has 16 or more annual estimates of the Gini tically significant. Until better and more surveys are
for more than half these countries. Because of the conducted in the region, imputing inequality mea-
lack of survey data in developing countries, the sures is fraught with serious concerns.
FIGURE B4.2.1 Standardized World Income Inequality Database (SWIID) estimates of the
Gini index show great variability
a. Côte d’Ivoire b. Mozambique
0.6 0.6
0.5 0.5
Gini index
Gini index
0.4 0.4
0.3 0.3
1990 1995 2000 2005 2010 1990 1995 2000 2005 2010
c. Uganda d. Zambia
0.6 0.6
0.5 0.5
Gini index
Gini index
0.4 0.4
0.3 0.3
1990 1995 2000 2005 2010 1990 1995 2000 2005 2010
Source: World Bank Africa Poverty database and Solt forthcoming a.
Note: Orange lines show the 95 percent confidence intervals on the SWIID Gini imputations. Blue dots are the survey-based Gini estimates
from the World Bank Africa Poverty database.
124 POVERTY IN A RISING AFRICA
FIGURE 4.3 The world’s most unequal countries are in Africa PovcalNet database) reveals that 7 of the 10
most unequal countries in the world are in
Africa (figure 4.3). All but two of the seven
countries (South Africa and Zambia) have
populations of less than 5 million.
The levels of inequality in Africa appear
even more remarkable if one considers that
many countries outside Africa—particularly
advanced economies and countries in Latin
São Tomé and Príncipe
America—use income rather than consump-
Niger
tion per capita to measure inequality. Relative
to consumption data, income data generally
produce higher levels of inequality.
Mali The heterogeneity in inequality across
Ethiopia
Burundi Africa is substantial and shows a geographi-
cal pattern (map 4.1). Inequality is higher
Guinea
Sierra Leone in Southern Africa (Botswana, Lesotho,
Namibia, South Africa, Swaziland, and Zam-
Sudan bia), where Gini indexes are above 0.5, as
Guinea-Bissau
well as in Central African Republic and the
Comoros. West African countries exhibit
Mauritius
lower levels of inequality, and countries in East
Liberia Africa are mixed. These findings are robust to
Mauritania
Tanzania other measures of inequality (box 4.1).
Some researchers have argued that these
Burkina Faso patterns in inequality have historical roots.
Congo, Rep.
Senegal In particular, the high levels of inequality in
Madagascar
Southern Africa are legacies of the land dis-
possession and racially discriminatory poli-
Gabon
Uganda
cies of the colonial period. There are notable
Angola
Ghana
Seychelles
differences in the history of communal land
Cameroon
Nigeria tenure systems in West and Central Africa
Côte de Ivoire
Chad compared with white settler economies (char-
Benin
Congo, Dem. Rep. acterized by privately owned small family
Mozambique
plots, large estates, and plantations) in East
Togo
and Southern Africa (Cornia 2014).
Malawi
There are few other discernable patterns
Cabo Verde
Gambia, The in terms of country traits and inequality.
Kenya Inequality levels do not differ statistically
Rwanda
Swaziland
between coastal and landlocked, fragile and
nonfragile, or resource-rich and resource-
poor countries, controlling for the four sub-
Lesotho
Zambia
Comoros
regions. Bhorat, Naidoo, and Pillay (2015)
Central African Republic
Botswana
also conclude that the average level of inequal-
Namibia
South Africa ity is not different between resource-rich and
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 other economies, but they note that a number
Gini index of resource-rich economies have high levels
of inequality. If the eight most unequal coun-
Source: PovcalNet for countries outside Africa; World Bank Africa Poverty database.
Note: Orange bars are African countries (based on consumption); light blue bars are other countries tries in the region (South Africa, Zambia, and
using consumption surveys; dark blue bars are other countries using income surveys. six small economies) are excluded and one
INEQUALITY IN AFRICA 125
Cabo Mauritania
Verde
Mali Niger
Senegal Sudan Eritrea
The Gambia Chad
Guinea-Bissau Burkina Faso
Guinea Benin
Nigeria
Côte Ethiopia
Sierra Leone d’Ivoire Ghana Central African South Sudan
Republic
Liberia Cameroon Somalia
Togo
Equatorial Guinea
Uganda
São Tomé and Princípe Rep. of Kenya
Gabon Congo
Rwanda
Dem. Rep. of Burundi
Congo
Tanzania
Seychelles
Comoros
Gini index Angola
Malawi
0.60–0.63 Zambia
0.50–0.59
0.46–0.49 Mozambique
Zimbabwe Madagascar Mauritius
0.41–0.45
Namibia
0.36–0.40 Botswana
0.31–0.35
No data Swaziland
South Lesotho
Africa
IBRD 41869
SEPTEMBER 2015
controls for country-level income, Africa has income status, or initial level of inequality in
inequality levels comparable to developing the first survey are evident.
countries in other parts of the world (Bho- The picture is the same if one looks at
rat, Naidoo, and Pillay 2015 draw the same the longest available time period for which
conclusion). comparable data are available. Cornia
Are African countries becoming more (2014) describes this pattern as “inequality
unequal? Analysis of 23 countries for which trend bifurcation.”8 Within-country trends
there are two comparable surveys to measure in inequality in Africa differ from trends in
inequality reveals that about half the coun- both Asia, where inequality is on the rise,
tries experienced a decline in inequality while and Latin America, where inequality has
the other half saw an increase (figure 4.4).7 been declining since the early 2000s (see
No clear patterns based on resource status, Ferreira and others 2013 for Latin America;
126 POVERTY IN A RISING AFRICA
FIGURE 4.4 Inequality rose in about half of the countries and fell in the other half
3
Annualized percentage change in Gini index
–1
–2
–3
–4
ag re
Rw nd
au a
m s
Za on
Cô en a
M d’Iv l
Gh r
Ni a
ria
ad
ra aso
Ta one
Ug ia
ts a
ng N ana
M m. ia
So bi .
h e
az a
Et o
M ia
i
am p
aw
ca
Ca ritiu
Bo and
Sw fric
M nd
S bi
an
ut qu
g
an
De ib
te eg
op
oz Re
ad oi
ge
Ch
o
To
ila
as
m
aF
al
w
o, m
a
Le
er
nz
hi
a
in
rk
er
Bu
Si
Co
Asian Development Bank 2014 and Rama countries in Southern Africa (Botswana,
and others 2015 for Asia). Namibia, and South Africa), which differ in
Should one expect a more systematic many ways (in addition to GDP per capita)
increase in inequality given Africa’s double from the rest of Africa. A more appropriate
decade of growth? One of the long-standing test of the Kuznets hypothesis is to compare
debates in economics is about the trends changes in inequality with changes in GDP
in inequality during periods of economic per capita using multiple observations per
growth. In the 1950s, Simon Kuznets for- country (panel b of figure 4.5). If the Kuznets
mulated the hypothesis that inequality first hypothesis holds, the data should trace out
increases and then declines as GDP per capita an inverted U-pattern or at least—given that
rises (Kuznets 1955). Because most countries most of the countries in the sample are poor
in Africa still have low levels of GDP, the and hence likely to be shifting along the ris-
Kuznets hypothesis suggests that inequality ing portion of the U—an upward slope to
should increase with rising GDP per capita. show inequality rising as GDP increases.
Empirical studies have not produced This is not the case: inequality is not moving
robust support for the Kuznets hypoth- in a clear direction and does not appear to
esis (Bruno, Ravallion, and Squire 1998; be systematically related to changes in GDP
Deininger and Squire 1996; Milanovic 2011). per capita. Other researchers have reached
The African data also fail to provide strong similar conclusions based on examination of
evidence for a Kuznets-type trajectory. recent data (Bhorat, Naidoo, and Pillay 2015)
Panel a of figure 4.5 compares the level and growth spells in the 1990s (Fields 2000).
of inequality (measured by the Gini) with All else constant, a reduction in inequal-
GDP per capita. Although there is a sig- ity is associated with a decline in poverty
nificant positive relationship between the (Bourguignon 2004; Klasen 2004). Many
level of GDP and inequality, it is driven countries in figure 4.6 are in quadrant 4,
almost entirely by the upper-middle-income where both inequality and poverty declined.
INEQUALITY IN AFRICA 127
FIGURE 4.5 There is no systematic relationship between growth and inequality in Africa
a. Correlation between Gini index and GDP per capita b. Changes in Gini index and GDP per capita
0.7 0.55 Zambia
South Africa Rwanda
Namibia Botswana 0.50
0.6
Zambia
Gini index
Gini index
0.45
0.5 Rwanda Swaziland
0.40 Nigeria
Nigeria Mauritania
0.4 Senegal
Mauritania 0.35
Mauritius Ethiopia
Ethiopia
0.3 0.30
0 5,000 10,000 15,000 20,000 0 1,000 2,000 3,000 4,000 5,000
GDP per capita GDP per capita
Sources: World Bank Africa Poverty database (subset of countries with comparable surveys); World Development Indicators database.
Note: Panel a is based on the most recent survey. Panel b excludes the five highest-income countries in panel a.
However, in a number of countries poverty But Africa as a whole has the highest level of
fell despite increasing inequality (quadrant 1 inequality of any region in the world.12 The
in figure 4.6). In these countries, the growth African Gini index rose by almost 9 percent
in mean consumption was large enough to between 1993 and 2008. By contrast, the
offset the rise in inequality. average country Gini fell by almost 5 percent,
and no change is observed if countries are
weighted by their population.
Inequality in Africa as a Whole
Combining survey data across countries
enables the study of the Africa-wide dis-
tribution of consumption.9 For this exer- FIGURE 4.6 Declining inequality is often associated with declining
cise, surveys are grouped into benchmark poverty
years (1993, 1998, 2003, and 2008).10 The
data cover 81 percent of regional GDP and
Annualized percentage change in Gini index
Quadrant 1
72 percent of the population, indicating 2 Ethiopia 04-10
Malawi
Zambia 98-04
that richer countries are more likely to be Chad
Rwanda 00-05 Togo Nigeria
Madagascar 05-10
Côte d’Ivoire
included.11 Given this coverage, the results Uganda 05-09 Ghana 98-05 Mozambique 96-02
Ghana 91-98 Zambia 04-06 Cameroon
0
probably represent a lower bound on Afri- Ethiopia 99-04
South Africa Rwanda 05-10 Senegal
Mozambique 02-09 Swaziland
Mauritania Namibia
Mauritius
Tanzania
The African Gini index is 0.52– 0.56 –2 Uganda 02-05
across the benchmark years, much higher Burkina Faso
Sierra Leone
The level of inequality in Africa is largely FIGURE 4.7 The richest households in Africa live
driven by within-country inequality, which mostly in the richer countries
explains considerably more than half of the
inequality measured by the mean log deriva- 100
tion (MLD). However, the increase in African 90
inequality was driven by a widening between 80
countries, as opposed to within-country
70
changes in inequality. Over time, a greater
60
share of African inequality is explained by
Percent
%
%
t5
t5
es
s
re
ch
inequality? To some extent, it does. Figure
o
Ri
Po
4.7 divides the African distribution of con- Upper-middle and high-income countries
sumption in 2008 into 20 ventiles, from Low-middle-income countries
Low-income countries
poor to rich, each representing 5 percent of
the African population. For each ventile, the Source: Jirasavetakul and Lakner 2015.
figure shows the share of the population in
low-, lower-middle-, and upper-middle-/
high-income countries. In 2008, 54 percent
Between-Group Inequality
of the population in the top 5 percent of the
African distribution were living in upper- This section explores the extent to which con-
middle-/high-income countries, 36 percent sumption levels differ across groups in an econ-
in lower-middle-income countries, and 10 omy based on some socioeconomic or other
percent in low-income countries. The share household trait. Between-group (or horizontal)
of the African population in upper-middle-/ inequality is measured by decomposing over-
high-income countries rises as one moves up all inequality into two parts: inequality attrib-
the distribution, while the share of the popu- uted to between-group (horizontal) differences
lation in lower-income countries declines. and inequality within groups. Horizontal
However, there is much overlap across these inequalities can come at a high cost to society.
country classifications, meaning there are Between-group inequalities can perpetuate
very rich households in poor countries and intergenerational persistence in poverty, and
vice versa. social exclusion and can limit socioeconomic
INEQUALITY IN AFRICA 129
mobility. They have been linked to violent con- To explore between-group inequality in
flict and social unrest and are therefore partic- Africa, seven groups are defined based on the
ularly detrimental for economic development consensus in the literature and the availabil-
and poverty reduction (Cramer 2005; Langer ity of information in the household surveys to
and Stewart 2015). In a similar vein, ethnic define groups.13 Of the seven groups exam-
fractionalization has been associated with ined, geographical location, education, and
poor outcomes in the provision of local public demographics are the most important drivers
goods (Miguel and Gugerty 2005) and lower of inequality (figure 4.8).14
levels of overall economic growth in Africa Spatial inequalities are important for both
(Easterly and Levine 1997). the urban-rural group and the regional group
FIGURE 4.8 Location, education, and demographics are the most important drivers of inequality
40 40 40
30 30 30
20 20 20
10 10 10
0 0 0
Low inequality High inequality Low inequality High inequality Low inequality High inequality
40 40 40
30 30 30
20 20 20
10 10 10
0 0 0
Low inequality High inequality Low inequality High inequality Low inequality High inequality
g. Age group
Percent of inequality explained
40
30
20
10
0
Low inequality High inequality
classifications.15 In Senegal one-third of have poverty rates that are more than 25 per-
total inequality is attributed to gaps between cent higher than the national average.
households in urban and rural areas. On Educational attainment of the household
the lower end of the spectrum are some of head is an even more important driver of
the small island states (the Comoros, São gaps in consumption across households. In
Tomé and Príncipe), where urban-rural gaps three countries (Rwanda, South Africa, and
are virtually nonexistent. A similar picture Zambia), educational attainment explains
holds for inequality between regions (first- about 40 percent of overall inequality. Higher
tier administration units). The two between- inequality is associated with greater inequal-
group components (urban-rural and regions) ity between education categories, an associa-
are correlated (0.73); countries with large tion that is not observed among most of the
urban-rural gaps in living standards also tend other socioeconomic groupings. Education
to have significant gaps across regions. Spa- tends to explain a greater share of inequality
tial inequalities may be even higher than cap- than the broad economic activity category of
tured by household consumption, because of the household head, an important driver of
the spatial aspects of public service provision inequality in some countries.
(the fact that the value of public services, such The demographic composition of the
as health services and schools, may be higher household also explains a large share of
in urban areas). inequality, up to 30 percent of overall inequal-
A nother way to view the extent of ity in Senegal and 32 percent in Botswana.
regional inequality is to compare mean con- This finding is consistent with the fact that
sumption per capita across areas. The gap larger households in Africa, especially house-
(as measured by the ratio between the rich- holds with many children, show significantly
est and the poorest regions) often shows that lower levels of consumption and higher levels
the richest regions have twice the mean con- of poverty than smaller households.
sumption of the poorest. The gap for first- Some demographic characteristics—
tier administration units is 2.1 in Ethiopia for example, the gender of the household
(regions), 3.4 in the Democratic Republic head—do not explain a substantial share of
of Congo (provinces), and more than 4.0 in total inequality. This finding is not surpris-
Nigeria (states).16 Inequality associated with ing, given that in many African countries,
geographical income segregation may be consumption per capita levels of male- and
more politically destabilizing than inequal- female-headed households do not differ
ity in which the poor and rich are equally widely. A shortcoming of this method of
dispersed geographically (Milanovic 2011), decomposing inequality is that the decom-
especially if geographical inequalities coin- position reveals nothing about the direction
cide with ethnicity or religion, as in northern of bias (that is, whether the disadvantage lies
and southern Nigeria. with female- or male-headed households).
In most household surveys, the samples Moreover, because consumption is measured
are too small to estimate inequality for geo- at the household level, the decomposition
graphical areas smaller than regions. Such does not provide any information about how
estimates can be made by combining house- consumption is distributed between men and
hold surveys with census data to yield poverty women within households (box 4.3).
maps (also known as small area poverty esti- For many countries, horizontal inequali-
mates). The poverty map of Zambia shows ties can be measured for more than one
that of the more than 1,400 constituencies in point in time. The main drivers of horizon-
the country, about one in seven has a poverty tal inequalities (geography, education, and
rate of less than half the country mean (de demographics) did not change during the
la Fuente, Murr, and Rascón 2015). At the period for which survey data are available
other extreme, 20 percent of constituencies (from the early 1990s to the present).
INEQUALITY IN AFRICA 131
BOX 4.3 Are resources within households shared equally? Evidence from Senegal
Little is known about interpersonal inequality in liv- consumption data are then collected at the cell level.
ing standards within households, including between Finally, expenditures that are shared by several cells
men and women, because consumption data are col- are collected and attributed equally to all household
lected at the household level and standard measures members. A measure of consumption per capita is
of poverty and inequality are calculated assuming then constructed at the cell level.
that resources are shared equally within the house- The results clearly show that not everyone in
hold (even if there is some normalization for size and the household gets the same resources. The ratio
demographic composition). between the consumption of the richest and poorest
The idea that individuals within a household do cells within a household can be as high as 23 (and
not always have the same living standards and that is still above 4 after trimming off the 5 percent of
income is not shared equally is not new (see Strauss, most unequal households). In general, food expen-
Beegle, and Mwabu 2000 and the ample evidence in ditures are equitably distributed, a critical insight
World Bank 2011). Gender and age are arguably the that underscores basic solidarity. In contrast, non-
most prominent individual attributes along which food expenditures are not divided equally. Over-
differentiation takes place within the household. all inequality is higher for cell-level consumption
The household structure in Senegal (as well as (Gini = 0.567) than for a household-level measure
in other West African countries) is unique in its that assumes equal consumption across household
complexity and offers opportunities to explore the members (Gini = 0.548).
extent of intrahousehold inequality. Households These unique consumption data also reveal a siz-
are structured like compounds. Within each house- able gender gap. Cells headed by men have signifi-
hold are “cells” made up of a head and unaccom- cantly higher consumption.
panied dependent members, while married broth- The poverty status of the household can hide
ers and each wife of the head and her children poverty within the household. About 1 nonpoor
form separate cells. Surveying and paying careful household in 10 has a poor cell within it (De Vreyer
attention to the compound structure and con- and Lambert 2014). There are also nonpoor cells in
sumption patterns among members reveals within- poor households. Targeting poor households would
household consumption patterns (De Vreyer and miss 6–14 percent of poor children (depending on
others 2008). Food expenditures are compiled based the poverty line), namely, children who reside in
on a detailed account of who shares which meal and poor cells within nonpoor households.
how much money is used to prepare it. Individual
Ta ar
ia
ria
a
ea
Rw a
da
os
aw
ge
an
d
an
c
or
in
ge
an
an
as
Ni
al
Gh
nz
Gu
m
Ug
Ni
ag
Co
ad
Intergenerational Persistence in
M
it did 50 years ago? 24 Is the occupation of countries (Ferreira and others 2013; Hertz
a farmer’s son less affected by his father’s and others 2007). These changes may partly
occupation than it was a generation ago? reflect the fact that since the 1990s, many
Using data from several recent household countries have eliminated school fees at
surveys in Africa and drawing on a set of the primary level (Bhalotra, Harttgen, and
surveys with information on adult children Klasen 2015). In terms of level of mobility in
and their fathers, the extent of intergenera- general, Africa has greater intergenerational
tional mobility in education and occupation educational mobility than Latin America.
is examined, as well as whether the extent However, mobility is lower than developed
of this mobility is changing among younger countries in Europe, the United States, and
generations.25 the former Eastern Bloc.
To measure educational mobility from Like education, one’s occupation may
the perspective of intergenerational persis- be determined largely by the occupations
tence, education is regressed on the edu- of one’s parents. The limited literature on
cational attainment of one’s parents. The this issue in Africa focuses on intergenera-
coefficient from this simple regression, ß, tional occupational persistence from farm
measures education persistence (see Black to nonfarm occupations. Here this analysis
and Devereux 2011 for a recent overview of is extended to look at three occupational
approaches to this measurement). Another classifications among men 20–65 (agricul-
measure of mobility is the correlation coef- ture, services, and other occupations) and
ficient between the outcomes of parents and their fathers. The analysis is restricted to
their children (ρ), which is the intergenera- the occupation of fathers because fewer sur-
tional gradient (ß), multiplied by the ratio of veys have information on the occupation of
the standard deviation across the two gen- mothers.
erations. 26 Three factors are explored—the Intergenerational occupational persis-
intergenerational gradient, the correlation tence in farming has been falling rapidly in
coefficient, and the ratio of standard devia- some countries (table 4.2). In the Comoros,
tions—for different cohorts to study inter- the share of farmers’ sons working in other
generational persistence in schooling across sectors is more than twice as large for the
generations (figure 4.10). youngest cohort as it is for older cohorts.
The correlation coefficient on intergener- Guinea is the most rigid economy in terms
ational mobility (the blue line in figure 4.10) of occupational shifting. There is substan-
slightly increased in most countries. Con- tial intergenerational mobility in work
versely, the intergenerational gradient, ß, is among people with fathers in services and
falling in most countries (the orange line in other sectors; generally less than half of
figure 4.10). An additional year of school- the youngest cohort are performing the
ing of one’s parent has a lower association same services or other sector work as their
with one’s own schooling than it used to. fathers. This change in intergenerational
This reflects, however, that the ratio of the occupational persistence is consistent with
standard deviations (the red line in figure the overall shifts in occupational structure
4.10) is rising, which in turn is related to in each country, specifically the falling
the low levels of schooling among parents in employment shares of agriculture (World
the oldest generation. For example, people Bank 2014a).
born in 1949 in Rwanda have on average To separate out economy-wide shifts,
1.5 years of schooling, while their parents the share of job mobility associated with
have only 0.1 years. The Africa intergen- expansion in nonagricultural sectors is net-
erational mobility trends are broadly com- ted out (following the approach of Bossuroy
parable to estimates in other developing and Cogneau 2013). Net mobility shows
134 POVERTY IN A RISING AFRICA
FIGURE 4.10 Intergenerational persistence in schooling is weaker among younger Africans than older
Africans
Persistence, SD ratio
Persistence, SD ratio
1.0 1.0 1.0
0 0 0
1943 1953 1963 1973 1983 1951 1961 1971 1981 1991 1952 1962 1972 1982 1992
Persistence, SD ratio
Persistence, SD ratio
1.0 1.0 1.0
0 0 0
1942 1952 1962 1972 1982 1945 1955 1965 1975 1985 1950 1960 1970 1980 1990
Persistence, SD ratio
Persistence, SD ratio
0 0 0
1950 1960 1970 1980 1990 1939 1949 1959 1969 1979 1949 1959 1969 1979 1989
Source: Azomahou and Yitbarek 2015. Data for other developing countries from Hertz and others (2007).
that shifts in the structure of occupations mobility (table 4.3). The Comoros, Rwanda,
in the economy (sometimes called struc- and Uganda exhibit the highest rates of
tural change) are not the only factor driving intergenerational mobility that is not attrib-
changes in intergenerational occupational utable to structural change.
INEQUALITY IN AFRICA 135
TABLE 4.2 Likelihood of remaining in one’s father’s sector in selected African countries
Sons of service sector Sons of other sector
Sons of farmers stay in sector employees stay in sector employees stay in sector
1 2 3 4 5 1 2 3 4 5 1 2 3 4 5
Country Oldest Youngest Oldest Youngest Oldest Youngest
Comoros 80 55 55 45 48 34 53 45 40 55 7 41 37 42 17
Ghana 76 65 64 59 71 47 50 51 60 52 21 22 32 25 32
Guinea 79 69 73 76 80 26 40 34 36 41 24 28 43 40 32
Rwanda 86 83 84 77 78 32 18 22 28 31 0 34 12 22 8
Uganda 78 72 66 60 72 33 39 40 37 27 32 28 34 43 33
Source: Azomahou and Yitbarek 2015.
Note: Table shows the percent of each cohort with the same occupation as their father. 1–5 are 10-year birth cohorts. The table should be read as follows: Among the youngest
cohort (cohort 5) in the Comoros, the son of a farmer has a 48 percent likelihood of also being a farmer. Members of the oldest cohort of farmers’ sons have a much higher chance of
being farmers (80 percent).
TABLE 4.3 Gross and net occupational intergenerational mobility out of farming in selected African
countries
Gross mobility Net mobility
1 2 3 4 5 1 2 3 4 5
Country Oldest Youngest Oldest Youngest
Comoros 29 47 49 56 57 15 24 24 29 28
Ghana 31 42 43 45 36 12 13 7 7 7
Guinea 30 38 34 35 30 16 19 11 8 8
Rwanda 17 22 21 29 31 12 14 14 17 13
Uganda 29 35 40 45 40 14 17 21 21 12
Source: Azomahou and Yitbarek 2015.
Note: Table shows the percent of each cohort with the same occupation as their father. 1–5 are 10-year birth cohorts. The table should be read as follows:
Among the youngest cohort (cohort 5) in the Comoros, for example, 57 percent of sons do not have the same occupations as their fathers. Net mobility is
computed as gross mobility minus the share of mobility associated with structural change in employment.
Extreme Wealth and Billionnaires inroads, but they still generally cover little of
Africa compared with other regions.
Household surveys are not suited for cap-
South Africa was the first African coun-
turing very high levels of income or wealth.
try to be represented on Forbes’ list, with
Missing information on extreme wealth leads
two billionaires in the late 1990s, followed
to underestimation of the extent of economic
inequality in a broader sense. Wealthy house- by Nigeria in 2008. By 2014 the region had
holds are often not surveyed and household 19 billionaires: 8 in South Africa, 7 in Nige-
surveys generally measure current consump- ria, and 1 each in Angola, Kenya, Tanzania,
tion or income (a flow measure) rather than and Uganda.27 Countries such as India expe-
the stock of household assets. Surveys are rienced a much sharper rise during a simi-
also likely to fail to capture rare income lar period. The number of billionaires there
events or income (and the wealth from it) that rose from 2 in the mid-1990s to 46 in 2012,
is obtained illegally (Africa Progress Panel according to Gandhi and Walton (2012).
2013). Data on holders of extreme wealth Although there are fewer billionaires in
are difficult to collect. The Forbes World’s Africa, their average aggregate net wealth in
Billionaires list, the World Top Incomes 2012 was higher ($5.2 billion per billionaire)
Database (currently covering South Africa than in India ($3.8 billion). Aggregate bil-
and ongoing in 15 other African countries), lionaire wealth as a percent of GDP increased
and the Global Wealth Databook have made steadily in Nigeria and South Africa from,
136 POVERTY IN A RISING AFRICA
FIGURE 4.11 Billionaire wealth in Africa is growing richest African (Aliko Dangote), whose for-
tune grew by a factor of 10 between 2010
and 2014.
(purchasing power parity constant 2011 dollars)
15
by 0.12 percent). Africa’s trend (not shown)
individuals 2004–14
richest people reside.) Oxfam International in their sample, the fraction of politically
(2015) estimates that globally 80 individuals connected billionaires in 1987, 1992, 1996,
possess as much wealth as half the world’s and 2002 ranges from 4 percent to 13 per-
population (the regional and global results cent. They conclude that politically con-
are not strictly comparable). 28 Few detailed nected wealth accumulation has a negative
studies explore the level of extreme wealth effect on economic growth worldwide. In
of nationals at the country level. One excep- resource-rich countries in Africa, there is
tion is the New World Wealth (2014) study concern that the elites gain wealth from
of Kenya, which estimates that about 8,300 resources through political connections (see
people own 62 percent of that country’s the examples and broad discussion in Burgis
wealth. 2015).
Does the source of this wealth matter?
Particularly in sectors where rent-seeking
behavior is more likely, the role of political
Concluding Remarks
connections in the wealth-generating pro- The latest evidence on inequality in Africa
cess could have implications for development paints a complicated picture. The most
and growth. Gandhi and Walton (2012) find unequal countries in the world are in Africa,
that in India in 2012, 60 percent of total net mostly in the southern part of the conti-
wealth was derived from “rent-thick” sectors, nent, but excluding the seven countries with
such as real estate, infrastructure, construc- extremely high inequality, inequality is not
tion, mining, telecommunications, cement, higher or lower than in other countries at
and media, where the influence of political similar income levels. In countries with com-
connections and the potential for rent extrac- parable surveys over time, inequality is fall-
tion are important (Rama and others 2015). ing in half and rising in half, without a clear
In Africa the share of extreme wealth derived association with factors such as resource-
from extractives has been declining. During richness, income level, or state fragility.
2011–14, about 20 percent of African billion- A clearer pattern emerges for horizontal
aires derived their wealth mainly or partially inequalities within countries, which continue
from telecommunications, and the share of to be dominated by unequal education levels
extreme wealth derived from services and the and high urban-rural and regional income
broad category of investment jumped from 1 disparities.
percent to 13 percent. From a regional perspective, inequal-
Forbes classifies the majority of net ity among Africans is rising and is high
wealth in Africa in 2014 as self-made as compared with other regions. This pattern
opposed to inherited. It estimates that self- reflects the range in national income levels
made aggregate net wealth in the region across countries and the fact that most of the
represented 74 percent of total net wealth poor in Africa reside in the poorest countries.
and that 81 percent of the billionaires in The income gap between African countries is
Africa reported being self-made. This clas- growing.
sification of self-made does not necessarily Another aspect of inequality—extreme
imply returns to successful entrepreneurship wealth—is missed altogether by household
and innovation (as opposed to accumulating surveys. Africa has seen a rise in billionaire
extreme wealth through political influence wealth, at least in countries for which data
or corrupt business practices). Bagchi and are available.
Svejnar (2015) assess wealth accumulation A portion of inequality in Africa can be
through political connections by looking attributed to inequality of opportunity, cir-
at evidence in news sources that suggests cumstances at birth that are major deter-
whether billionaires would have become minants of one’s poverty status as an adult.
billionaires in the absence of political con- Fortunately, at least in some countries,
nections. Among total billionaire wealth there has been a rise in intergenerational
138 POVERTY IN A RISING AFRICA
educational mobility, holding out hope that are interviewed. If prices differ spatially and
inequality of opportunity will decline. Never- temporally, deflated aggregates may produce
theless, intergenerational occupational per- different inequality measures and trends. For
sistence, at least as captured by three broad most of the surveys analyzed here, a deflated
(real) consumption measure is available. The
occupation categories, remains high in many
general findings on the levels and trends in
countries.
inequality are not substantially different if
Gini indexes are estimated using deflated
(real) consumption. One exception is the
Notes findings on between- and within-inequality
1. According to Olson (1965), if a public good by region or urban location, which tends to
is of interest to the rich, inequality could decline using spatially deflated aggregates.
facilitate collective action and allow the Székely and Hilgert (2007) analyze some of
poor to free ride. In fact, the evidence shows, these issues in Latin American countries.
more often the opposite occurs. Wealthy 5. Excluding these surveys has implications for
households, which can afford private pro- how the results compare with the results of
viders, opt out of financing public services other studies. For example, excluding the
such as schools and health care facilities and first of the three most recent national house-
redirect resources to efforts that do not serve hold surveys in Malawi (on the grounds
poor families. Mansuri and Rao (2013) pres- of incomparability in survey design), the
ent a range of evidence indicating that com- inequality trend in Malawi is not decreasing,
munities with high inequality have worse as Bhorat, Naidoo, and Pillay find (2015).
local development processes and outcomes. 6. Expenditure on consumer durables is not
They find that highly unequal incomes always included in the consumption mea-
amplify market failures. sure, because it represents highly irregular
2. Some studies find evidence that hig h purchases (Deaton and Zaidi 2002). The
inequality within ethnic groups rather than recommended practice is to include dura-
in the country as a whole is a driver of civil ble goods “use values” in the consumption
conflict (Huber and Mayoral 2014). Oth- measure.
ers find that it is inequality between ethnic 7. Bhorat, Naidoo, and Pillay (2015) use a dif-
groups that matters (Stewart 2008). Parallel ferent inequality measure but show simi-
with these efforts to explain civil conflict is lar results: of 34 countries in Sub-Saharan
the literature that explores how inequality, Africa, inequality rose in 18 and fell in 16.
especially ascriptive and horizontal inequali- Cornia (2014) and Fosu (2014) draw simi-
ties, explains crime rates (see, for example, lar conclusions. All three reports draw on
Blau and Blau 1982). the World Bank PovcalNet database. From
3. Similar contradictions in perceptions can be a population perspective, the results lean
found in views on inequality in the United toward increasing inequality; 57 percent of
States (Fitz 2015). the population in these countries are resid-
4. Purchasing power parity (PPP) adjustments ing in a country with increasing inequality.
to convert local currency units into U.S. dol- 8. Measuring polarization is another approach
lars do not affect national inequality mea- to looking at the consumption distribu-
sures. National temporal price adjustments tion, a concept related to but distinct from
(to bring a survey from year 1 to year 2 inequality. Polarization measures separa-
prices) also do not typically affect national tion (distance) across clustered groups in a
inequality measures. In contrast, within-sur- society. Clementi and others (forthcoming)
vey spatial price adjustments change inequal- show that Nigeria experienced both rising
ity measures. Both the World Bank Africa inequality and rising polarization between
Poverty Database (used here) and PovcalNet 2003/04 and 2012/13, which contributed
compute Gini indexes based on nominal to the eroding of the middle class. Keefer
consumption measures. They do not adjust and Knack (2002) argue that, in practice,
for the fact that the households interviewed polarization measures are strongly positively
pay different prices depending on where in correlated with inequality measures across
a country they live or the time of year they countries.
INEQUALITY IN AFRICA 139
9. For details on the calculations on African are evaluated. In addition, the between-
inequality, see Jirasavetakul and Lakner group component mechanically increases
(2015). This idea has also been pursued with the number of categories used. Elbers
globally, including in global inequality stud- and others propose an alternative decom-
ies by Anand and Segal (2015), Atkinson and position that compares between-group dif-
Brandolini (2010), and Milanovic (2005). ferences with the maximum inequality that
The analysis here draws heavily on Lakner would be obtainable if the number and size
and Milanovic (2015), who analyze the of groups were fixed at their actual levels,
global income distribution in 1988–2008. while the ranking of the groups is preserved.
10. Because of the limited availability of house- For instance, urban-rural inequality would
hold surveys, the analysis cannot start before be evaluated against a benchmark in which
1993, and there are not enough surveys for a all individuals living in rural areas appear at
benchmark year after 2008. the lower end of the distribution and all indi-
11. General coverage of Africa is good, but the viduals living in urban areas appear at the
coverage of fragile countries is low: on aver- upper end of the distribution, with the urban
age, the surveys cover only 28 percent of and rural population shares fixed at their
the population in fragile countries between actual levels. The decomposition thereby
1993 and 2003. The rate improves markedly takes into consideration the existing config-
in 2008 with the inclusion of the Democratic uration of population groups. Only the tra-
Republic of Congo and Sudan. ditional results are reported here but broadly
12. Inequality in Africa as a whole is higher than similar patterns result from the Elbers and
in Latin America (0.528); Asia other than others (2008) method (even though the esti-
China (0.450) and China (0.427); mature mated between-group shares are generally
economies (0.419); the Russian Federation, higher in the latter variant). For the analy-
Central Asia, and Southeast Europe (0.419); sis in this section, the mean log deviation is
and India (0.331) (Lakner and Milanovic the measure of inequality. Unlike the Gini, it
2015). Estimates by Pinkovskiy and Sala-i- is additively decomposable, a mathematical
Martín (2014) for Africa are even higher but property that is desirable in this context.
show a decrease. However, their estimates 14. Region typically refers to the administrative
are not drawn solely from a set of recent region (for example, province) in which the
surveys in Africa but rather from a combi- household resides. Education denotes the
nation of inequality measures from surveys, highest level of education of the household
national accounts, mean and growth rates, head (none, incomplete primary, completed
and interpolations and extrapolations for primary, completed lower secondary, uni-
missing inequality data (including impu- versity, other). Employment refers to the
tations of inequality measures from other main economic activity of the household
countries if no survey is available for a head (employee, employer/self-employed in
country). agriculture, employer/self-employed outside
13. There are several main approaches to decom- agriculture, other). Gender and age refer
posing inequality into within- and between- to the household head. The demographic
group inequality. The traditional version of categories are one or two adults without
the decomposition apportions total inequal- children, one or two adults and fewer than
ity into a component explained by differ- three children, one or two adults and three
ences in mean consumption between groups children or more, three adults or more with-
and a component that reflects inequality out children, three adults or more and up
within each group. The between-group com- to three children, three adults or more, and
ponent measures the share of overall inequal- four children or more.
ity that would be obtained if every individual 15. The results for urban/rural and education
had the average consumption level of his are less pronounced than those in Belhaj
or her group. However, as Elbers and oth- Hassine (2015), who studies 12 countries
ers (2008) note, in this approach between- in the Middle East and North Africa. She
group inequality reaches a maximum if each finds that gaps between regions account
individual constitutes a separate group—the for a larger share of inequality than gaps
yardstick against which between-group gaps within regions. Some of the inequality
140 POVERTY IN A RISING AFRICA
occupational mobility in Côte d’Ivoire in Stewarding Africa’s Natural Resources for All.
1985–88 (four waves), Ghana in 1988–2006 Geneva: Africa Progress Panel.
(five waves), Guinea in 1995, Madagascar in Alvaredo, Facundo, and Thomas Piketty. 2015.
1994, and Uganda in 1993; Hertz and oth- “Measuring Top Incomes and Inequality in the
ers (2007) examine educational mobility in Middle East.” Paris School of Economics.
Ethiopia in 1994, Ghana in 1998, and South Anand, Sudhir, and Paul Segal. 2015. “The
Africa in 1998; and Lambert, Ravallion, and Global Distribution of Income.” In Handbook
van de Walle (2014) analyze occupational of Income Distribution, vol. 2, part A, edited
mobility in Senegal. by Anthony B. Atkinson and François Bour-
26. These two measures of mobility—the inter- guignon, 937–79. Amsterdam: North-Holland.
generational gradient and the correlation Asian Development Bank. 2014. Inequality in
coefficient—can produce different findings Asia and the Pacific: Trends, Drivers, and
in the same setting. The intergenerational Policy Implications, edited by Ravi Kanbur,
gradient may decline over time (implying Changyong Rhee, and Juzhong Zhuang.
more mobility), but the correlation between Atkinson, Anthony B., and Andrea Brandolini.
the educational attainment of a child and 2010. “On Analyzing the World Distribution
parent can remain constant (implying no of Income.” World Bank Economic Review 24
change in mobility) (Hertz and others 2007). (1): 1–37.
This divergence may result from a reduc- Atkinson, Anthony B., Thomas Piketty, and
tion in the inequality of schooling in the Emmanuel Saez. 2011. “Top Incomes in the
child’s generation (for example, achieving Long Run of History.” Journal of Economic
universal primary education) relative to the Literature 49 (1): 3–71.
parents’ generation and a drop in the persis- Azomahou, Théophile, and Eleni A. Yitbarek.
tence effect—that is, education in the recent 2015. “Intergenerational Mobility in Africa:
birth cohort has become less dependent on Has Progress Been Inclusive?” Working
parental schooling than parental education paper, Maastricht University, Maastricht, the
was on the educational attainment of the Netherlands.
grandparents. Bagchi, Sutirtha, and Jan Svejnar. 2015. “Does
27. All billionaires included in this analysis are Wealth Inequality Matter for Growth? The
both citizens and residents of the region. Effect of Billionaire Wealth, Income Distribu-
Nathan Kirsh, a citizen of Swaziland who tion, and Poverty.” Journal of Comparative
resides in London, is thus excluded. Forbes Economics 43 (3): 505–30.
also excludes family fortunes, such as the Banerjee, Abhijit, and Thomas Piketty. 2005.
Chandaria family in Kenya and the Madh- “Top Indian Incomes, 1922–2000.” World
vanis in Uganda, if the wealth is believed to Bank Economic Review 19 (1): 1–20.
have been dispersed among family members. Belhaj Hassine, Nadia, 2015. “Economic Inequal-
28. The Oxfam estimates also draw on the data ity in the Arab Region.” World Development
from Credit Suisse (2014). Because infor- 66: 532–56.
mation on assets and debts from household Berg, Andrew, Jonathan D. Ostry, and Jeromin
survey data is rarely available in African Zettelmeyer. 2012. “What Makes Growth
countries, estimates for the region are based Sustained?” Journal of Development Econom-
largely on imputations from other low- ics 98 (2): 149–66.
income countries. In general, it is difficult to Bhalotra, Sonia, Kenneth Harttgen, and Stephan
compare assets because the appropriate data Klasen. 2015. “The Impact of School Fees on
are not available. For example, important Schooling Outcomes and the Intergenerational
assets held by the poor, such as landhold- Transmission of Education.” University of
ings, could be undervalued and assets held Bristol, United Kingdom.
Bhorat, Haroon, Karmen Naidoo, and Kavisha
by the rich may be hidden.
Pillay. 2015. “Growth, Poverty and Inequal-
ity Interactions in Africa: An Overview of
Key Issues.” University of Cape Town, South
References Africa.
Africa Progress Panel. 2013. Africa Prog- Black, Sandra E., and Paul J. Devereux. 2011.
ress Report 2013: Equity in Extractives; “Recent Developments in Intergenerational
142 POVERTY IN A RISING AFRICA
Mobility.” In Handbook of Labor Econom- Cobham, Alex, and Andy Sumner. 2013. “Is It
ics, vol. 4, part B, edited by Orley Ashenfel- All About the Tails? The Palma Measure of
ter and David Card, 1487–541. San Diego: Income Inequality.” Center for Global Devel-
North-Holland. opment Working Paper 343, Washington, DC.
Blau, Judith R., and Peter M. Blau. 1982. “The Cogneau, Denis, and Sandrine Mesplé-Somps.
Cost of Inequality: Metropolitan Structure 2008. “Inequality of Opportunity for Income
and Violent Crime.” American Sociological in Five Countries of Africa.” Research on Eco-
Review 47 (1): 114–29. nomic Inequality 16: 99–128.
Blundell, Richard, Luigi Pistaferri, and Ian Pres- Cornia, Giovanni A ndrea. 2014. “Income
ton. 2008. “Consumption Inequality and Par- Inequality Levels, Trends and Determinants in
tial Insurance.” American Economic Review Sub-Saharan Africa: An Overview of the Main
98 (5): 1887–921. Changes.” Università degli Studi di Firenze,
Bossuroy, Thomas, and Denis Cogneau. 2013. Florence.
“Social Mobility in Five African Coun- Cowell, Frank A. 2000. “Measurement of
tries.” Review of Income and Wealth 59 (S1): Inequality.” In Handbook of Income Distri-
S84–S110. bution, vol. 1, edited by Anthony B. Atkinson
Bourguignon, François. 2004. “The Poverty- and François Bourguignon, 87–166. Amster-
Growth-Inequality Triangle.” ICRIER Work- dam: North-Holland.
ing Paper 125, Indian Council for Research on Cramer, Christopher. 2005. “Inequality and
International Economic Relations, New Delhi. Conflict: A Review of an Age-Old Concern.”
Bruno, Michael, Martin Ravallion, and Lyn United Nations Identities, Conflict and Cohe-
Squire. 1998. “Equity and Growth in Devel- sion Programme Paper 11, United Nations
oping Countries: Old and New Perspectives on Research Institute for Social Development,
the Policy Issues.” In Income Distribution and Geneva.
High-Quality Growth, edited by Vito Tanzi and Credit Suisse. 2014. Global Wealth Databook
Ke-Young Chu, 117– 46. Cambridge, MA: 2014. Zürich: Research Institute, Credit
MIT Press. Suisse.
Brunori, Paolo, Francisco H. G. Ferreira, and Vito Dabalen, Andrew, A mbar Narayan, Jaime
Peragine. 2013. “Inequality of Opportunity, Saavedra-Chanduvi, and Alejandro Hoyos
Income Inequality and Mobility: Some Inter- Suarez, with Ana Abras and Sailesh Tiwari.
national Comparisons.” In Getting Devel- 2015. Do African Children Have an Equal
opment Right: Structural Transformation, Chance? A Human Opportunity Report for
Inclusion and Sustainability in the Post-Crisis Sub-Saharan Africa. Directions in Develop-
Era, edited by Eva Paus, 85–116. Basingstoke, ment Series. Washington, DC: World Bank.
United Kingdom: Palgrave Macmillan. Davies, James B., Susanna Sandström, Anthony
Brunori, Paolo, Flaviana Palmisano, and Vito Per- Shorrocks, and Edward N. Wolff. 2011. “The
agine. 2015a. “Inequality of Opportunity dur- Level and Distribution of Global Household
ing the Great Recession in Uganda.” WIDER Wealth.” The Economic Journal 121 (551):
Working Paper 2015/039, United Nations Uni- 223–54.
versity–World Institute for Development Eco- de la Fuente, Alejandro, Andreas Murr, and
nomics Research, Helsinki. Ericka Rascón. 2015. Mapping Subnational
———. 2015b. “Inequality of Opportunity in Poverty in Zambia. World Bank Group and
Sub-Saharan Africa.” University of Bari, Italy. Republic of Zambia Central Statistical Office.
Burgis, Tom. 2015. The Looting Machine. War- de Magalhães, Leandro, and Raül Santaeulàlia-
lords, Oligarchs, Corporations, Smugglers, Llopis. 2015. “The Consumption, Income, and
and the Theft of Africa’s Wealth. New York: Wealth of the Poorest: Cross-Sectional Facts
Public Affairs. of Rural and Urban Sub-Saharan Africa for
Checchi, Daniele, and Vito Peragine. 2010. Macroeconomists.” Policy Research Working
“Inequality of Opportunity in Italy.” Journal Paper 7337, World Bank, Washington, DC.
of Economic Inequality 8 (4): 429–50. Deaton, Angus S. 2005. “Measuring Poverty in
Clementi, Fabio, Andrew Dabalen, Vasco Molini, a Growing World (or Measuring Growth in a
and Francesco Schettino. Forthcoming. Poor World).” Review of Economics and Sta-
“When the Centre Cannot Hold: Patterns of tistics 87 (1): 1–19.
Polarization in Nigeria.” Review of Income Deaton, Angus S., and Salman Zaidi. 2002.
and Wealth. “Guidelines for Constructing Consumption
INEQUALITY IN AFRICA 143
Aggregates for Welfare Analysis.” Living Stan- Fitz, Nicholas. 2015. “Economic Inequality: It’s
dards Measurement Study Working Paper 135, Far Worse than You Think.” Scientific Ameri-
World Bank, Washington, DC. can March 31.
Deininger, Klaus, and Lyn Squire. 1996. “A Fleurbaey, Marc. 2008. Fairness, Responsibility,
New Data Set Measuring Income Inequality.” and Welfare. Oxford, UK: Oxford University
World Bank Economic Review 10 (3): 565–91. Press.
De Vreyer, Philippe, and Sylvie Lambert. 2014. Fleurbaey, Marc, and Vito Peragine. 2013. “Ex
“Intra-household Inequalities and Poverty in Ante versus Ex Post Equality of Opportunity.”
Senegal.” Unpublished working paper, Paris Economica 80: 118–30.
School of Economics. Forbes Media. “The World’s Billionaires.” Jersey
De Vreyer, Philippe, Sylvie Lambert, Abla Safir, City, NJ. http://www.forbes.com/billionaires
and Momar B. Sylla. 2008. “Pauvreté et /list/41/#version:static.
structure familiale: Pourquoi une nouvelle Fosu, Augustin Kwasi. 2014. “Growth, Inequal-
enquête?” Statéco 102. ity, and Poverty in Sub-Saharan Africa: Recent
Díaz-Giménez, Javier, Andy Glover, and José- Progress in a Global Context.” Oxford Devel-
Víctor Ríos-Rull. 2011. “Facts on the Distribu- opment Studies 43 (1): 44–59.
tions of Earnings, Income, and Wealth in the Gallup World Poll. 2013. http://www.gallup.com
United States: 2007 Update.” Federal Reserve /poll/174263/belief-work-ethic-strong-across
Bank of Minneapolis Quarterly Review 34 (1): -africa.aspx.
2–31. Gandhi, Aditi, and Michael Walton. 2012.
Easterly, William, and Ross Levine. 1997. “Africa’s “Where Do India’s Billionaires Get Their
Growth Tragedy: Policies and Ethnic Divi- Wealth?” Economic & Political Weekly 47
sions.” Quarterly Journal of Economics 112 (40): 10–14.
(4): 1203–50. Guénard, Charlotte, and Sandrine Mesplé-Somps.
Elbers, Chris, Peter Lanjouw, Johan A. Misti- 2010. “Measuring Inequalities: Do Household
aen, and Berk Özler. 2008. “Reinterpreting Surveys Paint a Realistic Picture?” Review of
Between-Group Inequality.” Journal of Eco- Income and Wealth 56 (3): 519–38.
nomic Inequality 6 (3): 231–45. Hertz, Tom., Tamara Jayasundera, Patrizio
Ferreira, Francisco H. G., and Jérémie Gignoux. Piraino, Sibel Selcuk, Nicole Smith, and Alina
2011. “The Measurement of Inequality of Verashchagina. 2007. “The Inheritance of
Opportunity: Theory and an Application Educational Inequality: International Com-
to Latin America.” Review of Income and parisons and Fifty-Year Trends.” B.E. Journal
Wealth 57 (4): 622–57. of Economic Analysis & Policy 7 (2): 1–46.
Ferreira, Francisco H. G., Christoph Lakner, Hlasny, Vladimir, and Paolo Verme. 2013. “Top
Maria Ana Lugo, and Berk Özler. 2014. Incomes and the Measurement of Inequality in
“Inequality of Opportunity and Economic Egypt.” Policy Research Working Paper 6557,
Growth: A Cross-Country Analysis.” Policy World Bank, Washington, DC.
Research Working Paper 6915, World Bank, Hoff, Karla. 2012. “The Effect of Inequality on
Washington, DC. Aspirations.” Background paper for Address-
Ferreira, Francisco, H. G., Julian Messina, Jamele ing Inequality in South Asia, edited by Martín
Rigolini, Luis-Felipe López-Calva, Maria Rama, Tara Béteille, Yue Li, Pradeep K. Mitra,
Ana Lugo, and Renos Vakis. 2013. Economic and John Lincoln Newman. Washington, DC:
Mobility and the Rise of the Latin American World Bank.
Middle Class. Washington, DC: World Bank. Huber, John D., and Laura Mayoral. 2014.
Ferreira, Francisco H. G., and Vito Peragine. “Inequality, Ethnicity and Civil Conflict.”
2015. “Equality of Opportunity: Theory and Unpublished working paper, Columbia Uni-
Evidence.” In Handbook of Well Being and versity, New York.
Public Policy, edited by Matthew D. Adler and Jenkins, Stephen P. 2014. “World Income Inequal-
Marc Fleurbaey. New York: Oxford University ity Databases: An Assessment of WIID and
Press. SWIID.” IZA Discussion Paper 8501, Institute
Fields, Gary. 2000. “The Dynamics of Poverty, for the Study of Labor, Bonn.
Inequality and Economic Well-being: African Jirasavetakul, La-Bhus, and Christoph Lakner.
Economic Growth in Comparative Perspec- 2015. “The African Distribution of Consump-
tive.” Journal of African Economies 9 (Supple- tion Expenditure.” Unpublished working
ment): 45–78. paper, World Bank, Washington, DC.
144 POVERTY IN A RISING AFRICA
Kanbur, Ravi, and Adam Wagstaff. 2014. “How Linton, Ralph. 1936. The Study of Man. New
Useful Is Inequality of Opportunity as a Policy York: Appleton-Century.
Construct?” Policy Research Working Paper Mansuri, Ghazala, and Vijayendra Rao. 2013.
6980, World Bank, Washington, DC. Localizing Development: Does Participation
Keefer, Philip, and Stephen Knack. 2002. “Polar- Work? Policy Research Report. Washington,
ization, Politics and Property Rights: Links DC: World Bank.
between Inequality and Growth.” Public Marrero, Gustavo A., and Juan G. Rodrí-
Choice 111: 127–54. guez. 2013. “Inequality of Opportunity and
Klasen, Stephan. 2004. “In Search of the Holy Growth.” Journal of Development Economics
Grail: How to Achieve Pro-Poor Growth.” In 104: 107–22.
Toward Pro-Poor Policies: Aid, Institutions, Miguel, Edward, and Mary Kay Gugerty. 2005.
and Globalization, edited by Bertil Tungod- “Ethnic Diversity, Social Sanctions, and Public
den, Nicholas Stern, and Ivar Kolstad, 63–93. Goods in Kenya.” Journal of Public Econom-
Washington, DC: World Bank. ics 89: 2325–68.
Knight, Frank. 2015. The Wealth Report 2015: Milanovic, Branko. 2005. Worlds Apart: Mea-
Global Perspectives on Prime Property and suring International and Global Inequality.
Wealth. London. Princeton, NJ: Princeton University Press.
Korinek, Anton, Johan A. Mistiaen, and Martin ———. 2011. The Haves and the Have-Nots:
Ravallion. 2006. “Survey Nonresponse and A Brief and Idiosyncratic History of Global
the Distribution of Income.” Journal of Eco- Inequality. New York: Basic Books.
nomic Inequality 4 (1): 33–55. Morival, Elodie. 2011. “Top Incomes and
Krueger, Dirk, Fabrizio Perri, Luigi Pistaferri, and
Racial Inequality in South Africa: Evidence
Giovanni L. Violante. 2010. “Cross-Sectional
from Tax Statistics and Household Surveys,
Facts for Macroeconomists.” Review of Eco-
1993–2008.” Master’s thesis, Paris School of
nomic Dynamics 13 (1): 1–14.
Economics.
Kuznets, Simon. 1955. “Economic Growth and
New World Wealth. 2014. Wealth in Kenya: The
Income Inequality.” American Economic
Future of Kenyan HNWIs. Johannesburg.
Review 45 (1): 1–28.
OECD (Organisation for Economic Co-operation
Lakner, Christoph. 2015. “The Ten Richest
and Development). 2015. In It Together: Why
Africans Own as Much as the Poorest Half
Less Inequality Benefits All. Paris.
of the Continent.” March 11, World Bank,
Washington, DC. http://blogs.worldbank.org Olinto, Pedro, Gabriel Lara Ibarra, and Jaime
/developmenttalk/ten-richest-africans-own Saavedra-Chanduvi. 2014. “Accelerating
-much-poorest-half-continent. Poverty Reduction in a Less Poor World:
Lakner, Christoph, and Branko Milanovic. 2015. The Roles of Growth and Inequality.” Policy
“Global Income Distribution: From the Fall of Research Working Paper 6855, World Bank,
the Berlin Wall to the Great Recession.” World Washington, DC.
Bank Economic Review. Advance Access pub- Olson, Mancur, Jr. 1965. The Logic of Collec-
lished September 26, 2015. tive Action: Public Goods and the Theory of
Lambert, Sylvie, Martin Ravallion, and Domi- Groups. Cambridge, MA: Harvard University
nique van de Walle. 2014. “Intergenerational Press.
Mobility and Interpersonal Inequality in an Oxfam International. 2015. “Wealth: Having It
African Economy.” Journal of Development All and Wanting More.” Oxfam Issue Briefing
Economics 110: 327–44. (January), Oxford.
Langer, Arnim, and Frances Stewart. 2015. Palma, José Gabriel. 2006. “Globalizing Inequal-
Regional Imbalances, Horizontal Inequali- ity: ‘Centrifugal’ and ‘Centripetal’ Forces at
ties, and Violent Conflicts: Insights from Four Work.” Revue Tiers-Monde 186: (2): 249–80.
West African Countries. Fragility, Conflict, ———. 2011. “Homogeneous Middles vs. Hetero-
and Violence Group, World Bank, Washing- geneous Tails, and the End of the ‘Inverted U’:
ton, DC. It’s All about the Share of the Rich.” Develop-
Lichbach, Mark Irving. 1989. “An Evaluation ment and Change 42 (1): 87–153.
of ‘Does Economic Inequality Breed Politi- Paris School of Economics. World Top Incomes
cal Conflict?’ Studies.” World Politics 41 (4): Database. http://topincomes.parisschoolof
431–70. economics.eu/#Database.
INEQUALITY IN AFRICA 145
Pew Research Center. 2013. Despite Challenges, ———. Forthcoming b. “On the Assessment
Africans Are Optimistic about the Future, and Use of Cross-National Income Inequality
Washington, DC. http://www.pewglobal. Datasets.” Journal of Economic Inequality.
org/2013/11/08/despite-challenges-africans Stewart, Frances, ed. 2008. Horizontal Inequali-
-are-optimistic-about-the-future/. ties and Conflict: Understanding Group Vio-
Piketty, Thomas 2014. Capital in the Twenty- lence in Multiethnic Societies. Basingstoke,
First Century. Cambridge, MA: Belknap Press UK: Palgrave Macmillan.
of Harvard University Press. Stiglitz, Joseph E. 2012. The Price of Inequality:
Pinkovskiy, Maxim, and Xavier Sala-i-Martín. How Today’s Divided Society Endangers Our
2014. “Africa Is on Time.” Journal of Eco- Future. New York: W. W. Norton.
nomic Growth 19 (3): 311–38. Strauss, John, Germano Mwabu, and Kathleen
Piraino, Patrizio. 2015. “Intergenerational Earn- Beegle. 2000. “Intra-household Allocations: A
ings Mobility and Equality of Opportunity Review of Theories and Empirical Evidence.”
in South Africa.” World Development 67: Journal of African Economies 9 (Supplement
396–405. 1): 83–143.
Rama, Martín, Tara Béteille, Yue Li, Pradeep Székely, Miguel, and Marianne Hilgert. 2007.
K. Mitra, and John Lincoln Newman. 2015. “What’s behind the Inequality We Measure?
Addressing Inequality in South Asia. South An Investigation Using Latin American Data.”
Asia Development Matters Series. Washing- Oxford Development Studies 35 (2): 197–217.
ton, DC: World Bank. U N DP (United Nations Development Pro-
Ravallion, Martin. 2001. “Growth, Inequal- gramme). 2014. Humanity Divided: Con-
ity and Poverty: Looking Beyond Averages.” fronting Inequality in Developing Countries.
World Development 29 (11): 1803–15. Bureau for Development Policy. New York:
Roemer, John. 2000. Equality of Opportunity. UNDP.
Cambridge, MA: Harvard University Press. Van de Gaer, Dirk, and Xavier Ramos. 2015.
Roemer, John, and Alain Trannoy. 2015. “Equal- “Measurement of Inequality of Opportunity
ity of Opportunity” In Handbook of Income Based on Counterfactuals.” Paper presented at
Distribution, vol. 2, edited by Anthony B. the Sixth Meeting of the Society for the Study
Atkinson and François Bourguignon, 217– of Economic Inequality (ECINEQ), Luxem-
300. Amsterdam: North-Holland. bourg, July 13–15.
Santaeulàlia-Llopis, Raül, and Yu Zheng 2015. World Bank. 2005. World Development Report
“The Price of Growth: Consumption Insur- 2006: Equity and Development. Washington,
ance in China 1989-2009.” Unpublished work- DC: World Bank.
ing paper, Washington University, St. Louis. ———. 2011. World Development Report 2012:
Shepherd, Andrew, Lucy Scott, Chiara Mari- Gender Equality and Development. Washing-
otti, Flora Kessy, Raghav Gaiha, Lucia da ton, DC: World Bank.
Corta, Katharina Hanifnia, and others. 2014. ———. 2014a. Africa’s Pulse, vol. 10 (October),
Chronic Poverty Report 2014 –2015: The Washington, DC: World Bank.
Road to Zero Extreme Poverty. London: ———. 2014b. Taking Stock: An Update on
Chronic Poverty Advisory Network, Overseas Vietnam’s Recent Economic Developments.
Development Institute. Washington, DC: World Bank.
Solt, Frederick. Forthcoming a. “The Standard-
ized World Income Inequality Database.”
Social Science Quarterly.
ECO-AUDIT
Environmental Benefits Statement
The World Bank Group is committed to reducing its environmental footprint. In
support of this commitment, the Publishing and Knowledge Division leverages
electronic publishing options and print-on-demand technology, which is located in
regional hubs worldwide. Together, these initiatives enable print runs to be lowered
and shipping distances decreased, resulting in reduced paper consumption, chemi-
cal use, greenhouse gas emissions, and waste.
The Publishing and Knowledge Division follows the recommended stan-
dards for paper use set by the Green Press Initiative. The majority of our books
are printed on Forest Stewardship Council (FSC)–certified paper, with nearly all
containing 50–100 percent recycled content. The recycled fiber in our book paper
is either unbleached or bleached using totally chlorine-free (TCF), processed
chlorine-free (PCF), or enhanced elemental chlorine-free (EECF) processes.
More information about the Bank’s environmental philosophy can be found at
http://www.worldbank.org/corporateresponsibility.