Poverty 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 Some rights reserved 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. Rights and Permissions 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 Library of Congress Cataloging-in-Publication Data Names: Beegle, Kathleen, 1969- author. | World Bank, issuing body. Title: Poverty in a rising Africa / [Kathleen Beegle, Luc Christiaensen, Andrew Dabalen, Isis Gaddis]. Description: Washington DC : World Bank, [2016] Identifiers: LCCN 2016009159 | ISBN 9781464807237 Subjects: LCSH: Poverty—Africa. | Economic development—Africa. | Africa—Economic conditions. Classification: LCC HC800.Z9 P6187 2016 | DDC 339.4/6096—dc23 LC record available at http://lccn.loc.gov/2016009159 Cover design: Bill Pragluski, Critical Stages LLC. Cover image: Africa Footprints, lithograph © Richard Long/World Bank Art Program. Further permission required for reuse. Contents Foreword . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi Acknowledgments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii About the Authors and Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv 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 2. Revisiting Poverty Trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 Trends Using Comparable and Better-Quality Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 Robustness to Reliance on GDP Imputation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 Profiling the Poor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 The Movement of People into and out of Poverty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 Concluding Remarks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 3. Poverty from a Nonmonetary Perspective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 The Capability Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 Levels of and Trends in Well-Being . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 Multiple Deprivation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 4. Inequality in Africa . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 Perceptions of Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 Measurement of Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120 Inequality Patterns and Trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 Unequal Opportunities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 Extreme Wealth and Billionaires . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 Concluding Remarks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141 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 3.5 Many sixth graders in Africa lack basic reading skills . . . . . . . . . . . . . . . . . . . . . . . . 90 3.6 Life expectancy in Africa is rising, but it remains the lowest in the world . . . . . . . . . . 90 3.7 Healthy life expectancy at birth ranges widely . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 3.8 Healthy life expectancy is lower in resource-rich countries . . . . . . . . . . . . . . . . . . . . 91 3.9 Vaccination rates rose and child mortality from malaria fell . . . . . . . . . . . . . . . . . . . . 92 3.10 Many factors contribute to underweight and obesity in African women . . . . . . . . . . . 95 3.11 About 1 in 10 Africans suffers from a disability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 3.12 Conflict slows progress in reducing under-five mortality and increasing life expectancy in Africa . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 3.13 The incidence and acceptance of domestic violence in Africa has declined . . . . . . . . . 99 3.14 Acceptance of domestic violence is twice as high in Africa as in other developing countries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 3.15 Women’s acceptance of domestic violence varies widely across countries in Africa . . 100 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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 3.17 Voice and accountability levels remain low in Africa . . . . . . . . . . . . . . . . . . . . . . . . . 102 3.18 Voice and accountability are stronger in middle-income countries, and often lower in resource-rich economies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 3.19 Less than half of Africa’s population has regular access to mass media . . . . . . . . . . . 103 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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104 3.21 A large share of African women suffers from multiple deprivations . . . . . . . . . . . . . 106 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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 3.23 Country ranking changes only slightly when the dimension threshold changes . . . . . 109 4.1 Views on inequality differ within and across countries . . . . . . . . . . . . . . . . . . . . . . . 119 4.2 Survey respondents’ perceptions of the adequacy of their government’s efforts to narrow the income gap differ across countries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 B4.1.1 The Lorenz curve illustrates the Gini measure of inequality . . . . . . . . . . . . . . . . . . . 120 B4.1.2 Different inequality measures reveal a similar story. . . . . . . . . . . . . . . . . . . . . . . . . . 121 B4.2.1 Standardized World Income Inequality Database (SWIID) estimates of the Gini index show great variability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 4.3 The world’s most unequal countries are in Africa . . . . . . . . . . . . . . . . . . . . . . . . . . . 124 4.4 Inequality rose in about half of the countries and fell in the other half . . . . . . . . . . . 126 4.5 There is no systematic relationship between growth and inequality in Africa . . . . . . 127 4.6 Declining inequality is often associated with declining poverty . . . . . . . . . . . . . . . . . 127 4.7 The richest households in Africa live mostly in the richer countries. . . . . . . . . . . . . . 128 4.8 Location, education, and demographics are the most important drivers of inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129 4.9 Unequal opportunities account for up to 20 percent of inequality in Africa . . . . . . . 132 4.10 Intergenerational persistence in schooling is weaker among younger Africans than older Africans . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134 4.11 Billionaire wealth in Africa is growing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136 4.12 Extreme wealth increases with GDP in Africa and elsewhere . . . . . . . . . . . . . . . . . . 136 Contents ix 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 AIDS acquired immune deficiency syndrome BMI body mass index CPI consumer price index DHS Demographic and Health Survey GDP gross domestic product GHS General Household Survey HIV human immunodeficiency virus ICP International Comparison Program IDP internally displaced person MDG Millennium Development Goal MICS Multiple Indicator Cluster Survey MLD mean log deviation PPP purchasing power parity S2S survey-to-survey SDG Sustainable Development Goal SWIID Standardized World Income Inequality Database WGI Worldwide Governance Indicators xvii Key Messages Measuring poverty in Africa remains a challenge. • The coverage, comparability, and quality of household surveys to monitor living standards have improved. Still, by 2012, only 27 of the region’s 48 countries had conducted at least two comparable surveys since 1990 to track poverty. • Regular and good-quality GDP, price, and census data are also lacking. • Technical approaches can fill in some gaps, but there is no good alternative to regular and good-quality data. A regionwide effort to strengthen Africa’s statistics is called for. Poverty in Africa may be lower than current estimates suggest, but more people are poor today than in 1990. • The latest estimates from the World Bank show that the share of Africans who are poor fell from 57 percent in 1990 to 43 percent in 2012. Limiting estimates to comparable surveys, drawing on nonconsumption surveys, and applying alternative price deflators suggest that poverty may have declined by even more. • Nonetheless, even given the most optimistic estimates, still many more people are poor because of population growth: more than 330 million in 2012, up from about 280 million in 1990. • Poverty reduction has been slowest in fragile countries, and rural areas remain much poorer, although the urban-rural gap has narrowed. Chronic poverty is substantial. Nonmonetary dimensions of poverty have been improving. • Health, nutrition, education, and empowerment have improved; and violence has diminished. • But the challenges remain enormous: more than two in five adults are still illiterate, and the quality of schooling is often low; after a decade of relative peace, conflict is on the rise. • Nonmonetary welfare indicators are weaker in resource-rich countries, conditional on income, pointing to the unmet potential of natural resource wealth. Inequality in Africa has many dimensions. • The data do not reveal a systematic increase in inequality across countries in Africa. But these data do not capture extremely wealthy Africans, whose numbers and wealth are increasing. • Spatial inequalities (differences between urban and rural areas and across regions) are large. • Intergenerational mobility in areas such as education and occupation has improved, but mobility is still low and perpetuates inequality. 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 comparable surveys during 1990–2012 (map is driven by capital-intensive sectors such as O.1). On the upside, they represent more than mining and oil production (Loayza and Rad- three-quarters of Africa’s population. datz 2010) and may lead to poverty reduc- The estimation of poverty also requires tion being overestimated. Caution is therefore data on price changes. For cross-country counseled, especially when extrapolating to a comparisons of poverty in a base year, distant future (or past). 2011 in this case, nominal consump- tion must be converted to 2011 price lev- els. The main method used to make this Improving Data on Poverty adjustment is the consumer price index Lack of funding and low capacity are often (CPI), which relies on both the collection of cited as main drivers for the data gaps in country-specific price data and basket weights Africa. But middle-income status is not associ- of consumer items to measure inflation. The ated with the number of consumption surveys CPI suffers from three specific problems in a country conducts, and countries receiv- Africa, in addition to the more general techni- ing more development aid do not have more cal difficulties. First, in many countries prices or higher-quality poverty data. In terms of are collected only from urban markets. Sec- capacity, the production of high-quality con- ond, the basket weights rely on dated house- sumption surveys and statistics is technically hold surveys and sometimes only on market complex, involving the mobilization of finan- purchases (excluding home-produced foods). cial and human resources on a large scale Third, computational errors sometimes bias and requiring the establishment of robust the data, as in Tanzania (World Bank 2007) quality-control mechanisms. But many coun- and Ghana (IMF 2003, 2007).3 tries that do not conduct household surveys to Across the globe, when surveys are not available in a given year, researchers use GDP measure poverty at the same time undertake to compute annual poverty estimates. Mis- other activities that are more or equally com- sing data are interpolated (between surveys) plex (delivering antiretroviral drugs to people and extrapolated (to years before and after with AIDS and conducting national elections, the previous and latest surveys) using GDP for example) (Hoogeveen and Nguyen 2015). growth rates (see World Bank 2015a). Not Good governance is strongly correlated with all of these GDP data are reliable, however. higher-quality data (figure O.1). Countries Ghana, for example, leapt from low-income that have better scores on safety and rule of to low-middle-income country classifica- law also have superior statistical capacity. tion after rebasing its GDP in 2010; follow- Many researchers have recently suggested ing rebasing, Nigeria surpassed South Africa that problems with the availability, compara- overnight as the biggest economy in Africa. bility, and quality of data reflect the political These examples suggest that GDP growth preferences of elites (Carletto, Jolliffe, and rates—and by extension the extrapolated Banerjee 2015; CGD 2014; Devarajan 2013; poverty reductions—may be underestimated. Florian and Byiers 2014; Hoogeveen and Another issue is that imputation based on Nguyen 2015). Political elites may not favor GDP growth rates assumes that GDP growth good-quality statistics for several reasons. translates one-to-one into household con- First, where clientelism and access to poli- sumption and that all people see their con- tics are limited, a record of achievement that sumption expand at the same pace. But GDP can be supported by good-quality statistics includes much more than household con- is unnecessary because support from a small sumption: on average across a large sample group of power brokers suffices. Second, of African countries, household consumption maintaining a patronage network is costly, surveys captured just 61 percent of GDP per and high-quality statistics come at a high capita. The assumption that growth is evenly opportunity cost. Third, poor-quality sta- distributed can also be tenuous when growth tistics reduce accountability. The prevailing OVERVIEW 7 FIGURE O.1 Good governance and statistical capacity go together 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 Uganda Benin Mali Chad Madagascar Ghana Togo 60 Central African Zimbabwe Ethiopia Swaziland Seychelles Mauritania Zambia Republic Congo, Sierra Leone Dem. Rep. Cameroon Burundi Guinea Kenya 50 Namibia Botswana Angola Côte d’Ivoire Congo, Rep. Liberia 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 Source: Hoogeveen and Nguyen 2015. 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 Poverty rate (percent) 55 in tracking poverty, although, as with GDP extrapolation, caution is counseled when pre- dicting farther out in the past or the future (Christiaensen and others 2012; Newhouse 45 and others 2014; World Bank 2015a). Apply- ing this method to the 23 largest countries in Africa (which account for 88 percent of both the population and the poor) and keeping 35 only good-quality and comparable consump- 1990–94 1995–99 2000–04 2005–09 2010–12 tion surveys suggests that poverty declined Survey to survey PovcalNet from 55 percent in 1990–94 to 40 percent in Comparable and good-quality surveys 2010–12 (figure O.3, blue line). This decline is slightly larger than the one obtained from Source: World Bank Africa Poverty database; calculations using additional household surveys for the 23 largest countries in Africa. the World Bank’s PovcalNet for the same 23 countries (which showed the poverty rate falling from 57 percent to 43 percent) (green Engel curve (which shows households’ food line) but smaller than the 19 percentage point budget share declining as real consumption reduction obtained using the comparable and rises) remains constant over time, so that good-quality surveys and GDP imputation deviations indicate over- or underestima- for these countries (red line). tion of the price deflator used. Application Another approach to addressing consump- to urban households in 16 African countries tion data gaps is to forgo using consumption with comparable surveys during the 2000s data entirely and examine changes in house- suggests that CPIs in Africa tend to overstate hold assets. However, although changes in increases in the (urban) cost of living. Poverty asset holdings may be indicative of some in many African countries may have declined aspects of household material well-being, this faster than the data indicate if the CPI is approach does not yet serve well as a proxy or overestimated. Research on many more coun- replacement for what consumption measures. tries as well as rural areas and time periods is A final issue concerns how consumption needed to confirm these results. data from a given survey year are adjusted Taken together, this set of results sug- to the year of the international poverty line, gests that poverty declined at least as much which is 2011. National CPIs are typically as reported using the World Bank database used to inflate/deflate nominal consumption PovcalNet and that the poverty rate in Africa to this benchmark year. To address concerns may be less than 43 percent. This news is about applying CPI to adjust consumption of encouraging. Nonetheless, the challenges households, researchers can look for evidence posed by poverty remain enormous. As a of the potential level of CPI bias and the result of rapid population growth, there are implications of any bias for poverty trends. still substantially more poor people today An overestimated (underestimated) CPI will (more than 330 million in 2012) than there result in flatter (steeper) poverty trends. were in 1990 (about 280 million), even under One way to assess CPI bias is by using the most optimistic poverty reduction sce- the Engel approach (Costa 2001; Hamilton nario (that is, using comparable and good- 2001). It is based on the assumption that the quality surveys only). 10 POVERTY IN A RISING AFRICA 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 households (48 percent) are higher than Taking a Nonmonetary in female-headed households (40 percent), Perspective except in Southern Africa, where poverty among female-headed households is higher Many aspects of well-being cannot be prop- (Milazzo and van de Walle 2015). erly priced or monetarily valued (Sandel Two caveats are warranted. First, the 2012; Sen 1985), such as the ability to read smaller household size of female-headed and write, longevity and good health, secu- households (3.9 people versus 5.1) means that rity, political freedoms, social acceptance and using per capita household consumption as status, and the ability to move about and con- the welfare indicator tends to overestimate nect. Recognizing the irreducibility of these the poverty of male-headed households rela- aspects of well-being, the Human Develop- tive to female-headed households if there are ment Index (HDI) and the Multidimensional economies of scale among larger households Poverty Index (MPI) (Alkire and Santos (Lanjouw and Ravallion 1995; van de Walle 2014) focus on achievements in education, and Milazzo 2015). But household composi- longevity and health, and living standards tion also differs: the dependency ratio is 1.2 (through income, assets, or both), which they among households headed by women and 1.0 subsequently combine into a single index. among households headed by men. Counting This study expands the scope to include children as equivalent to adults can lead to freedom from violence and freedom to decide an underestimation of poverty in male versus (a proxy for the notion of self-determination female-headed households. Understanding that is critical to Sen’s capability approach).5 the differences in poverty associated with the It also examines jointness in deprivation, gender of the household head is intertwined by counting the share of people deprived in with how one defines the consumption indi- one, two, or more dimensions of poverty. cator used in measuring poverty. Second, This approach achieves a middle ground woman household heads are a diverse group. between a single index of nonmonetary pov- Widows, divorced or separated women, and erty (which requires weighting achievements single women frequently head households in the various dimensions) and a dashboard that are relatively disadvantaged , as opposed approach (which simply lists achievements to households with a temporarily absent male dimension by dimension, ignoring jointness head (van de Walle and Milazzo 2015). in deprivation) (Ferreira and Lugo 2013). The evidence examined above captures The focus in selecting indicators was on snapshots of poverty. Looking at the body outcomes (not inputs) that are measured at of evidence on the evolution of households’ the individual (not the household) level. Infor- poverty over time (that is, taking movies of mation on these indicators is now much more people’s poverty status) reveals large varia- widely available than it once was, although tion across countries. Panel data estimates of some of the comparability and quality issues chronic poverty (the share of households stay- highlighted above also apply (see, for exam- ing poor throughout) range from 6 percent ple, UNESCO 2015 for a review of data chal- to almost 70 percent. Countries with similar lenges in tracking adult literacy). poverty rates can also be quite dissimilar in Overall, Africa’s population saw substan- terms of their poverty dynamics. A system- tial progress in most nonmonetary dimen- atic assessment using synthetic two-period sions of well-being, particularly health and panels (which are less prone to measurement freedom from violence. Between 1995 and errors) constructed for 21 countries reveals 2012, adult literacy rates rose by 4 percent- that about 58 percent of the poor population age points. Gross primary enrollment rates was chronically poor (poor in every period), increased dramatically, and the gender gap with the remaining poor being poor only in education shrank. Life expectancy at transiently (in only one period) (Dabalen and birth rose 6.2 years, and the prevalence of Dang 2015). Chronic poverty remains perva- chronic malnutrition among children under sive in the region. 5 fell by 6 percentage points. The number 12 POVERTY IN A RISING AFRICA 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 50–300 (6) 50–400 (6) 50– 650 (9) 10–50 (12) 10–50 (9) 10–50 (14) 0–10 (25) 0–10 (28) 0–10 (20) 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 MAP O.3 Inequality in Africa shows a geographical pattern 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 Source: World Bank Africa Poverty database. 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 Quadrant 1 40 percent of overall inequality. Countries 2 Malawi Ethiopia 04-10 Rwanda 00-05 Togo Nigeria Zambia 98-04 with higher inequality tend to have a high Chad Ghana 98-05 Mozambique 96-02 Madagascar 05-10 Côte d’Ivoire share of their inequality driven by unequal Uganda 05-09 0 Ghana 91-98 Zambia 04-06 Cameroon education, which is an association that is not South Africa Rwanda 05-10 Senegal Ethiopia 99-04 Mozambique 02-09 Swaziland Mauritania Namibia Dem. Rep. Congo Mauritius observed for most of the other socioeconomic Uganda 09-12 Botswana Tanzania groupings. –2 Uganda 02-05 The demographic composition of the Sierra Leone Burkina Faso household also explains a large share of –4 inequality (30 percent in Senegal and 32 per- Quadrant 4 Madagascar 01-05 cent in Botswana). In countries for which data –10 0 are available to study trends in horizontal –5 5 Annualized percentage change in poverty rate inequality from the mid-1990s to the present, the main drivers—geography, education, and Survey mean increased Survey mean decreased demographics—have not changed, though Source: Countries in World Bank Africa Poverty database with comparable surveys. some variations exist at the country level. Note: Ethiopia 1995–99, an outlier, is excluded. Survey years are indicated for countries with more Inequality in Africa is the product of many than one pair of comparable surveys. forces. The circumstances in which one is born (for example, in a rural area, to unedu- cated parents) can be critical. Inequality of 2015). Not surprisingly, the wealthiest Afri- opportunity (what sociologists call ascrip- can households are much more likely to live tive inequality)—the extent to which such in countries with higher per capita GDP. circumstances dictate a large part of the out- Inequality can be decomposed into two comes among individuals in adulthood—vio- parts: inequality between groups (horizon- lates principles of fairness. tal inequality) and inequality within groups The evidence on inequality of economic (vertical inequality). Among the range of opportunity in Africa has been limited. groups one can examine, geography, educa- This report draws on surveys of 10 African tion, and demography stand out as groups for countries to explore the level of inequality which a large share of overall inequality is of economic opportunity by looking at such explained by the group to which one belongs. circumstances as ethnicity, parental educa- From the decomposition method, spatial tion and occupation, and region of birth. inequalities (by region, urban or rural, and The share of consumption inequality that so forth) explain as much as 30 percent of is attributed to inequality of opportunity is total inequality in some countries. Perhaps a as high as 20 percent (in Malawi) (because more straightforward approach to assessing of data limitations, this estimate is a lower spatial inequality is simply to look at mean bound). But inequality of opportunity is not consumption per capita across geographic necessarily associated with higher overall domains. The ratio of mean consumption inequality. between the richest and the poorest regions Another approach to measuring inequality is 2.1 in Ethiopia (regions), 3.4 in the Demo- of opportunity is to examine persistence in cratic Republic of Congo (provinces), and intergenerational education and occupation. more than 4.0 in Nigeria (states). Price differ- Does the educational attainment of a child’s ences across geographic areas drive some of parents affect a child’s schooling less than it this gap; adjusted for price differences, spa- did 50 years ago? Is a farmer’s son less likely tial inequalities are lower but are still large. to be a farmer than he was a generation ago? Education of the household head is asso- Among recent cohorts, an additional year ciated with even larger consumption gaps of schooling of one’s parents has a lower OVERVIEW 17 association with one’s own schooling than Notes it did for older generations, suggesting more 1. Throughout this report, Africa refers to Sub- equal educational opportunities for younger Saharan Africa. cohorts. Intergenerational mobility trends 2. The focus is on a range of measurement issues, are comparable to trends estimated for other including the limited availability, comparabil- developing countries. For occupation the ity, and quality of consumption data and the findings are more mixed for the five coun- remedies used to overcome these constraints. tries for which data are available. Intergener- For a range of other measurement issues— ational occupational mobility has been rising including the measurement of service flows rapidly in the Comoros and Rwanda. In con- from housing and durable goods, the conver- trast, it remains rigid in Guinea. The shift in sion of household into individual consumption the structure of occupations in the economy measures (to account for differential needs (sometimes called structural change) is not and economies of scale), and methodological the sole reason for changes in intergenera- differences in constructing poverty lines—the tional occupational mobility. Other factors, report adopts standard approaches. such as discrimination, social norms, and 3. An additional aspect to measuring cross- country poverty is converting local currency impediments to mobility (poor infrastruc- measures into a common currency. This report ture, conflict, and so forth), are also chang- adopts the new international poverty line of ing in ways that can affect mobility. $1.90/day in 2011 based on the latest round These results tell only part of the story of the purchasing power parity (PPP) exercise because household surveys are not suited to and discusses the complicated set of issues that measuring extreme wealth. Data on holders PPPs entail. of extreme wealth are difficult to collect, but 4. The five countries for which no survey data are such people are increasingly on the radar in available to estimate poverty (Eritrea, Equato- discussions of inequality around the globe. rial Guinea, Somalia, South Sudan, and Zim- Africa had 19 billionaires in 2014 accord- babwe) were assigned the regional poverty rate ing to the Forbe’s list of “The World’s Bil- based on the other 43 countries. lionaires.” Aggregate billionaire wealth 5. Sen’s capability approach provides the philo- increased steadily between 2010 and 2014 in sophical foundations for the nonmonetary Nigeria (from 0.3 percent to 3.2 percent of perspective. 6. Below-average performance in Africa’s three GDP) and South Africa (from 1.6 percent to most populous countries (Nigeria, the Demo- 3.9 percent). The number of ultra-high-net- cratic Republic of Congo, and Ethiopia) partly worth individuals (people witha net worth of drives the high levels of nonmonetary poverty at least $30 million) also rose. Few detailed in the region. studies explore the level of extreme wealth of 7. Higher life expectancy for women is possible nationals. One exception comes from Kenya, even in an environment that is disadvantageous where 8,300 people are estimated to own 62 to them, given that women are genetically pre- percent of the country’s wealth (New World disposed to live longer (Sen 2002; World Bank Wealth 2014). The share of extreme wealth 2011). derived from areas prone to political cap- ture, including extractives, has been declin- ing, while the share derived from services References and investment has been increasing. Between ACLED (Armed Conflict Location and Event 2011 and 2014, 4 out of 20 billionaires in Data Project) http://www.acleddata.com Africa derived their wealth mainly or par- /about-acled/. tially from telecommunications. Data limi- Alkire, Sabina, and Maria Emma Santos. 2014. tations make it difficult to draw conclusions “Measuring Acute Poverty in the Developing about whether the emergence of extreme World: Robustness and Scope of the Multidi- wealth in Africa is driven less by political mensional Poverty Index.” World Develop- connections than it used to be. ment 59: 251–74. 18 POVERTY IN A RISING AFRICA 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 Surveys per country per year 2.0 multiple survey sources, including inte- grated household surveys, often with a focus on consumption; the DHS, which focus 1.5 on women’s fertility decisions, health, and nutrition; the MICS, which are designed 1.0 to monitor human development outcomes, particularly among women and children; the Core Welfare Indicators Question- 0.5 naire (CWIQ) Surveys, which emphasize poverty-related indicators and service- 0 delivery outcomes; population and housing 1980 1990 2000 2010 censuses; and labor force surveys. In addi- South Asia East Asia and Paci c tion, specialized surveys conducted outside Africa Latin America and the Caribbean the national statistical system (Barometer, Europe and Central Asia Middle East and North Africa Gallup, the World Values Surveys) solicit Source: Demombynes and Sandefur 2014. citizens’ opinions on governance, leadership, political stability, corruption, and a range of social issues, including crime, social capital, analysis that would not have been possible and religious practices (box 1.1). even a decade ago. The impressive improvement in survey Consumption surveys, the building blocks data collection depicted in figure 1.1 has for measuring monetary poverty and inequal- arisen almost entirely because of the expan- ity, have not witnessed similar growth. There sion of surveys that do not collect consump- are not more surveys available today to tion data.4 Figure 1.2 provides a breakdown measure monetary poverty than there were of the types of surveys conducted in Africa in the early 1990s. The average number of in five-year periods since the 1990s. It shows consumption surveys per five-year period steady growth in the number of nonconsump- has been just under 40 since 1990, with only tion surveys during the 1990s. The number small variations around the mean. of such surveys peaked in 2000–04 but still An average of 40 consumption surveys numbered 92 in 2010–14. every five years for Africa results in less than The increase in the number of noncon- one survey per country every five years with sumption surveys has enriched knowledge which to measure poverty. Even more trou- of nonincome dimensions of poverty, such as blesome is the uneven coverage across coun- child nutrition, women’s empowerment, and tries. Between 1990 and 1999, there is not a access to services in many sectors as well as single survey with consumption data to mon- on joint deprivation across dimensions. Many itor poverty for 18 of 47 countries in Africa of these indicators are collected at the indi- (figure 1.3). Among the remaining 29 coun- vidual level and hence provide information tries, 16 each have just a single survey. As a on differences in the experiences of poverty result, for 34 of 47 countries in the region and deprivation of men and women, insights (covering 42 percent of the population), there that cannot be gained from household-level are no data on changes in poverty or con- consumption data. Chapter 3 makes exten- sumption for an entire decade. Coverage has sive use of these datasets to conduct an improved since. Data are unavailable for only 28 POVERTY IN A RISING AFRICA 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. Afrobarometer Nonsurvey Methods of Data Collection Afrobarometer is a nonpartisan research project Satellites, run mostly by the U.S. National Aeronau- that gathers data on social, political, and economic tics and Space Administration (NASA), collect data attitudes. It has conducted surveys in more than 30 on metrics such as night lights, vegetation cover, and African countries. A key feature of these surveys precipitation. The unique features of these datasets is the harmonized set of questions, which allows are their high resolution and geo-referencing. The comparison across countries and within countries data are collected from small areas at high frequency. over time. Survey questions probe attitudes toward The use of satellite data is flourishing. They democracy, governance, elections, macroeconomics have been used to study urbanization, the accuracy and markets, poverty, social capital, conflict and of GDP information, deforestation, and impend- crime, participation, and national identity. The latest ing drought or crop failure. There have also been round introduced modules on corruption, access to attempts to extend their use to understand the evo- justice, the role of China in Africa, pan-Africanism lution of poverty and inequality (Elvidge and oth- and regional integration, energy supply, tolerance, ers 2009; Mveyange 2015; Noor and others 2008; and citizenship. Data from these surveys are used Pinkovskiy and Sala-i-Martín 2015). to construct the lived poverty index (LPI), which is based on experiential measures, such as how often a. Like household surveys conducted by national households go without basic necessities (Dulani, statistics offices, these surveys rely on face-to- Mattes, and Logan 2013). Barometer surveys are face interviews with household members. Wide- spread cell phone ownership in Africa has opened also conducted in other regions of the world. up opportunities for collecting data by phone, obviating the need for face-to-face interviews. If Gallup World Poll executed well, phone surveys can collect repre- Since 2005 the Gallup World Poll has tracked issues sentative data on a wide range of topics more fre- such as economic confidence; life satisfaction; quently and at lower cost than traditional face- employment; confidence in the leadership, military, to-face surveys (Hoogeveen and others 2014). and police; religion; access to food; the environ- This approach generally relies on a baseline ment; migration; media freedom; human suffering; survey of face-to-face household interviews. The and corruption. Surveys are standardized to allow World Bank’s Listening to Africa Initiative, for example, combines a face-to-face baseline house- comparisons across countries and within countries hold survey with follow-up phone interviews of over time. Gallup recently added a question about selected respondents. This approach allows the self-reported household income to measure poverty collection of a rich dataset at baseline and a few (Phelps and Crabtree 2013). selected questions about specific issues (educa- tion, health, labor markets, and so on) at higher World Value Surveys frequency (monthly, twice a week) and at later The World Values Survey, established in 1981, is a points in order to gauge changes in the funda- global research project that explores people’s val- mental dimensions of well-being. In addition to ues and beliefs and their social and political impact collecting data for policy analysis and research, in almost 100 countries. Topics include support cell phone surveys have proven to be effective tools for monitoring service delivery failures, for democracy, tolerance of foreigners and ethnic corruption, and the breakout of conflict and epi- minorities, support for gender equality, the role of demics. Cell phone surveys have been used to religion and changing levels of religiosity, work, monitor the impacts of Ebola in Guinea, Liberia, family, politics, national identity, culture, diversity, and Sierra Leone (World Bank 2015c) and the insecurity, attitudes toward the environment, the welfare of refugees in Mali (Etang-Ndip, Hoo- impact of globalization, and subjective well-being. geveen, and Lendorfer 2015). THE STATE OF DATA FOR MEASURING POVER T Y 29 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. Number of surveys per ve-year period 140 Many fragile states, including Chad, the Democratic Republic of Congo, Sierra Leone, 120 105 103 and Togo, were part of this wave. Twenty 96 92 100 seven countries have done a survey since 2011 (map 1.1). 80 70 72 Conducting a survey does not necessarily 60 mean that the data collected are available. If 43 46 42 37 the microdata collected in a survey are not 40 33 included in the World Bank database, the data are deemed inaccessible in this report. This 20 28 definition of accessibility is a narrow one, 0 because it does not address access by the gen- 1990–94 1995–99 2000–04 2005–09 2010–14 eral public or whether users have to pay for the data, two important factors that significantly Total number of surveys Number of consumption surveys curb the usefulness of household survey data Number of nonconsumption surveys to the public and hence undermine knowledge about poverty trends and drivers in Africa. Sources: Data from the World Bank microdata library, PovcalNet, World Development Indicators, and For three countries (Equatorial Guinea, the International Household Survey Network. Note: Consumption surveys include surveys that may not be the source of the official poverty South Sudan, and Zimbabwe), recent data estimates. Nonconsumption surveys include Demographic and Health Surveys (DHSs), Multiple are not available even though surveys were Indicator Cluster Surveys (MICSs), labor force surveys, and other ad hoc surveys. 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 0 1 2 3 4 5 6 7 8 9 0 1 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 No survey One survey Two surveys (interval of six years or more) Two surveys (interval of ve years or less) Three or more surveys Source: Data from the World Bank microdata library. Note: In 2011 the number of countries increased from 47 to 48 with the independence of South Sudan. Surveys for which microdata could not be accessed are counted as not available. Four countries (rather than five) have no data for the period 2003–12. Although the Zimbabwe 2007–08 survey is available, its consumption data cannot be used for monetary measures of poverty because it was conducted at a time of hyperinflation. See also endnote 5. 30 POVERTY IN A RISING AFRICA 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 Source: Data from the World Bank microdata library. 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 • Nationally representative sample : A Between 1990 and 2012, only 27 of 48 nationally representative sample is neces- countries conducted two or more compa- sary to obtain statistics that apply to the rable surveys (map 1.2). As a result, even whole population, not merely a subgroup. some countries that have multiple surveys are Comparability is obviously impossible if unable to track poverty reliably over time. one round covers only urban households Guinea and Mali, for example, each con- and the next covers only rural areas. ducted four surveys, but none of them is com- • Seasonality: Many consumption patterns parable (box 1.2). vary over the year, which has implica- Second, there was a slight improvement in tions for measuring poverty (Kaminski, comparability between 2000 and 2014. More Christiaensen, and Gilbert 2014; Muller surveys were implemented after 2000, and 2008). In Africa, for instance, food and more of them were comparable than before cash income among farmers is plentiful 2000. after harvests and dwindles during the The picture of comparability would lean season. Comparability may be lost if appear even bleaker if a more stringent defi- survey rounds are conducted during dif- nition of comparability had been adopted. ferent months. For instance, the list of consumption items • Reporting instrument and period: Con- on which household members are asked to sumption data can be collected either report can be long (a list of specific foods) or by asking household members to recall short (if foods are grouped). It is not unusual their purchases and consumption from for surveys in the same country to change own production (farm harvest) (in the these lists dramatically from one round to the past seven days, past two weeks, the last next (from well under 100 to well over 100).8 month, and so on) or to keep a diary of In general, respondents recall more when pre- such activities (for two weeks, a month, sented with a more disaggregated list, so that or longer). A body of evidence shows that reported consumption is generally higher; a the method used matters (see Beegle and condensed list may lead to more reporting others 2012). Both the reporting period errors. Changing the list over time thus com- and the instrument (recall or diary form) promises consistency. If other factors—such should remain consistent. as the quality of fieldwork and supervision— are also taken into consideration, even fewer Based on these three criteria, 148 con- household surveys in Africa would be consid- sumption surveys conducted in Africa ered comparable. between 1990 and 2012 were reviewed Lack of comparability, combined with the for comparability.7 Figure 1.4 displays the long gap between surveys (often five years results. Blue dots indicate surveys that are or more) hampers the ability to understand comparable within the country; solid black changes in welfare over time. Although diamonds indicate noncomparable surveys. Africa is doing well in terms of the number Dotted lines connect comparable surveys. of countries on which data are available and Hollow black diamonds indicate surveys that compares reasonably well with other poor are not available. In some instances two or regions in the number of surveys per country, more cross-sections in a country with four or the region trails most other country group- more cross-sections are comparable but the ings in terms of comparable surveys, falling in other two or more are not. (South Africa, for the bottom half of the World Bank’s regional example, has two pairs of surveys that are grouping of countries (table 1.1). Since 1990 comparable with each other, but it does not the average African country conducted only have four comparable surveys). 3.8 consumption surveys (about one survey Several observations emerge from the every six years), 2.2 fewer than the devel- findings presented in figure 1.4. First, many oping world average. The average develop- consumption surveys are not comparable. ing country conducts one survey every four 32 POVERTY IN A RISING AFRICA 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 Comparable Not Not surveys comparable available • • ◆ ◆ Note: Figure is based on all household surveys conducted in Africa between 1990 and 2012. It excludes consumption surveys not used for official poverty monitoring. Not available refers to surveys for which the microdata and/or documentation could not be accessed. THE STATE OF DATA FOR MEASURING POVER T Y 33 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 25 scant. Empirical studies for developing coun- tries in other regions suggest that within- 20 country price variation can be important, at least in larger countries (Deaton and Dupriez 15 2011; Majumder, Ray, and Sinha 2012). Despite the importance of adjusting for 10 differences in the cost of living across regions in a country for capturing true living stan- 5 dards, such adjustments are not widespread. 0 In Africa, PovcalNet, which has the largest Consumption Number of collection of consumption data from house- (in million consumption items hold surveys across countries of the world, Leones) (Average rural + urban) adjusts for spatial price differences only in January ( rst month of eldwork) Angola, Burkina Faso, and South Africa. December (last month of eldwork) There is no explanation for why the adjust- ment is made only in these countries. Outside Source: Data from the 2011 Sierra Leone Integrated Household Survey. of Africa, PovcalNet data on consumption are adjusted for within-country spatial price differences in countries in Latin America and over 12 months, specifically for the House- the Caribbean, China, India, Indonesia, and, hold Budget Surveys 2000/01 and 2007, most for food only, in countries in Europe and likely reflecting enumerator fatigue (NBS Central Asia. This report uses the consump- 2009). In Sierra Leone, where households tion measure used in PovcalNet for Africa, were randomly interviewed, both the number meaning that for most countries it has not of food items and the level of consumption fell been adjusted for spatial price differences.11 steadily during the 12 months of fieldwork (figure 1.6). The number of reported food Adjusting for price changes using the CPI purchases among urban respondents fell by The CPI is used to track inflation in con- one-third over the course of the survey, a drop sumer prices. This core economic indicator that is explained only partly by seasonality. is used to index pensions, wages, taxes, and The reported urban-rural gap also narrowed, social security benefits and to anchor mon- possibly because of data quality issues. etary policy. The largest consumer price data collec- tion exercise in Africa is conducted by Sta- Price Data tistics South Africa, which regularly collects Price data are indispensable to poverty mea- 65,000 price quotations from 27,000 outlets surement. Global poverty estimates reported (ILO 2013). In other African countries, the in PovcalNet rely on two types of price number of CPI price quotations ranges from indexes: national CPIs to deflate nominal 1,150 (São Tomé and Príncipe) to 51,170 consumption to a common base year and PPP (Ethiopia). exchange rates to convert local currencies CPI calculation requires weights to aggre- into a common currency. gate the price data across items into an index. Because people living in different countries These weights typically come from budget face different prices, comparison of living share estimates from household surveys. THE STATE OF DATA FOR MEASURING POVER T Y 37 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 Source: ILO 2013. Note: Figures are weighted using the population in 2013. THE STATE OF DATA FOR MEASURING POVER T Y 39 population lives in countries that exclude subcomponents to reflect the consumption home production from weights; for another patterns of the poor or construct survey-based third, it is not clear whether the weights price deflators so that prices and weights are include home production. The CPI guidelines computed directly from household surveys. issued by the United Nations (UN 2009) leave Since there is little agreement or technical the decision on the inclusion of own produc- guidance on how to adjust nominal consump- tion in weights to the discretion of countries, tion data for price changes, countries often because the decision depends partly on what use ad hoc and context-specific methods. the index is used for. For the purpose of pov- Another approach is the Engel curve erty analysis, where own-consumed goods method, pioneered by Costa (2001) and are typically included in the consumption Hamilton (2001). It is based on the notion aggregate and valued at (proximate) market that changes in food budget shares over time prices, the weights for price indexes should reflect changes in real incomes. Chapter 2 include consumption of own production. As takes a closer look at what this method sug- with urban and plutocratic bias, whether this gests about the magnitude and direction of bias matters in measuring poverty depends the CPI bias and the implications for measur- on whether the inflation associated with ing poverty in Africa. these goods differs from the inflation associ- Despite the caveats, national CPIs are ated with other items. applied almost uniformly for across-survey Biases arising from computational and price adjustment in the context of global pov- similar errors also reduce the accuracy of erty measurement (although in cases where the CPI. In Tanzania, for instance, the CPI CPI-measured inflation rates appear highly underestimated inflation in 2002–05 because implausible, alternative inflation estimates of defective protocols for removing outliers are occasionally used). and other computational errors. The mis- takes were eventually corrected and the CPI Using purchasing power parities to measure series revised, though concerns remained global poverty that the series continued to underestimate For cross-country analysis, it is necessary inflation (Adam and others 2012; World to convert local currency values into a com- Bank 2007). Similar evidence is reported for mon currency. The approach has typically Ghana in 1999–2001 (IMF 2003, 2007). involved using PPP rather than traditional In situations where price changes are currency exchange rates to compare both politically sensitive, governments may have poverty and GDP across countries. an incentive to exert pressure on statistical The PPP exchange rate is based on a large- agencies to misreport inflation or strategi- scale effort to collect and compare prices for cally time methodological changes to reduce a set of items across all countries (see World measured inflation. If statistical agencies Bank 2014 for a detailed discussion of PPPs). are not independent, CPI-measured infla- The International Comparison Program tion may be biased downward, leading to (ICP), which is in charge of the PPP calcula- an overestimation of poverty reduction (Bar- tions, is a massive global undertaking that rionuevo 2011; Berumen and Beker 2011). covers thousands of goods and services in Although the notion of political economy 200 countries.13 About 199 countries, with bias is plausible, political influence on the 97 percent of the world’s population and 90 computation of the rate of inflation is diffi- percent of the world’s economy, participated cult to document. in the latest round (2011). In Africa 45 of 48 Because of these shortcomings of the CPI, countries (all but Eritrea, Somalia, and South poverty estimates at the national level often Sudan) participated, up from 19 in 1993 and use alternative approaches to adjust for spatial 44 in 2005. or temporal price differences. Some statistical A controversy erupted in 2014 follow- agencies and academic studies reweight CPI ing the release of the 2011 PPPs. The debate 40 POVERTY IN A RISING AFRICA 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 Jarocinski´ 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 M mb ala c oz ia wi Si amb The 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 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 Source: World Bank 2014. Note: Countries are ranked by their 2005 PPP estimate of GDP per capita. GDP per capita of Equatorial Guinea using 2011 PPP was $39,440; in the figure it is capped at $18,000, so that incomes for the other countries are distinguishable. (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 BOX 1.3 Many kinds of data in Africa are unreliable 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 (1a) (1b) (2a) (2b) (3a) (3b) 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 Number of observations 133 93 133 93 132 93 R–squared 0.251 0.432 0.098 0.390 0.096 0.189 Sources: Survey counts: International Household Survey Network, World Bank microdata library, and PovcalNet. Government effectiveness variable: Worldwide Governance Indicators. Freedom index: Freedom House. Other control variables: World Development Indicators. Note: The data set consists of one observation per country. In columns 1a and b, the dependent variable is the total number of consumption surveys con- ducted between 1990 and 2012. In columns 2a and b, the dependent variable is the share of consumption surveys that are comparable. In columns 3a and b, the dependent variable is the number of surveys that are open (that is, available to the public). The freedom index is Freedom House’s freedom of politi- cal rights and civil liberties. It ranges from 1 to 7, where 1 is the most free and 7 is the least free. Regressions control for population and land area. Standard errors are clustered at the country level. The constant term is not shown. The R-squared is for a pooled regression (African and non-African countries) with interaction terms. Significance level: * = 10 percent, ** = 5 percent, *** = 1 percent. 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 FIGURE 1.11 Good governance and statistical capacity go together 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 Uganda Benin Mali Chad Madagascar Ghana Togo 60 Central African Zimbabwe Ethiopia Swaziland Seychelles Mauritania Zambia Republic Congo, Sierra Leone Dem. Rep. Cameroon Burundi Guinea Kenya 50 Namibia Botswana Angola Côte d’Ivoire Congo, Rep. Liberia 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 Source: Hoogeveen and Nguyen 2015. 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 Recognizing Other Challenges in families own specific consumer durables, but Measuring Poverty few collect information on the (current or past) value of these items. Many consumption Several other challenges make measuring pov- measures include expenditures on education erty difficult.18 First, it is difficult to monetize the consumption of many goods and services. and health, but they understate the “true” For example, the market price of food grown consumption value if those services are subsi- and consumed by the household (or given as dized or publicly provided. a gift or wage payment) must be estimated in Second, the global monitoring of poverty order to monetize the value of that food. The uses consumption per capita as the mea- use value of housing and durable goods, when sure of welfare comparisons, dividing total included in the consumption measure, must household consumption by the number of also be estimated. Although econometric household members. Such a practice ignores techniques can be used to estimate the rental differences in consumption across household price when a home is owned by a household, members and economies of scale in house- for example, the estimates are reliable only if hold consumption. Failure to address both a robust rental market exists, which is not the issues may affect poverty comparisons across case in many rural areas of Africa. The prob- groups within and across countries. lem of imputing a use value is complicated Third, having chosen consumption as by the fact that the typical data collected in the welfare measure, a standard needs to be surveys do not always reflect the information set to determine who is poor and who is not; needed to calculate use values. For instance, different approaches exist to determine such many surveys collect information on whether a poverty line (box 1.5). BOX 1.5 What is the threshold for being poor? 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 Concluding Remarks and Rethink the financing model. The most desirable and sustainable arrangement Recommendations for financing a country’s statistical needs The production of social and economic sta- is through domestic resources. Doing so tistics in Africa has been improving over the requires elites to embrace the benefits of evi- past 20 years. More household surveys are dence-based decision making and make the being conducted. Participation in decennial collection of statistics the responsibility of census rounds is rising. More countries are an autonomous agency, run by an indepen- updating their GDP base years. Participation dent governance board and professionalized by African countries in the latest Interna- staff. The agency should have a clear mandate tional Comparison Program round reached regarding the types of data it is to collect, ded- the highest level ever. Data on governance, icated funding from general appropriations, political attitudes, and other nonmonetary and clear reporting arrangements to institu- aspects of poverty are being collected in tions that represent the electorate, such as par- greater volume, as are gender-disaggregated liament. Current political arrangements often data on health, violence, and empower- favor limited funding for statistics, perhaps ment-related issues. These data have helped to exercise influence over statistical agencies. researchers examine poverty from a broader The replacement of domestic financing by perspective. donor financing has not always been effective These improvements are welcome, but because the interests of donors are not always there is cause for concern, for three main aligned with the interests of governments. reasons. First, data production has increased Alternative financing models are therefore from a very low base. A sustained effort in needed. One model would require donors, producing data will continue to be important such as the World Bank, to finance statisti- if the region hopes to catch up with other cal production in perpetuity through grant regions. programs in countries that are unwilling to Second, many of the data that have been produce good-quality statistics. This model produced, especially consumption data, would be akin to the model the U.S. Agency are of poor quality; in the worst cases, they for International Development (USAID) fol- are unusable. For instance, of the 148 sur- lows with the DHS. Where there is domestic veys reviewed, only 78 were comparable to interest in improving the volume and quality another survey in order to track poverty. of statistics but financing is a constraint, a Only 11 countries rely on GDP base years cofinancing arrangement could be pursued. that are no more than five years old, the rec- For instance, donors could finance a larger ommended frequency of updating. share of the costs in the early stages of data Third, data problems are more than tech- production. As domestic resources expand nical. An important, often underappreciated and institutional capacity grows, that share reason for low investment in statistics in would decline. Additional incentives to Africa is that frequent and high-quality sta- increase demand through open data access, tistics do not enjoy strong support from poli- participation in regional programs for stan- ticians and policy makers. Once produced, dard setting, and additional capacity support its use does not preclude another person’s use could be built into the compact. of it. As such they can be used by indepen- Focus on results and open data access. Too dent researchers, advocacy groups, and rival many statistical support programs focus on politicians to illuminate progress but also to inputs and outputs rather than results. There audit performance of incumbents. is also weak demand for data production. Because of these problems, the founda- Opening data to public access could address tion on which to make policy and demand both problems. Public scrutiny by users and accountability for results is weak. What can policy makers could help improve qual- be done? ity and increase accountability. Knowledge THE STATE OF DATA FOR MEASURING POVER T Y 51 production externalities would follow, as challenging to use any monetary measures. research using the data expands. The survey has been used to measure other Develop and enforce methodological and aspects of well-being. operational standards. The ultimate aim of 6. Other survey design and implementation fea- tures can also render survey-based consump- improving the capacity of national statisti- tion estimates incomparable. The focus here cal offices should be to enable them to collect is on the most common types of comparabil- more frequent and higher-quality data. But ity problems. better outcomes are possible even without 7. Even though 180 surveys were identified, more frequent data collection. The average only 148 were available in the World Bank’s African country implemented four consump- microdata library and could be included in tion surveys in the past two decades, but the review. Not all of these 148 surveys were many of them cannot be used because of com- available for use by the report team, however. parability and quality concerns. Had survey Some surveys do not include a consumption methods been consistent, the data collected aggregate with which to measure poverty. could have been useful. Developing consen- Some include consumption measures but sus on international standards for measuring have not gone through a vetting process used by the World Bank. Others (such as South monetary poverty would help guide countries Africa 2000) have consumption aggregates on international best practices for measuring that are available only as grouped data. The monetary poverty. team was able to use 113 of the 148 surveys for the analysis of poverty trends. 8. In Kenya, for instance, the number of food Notes items increased from about 80 in the 1997 1. PovcalNet is the World Bank’s online analysis Welfare Monitoring Survey to more than 150 tool. It is available at http://iresearch.world in the 2005/06 Kenya Integrated Household bank.org/PovcalNet/. Budget Survey. In Zambia the number of 2. Latin American and some Europe and Central food items rose from less than 40 to more Asian countries traditionally use income than 130 between the 2006 and 2010 rounds instead of consumption to measure poverty. of the Living Conditions Monitoring Survey. Measuring household income in economies 9. The Kenya and Niger studies do not offer a dominated by subsistence agriculture and benchmark for consumption that is taken as informal self-employment, which includes true consumption. The Tanzania study pro- most African countries, is complicated. For poses that the intensive personal diary is such this reason, consumption is generally the pre- a benchmark. Both the Kenya and the Niger ferred indicator of monetary living standards studies find that diary consumption is lower and poverty. than recall consumption, but it is not clear 3. This result is based on reviews of the inven- whether the finding indicates underreporting tory in the International Household Survey in the diary survey or overestimation in the Network, a voluntary association of devel- recall survey. opment partners and member countries that 10. Poor questionnaire design (flow and question aims to improve the availability, accessibility, wording or content) is an important aspect of and quality of household surveys. quality that is not related to process. 4. Consumption surveys collect data on more 11. Using spatially deflated consumption mea- than just consumption. If they are carried out sures does not change the overall story in as integrated surveys, they provide informa- chapter 2. Some poverty estimates are lower, tion on income sources, labor, use of educa- some are higher, and many show no change tion and health care services, remittances, when consumption is adjusted for price dif- social assistance, and other socioeconomic ferences. Likewise, the inequality analysis in dimensions of households. chapter 4 is robust to using spatially deflated 5. Data from the Zimbabwe 2007–08 Income consumption measures. Consumption Expenditure Survey are avail- 12. An influential National Research Council able, but that survey was conducted during report (Schultze and Mackie 2002) argues a period of hyperinflation, making it very against including the virtual price reduc- 52 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 Jaroci ń ski. 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) Estimates 1990–94 1995–99 2000–04 2005–09 2010–12 PovcalNet 13.5 11.6 15.3 16.7 17.8 Comparable surveys only 1.4 4.7 9.3 13.5 14.7 Comparable and good-quality surveys only 1.4 3.7 7.0 12.1 17.1 Note: Number of data points needed in all periods was 215, except in 2010–12, when 129 were needed because there are 3 instead of 5 years in the period. exaggerate poverty reduction. One is the Robustness to Reliance influence of the adjustments for poor-quality on GDP Imputation surveys in Nigeria (box 2.1), which affects the level of poverty. The other is the extent To check the robustness of alternative esti- to which GDP imputations are used to fill mates of poverty trends to the reliance on in gaps, which has the potential to influence GDP imputations, we present three illustra- trends. tive sources of evidence on trends. The first The number of survey-based estimates for is the selection of a sample of countries in annual poverty rates in Africa is small (table which two or more comparable and rela- 2.1). Between 1990 and 1994, for example, tively good-quality surveys are available. only 13 percent of the 215 data points needed The second approach, survey-to-survey (S2S) for 43 countries were based on a survey esti- imputations, also entails imputations, but mate under PovcalNet—and the share is of a type that does not rely at all on GDP. much smaller if comparability and quality are The last illustration addresses one additional taken into account. Coverage rates are low in potential source of bias in the trends: the role other periods as well, although since 2005 the that prices have played since 2002. share of actual data used in PovcalNet and the adjusted data has converged. Restricting Comparable Spells Data as a the revised poverty estimates to comparable Robustness Check surveys of reasonable quality reduces the number of surveys used from 143 to 74. By Between 1990 and 2012, very few countries design, the removal of noncomparable and in Africa conducted more than two con- poor-quality data increases the number of sumption surveys that are comparable and of imputations and reliance on GDP. good quality. Having a large pool of coun- Relying on GDP estimates to fill survey tries with such data would have allowed us data gaps entails several important assump- to assess the GDP-heavy imputation trends tions. First, models assume that all aggregate against actual data. Only three countries income growth is consumed. This assump- (Ethiopia, Ghana, and Uganda) have data tion may overestimate the magnitude of a that pass this test, which is too small a sam- decrease in poverty during periods of high ple to make general conclusions. However, growth (when savings result) or overesti- data for 24 countries—out of the 27 coun- mate the increase during periods of major tries that conducted at least two comparable downturns (when people can draw on sav- surveys during this period—are available.7 ings to smooth consumption). Second, they Figure 2.2 shows the average annual percent- take for granted that growth is shared equi- age point reduction in poverty between com- tably across households, either nationwide parable surveys for these countries. or within sectors of activity, an assumption Poverty reduction varied widely across that is not always supported by empirical cou ntries. I n fou r cou ntries pover t y evidence. Third, GDP data are prone to their increased, 8 in three it stagnated; 9 in the own quality and measurement problems (Jer- other two-thirds, it fell 0.3–4.9 percentage ven 2013; Deaton 2005). points a year. More than half of the countries REVISITING POVERTY TRENDS 61 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 o 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 n an nt er ila Ca as oz za ne (G . R o 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 0 –1 –2 –3 –4 Comparable surveys Noncomparable surveys –5 2 10 05 0 10 9 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 0 00 0 00 1– – – – – 7– 1– 20 5 5 20 98 0 97 93 20 ,2 99 9 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 5 m Ta 99 So Bu ad Ca 20 (1 M ( ia ia op op hi hi Et Et Actual S2S Source: World Bank Africa Poverty database. Note: The end year is used to impute the start year, except in Ethiopia, where, because the imputation was sensitive to the choice of the base year, both results are reported. The set of covariates used to model consumption includes traits of the household head (education, occupation, employment status), household demographics, housing and asset ownership, location (rural and urban), and interactions with other variables. For S2S, the povimp stata com- mand was used (for details see Dang and Nguyen 2014). REVISITING POVERTY TRENDS 63 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 Poverty rate (percent) 55 out of 5 spells). This finding provides some validation of the S2S method. Second, trends derived from noncompa- rable surveys are not very reliable. Estimates 45 based on the S2S imputation reverse the poverty trends in 4 of the 10 countries, and the size of the gap between actual and S2S predicted poverty is substantially larger for 35 noncomparable surveys than for comparable 1990–94 1995–99 2000–04 2005–09 2010–12 surveys. These findings underscore the poten- Survey to survey PovcalNet tial importance of the exercise underpinning Comparable and good-quality surveys figure 2.1, where comparability is taken into account. Source: World Bank Africa Poverty database. Note: Sample includes the 23 largest countries in the region. The S2S line shows the estimate in the We applied the S2S approach to the larg- period based on available surveys and the S2S method described in the text. The comparable and est 23 countries in Africa in order to check good-quality line shows the trend using corrected data for these 23 countries, and PovcalNet line shows the PovcalNet estimate for these 23 countries. the robustness of trends that are largely dependent on GDP imputations.10 For these countries, the S2S model was calibrated on a countries obtained using five-year averages recent good-quality consumption survey and for each period using PovcalNet and data the estimated parameters applied to the pov- from comparable and good-quality surveys.13 erty predicting (nonconsumption) variables The S2S approach suggests a 16 percentage from other consumption and nonconsump- point decline in the poverty rate (from 55 per- tion surveys (including, for example, from the cent in 1990–94 to 39 percent in 2010–12), Demographic and Health Surveys [DHS]). only slightly higher than the 13 percentage For each country, at least one data point was point reduction estimates from PovcalNet obtained for each of five periods: 1990–94, (from 57 percent to 44 percent) but lower than 1995–99, 2000–04, 2005–09, and 2010–15. the 20 percentage point estimate based on the Annual estimates were not possible, because data corrected for comparability and quality. of insufficient survey coverage. When there The regional poverty estimates obtained was no suitable survey in a period, the most from the S2S lead to two additional obser- recent available estimate in the preceding or vations. First, the S2S imputation approach subsequent period was used.11 If neither was predicts lower poverty rates throughout the available, a regional average poverty rate period. Second, discrepancies between the from countries with available imputations poverty rates estimated using S2S, the rates was assigned.12 based on the PovcalNet and comparable Figure 2.4 shows population-weighted and quality-corrected data are largest in the average poverty rates for each period. Because late-1990s; they narrow in the 2000s. The there is only one data point per country in S2S results hint at the possibility that the each period for the S2S, a period’s point esti- results from both PovcalNet and the com- mate is assumed to be the average for a coun- parable and quality-corrected data provide try in that period. These averages were then a distorted picture of the extent of poverty used to obtain an average regional estimate reduction in the region—PovcalNet because for the period. These estimates are compared it fails to account for the noncomparability with the regional estimates for the largest 23 and poor quality of surveys and the corrected 64 POVERTY IN A RISING AFRICA 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 in ation rate (percent) Annual in ation 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 a 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 a Ni hi M a M in m in m rk rk Bu Bu M M Food CPI Overall CPI 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. Di erence 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 BOX 2.3 Can wealth indexes be used to measure changes in poverty? 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 Differences in Poverty Reduction by –7.1 Landlocked Country Typologies What distinguishes countries that have suc- ceeded in reducing poverty from countries –12.6*** Resource rich that have not? To answer this question, this section uses corrected data for all African countries and a classification of countries Fragile 15.1*** along four dimensions: fragility, resource richness, landlockedness, and low national –15 –10 –5 0 5 10 15 20 income. It first examines simple changes Change in poverty rate (percentage points) in poverty rates between 1996 and 2012 compared to alternative category for each country type. It then examines the relationship between country type and Source: World Bank Africa Poverty database. changes in poverty conditional on the other Note: Figure shows results of a regression of the change in the poverty rate on country character- istics. Based on estimated poverty rates for 43 countries (1996–2012) using comparable and good- classifications, using a simple regression quality surveys. specification. *** Statistically significant at the 1% level. Fragility. The results show that poverty fell even in fragile states, albeit by less than on average resource-rich countries reduced in nonfragile states. Between 1996 and 2012, poverty by about 13 percentage points more the poverty rate in fragile countries declined than non-resource-rich countries. However, from 65 percent to 53 percent (a 12 percent- a number of surveys were dropped from the age point change). This decline was much set of resource-rich countries because of lack more modest than the 24 percentage point of comparability and quality, increasing reli- drop in nonfragile economies (from 56 per- ance on GDP for imputations. To the extent cent to 32 percent). Some fragile countries are that GDP tracks with consumption surveys resource rich, landlocked, or both. Therefore, less well in resource-rich countries, the rate a simple binary comparison between fragile of poverty reduction will be overestimated. and nonfragile countries is unlikely to cap- Empirical evidence on the latter is mixed. ture the contribution to poverty reduction of For Zambia imputations relying on GDP fragility alone. Conditional on the three other indicate more rapid poverty reduction, traits (resource richness, landlockedness, and whereas S2S imputations show an increase income), poverty reduction for fragile coun- in poverty. In Nigeria both methods pre- tries was lower than for nonfragile countries dict roughly the same magnitude of poverty by 15 percentage points, and the difference reduction. The main driver of the difference was statistically significant (figure 2.6). in poverty reduction between resource-rich Resource richness. Resource-rich coun- and resource-poor countries is corrections to tries experience more poverty reduction the Nigeria data. Nigeria’s population share than non-resource-rich countries: the pov- among resource-rich countries (44 percent) erty rate fell 26 percentage points (from 62 is even larger than for the region as a whole percent to 36 percent), compared with 18 (18 percent). Before corrections for compara- percentage points (from 55 percent to 37 bility and quality, Nigeria’s surveys showed percent) in non-resource-rich economies. slow poverty reduction, despite relatively Conditional on the other characteristics, high GDP growth for more than a decade. 70 POVERTY IN A RISING AFRICA Poverty levels were higher in Nigeria than in Low-income status . Middle-income many countries at much lower income lev- countries reduced poverty by 26 percent- els in Africa and the rest of the world. This age points—7 percentage points more than stagnant poverty rate has been considered an low-income countries. Conditional on other artifact of poor-quality data (World Bank traits, however, they did not perform bet- 2014c). With the corrected data, Nigeria’s ter than low-income countries (1 percentage poverty rates are much lower (and closer to point difference). countries in its income group) and the decline in poverty steeper, changing the performance Differences in Poverty Reduction by of resource-rich countries (see box 2.1). Setting and Gender Landlockeness. Some researchers have posited that landlocked countries perform Although Africa is urbanizing heavily, its worse than coastal countries because trans- population remains predominantly rural: in port costs impede trade and lower com- the majority of countries, 65–70 percent of petitiveness (Bloom and Sachs 1998; Luke, the population resides in rural areas (World Sachs, and Mellinger 1999). The results Bank 2015a). Rural residents have higher presented here provide no support for this poverty rates across countries (figure 2.7). hypothesis. Landlocked countries reduced The corrected data for all countries reveal poverty by 24 percentage points (from 65 that both urban and rural populations expe- percent to 41 percent)—3 percentage points rienced declines in poverty between 1996 more than coastal countries, where poverty and 2012. Urban poverty rates dropped 16 fell from 56 percent to 35 percent. When percentage points (a 48 percent decline), and resource richness, fragility, and income sta- rural poverty rates fell 23 percentage points tus are controlled for, the difference in favor (a 33 percent decline). The gap in the pov- of landlocked countries widens to 7 percent- erty rate between urban and rural areas also age points, but this difference remains sta- declined (from 35 percentage points to 28 tistically insignificant. percentage points). Among the four geographic regions, three of four (East, Southern, and West) have FIGURE 2.7 Urban poverty in Southern and West Africa fell by halved (or almost halved) poverty. No rural almost half between 1996 and 2012 areas halved poverty. Rural populations in West and Southern Africa experienced declines in poverty of about 40 percent. East Africa 1996 Africa is distinguished by the large share 2012 of female-headed households (26 percent of West Africa 1996 all households and 20 percent of all people). 2012 Among these households, 62 percent contain 1996 Central Africa 2012 no adult men (15 or older). These statistics hide large variations across 1996 Southern Africa 2012 countries and regions in Africa (Milazzo and van de Walle 2015). Southern Africa has the 1996 All Africa 2012 highest rate of female-headed households (43 percent). West Africa exhibits the low- 0 20 40 60 80 est incidence: one household in five is headed Poverty headcount (percent) by a woman, and female-headed households Urban Rural account for 15 percent of the population. The relatively low rate in West Africa reflects Source: World Bank Africa Poverty database. Estimates based on data corrected for comparability and quality. both polygamy and high remarriage rates Note: Data are population weighted. among widows. Except in Southern Africa, REVISITING POVERTY TRENDS 71 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 a. East Africa b. Central Africa 60 60 Share of households headed Share of households headed by women (perccent) 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 c. Southern Africa d. West Africa 60 60 Share of households headed Share of households headed by women (perccent) 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 Source: Milazzo and van de Walle 2015. Note: Estimates are from several rounds of Demographic and Health Surveys. Earliest refers to first survey, latest to last survey. East Africa includes Comoros, Ethiopia, Kenya, Madagascar, Malawi, Mozambique, Rwanda, Tanzania, Uganda, and Zambia. Central Africa includes Cameroon, Chad, the Republic of Congo, and Gabon. Southern Africa includes Lesotho, Namibia, and Zimbabwe. West Africa includes Benin, Burkina Faso, Côte d’Ivoire, Ghana, Guinea, Mali, Niger, Nigeria, and Senegal. Zimbabwe is classified here as Southern (instead of East) Africa in order to create large enough country samples for each subregion. 72 POVERTY IN A RISING AFRICA 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 South Africa 2008–12 29 28 43 Nigeria 2010/11–2012/13 27 20 53 Malawi 2010/11–2013/14 23 32 44 South Africa Kwazulu-Natal 1993–2004 26.6 45.5 27.9 South Africa Kwazulu-Natal 1993–2004 22.8 42.3 34.9 South Africa Free State non-HIV 2000–01 10 35 55 South Africa Free State-HIV 2000–01 20 31 49 Lesotho 1993–2002 26 42 32 Kenya Tegemeo 1997–2007 12.94 70.8 16.23 Kenya Tegemeo 1997–2000 37.2 30.1 32.7 Zimbabwe 1993–96 10.6 59.6 29.8 Ethiopia 2006–2009 26.6 25.24 48 Madagascar 1997–1999 64.9 26 9.1 Non-nationally Representative Uganda 1992–99 18.9 39.4 41.7 South Africa Kwazulu-Natal 1993–98 22.7 31.5 45.8 South Africa Kwazulu-Natal 1993–98 62.2 23.3 14.5 South Africa Kwazulu-Natal 1993–98 18 34 48 South Africa Gauteng 1997–2001 35.9 58.3 5.8 Kenya 1993–95 70.8 22.5 6.8 Ethiopia rural 1994–2004 20 43 37 Ethiopia rural 1994–2004 6 79 15 Ethiopia rural 1994–97 13 46 41 Ethiopia rural 1994–97 7 63 30 Ethiopia urban 1994–2009 8.49 29.04 62.47 Ethiopia urban 1994–97 21.5 36.2 42.2 Ethiopia urban 1994–95 24.8 30.1 45.1 Côte d’Ivoire 1987–88 25 22 53 Côte d’Ivoire 1986–87 13 22.9 64.1 Côte d’Ivoire 1985–86 14.5 20.2 65.3 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 Source: Dang and Dabalen 2015. REVISITING POVERTY TRENDS 75 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- Poverty headcount (percent) ponents: households that were upwardly mobile (poor in the first period but not poor 60 in the second period) and households that were never poor (nonpoor in both periods). Figure 2.10 reveals three aspects of pov- 50 erty dynamics in Africa. First, on average about 35 percent of the population of a 40 country is chronically poor. These people account for 58 percent of the poor. About 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 Deaton, Angus. 2005. “Measuring Poverty in a in Vietnam.” University of Waikato, New Growing World (or Measuring Growth in a Zealand. Poor World).” The Review of Economics and Gibson, John, Steven Stillman, and Trinh Le. Statistics LXXXVII (1): 1–19. 2008. “CPI Bias and Real Living Standards de Carvalho Filho, Irineu, and Marcos Chamon. in Russia during the Transition.” Journal of 2012. “The Myth of Post-Reform Income Development Economics 87 (1): 140–61. Stagnation: Evidence from Brazil and Mex- Glewwe, Paul. 2012. “How Much of Observed ico.” Journal of Development Economics 97 Economic Mobility Is Measurement Error? IV (2): 368–86. Methods to Reduce Measurement Error Bias, Dercon, Stefan, and Pramila Krishnan. 2000. with an Application to Vietnam.” World Bank “Vulnerability, Seasonality and Poverty in Economic Review 26 (2): 236–64. Ethiopia.” Journal of Development Studies 36 Gwatkin, Davidson R., Shea Rutstein, Kiersten (6): 25–53. Johnson, Rohini Pande, and Adam Wagstaff. Diewert, W. Erwin. 1998. “Index Number Issues 2000. “Socio-Economic Differences in Health, in the Consumer Price Index.” Journal of Eco- Nutrition, and Population.” World Bank, nomic Perspectives 12 (1): 47–58. HNP/Poverty Thematic Group, Washington, Douidich, Mohamed, Abdeljaouad Ezzrari, Roy DC. Van der Weide, and Paolo Verme. 2013. “Esti- Gwatkin, Davidson, Shea Rutstein, Kiersten mating Quarterly Poverty Rates Using Labor Johnson, Eldaw Suliman, Adam Wagstaff, Force Surveys: A Primer.” Policy Research and Agbessi Amouzou. 2007. Socio-Economic Working Paper 6466, World Bank, Washing- Differences in Health, Nutrition, and Popula- ton, DC. tion. Washington, DC: World Bank. Duponchel, Marguerite, Andy McKay, and Sarah Hamilton, Bruce W. 2001. “Using Engel’s Law Ssewanyana. 2014. “The Dynamics of Poverty to Estimate CPI Bias.” American Economic in Uganda, 2005/6 to 2011/12: Has the Prog- Review 91 (3): 619–30. ress Stalled?” Working Paper, Centre for the Harttgen, Kenneth, Stephan Klasen, and Sebas- Study of African Economies (CSAE), Oxford, tian Vollmer. 2013. “An African Growth Mir- United Kingdom. acle? Or: What Do Asset Indices Tell Us about Filmer, Deon, and Lant Pritchett. 1999. “The Trends in Economic Performance?” Review of Effect of Household Wealth on Educational Income and Wealth 59 (S1): S37–S61. Attainment: Evidence from 35 Countries.” Hausman, Jerry. 2003. “Sources of Bias and Solu- Population and Development Review 25 (1): tions to Bias in the CPI.” Journal of Economic 85–120. Perspectives 17 (1): 23–44. ———. 2001. “Estimating Wealth Effects with- Hobijn, Bart and David Lagakos. 2005. “Infla- out Expenditure Data or Tears: An Applica- tion Inequality in the United States,” Review tion to Educational Enrollments in States of of Income and Wealth, 2005, 51 (4), 581–606. India.” Demography 38 (1): 115–32. Horrell, S., and Krishnan P. 2007. “Poverty and Filmer, Deon, and Kinnon Scott. 2012. “Assessing Productivity in Female-Headed Households in Asset Indices.” Demography 49 (1): 359–92. Zimbabwe.” Journal of Development Studies Finn, Arden, and Murray Leibbrandt. 2013. “The 43 (8): 1351–80. Dynamics of Poverty in the First Three Waves Howe, Laura, James R. Jargreaves, Sabrine of NIDS.” NIDS Discussion Paper 2013/1, Gabrysch, and Sharon Huttly. 2009. “Is National Income Dynamics Study, Southern the Wealth Index a Proxy for Consumption Africa Labour and Development Research Expenditure? A Systematic Review.” Journal Unit (SALDRU), University of Cape Town, of Epidemiology and Community Health 63 Rondebosch, South Africa. (11): 871–80. Gaddis, Isis. 2015. “Prices for Poverty Analysis Jerven, Morten. 2013. Poor Numbers: How We in Africa.” Background paper prepared for the Are Misled by African Development Statistics Poverty in a Rising Africa report, World Bank, and What to Do about It. Ithaca, NY: Cornell Washington, DC. University Press. Gibson, John, Trinh Le, and Bonggeun Kim. Lanjouw, Peter, and Martin Ravallion. 1995. 2014. “Prices, Engel Curves and Time-Space “Poverty and Household Size.” Economic Deflation: Impacts on Poverty and Inequality Journal 105 (433): 1415–34. REVISITING POVERTY TRENDS 81 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. BOX 3.1 How useful are subjective data in monitoring 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 Levels of and Trends in points, despite a rapid increase in primary enrollment rates since 2000. This change Well-Being compares unfavorably with the 17 percent- age point increase in South Asia and in the Education and Literacy Middle East and North Africa and the 10 Education can expand people’s capabilities. percentage point increase in East Asia and It helps people access and digest informa- Pacific, where the literacy rate is approaching tion and knowledge. Doing so requires at a 93 percent. minimum being literate. The focus here is This low average for Africa masks sub- mainly on adult literacy rates: the percentage stantial variation within the region. More of adults who can, with understanding, read than half the population is illiterate in and write a short, simple statement about seven countries, almost all of them in West their everyday lives. Africa (figure 3.2). Niger (with an adult lit- Adult literacy rates evolve slowly in the eracy rate of only 15 percent) and Guinea absence of effective large-scale adult educa- (where the rate is just 25 percent) have the tion programs, because their evolution is lowest literacy levels in Africa. At the other influenced mainly by the literacy levels of extreme, literacy levels exceed 90 percent in younger cohorts. Current school enrollment Equatorial Guinea and South Africa, and rates and test scores are therefore also con- they exceed 70 percent in some poor and sidered, to assess how adult literacy is likely fragile countries as well, such as Eritrea and to evolve. Zimbabwe. Africa’s literacy rate stood at 58 percent in The gender literacy gap remains high, aver- 2012: more than two in five Africans cannot aging about 25 percentage points, although read or write a sentence (figure 3.1 and box gender parity in education is one of the Mil- 3.2). There has been progress, but it has been lennium Development Goals on which Africa slow. Between 1995 and 2012, the literacy has performed best. Gender parity in literacy level in the region increased by 4 percentage is especially low in West Africa (figure 3.3). FIGURE 3.1 Africa’s literacy rate is the lowest in the world 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 Paci c Middle East and North Africa Sub-Saharan Africa Latin America and the Caribbean South Asia Source: EDSTAT data. Note: The adult literacy rate is the percent of the population 15 and older that can, with understanding, read and write a short, simple statement about their everyday lives. Missing years were inter- or extrapolated. POVERT Y FROM A NONMONE TARY PERSPEC TIVE 87 BOX 3.2 Tracking adult literacy with data remains challenging 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 Ratio of female to male literacy rate 100 1.6 90 1.4 Male adult literacy rate (%) 80 ࡗ 1.2 70 ࡗ 60 ࡗࡗࡗࡗࡗ 1.0 ࡗࡗࡗࡗ ࡗࡗࡗࡗࡗࡗࡗࡗ 50 ࡗࡗ ࡗ 0.8 ࡗࡗࡗࡗࡗࡗ 40 ࡗࡗࡗࡗࡗࡗ ࡗࡗࡗࡗࡗࡗ 0.6 30 ࡗ ࡗࡗ ࡗࡗ 0.4 20 ࡗ 10 0.2 0 0 S e am o ua Mh A nd 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 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 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 n g h an Pr tre bo bw 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 b be d Co Ver ad ab u t zila o 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 ࡗ Ratio of female to male literacy rate Source: EDSTAT data. Africa average is population-weighted. FIGURE 3.4 Illiteracy is higher among poorer people, older people, rural dwellers, and people in resource-rich and landlocked countries Landlocked compared with coastal 2.8 Low-middle income compared with low income –13.8 Upper-middle and high income compared with low income –31.9 Fragile compared with nonfragile –8.4 Resource-rich compared with resource-poor 3.1 Urban compared with rural –17.1 Poor compared with nonpoor 18.4 Divorced, widowed, no longer living together, never married –20.4 compared with married Female-headed household compared with male-headed household Change per additional household member 0.58 Age group compared with age group 15–19 45–49 18.8 40–44 12.8 35–39 11.5 30–34 9.2 25–29 6.8 20–24 2.2 –40 –30 –20 –10 0 10 20 30 Percentage point di erence in rate of illiteracy among women Source: Data from Demographic and Health Surveys 2005–13. Note: Results are from ordinary least square regression. All estimated coefficients are statistically significant. 90 POVERTY IN A RISING AFRICA 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 FIGURE 3.7 Healthy life expectancy at birth ranges widely 70 60 50 40 Years 30 20 10 0 Libana ut za a Af a T ica Co B ogo rk , R n aF . 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 Gh on Er ana M mo ea ri s e ud ia G a an Bo Ken ia ts y a Ga nda Le ic e in ep nd hi a Se asca an o iu au ro Gufric az eri o, m ali So Tan eri h ni 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 b e a a Co itr m n o b r é b it b .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 Source: WHO 2015. Africa average is population-weighted. 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) People in upper-middle-income and high- 6 income countries can expect to live in good health 6.5 years longer than people in low- 4 income countries, after controlling for the 2.5 other country traits. People in coastal coun- 2 tries also have higher healthy life expectancy. Under-five mortality and HIV prevalence. 0 Two mortality indicators are significant driv- ers of changes in life expectancy in Africa: –2 –1.6 under-five mortality rates and HIV preva- lence rates. For every 10 additional children –4 –3.8 per 1,000 live births surviving to the age of –4.5 five, life expectancy increased by 0.7 years; –6 for every percentage point increase in HIV Resource- Upper- Lower- Fragile Landlocked rich middle middle states prevalence, life expectancy decreased by 1 income income year. These two factors alone explain more than three-quarters of the variation in life Source: Data from WHO 2015. expectancy in the region (under-five mortal- Note: Results are from ordinary least squares regression of data for 2000–12 ity explains 50 percentage points and HIV including these four country traits. 92 POVERTY IN A RISING AFRICA FIGURE 3.9 Vaccination rates rose and child mortality from malaria fell a. Under- ve mortality (1,000s deaths) b. Under- ve mortality from malaria, 2000–13 and immunization (%), 2000–13 500 100 800 450 90 700 400 80 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 Pertussis 0–4 years DTP3 immunization 20 Tetanus 0–4 years Measles immunization Measles 1–59 months 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. prevalence explains 28 percentage points). partly as a result of the expanded use of Country gross domestic product (GDP) lev- insecticide-treated bednets. 6 Many more els and the number of deaths from conflict in children still die annually from malaria previous years do not have important effects than from measles, tetanus, and pertussis on life expectancy beyond their effects on together, however (figure 3.9). The risk of a child mortality or HIV prevalence.5 child under five dying from malaria is low The decline in the under-five mortality in Southern Africa (excluding Malawi and rate—from 173 in 1995 to 92 in 2013—went Zambia), partly because of climatic condi- hand-in-hand with the increase in immuniza- tions. It exceeds 20 deaths per 1,000 live tion rates and the decline in the incidence of births in Angola (21), Nigeria (24), Guinea malaria-related deaths (figure 3.9). Room for and Sierra Leone (27), Chad (28), and the further decline through expansion of immu- Central African Republic (35). nization remains—the immunization rate The second-most important disease hold- against measles is still only about 60 percent ing back Africa’s life expectancy is HIV/ in some of the region’s most populous (Ethio- AIDS. In 2012, 1.1 million people in the pia) and resource-rich (the Democratic Repub- region died of AIDS—almost four times as lic of Congo, Nigeria, South Africa) countries. many as in the rest of world combined (about In Equatorial Guinea, where GDP per capita 300,000). The continent’s HIV prevalence is more than $15,000 a year, only about one rate peaked at 5.8 percent in 2002, declining child in two is vaccinated against measles. to 4.5 percent in 2013 (World Development In 2000 less than $100 million was dis- Indicators). bursed to malaria-endemic countries to fight Southern Africa has been especially hard malaria; in 2013 the figure was $1.97 bil- hit by HIV/AIDS. At least 10 percent of 15- lion. As a consequence, the number of chil- to 49-year-olds there are HIV-positive (10.3 dren dying from malaria fell dramatically, percent in Malawi, 10.8 in Mozambique, POVERT Y FROM A NONMONE TARY PERSPEC TIVE 93 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 MAP 3.1 HIV prevalence remains very high in Southern Africa 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 a. Factors associated with underweight b. Factors associated with obesity –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 Fragile compared Fragile compared with nonfragile 2.0 with nonfragile 0.1 Resource-rich compared Resource-rich compared with resource-poor 3.7 with resource-poor –2.9 Urban compared with rural –1.6 Urban compared with rural 4.1 Poor compared with nonpoor 3.2 Poor compared with nonpoor –4.6 Divorced, widowed, no longer Divorced, widowed, no longer living together, never married 2.7 living together, never married –3.0 compared with married compared with married Female-headed household Female-headed household compared with male- –1.2 compared with male- 0.9 headed household headed household Education compared Education compared with no education with no education –1.9 3.6 Attended primary education Attended primary education Attended secondary education –4.1 Attended secondary education 5.0 Attended higher education –5.3 Attended higher education 8.4 –8 –6 –4 –2 0 2 4 6 8 10 –8 –6 –4 –2 0 2 4 6 8 10 Percentage point di erence Percentage point di erence Source: Data from Demographic and Health Surveys 2005–13. Note: Results are from ordinary least squares regression of an indicator variable of an adult woman being underweight (1 if the body mass index is less than 18.5, 0 otherwise) or over- weight (1 if the body mass index is more than 30, 0 otherwise). Sample includes nonpregnant women who did not give birth in the three months before the interview. All estimated coefficients (except on Fragile) are statistically significant. 96 POVERTY IN A RISING AFRICA 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 o a a i e aw ny an bi bw as tional conflicts and civil wars of the 1990s rit m aF al Ke Gh ba au Za M in (in Angola, Liberia, Mozambique, Rwanda, M m rk Zi Bu and Sierra Leone) have receded in scale Source: Mitra, Posärac, and Vick 2013, based on data from World Health and intensity, but election-related violence, Surveys. Note: Disability is defined as having severe difficulty moving about, con- extremism, terror attacks, drug trafficking, centrating, remembering, seeing or recognizing people across the road maritime piracy, and criminality have been (while wearing glasses), or taking care of oneself. growing. Wars are increasingly being fought by armed insurgents on the periphery of fac- to 13.0 percent in Malawi. The numbers are tionalized and militarily weak states, such as higher among women (10.6 percent) than the Arab and Tuareg uprisings in Mali and men (7.3 percent). They are also higher in Boko Haram in Nigeria. West Africa has rural areas (9.9 percent) than urban areas emerged as a key transit point in the traffick- (6.9 percent). Disability prevalence rates in ing of narcotics between Latin America and Africa are similar to the average rates in the Europe, and piracy has expanded in the Gulf Asian and Latin American countries exam- of Guinea. ined by Mitra, Posärac, and Vick (2013). In addition to undermining the basic func- tioning of being secure, conflict also affects many of the other functionings and opportu- Freedom from Violence nities that are critical to self-determination. It The ability to live free from violence affects affects not only the people directly affected people’s survival, dignity, and daily life. Inse- but also the broader population inside and curity significantly reduces the choices a per- outside the country (by, for example, creat- son can make regarding what to do and who ing internally displaced persons and refugees to be (capabilities). [box 3.3]). Countries suffering more than Afrobarometer data from 2010–12 indi- 100 casualties in a year experience a decline cate that insecurity is pervasive in Africa. in economic growth of 2.3 percent. These In these surveys, 12 percent of respondents effects can be long-lasting. Annual economic indicate that either they or a family member growth in Burundi has hovered around 4 had been physically attacked at least once percent since the civil war ended in the early during the past year. Fifty-three percent indi- 2000s. But panel data indicate that the share cated that they feared political intimidation of households that reported being (mon- or violence at least once during election cam- etarily) poor rose from 21 percent in 1993 paigns; 40 percent indicated that they or a (before the civil war) to 46 percent in 1998 family member had felt unsafe at least once (during civil war) and 64 percent in 2007 while walking in the neighborhood during (several years after the end of the civil war) POVERT Y FROM A NONMONE TARY PERSPEC TIVE 97 MAP 3.2 The number of violent events against civilians is increasing, especially in Central Africa and the Horn a. 1997–99 b. 2009–11 c. 2014 50–300 (6) 50–400 (6) 50– 650 (9) 10–50 (12) 10–50 (9) 10–50 (14) 0–10 (25) 0–10 (28) 0–10 (20) 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 BOX 3.3 What happens to Africans who flee their homes? 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 a. Change in under-five mortality b. Change in life expectancy and mean fatalities per year and mean fatalities per year Average annual number of fatalities Average annual number of fatalities 600 1,000 800 400 600 400 200 200 0 0 −150 −100 −50 0 0 5 10 15 20 Change in under-five mortality 2000−10 Change in life expectancy 2000−12 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% 2000–06 2007–13 2000–06 2007–13 Incidence of domestic violence Other developing countries Acceptance of domestic violence Africa 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 Source: Data from Demographic and Health Surveys 2007–13. 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 a. Acceptance of domestic violence b. Incidence of domestic violence Upper-middle and high income Upper-middle and high income –7.6 compared with low income 7.2 compared with low income Low-middle income Low-middle income –10 –1 compared with low income compared with low income Landlocked compared with coastal 6.1 Landlocked compared with coastal –4.1 Resource-rich compared Resource-rich compared with resource-poor 16 3.6 with resource-poor Fragile compared with nonfragile 9.2 Fragile compared with nonfragile 5.2 Annual trend –1.7 Annual trend –0.6 Change per additional children born 0.9 Change per additional children born 1.4 No longer living together or separated –1.7 No longer living together or compared with never married 22.3 separated compared with married Divorced compared with never married –0.3 Divorced compared with married 25 Widowed compared with never married –1.2 Widowed compared with married 3.7 Living together compared with never married 0.4 Living together compared with married 9.1 Married compared with never married –1.6 Currently working compared 5.3 Currently working compared with not working with not working 0.2 Education compared with no education Education compared with no education Attended higher education 1.5 Attended higher education –31.2 Attended secondary education 10.4 Attended secondary education –15.8 Attended primary education 14.6 Attended primary education –8.4 Urban compared with rural 1.2 Urban compared with rural –1 Poverty quintiles compared Poverty quintiles with poorest quintile compared with poorest quintile Richest –7.1 Richest –3.9 Richer –1.6 Richer –1.3 Middle 0.6 Middle –0.3 Poorer Poorer –0.5 0.6 Age group compared Age group compared with age group 15–19 with age group 15–19 45–49 –13.7 45–49 –3.1 40–44 –12.1 40–44 –1.3 35–39 –10.8 35–39 –0.1 30–34 –9.1 30–34 2.6 25–29 –7 25–29 4.6 20–24 –3.3 20–24 4.5 –40 –30 –20 –10 0 10 20 –15 –10 –5 0 5 10 15 20 25 30 Percentage point di erence Percentage point di erence Source: Data from Demographic and Health Surveys 2000–13. Note: Results are from ordinary least squares regressions. All estimated coefficients are statistically significant except coefficients on divorced in acceptance of domestic violence and age 35–39 and poorer and middle-income quintiles in incidence of domestic violence. 102 POVERTY IN A RISING AFRICA 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 1.5 Almost 40 percent of Africans do not reg- 1.0 ularly listen to the radio, watch television, or 0.5 read a newspaper at least once a week (fig- 0 ure 3.19). Exposure to the media is lower in –0.5 Africa than in the rest of the developing world (excluding China), where only 25 percent of –1.0 the population lacks regular media exposure. –1.5 African countries with high media exposure –2,0 (more than 80 percent of the population –2.5 exposed) include Gabon, Ghana, and Kenya. Media exposure is typically lower around 96 98 00 02 03 04 05 06 07 08 09 10 11 12 13 19 19 20 20 20 20 20 20 20 20 20 20 20 20 20 the Sahel, in many of the coastal countries of East Asia and Paci c North America West Africa, and in Africa’s populous coun- Europe and Central Asia South Asia tries (the Democratic Republic of Congo and Latin America and the Caribbean Sub-Saharan Africa Middle East and North Africa Ethiopia), where only about 40 percent of the population have regular access to the media. Source: Worldwide Governance Indicators. There is also an important gender gap. On POVERT Y FROM A NONMONE TARY PERSPEC TIVE 103 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 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 ria Su lic ui n hi ia n .R u E r ea í n Se en ho an a aF e o er N ali Le er Co ibe e m ria an s 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. zil na d Be d s e rk biq al 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 S Z an to Si ua é m Eq To o Sã Source: Worldwide Governance Indicators 2013. Africa average is population-weighted. 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 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 í n 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 a e G o, éa M ve fri m lA De To ra o nt Sã Ce Women Men Source: Latest available Demographic and Health Surveys 1994–2013. Note: Exposure to mass media means a person listened to the radio, watched TV, or read the newspaper at least once a week. Developing world excludes Africa and is population-weighted. Africa average is population-weighted. 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 Landlocked compared with coastal 14.9 Low-middle income compared with low income 15.9 Upper-middle and high income –5.2 compared with low income Fragile compared with nonfragile –7.4 Resource-rich compared with resource-poor 7.2 Urban compared with rural –6.9 Poor compared with nonpoor 5.0 Female-headed household compared –10.6 with male-headed household Change per additional household member 1.3 Age group compared with age group 15–19 45–49 –20.3 40–44 –20.0 35–39 –18.2 30–34 –15.6 25–29 –12.5 20–24 –7.6 –25 –20 –15 –10 –5 0 5 10 15 20 Percentage point di erence Source: Data from Demographic and Health Surveys 2005–13. Note: Results are from ordinary least squares regression. All estimated coefficients are statistically significant. 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 To measure deprivation in multiple dimensions, we Freedom to decide is measured by an indicator use data from the Demographic and Health Surveys capturing lack of frequent media exposure (not using (DHS) on 25 countries, covering 72 percent of the at least one media channel [newspaper, television, population of Africa. We focus on the four areas of radio] at least once a week) or not being involved in deprivation discussed earlier. Illiteracy is defined as decisions regarding own health care, family visits, being unable to read a full sentence, being blind, or or spending. Both indicators are correlated with having no reading card for the required language. the Worldwide Governance Indicator of voice and More than half (56 percent) of women in the sample accountability (correlation coefficient is 0.4). countries are illiterate. For comparison, we augment these dimensions by Women are classified as deprived in health if they adding a fifth aspect: asset poverty. We use the DHS are undernourished (BMI below 18.5). There is no asset index to classify women as asset poor or nonpoor direct information on life expectancy. The correla- (Christiaensen and Stifel 2007; Filmer and Scott 2012; tion coefficient between country life expectancy and Sahn and Stifel 2000 establish correlations with con- the proportion of undernourished women is 0.3. sumption). Country cutoffs are defined based on the Women’s attitudes toward domestic violence are share of the population living below $1.90 for the cor- used as an indicator of physical security. Across responding survey year. The correlation of this indica- countries, social norms toward domestic violence tor with the other dimensions is 0.33, underscoring and the incidence of casualties from political violence the fact that asset wealth does not capture deprivation are correlated (the correlation coefficient is 0.4). in many basic functionings and capabilities. 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 Landlocked compared with coastal 0.3 Upper-middle and high income –0.8 compared with low income Low-middle income compared with low income –0.4 Fragile compared with nonfragile 0.2 Resource-rich compared with resource-poor 0.4 Urban compared with rural –0.5 Poor compared with nonpoor 0.6 Divorced, widowed, no longer living together, 0.6 never married compared with married Female-headed household compared –0.1 with male-headed household Change per additional household member 0.03 Age group compared with age group 15–19 45–49 –0.1 40–44 –0.2 35–39 –0.2 30–34 –0.2 25–29 –0.2 20–24 –0.1 –1 –0.8 –0.6 –0.4 –0.2 0 –0.2 –0.4 0.6 0.8 Number of deprivations Source: Data from Demographic and Health Surveys 2005–13. Note: Results are from ordinary least squares regression on the number of deprivations out of a total of four deprivations. All estimated coefficients except the annual trend are statistically significant. 108 POVERTY IN A RISING AFRICA 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 Source: Data from Demographic and Health Surveys 2005–13. 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 BOX 3.5 What is the multidimensional poverty index (MPI)? 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 n 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 r E t ea ia Gu i te nd al ge o an Ca ani bi d Sw cip on bo ni De as n r op ib er R e Co urk Ivoi or o z K en in ge M an e t ila o 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 Not deprived Deprived in three dimensions Deprived in one dimension Deprived in four dimensions Deprived in two dimensions Source: Data from Demographic and Health Surveys 2005–13. Note: Countries are ranked by the share of the female adult population deprived in at least three out of four dimensions. 110 POVERTY IN A RISING AFRICA 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 and it may affect growth prospects. With Perceptions of Inequality respect to poverty reduction, when ini- tial inequality is higher, a larger share of Several survey efforts capture the perceptions poor households will have incomes farther and attitudes of citizens toward inequality. below the poverty line, so that growth (the The picture that emerges is not clear, in part increase in income) will result in less pov- because the survey questions differ.3 erty reduction (Bourguignon 2004; Klasen The World Values Survey asks respon- 2004; Ravallion 2001). Tentative evidence dents if more or less inequality is needed in also suggests that inequality leads to lower their country. Its results reveal polarization: and less durable sustainable growth pro- in some countries, more than 20 percent of cesses and thus less poverty reduction (Berg, respondents indicate that more inequality Ostry, and Zettelmeyer 2012; OECD 2015) is needed and more than 20 percent indi- if, for example, wealth is used to engage in cate that less inequality is needed. Figure rent seeking and other distortionary eco- 4.1 shows the results for four countries; the nomic behaviors (Stiglitz 2012). The path- results are similar for the seven other African way by which inequality evolves thus matters countries covered by the World Values Sur- for growth. Marrero and Rodriguez (2013) vey, and the pattern does not change mark- find a robust negative relationship between edly between country survey rounds. These growth and inequality of opportunity in the results are consistent with the point made in United States. Ferreira and others (2014) find World Development Report 2006: Equity suggestive evidence of a negative association and Development (World Bank 2005) that, between inequality and growth but conclude contrary to preconceived notions, citizens do that the data do not show a robust negative not by and large view inequality negatively. association between inequality of opportu- The share of the population in the African nity and growth. Other studies conclude that countries that indicated that income should as countries reach higher levels of develop- be more equal was just 21 percent—lower ment, greater emphasis should be given to than the 28 percent for all countries included reducing inequality over spurring growth in the World Value Survey. to reduce poverty (Olinto, Lara Ibarra, Afrobarometer surveys find that among and Saavedra-Chanduvi 2014). For high- a list of more than 30 possible responses, inequality, low-income countries, then, there respondents rarely cite inequality as one of is tension between a focus on growth and an the most important problems facing their emphasis on addressing inequality countries. In these surveys, poverty and On the basis of the growing body of lit- employment are the primary concerns of erature on the effects of initial and chang- respondents in most countries. In the major- ing inequality on growth and poverty, some ity of the 30 African countries surveyed in observers argue that reducing inequality the Gallup Poll (2013), most respondents should be an explicit development goal (Shep- report that individuals can get ahead by herd and others 2014). For the few African working hard. countries for which there is evidence, this In contrast, in the Pew Global Attitudes notion seems to resonate with policy mak- Survey, 70–81 percent of respondents in the ers. In its survey of 15 developing countries six African countries covered agreed that (in Africa, this included Cameroon, Malawi, inequality is a major problem in their coun- Nigeria, and South Africa), the United try (Pew Research Center 2013). Similarly, Nations Development Programme (UNDP Afrobarometer surveys show that most Afri- 2014) finds that 77 percent of policy mak- cans respond that their government is doing ers perceive the current level of inequality as quite or very badly at narrowing the income a threat to long-term national development. gap between the rich and poor. These sen- Only 10 percent consider inequality condu- timents do not correlate with the level of cive to long-term development. inequality in the country (figure 4.2). INEQUALITY IN AFRICA 119 FIGURE 4.1 Views on inequality differ within and across countries 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 Ni nya ri a N e Na en i Ke ar 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 i 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 Measurement of Inequality typically lower than other monetary inequality measures. Like the poverty analysis in chapter 2, the This chapter focuses on inequality as mea- analysis of inequality in this chapter is based sured by the Gini index (box 4.1) in consump- on data on consumption from nationally tion per capita, the same metric used to assess representative household surveys. With few poverty in chapter 2. Consumption is gener- exceptions, the factors that make measuring ally regarded as easier to measure than income poverty a challenge also complicate the mea- in low-income economies (Deaton and Zaidi surement of inequality.4 Changes in the ques- 2002). Current consumption generally does tionnaire or the seasonal timing of fieldwork not reveal the full extent of economic inequal- can distort apparent trends in inequality. To ity, however, because consumption does not prevent this problem, the analysis presented capture savings and wealth. in this chapter excludes surveys that are not Income and wealth inequality are alter- comparable (as defined in chapter 1).5 natives to consumption-based measures. In Surveys in Africa measure inequal- most economies, income-based measures ity based on consumption. Cross-regional of inequality are higher than consumption- comparisons typically ignore the differ- based measures (Blundell, Pistaferri, and ence between income and consumption Preston 2008; Krueger and others 2010; measures of inequality (as noted in the Santaeulàlia-Llopis and Zheng 2015), and next section), but it is an important dis- wealth inequality is typically higher than tinction because consumption inequality is income inequality (Davies and others 2011; BOX 4.1 A Primer on the Gini Index 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) tom X percent of the population accounts for X per- Cumulative share of cent of consumption (or earns X percent of income), Line of equality A and the Lorenz curve coincides with the diagonal. (45 degree) If there is some degree of inequality, the bottom X B percent of the population accounts for less than X percent of consumption. The Lorenz curve bows Lorenz curve outward; the farther it is from the diagonal line, of consumption the higher the degree of inequality. In the extreme (or income) case of perfect inequality, all consumption is con- 0 100% Cumulative share of the X centrated in the hands of the richest individual, and population, from poorest to richest the Lorenz curve coincides with the line from 0 to X to Y. The Gini index reflects the area between the line of perfect equality (the diagonal) and the Lorenz which is part of the generalized entropy class of curve (A), relative to the maximum area that would inequality indexes (Cowell 2000). As in the Gini be attained under perfect inequality (A + B). index, higher values of the MLD represent higher An alternative measure of inequality is the mean levels of inequality, but unlike the Gini index, the log deviation (MLD), also called Theil’s L index, MLD is not bound by 1. The MLD shows the per- (Box continues next page) INEQUALITY IN AFRICA 121 BOX 4.1 A Primer on the Gini Index (continued) 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). FIGURE B4.1.2 Different inequality measures reveal a similar story a. Rank of countries by Gini and b. Rank of countries by mean log derivation (MLD) Gini and Palma 50 50 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 Diaz-Gimenez, Glover, and Rios-Rull 2011; Magalhães and Santaeulàlia-Llopis (2015) Piketty 2014; Rama and others 2015; for compare inequality in consumption, income, Africa, see de Magalhães and Santaeulàlia- and wealth in Malawi, Tanzania, and Llopis 2015). Consumption and income are Uganda. Their measure of wealth includes flow measures that relate to a specific period land, housing, livestock, agricultural equip- (for example, one year); wealth is a stock ment, and household durable goods, net of measure that reflects assets accumulated over any debt (their data exclude housing in Tan- a lifetime (through savings) and across gen- zania and debt in Uganda). Because finan- erations (through bequests). cial assets are not included, total wealth Most household surveys in Africa lack is understated, particularly among urban detailed data on the value of household households. wealth. Taking advantage of the select Their results show the pattern observed in few that include partial wealth data, de other regions. In Malawi wealth inequality 122 POVERTY IN A RISING AFRICA 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 BOX 4.2 Can the Gini index be estimated without a survey? 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 MAP 4.1 Inequality in Africa shows a geographical pattern 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 Source: World Bank Africa Poverty database. 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 2 1 0 –1 –2 –3 –4 ag re ts a ng N ana M m. ia So bi . h e az a 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 Et o M ia i am p a 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 A hi a in rk er Bu Si Co Source: World Bank Africa Poverty database. Note: Annualized percentage change in the Gini index is based on the two most recent and comparable surveys available. 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 can inequality. Uganda 09-12 Botswana Dem. Rep. Congo Tanzania The African Gini index is 0.52– 0.56 –2 Uganda 02-05 across the benchmark years, much higher Burkina Faso Sierra Leone than individual-country inequality measures –4 (table 4.1). Only four countries (Botswana, Quadrant 4 Madagascar 01-05 the Central African Republic, Namibia, and South Africa) have Gini indexes that –10 –5 0 5 Annualized percentage change in poverty rate are higher than the African Gini in 2008. As discussed earlier, by and large, African Survey mean increased Survey mean decreased countries have levels of inequality that are Source: Countries in World Bank Africa Poverty database with comparable surveys. similar to other developing countries if mea- Note: Ethiopia 1995–99, an outlier, is excluded. Survey years are indicated for countries with more sured in terms of average country inequality. than one pair of comparable surveys. 128 POVERTY IN A RISING AFRICA TABLE 4.1 Inequality in Africa, 1993–2008 Benchmark year Percentage change Indicator 1993 1998 2003 2008 1993–2008 Gini index for Africa 0.52 0.52 0.54 0.56 8.6 Average country Gini index 0.47 0.45 0.45 0.45 3.8 Average country Gini index, population weighted 0.44 0.44 0.43 0.44 −0.5 African mean log deviation 0.47 0.47 0.51 0.57 20.0 Within-country contribution to African mean log deviation (percent) 73.4 71.3 64.3 59.7 Source: Jirasavetakul and Lakner 2015. 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 gaps across countries. These results stand 50 in sharp contrast to global inequality, where 40 within-country inequality increased both in 30 the level of inequality and as a share of total 20 inequality (even though between-country 10 differences remain the dominant source of 0 global inequality). % % Does country GDP explain African 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 a. Region b. Urban c. Education Percent inequality explained Percent inequality explained Percent of inequality explained 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 d. Demographics e. Employment f. Gender Percent of inequality explained Percent of inequality explained Percent of inequality explained 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 Source: World Bank Africa Poverty database. Note: Figure shows the percent of total inequality explained by differences in mean consumption between the groups indicated for 26 countries in Africa. For employment, the num- ber of countries included is 17 because of missing data on employment status or industry. Inequality is measured by the mean log deviation. Countries are sorted by the overall level of inequality, from low to high (left to right). 130 POVERTY IN A RISING AFRICA 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 Unequal Opportunities is the concept of achieved versus ascribed sta- tus (Linton 1936) and ascriptive inequality. It Inequality across households is the product can exacerbate overall inequality and violate of many forces. The circumstances in which principles of fairness and equal opportunity. one is born—in a rural area, to uneducated A growing body of literature in the past 15 parents—are one important force. Inequal- years tries to assess the degree of inequality ity of opportunity is the extent to which such of opportunity and evaluate the opportunity- circumstances dictate the outcomes of indi- equalizing effects of public policies (see the viduals in adulthood. In economics this con- recent surveys by Ferreira and Peragine 2015 cept has been articulated by Fleurbaey (2008) and Roemer and Trannoy 2015)—efforts and Roemer (2000), among others. In the that face a number of challenges (Kanbur field of sociology, inequality of opportunity and Wagstaff 2014). Building on the previous 132 POVERTY IN A RISING AFRICA discussion of horizontal inequalities, which The circumstances used to measure inequal- described the contribution of different indi- ity of economic opportunity include ethnic- vidual characteristics to total inequality, this ity, parental education and occupation, and section presents evidence on inequality of eco- region of birth. 21 The analysis focuses on nomic opportunity and the intergenerational individuals 15 years and older. Like other transmission of education and occupation.17 researchers in this field (see, for example, Fer- reira and Gignoux 2011), inequality is mea- sured using the MLD.22 Inequality of Economic Opportunity The share of inequality that can be attrib- The approach to measuring inequality of uted to inequality of opportunity ranges economic opportunity entails unpack- from 8 percent (Madagascar) to 20 percent ing how much of current consumption can (Malawi) (figure 4.9). The ranking of coun- be explained by a person’s circumstances tries changes considerably if one looks at in childhood and how much is explained inequality of opportunity rather than overall by individual responsibility, luck, or effort inequality (note that the countries in figure (obtained as the residual).18 Such estimates 4.9 are sorted by inequality): Countries with of inequality of economic opportunity are higher inequality in outcomes are not neces- available for many countries worldwide, sarily characterized by a larger share of the but evidence for Africa has been limited to inequality attributed to inequality of opportu- date.19 Drawing on surveys from 10 countries nity. The Comoros, for instance, has the high- (the Comoros, Ghana, Guinea, Madagascar, est overall level of inequality, but its share of Malawi, Niger, Nigeria, Rwanda, Tanzania, inequality of opportunity is among the low- and Uganda), this section presents more com- est. Furthermore, the magnitude of inequal- prehensive evidence for countries in Africa.20 ity of opportunity is only partly correlated with the number of circumstances available in the data, suggesting that the differences FIGURE 4.9 Unequal opportunities account for up to 20 percent of observed across countries do not solely reflect inequality in Africa differences in the availability of circumstance 25 variables but say something meaningful about the structure of inequality (however, more circumstances are also typically expected to Percent of inequality attributed yield greater inequality of opportunity). to unequal opportunities Estimates of inequality of opportunity 15 calculated in this manner represent a lower bound, because many circumstance variables (family wealth, parenting time, the quality of education) are not observed in household surveys and hence cannot be considered in 5 the estimation.23 This issue also complicates comparisons across countries, because the surveys differ in the number and granularity 0 of the circumstance variables. r Ta ar ia ria a ea Rw a da i os aw ge an d an c or in an ge an as Ni al Gh nz Gu m Ug Ni ag M Co ad Intergenerational Persistence in M Source: Brunori, Palmisano, and Peragine 2015b. Education and Occupation Note: The figure shows the share of total mean log deviation (MLD) that is attributed to inequality of economic opportunity. Countries are ordered by their level of inequality measured by the MLD, Does the educational attainment of parents with the least unequal countries on the left and the most unequal on the right. matter less today to a child’s schooling than INEQUALITY IN AFRICA 133 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 a. Comoros, 2004 b. Congo, Dem. Rep., 2011 c. Ghana, 2013 1.5 1.5 1.5 Persistence, SD ratio Persistence, SD ratio Persistence, SD ratio 1.0 1.0 1.0 0.5 0.5 0.5 0 0 0 1943 1953 1963 1973 1983 1951 1961 1971 1981 1991 1952 1962 1972 1982 1992 d. Guinea, 2003 e. Madagascar, 2005 f. Malawi, 2010 1.5 1.5 1.5 Persistence, SD ratio Persistence, SD ratio Persistence, SD ratio 1.0 1.0 1.0 0.5 0.5 0.5 0 0 0 1942 1952 1962 1972 1982 1945 1955 1965 1975 1985 1950 1960 1970 1980 1990 g. Nigeria, 2010 h. Rwanda, 2000 i. Tanzania, 2009 1.5 1.5 1.5 Persistence, SD ratio Persistence, SD ratio Persistence, SD ratio 1.0 1.0 1.0 0.5 0.5 0.5 0 0 0 1950 1960 1970 1980 1990 1939 1949 1959 1969 1979 1949 1959 1969 1979 1989 j. Uganda, 2005 k. Other developing countries 1.5 1.5 Persistence, SD ratio Persistence, SD ratio 1.0 1.0 Correlation coe cient Intergenerational gradient 0.5 0.5 Standard deviation ratio (parents/children) 0 0 1945 1955 1965 1975 1985 1940 1950 1960 1970 1980 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) 5 The growth in extreme wealth in the Aggregate net wealth as percent of GDP region since 2010 can be decomposed into 4 two components: the increase in the wealth of veteran billionaires and the addition of new 3 billionaires. More than half of the growth in Nigeria’s extreme wealth is explained by the growth in the wealth of the veterans. 2 The weight of newcomers in the growth in extreme wealth in South Africa rose from 40 1 percent in 2011 to 54 percent in 2013. Across the set of six countries, the contribution of 0 newcomers to the growth in extreme wealth 2006 2007 2008 2009 2010 2011 2012 2013 jumped from 37 percent in 2011 to 61 per- Angola Kenya Nigeria South Africa cent in 2013. Tanzania Uganda Total With a focus on billionaires, the Forbes list captures only the very top of extreme wealth. In 2013 Forbes reported on 50 Afri- Sources: Aggregate net wealth: Forbes’ “The World’s Billionaires.” GDP: World Development Indicators. cans worth at least $400 million. This list still leaves out lower levels of wealth that are high by any standard. 0.3 and 1.6 percent in 2010 to 3.2 and 3.9 Knight Frank (2015) surveys private bank- percent in 2013 (figure 4.11). The increase is ers and wealth advisors to collect data on partly explained by the rise in the number of ultra-high-net-worth individuals (individuals billionaires in both countries over the period. whose net worth exceeds $30 million) in 90 Nigeria’s rapid climb also stems from the fact countries, of which 14 are in Africa. Across that, since 2011, it has been the home of the countries, the number of ultra-high-net- worth individuals increases with GDP per capita growth. The number tends to increase FIGURE 4.12 Extreme wealth increases with GDP in Africa even where economies are in decline or stag- and elsewhere nating (in Zimbabwe, for example, the num- ber of ultra-high-net-worth individuals rose 20 in number of ultra-high-net-worth by 5.2 percent while GDP per capita declined Annualized percentage change 15 by 0.12 percent). Africa’s trend (not shown) individuals 2004–14 is very similar to the global trend (the gray 10 dotted line in figure 4.12). What do these data reveal about inequal- 5 ity? Given the limited data on these extremely wealthy individuals, there is no straightfor- 0 ward answer. Credit Suisse (2014) presents –5 estimates of the distribution of wealth using –4 –2 0 2 4 6 8 10 12 the Forbes list and imputations based on Annualized percentage change in GDP per capita 2004–13 cross-country relationships and consump- non-African countries African countries tion surveys. Using these data, Lakner (2015) finds that the 10 richest people in Africa pos- Sources: Data on number of ultra-high-net-worth individuals are adapted from World Bank 2014b, sess wealth equivalent to the wealth of the based on Knight Frank 2015. GDP data are from World Development Indicators. Note: GDP is measure in purchasing power parity constant 2011 dollars. Black dotted line shows the poorest half of the population. (His find- global trend (African and non-African countries). ings include North Africa, where 3 of the 10 INEQUALITY IN AFRICA 137 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 across geographic areas reflects differences empirical evidence on the consequences of in the cost of living. Between-group inequal- being orphaned focuses on health indicators ity for regions and for urban/rural areas is and education, with some studies showing a lower if a delated measure of consumption is causal impact of orphanhood on schooling used rather than a nominal measure. Across outcomes. Because orphanhood is not a ran- countries, it declines by about 15 percent on dom event (it is correlated with other house- average for both regions and urban/rural. hold measures, such as urban status and 16. If a deflated measure of consumption is household education and wealth), the cor- used, these ratios fall to 1.3 (Ethiopia), 2.2 relation does not imply that education levels (Democratic Republic of Congo), and 3.9 are generally worse among orphans. Indeed, (Nigeria). the most recent Demographic and Health 17. The focus here is on inequality of opportu- surveys show that in half of the countries nity from the perspective of economic out- surveyed, orphans are no less likely than comes in adulthood. A third domain is the nonorphans to be enrolled in school. In human opportunity index, which captures Nigeria and Chad, orphans are more likely the extent to which circumstances such as than other children to be in school. school attendance, immunizations, and 22. T h e r e a r e d i f f e r e nt m e t ho d o lo g i c a l household infrastructure, including access approaches to measuring inequality of to sanitation and water, contribute to gaps opportunity, including the choice of inequal- in outcomes for children. Dabalen and oth- ity measure (Gini or MLD), the estimation ers (2015) present detailed analysis of the approach (parametric or nonparametric), human opportunity index for Africa across and the choice of circumstances to use if the many countries and years. They find that set of circumstances differs across surveys. greater coverage for all was more important All circumstances available for each country than changes in equity for improvements in are used. This choice is the best in analyz- human opportunities. ing a single country, but it poses some dif- 18. This approach is described as the ex ante ficulties in terms of comparability across approach to measuring inequality of oppor- countries. There is a trade-off between the tunity, as opposed to the ex post approach robustness and usefulness of the analysis (Checchi and Peragine 2010; Fleurbaey and in each country and the demands of com- Peragine 2013). In the ex post approach, parability across countries. As the number there is no inequality of opportunity if peo- of circumstances increases, the estimate of ple who exert the same effort end up with inequality of opportunity will also increase. the same outcome. Inequality of opportunity The estimates reported here are based on a in this approach is measured as inequality nonparametric estimation approach (Fer- within responsibility classes (that is, within reira and Gignoux 2011). The MLD is com- the set of individuals at the same effort monly used as the measure of inequality in level). this literature, although some researchers 19. Exceptions are Cogneau and Mesplé-Somps propose using the Gini from a theoreti- (2008) for Côte d’Ivoire (1985–88), Ghana cal point of view (van de Gaer and Ramos (1988 and 1998), Guinea (1994), Mada- 2015) and an empirical perspective (Brunori, gascar (1993), and Uganda (1992); Piraino Palmisano, and Peragine 2015b). (2015) for South Africa; and Brunori, Palmi- 23. If data on these unobserved circumstances sano, and Peragine (2015a) for Uganda. were available, the share of inequality attrib- Broader international comparisons of uted to circumstances would go up, as would inequality of economic opportunity are pre- the level and share of inequality of economic sented in Ferreira and Gignoux (2011) for opportunity (though the extent of underes- Latin America and in Brunori, Ferreira, and timation also depends on the degree of cor- Peragine (2013) for 41 countries. relation between unobserved and observed 20. This subsection draws on Brunori, Palmi- circumstances). sano, and Peragine (2015b). 24. This subsection draws on Azomahou and 21. Another circumstance that is particularly Yitbarek (2015). relevant for countries in the region hard 25. Related studies include the following: hit by HI V/AIDS is orphan status. The B ossu roy a nd C og ne au (2013) cover INEQUALITY IN AFRICA 141 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.