Achieving the Millennium Development Goals (Studies in Development Economics and Policy)

Achieving the Millennium Development Goals Edited by

Mark McGillivray

Studies in Development Economics and Policy General Editor: Anthony Shorrocks UNU WORLD INSTITUTE FOR DEVELOPMENT ECONOMICS RESEARCH (UNU-WIDER) was established by the United Nations University as its first research and training centre and started work in Helsinki, Finland, in 1985. The purpose of the Institute is to undertake applied research and policy analysis on structural changes affecting the developing and transitional economies, to provide a forum for the advocacy of policies leading to robust, equitable and environmentally sustainable growth, and to promote capacity strengthening and training in the field of economic and social policy-making. Its work is carried out by staff researchers and visiting scholars in Helsinki and through networks of collaborating scholars and institutions around the world. UNU World Institute for Development Economics Research (UNU-WIDER) Katajanokanlaituri 6B, FIN-00160 Helsinki, Finland

Titles include: Tony Addison, Henrik Hansen and Finn Tarp (editors) DEBT RELIEF FOR POOR COUNTRIES Tony Addison and George Mavrotas (editors) DEVELOPMENT FINANCE IN THE GLOBAL ECONOMY The Road Ahead Tony Addison and Alan Roe (editors) FISCAL POLICY FOR DEVELOPMENT Poverty, Reconstruction and Growth George G. Borjas and Jeff Crisp (editors) POVERTY, INTERNATIONAL MIGRATION AND ASYLUM Ricardo Ffrench-Davis and Stephany Griffith-Jones (editors) FROM CAPITAL SURGES TO DROUGHT Seeking Stability for Emerging Economies David Fielding (editor) MACROECONOMIC POLICY IN THE FRANC ZONE Basudeb Guha-Khasnobis (editor) THE WTO, DEVELOPING COUNTRIES AND THE DOHA DEVELOPMENT AGENDA Prospects and Challenges for Trade-led Growth Basudeb Guha-Khasnobis, Shabd S. Acharya and Benjamin Davis (editors) FOOD INSECURITY, VULNERABILITY AND HUMAN RIGHTS FAILURE Basudeb Guha-Khasnobis and Ravi Kanbur (editors) INFORMAL LABOUR MARKETS AND DEVELOPMENT Basudeb Guha-Khasnobis and George Mavrotas (editors) FINANCIAL DEVELOPMENT, INSTITUTIONS, GROWTH AND POVERTY REDUCTION Aiguo Lu and Manuel F. Montes (editors) POVERTY, INCOME DISTRIBUTION AND WELL-BEING IN ASIA DURING THE TRANSITION

George Mavrotas (editor) DOMESTIC RESOURCE MOBILIZATION AND FINANCIAL DEVELOPMENT George Mavrotas and Anthony Shorrocks (editors) ADVANCING DEVELOPMENT Core Themes in Global Economics Mark McGillivray (editor) ACHIEVING THE MILLENNIUM DEVELOPMENT GOALS Mark McGillivray (editor) HUMAN WELL-BEING Concept and Measurement Mark McGillivray (editor) INEQUALITY, POVERTY AND WELL-BEING Robert J. McIntyre and Bruno Dallago (editors) SMALL AND MEDIUM ENTERPRISES IN TRANSITIONAL ECONOMIES Vladimir Mikhalev (editor) INEQUALITY AND SOCIAL STRUCTURE DURING THE TRANSITION E. Wayne Nafziger and Raimo Väyrynen (editors) THE PREVENTION OF HUMANITARIAN EMERGENCIES Machiko Nissanke and Erik Thorbecke (editors) GLOBALIZATION AND THE POOR IN ASIA Can Shared Growth be Sustained? Machiko Nissanke and Erik Thorbecke (editors) THE IMPACT OF GLOBALIZATION ON THE WORLD’S POOR Transmission Mechanisms Matthew Odedokun (editor) EXTERNAL FINANCE FOR PRIVATE SECTOR DEVELOPMENT Appraisals and Issues Laixiang Sun (editor) OWNERSHIP AND GOVERNANCE OF ENTERPRISES Recent Innovative Developments Guanghua Wan (editor) UNDERSTANDING INEQUALITY AND POVERTY IN CHINA Methods and Applications Studies in Development Economics and Policy Series Standing Order ISBN: 978 0 333 96424 8 hardback Series Standing Order ISBN: 978 0 230 20041 8 paperback (outside North America only) You can receive future titles in this series as they are published by placing a standing order. Please contact your bookseller or, in case of difficulty, write to us at the address below with your name and address, the title of the series and one of the ISBNs quoted above. Customer Services Department, Macmillan Distribution Ltd, Houndmills, Basingstoke, Hampshire RG21 6XS, England

Achieving the Millennium Development Goals Edited by

Mark McGillivray

in association with the United Nations University – World Institute for Development Economics Research

© United Nations University 2008 All rights reserved. No reproduction, copy or transmission of this publication may be made without written permission. No portion of this publication may be reproduced, copied or transmitted save with written permission or in accordance with the provisions of the Copyright, Designs and Patents Act 1988, or under the terms of any licence permitting limited copying issued by the Copyright Licensing Agency, Saffron House, 6–10 Kirby Street, London EC1N 8TS. Any person who does any unauthorized act in relation to this publication may be liable to criminal prosecution and civil claims for damages. The authors have asserted their rights to be identified as the authors of this work in accordance with the Copyright, Designs and Patents Act 1988. First published 2008 by PALGRAVE MACMILLAN Palgrave Macmillan in the UK is an imprint of Macmillan Publishers Limited, registered in England, company number 785998, of Houndmills, Basingstoke, Hampshire RG21 6XS. Palgrave Macmillan in the US is a division of St Martin’s Press LLC, 175 Fifth Avenue, New York, NY 10010. Palgrave Macmillan is the global academic imprint of the above companies and has companies and representatives throughout the world. Palgrave® and Macmillan® are registered trademarks in the United States, the United Kingdom, Europe and other countries ISBN-13: 978−0−230−21723−2 ISBN-10: 0−230−21723−0 This book is printed on paper suitable for recycling and made from fully managed and sustained forest sources. Logging, pulping and manufacturing processes are expected to conform to the environmental regulations of the country of origin. A catalogue record for this book is available from the British Library. A catalog record for this book is available from the Library of Congress. 10 17

9 16

8 15

7 14

6 13

5 12

4 11

3 10

2 09

Printed and bound in Great Britain by CPI Antony Rowe, Chippenham and Eastbourne

1 08

Contents

List of Tables and Figures

viii

Acknowledgements

xiii

Notes on the Contributors

xiv

List of Abbreviations

xvi

Foreword by Anthony Shorrocks

1

2

3

xviii

The Millennium Development Goals: Overview, Progress and Prospects Mark McGillivray Introduction The MDGs and progress towards them Volume structure and contents Projecting Progress Towards the Millennium Development Goals Howard White and Nina Blöndal Introduction Approaches to making projections Income poverty Mortality Education Nutrition HIV/AIDS Conclusions Appendix: projection methods Achieving Health, Wealth and Wisdom: Links between Aid and the Millennium Development Goals David Fielding, Mark McGillivray and Sebastián Torres Introduction A brief literature review Data definition and measurement v

1 1 2 14

20 20 20 23 28 34 39 42 43 45

55 55 57 60

vi

Contents

Empirical analysis Conclusion Appendix 4

5

6

7

Achieving the Water and Sanitation Millennium Development Goal P. B. Anand Introduction Hypotheses, methodology, data and limitations Analysis Conclusions and further issues Appendix Measuring Pro-Poor Progress Towards the Non-Income Millennium Development Goals Melanie Grosse, Kenneth Harttgen and Stephan Klasen Introduction The concept of pro-poor progress Methodology Data Results Conclusion Links between Childhood Mortality and Economic Growth and Their Implications for Achieving the Millennium Development Goals in India Sonia Bhalotra Introduction Why growth? Related research Data Descriptive statistics The econometric model Results Conclusions Achieving the Millennium Development Goal for Primary Schooling in India Sonia Bhalotra and Bernarda Zamora Introduction Data and definitions Related literature and contributions

65 84 86

90 90 92 95 111 113

123 123 125 126 127 129 144

150 150 151 153 156 157 160 164 169

174 174 175 179

Contents

Analytical approach Empirical model Results Decomposition and simulation Conclusions Appendix 8

The Burden of Government Debt in the Indian States: Implications for the MDG Poverty Target Indranil Dutta Introduction Debt and MDGs Methodology Results and analysis Simulation Conclusion

Index

vii

182 182 184 194 199 200

208 208 210 212 214 223 226 229

List of Tables and Figures Tables 1.1 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 2.10 2.11 2.12 2.13 2.14 2.15 2.A1 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8

The Millennium Development Goals Income poverty forecasts (% living below $1 a day) for 2015 from three studies Subregional income poverty estimates 2015 ($1 a day) Rural and urban poverty in 2015 headcount ratio and millions of people ($1 a day) Mortality estimates 2015 (rates per 1000) Subregional estimates of infant and under-5 mortality, 2015 Rural urban differentials in under-5 mortality, 2015 Access to water in 2015 Net primary enrolment rate, 2015 (proportion of age cohort enrolled) Subregional primary enrolment rates (NER) and numbers out of school, 2015 (millions) Numbers of children out of school, 2015 (millions) Gender equality in education, female/male ratio, 2015 Literacy in 2015 by region Alternative estimates of underweight Per capita food consumption (calories per person per day) Number of HIV/AIDS-related deaths Summary of main approaches to projecting MDG indicators Countries included in the analysis Descriptive statistics for the asset weights Summary statistics Variable definitions and model structure Fitted regression coefficients Main econometric results Equilibrium effects on each variable of 1 standard error shock to each equation Predicted percentage change in each variable for a 100 per cent increase in aid

viii

4 24 26 26 29 30 33 34 35 36 38 38 38 39 41 43 50 61 62 66 68 70 72 76 79

List of Tables and Figures

3.9

3.10

3.A1 4.1

4.2 4.3 4.4 4.5 4.6

4.7 4.8 4.9 4.10 4.11 4.12 4.13 4.A1 4.A2 5.1 5.2 5.3 6.1

Percentage increases in each wellbeing indicator with 10, 50 and 100 per cent increases in aid, disaggregated by quintiles Percentage increases in each wellbeing indicator with 10, 50 and 100 per cent increases in aid, disaggregated by initial wellbeing levels Data sources for the conditioning variables Mean and standard deviation of percentage of population having access to water and sanitation, 1990 and 2000 Distribution of number of countries as per access to water and sanitation, 2000 Access to water or sanitation: is it a function of per capita GDP? HDI and access to water and sanitation Synergy in providing access to water and access to sanitation Proportion of population with access to water and sanitation in 2000 is a function of the proportion of population having such access in 1990 Starting point effect and being in sub-Saharan Africa Top ten countries in terms of progress with regard to access to water, 1990–2000 (% of population) Top 10 movers with regard to per cent of population having access to sanitation, 1990 to 2000 Were high achievers different? Results of multiple regression analysis: dependent variable is access to water (% of population), 2000 Results of multiple regression analysis: dependent variable is access to sanitation (% of population), 2000 Whether child mortality rate (2000) is affected by access to water and sanitation Access to water: forecasts for year 2015 by country based on regression models Access to sanitation: forecasts by country based on regression models Selected MDGs in Bolivia (1989 and 1998) Non-income achievements by income decile (conditional on income, Bolivia, 1989 and 1998) Pro-poor growth and pro-poor progress in Bolivia (between 1989 and 1998) Level and change of under-5 mortality and income: all-India and states, 1970–98

ix

82

83 86

95 96 97 100 102

104 105 105 106 106 107 109 110 114 117 131 133 136 158

x

List of Tables and Figures

6.2 6.3 6.4 6.5 7.1 7.2 7.3 7.4 7.5 7.6 7.7 7.A1 7.A2 8.1 8.2 8.3 8.4 8.5 8.6 8.7

Changes in under-5 mortality and income: all-India and states, 1970–81 and 1982–94 The income elasticity of under-5 mortality: alternative sets of control variables Growth elasticities and fixed effects by state Was there a ‘structural break’ in the income elasticity? Selection of the samples for analysis Microdata sample statistics (weighted by all-India sample weight) Annual growth rates of state level variables (per cent p.a.), 1982–99 Probit estimates of school attendance among 6–11-year-old children Probit estimates of school completion among 12-year-old children Stage-2 results: regression of the state fixed effects on state-level variables Decomposition and prediction based on the linear probability model Linear probability model of school attendance for children aged 6–11 Linear probability model of school completion for children aged 12 years Random-effect models on log of the head count ratio Fixed-effect model on log of the headcount ratio 2SLS and random-effect models on log of the headcount ratio Predicted values of poverty in 2007 Predicted values of poverty in 2015 Simulated values of poverty in 2005, with varied levels of debt ratio Simulated values of poverty in 2015, with varied levels of health expenditure per capita

160 165 166 168 177 178 180 185 189 193 197 200 203 215 216 219 220 222 224 225

Figures 1.1 1.2 1.3 1.4 1.5 1.6 2.1

Progress towards MDG1, MDG2, MDG4 and MDG6 MDG progress in sub-Saharan Africa MDG progress in South Asia Progress towards MDG7 Progress towards MDG8 Total ODA from DAC member countries, 1990–2010 Share of global poverty (proportion of poor in each region using $US1 a day poverty line)

6 7 10 11 12 13 24

List of Tables and Figures

2.2 2.3 2.4 2.5 2.6 2.7 2.8 3.1 4.1

4.2

4.3 4.4 4.5 4.6 4.7 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 5.10

Income poverty 2015 (growth-based projections, differential elasticities) Under-5 mortality 2015 (UN Population Projection, 2004 Revision) Mortality trends Net enrolment rates 2015 (naïve projections) Patterns and trends from FAO data Trends in agricultural output per person (2000 = 100) HIV/AIDS three scenarios for Africa Frequencies of values of the five wellbeing indicators GDP per capita (average for 1997–9 in $US, 2000 prices) and percentage of people having access to water in 2000 GDP per capita (average for 1997–9 in $US, 2000 prices) and percentage of people having access to sanitation in 2000 Change in HDI and change in percentage of population with access to water, 1990–2000 Change in HDI and change in percentage of population with access to sanitation, 1990–2000 Countries as percentages with access to water and sanitation in 2000 Access to water in 1990 and in 2000 Access to sanitation in 1990 and in 2000 Growth incidence curve and absolute change for income Conditional NIGIC and absolute change for stunting (z-score ∗ 100) Conditional NIGIC and absolute change for primary completion rate Conditional NIGIC and absolute change for literacy rate Conditional NIGIC and absolute change for ratio of education of women to men Conditional NIGIC and absolute change for share of women in wage employment Conditional NIGIC and absolute change for under-5 survival rate Conditional NIGIC and absolute change for under-1 survival rate Conditional NIGIC and absolute change for vaccinations Conditional NIGIC and absolute change for birth attendance rate

xi

27 31 32 37 40 41 42 63

98

99 100 101 102 103 104 135 137 138 138 139 139 140 141 142 142

xii

List of Tables and Figures

5.11 5.12 6.1 6.2 6.3 6.4

Conditional NIGIC and absolute change for access to water Conditional NIGIC and absolute change for access to sanitation Trends in under-5 mortality by state Trends in real log income per capita by state The relationship of under-5 mortality and state income: quadratic fit by state All-India trends in under-5 mortality and state income: population-weighted averages

143 143 159 161 161 162

Acknowledgements This volume originates from the UNU-WIDER research project entitled ‘The Millennium Development Goals: Assessing and Forecasting Progress’. The Board of UNU-WIDER provided valuable suggestions during the early stages of the project. Tony Shorrocks, Director of UNU-WIDER, provided considerable encouragement, advice and support throughout the life of the project. Tony Addison, former Deputy Director of UNU-WIDER, provided very useful advice, especially at the early stages of the project. Valuable comments from anonymous referees on a previous draft of the volume were most useful in shaping the final product. The Millennium Development Goals: Assessing and Forecasting Progress project was supported by many UNU-WIDER staff. Special thanks are due to Anne Ruohonen and Adam Swallow. Anne served as the project secretary, providing extremely efficient support and consistently good humour. The book could not have ever come to completion without Anne’s inputs. Adam provided incisive and timely publications advice, handling often complex matters with sound judgement and skill. UNU-WIDER gratefully acknowledges the financial contributions to the research programme by the governments of Denmark (Royal Ministry of Foreign Affairs), Finland (Ministry for Foreign Affairs), Norway (Royal Ministry of Foreign Affairs), Sweden (Swedish International Development Cooperation Agency – Sida) and the United Kingdom (Department for International Development).

xiii

Notes on the Contributors P. B. Anand is Reader in Environmental Economics and Public Policy at the Centre for International Development, University of Bradford, UK. His research interests include collective action, new institutional economics, human rights and capability approach. He is currently researching agency and governance issues with a focus on the interaction between civil society and local governance institutions. Sonia Bhalotra is Reader in Economics at the University of Bristol, UK. She obtained an MPhil and a DPhil in Economics from the University of Oxford, UK, and a BSc in Economics from Delhi University, India. She has previously worked at the Universities of Cambridge and Sussex, UK, and has consulted for UNICEF, the World Bank, UNU-WIDER, UNESCO and the ILO. Nina Blöndal obtained her MSc from the London School of Economics, UK, in 2002 and has subsequently worked within the field of international development, focusing on development effectiveness and impact evaluation. She has worked on several impact evaluations for the World Bank’s Independent Evaluation Group and as a technical adviser for Danida and is currently studying for a PhD in economics at the University of Copenhagen, Denmark. Indranil Dutta is Lecturer in Economics at the University of Manchester, UK. He was previously a research fellow with UNU-WIDER. His research interests are mainly in the area of development economics, particularly on issues of poverty and inequality. David Fielding is Professor of Economics at the University of Otago, New Zealand. His research interests are in development macroeconomics and quantitative political economy. He has previously held positions at Oxford, Nottingham and Leicester Universities in the UK. His most recent Palgrave Macmillan book is Macroeconomic Policy in the Franc Zone (2005). Melanie Grosse is a PhD student in development economics at the University of Göttingen, Germany. She holds a first-class degree in economics at the University of Göttingen. Her work focuses on measuring poverty, inequality and pro-poor growth in income and non-income dimensions, with applications to Latin American countries. She is also the research manager of an interdisciplinary research group focusing on sustainable resource use in Indonesia. xiv

Notes on the Contributors

xv

Kenneth Harttgen studied economics at the University of Göttingen, Germany, and recently completed his PhD there on empirical dynamics of determinants, distribution and dynamics of poverty, with case studies from sub-Saharan Africa and Latin America. He currently holds a post-doctoral position at the University of Göttingen and has worked as a consultant to several international development agencies including UNDP, UNESCO, GTZ and the OECD. Stephan Klasen is Professor of Development Economics at the University of Göttingen, Germany. He holds a PhD from Harvard University and has since held positions at the World Bank, the University of Cambridge, UK, and the University of Munich, Germany. His research focuses on measurement and analysis of poverty and inequality in developing countries. Mark McGillivray is Chief Economist of the Australian Agency for International Development. He was previously Deputy Director of UNUWIDER. Mark is also honorary Professor of Development Economics at the University of Glasgow, UK, an External Fellow of the Centre for Economic Development and International Trade at the University of Nottingham, UK, and an Inaugural Fellow of the Human Development and Capability Association. His research focuses on aid effectiveness and allocation and measuring achieved human well-being. Sebastián Torres studied economics at the University of the Republic of Uruguay, development economics at the Institute of Social Studies, The Hague, The Netherlands, and has completed his PhD thesis on simultaneous equation models of economic development and income inequality at the University of Leicester, UK. He holds a post-doctoral grant from the Economic and Social Research Council and has worked as a consultant to UNESCO, UNDP, UNIDO and UNU-WIDER. Howard White is Fellow of the Institute of Development Studies, University of Sussex, UK. He has worked for many different agencies in countries across Africa and Asia and has over fifty publications in refereed journals, focusing on aid effectiveness and poverty reduction. His books include Targeting Development: Critical Perspectives on the MDGs and Econometrics and Data Analysis for Developing Countries. Bernarda Zamora holds a PhD in Economics from University Carlos III of Madrid, Spain. She has worked as an economist at the International Monetary Fund. Since 2002 she has worked in Spain as a Visiting Professor at the University of Alicante and University Jaume I in Castellón, and is currently a Research Associate at the School for Policy Studies of the University of Bristol, UK.

List of Abbreviations CHIM DAC Danida DHS EKC FAO GEP GIC GLS GMM GNI GRIM HDI HNP IDS IFPRI IIPS INFHS IMR ISS LAC LDCs MAMS MDGs MPTFE NER NFHS NIGIC NGO NSS ODA ODI OECD PPCH

change in mean Development Assistance Committee, OECD Danish International Development Agency demographic and health survey, World Bank environmental Kuznets curve United Nations Food and Agriculture Organization Global Economic Prospects, World Bank growth incidence curve generalized least squares generalized method of moments gross national income growth rate in the mean Human Development Index health, nutrition, population (poverty data), World Bank Institute of Development Studies, UK International Food Policy Research Institute, USA International Institute for Population Sciences, India Indian National Family Health Survey infant mortality rate Institute of Social Studies, The Netherlands Latin America and the Caribbean least-developed countries Maquette for MDG Simulations, World Bank Millennium Development Goals Millennium Project Task Force on Education net enrolment rate National Family Health Survey of India non-income growth incidence curve non-governmental organization national sample survey official development assistance Overseas Development Institute, UK Organisation for Economic Cooperation and Development pro-poor change

xvi

List of Abbreviations

PPGR Sida SimSIP SSA TI UNAIDS UN-DESA UNDP UNICEF UNFPA UNU-WIDER UPE WHO

pro-poor growth rate Swedish International Development Cooperation Agency Simulations for Social Indicators and Poverty, World Bank sub-Saharan Africa Transparency International Joint United Nations Programme on HIV/AIDS United Nations Department of Economic and Social Affairs United Nations Development Programme United Nations Childrens Fund United Nations Population Fund United Nations University World Institute for Development Economics Research universal primary education World Health Organization

xvii

Foreword

The Millennium Development Goals (MDGs) agreed at the United Nations Millennium Summit in September 2000 constitute the most ambitious enterprise ever undertaken by the international development community. They aim to reduce extreme income poverty and hunger, achieve universal primary education, promote gender equality and empower women, reduce child mortality, improve maternal health, combat HIV/AIDS and other major diseases, provide access to water and sanitation, ensure environmental sustainability and establish a global partnership for development. Each of these goals is defined in terms of a specific target to be achieved by 2015. While there has been tremendous progress in many countries, overall global progress towards achieving the MDG targets has been mixed, with many countries well behind the schedule for 2015. This volume results from the UNU-WIDER research project on ‘The Millennium Development Goals: Assessing and Forecasting Progress’. It suggests how the MDGs can be achieved, by presenting empirical analyses of the core determinants of MDG target variables. A key insight is that most of the MDG targets are causally related in one way or another. Recognition of these interdependencies is crucial not only in analysing the MDGs but also in devising strategies for their achievement. The book provides both broad mapping and details of each of the MDGs and their corresponding targets. It also reports actual and projected progress towards their achievement, examines the role that aid can play in relation to the education and child health objectives, analyses the determinants of the water and sanitation targets, and examines how the measurement of pro-poor growth can be improved with respect to the non-income dimensions of poverty targeted by the MDGs. Special attention is given to achieving the income poverty, education, and child health targets in India, the country which is home to more than 20 per cent of the world’s poor. This book will be of value to a wide audience, including not only those involved in designing and implementing strategies for the MDGs, but also readers with a general interest in poverty and human wellbeing in developing countries. ANTHONY SHORROCKS Director, UNU-WIDER xviii

1 The Millennium Development Goals: Overview, Progress and Prospects Mark McGillivray

Introduction The international community has over the years embraced various campaigns that have aimed to achieve certain development goals, variously defined. These include the United Nations ‘Education for All’ and ‘Health for All’ campaigns adopted in 1978 and 1990 respectively. Education for All aimed to achieve universal worldwide access to primary education by the year 2000. Health for All aimed to provide universal primary education and universal access to healthcare by the same year. Unanimously adopted by all member states at the United Nations Millennium Summit in September 2000, the Millennium Development Goals (MDGs) are to date the most ambitious and comprehensive developmental undertaking ever embraced by the international community. The MDGs involve the eradication of extreme income poverty and hunger, achieving universal primary education, promoting gender equality and empowering women, reducing child mortality, improving maternal health, combating HIV/AIDS and other major diseases, ensuring environmental sustainability and developing a global partnership for development. In most cases the corresponding targets are to be achieved by 2015. A key component of the United Nations strategy to achieve these highly ambitious goals is the doubling of official development assistance (ODA) from its 2003 level, to approximately $US135 billion per year by 2006 and to further increase it to $US195 billion per year by 2015 (United Nations Millennium Project 2005). The official donor community certainly appears to be taking both these calls and the MDGs seriously, to the extent that it has responded both with enthusiastic words at the 2002 Monterrey Conference on Financing for Development and, more importantly, with substantial increases in ODA in subsequent 1

2

Mark McGillivray

years. Many core activities of a number of key international organizations (OECD, UNDP, the World Bank and others within the United Nations system) have also been geared towards the MDGs, and civil society worldwide has embraced the goals. There was also the Second United Nations Millennium Summit in September 2005, at which the international community reaffirmed its commitment to achieving the MDGs. The preceding factors should not be taken to imply that the MDGs have been enthusiastically embraced by all concerned or that they have not been the subject of criticisms, valid or otherwise. Among the criticisms of the MDGs is that they reflect a poor level of analysis, hide more than they reveal about global development challenges, carry the potential to distort meaningful intellectual and research agendas and could serve as a harmful vehicle for a realignment of the political economy of development at the global level (Saith 2006). These factors do, however, combine to suggest that one should not reject the MDGs as merely another symbolic or hollow gesture of the international community, the apparent failure of the Education for All and Health for All campaigns notwithstanding. Achieving the Millennium Development Goals aims to provide analytical insights into how the MDG targets might be achieved. It does this by presenting original and rigorous empirical analyses of key behavioural relationships and how they are likely to impact on progress towards the MDGs. A key recognition is that most of the MDG targets are casually related in one way or another. No one goal can be looked at in isolation from the others, nor from key macroeconomic outcomes not built directly into or recognized within the MDGs. Central to achieving the MDGs is a recognition of these interpendencies, and any robust and insightful analysis of them must take this into account. This first chapter has two remaining aims. The first is to outline briefly each MDG and the progress made towards the targets on which it is based. To this extent, it provides a broad informational context for the remaining chapters. The second is to provide an overview of the volume and briefly to outline the contents of those chapters.

The MDGs and progress towards them The MDGs The United Nations General Assembly, at the 2000 United Nations Millennium Summit, adopted unanimously what is known as the Millennium Declaration. The MDGs were a component of this Declaration.The Declaration itself is much broader than the MDGs, containing inter alia

Overview, Progress and Prospects

3

statements of principle relating to freedom, equality, solidarity, tolerance, respect for nature and shared responsibility (United Nations 2000). The MDGs actually emerge from the section of the Declaration addressing development and poverty eradication. They can be seen largely as a response by the international community to the intolerably low levels of wellbeing experienced by so many people living in developing countries, and the growing gaps in living standards achieved between the richest developed and poorest developing countries. It should come as no surprise, therefore, that seven of the eight MDGs focus on achieving living standards or wellbeing outcomes primarily in developing countries. These MDGs, MDG1 through to MDG7, can be interpreted as intrinsic, in that they involve outcomes worth achieving in their own right. Nor should it come as a surprise that the remaining goal, MDG8, mainly has instrumental roles vis-à-vis these outcomes, calling on certain actions from developed countries. The articulation of each MDG is a statement of broad principle or intent that at best is open to interpretation or at worst vague without clear and precise meaning. MDG1, for example, is to ‘eradicate extreme poverty and hunger’. For most goals this vagueness is removed by one or more targets with which each is associated. Each target and the MDG to which it corresponds are outlined in Table 1.1. There are 18 targets in total. Some of them address long-held priorities of the UN, for which strategies have been in place for some time, such as addressing the needs of the least-developed countries, landlocked countries and small island developing states (MDG8, Targets 8.2 and 8.3). The target for MDG2, ensuring that by 2015 all children complete a full course of primary schooling, builds on the above-stated aim of the UN Education for All campaign. Ten of the MDG targets are time-bound and defined in reasonably precise quantitative terms. Nine targets are intended to be achieved by 2015, and one, improving the lives of at least 100 million slumdwellers (Target 7.3), is to be achieved by 2020. The main goal, that which receives most attention and emphasis, is the income poverty reduction target. This target is to reduce by 2015 the proportion of people whose income is less than one dollar per day, measured in terms of international purchasing power dollars ($PPP), to half of what it was in 1990. It follows that the measure of extreme income poverty on which MDG1 is partly based is the World Bank poverty headcount threshold of $PPP1 per day. The remaining eight MDG targets are more qualitative in nature and are perhaps better described as statements of principle or intent. With the exception of Target 7.1, which corresponds to MDG7, all of these

4 Table 1.1 The Millennium Development Goals MDG1: Eradicate Extreme Poverty and Hunger Target 1.1: halve, between 1990 and 2015, the proportion of people living on less than a dollar a day. Target 1.2: halve, between 1990 and 2015, the proportion of people who suffer from hunger. MDG2: Achieve Universal Primary Education Target 2: ensure by 2015 that all boys and girls complete a full course of primary schooling. MDG3: Promote Gender Equality and Empower Women Target 3: eliminate gender disparity in primary and secondary education preferably by 2005 and in all levels of education by 2015. MDG4: Reduce Child Mortality Target 4: reduce by two thirds, between 1990 and 2015, the mortality rate among children under five. MDG5: Improve Maternal Health Target 5: reduce by three quarters, between 1990 and 2015, the maternal mortality ratio. MDG6: Combat HIV/AIDS, Malaria and Other Diseases Target 6.1: halt by 2015 and begin to reverse the spread of HIV/AIDS. Target 6.2: halt by 2015 and begin to reverse the incidence of malaria and other major diseases. MDG7: Ensure Environmental Sustainability Target 7.1: integrate the principles of sustainable development into country policies and programmes and reverse loss of environmental resources. Target 7.2: halve, between 1990 and 2015, the proportion of people without sustainable access to safe drinking water and basic sanitation. Target 7.3: to improve the lives of at least 100 million slum dwellers by 2020. MDG8: Develop a Global Partnership for Development Target 8.1: develop further an open, rule-based, predictable, non-discriminatory trading and financial system, including a commitment to good governance, development, and poverty reduction both nationally and internationally. Target 8.2: address the special needs of the least developed countries, including tariff and quota free access for least developed countries’ exports; enhanced programme of debt relief for HIPCs and cancellation of official bilateral debt; and more generous ODA for countries committed to poverty reduction. Target 8.3: address the special needs of landlocked countries and small island developing states. Target 8.4: deal comprehensively with the debt problems of developing countries through national and international measures in order to make debt sustainable in the long term. Target 8.5: in cooperation with developing countries, develop and implement strategies for decent and productive work for youth. Target 8.6: in cooperation with pharmaceutical companies, provide access to affordable essential drugs in developing countries. Target 8.7: in cooperation with the private sector, make available the benefits of new technologies, especially information and communications. Source: United Nations (2007).

Overview, Progress and Prospects

5

qualitative targets correspond to MDG8, ‘Developing a Global Partnership for Development’. Target 8.2, for instance, involves ‘more generous ODA for countries committed to poverty reduction’. Putting aside the issue of recipient-country commitments to poverty reduction, there are various ways ‘more generous ODA’ can be interpreted. For example, it could involve donors giving larger proportions of their gross national incomes (GNIs) as ODA. Likewise, more generous ODA might simply be interpreting in absolute rather than relative volumes or in terms of more aid provided in the forms of grants rather than loans. Each of these interpretations points to obvious measures on the basis of which quantitative targets could be defined, such as the level of ODA as a percentage of donor gross national income. This measure is the basis of the well-known and longstanding 0.7 per cent target, to which there is no reference in the Millennium Declaration. The international community could have easily adopted precise, quantitative targets for each of the areas addressed in MDG8 had there been the commitment to do so. Such commitment is lacking, it seems. Progress towards the MDGs The MDGs, as mentioned, are a response to the intolerably low living standards of so many of the world’s population. The vast majority of these people – practically all, according to some indicators – live in developing countries. With the possible exception of tackling the spread of HIV/AIDS and tuberculosis, which are essentially global epidemics, achieving the first seven MDGs is essentially about progress in these countries. It therefore follows that tracking performance towards them, and establishing in which parts of the world the greatest challenges will be faced, requires us to focus on the developing world. Now we have made these points, it is clear that of effort required to meet the MDGs is very unevenly distributed across the regions of the developing world. Most of these regions will achieve most of the MDGs. But achieving most of the MDGs in sub-Saharan Africa (SSA) is unlikely in the extreme, it seems, based on a simple examination of the relevant data. Indeed, if the developing world as a whole does not achieve some MDGs, it will be due to a lack of progress in SSA. A lack of progress in South Asia will also play a part in this outcome. Figures 1.1 and 1.2 help illustrate the comments just made regarding overall developing-country and SSA progress towards the MDGs. We return to the progress in South Asia later. All data shown in these figures, and those appearing later in the chapter, have been either taken from or calculated using information in UNAIDS (2006), OECD (2007a; 2007c),

6

1990

1999

2005

2015 (Target)

100 80 60 40 20 0 MDG1: Income Poverty (% of population)

Figure 1.1

MDG1: Hunger (% of children)

MDG2: Primary Schooling (% of school age population)

MDG4: Child Mortality (deaths per 1000 live births)

MDG6: HIV Prevalence (millions of people)

MDG6: Tuberculosis (millions of people)

Progress towards MDG1, MDG2, MDG4 and MDG6

Notes: (i) Data on MDG1, MDG2 and MDG4 are for developing regions only, while data for MDG6 are for the world. (ii) 2005 data for income poverty and child mortality are actually for 2004. (iii) Hunger and child mortality data are not available for 1999.

1990

1999

200 180 160 140 120 100 80 60 40 20 0 MDG1: Income Poverty (% of population)

Figure 1.2

MDG1: Hunger (% of children)

MDG2: Primary Schooling (% of school age population)

MDG4: Child Mortality (deaths per 1000 live births)

MDG6: HIV Prevalence (millions of people)

MDG progress in sub-Saharan Africa

Notes: (i) 2005 data for income poverty and child mortality are actually for 2004. (ii) Hunger and child mortality data are not available for 1999.

7

8

Mark McGillivray

UN (2007), World Bank (2007) and WHO (2007). Only progress towards those targets for which sufficient statistical data are available are shown. The targets shown in these figures have been calculated directly from published 1990 data or, in cases where the target is not precisely articulated, have been inferred. The income poverty target, for instance, has been calculated by taking 50 per cent of the number of people living in poverty worldwide according to the information reported in UN (2007). Likewise, the child mortality target has been obtained by taking one-third of the 1990 child mortality rate reported in this source. The official targets for MDG6 are a little vague in that they simply mention halting and beginning to reverse the spread of the diseases in question. The MDG6 targets shown in Figures 1.1 and 1.2 are inferred, therefore, being premissed on the possibly generous assumption that ‘halting and beginning to reverse the spread’ involves keeping the incidence of these diseases at their 1999 levels. Consider first the income poverty target. The income poverty data shown in Figures 1.1 and 1.2 are for the percentage of the world and SSA population living on less than $PPP1 per day. In developing regions as a whole, this percentage fell by 13 points, from 32 in 1990 to 19 in 2004. Such a fall corresponds to the number of people living on less than this income dropping from 1.25 billion in 1990 to 980 million in 2004 (UN 2007). The target is 16 per cent by 2015 and, if this progress continues, the MDG1 income poverty target will, in all probability, be met in the developing world as a whole. Africa’s march to the income poverty target is much less certain. In 1990, 47 per cent of the population in SSA lived below the $PPP1 per day poverty line. This percentage fell to 41 in 2004. But meeting the income poverty target requires a reduction of 18 percentage points, to 23 per cent, which on face value would appear unlikely given the trend since 1990. Similar scenarios exist with respect to the hunger target for MDG1, and the targets for MDG2, MDG4 and MDG6, as Figures 1.1 and 1.2 demonstrate. The hunger data relate to the percentage of children underweight, which is interpreted as an indicator of hunger and is used by international agencies to monitor progress with respect to the second MDG1 target. A 10 percentage point reduction between 2005 and 2015 is required if the developing regions as a whole are to reach the hunger target. A 12-point reduction over the same period is required in SSA. Developing regions collectively need to achieve a further 12 percentage point increase in primary school enrolments if MDG2 is to be achieved.1 A 20-point increase is required in SSA. A seemingly unlikely 48-unit reduction in the child mortality rate is required between 2004 and 2015

Overview, Progress and Prospects

9

if MDG4 is to be achieved in developing countries as a whole. A seemingly impossible 104-unit reduction is required over the same period in sub-Saharan Africa. The most striking trends are those in HIV prevalence. While all other target variables shown in Figure 1.1 show progress, albeit not apparently sufficient in some cases to achieve the corresponding target, the number of people worldwide with HIV continues to rise. Between 1990 and 2005, the number of people worldwide with HIV increased from 8 million to 39 million. Keeping the spread of HIV at its 1999 level is clearly not happening, as its global prevalence rose by almost 10 million people between 1999 and 2005. A close inspection of Figure 1.2 shows that the majority of the world’s HIV-infected people are actually in SSA. While this region’s contribution to world infections has declined over time, 64 per cent of people in the world with HIV are in SSA. Little more needs to be said about these numbers. This is clearly a problem of enormous magnitude. Better news, not shown in Figure 1.1, is that the global rate of increase in HIV prevalence among 15-to-49-year-olds has begun to taper off in recent years. While rising from 0.3 per cent in 1990 to just under 1 per cent in 2002, since 2003 it has remained at 1 per cent. A similar trend is evident in SSA, where 6 per cent of those in the 15-to-49-year age group are infected with HIV (UNAIDS 2006). Similar news applies to tuberculosis, as Figure 1.1 suggests. While an estimated 8.8 million new cases were reported in 2005, the prevalence of tuberculosis fell from 16.6 million people in 1999 to 14.4 million in 2005 (UN 2007; WHO 2007). If the target for tuberculosis is interpreted as maintaining its incidence at the 1999 level, then the target will be achieved, provided its incidence can be kept at its current level or slightly higher. It is widely recognized that the MDGs will be hardest to achieve in SSA, consistent with the evidence just presented. This should not imply however that there will be little or no difficulty in reaching some MDGs in other parts of the world. There are, indeed, widespread concerns that a number of MDGs may not be met in South Asia, as was alluded to above. Figure 1.3 helps explain why. There are some concerns about whether the income poverty target will be achieved, and a recognition that progress in India is important in this regard (UN 2007). The most profound concerns are for the MDG1 hunger and MDG4 child mortality targets. There has been comparatively little progress towards the first of these targets. As Figure 1.3 shows, the proportion of children underweight fell by 7 percentage points between 1990 and 2005. A further drop of 25 percentage points, from 46 to 21 per cent, is required if the MDG1 hunger target is to be reached. Child mortality fell by 44 deaths per 1000 live

10 Mark McGillivray 1990

1999

2005

2015 (Target)

140 120 100 80 60 40 20 0 MDG1: Income Poverty (% of population)

MDG1: Hunger (% of children)

MDG2: Primary Schooling (% of school age population)

MDG4: Child Mortality (deaths per 1000 live births)

Figure 1.3 MDG progress in South Asia Notes: (i) 2005 data for income poverty and child mortality are actually for 2004. (ii) Hunger and child mortality data are not available for 1999.

births between 1990 and 2005. This fall is substantial, but it comes from a very high base of 126 deaths per 1000 live births, and a seemingly improbable further decline of 42 deaths per 1000 live births is required by 2015 if MDG4 is to be achieved. Progress with respect to MDG7 and MDG8 is shown in Figures 1.4 and 1.5. Progress in the context of these goals is necessarily vague, given the absence of clearly specified targets. The exception is the MDG sanitation target, which is specified precisely. On face value, progress towards this target needs to be accelerated if it is to be reached. What can be said about the remaining MDG7 and MDG8 targets? While falling between 1990 and 1999 in overall volume terms, the level of ODA provided by countries that are members of the OECD Development Assistance Committee (DAC) is higher in 2005 than in 1990. This is shown in Figure 1.5. This applies to total ODA and to that allocated to leastdeveloped countries (LDCs). It is evident that developing-country access to developed-country markets increased between 1999 and 2005; the debt of developing countries fell between 1990 and 2005, and both youth unemployment and the use of new technology increased between 1999 and 2005 (see Figure 1.5). While the proportion of slum-dwellers has decreased, a less pleasing picture emerges for the remaining MDG7 target variables (see Figure 1.4). This is to the extent to which deforestation has remained virtually constant between 1990 and 2005.2

1990

1999

2005

2015 (Target)

80 70 60 50 40 30 20 10 0 MDG7: Basic Sanitation (% of population)

Figure 1.4

MDG7: Deforestation (% of land area)

MDG7: Greenhouse Gas Emissions (Developing Regions) (billions of tonnes)

MDG7: Greenhouse Gas Emissions (Developed Regions) (billions of tonnes)

MDG7: Slum Dwellers (% of urban population)

Progress towards MDG7

Notes: (i) 2005 data for basic sanitation and greenhouse gas emissions are actually for 2004. (ii) Sanitation and greenhouse gas emission data are not available for 1999.

11

12

1990

1999

2005

90 80 70 60 50 40 30 20 10 0 MDG8: ODA (Total) (billions of dollars)

Figure 1.5

MDG8: ODA (LDCs) (billions of dollars)

MDG8: Market Access (% of imports)

MDG8: Debt (% of exports)

MDG8: Youth (% unempolyed)

MDG8: New MDG8: New Technology Technology (Developed Regions) (Developing Regions) (users per 100 (users per 100 population) population)

Progress towards MDG8

Notes: (i) 2005 youth unemployment data are actually for 2006. (ii) Youth and new technology not available for 1990.

Overview, Progress and Prospects

13

2000

2010

140000 Actual

Predicted

Net disbursements ($US millions, 2005 prices)

120000 100000 80000 60000 40000

2009

2008

2007

2006

2005

2004

2003

2002

2001

1999

1998

1997

1996

1995

1994

1993

1992

1991

0

1990

20000

Figure 1.6 Total ODA from DAC member countries, 1990–2010

It was mentioned at the outset of this chapter that a key component of the strategy to achieve the MDGs is the doubling of ODA from its 2003 level to approximately $US135 billion per year by 2006 and to further increase it to $US195 billion by 2015. It was also mentioned that the UN has called on donors to actually provide these levels of ODA. Further information on the level of ODA is provided in Figure 1.6, which shows the actual annual levels of global aid from 1990 to 2006 as well as anticipated or projected levels, based on donor pledges and forward commitments, to 2010. The ODA data shown in Figure 1.6 have been obtained from OECD sources (2007a; 2007b; 2007c). A scaling up of ODA is clearly evident and the level of these flows in 2005, which amounted to $US107 billion, is the highest ever provided by OECD DAC members. The scaling up did not, however, result in the hoped-for doubling of ODA by 2006. In fact the level of ODA in 2006 fell back from its 2005 level, albeit slightly, to $US104 billion. This in part reflects the nature of the very much increased ODA in 2005, which was driven largely by increases in debt forgiveness that do not wholly reflect a high real allocation of public funds to aid budgets. If the trend in anticipated levels is sustained and if anticipations are correct, with donors fully delivering on pledges and commitments, ODA will reach in 2005 prices somewhere between $US160 billion and $US170 billion by 2015. This is clearly well short of the call from the UN for ODA to reach $US195 billion by 2015. The implications of this shortfall for obtaining the MDGs remain to be seen.

14 Mark McGillivray

Finally, having examined ‘progress towards the MDGs’ it would be remiss, at this stage of the chapter, not to consider what this actually means. What does ‘progress towards the MDGs’ or for that matter ‘MDG achievement’ actually mean? The preceding discussion has, to some extent, been vague with respect to these questions. Answers to them logically follow from each other, so let us focus on the meaning of the second question. There are three possible interpretations for any given goal: (i) achievement in all countries and therefore worldwide; (ii) achievement in all regions of the world but not necessarily achievement in each country within each region; or (iii) achievement in the world as a whole but not necessarily in each region or country. The second and third scenarios could be interpreted as referring to average achievements, in that overachievement in some countries compensates for failure to achieve the goals in others. Some associated with the design and implementation of strategies to achieve the MDGs have recently sought to provide clarification on precisely what ‘MDG achievement’ actually means. One position is that assessing whether progress is ‘on track’ for meeting the 2015 targets can be done only at the global level and cannot, therefore, be done for any specific region or particular country. This corresponds to interpretation (iii) above. Vandemoortele (2007:6) adopts this position, specifically asserting that ‘it is erroneous . . . to lament that sub-Saharan Africa will not meet the MDGs’. The current chapter does not seek to resolve this issue, but one point is worth making. To be educated, to be healthy and to have an adequate material standard of living reflects universal human values. They are identified in the United Nations Charter on human rights: each is, in fact, a universal, unalienable human right. This is why most of the MDGs can be viewed as having intrinsic value. These recognitions, which are reflected in the Millennium Declaration, provide a case for defining the MDGs as targets that are to be met within each country. To claim, for instance, that the MDGs have succeeded in eradicating extreme poverty and hunger when at the same time these conditions persist across an entire region or in a number of countries would appear to be inconsistent with the spirit of the Millennium Declaration. Put differently, relying purely on global aggregates seems inconsistent with the principles on which the MDGs are founded and would appear to be reflecting somewhat shaky ethical grounds.

Volume structure and contents Achieving the Millennium Development Goals consists of seven more chapters, each of which examines or uses empirical research methods.

Overview, Progress and Prospects

15

Chapters 2 to 5 look at generic issues that are notnecessarily specific to any single country or developing-country group. Among the issues considered are projections of progress towards the goals, the impact of aid and interrelations between goals, the attainment of the targets relating to water and sanitation, and pro-poor growth measurement in non-income dimensions of poverty. Chapters 6 and 7 examine the health and education outcomes on which the MDGs focus – in particular child mortality and school enrolment in particular – using household data for Indian states. Chapter 8 looks at links between efforts within Indian states to service foreign debt and progress toward the MDG income poverty reduction target. More detailed descriptions of each chapter follow below. These descriptions highlight the main finding or findings from each chapter, in particular those that are policy-relevant. Why do three of this volume’s eight chapters focus on India, especially given that the greatest challenges in meeting the MDGs are in sub-Saharan Africa? There are four reasons for this. First, in the early 2000s roughly a quarter of the world’s poor – those living on less than $PPP1 per day, lived in India. Approximately 850 million Indians – 80 per cent of the country’s population – lived on less than $PPP2 per day in the early 2000s (UNDP 2006). Second, despite India’s considerable economic progress in recent years, there are widespread concerns that India will not achieve some of the MDGs. This point was partly alluded to above, when reference was made to the slow progress towards the MDG1 hunger target and MDG4 in the South Asian region as a whole. The rate of progress towards these goals in India is a factor contributing to this, due in part to the size of the Indian population. Third, much of the country- or region-specific research on the MDGs has tended, rightly, to focus on SSA. Comparatively little research has been undertaken for other parts of the world. But achieving the MDGs in other regions of the world is important and here there is an apparent void in the literature. Fourth, data sets relating to living conditions in India are much better than those for most other developing countries. More rigorous investigation into achieving the MDGs is thus possible for India, and many questions can be addressed. Chapter 2 assesses progress towards the millennium development goals. Since the adoption of the International Development Targets, and their successors the MDGs, a growing number of publications have presented estimates of development outcomes in 2015 which, as mentioned, is the target year for most of the goals. What most of these projections show is that the developing world as a whole is ‘off track’ with respect to a number of targets and many countries will fall far short. Chapter 2 examines the soundness of these somewhat dire projections. On the

16 Mark McGillivray

basis of this examination, it concludes inter alia that many of the gains achieved in the outcomes targeted by the MDGs will continue into the new millennium, although not usually fast enough to achieve the ambitious targets set by goals. Neither the goal for income poverty reduction nor that for lower mortality will be met in the vast majority of countries. Attaining universal primary education is the one area where the goal looks achievable in many countries, though by no means all. The chapter warns, however, that these projections are based on various assumptions, including the assumption of business as usual and that various adverse shocks may result in far worse scenarios. Chapter 3 examines aid and the MDGs in terms of health, wealth and education. It uses a relatively new cross-country data set to estimate: (i) the strength of the links between a number of MDG target and related variables, including health, educational status and access to water and sanitation; and (ii) the extent to which aid impacts on these variables. The chapter differs from previous studies of links between wellbeing variables and investigations of aid effectiveness by analysing data for different population subgroups in each country, thus avoiding a number of drawbacks of using national-level data. Among the chapter’s findings is that child mortality is the central variable, where decreases lead to the largest beneficial changes in the other MDG or MDG-similar variables under consideration. It is also the variable on which aid has the largest quantitative impact. This implies that if aid flows are to achieve the maximum benefit, donors should prioritize primarily the MDG4 target for child mortality. The authors also find that while aid is effective overall, the poorest subgroups within each country are typically not the principal beneficiaries of these inflows. This suggests that if the wellbeing of these groups and inequality reduction are priorities, donors need to try harder to target these groups more effectively. Failure to do so will result in a more inequitable world, even if the MDGs are achieved. Chapter 4 uses cross-country regression analysis to develop models to forecast the projected proportion of population with access to water and sanitation in 2015, based on current variables. This study also revisits the issue of whether per capita GDP, levels of human development and governance impact on access to water and sanitation. Further, an attempt is made to explore whether the synergy effect is significant in a statistical sense. This involves examining whether the countries that have made significant progress with one target are more likely to make significant progress with other related targets, and whether and to what extent the achievement or lack of progress on these two targets can impinge on performance in relation to other MDGs or targets. The author finds that on

Overview, Progress and Prospects

17

current trends the water target will be either just barely achieved or else narrowly missed in the majority of countries, while the sanitation target will be missed in the great majority of countries. The chapter also points to a strong relationship between access to water and sanitation and child mortality, which suggests that the international community needs to address more seriously the prospect that the MDG7 might not be met. Chapter 5 looks at the non-income-related MDGs. As noted above, all but one of the MDGs involve a target that is not defined in terms of income. While there are plenty of measures designed to track progress in incomes, there are no corresponding measures for tracking the distribution of progress in non-income dimensions of poverty, and thus the distribution of progress towards MDGs 2–7. Chapter 5 proposes to extend the pro-poor growth measurement to non-income dimensions of poverty, particularly health and education. It illustrates empirically the proposed approach for Bolivia and shows that it allows a much more detailed assessment of progress towards MDGs 2–7 by focusing on the distribution of progress. Furthermore, this extension also allows an explicit assessment of the linkage between progress in MDG1 and MDGs 2–7 as well as extending traditional incidence analysis by quantifying outcomes in non-income dimensions of poverty along the income distribution. Chapter 6 links childhood mortality and economic growth in India. As such, it addresses MDG4 and picks up on the concerns that South Asia might not achieve this goal. Using state-level data obtained from the Indian National Family Health Survey (INFHS), the chapter investigates the extent to which the decline in child mortality in India over the last three decades can be attributed to economic growth. In doing this, it exploits the considerable variation in growth over this period, across states and over time. Empirical estimates reported in Chapter 6 are then used to produce a crude estimate of the rate of economic growth that would be necessary to achieve the MDG of reducing the under-5 mortality, by the year 2015,to a third of its level in 1990. The main conclusion is that while growth does have a significant impact on mortality risk, growth alone cannot be relied upon to achieve the goal. Chapter 7 addresses achievement of the MDG for primary schooling in India, and uses two large repeated cross-sections, one for the early 1990s and one for the late 1990s, to describe growth in school enrolment and completion rates for boys and girls, and to explore the extent to which enrolment and completion rates have developed over time. The data are also taken from the INFHS. It decomposes this growth into one component due to changes in the characteristics that determine schooling, and a second associated with changes in the responsiveness of

18 Mark McGillivray

schooling to given characteristics. The chapter’s analysis performs illustrative simulations relevant to the question of whether India will be able to achieve the universal primary education target by the year 2015. The simulations suggest that India will achieve universal attendance, but that primary-school completion rates will not exhibit much progress. Chapter 8 explores what impact, if any, Indian state government debts have on achieving the income poverty target of the MDGs. To fulfil this and many of the other MDG targets, national governments, especially in the developing world, have to undertake major investments in the social sector; but how much they will really be able to do so will depend on the conditions of their finances. The chapter finds that government investment in the social sector is extremely important for the Indian states in reducing poverty, but the government’s debt burden is actually stopping several states from attaining the poverty target. Specifically the chapter finds that while the impact of the debt on poverty is not very harmful in the medium term, it has significant negative impact in the longer run. The chapter’s main conclusion is, therefore, that for policy purposes, reductions in debt should be given priority. The topics covered in this book address important issues relating to the achievement of the MDGs, both in India and elsewhere. They also attempt to give some insight into the state of MDG-relevant research. While the chapters are useful in their individual focus, it is also hoped that they will stimulate further discussion aimed at better, more effective progress towards worldwide achievement of the MDG targets and, more generally, towards a more equitable and stable world.

Acknowledgements The author is grateful to two anonymous referees for useful comments on this chapter. The usual disclaimer applies.

Notes 1. The primary school enrolment data shown in Figures 1.1 and 1.2 are the number of students of primary school age, enrolled in either primary or secondary school, as a percentage of the total population in that age group. See UN (2007) for further details. 2. The market access data shown in Figure 1.5 are the percentage of imports (excluding arms and oil) from developing countries admitted duty-free to developed countries. Debt data are external debt payments as a percentage of export revenue. New technology data are the number of internet users.

Overview, Progress and Prospects

19

References OECD (Organisation for Economic Co-operation and Development) (2007a) OECD Journal on Development: OECD DAC Development Co-operation Report 2006. Paris. OECD (Organisation for Economic Co-operation and Development) (2007b) ‘Development Aid from OECD Countries Fell 5.1% in 2006’. Press release. Paris. OECD (Organisation for Economic Co-operation and Development) (2007c) International Development Statistics On-line. Paris. Saith, A. (2006) ‘From Universal Values to Millennium Development Goals: Lost in Translation’. Development and Change, 37(6): 1167–99. UNAIDS (Joint United Nations Programme on HIV/AIDS) (2006) 2006 Report of the Global AIDS Epidemic. Geneva. UNDP (United Nations Development Programme) (2006) Human Development Report, 2006. New York: Palgrave Macmillan. United Nations (2000) United Nations Millennium Declaration: Resolution Adopted by the General Assembly, Fifty-fifth Session, Agenda Item 60(b) New York. United Nations (2007) The Millennium Development Goals Report 2007. New York. United Nations Millennium Project (2005) Investing in Development: A Practical Plan for Achieving the Millennium Development Goals. New York: UNDP. Vandemoortele, J. (2007) ‘The MDGs: “M” for Misunderstood?’. WIDER Angle 2007/1: pp. 6–7, Helsinki: UNU-WIDER. WHO (World Health Organization) (2007) Global TB Database. Geneva. World Bank (2007) World Development Indicators, 2007. Washington, DC.

2 Projecting Progress Towards the Millennium Development Goals Howard White and Nina Blöndal

Introduction Since the adoption of the International Development Targets, and their successor Millennium Development Goals (MDGs), a growing number of publications have presented estimates of development outcomes in 2015, the target year for most of the goals. What the majority of these projections show is that the developing world as a whole is ‘off track’ with respect to most targets. They will not, in aggregate, be met and many countries will fall far short. The MDGs seem set to pass into history as another set of missed development targets. This chapter looks at the soundness of these projections. We discuss briefly the basis for making such projections and then examine the findings for the major MDG targets: income poverty, mortality, education, nutrition and HIV/AIDS.

Approaches to making projections1 Projections may vary because of differences in assumptions regarding (i) determinants of the outcome of interest, (ii) model parameters or (iii) future values of the determinants. The simplest models, and by far the most common approach, take time as the only determinant – that is, the future is forecast based on historical trends, an approach called here naïve projections. Naïve projections tell us if a country is on track or not to meet the relevant goal, which is certainly of policy interest, but are of less use in predicting actual expected values at some point in the future unless its determinants really are uncertain or unforecastable. The next most common approach is to use an outcome–income elasticity to base the forecast on projections of economic growth, the 20

Projecting Progress

21

latter usually being taken from some other source, such as the World Bank’s Global Economic Prospects. Since income is highly correlated with most of the outcomes of interest, this approach can be expected to give a reasonable first approximation. However the correlation with income is imperfect, so adding more variables to the right-hand side of the determinants equation will help, at least if the future values of these determinants can be predicted with any degree of confidence. More sophisticated versions utilize multi-equation models. There are undoubted advantages to such models, which allow a wider range of policy simulations, but they are quite resource-intensive, the assumptions more hidden than is the case for simpler projections. Results from the most recent multi-equation model, the World Bank Maquette for MDG Simulations (MAMS), were not available at the time of writing, and such approaches are not the basis of the projections presented here. However, models allow the flexibility to vary the underlying assumptions and so conduct policy experiments. Such an approach should be of interest to policy-makers, with the caveat that many are wary of CGEs because of their apparent complexity. This is no reason not to use them, but is a reason to be as explicit about the assumptions as possible. Some indicators have risen in recent years at a far greater rate than their historical precedent – primary enrolment being the most obvious case. In these cases, if naïve projections are based on very recent trends, then such forecasts may be superior to those which are model-based. But naïve projections will underestimate future growth if the period used to calculate expected growth includes many years prior to any recent expansion. The main point is that policy can make a difference to the achievement of goals – as shown by the experience of countries that do not follow the expected outcome–income trajectory (see World Bank 2005b for a discussion of Bangladesh, which has been remarkably successful in reducing fertility and mortality faster than income growth would suggest). Were the right policies to be implemented, then all future outcomes would be better than suggested here, though this of course begs the question as to what the ‘right policies’ are. A further source in variation of projections is the level of disaggregation at which they are made. Although most sources report results at the regional level, it is preferable that these regional aggregates be based on country-level projections using country-specific data and parameters. This is not always the case. All methods are of course dependent on the quality of data used. Data availability has improved greatly in the last two decades, though most outcomes are not available on an annual basis.2 There are also exceptions to the general increase in data availability, such

22 Howard White and Nina Blöndal

as maternal mortality and literacy, for which data are less reliable. Moreover, even where data are good, there are frequently groups missed from the data, notably those not in fixed households (nomads, street children, those in institutions, and so on). Basis of own projections Our own projections are made at the country level and are reported using either the UN or the World Bank regional classifications. The baseline data (that is, pre-projection) are from the UN Millennium Indicators Database, supplemented by the World Bank’s World Development Indicators and Demographic and Health Survey (DHS) data as necessary. Population data are from the UN Population Database, 2004 Revision. Projections are made using one of three methods: (i) naïve projections, (ii) growth-based estimates, and (iii) model-based estimates (that is, growth plus other factors). The latter two approaches use differential elasticities to capture the greater poverty-reducing effect of growth in low-inequality settings. Rural–urban rates are based on countryspecific estimates of the rural–urban differential. Specific details of the projections are as follows:

• Income poverty is modelled as a function of income growth alone,

•

• • •

with the elasticity varying according the level of initial inequality. No allowance is made for the fact that the elasticity also changes over time; while this would be a more realistic approach, too little information is available to make such estimates in a sufficiently informed manner. Under-5 mortality is modelled as a function of income poverty (elasticity = 0.4) and female enrolments (elasticity = −1.3) and an autonomous decline of 1.5 per cent per annum. These elasticities are taken from a review of papers on the cross-country determinants of under-5 mortality (for example Hanmer et al. 2003). Undernutrition is modelled as a function of income alone, with the elasticity again varying by the level of inequality. Net enrolments are based on naïve projections, made separately for male and female, capped at 98 and the female rate capped at the male rate.3 Literacy is calculated in two ways: (i) naïve projections, done separately for male and female, and capped at 99.5, and (ii) an accounting approach based on age structure and net enrolments, removing older, less-literate people from the population and adding those graduating

Projecting Progress

23

from school who are assumed to be literate. Again the data are capped at 99.5. • Separate rural and urban estimates are made by using the ratio of rural to urban incidence, based on observed values where available and modelled (as a function of poverty, urbanization, education and regional dummies) when not available.

Income poverty Early estimates of progress towards reducing income poverty come from studies conducted by the Institute of Social Studies for the Sida 2015 project (hereinafter the ISS study, Hanmer et al. 1997a, 1997b), and work undertaken by the Overseas Development Institute (ODI).4 These two studies made forecasts based on estimates of the elasticity of poverty with respect to income, allowing the elasticity to vary according to the degree of inequality.5 The World Bank has since begun publishing annual estimates of income poverty in 2015 in Global Economic Prospects. These estimates are the same as those to be found in the Bank’s MDG Global Monitoring Report. The methodology used in these studies is not public, but appears to be based on household data rather than cross-country elasticities. Elasticities based on household data are generally higher than those from cross-country regressions, so the estimates of future numbers of poor will be lower than if elasticities estimated using cross-country data are used. Our own estimates follow the ODI–Sida approach applied on a countryby-country basis. Projected growth rates are based on regional averages from Global Economic Prospects, with a few exceptions. The poverty elasticity is assumed to vary according to initial inequality.6 To calculate regional figures of absolute numbers of poor, the total for the region is scaled up to capture the population not covered by the data. Table 2.1 summarizes the various results. The most recent World Bank estimates present a more positive picture than do the earlier studies, except with respect to sub-Saharan Africa (SSA). Our own estimates give figures more similar to those of the earlier studies. In general this will be because we assume lower elasticities than those used by the World Bank, especially for high-inequality countries, so the rate of poverty reduction as a consequence of growth is less. However, in the case of Africa, our estimates are more positive. This discrepancy is most likely caused by the World Bank using recent country-specific poverty estimates not available to us. Discrepancies also arise from both differing assumed growth rates and data revisions to poverty estimates in the base year (1990). More

24 Howard White and Nina Blöndal Table 2.1 Income poverty forecasts (% living below $1 a day) for 2015 from three studies

Sub-Saharan Africa Middle East and North Africa East Asia and Pacific South Asia Latin America and Caribbean Eastern Europe and CIS

Sida 2015

Hanmer and Naschold

World Bank (2005)

Own estimates

32.4 3.4 7.2 18.6 14.3 2.7

33.2 1.9 10.1 23.3 15.7 3.9

38.4 0.9 0.9 12.8 6.9 0.4

30.0 4.2 8.4 16.8 16.1 4.3

Source: See text.

Share of global poverty (%)

60 50 40 30 20 10 0

East Asia and Pacific

Europe and Central Asia

1990

Latin America and Caribbean

Middle East and North Africa

2001

2015

South Asia

Sub-Saharan Africa

Figure 2.1 Share of global poverty (proportion of poor in each region using $US1 a day poverty line) Source: Calculated from World Bank Global Economic Prospects data.

recent growth forecasts are generally more positive than the older ones7 and estimates of poverty in 1990 have been revised downward, especially for Latin America and the Caribbean (LAC). Income poverty will be concentrated increasingly in Africa (Figure 2.1), with absolute numbers of poor under the $US1 a day standard in Africa overtaking South Asia at around the current time. If a $US2 a day poverty line is used then South Asia will continue to have the bulk of the world’s income-poor in 2015 (46 per cent), followed by SSA (31 per cent).

Projecting Progress

25

Subregional estimates Subregional figures (Table 2.2) are based on our own income poverty estimates. The total number of poor in 2015 (736 million) sits well within the range of earlier World Bank estimates, though it is higher than their current projection. The headcount ratio is highest in West and East Africa. Central Africa appears better off than Central and South America but it should be recalled that most of the African countries do not have data, especially those countries affected by conflict, where poverty will be high. These patterns are also evident in the map, although the data limitations also show up there (Figure 2.2). As shown in Table 2.2, the reasonable progress in reducing income poverty is not usually sufficient to meet the MDG of halving poverty by 2015. Of the 71 countries for which there are data, less than a third (21) are expected to achieve the goal (using our own growth-based projections), with Asian countries accounting for a disproportionate share of success. India, China, Indonesia, Bangladesh and Vietnam are all predicted to meet the target. Together these five countries will account for close to half (46 per cent) of the population of the developing world in 2015. Hence expressed in terms of the world’s population the prospects appear better (about half will reach the goal, which of course means that half will not) than when expressed in terms of countries (over two-thirds will not meet the target). Rural versus urban poverty The poverty rate is generally higher in rural areas than in urban ones. Data were available on rural and urban poverty rates using the national poverty line for 56 countries for various years in the 1990s. The mean ratio of numbers of rural and urban poor was 1.86 (and the median 1.53). For only 7 of the 56 countries was urban poverty higher than rural, compared with 18 in which it was over twice as high in rural areas. In order to generate urban and rural poverty estimates for a larger number of countries, this ratio was imputed based on a regression of determinants.8 The estimates are made at the country level, using (i) mostly regional growth rates with some country-specific adjustments, and (ii) elasticities which vary according to the degree of initial inequality. The main finding is that in 2015 two-thirds of the world’s income-poor will be in rural areas. The majority of the populations in Africa and Asia will be rural until after 2020. Given the higher poverty rate in rural areas (16 per cent globally compared with 7 per cent in urban areas), it follows that these areas must account for a disproportionate number of the poor (Table 2.3).

26 Table 2.2 Subregional income poverty estimates 2015 ($1 a day) Progress in halving income poverty

Income poverty in 2015

No. of countries Headcount Population Contribution 2015 expected ratio Number share to poverty headcount to meet (% poor) (millions) (%) (%) as % 19901 goal2 East Africa Central Africa North Africa Southern Africa West Africa Eastern Asia South-Central Asia South-East Asia Western Asia Eastern Europe Northern Europe Southern Europe Caribbean Central America South America Total

33.8 7.5 1.4 11.3 37.5 6.1 14.5

122 11 3 6 124 98 271

4.6 3.3 3.0 1.0 1.0 1.7 9.7 8.5 10.2

29 9 9 1 2 1 16 36 736.3

5.0 2.0 3.1 0.8 4.6 22.2 25.8

16.6 1.5 0.4 0.8 16.8 13.3 36.8

84 – 64 89 78 21 37

0/7 0/0 0/3 1/5 1/9 1/1 6/9

8.6 3.6 3.9 1.4 2.1 0.6 2.3 5.9 100.0

3.9 1.2 1.2 0.1 0.2 0.1 2.2 4.9 100.0

29 141 74 25 52 55 51 97 –

4/6 0/2 2/9 1/2 – 1/3 2/5 2/10 21/71

Notes: 1. Calculated using only countries for which estimates available for both 1990 and 2015. 2. Second figure is the number of countries for which early 1990s income poverty data were available. (Data availability for this column is less than that used for poverty projections, as more recent poverty data were usually used for the projections.)

Table 2.3 Rural and urban poverty in 2015 headcount ratio and millions of people ($1 a day)

Sub-Saharan Africa Middle East and North Africa Eastern Europe and Central Asia East Asia and Pacific South Asia Latin America and Caribbean Total

Headcount ratio

Millions of people

Total

Urban

Rural

Total

Urban

Rural

% Rural

30.0 2.8

21.6 1.6

37.4 4.2

282 10

86 4

196 6

69.6 63.6

3.2

2.5

4.3

15

8

7

47.9

5.7 14.8 8.7

3.1 10.6 7.0

8.4 16.8 16.1

120 255 54

33 59 35

87 196 19

72.5 76.8 35.3

11.8

7.4

16.3

737

225

512

69.5

Percent of population ⬍ 4.3 ⬎ 4.3 to 10.3 ⬎ 10.3 to 24.4 ⬎ 24.4 Not available

Figure 2.2

Income poverty 2015 (growth-based projections, differential elasticities)

27

28 Howard White and Nina Blöndal

Summary Recent estimates show that Africa is overtaking South Asia as the region with the largest number of those living on less than a dollar a day, and by 2015 half of those below this poverty line globally will be in that region. The prospects for most of Africa meeting the MDG of halving income poverty seem remote. These trends reflect both the region’s poorer growth performance and the fact that it has high and growing levels of inequality. By contrast the share of East Asia in world poverty is falling rapidly and will be only 3 per cent by 2015; this region will meet the MDG as will other parts of Asia. Latin American countries will be close to achieving the goal, but excluding several in South America. If a higher poverty line of $US2 a day is used then the global poverty profile shifts, with South Asia again having most of the world’s poor, followed by Africa. Income poverty is higher in rural areas than urban ones. In 2015, 70 per cent of the income-poor will live in rural areas.

Mortality Different estimates Hanmer and Naschold (2000) provide regression-based estimates of infant and under-5 mortality allowing for income, HIV/AIDS, education and health services.9 However, there have been substantial upward revisions to mortality estimates in the base year (1990) since they carried out their analysis, so it is to be expected that their results should have a downward bias. Indeed, the figures are lower than those available from other sources. Using data from the World Bank Global Monitoring Report, we made our own naïve projections at the regional level, and then adjusted these to allow for expected higher future growth. We also made country-specific estimates using a model based on poverty levels and female primary enrolment. The final estimates are those from the UN World Population Prospects, 2004 revision. These estimates are based on demographic trends, without reference to economic conditions, but appear nonetheless broadly consistent with the growth-adjusted estimates we made. However, the model-based estimates show a stronger reduction in mortality, notably in SSA and South Asia, this discrepancy being driven by the rapid growth of female enrolments in some countries and the fact that the model does not allow for additional AIDS-related deaths (Table 2.4).

Projecting Progress Table 2.4

29

Mortality estimates 2015 (rates per 1000) Own projections Hanmer and Naïve Growth Model UN population Naschold projection adjusted based projections

Under-5 mortality Sub-Saharan Africa Middle East and North Africa East Asia and Pacific South Asia Latin America and Caribbean Eastern Europe and Central Asia

118 34 8 55 15 6

168 42 30 78 25 29

138 30 32 75 10 30

97 31 27 50 24 25

130 28 30 69 22 30

Infant mortality Sub-Saharan Africa Middle East and North Africa East Asia and Pacific South Asia Latin America and Caribbean Eastern Europe and Central Asia

67 22 7 43 14 10

96 53 24 54 19 25

89 28 27 56 29 31

89 28 27 56 29 31

80 25 25 51 17 6

Note: Own projections are at regional level for model-based. Source: See text.

Subregional and country data Table 2.5 shows the subregional figures using the UN Population projection estimates. Rates are highest in West and Central Africa. These are both areas in which tropical diseases remain important and so have unusually high ratios of child to infant deaths. But they are also areas in which conflict has adversely affected mortality. The maps show the scar of premature death running across Africa (Figure 2.3). Some trends All three major developing country regions have seen declining under-5 mortality, though it has been more rapid in Asia and LAC than Africa (Figure 2.4). Indeed, the rate of reduction in Africa has been insufficient to keep up with population growth, so the number of deaths is continuing to climb. As a result, Africa’s share of under-5 deaths is growing. In the case of infant deaths this trend represents a dramatic reversal. In the 1960s three-quarters of all infant deaths were in Asia, but by 2015 Africa will account for the majority, as it has done for child deaths for some years now. However, as with income poverty reduction, the rate of decline is rarely sufficient to meet the ambitious goal of a two-thirds reduction by 2015. Our own projections suggest that only five countries will

Table 2.5

Subregional estimates of infant and under-5 mortality, 2015 Progress toward reducing Under-5 mortality (U5M) in 2015 U5M by two-thirds

IMR Number U5M Number (per 000) (millions) Share (per 000) (millions) Share East Africa Central Africa North Africa Southern Africa West Africa Eastern Asia South-Central Asia South-Eastern Asia Western Asia Eastern Europe Northern Europe Southern Europe Western Europe Caribbean Central America South America Northern America Australia/New Zealand Melanesia Micronesia Polynesia

71 95 28 29 93 23 50 26 27 12 4 6 4 27 14 18 6 4 46 35 17

0.9 0.6 0.1 0.0 1.1 0.5 1.9 0.3 0.2 0.0 0.0 0.0 0.0 0.0 0.0 0.1 0.0 0.0 0.0 0.0 0.0

15.8 10.2 2.3 0.5 18.8 7.7 32.7 4.6 2.6 0.5 0.1 0.1 0.1 0.6 0.7 2.0 0.5 0.0 0.2 0.0 0.0

113 162 37 47 149 27 67 32 32 15 6 7 5 39 18 22 5 5 n.a. 8 14

1.3 0.8 0.2 0.1 1.5 0.5 2.5 0.3 0.2 0.0 0.0 0.0 0.0 0.0 0.1 0.2 0.0 0.0 0.0 0.0 0.0

16.9 10.4 2.6 1.3 19.5 6.5 32.5 3.9 2.6 0.0 0.0 0.0 0.0 0.0 1.3 2.6 0.0 0.0 0.0 0.0 0.0

Total

44

6.0

100.0

60

7.8

100.0

Source: UN Population Projections 2004.

U5M in 2015 Number of as a per cent of countries expected rate in 1990 to meet goal 65 82 39 73 69 58 55 41 47 71 – – – 52 36 42 50 51 68 26 33

0/16 0/9 0/6 0/5 0/17 0/5 0/14 3/11 0/18 2/10 – – – 0/12 0/8 0/12 – – – – – 5/143

30

Infant mortality in 2015

Deaths per 1,000 live births ⬍ 27 ⬎ 27 to 64 ⬎ 64 to 118 ⬎ 118 Not available

Figure 2.3

Under-5 mortality 2015 (UN Population Projection, 2004 Revision)

31

Infant mortality rate

32 Howard White and Nina Blöndal 200 180 160 140 120 100 80 60 40 20 0 1950

1970

No. of deaths (millions)

2010 Asia

2030 LAC

3.0

12

2.5

10

2.0

8

Asia

2025

2035

2045

LAC

0.5

2

(e)

2015

1.0

4

90 80 70 60 50 40 30 20 10 0 1940

2005 Africa

1.5

6

0 1940

1995

(b)

14

1960

1980 Africa

(c)

Share of deaths

1990

Africa

(a)

90 80 70 60 50 40 30 20 10 0 1985

2000

2020

Asia

2040

2060

LAC

0.0 1985

1995

2005 Africa

(d)

2015 Asia

2025

2035

2045

2035

2045

LAC

80 70 60 50

1960

1980 Africa

2000 Asia

2020

2040 LAC

2060

40 30 20 10 0 1985

1995

(f)

2005 Africa

2015 Asia

2025 LAC

Figure 2.4 Mortality trends: (a) infant mortality rate; (b) child mortality rate; (c) number of infant deaths; (d) number of child deaths; (e) regional shares of infant deaths (per cent); (f) regional shares of child deaths (per cent) Source: Based on UN Population Projections 2004 revision.

do so: Indonesia and Vietnam are the only two of significance (the others are Montserrat and Vanuatu; outside the sample, the Czech Republic, Moldova and Portugal will also do so). The UN Population projections are more optimistic, showing 10 countries expected to reach the target, with Egypt, Syria and Tunisia being among those added to the list. Rural and urban differences Mortality is higher in rural areas than urban ones in virtually all countries. Data from the last 15 years from 126 Demographic and Health Surveys (DHS) were examined. In only eight countries was infant mortality lower in rural areas than urban ones: on average the infant mortality

Projecting Progress Table 2.6

33

Rural urban differentials in under-5 mortality, 2015

Sub-Saharan Africa Middle East and North Africa Europe and Central Asia East Asia and Pacific South Asia Latin America and Caribbean Total

U5M rate (per 000)

Number (millions)

Urban

Rural

Total

Urban

Rural

Total

81 24 21 19 36 22 37

108 39 32 35 57 36 63

97 31 25 27 50 24 51

4.9 0.6 0.4 1.4 2.0 0.9 10.3

9.2 0.6 0.3 2.5 6.7 0.4 19.8

14.1 1.2 0.7 3.9 8.7 1.3 30.0

rate (IMR) was just over a third higher in rural areas, though the ratio was over 2 for Peru. The contrast is stronger for child mortality, with the average ratio being 1.7 (median 1.6); in only 5 cases was rural mortality lower than that in urban areas, and rural mortality was double that in urban areas in 30 cases, reaching a ratio of 4.9 in Armenia. There are no clear regional patterns in these differentials. For simulation purposes, actual values of the ratio for under-5 mortality are used for those countries covered by the survey, with remaining countries set at the sample median (1.4).10 The results are shown in Table 2.6 (applied to the modelbased estimates). Of the 30 million under-5 deaths in 2015, 20 million (that is, two-thirds) will take place in rural areas. Water supply Lack of access to water will be a largely rural phenomenon by 2015: three-quarters of the 761 million without access to water will be in rural areas. Lack of access will be concentrated in East Asia and SSA, which will together account for 80 per cent of those without access (Table 2.7). Summary All three major developing country regions have seen declining mortality, though it has been more rapid in Asia and LAC than in Africa. But only in rare cases has the rate of reduction been at the 1.6 per cent a year required to achieve the MDG of a two-thirds reduction by 2015. Indeed, the rate of reduction in Africa has been insufficient to keep up with population growth, so the number of deaths is continuing to climb. As a result, Africa’s share of under-5 deaths will continue to grow. In the case of infant deaths this trend represents a dramatic reversal. In the 1960s

34 Howard White and Nina Blöndal Table 2.7 Access to water in 2015

Access to water (% of population)

Share of total number Urban Rural Total Urban Rural Total globally Sub-Saharan Africa Middle East and N. Africa Europe and Central Asia East Asia and Pacific South Asia L. America and Caribbean Total

Number without access (millions)

Share of those without access in rural areas

82.2 96.0

57.6 79.4

69.3 89.7

72 9

227 30

288 37

38 5

79 80

99.4

83.2

94.1

2

28

29

4

98

86.3 95.7 98.0

78.2 94.9 85.4

85.4 95.6 96.5

147 24 10

225 59 18

308 77 22

41 10 3

73 77 81

91.4

81.5

87.8

264

586

761

100

77

three-quarters of all infant deaths were in Asia, but by 2015 Africa will account for the majority, as it has done for child deaths for some years now, with the highest rates in Western and Central Africa. Two-thirds of these deaths will be in the rural areas of Africa, which is a similar share of under-5 deaths in rural areas across the world. Similar patterns reveal themselves with respect to critical determinants of under-5 mortality: access to water and immunization. For example, West and Central Africa are also the two subregions with the lowest access to water expected in 2015. Over three-quarters of those without water in 2015 will be in rural areas. Immunization coverage fell in some countries in the 1990s and there remain a significant minority of countries with unacceptably low coverage rates. Falling under-5 mortality, other health improvements which increase longevity and the demographic transition are increasing the share of the elderly in the population of all regions. This trend will in time create new problems for developing countries.

Education The Millennium Project Task Force on Education (MPTFE) is the only study to estimate primary net enrolment rates.11 Projections are made based on predictive power of the s-curve (that is, that later enrolments are harder to achieve) and suggest that SSA will continue to lag behind

Projecting Progress Table 2.8

35

Net primary enrolment rate, 2015 (proportion of age cohort enrolled)

Sub-Saharan Africa South Asia Middle East and North Africa Latin America and Caribbean Eastern Europe and Central Asia East Asia and Pacific

MPTFE

Own projections

79.6 86.1 88.0 95.6 96.3 97.0

73.9 92.4 89.3 98.0 91.3 90.8

Source: Millennium Project Task Force on Education (MPTFE) and own projections.

other regions with an estimated enrolment rate of just under 80 per cent. South Asia and the Middle East and North Africa are also estimated to lie below 90 per cent while other regions are expected to lie within 5 per cent of universal enrolment. This ‘last 5 per cent’ is increasingly recognized as being ‘problem groups’ requiring different policies. Our own countrylevel naïve projections do not allow for this effect, which may explain the higher figures in South Asia and LAC. Naïve projections seem appropriate since the policy push behind primary enrolments has enabled increases over and above those expected by economic performance or other possible determinants. It is the possibility of such a push in Africa which explains the higher rates there estimated by MPTFE. However, India began a very recent push, not yet shown in the data and so not picked up by naïve projections (Table 2.8). The subregional breakdown shows the problem in Africa to be least in Southern Africa (Table 2.9), which is confirmed partially by the map, but it is patchy owing to limited data (Figure 2.5). Defining universal primary enrolment as a net enrolment rate (NER) of 98 per cent shows that the majority of countries appear on track to achieve this goal. Breaking down our own projections we see that SSA will account for nearly half of all the 80 million children out of primary school by 2015 (Table 2.10), reflecting the lower enrolment rates across the subcontinent (Figure 2.5). Gender dimension Areas with low enrolments have a disproportionate number of girls out of school (see Table 2.10). The MPTFE study also provides estimates of gender equality in primary education measured by female-to-male net primary enrolment. Sub-Saharan Africa is estimated to have the lowest ratio, with 93 girls to 100 boys (Table 2.11).

36 Howard White and Nina Blöndal Table 2.9 Subregional primary enrolment rates (NER) and numbers out of school, 2015 (millions) No. of countries with UPE No. NER by 2015 East Africa 13.9 76.5 Central Africa 5.6 77.3 North Africa 4.2 84.8 Southern Africa 0.6 90.5 West Africa 15.4 70.8 Eastern Asia 9.3 91.8 South-Central 19.1 91.2 Asia South-Eastern Asia 6.3 90.2 Western Asia 2.0 93.6 Eastern Europe 0.8 95.2 Northern Europe 0.2 97.6

9/17 7/9 6/7 1/5 12/18 4/7 10/14 7/11 13/18 5/10 11/13

No. of countries with UPE No. NER by 2015 Southern Europe 0.3 Western Europe 0.2 Caribbean 0.1 Central America 0.4 South America 0.9 North America 0.6 Australia/ 0.0 New Zealand Melanesia 0.2 Micronesia 0.0 Polynesia 0.0 Total 80.0

96.3 98.0 98.0 98.0 98.0 98.0 98.0

5/10 – 11/13 24/24 8/8 – –

85.8 – 98.0 – 98.0 – 89.1 182/230

Notes: NER = net enrolment rates; UPE = universal primary education.

Literacy No estimates of literacy were located. Country-level projections were made for this report using two methods: (i) naïve projections, and (ii) an accounting approach based on the net enrolment rate.12 The two approaches are not entirely consistent, with naïve projections predicting lower literacy. This is because recent rapid increases in enrolments are not yet reflected in higher literacy, but the accounting approach picks up these increases. However, both methods find that the vast majority of illiterates in 2015 will be in South Asia, with the majority of the remainder in SSA (Table 2.12). Summary Sub-Saharan Africa will continue to lag behind other regions with respect to primary-school enrolments, with an estimated enrolment rate of just under 80 per cent, though the problem is less in Southern Africa than elsewhere. South Asia and the Middle East and North Africa are also estimated to lie below 90 per cent while other regions are expected to lie within 5 per cent of universal enrolment. However, this ‘last 5 per cent’ is increasingly recognized as consisting of ‘problem groups’ requiring different policies.

Percent ⬍ 65 ⬎ 65 to 85 ⬎ 85 to 95 ⬎ 95 Not available

Figure 2.5

Net enrolment rates 2015 (naïve projections)

37

38 Table 2.10 Numbers of children out of school, 2015 (millions) Proportion Share of total Boys Girls Total girls out of school Sub-Saharan Africa 17.9 21.4 Middle East and North Africa 2.4 2.4 Eastern Europe and Central Asia 1.3 1.8 East Asia and Pacific 9.5 5.9 South Asia 6.8 8.5 Upper middle income 0.7 0.7 High income 0.7 0.7 Total 39.2 41.4

39.3 4.7 3.1 15.4 15.3 1.3 1.4 80.6

54.5 49.8 59.0 38.4 55.7 50.0 50.0 51.4

48.8 5.9 3.9 19.1 19.0 1.7 1.7 100.0

Table 2.11 Gender equality in education, female/male ratio, 2015

Region South Asia Sub-Saharan Africa Latin America and Caribbean Europe and Central Asia East Asia and Pacific Middle East and North Africa

MPTF on education (primary)

MPTF on gender (primary, gross)

Own estimates (primary)

95.3 93.2 99.1 99.3 98.8 96.0

111.0 94.6 96.2 98.3 98.4 100.0

99.6 98.8 100.0 99.2 101.1 100.0

Note: MPTF = Millennium Project Task Force. Source: UN Millennium Project (2005b), and own estimates.

Table 2.12 Literacy in 2015 by region Naïve projections

Accounting based

No. of Literacy No. of illiterates Literacy illiterates rate (millions) rate (millions) Sub-Saharan Africa Middle East and North Africa Europe and Central Asia East Asia and Pacific South Asia Latin America and Caribbean Developed countries Total

81.7 86.6 99.1 99.1 68.1 98.1 99.1 89.6

99 33 4 16 387 9 8 555

93.7 91.9 99.3 99.3 80.9 99.4 96.8 93.8

34 20 3 11 232 3 26 329

Projecting Progress

39

Table 2.13 Alternative estimates of underweight Children Adults Smith and FAO (2015) Haddad (2020) Onis et al. (2015) Prevalence South Asia Sub-Saharan Africa East Asia Near East and North Africa Latin America and Caribbean

12 23 6 7 6

37.4 28.8 12.8 5 1.9

26.2 29.2 3.0 7.4 3.4

Numbers South Asia Sub-Saharan Africa East Asia Near East and North Africa Latin America and Caribbean Developing countries

195 205 135 37 40 612

66 48.7 21.4 3.2 1.1 140.4

61.8 42.7 3.0 3.4 2.0 112.9

Shares South Asia Sub-Saharan Africa East Asia Near East and North Africa Latin America and Caribbean Developing countries

31.9 33.5 22.1 6.0 6.5 100.0

47.0 34.7 15.2 2.3 0.8 100.0

54.7 37.8 2.7 3.0 1.8 100.0

Source: See text.

Nutrition Different estimates Nutrition outcomes have been comprehensively modelled in three different sources: two IFPRI studies (Smith and Haddad 2000; von Braun et al. 2005) and FAO’s agricultural projections (FAO 2002).13 In addition Onis et al. (2004) present subregional-based naïve projections. The FAO study refers to adults who are undernourished (defined with reference to a calorie requirement of 1900 kcal per day), whereas the IFPRI studies and Onis et al. refer to children only (using anthropometric measurement). The IFPRI results shown forecast for 2020 rather than for 2015 (Table 2.13). The different estimates do not, in this case, give a similar picture. Smith and Haddad’s estimates are in general higher than those of Onis, and there is a marked discrepancy in the case of East Asia – which Smith and

350 Millions underweight

40 35 30 25

20 15 10 5 0 (c)

East Asia

(b)

South Asia

East Asia

South Asia

Latin America and Caribbean

Near East/ North Africa

5

Latin America and Caribbean

10

40 35 30 25 20 15 10 5 0

Sub-Saharan Africa

15

0 (a)

Per cent underweight

Share of undernutrition

20

Sub-Saharan Africa

Per cent underweight

25

Near East/ North Africa

40 Howard White and Nina Blöndal

300 250 200 150 100 50 0

1990/92

1997/92

Sub-Saharan Africa LAC East Asa

2015

2030

North-East North Africa South Asia

(d)

1990/92

1997/99

Sub-Saharan Africa LAC East Asa

2015

2030

North-East North Africa South Asia

Figure 2.6 Patterns and trends from FAO data: (a) prevalence of adult underweight, 2015; (b) share of adult underweight, 2015; (c) prevalence of adult underweight, 1990–2030; (d) number of adult underweight, 1990–2030

Haddad have as accounting for 15 per cent of children underweight by 2020, compared with Onis’s 3 per cent in 2015. While both have South Asia’s share as exceeding that of Africa, Smith and Haddad predict higher prevalence in South Asia than Africa, but Onis the reverse. FAO forecasts for adults give the two regions an equal share. The different studies do give broadly similar results for prevalence in SSA, as does another recent IFPRI study which suggests prevalence of child malnutrition to be 30.4 per cent in 2015 by their baseline ‘business as usual’ scenario (Rosegrant et al. 2005). However, the point of the study is to show how different policy interventions – such as rural infrastructure and investing in agricultural research – can make an enormous difference to nutritional outcomes. The different scenarios place the number of malnourished children in 2025 at between 9.4 million and 55.1 million.

Some trends Figure 2.6 shows trends from the FAO data. As mentioned, the shares of Africa and Asia are about the same in 2015, reflecting a growing number

Projecting Progress Table 2.14

41

Per capita food consumption (calories per person per day) 1964–6

1997–9

2015

2054 2058 2290 2393 2017 1957

2681 2195 3006 2824 2403 2921

2850 2360 3090 2980 2700 3060

Developing countries Sub-Saharan Africa Middle East and North Africa Latin America and the Caribbean South Asia East Asia

Index of agricultural production

Source: FAOSTAT; see http://faostat.fao.org/.

140 120 100 80 60 40 20 0 1960

1970

1980

1990

Sub-Saharan Africa

East Asia

LAC

South Asia

2000

Figure 2.7 Trends in agricultural output per person (2000 = 100) Source: Calculated from FAOSTAT; http://faostat.fao.org/.

of undernourished people in Africa, compared with a steady decline in South Asia. These trends reflect differing trends in food availability. The FAO report data on food consumption, based on calculations of food production and the net trade balance. Food consumption in Africa is set to rise slowly, falling further behind that of other regions (Table 2.14). This trend reflects the region’s dismal agricultural performance. In contrast to other regions which have experienced rising agricultural productivity since the 1960s, that in Africa has declined (Figure 2.7). Inequality The FAO (2002) report graphs differing levels of malnutrition for three different distributions against average food consumption. With high

42 Howard White and Nina Blöndal 40

6.0

35 30

4.0

Millions

Prevalence (%)

5.0

3.0 2.0

25 20 15 10

1.0

5

0.0 (a) 1980

1990

Baseline

2000 Optimistic

2010

2020 Pessimistic

0 1980

(b)

1990 Baseline

2000 Optimistic

2010

2020 Pessimistic

Figure 2.8 HIV/AIDS three scenarios for Africa: (a) adult HIV prevalence in Africa by three different scenarios; (b) number of adults living with HIV/AIDS in Africa Source: UNAIDS (2005).

inequality, which is experienced in SSA, the region’s average food availability yields undernutrition of 23 per cent. But if the region were to have low inequality then, with the same level of food availability, undernutrition would be just 10 per cent.

HIV/AIDS Projections related to diseases are plagued by additional uncertainty regarding both current prevalence rates, the appropriate models to be used for prediction purposes and the extent to which effective action will be taken to combat the disease. Based on current trends, UNFPA (no date) estimate that there will be 46 million new HIV infections by 2010, but only a third that number (17 million) with an appropriate policy response. The three scenarios in a recent UNAIDS (2005) report on AIDS in Africa show that the current figure of 25 million adults living with HIV/AIDS may either fall to 14 million by 2025 or rise to 38 million – nearly two-thirds more than the optimistic estimate (Figure 2.8). Even the most pessimistic estimates suggest nonetheless that prevalence will decline, reflecting the belief that within 15 years (by around 2020) there will be no new AIDS cases. As AIDS in Africa is brought somewhat under control, attention is shifting to Asia, where the epidemic is in its early stages: the number of AIDS-related deaths in Asia will quadruple in the next 15 to 20 years, compared with an increase of about a third in Africa. But, despite its larger population share, Asia will never account for even half of all deaths, a dubious honour that remains with Africa into the longer term

Projecting Progress

43

Table 2.15 Number of HIV/AIDS-related deaths 1990–5

2000–5

2010–15

Absolute number of excess deaths from HIV/AIDS (000s) Sub-Saharan Africa 3,216 14,807 18,933 Asia 1,314 3,461 10,872 LAC 201 697 774 More developed 427 789 1,692 Total 5,158 19,754 32,271 Share of excess deaths (per cent) Sub-Saharan Africa 62.3 Asia 25.5 LAC 3.9 More developed 8.3 Total 100.0

75.0 17.5 3.5 4.0 100.0

2020–5

2045–50

18,585 17,078 730 1,557 37,950

7,510 4,821 37 49 12,417

58.7 33.7 2.4 5.2 100.0

49.0 45.0 1.9 4.1 100.0

60.5 38.8 0.3 0.4 100.0

Note: More developed are Russia and USA. Source: UN Population Division.

(Table 2.15).14 Africa’s share is likely understated in this table given the efficacy with which Asian countries have been tackling the epidemic, and the continuing opportunity for some of these countries to head it off, compared with the more mixed picture in Africa. Those data that are available, though not systematic, suggest that the incidence of HIV/AIDS is higher in urban areas,15 and many affected rural residents go to urban areas for treatment. However, since most of the population in Africa and Asia are rural it is possible that the absolute number of HIV/AIDS cases is higher in rural areas than urban ones.

Conclusions There has been substantial progress in poverty reduction . . . The second half of the last century witnessed remarkable gains in the reduction of many forms of poverty. Mortality fell, and life expectancy rose, across the developing world at historically unprecedented rates – much faster than had been achieved in the now developed countries. With the exception of countries affected by HIV/AIDS, under-5 mortality has continued to decline in most countries even in times of economic stagnation or decline, thanks to increased immunization coverage, access to safe water and so on. The last decade has seen many countries adopt programmes to ensure universal primary education.

44 Howard White and Nina Blöndal

. . . and this progress will continue, though not fast enough to achieve the MDGs . . . All projections suggest that these gains will continue into the new millennium, although not usually fast enough to achieve the ambitious targets set by the MDGs. Neither the goal for income poverty reduction nor that for lower mortality will be met in the vast majority of countries. Attaining universal primary education is the one area where the goal looks achievable in many countries, though by no means all. . . . so there will still be substantial poverty . . . Even where these goals are achieved substantial poverty will remain – the aim is only to halve poverty, not eliminate it. Current projects suggest there will be around three-quarters of a billion people living on less than a $1 a day in 2015. In absolute terms the number of children dying prematurely in Africa will continue to rise for some years to come. . . . of which an increasing share will be in Africa Progress has been, and will continue to be, slowest in Africa, opening up a growing gap between the region and the rest of the developing world. South Asia, formerly the poorest region, continues to have substantial numbers of undernourished children, and the most children out of school (though recent efforts in India are not captured in the data). And while more poverty will be in urban areas, it will still be predominately rural Urban areas are growing rapidly, and with them slums. So poverty will become more and more an urban issue. But, other than HIV/AIDS, poverty indicators are worse in rural areas in virtually all countries. Since rural residents will remain the majority of the world’s population, the bulk of the poor (60–70 per cent, depending on the indicator) will still be in rural areas in 2015. The nature of problems changes . . . As poverty falls so does its character. The policies required when the bulk of the population are poor differ from those needed when only a minority are affected, especially as many of these remaining poor may share common characteristics (remote, ethnic minority, nomadic, and so on). As under-5 mortality falls it becomes more and more concentrated in the first months or even days, of life. General socioeconomic development and public health interventions can reduce high mortality rates, but once they are lower the required policies become more medically

Projecting Progress

45

intensive. ‘The last 5 per cent’ not in school – children of nomads, parents who think girls should not go to school, street children – are harder to reach, with simply providing schools not being enough. . . . new problems emerge . . . Development brings with it new problems. The demographic transition will increase the share of the elderly in the population. Obesity rises with urbanization. Smoking becomes a major health problem. . . . some remain hidden But some age-old problems – notably the plight of people with disabilities – remain with us and are largely ignored. Finally, we should try to expect the unexpected Predictions of progress assume ‘business as usual’. But shocks of various kinds should be expected, though their precise timing and location cannot be predicted with accuracy. Smaller shocks are frequent, but local in their effects and so with little impact on aggregate trends. Except for protracted conflict, smaller shocks also have short-lasting effects. But many low-income countries are vulnerable to very frequent small shocks, the cumulative impact of which seriously undermines poverty reduction efforts. More seriously still, there are strong grounds for believing there will be a major reversal, possibly related to either global conflict or environmental catastrophe, before the middle of this century.

Appendix: projection methods The value of an outcome indicator, Yi,t , for country or region at time t is given in general by Yi,t = β0,i + β1,i Xi,t

(2.A1)

where X is a vector of determinants comprising one or more variables. Hence the prediction of Y when t is at some point in the future (for example, 2015) involves three unknowns: (i) Model specification: which variables to include in X. (ii) Projections of independent variables: the projected value of X at time t.

46 Howard White and Nina Blöndal

(iii) Parameterization: The model parameters (estimation of which is usually based on historical data for X and Y). The general specification given in equation (2.A1) allows the parameter values to vary between countries/regions. Nearly all projections may be written in a form equivalent to equation (2.A1).16 The main exceptions are those based on demographic models, such as those produced for the UN and some analyses of HIV/AIDS, which rely on demographic accounting identities.17 However, even for these exceptions, the formulation in equation (2.A1) identifies the three potential sources of differences in projections. Different model specifications (sets of X-variables) are now discussed in turn. Constant time trend: naïve projections The simplest model is to assume that the future will be like the past, that is, to pass future trends on historical ones. The most common way of doing this is to use a version of equation (2.A1) in which the outcome (Y)-variable is logged and T is simply a time trend: ln Yi,t = β0,i + β1,i T

(2.A2)

so β1 = the annual rate of growth, and β0 is the constant.18 Estimates of β1 may be made in two ways: (i) The most usual method is to estimate the rate of growth based on past data, usually from 1990 to the most recently available year. Hence it is being assumed that the indicator will continue to change at the same rate as it has done since 1990. This approach is implicit in all discussions as to whether countries are ‘on track’ to meet the MDGs, and is the most commonly used with respect to social indicators (for example, the Global Monitoring Report 2005, World Bank 2005a, presentations for primary school completion and under-5 mortality). (ii) Demery and Walton (1998) assume a decline in under-5 mortality of 1.5 per cent a year, which they say has been identified historically as an autonomous element in mortality reduction. These estimates are then augmented to allow for the effects of female education and income. The first of these approaches is labelled as ‘naïve projections’,19 since it takes no account of possible changes in the determinant variables

Projecting Progress

47

between the previous period and the next.20 However, the method does have the virtue of simplicity: (i) there is no need to project the X-variable (since the value of the time trend is of course known in the future); and (ii) the underlying model is easy to grasp: policy-makers can readily understand a statement such as ‘at the current rate of progress, the target of halving the proportion of the people who suffered from hunger in 1990 will not be met by 2015’ (UN MDG Report, Goal 1; United Nations 2005:5). Naïve projections do give information of value. If a country is off track there have to be compelling reasons for believing that performance will change if it is thought the target might still be met. However, they are of dubious value in predicting expected values for future years, which is the main purpose of this report, for variables which are linked closely to underlying determinants. But for variables which can be autonomously driven by policy then naïve projections may prove superior predictors to the models discussed below. Income-based projections The next most common approach, and the dominant one for income poverty projections, is to model Y as depending solely on income (GDP per capita, preferably adjusted for PPP). The form of equation (2.A1) takes logs of both dependent and independent variables: ln Yi,t = β0,i + β1,i ln (INCi,t )

(2.A3)

where INCi,t is GDP per capita for country/region i at time t. Given equation (2.A3), β1 is the elasticity of Y with respect to income, that is the percentage change in Y given a 1 per cent change in income per capita. The elasticity can be obtained from cross-country regressions. This procedure means that a single value of β1 is used for all countries. However β1 may be allowed to vary across subsamples, either by running subsample regressions or else by allowing for a slope dummy for the required categories. For example, Hanmer and Naschold (2000) estimate different income elasticities for income poverty according to the level of inequality. While in principle time series data could be used to produce countryspecific estimates of β1 there are in practice few countries with sufficient time series of the required indicators.21 But in the case of income poverty (see pages 23–8) the elasticity is estimated by the slope of cumulative distribution function at the poverty line. Hence a single household income survey will allow a country-specific estimate of this elasticity.22

48 Howard White and Nina Blöndal

This model specification assumes distribution-neutral growth, that is the income of all income groups grows at the same rate (equal to the overall rate of growth). Such occurrences are a historical rarity; while it is difficult to detect any systematic statistical relationship between the rate of growth and inequality,23 this is not the same as saying that distribution does not change during growth episodes. It does, falling in roughly equal measure between growth episodes which are pro-poor (growth of income among the poor exceeds the average) and those which are antipoor (growth of income among the poor is less than the average). Some analyses allow for changes in distribution during growth;24 for example, Demery and Walton calculated the growth required to halve poverty with distribution-neutral growth and, for selected countries, assuming distribution worsens to a specified level (given by the current level in a comparator country). There remains some debate as to the relative role of income versus other factors in determining social indicators. However, the forecaster is not so concerned with channels. The correlation between income per capita and most social indicators is high, with the R2 from the simple regression being typically in the range 0.6–0.8. It is undoubtedly the case that the income term is picking up the effect of other determinants which are correlated with income, such as female education. But for forecasting purposes it is not necessary to separate these effects. (Of course if female education has an effect independent from that of income, and education is not perfectly correlated with income, then the fit and therefore the forecast would be improved by adding this variable.) Given this reasonably high R2 from the simple regression, income-based forecasts can be taken as good first approximations. However, as will emerge in the subsequent discussions, ignoring distributional issues is a major shortcoming.

Adding more explanatory variables Additional explanatory variables will improve the fit of the equation, and so also the reliability of the forecast, provided that the predictions of the X-variables are not too wildly inaccurate. Several studies have forecast various indicators using more explanatory variables. The main example is the World Bank project, Simulations for Social Indicators and Poverty (SimSIP; www.worldbank.org/simsip). This project, which grew out of projections made for the LAC region, provides spreadsheets which can be downloaded and used to make country-level projections of a range of MDG-type indicators. The projections may be

Projecting Progress

49

based on either historical trends (using the best-fitting of four possible ways of fitting a trend) or model-based elasticities, where the independent variables are economic growth, population growth, urbanization and a time trend. Income poverty measures are disaggregated by rural and urban. In addition to income, Hanmer and Naschold (2000) used HIV prevalence in the under-5 and maternal mortality equations, and the number of physicians in the former and literacy in the latter. Demery and Walton used female literacy in their under-5 mortality equation. Multi-equation models Any solvable multi-equation model can, in principle, be written as a reduced-form equation in the syle of equation (2.A1), that is, expressing the outcome as a function of the exogenous variables in the model. In practice, however, such models are run as computer simulations, showing the effect of different assumptions regarding the trajectories of the exogenous variables (and possibly under different parameter assumptions reflecting different policy scenarios). The World Bank has developed a model called MAMS, which is a CGE model incorporating social indicators. At present MAMS has been applied only to the case of Ethiopia. While the CGE approach allows far more detailed modelling, critics of CGE-based analyses are wary of the extent to which the results are driven by model assumptions, which may in the end be derived by fairly crude methods. Having said that, multi-equation approaches can pick up the complementarities between the different indicators, which single-equation estimates will not.25 Other multi-equation models have been used with respect to nutrition. FAO use country-level modelling of agricultural production and trade, from which they calculate nutrition outcomes. Two IFPRI studies have calculated child nutrition outcomes from different models (Smith and Haddad 2002; von Braun et al. 2005). The level of analysis Analysis is preferably carried out at country level, and the result aggregated to present regional and global forecasts. No study can do this in its entirety, since there are countries for which data do not exist. Thus, either implicitly or explicitly, values are assigned to countries for which there are no data, most usually assuming their performance will equal the average for countries in that region (this is unlikely to be a good assumption since countries without data are more likely poor performers).

50 Howard White and Nina Blöndal Table 2.A1

Summary of main approaches to projecting MDG indicators

Model specification Parameterization Naïve projections

Projection of independent variables

Calculation of historical growth rate (1) Elasticities taken from existing studies;

Not necessary (only independent variable is time trend, which is known) Income-based (1) Use of regional growth forecasts by recognized authority (usually World Bank Global Economic Prospects); (2) regression-based (2) regression-based growth estimates estimate based on initial values Income plus models As for income-based As for income-based Multi-equation (1) Regression-based; Assumed values for different models (2) CGE approach policy simulations Source: See text.

The level of analysis matters not only because each country may require different parameters and exogenous values, but because aggregation may conceal country-level limits. This is the case for many of the MDGs for which it is true that the world as a whole is on track to meet the goal simply by virtue of China’s performance. It would therefore be wrong to say that if current trends continue then the goal will be met – not only because it will be met in China but not many other places (though this is true), but because China cannot logically sustain its rate of progress (more than 100 per cent of children cannot go to school, and less than 0 per cent cannot live below the poverty line). Summary Projections may vary because of differences in assumptions regarding (i) determinants, (ii) model parameters or (iii) future values of the determinants. The simplest models – and by far the most common approach – take time as the only determinant. The result tells us if a country is on track or not to meet the relevant goal, which is certainly of policy interest, but of less use in predicting the actual expected value at some point in the future. Since income is highly correlated with most of the outcomes of interest, a first approximation can be given by applying the relevant elasticity to projected growth rates. This is the basis for many projections. However the correlation with income is imperfect, so adding more variables to the right-hand side will help, at least if their future values

Projecting Progress

51

can be predicted with any degree of confidence. It is argued below that distribution is a right-hand-side variable which should not be ignored. More sophisticated versions utilize multi-equation models. There are undoubted advantages to such models, which allow a wider range of policy simulations, but they are quite resource-intensive and assumption-dependent.

Acknowledgements This chapter is based on a report prepared for the Rockefeller Foundation. However, the views expressed here are those of the authors.

Notes 1. The Appendix presents a fuller discussion of the different approaches. 2. They may be reported on an annual basis in sources such as UNDP’s Human Development Report, but this is not the same thing. These figures are based on interpolation between available data points. 3. While the latter is generally a reasonable assumption, the emergence of stipend schemes in recent years has resulted in girls’ enrolment exceeding that of boys in some countries, for example Bangladesh. 4. The same person, Lucia Hanmer, was responsible for the design of the quantitative aspects of both these studies. 5. Forecasting was also carried out by Demery and Walton (1998) from the World Bank using an approach similar to that used in the ISS and ODI studies, though the results were not presented in a way allowing regional tabulations. 6. The elasticity is taken as −0.5, −0.8 and −1.2 for high, middle- and lowinequality countries respectively, where the former has a Gini coefficient of ≥0.54 and the latter one of ≤0.40. 7. These more positive estimates reflect the relative strength of recent growth performance. Making estimates in this way ignores the cyclical nature of growth and possible threats to growth in some regions, such as the growing dominance of Asian economies. 8. The model contains net enrolments, the degree of urbanization and subregional dummies. 9. Infant mortality is the probability of death in the first year of life and under-5 mortality that before the fifth birthday. These rates are expressed per thousand live births. Child mortality is the probability of death between first and fifth birthdays. 10. Given the importance of China, which is not covered by the DHS data, other sources were consulted. The 1997 China Human Development Report gives under-5 mortality in rural area as 71 compared with 21 in urban areas, a ratio of 3.5. This ratio is much larger than the sample maximum (2.2 for Peru), which might just reflect data differences, but may also be a real disparity on account of the strong differences in development in the western provinces

52 Howard White and Nina Blöndal

11. 12.

13. 14.

15.

16.

17. 18.

19.

20.

21.

22.

23.

compared with the eastern seaboard and the stronger enforcement of the one-child policy in urban areas (personal communication: Hilary Standing and Gerry Bloom, IDS). The ratio for China is set to 2 for the simulation. UN Millennium Project (2005a). Older people with lower literacy were removed from the population and younger people who have been to school added to it. It is being assumed that all those enrolled are literate on completion. This is certainly a rather heroic assumption, but it is one that already lays behind the literacy data themselves. Literacy is capped at 99.5 in all countries. The forecasts from the latter IFPRI study are not given here, as their results are available only in graphical form. The table shows HIV/AIDS deaths from the UN–DESA Population Division medium variant. No global projections of the number of people living with HIV/AIDS have been found. The HIV/AIDS Surveillance Database of the US Census Bureau (http:// www.census.gov/ipc/www/hivtable.html) reports data as ‘major city’ and ‘outside major city’, which is not the same as urban–rural. Nonetheless, these data show higher incidence in cities than outside in virtually all cases for which there are data. The most sophisticated models are multi-equation models, such as MAMS. However, even these approaches can be written as a single reduced-form equation such as equation (2.A1). There is also an exception when calculating income poverty, which is noted below. An alternative specification would be to leave Y unlogged, which would imply a constant absolute change in Y. This is not a very plausible specification, but is one of the four used by Ramadas et al. (2002) for the World Bank SimSIP project (see pages 48–9). De Onis et al. (2004) use a logistic formulation for their analysis of malnutrition, thus allowing for the fact that the dependent variable is bounded between zero and one. The terminology follows that used in economics; naïve expectations are the belief that the variable of interest will not change value from one period to the next. Since the dependent variable in equation (2.2) is logged, this means that the rate of change of determinants is being assumed constant in making naïve projections, not their level (which would be worse than naïve). Data sources such as World Development Indicators and the Human Development Report give a misleading impression of the availability annual estimates for most indicators. However, these series are based on less than annual data collection. The intervening years are obtained by interpolation. Estimates obtained in this way enable cross-country comparisons on an annual basis, but cannot be used meaningfully for time series analysis. A non-regression-based approach can thus be used for income poverty estimates using household survey data. Assuming distribution neutral growth, the income of all households is increased by the same amount as the overall rate of GDP growth and the poverty measures recalculated. It is the absence of such a relationship which is in fact being demonstrated in the influential Dollar and Kraay paper ‘Growth Is Good for the Poor’ (2004),

Projecting Progress

53

rather than the conclusion, widely drawn, that growth generally (or even necessarily) benefits the poor. See White and Anderson (2001) for an elaboration of this point, and for operationalization of the definitions of pro-poor growth given here. 24. This is different from the approach of Hanmer and Naschold mentioned above (and Hanmer et al. 2000), in which the elasticity varies according to the initial level of inequality. 25. This has elsewhere been called this the ‘basic needs multiplier’ (White 1997) and in more recent work on Ghana identified the importance of the ‘inter-generational multiplier’ whereby parents who have themselves been educated are more likely to send their own children to school (World Bank 2004).

References De Onis, M., M. Blossner, E. Borghi, E. A. Frongillo and R. Morris (2004) ‘Estimates of Global Prevalence of Childhood Underweight in 1990 and 2015’. Journal of the American Medical Association, 291(21):2600–06. Demery, L. and M. Walton (1998) Are Poverty Reduction and Other 21st Century Social Goals Attainable? Washington, DC: World Bank. Dollar, D. and A. Kraay (2004) ‘Growth Is Good for the Poor’. In A. Shorrocks and R. van der Hoeven (eds), Growth, Inequality and Poverty: Prospects for Pro-Poor Economic Development. Oxford University Press for UNU-WIDER. FAO (2002) World Agriculture: Towards 2015/2030. Rome. Hanmer, L. and F. Naschold (2000) ‘Attaining the International Development Targets: Will Growth Be Enough?’. Development Policy Review, 18(1):11–36. Hanmer, L., J. Healey and F. Naschold (2000) ‘Will Growth Halve Poverty by 2015?’ Overseas Development Institute Policy Briefing 8. London. Hanmer, L., N. de Jong, R. Kurian and J. Mooij (1997a) ‘Social Development: Past Trends and Future Scenarios’. Sida Project 2015. The Hague: Institute of Social Studies. Hanmer, L., N. de Jong, R. Kurian and J. Mooij (1997b) ‘Poverty and Human Development: What Does the Future Hold?’ Institute of Social Studies Working Paper 259. The Hague. Hanmer, L., R. Lensink and H. White (2003) ‘Infant and Child Mortality in Developing Countries: Analysing the Data for Robust Determinants’. Journal of Development Studies, 40(1):101–18. Ramadas, K., D. van der Mensbrugghe and Q. Wodon (2002) SimSIP Poverty: Poverty and Inequality Comparisons Using Group Data, Washington, DC: World Bank. Rosegrant, M. W., S. Cline, W. Li, T. Sulser and R. Valmonte-Santos (2005) ‘Looking Ahead: Long-Term Prospects for Africa’s Agricultural Development and Food Security’. International Food Policy Research Institute 2020 Discussion Paper 41. Washington, DC. Smith, L. C. and L. Haddad (2000) ‘Overcoming Child Malnutrition in Developing Countries: Past Achievements and Future Choices’. Food, Agriculture and Environment Discussion Paper. Washington, DC: International Food Policy Research Institute.

54 Howard White and Nina Blöndal UN Millennium Project (2005a) Toward Universal Primary Education: Investments Incentives and Institutions, Millennium Project Task Force on Education and Gender Inequality Report. New York: United Nations. UN Millennium Project (2005b) Taking Action: Achieving Gender Equality and Empowering Women. Millennium Project Task Force on Education and Gender Inequality Report. London: Earthscan. UNAIDS (Joint United Nations Programme on HIV/AIDS) (2005) AIDS in Africa: Three Scenarios to 2025. Geneva. UNFPA (United Nations Population Fund) (no date) ‘The Price of Inaction’. Available at: www.unfpa.org/icpd/inaction.htm United Nations (2005) The Millennium Development Goals Report 2005. New York: United Nations. Von Braun, J. et al. (2005) ‘New Risks and Opportunities for Food Security Scenario Analyses for 2015 and 2050’. International Food Policy Research Institute 2020 Discussion Paper 39. Washington, DC. White, H. (1997) ‘The Economic and Social Impact of Adjustment in Africa: Further Empirical Analysis’. Institute of Social Studies Working Paper 245. The Hague. White, H. and E. Anderson (2001) ‘Growth versus Distribution: Does the Pattern of Growth Matter?’ Development Policy Review, 19(3):267–89. World Bank (2004) Books, Buildings, and Learning Outcomes: An Impact Evaluation of World Bank Support to Basic Education in Ghana. Washington, DC: Operations Evaluation Department, World Bank. World Bank (2005a) Global Monitoring Report 2005. Washington, DC: World Bank. World Bank (2005b) Maintaining Momentum to 2015? An Impact Evaluation of Interventions to Support Maternal and Child Health and Nutrition in Bangladesh. Washington, DC: Operations Evaluation Department, World Bank. World Bank (annual) Global Economic Prospects. Washington, DC: World Bank. Available at: www.worldbank.org/prospects

3 Achieving Health, Wealth and Wisdom: Links between Aid and the Millennium Development Goals David Fielding, Mark McGillivray and Sebastián Torres

Introduction Achieving the MDGs is an ambitious undertaking and one that in planning relies heavily on the developmental impact of aid. The achievement of the MDGs by 2015, or later, requires inter alia a recognition and understanding of their interdependence and the efficient intra-country allocation of development aid. This chapter addresses both issues. Using a new cross-country data set, it empirically examines: (i) the strength of the links between a number of MDG target and related variables, including health, educational status and access to water and sanitation; and (ii) the extent to which aid impacts on these variables. Identification of the key links between the different target variables can help to inform prioritization of the MDGs, by suggesting the areas of endeavour that are likely to have the largest and widest impact. A particular interest of the chapter is whether there is a central or pivotal variable, which, if targeted, leads to the greatest beneficial impact on the remaining variables. Importantly, the chapter’s data set provides information on population subgroups within each country. This allows the chapter also to draw inferences regarding the impact of aid on the poorest groups in each country. Such information is clearly crucial for strategies to move towards achieving the MDGs, especially in countries that are a long way from or not tracking towards achieving the goals. Overall, the chapter seeks not only to provide policy-relevant information on achieving the MDGs, but also to contribute to the literatures on the determinants of cross-country wellbeing achievement and aid effectiveness. This analysis differs in three ways from previous work on the determinants of the cross-country variation in the level of wellbeing. First, we 55

56 David Fielding, Mark McGillivray and Sebastián Torres

model simultaneously five variables capturing different aspects of wellbeing, along with a measure of aid. They relate respectively to the level of material prosperity, the supply of water, educational attainment, fertility and health. One of these variables, the health indicator, is an MDG target variable. Three (water supply, sanitation and educational attainment) are very similar to three other target variables. The water supply variable, for example, is access to piped water. The corresponding MDG target is the proportion of the population with access to an improved water source. Each of these wellbeing variables has the clear potential to impact on each of the others. Modelling all simultaneously permits us to identify the linkages that are the most quantitatively important, including identifying whether there is a pivotal variable, as defined above. Second, we are using a newly compiled data set that reports observations on a wide range of wellbeing indicators for quintiles within developing countries, rather than just overall averages for each country. As a consequence, our model gives equal weight to the wellbeing outcomes of the rich and the poor within a country. Third, our measure of material wellbeing is based on a household survey recording each household’s possessions. We make no reference to per capita income or wealth: instead, our model employs a basic measure of visible material prosperity at the household rather than the personal level. Appropriate for analysing wellbeing in developing countries, this measure is based on the basic assets a household possesses. These assets are basic enough for differences in quality across countries not to be a major worry. Our chapter differs from previous research on aid effectiveness in two main ways. The first is in the use of subnational data to assess aid impact, rather than national averages. This allows us to assess simultaneously the impacts of aid across population subgroups, rich and poor. These impacts have not previously been addressed in empirical research, and the fact that they have not is indicative of a huge void in the aid literature. The second difference is that our chapter provides estimates of a structural model of aid and wellbeing outcomes. This also has not previously been attempted, and allows us to identify a range of direct and indirect aid impacts. There is obvious relevance here for MDG achievement, especially regarding the intra-country allocation of wellbeing aid most conducive to achieving these goals. To outline this chapter, we first review briefly strands of the wellbeing and aid effectiveness literatures,1 before outlining the ways our variables are defined and measured. Next we present descriptive statistics obtained from the data we use, outlining the econometric methods employed to analyse these data, and discuss the results obtained from this analysis.

Aid and the MDGs

57

Then we report results from a number of simulations, mainly involving the effects predicted by our model of increasing aid or increasing its effectiveness. Among the increases under consideration is a doubling of aid levels from those that were provided in the periods under review. Among the chapter’s findings is that child mortality is the central variable, in that a decrease in child mortality leads to the largest beneficial changes in the other MDG or MDG-similar variables under consideration. It is also the variable on which aid has the largest quantitative impact. Our examination further finds that while aid is effective overall, the poorest subgroups within each country are typically not the principal beneficiaries of these inflows.

A brief literature review Economists have long been aware of the importance of links between the various wellbeing dimensions and their implications for poverty. There is a large literature tracing the theory and evidence relating to the ways the income of a population is connected to standards of education and health, and also to fertility. Average standards of education and health are elements of human capital that are likely to determine a region’s overall productivity level, and hence its per capita income. Moreover, with decreasing returns to scale, higher fertility and population growth will result in lower labour productivity. On the other hand, a household’s decisions about human capital investment and the number of children to produce may depend on its current income level, especially with imperfect capital markets (Becker 1981). While the literature on links between wellbeing indicators offers a number of important insights, we believe that it embodies a number of limitations. First, empirical studies relating to the connections between different dimensions of wellbeing focus typically on a single link in the chain. There are studies of the impact of a region’s education on its income (for example, Teulings and van Rens 2003), of income on education (for example, Fernandez and Rogerson 1997), of health on income (for example, Pritchett and Summers 1993), of income on health (for example, Bloom et al. 2004), of fertility on income (for example, Ahlburg 1996) and of income on fertility (for example, Strulik and Siddiqui 2002).2 Many of these studies present careful and compelling evidence on their chosen area of research, but taken as a whole they embody certain limitations. The heterogeneity of statistical methodologies and data sets across these chapters means that they do not shed any collective light on the relative importance of the different causal links in the overall

58 David Fielding, Mark McGillivray and Sebastián Torres

wellbeing achievement process. It would be useful to know, for example, if any one link is particularly strong, and therefore a potential focus for wellbeing policy interventions. Moreover, while authors are aware of the likely simultaneity of different wellbeing indicators, the focus on a single link in the chain means they never venture beyond an instrumental variables approach to estimation. Such an approach neglects the correlation of errors across equations for different indicators, which may be of interest in itself as well as affecting the statistical efficiency of the estimates. Second, most existing cross-country studies use data on the average value of the wellbeing indicators in each country. The main aim of most empirical economic research has been to explain correlations in these indicators at the national level. Researchers in education and health sciences have often been more sensitive to the drawbacks of such an approach.3 They point out that using mean income places a large weight on the income of the rich, because income distributions are left-skewed, so the mean figure reported for a country is higher than the median. Looking at the link between variations in mean income and, say, variations in infant mortality might be misleading, because high infant mortality is a consequence of the poverty of middle- and low-income groups in a developing country. One way of addressing this problem might be to include a measure of income distribution in the empirical model; however, a more direct approach would be to measure separately the income and health status of the rich and poor within a country. Economists have also long had an interest in aid effectiveness. Dating back to the 1950s, the literature on this topic has been dominated by empirical studies looking at links between aid and per capita income growth. Studies conducted from the mid-1990s generally conclude that per capita growth in recipient countries would be lower in the absence of aid. Some studies find that this outcome is contingent on the quality of recipient country policies, while others point to other contingencies such as recipient country structural vulnerability, institutions, climatic conditions and political stability. The best-known and arguably the most influential aid growth study is Burnside and Dollar (2000). Others include Guillaumont and Chauvet (2001); Hansen and Tarp (2001); Hudson and Mosley (2001); Chauvet and Guillaumont (2002); Collier and Hoeffler (2004); Dalgaard et al. (2004).4 While the recent aid-growth literature has provided a number of useful insights, it must be recalled that in principle aid is primarily about enhancing wellbeing in developing countries, especially that of the poorest people. Recognizing that wellbeing has both economic and

Aid and the MDGs

59

non-economic dimensions, this involves not only increases in incomes but other outcomes, especially in health and education.5 The MDGs are a reaffirmation of these points, given the role that aid is expected to play in their achievement and the range of target variables they identify. Levels of achievement in many wellbeing dimensions tend to be linked. Countries that achieve higher incomes tend to achieve better health and higher levels of education, although this rule of thumb can break down if one looks at specific country groups or subnational data (McGillivray 2005). To this extent one might be able to infer from the recent aid-growth literature that aid has also led to improvements in health and education. However, it would be better to depend not on inferences but on direct evidence, especially given the non-uniformity of income, health and education outcomes. The case for such evidence grows if one accepts the widespread belief that per capita income growth alone is insufficient to attain the MDGs. Another stream within the aid effectiveness literature looks inter alia at the impact of aid on various categories of government expenditure6 This literature is based on a recognition that most aid goes typically to the public sector of recipient countries as an intended augmentation of public expenditure. Better-known studies of this type include Heller (1975); Pack and Pack (1990, 1993); Gang and Khan (1991); Franco-Rodriguez et al. (1998); Feyzioglu et al. (1998). Most of these studies estimate a system of equations that allows for the simultaneous determination of expenditures. The categories of expenditure include those relevant to achievement in non-economic wellbeing dimensions, including health and education. While results do vary between studies, aid is typically associated with increases in expenditure on health and education. Other studies have reported a positive association between aid and broader expenditure categories that include spending on health and education.7 While these results are encouraging, there remains the problem of identifying connections between changes in these expenditures and actual wellbeing outcomes.8 Without considerable additional information this is a highly speculative exercise, given the complexity of links between wellbeing-related expenditures and outcomes such as life expectancy, infant mortality and literacy. Similarly, World Bank (2003) is one of many studies to show that changes in public expenditure on health and education are often weakly related with health and education outcomes. One of the possible reasons is that the efficiency of these expenditures varies among countries. Increasing these expenditures appears to be a necessary but not sufficient condition for improving health and education outcomes. We return to this important issue below.

60 David Fielding, Mark McGillivray and Sebastián Torres

A much more recent stream in the aid effectiveness literature is not subject to the preceding criticism, as the studies within it do actually look at links between aid and wellbeing outcomes. These studies include Boone (1996); Kosack (2003); Mosley et al. (2004); Gomanee et al. (2005a); and Gomanee et al. (2005b). Each looks at the impact of aid on the infant mortality rate and, with the exception of Boone (1996), the human development index (HDI).9 The majority finding of these studies is that aid is positively associated with national wellbeing outcomes in health and possibly also education. This is perhaps the most encouraging result emanating from the aid literature, given the objective that such transfers are intended to achieve. But this research is subject to a fundamental criticism, one that it shares with the two other streams just discussed: it employs national data, either averaged or aggregated in some way. The primary beneficiaries, in terms of wellbeing outcomes, might not be the poorest within aid-receiving countries. Such an outcome is counter to the principles of aid, and to what the MDGs are intended to achieve. This speculation is fuelled (although clearly not confirmed) by the fact that within many developing countries public spending on health and education, including that on primary health and primary education, has a pro-rich orientation (World Bank 2003). What is required is information on the impact of aid by population subgroup. The literature on aid effectiveness, discussed above, ignores this question.

Data definition and measurement The wellbeing indicators that are the focus of our econometric analysis are taken from the World Bank’s HNP Poverty Data (World Bank n.d.), which aggregates household survey data from 55 countries. Of these countries, 48 are included in our analysis; they are listed in Table 3.1.10 As can be seen from the table, the year of measurement varies slightly from one country to another. An innovative characteristic of the HNP (health, nutrition, population) data set is the way it measures material wellbeing. Its material wellbeing measure is based on the presence or absence of various durable assets in the household, and of certain characteristics of the household’s dwelling place. The assets in question vary from one country to another, depending on the material possessions specific to a certain culture. Every household in the country survey is ascribed a value of zero or one for each asset or dwelling attribute, depending on whether that asset or attribute is present in the household. A household-specific prosperity index is then constructed as the weighted sum of all the binary asset variables.

Aid and the MDGs Table 3.1

61

Countries included in the analysis Survey year

Bangladesh Benin Bolivia Brazil Burkina Faso Cambodia Cameroon C.A.R. Chad Colombia Comoros Côte d’Ivoire Dom. Rep. Egypt Ethiopia Gabon Ghana Guatemala Guinea Haiti India Indonesia Jordan Kenya

2000 2001 1998 1996 1999 2000 1998 1995 1997 2000 1996 1994 1996 2000 2000 2000 1998 1999 1999 2000 1999 1997 1997 1998

Survey year Madagascar Malawi Mali Mauritania Morocco Mozambique Namibia Nepal Nicaragua Niger Nigeria Pakistan Paraguay Peru Philippines Rwanda S. Africa Tanzania Togo Uganda Vietnam Yemen Zambia Zimbabwe

1997 2000 2001 2001 1992 1997 2000 2001 2001 1998 1990 1990 1990 2000 1998 2000 1998 1999 1998 2001 2000 1997 2002 1999

The weights are the coefficients in the first principal component of the whole set of asset variables, scaled so as to sum to unity. (A few of the weights are negative, and in these cases one might conclude that the presence of that characteristic is a sign of poverty.) Households are then ranked by the index and divided into quintiles; average health and education statistics are reported for the households in each quintile. We wish to construct a cross-country measure of material wellbeing. The asset indices reported in the HNP data set are not appropriate for this purpose, because they are based on country-specific sets of assets. Nevertheless, there is a subset of eight assets and attributes common to all countries in the database.11 These are: the presence of an electricity supply; possession of a radio, of a television, of a refrigerator, of a car; access to a flush toilet; use of a ‘bush or field latrine’ (a euphemism for the complete absence of sanitary facilities); and the presence of a dirt or

62 David Fielding, Mark McGillivray and Sebastián Torres Table 3.2

Descriptive statistics for the asset weights

Asset

Mean

Median/mean

Std dev./mean

Electricity Radio Television Refrigerator Car Flush toilet Bush/field latrine (−) Dirt/sand floor (−)

0.149 0.095 0.144 0.146 0.090 0.097 0.128 0.149

1.03 1.01 1.04 1.00 1.07 0.99 1.02 1.07

0.18 0.27 0.11 0.16 0.26 0.38 0.43 0.31

Notes: The numbers in the table are subject to rounding error.

sand floor in the house. The last two of these characteristics are signs of poverty and take a negative weight in all countries. If we look at the relative importance of each of these characteristics in each country, we find very little variation from one country to another. Table 3.2 reports the cross-country means of the weights on the eight characteristics (scaled so that these mean weights sum to unity; subject to rounding error), along with the ratios of each median and standard deviation to its respective mean. The table shows that the standard deviations are quite small, and that the medians are close to the means, indicating an approximately symmetrical distribution. Therefore, we will construct a cross-country asset measure for the kth quintile of the nth country as follows: asskn =
h sh · zhkn

(3.1)

where h = 1, . . . , 8 indexes the assets, sh is the weight on the hth asset, taken from the first column of Table 3.2 (
h sh = 1) and zhkn indicates the fraction of households in the quintile possessing the asset. In the case of ‘bush latrines’ and dirt floors, zhkn indicates the fraction of houses without the characteristic. As can be seen from Table 3.2, there is not a great deal of variation in the sh , so results from an alternative definition of material wellbeing with ∀hsh = 0.125 yields results very similar to the ones reported below. Our other four endogenous wellbeing outcome indicators capture average levels of sanitation, education, fertility and health of each quintile in each country. Sanitation (wtrkn ) is measured as the fraction of the quintile with access to piped water. (One alternative modelling strategy is to aggregate the lavatorial elements of the assets index with wtr.

Aid and the MDGs 60 ass

150 wtr

40

100

20

50 0.00 sch

0.25

0.50

0.75

1.00

fer

0.00

0.25

0.50

0.75

63

1.00

40

30 20

20

10

mor

0.00

0.25

0.50

0.75

1.00

0

2

4

6

8

50 25

⫺0.1

0.0

0.1

0.2

0.3

0.4

0.5

0.6

Figure 3.1 Frequencies of values of the five wellbeing indicators

However, the resulting measure of sanitation leads to a model that fits the data more poorly. It seems that high-quality lavatory facilities are more an indication of material prosperity than they are of good sanitation.) In the HNP data set, among the measures of educational attainment is the fraction of adults aged 15–49 who have completed grade 5. We denote this measure as schkn . Fertility (ferkn ) is measured in the database as the average number of live births per woman aged 15– 49. A wide range of family health indicators are reported, though not all are reported for every country. In the results reported below, we use the mortality rate for children under five years (morkn ). To summarize, our wellbeing outcome indicators are asskn , wtrkn , schkn , ferkn and morkn . The distributions of these five variables are illustrated in Figure 3.1. We did also consider alternative definitions of material wellbeing, education and health, using (i) uniform asset weights to define material wellbeing, (ii) the fraction of women reading a newspaper at least once a week to measure education and (iii) the mortality rate for children under 12 months to measure family health. The seven alternative regression specifications combining the different measures produced results similar to the ones reported below. Alternative measures of sanitation (for example, the fraction of households with access to a

64 David Fielding, Mark McGillivray and Sebastián Torres

purpose-built latrine or flush lavatory) produce results broadly similar to the ones here, but with rather higher standard errors. It can be seen from Figure 3.1 that the wellbeing indicators are not normally distributed. In the case of ass, wtr and sch there are some observations close to the theoretical bound where all of the households in the quintile possess all of the assets (or, at the other extreme, none of them) in the material wellbeing index; some observations record quintiles where all or none of the households have access to piped water or primary education. This problem does not occur for the other two variables, fer and mor, but in the latter case the distribution does seem to be left-skewed. For this reason the ass, wtr and sch equations are fitted in Probit form, as described below. The fer and mor equations will be fitted in log-linear form. In order to identify the impact of one wellbeing indicator on another, we need to include a range of exogenous conditioning variables in our regression equations. Restrictions on the coefficients of the conditioning variables will permit us to identify the links between the wellbeing indicators. Note that these exogenous national characteristics vary across countries but not across quintiles within a country. We include in our model variables to capture factors relating to (i) geography, (ii) history and (iii) culture. Data sources for these variables are listed in the Appendix. Included in (i) are: the country’s surface area in square kilometres (size), a logarithmic measure of the value of its natural resources in US dollars (natres), a dummy for whether it has a maritime coastline (coast), its mean annual temperature in 0.1 degrees centigrade (temp), and the fraction of the population at risk from malarial infection (malfal). Given its significance in previous studies (for example, Easterly and Levine 1997), we also include a dummy for countries in Africa (africa). Included in (ii) are dummy variables for whether the country was colonized by Great Britain (britain) or by France (france). Included in (iii) are the fraction of the population that are Christian (chrs), the fraction that are Muslim (mus) and an index of ethno-linguistic fractionalization (ethno). We introduce into our analysis a measure of the level of aid to a country, aid, to test whether aid impacts on the above-mentioned endogenous wellbeing variables. This variable is measured as the average ratio of net ODA disbursements to GDP in the five years up to and including the measurement of the wellbeing indicators. Data are taken from the OECD–DAC online database (OECD 2005). There are many reasons why aid might influence these variables, including: (i) its economy-wide impact on growth; (ii) its impact on government expenditure on health,

Aid and the MDGs

65

education and water (in cases where these expenditures are productive); (iii) aid-funded projects that boost private incomes and support the provision of public goods and services; (iv) donor initiatives to improve the productivity of the above expenditures through the provision of technical assistance and capacity-building; and (v) conditions attached to aid inflows requiring recipients to pay more attention to health, education and water in their own policy agendas. On the basis of these factors, combined with evidence from the existing literature discussed above that the effects mentioned in points (i) and (ii) are in general positive, our expectation a priori is that aid will have a positive impact on the endogenous wellbeing variables. We have no firm expectations a priori regarding the relative impacts of aid on each population subgroup. One might expect that the primary beneficiaries of aid are the poorest quintiles, especially when the interaction between donors and recipient governments encourages the latter to give the wellbeing of the poorest a greater weight in policy. However, given the range and complexity of aid effects, it remains uncertain whether these groups will benefit most from aid. Finally, we need an instrument for our potentially endogenous aid variable, in line with the recent literature on aid. The chosen instrument, discom, is the ratio of ODA disbursements to ODA commitments over the five years before the first year of measurement of aid. An ODA commitment is the amount of funds donors make available to recipients for disbursement. Recipients that do not fully disburse commitments in the current year can be expected to receive less aid in the subsequent year. More generally, if a smaller proportion of the ODA commitment is disbursed in period t − 1, less aid will be provided in period t. All data required to calculate discom were taken from OECD (2005).

Empirical analysis Descriptive statistics Table 3.3 provides data on the unconditional correlations of the wellbeing indicators, again disaggregating by quintile. The signs on individual correlation coefficients are what one might expect. Assets, water and education (the ‘goods’) are positively correlated; fertility and child mortality (the ‘bads’) are also positively correlated. Correlations across these two pairs are always negative. The correlations are generally highest for quintiles 4–5 (the richest) and lowest for quintiles 1–2 (the poorest). This suggests that the variation in outcomes for richer households is more systematic, and may be more closely correlated with observable

66 Table 3.3 Summary statistics Means

ass

wtr

sch

fer

mor

Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5

0.1595 0.2511 0.3379 0.4481 0.6534

0.0411 0.1054 0.1551 0.2820 0.5277

0.3075 0.4083 0.4974 0.6110 0.7851

6.0625 5.3417 4.8542 4.2646 3.2146

0.1781 0.1682 0.1556 0.1294 0.0876

Standard deviations Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5

0.1175 0.1610 0.2013 0.2250 0.1822

0.1322 0.2289 0.2711 0.3150 0.3109

0.2224 0.2592 0.2691 0.2580 0.1642

1.3731 1.4158 1.5492 1.5444 1.1897

0.0914 0.1013 0.1023 0.0929 0.0582

Quintile 1 correlations

ass

wtr

sch

fer

wtr sch fer mor

0.7932 0.4772 −0.3617 −0.5057

0.2931 −0.1536 −0.3641

−0.2518 −0.5494

0.4271

Quintile 2 correlations wtr sch fer mor

0.7703 0.6525 −0.5455 −0.6679

0.3558 −0.3125 −0.4482

−0.4803 −0.6694

0.6669

Quintile 3 correlations wtr sch fer mor

0.8205 0.7197 −0.6858 −0.7425

0.4443 −0.4690 −0.5123

−0.6550 −0.7393

0.8126

Quintile 4 correlations wtr sch fer mor

0.8391 0.7498 −0.7453 −0.7874

0.5373 −0.5862 −0.6281

−0.7558 −0.7893

0.8747

Quintile 5 correlations wtr sch fer mor

0.8285 0.7128 −0.6536 −0.8270

0.5697 −0.5301 −0.7640

−0.7281 −0.8186

0.8613

Aid and the MDGs

67

independent characteristics; the variation among poorer households may have a larger stochastic element. These characteristics indicate that in our econometric model it would be unwise to try to impose a priori any structure on the covariance matrix of residuals for each wellbeing indicator and each quintile. Variances and covariances are unlikely to be uniform across quintiles, let alone across indicators. Outcomes at the upper end of the assets distribution are likely to be somewhat more predictable than those at the lower end. Model structure The descriptive statistics suggest strong interrelations between our four wellbeing indicators. However, the descriptive statistics also suggest that conditional variances are unlikely to be constant across indicators or across quintiles, and it would be unwise to make any assumptions a priori about the corresponding covariances. So our model will take the following general form. Let the jth wellbeing indicator for the kth quintile in the nth country (j = 5, k = 5, n = 48) be denoted yjkn . Then our regression equation for the jth indicator is yjkn = F(αjk +
i=j βij · yikn +
p ϕjp · xnp + θj ln aidn ) + ujkn

(3.2a)

for j = (ass, wtr, sch) and ln (yjkn ) = αjk +
i=j βij · yikn +
p ϕjp · xnp + θj ln aidn + ujkn

(3.2b)

for j = (fer, mor). F(.) is the Normal cumulative density function. xnp is the value of the pth exogenous conditioning variable in the nth country and ujkn is a residual. Our aid equation is ln (aidn ) = αAID +
p ϕAIDp xnp + θAID · ln discomn + uAIDn

(3.3)

A priori restrictions on the φjp -coefficients allow us to identify (most of) the βij -coefficients that capture the interactions between the five wellbeing indicators. We allow the conditional cross-country mean of each wellbeing indicator, αjk , to vary across quintiles, so that we are in fact fitting a fixed-effects model. We have 5 × 5 × 48 = 1200 observations of yjkn , and hence 1200 observations of the residuals ujkn . We do not wish to assume any restriction on the correlation of residuals across indicators or across quintiles, so the model is fitted by stacking 26 regression equations – one for each j and each k, plus one for aid – and estimating the coefficients in each equation simultaneously by 3SLS. With only 48

68 David Fielding, Mark McGillivray and Sebastián Torres Table 3.4 Variable definitions and model structure y-variables ass wtr sch fer mor aid

The assets index The fraction of households with access to piped water The fraction of household members with primary education Live births per woman Under–5 mortality rate Ratio of aid to GDP

x-variables for: africa dummy = 1 if in Africa britain dummy = 1 if colonized by Britain france dummy = 1 if colonized by France ethno ethno-linguistic fractionalization index ln(siz) log country surface area ln(nat) log natural resource capital value coast dummy = 1 if country has a coastline chrs fraction of the population that is Christian musl fraction of the population that is Muslim temp temperature (in 0.1 degrees C) tems tmp2 /100 malfal fraction of population at risk from malaria

Appearing in the equations ass wtr sch fer mor ass wtr sch fer mor ass wtr sch fer mor ass wtr ass ass ass

wtr wtr wtr

mor

sch

fer

sch

fer mor mor mor

Note: All of the x variables also appear in the aid equation, along with the disbursement– commitment ratio discom.

countries, we do not have enough degrees of freedom to allow the slope coefficients (βij , φjp ) to vary across quintiles, so each of these should be interpreted as the mean effect of a particular explanatory variable across all countries and all quintiles. It is possible to fit a quintile-specific model, but with 48 observations, standard errors on individual coefficients are so high as to preclude much economic interpretation. Identification of the β-coefficients requires some a priori restrictions on the φ-coefficients. These restrictions, summarized in Table 3.4, are as follows. First, some of the geographical characteristics are unlikely to have a direct impact on anything other than material resources (ass, wtr) through an effect on factor productivity. These characteristics are country size (size), natural resource wealth (natres) and ethno-linguistic fractionalization (ethno). Similarly, other geographical characteristics

Aid and the MDGs

69

are unlikely to have a direct impact on anything other than health. These characteristics are temperature (temp)12 and malaria risk (malfal). Whether a country has a coastline (coast) might affect health and material prosperity, but it is unlikely to affect education or fertility directly, and so it can be excluded from the equations for these two indicators. These restrictions together allow us to identify the effects of material wellbeing (ass), water access (wtr) and health (mor) in each of the other four equations, except that the effects of ass on wtr and of wtr on ass are unidentified. The effects of fertility (fer) and education (sch) in the assets, water and health equations are identified by assuming that religious adherence, as captured by chrs and musl, has no direct effect on assets and health. However, it might affect attitudes towards contraception or the value of education (especially female education), and so have a role in determining fer and sch. The other effects we do not attempt to identify – because of an absence of any obvious instrument – are of fer in the sch equation and of sch in the fer equation. This will turn out to be important when we come to interpret the coefficients in these two equations. Equation (3.3) allows us to identify the effects of aid in all five of the wellbeing indicator equations. The θj-coefficients will then indicate the impact of an increase in aid on each indicator, holding the other indicators constant. That is, they will determine the partial derivatives ∂yj /∂ ln (aid). (However, in the case of ass, wtr and sch the partial elasticities will not be constant because of the logistic transformation.) In addition, we can solve the equations and calculate the total derivative effects dyj /d ln (aid). Again, these derivatives will not be constant, but we are able to use the model to predict the consequence for each indicator in each quintile in each country of an increase in aid to that country by a certain percentage amount, starting at the observed levels of aid and of the wellbeing indicators. Regression results The regression results for the four wellbeing indicators are reported in Tables 3.5 and 3.6. Note that the first three equations are in Probit form,13 so that the coefficients tell us, approximately, how proportionate changes in the explanatory variables translate into proportionate changes in y/(1 − y). The other equations are log-linear, so that coefficient values translate into proportionate changes in y. The first part of Table 3.5 shows the β- and φ-coefficients in the logistic ass equation, along with corresponding standard errors and t-ratios. The significant φ-coefficients are those on size, coast, africa and ethno. Ceteris paribus,

70 Table 3.5 Fitted regression coefficients coeff.

Std error

t-ratio

ass equation natres ethno size coast africa britain france sch fer mor aid

0.0130 −0.2025 0.0489 0.1025 −0.1659 −0.0047 0.1067 0.8900 0.1068 −0.4146 0.0697

0.0225 0.0988 0.0222 0.0474 0.0719 0.0615 0.0845 0.1793 0.0780 0.0448 0.0393

0.5786 −2.0500 2.2032 2.1640 −2.3071 −0.0770 1.2635 4.9640 1.3701 −9.2497 1.7732

wtr equation natres ethno size coast africa britain france sch fer mor aid

−0.0561 −0.5033 0.2920 0.0463 −0.2788 0.5123 0.5081 −0.4526 0.7836 −1.5108 0.4437

0.0658 0.3358 0.0600 0.1343 0.2609 0.2357 0.2799 0.3975 0.2159 0.1526 0.1182

−0.8528 −1.4989 4.8688 0.3451 −1.0687 2.1734 1.8156 −1.1384 3.6301 −9.9025 3.7524

sch equation chr mus africa britain france ass wtr mor aid

0.3867 −0.2141 0.1136 0.5751 0.0545 2.3295 −0.7767 −0.3486 0.0423

0.1150 0.0873 0.1030 0.0725 0.0691 0.2288 0.1050 0.0546 0.0436

3.3640 −2.4526 1.1029 7.9383 0.7890 10.1799 −7.3970 −6.3894 0.9696

fer equation chr mus africa britain france ass wtr mor aid

0.4504 0.2654 −0.1506 0.1188 0.0578 −0.8459 0.1393 0.3700 0.1037

0.0421 0.0375 0.0533 0.0518 0.0428 0.1167 0.0802 0.0313 0.0318

10.7058 7.0833 −2.8236 2.2931 1.3514 −7.2497 1.7360 11.8133 3.2605 (Continued)

Aid and the MDGs Table 3.5

71

(Continued) coeff.

Std error

t-ratio

mor equation temp tems/100 malfal coast africa britain france ass wtr sch fer aid

−0.4122 0.0766 0.4375 −0.1305 0.1770 0.1439 0.0809 0.0930 −0.5388 −1.0602 0.6854 −0.1208

0.2360 0.0583 0.0705 0.0389 0.0840 0.0817 0.0878 0.2642 0.1097 0.1619 0.0931 0.0464

−1.7462 1.3152 6.2065 −3.3551 2.1078 1.7618 0.9213 0.3519 −4.9106 −6.5503 7.3647 −2.6060

aid equation natres ethno size coast chr mus temp tems/100 malfal britain africa france discom

0.2672 0.1513 −0.3787 −0.2652 0.1304 0.4902 −1.6383 0.2680 0.6492 0.2348 −0.1308 −0.1795 0.4436

0.0675 0.4695 0.1083 0.1609 0.4248 0.2555 1.4365 0.3412 0.3164 0.2717 0.3356 0.3613 0.2031

3.9566 0.3222 −3.4955 −1.6480 0.3069 1.9188 −1.1405 0.7854 2.0520 0.8644 −0.3898 −0.4969 2.1843

large countries with a maritime coastline can be expected to have a higher level of material prosperity. Ethno-linguistic diversity and location in Africa have a negative impact on material prosperity, as in Easterly and Levine (1997). Two of the three identified β-coefficients are large and statistically significant. As expected, better standards of education (higher sch) and health (lower mor) lead to higher levels of material prosperity: this is the human-capital effect. An increase of 1 percentage point in the fraction of household members with primary education can be expected to raise ass/(1 − ass) by just under 1 per cent. A 1 per cent reduction in child mortality can be expected to lower ass/(1 − ass) by just under 0.5 per cent. These effects do not take into account any feedback from the effects of higher material prosperity on education and health, which is discussed later.

72 Table 3.6 Main econometric results (a) Regression R2 statistics

ass wtr sch fer mor aid

Quintile 1

Quintile 2

Quintile 3

Quintile 4

Quintile 5

0.4600 0.4440 0.6469 0.4410 0.6912

0.6805 0.3202 0.7838 0.6534 0.8528

0.7397 0.3234 0.8545 0.7957 0.8770 0.4320

0.7980 0.5124 0.8592 0.8098 0.8866

0.7166 0.6704 0.8480 0.7719 0.9113

(b) Regression standard errors

ass wtr sch fer mor aid

Quintile 1

Quintile 2

Quintile 3

Quintile 4

Quintile 5

0.0901 0.1004 0.1319 0.2076 0.3159

0.0930 0.1984 0.1234 0.1871 0.2457

0.1038 0.2376 0.1048 0.1660 0.2582 0.7862

0.1010 0.2223 0.0966 0.1760 0.2681

0.0970 0.1809 0.0641 0.1758 0.2279

(c) Cross-quintile averages of cross-variable residual correlations

wtr sch fer mor

ass

wtr

sch

fer

0.6363 −0.4511 0.3470 0.1632

−0.0231 0.0993 0.3121

−0.2774 0.3273

−0.5908

(d) Within-variable residual correlations

ass

wtr

sch

Quintile 1

Quintile 2

Quintile 3

Quintile 4

Quintile 2 Quintile 3 Quintile 4 Quintile 5

0.8807 0.6502 0.2000 0.0148

0.8756 0.4934 0.2487

0.7617 0.4746

0.8141

Quintile 2 Quintile 3 Quintile 4 Quintile 5

0.7876 0.7975 0.5862 0.2343

0.9107 0.7083 0.4440

0.7817 0.4716

0.6498

Quintile 2 Quintile 3

0.8947 0.7603

0.9082 (Continued)

Aid and the MDGs Table 3.6

fer

mor

73

(Continued) Quintile 1

Quintile 2

Quintile 3

Quintile 4

Quintile 4 Quintile 5

0.5552 0.4061

0.6364 0.4832

0.8128 0.6135

0.7768

Quintile 2 Quintile 3 Quintile 4 Quintile 5

0.8232 0.6786 0.5435 0.1627

0.8673 0.7500 0.3731

0.8059 0.5271

0.6933

Quintile 2 Quintile 3 Quintile 4 Quintile 5

0.7586 0.6003 0.5321 0.3678

0.7424 0.6263 0.3120

0.7643 0.5103

0.5475

The coefficient on fer is positive. Although the coefficient is statistically insignificant, its sign contradicts both received wisdom and the negative unconditional correlation between ass and fer. One explanation for the positive coefficient (or at least, for an effect offsetting the usual negative impact of high fertility, leading to a coefficient close to zero) is that for a given level of education and health, higher fertility leads to larger households, and larger households are able to acquire more assets. It is not possible to test this hypothesis directly, because household size is not reported in the data set. This is magnified if there are scale economies in some types of household production. None of this implies that higher fertility is good for material prosperity in equilibrium, because – as we see shortly – higher fertility could be bad for education and health, and therefore bad for material prosperity overall. The second part of Table 3.5 reports the results for the wtr equation. Here, the significant φ-coefficients – all positive – are those on size, britain and france. The first of these might reflect the fact that ass is not identified in the wtr equation: it is likely that access to piped water is more extensive in countries where households have access to other material assets. The second two suggest some positive influence of British and French colonization. There are significant β-coefficients on fertility and mortality, but not on schooling. Households with poorer health outcomes are less likely to be able to afford piped water. A 1 per cent increase in the mortality variable is associated with a fall in w/(1 − w) of around 1.5 per cent. More surprisingly, a 1 per cent increase in fertility is associated with an increase in w/(1 − w) of around 0.75 per cent. One explanation for this

74 David Fielding, Mark McGillivray and Sebastián Torres

effect is household economies of scale, as discussed above. Another is that households with an unexpectedly large number of children value sanitation more at the margin, and use resources that would otherwise be spent elsewhere to gain access to clean water. The third part of Table 3.5 reports the results for the sch equation. Here, the statistically significant φ-coefficients are those on chrs, musl and britain. Ceteris paribus, countries with a relatively large Christian population and those colonized by Britain can expect to have relatively high education levels. All three of the identified β-coefficients are large and statistically significant. On average, more prosperous households invest in more education: a rise in the assets index by 1 percentage point is associated with a level of sch/(1 − sch) about 2 per cent higher. But for a given level of assets, healthier households also invest in more education. A 1 per cent reduction in mor is associated with a level of sch/(1 − sch) that is about 0.3 per cent higher. Why is this so? Surely education and healthcare make competing claims on household resources? One reason for this is that caring for the sick and dying takes up time that would otherwise be spent learning. Another is that a high rate of child mortality reflects the poor health status of the parents, in whose education few resources have been invested because sickness reduces the returns to schooling. Unfortunately, there is no information in the data set that would shed light on which of these reasons is more important. By contrast, an increase in wtr by 1 percentage point reduces sch/(1 − sch) by about 0.75 per cent. This may reflect that fact that, ceteris paribus, a higher level of expenditure on sanitation leaves fewer resources for education. Another potential explanation is that, for a given level of material prosperity and health, households that already have access to clean water are more likely to choose a large number of children, and that a higher fertility rate is associated with less education. The impact of wtr in the fer equation is discussed below; remember that fer is not identified in the sch equation. The fourth part of Table 3.5 reports the results for the fer equation. Here, the statistically significant φ-coefficients are those on chrs, musl, africa and britain. Fertility is higher in countries with large Christian and Muslim populations, especially if they were British colonies, but, given these characteristics, lower in Africa than elsewhere. There is a large and significant β-coefficient on mor. That is, a higher level of child mortality leads to a higher fertility rate: to some extent, parents will seek to replace the children they have lost. A 1 per cent increase in mortality leads to an increase in fertility of around 0.3 per cent. There is also a negative coefficient on ass, a 1 per cent increase in the assets index being associated with a reduction in fertility of just under 1 per cent. On average, families

Aid and the MDGs

75

with more material assets have fewer children. However, a percentage point increase in the fraction of houses with piped water increases fertility slightly. For a given level of assets and a given child mortality rate, better sanitation facilities are associated with a decision to have more children. This suggests that better sanitation reduces the costs of bearing and raising an infant. It also ties in with the negative coefficient on wtr in the sch equation, discussed above. The fifth part of Table 3.5 reports the results for the mor equation. Here, the statistically significant φ-coefficients are those on coast, africa, britain and malfal. Mortality rates are higher in countries with a climate favourable to malaria-bearing mosquitoes but lower in a maritime environment. Given these characteristics, they are still higher in Africa. (Here is the part of explanation for the negative Africa dummy in growth regressions: Africa is not just unusually inefficient, it is also unusually unhealthy.) Three of the four β-coefficients are statistically significant. Higher levels of sanitation and education are associated with lower mortality rates. The first effect reflects the importance of sanitation for health: a percentage point increase in the fraction of houses with piped water reduces mortality by around 0.5 per cent. The second reflects either the complementarity of investment in education and investment in health, or a beneficial effect of education on household hygiene and therefore health outcomes. A percentage point increase in the fraction of households with primary education leads to a reduction in mortality of around 1 per cent. Finally, a 1 per cent increase in fertility increases child mortality by around 0.67 per cent. A higher birthrate increases the risks facing each individual child. Conditional on these effects, variations in the level of household assets have no impact on mortality. Table 3.6 presents some descriptive statistics for the Table 3.5 model. Parts (a) and (b) of Table 3.6 show that the model explains a relatively small fraction of the sample variation in the characteristics of households at the bottom of their national asset distributions (quintiles 1 and 2), and a relatively large fraction of the corresponding variation for their more prosperous neighbours (quintiles 4 and 5). This difference is manifested in a systematic pattern in the R2 statistics for our 20 regressions, and consequently in a systematic pattern in the corresponding equation standard errors, which are lower for the higher-asset quintiles. Non-modelled country-specific effects play a larger role in determining the outcomes for the poor than they do in determining the outcomes for the rich. The reasons for this discrepancy are an important subject for future study. Parts (c) and (d) of Table 3.6 report some of the residual correlations from the fitted model. Part (c) reports the average value of between-indicator

76 David Fielding, Mark McGillivray and Sebastián Torres Table 3.7 Equilibrium effects on each variable of 1 standard error shock to each equation Equation

ass increase Std deviation wtr increase Std deviation sch increase Std deviation fer decrease Std deviation mor decrease Std deviation Sum

ass ↑ 1 s.e.

wtr ↑ 1 s.e.

sch ↑ 1 s.e.

fer ↓ 1 s.e.

mor ↓ 1 s.e.

0.1988 0.0658 0.0732 0.0669 0.1776 0.0933 0.3153 0.1225 0.4254 0.2015 1.1903

−0.0157 0.0194 0.2055 0.0577 −0.0630 0.0371 −0.0336 0.0392 0.0224 0.0621 0.1156

0.1250 0.0690 0.0890 0.0733 0.2339 0.0963 0.2660 0.1267 0.4666 0.2052 1.1805

0.0524 0.0309 0.0429 0.0363 0.0676 0.0401 0.3404 0.0581 0.3232 0.0905 0.8265

0.1334 0.0678 0.1892 0.1252 0.1414 0.0824 0.3650 0.1238 0.7529 0.2116 1.5819

Note: Each column relates to a shock to a particular equation, and each row to the impact on a particular variable. Figures in bold record the mean effect across all countries and quintiles; figures in italic record the corresponding standard deviations.

correlations. These are generally quite small, so there is no substantial unexplained co-movement in our wellbeing indicators. However, in part (d) we see a positive correlation coefficient for the individual dependent variables across quintiles, which suggests that random variations in country-specific characteristics do play a role in determining outcomes for each particular wellbeing indicator, conditional on the observed levels of the others. Some implications of the results The φ-coefficients in Table 3.5 indicate the partial derivatives of each wellbeing indicator with respect to the others. They generally show beneficial associations: that is, improvements in one indicator are associated typically with improvements in the others. (Remember, of course, that in the case of fertility and mortality, improvements correspond to a fall in the indicator.) However, there are a number of exceptions to this general pattern. In particular, wider access to piped water appears to be associated with higher fertility rates and less education. Table 3.7 provides some information on the relative importance of the different interactions. It shows the impact on every variable in the system (ass, wtr, sch, fer, mor) of a shock to each equation in turn.

Aid and the MDGs

77

The effects reported are equilibrium effects, allowing for all of the interactions between the different wellbeing indicators. The shocks are normalized on the standard errors in the Table 3.5 equations. This normalization is an important part of the interpretation of the results: we are using the equation standard errors as an indication of the conditional variation that we can typically expect in each variable. In all cases we consider ‘beneficial’ shocks, so the shocks to the ass, wtr and sch equations are positive, while the shocks to the fer and mor equations are negative. The effects recorded are the predicted increases in ass, wtr and sch, and the predicted reductions in fer and mor. Each column in the table corresponds to a certain shock, while each row corresponds to a certain variable affected by the different shocks. In this context, a shock can be interpreted as an idiosyncratic effect that makes one of the wellbeing indicators for a particular quintile in a particular country better than one could expect, given the characteristics of that quintile and that country. The nonlinearity of the model means the effects depend on the initial values of the variables. For the purposes of Table 3.7, we compute the effects in 240 different cases; in each case, as starting values we use the observations actually recorded for one point in our data set. The main figures in Table 3.7 are means for the 240 observations; the table also includes the corresponding standard deviations. It can be seen from the table that, in general, the shocks with the largest beneficial effect are those to the mortality equation. The effects of shocks to the ass and sch equations are a little smaller – about three-quarters the size of the mor effects on average. The effects of shocks to the fer equation are slightly smaller again. However, the outstanding feature in the table is the impact of shocks to the wtr equation. As we have already seen, improvements in access to piped water are associated with higher fertility and lower standards of schooling. This means that in equilibrium, a beneficial shock to the wtr equation actually worsens the schooling and fertility outcomes on average, and as a consequence also the material assets outcome: the relevant figures in the wtr column in Table 3.7 are negative. Nevertheless, the detrimental feedback between wtr, sch and fer is offset by beneficial feedback between wtr and mor, so all of the equilibrium effects of a shock to wtr on the other variables are close to zero on average. Improving access to piped water may be valuable in its own right, but it does not seem to facilitate other dimensions of wellbeing, on average. By contrast, anything that helps to generate idiosyncratic improvements in one of the other wellbeing indicators will, on average, improve all of the rest; these effects are largest for improvements in mortality. The conflicting feedback effects in wtr also explain why the figures

78 David Fielding, Mark McGillivray and Sebastián Torres

in the wtr row in Table 3.7 are smaller than those in other rows: beneficial shocks to other variables do not improve water access so much, on average. The impact of aid In this subsection we investigate what the fitted model tells us about the changes in wellbeing outcomes we can expect to see if the amount of aid allocation to each country increases. However, all of the following remarks apply equally as predictions of the consequences of an improvement in the ‘effectiveness’ of a given amount of aid, which would be captured by a proportional change in the size of the aid coefficient in each equation. While the impact of aid on wellbeing outcomes is an interesting topic in its own right, we are specifically interested in whether there is empirical support for the international community’s strategy of achieving the MDGs through, inter alia, increased aid flows. Such support does appear to exist, based on the results we now report and discuss. Table 3.5 shows that there is a statistically significant direct (partial derivative) effect of aid on 4 out of the 5 wellbeing indicators; the exception is schooling. Ceteris paribus, a 1 per cent increase in aid will raise ass/(1 − ass) by somewhat less than 0.1 per cent and wtr/(1 − wtr) by about 0.4 per cent; it will reduce mor by a little over 0.1 per cent. Household assets, sanitation and mortality do improve significantly if aid rises. However, there is also a 0.1 per cent increase in fer: the extra income associated with aid inflows tends to raise birthrates on average. To the extent that higher fertility is detrimental to other wellbeing objectives, this effect will tend to undermine the effectiveness of aid flows. The estimates of these direct effects do not take into account any of the beneficial or detrimental interactions between the wellbeing indicators. In order to see how the interactions are likely to affect the consequences of an increase in aid, we need to solve out the Table 3.5 model. Again, it is important to note that the nonlinearity of the equations means that the impact of an increase in aid will not be uniform across all quintiles and countries. The magnitude of the response of the different wellbeing indicators will depend on their initial values. For this reason, Table 3.8 shows the predicted change in each of our wellbeing indicators with a doubling of aid in each quintile and in each country. In this table, each figure measures the difference in percentage terms14 between the current value of the indicator and the value predicted with a doubling of aid in the solved-out model. The table shows improvements in ass, wtr, sch and mor in all 240 cases. Schooling improves as a result of improvements in

Table 3.8

Predicted percentage change in each variable for a 100 per cent increase in aid Assets Q1

Q2

Bangladesh 2.8 6.0 Benin 1.7 3.4 Bolivia 5.7 10.1 Brazil 5.2 10.4 Burkina 1.0 1.6 Cambodia 3.6 6.3 Cameroon 2.7 7.2 C.A.R. 1.9 3.6 Chad 1.7 2.0 Colombia 9.0 7.7 Comoros 2.5 3.7 Côte d’Ivoire 2.2 4.7 Dom. Rep. 8.4 10.0 Egypt 6.9 9.7 Ethiopia 1.5 2.0 Gabon 11.9 10.4 Ghana 3.9 7.8 Guatemala 4.8 7.2 Guinea 1.4 2.3 Haiti 2.8 5.4 India 2.6 7.0 Indonesia 6.3 13.4 Jordan 8.3 5.9 Kenya 6.8 9.4

Water Q5

Q1

Q2

Schooling

Q3

Q4

Q3

Q4

Q5

Q1

11.2 5.4 8.4 8.1 2.5 9.1 12.4 6.4 2.9 5.2 8.5 7.1 8.3 8.7 2.3 8.2 8.8 9.8 4.2 8.6 11.5 11.7 4.5 9.2

10.7 5.7 1.5 7.7 17.2 24.6 25.9 4.2 9.5 7.5 0.7 5.7 8.4 20.8 27.4 2.6 4.7 2.3 8.8 27.5 27.0 15.8 4.3 5.9 5.4 4.1 0.6 9.0 16.9 25.2 24.9 8.5 3.5 10.1 0.4 2.0 5.6 8.0 26.5 1.5 10.3 5.0 2.7 14.6 24.2 28.8 20.2 4.1 9.8 5.5 0.6 7.7 17.3 21.5 25.3 4.0 8.0 9.4 0.9 6.5 13.8 20.7 25.4 3.1 4.7 9.5 3.2 4.1 7.2 14.6 29.1 1.9 2.8 2.2 22.8 27.6 22.0 8.8 8.8 6.4 11.4 9.1 0.9 3.3 8.5 23.7 23.8 3.2 9.1 7.3 1.9 12.0 20.8 27.5 24.8 2.8 5.6 3.0 2.4 9.9 16.6 25.4 17.4 12.2 5.7 2.1 23.9 25.5 19.7 9.4 1.0 4.2 3.8 12.3 0.5 1.4 1.2 3.3 14.9 2.3 7.4 4.2 9.4 13.4 16.4 18.7 20.8 14.6 7.2 4.8 2.3 12.1 12.8 20.5 22.0 6.4 6.7 3.5 10.9 25.5 25.8 25.7 18.1 4.5 8.6 9.2 0.1 1.7 5.2 13.6 24.2 1.6 11.1 6.3 2.4 12.8 16.8 25.5 24.8 3.0 8.7 4.7 0.4 4.8 9.8 20.3 24.4 4.5 7.0 3.1 1.1 16.3 21.4 26.8 13.1 7.5 3.6 1.8 24.3 7.7 5.9 4.5 0.6 4.5 8.2 5.1 3.2 12.2 24.1 26.8 21.0 10.0

Fertility

Q2

Q3

Q4

Q5

Q1

Q2

6.9 3.9 7.2 11.2 2.0 4.8 8.3 4.4 2.0 3.5 3.6 4.2 10.1 6.7 2.5 10.0 7.6 4.0 2.5 4.2 9.6 14.5 4.5 10.1

11.5 8.5 1.4 0.1 −5.3 6.1 8.4 3.3 2.2 −0.7 4.4 1.8 1.4 −4.6 −11.5 5.7 1.1 0.2 −4.7 −12.5 2.5 3.5 7.5 3.6 2.5 6.3 7.1 2.3 −0.8 −5.1 13.0 7.6 1.2 0.4 −7.3 6.2 5.4 6.9 1.7 −1.3 2.4 3.2 5.9 2.4 1.9 1.7 1.3 0.8 −9.5 −7.0 9.4 10.0 6.5 1.0 −0.7 5.3 5.9 3.6 1.4 −2.8 6.0 1.4 0.4 −10.2 −11.5 6.5 4.5 1.6 −6.2 −10.6 2.9 4.6 12.8 2.6 1.9 5.8 4.3 0.7 −15.9 −12.1 8.0 3.3 0.6 −2.3 −8.0 7.1 2.4 0.4 −3.0 −6.5 4.4 8.8 6.1 3.1 1.5 7.5 9.0 2.5 0.6 −3.6 13.5 5.8 0.6 0.2 −7.6 10.9 3.2 1.7 −5.4 −18.0 3.2 2.3 1.1 −7.8 −3.9 6.1 4.1 1.3 −7.5 −11.1

Mortality Q3

Q4

Q5

Q1

Q2

Q3

Q4

Q5

−14.2 −4.3 −8.2 −7.7 1.0 −9.6 −16.2 −5.9 0.5 −2.8 −9.4 −6.7 −8.1 −9.0 1.3 −7.9 −9.3 −11.0 −1.8 −9.2 −14.9 −14.7 −1.6 −9.7

−12.6 −11.0 −1.9 −3.0 −0.8 −11.7 −10.9 −7.7 −2.4 1.1 −14.1 −9.7 −3.4 −3.9 −1.4 −6.4 −5.7 −5.2 −9.6 −13.4 −8.6 −6.0 0.0 −7.8

−3.6 −6.6 1.9 −1.0 −11.6 −2.8 −3.2 −10.4 −10.3 2.0 −9.7 −6.4 0.6 2.3 −15.8 −1.1 −1.8 0.0 −9.7 −4.7 −2.0 0.2 2.8 −2.5

−13.3 −9.8 −22.0 −20.4 −7.6 −14.4 −12.4 −10.8 −10.4 −33.2 −11.4 −11.2 −28.8 −29.4 −9.1 −38.8 −17.7 −20.7 −7.9 −12.2 −12.9 −20.0 −30.9 −25.2

−22.9 −15.8 −37.8 −32.6 −9.7 −24.3 −25.7 −17.0 −11.2 −31.0 −14.2 −20.7 −31.3 −35.6 −10.3 −33.6 −27.7 −30.1 −10.7 −21.7 −25.7 −43.6 −19.5 −32.4

−38.5 −21.8 −32.5 −28.1 −13.1 −33.8 −41.5 −25.8 −14.3 −23.5 −28.6 −29.2 −28.5 −31.2 −10.9 −28.0 −29.3 −36.4 −16.6 −31.0 −37.1 −40.5 −15.7 −33.6

−38.3 −35.1 −19.7 −24.7 −16.6 −38.5 −34.6 −29.9 −20.8 −13.5 −40.4 −35.2 −25.4 −20.3 −15.7 −26.7 −26.1 −27.8 −30.8 −39.9 −30.6 −29.7 −12.9 −31.7

−25.8 −30.4 −10.7 −22.3 −37.6 −23.2 −24.9 −35.6 −36.5 −12.4 −33.8 −29.3 −17.6 −8.8 −39.7 −20.7 −21.7 −18.2 −33.7 −27.0 −23.1 −16.8 −7.8 −22.3

(Continued)

Table 3.8

(Continued) Assets Q1

Water

Q2

Q3

Q4

Q5

Q1

Q2

Schooling Q3

Q4

Q5

Q1

Fertility

Q2 Q3

Q4

Q5

Q1

Mortality

Q2

Madagascar 2.2 3.6 5.4 8.6 7.4 1.9 7.5 14.5 26.8 27.4 3.4 4.0 4.6 5.5 3.4 1.1 −1.2 Malawi 1.8 3.8 5.1 7.3 7.4 0.2 2.7 3.4 7.9 19.7 4.2 6.3 7.7 8.6 3.9 1.3 −2.2 Mali 1.2 1.8 2.5 4.0 9.8 1.0 4.4 7.0 12.1 29.2 1.7 1.9 2.0 2.9 6.3 3.2 2.2 Mauritania 3.9 5.5 6.6 9.8 7.1 18.2 25.8 25.6 27.6 15.2 1.1 1.4 2.5 6.5 6.3 −0.8 −3.3 Morocco 3.8 6.5 10.8 9.5 3.8 16.5 26.7 27.3 25.9 5.5 1.3 2.5 8.2 6.8 4.0 −0.6 −5.0 Mozambique 1.1 2.6 3.7 6.4 10.4 0.1 2.8 5.1 13.7 28.6 1.9 3.3 4.1 5.6 7.3 3.3 0.7 Namibia 8.8 9.1 7.3 5.4 2.1 28.6 28.4 20.1 18.2 2.7 4.9 5.4 4.0 1.3 0.4 −9.0 −9.6 Nepal 2.7 4.4 6.7 8.8 6.7 2.2 9.7 20.0 24.8 26.2 4.3 4.8 5.1 6.4 2.0 0.1 −2.6 Nicaragua 6.5 9.2 6.4 4.4 1.6 24.6 26.4 18.1 13.3 1.6 3.5 5.8 3.3 2.0 1.1 −5.3 −9.7 Niger 1.0 1.1 1.7 2.6 9.4 0.6 0.9 2.0 4.9 28.5 1.0 1.1 1.4 2.1 5.9 3.9 3.6 Nigeria 1.3 3.0 6.0 7.7 5.6 0.0 0.4 1.7 5.6 14.1 3.2 5.7 9.2 8.7 1.8 2.3 −0.8 Pakistan 3.1 5.0 7.4 8.9 5.3 6.8 14.5 22.4 28.6 16.2 3.2 4.0 5.2 5.1 2.4 −0.2 −3.2 Paraguay 9.4 9.5 7.6 4.5 1.7 27.6 25.7 27.4 16.6 2.3 6.3 6.4 3.4 1.3 0.9 −10.3 −10.3 Peru 10.6 11.4 6.0 4.3 1.6 9.8 24.3 26.2 21.5 3.3 12.1 9.8 2.0 1.0 0.9 −13.2 −14.0 Philippines 13.0 9.6 6.4 4.7 3.1 6.8 20.0 22.6 24.4 17.6 16.7 7.6 2.5 0.5 0.4 −17.9 −10.6 Rwanda 1.4 2.7 4.5 7.6 11.0 0.0 0.4 0.9 2.1 13.0 2.9 4.7 6.9 11.0 10.9 2.4 0.0 South Africa 8.4 6.9 5.9 5.2 2.3 5.6 17.0 24.6 24.7 6.6 10.1 3.6 0.8 0.3 0.3 −9.5 −5.2 Tanzania 4.9 7.2 8.9 10.6 7.5 3.9 12.5 11.1 20.5 24.8 7.1 7.9 10.4 9.6 3.4 −3.9 −7.5 Togo 2.0 4.0 7.3 12.6 7.9 0.6 4.1 6.7 13.7 21.4 2.9 5.2 9.0 14.2 4.6 1.7 −1.9 Uganda 2.4 4.5 6.7 8.1 6.5 1.0 7.1 11.3 18.7 25.1 5.0 5.9 7.1 6.0 1.5 0.1 −3.1 Vietnam 11.4 8.0 4.6 3.5 1.6 20.5 27.1 13.0 9.6 1.5 10.8 4.4 3.0 2.3 1.6 −14.2 −7.6 Yemen 3.9 6.9 9.6 9.7 5.1 5.2 22.3 29.4 28.1 12.8 4.5 4.5 6.1 6.2 2.9 −1.5 −6.1 Zambia 3.8 6.5 9.0 7.9 5.2 2.2 11.2 12.8 16.8 23.3 6.4 7.3 9.5 5.3 0.4 −2.2 −6.4 Zimbabwe 9.6 9.3 8.0 6.7 4.7 7.1 20.4 20.2 23.8 21.8 12.4 7.4 4.8 2.0 0.4 −12.1 −10.2 Average

4.7

6.2

7

7.1

5.7

6.6 12.8 15.4 18.6 17.9

5.3 5.6

5.8

5

3

−2.9

−5.3

Q3

Q4

Q5

Q1

Q2

Q3

Q4

Q5

−3.9 −4.4 1.1 −5.0 −12.8 −1.1 −6.2 −6.1 −4.7 2.8 −6.0 −7.1 −6.8 −4.2 −4.7 −3.1 −3.4 −10.5 −7.8 −6.4 −2.1 −10.5 −10.3 −7.3

−8.9 −7.6 −1.3 −10.8 −10.5 −5.6 −2.5 −9.5 −1.5 1.1 −8.0 −9.2 −1.4 −1.5 −1.8 −8.7 −2.4 −12.8 −16.8 −7.9 −0.3 −10.6 −7.2 −4.8

−6.6 −6.3 −10.9 −6.7 −1.2 −12.0 2.7 −5.0 3.0 −10.2 −2.7 −2.9 2.9 3.0 0.5 −13.2 2.2 −6.5 −7.3 −4.4 2.8 −2.6 −2.3 −1.7

−12.0 −12.0 −8.3 −19.6 −18.7 −8.1 −34.4 −13.7 −28.4 −7.0 −10.1 −15.2 −36.2 −34.6 −40.8 −9.7 −27.9 −20.2 −10.4 −13.9 −39.6 −16.7 −17.5 −32.7

−17.2 −17.7 −11.0 −25.5 −28.2 −12.7 −35.1 −20.0 −34.5 −7.4 −14.9 −22.3 −35.2 −40.4 −33.6 −13.3 −24.2 −27.9 −17.0 −20.1 −32.1 −28.7 −25.9 −33.3

−23.2 −20.9 −13.3 −27.6 −39.6 −15.9 −27.1 −28.1 −24.3 −8.9 −22.7 −30.2 −30.7 −27.0 −25.8 −17.9 −24.2 −31.7 −26.3 −25.7 −19.6 −36.9 −31.5 −28.6

−34.0 −26.2 −18.5 −36.7 −35.9 −24.9 −20.8 −34.2 −18.3 −12.2 −25.4 −34.7 −19.2 −21.6 −22.9 −26.5 −23.2 −37.3 −41.2 −29.5 −15.9 −36.5 −27.2 −26.0

−30.5 −26.7 −37.3 −27.2 −16.0 −38.8 −8.2 −27.5 −8.1 −36.2 −19.2 −21.2 −8.4 −8.9 −17.6 −35.0 −10.5 −29.1 −29.1 −25.9 −8.8 −19.7 −22.5 −21.3

−6.5

−6.4

−3.7 −19

−24.2 −26.6 −27.4 −23.2

Aid and the MDGs

81

the other wellbeing indicators, even though aid has no significant direct effect on this variable. The largest improvements in sch (14 per cent or more) are predicted for quintile 1 in the Philippines and Gabon. The largest improvements in wtr (28 per cent or more) are predicted for quintile 1 in the Philippines, quintiles 3–4 in the Yemen, quintile 4 in Cambodia and quintile 5 in Niger. The largest improvements in mor (40 per cent or more) are predicted for quintile 1 in the Philippines, quintile 3 in Indonesia and quintile 4 in Comoros and Togo. The overall effects on fertility are quite varied. As we have already seen, the direct effect of aid on fertility is positive. However, this is offset by a strong bidirectional link between fertility and mortality. To the extent that aid reduces child mortality, it induces a virtuous spiral in which lower mortality reduces the number of children a family needs to have, which in turn improves child health. So the predicted equilibrium effect of aid is positive in some cases and negative in others. In the average country, fertility falls slightly. Table 3.9 shows the average impact of the 100 per cent increase in aid by quintile, along with corresponding figures for 10 per cent and 50 per cent increases. Either quintile 3 or quintile 4 benefits most on average from the 100 per cent increase with respect to all wellbeing outcomes. Quintile 1, the poorest, does worst from the 10 per cent and 50 per cent increases with respect to all five wellbeing indicators under question; quintiles 3 and 4 do best. Moreover, in the majority of instances the standard deviations for quintile 1 are larger than standard deviations for other quintiles. Overall, the poorest in each country benefit least from increases in aid, and their experience is more varied. Table 3.10 further summarizes the predicted consequences of an increase in aid. Like Table 3.9, Table 3.10 shows, for each wellbeing indicator, the average change predicted with a 100 per cent increase in aid. But this time, mean effects are calculated for a group of absolutely rather than relatively poor (and not-so-poor) quintiles. That is, our first aggregate is formed from a group of quintiles with poor outcomes relative to the whole sample rather than to the respective countries in which they are located. Also, poverty is defined in terms of observed values of the wellbeing indicators, with a different aggregation for each indicator. Two further aggregates are constructed for each indicator, one for a ‘middle’ group and one for a ‘high’ group. For assets, water and schooling, the ‘low’ group consists of quintiles for which the existing measure is less than 20 per cent, the ‘middle’ group consists of quintiles for which the existing measure is between 20 per cent and 50 per cent, and the ‘high’

Percentage increases in each wellbeing indicator with 10, 50 and 100 per cent increases in aid, disaggregated by

Assets 10% increase in aid 50% increase in aid 100% increase in aid Water 10% increase in aid 50% increase in aid 100% increase in aid Schooling 10% increase in aid 50% increase in aid 100% increase in aid Fertility 10% increase in aid 50% increase in aid 100% increase in aid Mortality 10% increase in aid 50% increase in aid 100% increase in aid

Quintile 1

Quintile 2

Quintile 3

Quintile 4

Quintile 5

Mean

Mean

Mean

Mean

Mean

Std dev.

Std dev.

Std dev.

Std dev.

82

Table 3.9 quintiles

Std dev.

0.6 2.6 4.7

0.4 1.9 3.3

0.8 3.6 6.2

0.5 1.9 3.0

1.0 4.1 7.0

0.4 1.6 2.6

1.0 4.2 7.1

0.4 1.5 2.5

0.8 3.4 5.7

0.4 1.8 2.9

0.7 3.4 6.6

1.1 4.8 8.6

1.5 6.9 12.8

1.3 5.4 9.2

1.8 8.4 15.4

1.2 4.9 8.3

2.3 10.5 18.6

1.1 4.6 7.8

2.5 10.7 17.9

1.2 5.2 9.1

0.7 3.1 5.3

0.5 2.1 3.7

0.8 3.4 5.6

0.5 1.9 2.9

0.9 3.5 5.8

0.5 1.9 3.1

0.8 3.1 5.0

0.5 2.0 3.2

0.5 1.9 3.0

0.5 1.9 2.9

−0.3 −1.5 −2.9

0.7 3.3 5.8

−0.7 −3.1 −5.3

0.8 3.1 5.0

−0.9 −3.8 −6.5

0.7 2.7 4.4

−0.9 −3.9 −6.4

0.7 2.7 4.5

−0.6 −2.4 −3.7

0.8 3.0 5.0

−2.4 −10.8 −19.1

1.3 5.6 10.0

−3.2 −14.0 −24.2

1.4 5.6 9.2

−3.6 −15.4 −26.6

1.1 4.7 7.9

−3.8 −16.1 −27.4

1.1 4.7 8.0

−3.4 −14.0 −23.2

1.3 5.6 9.5

Notes: Figures in bold indicate mean percentage changes in each variable as a result of the increase in aid (that is, percentage point changes for ass, wtr and sch, and percentage growth for fer and mor). Figures in italic indicate corresponding standard deviations.

Aid and the MDGs

83

Table 3.10 Percentage increases in each wellbeing indicator with 10, 50 and 100 per cent increases in aid, disaggregated by initial wellbeing levels Aid increase ass change low initial ass (<20%) mid initial ass (<50%) high initial ass (≥50%) sch change low initial sch (<20%) mid initial sch (<50%) high initial sch (≥50%) mor change

Aid increase 10% 0.54 0.34 1.20 0.35 0.82 0.35 10% 0.39 0.17 0.88 0.31 0.79 0.62 10%

50% 2.40 1.49 5.11 1.37 3.37 1.41 50% 1.67 0.75 3.70 1.27 3.13 2.47 50%

100% 4.28 2.60 8.69 2.12 5.56 2.35 100% 2.87 1.34 6.29 2.12 5.01 3.88

wtr change low initial wtr (<20%) mid initial wtr (<50%) high initial wtr (≥50%) fer change low initial fer (<3.00) mid initial fer (<5.00) high initial fer (≥5.00)

10%

50%

100%

1.30 6.26 11.87 1.19 5.31 9.27 2.89 12.75 22.05 0.83 3.49 5.95 2.42 9.83 15.86 1.33 5.63 9.36 10%

50%

100%

−0.59 −2.27 −3.47 0.82 3.36 5.45 −1.03 −4.24 −7.01 0.63 2.50 4.04 −0.50 −2.27 −4.08 0.74 3.12 5.27

100%

low initial mor −3.59 −14.97 −24.97 (<10%) 1.22 5.25 9.02 mid initial −3.74 −16.21 −28.14 mor (<20%) 1.16 4.72 7.71 high initial −2.02 −8.98 −15.99 mor (≥20%) 0.92 4.07 7.21 Note: Figures in bold indicate mean percentage changes in each variable as a result of the increase in aid (that is, percentage point changes for ass, wtr and sch, and percentage growth for fer and mor). Figures in italic indicate corresponding standard deviations.

group consists of quintiles for which the measure is more than 50 per cent. For fertility, the three groups are for less than 3 children per woman, 3 to 5 children and more than 5 children. For child mortality, the three groups are for less than 100 per thousand, 100 to 200 per thousand and over 200 per thousand. There is some variation across low, medium and high groups in terms of the size of the predicted effect of an increase in aid. There is also substantial variation across the different indicators. The largest predicted improvements are in child mortality. With a 100 per cent increase in aid, mortality falls by 25 per cent or more on average for the low and medium groups and by over 15 per cent on average for the high group. Predicted improvements in piped-water access are slightly

84 David Fielding, Mark McGillivray and Sebastián Torres

smaller than this for middle and high groups, and substantially smaller for the low group. The average improvements predicted for schooling and fertility are smaller again, reflecting some of the conflicting effects discussed above, and for fertility there is a great deal of variation around the mean.

Conclusion This chapter presents the results of a cross-country empirical model of various wellbeing outcomes that estimates the relative importance of the many different causal links between material prosperity, water supply, education, fertility and health in a simultaneous equations system. Three of these outcomes – water, education and health – are targeted by the MDGs. Wellbeing indicators are measured separately for different material wellbeing quintiles within each country, and material wellbeing is defined in terms of the presence of a set of material attributes within the household. We also condition each wellbeing indicator on a measure of aid flows to the country, allowing for the potential endogeneity of aid. The interactions between the different wellbeing indicators are numerous and, on the whole, reinforcing. An improvement in one is typically – though not always – associated with an improvement in another. The variable in which a given improvement was found to be associated with the largest and widest improvements in the others was the health outcome variable, child mortality. This suggests that, of the MDG targets under consideration, that for child mortality should be given priority in strategies on these internationally agreed developmental goals. That the improvements are on the whole reinforcing means that a beneficial impact of aid on a wellbeing outcome does not rely on aid having a direct effect on that outcome. This turns out to be particularly important in the case of schooling. We do not find any statistically significant direct effect of aid on schooling, but the indirect effects through improvements elsewhere are often substantial. Our results suggest that aid can be expected to improve outcomes across a wide variety of wellbeing indicators, including sanitation and child health and basic household assets along with schooling. The size of the predicted effect varies across countries, across quintiles and across the indicators, but in almost all cases we predict an improvement. The largest gains are in child health. A doubling of aid allocated to

Aid and the MDGs

85

each country can be expected to reduce child mortality rates by at least 20 per cent in the majority of cases, and by over 40 per cent in the most extreme case. This reinforces the case for targeting child mortality in seeking to achieve the MDGs. It follows that our estimates of the impact of aid are predicated on the assumption that the composition of public expenditure and its relative incremental impact or efficiency among the wellbeing outcomes under consideration do not change as aid increases. If these factors change, then so may also the direct impact of aid on wellbeing indicators, and therefore also the effect on wellbeing outcomes in equilibrium, after allowing for interactions between the different indicators. However, we can also make some tentative predictions about the consequences of focusing aid on a particular wellbeing goal. Among the wellbeing outcomes we model, the largest positive spillover effects are from reductions in child mortality. Commenting on optimal aid allocation is beyond the scope of this chapter, because it requires the valuation of alternative wellbeing outcomes. Nevertheless, our results suggest that aid improving health outcomes will have a positive impact at least as large and widespread as improvements in any other direction, and that any reduction in health-related aid expenditure will have serious detrimental consequences for other aspects of social and economic wellbeing. Our finding that aid has substantial beneficial effects on indicators of wellbeing measured at the household level is consistent with the emerging consensus in international wellbeing circles that aid is indeed effective at the national level. It also provides some support for the international community’s strategy of looking to aid to help achieve the MDGs. Our finding that the impact of aid differs among households, with the poorest households seemingly benefiting least from these inflows, is one that warrants further research. This is a critical finding and if supported by subsequent research findings has enormous implications for aid-supported interventions such as the MDGs. One obvious implication is that while such interventions might increase overall living standards in developing countries, this will be at the cost of the living standards of the poor falling further behind those of the rich in these countries. More generally, the potential impact of aid on population subgroups within recipient countries is still a substantial void in the aid effectiveness literature. Very little is known about these impacts and the literature should devote much more attention to them.

86 David Fielding, Mark McGillivray and Sebastián Torres

Appendix Table 3.A1

Data sources for the conditioning variables

Variable Definition britain france ethno size natres chrs musl temp malfal

Dummy = 1 if colonized by Britain Dummy = 1 if colonized by France Ethno-linguistic fractionalization index Country surface area Natural resource capital value Fraction of the population that is Christian Fraction of the population that is Muslim Temperature (in 0.1◦ C) Fraction of population at risk from malaria

Source La Porta et al. (1998) La Porta et al. (1998) Krain (1997) CIA (1997) Dixon and Hamilton (1996) La Porta et al. (1998) La Porta et al. (1998) Hoare (2005) McArthur and Sachs (2001)

Notes 1. A review of the MDG literature is not attempted. Such a review is premature, given that the topic is still only an emerging one. Suffice to note that these studies, which mainly consist of reports from official donor and related agencies, have not sufficiently addressed the specific focus of this chapter. 2. Briefly, the theoretical rationale for the effects is as follows. Higher standards of education and health embody human capital investments that increase productivity and so per capita income. Higher fertility entails a higher rate of population growth, and so a lower capital–labour ratio and (with decreasing returns to labour) lower productivity. Education and health are also normal consumption goods, so expenditure on them increases with per capita income. High fertility is a consequence of a low opportunity cost of labour (especially female labour), and is therefore decreasing in per capita income. 3. See, for example, Dean Jamison’s comments at the IMF Economic Forum, Health, Wealth and Welfare, 15 April 2004, available at: www.imf.org/ external/np/tr/2004/tr040415.htm 4. Surveys of the aid growth literature can be found in Hansen and Tarp (2000); Morrissey (2001); Clemens et al. (2004); Addison et al. (2005); McGillivray et al. (2006). 5. It should be acknowledged that the terms ‘economic’ and ‘non-economic’ wellbeing are used rather loosely. Greater achievement in a non-economic wellbeing dimension is taken simply to be something other than an increase in income. 6. The work by McGillivray and Morrissey (2004) provides a survey of these studies.

Aid and the MDGs

87

7. It should, though, be emphasized that these studies usually adopt a twofold disaggregation of expenditure, such as recurrent and capita expenditures. The level of aggregation is such that it might be dubious to infer anything regarding health and education expenditures. 8. One of the many criticisms of income per capita as an achieved wellbeing measure is that it does not capture outcomes. To this extent, the criticism just outlined applies also to the aid growth literature. 9. A composite index, the HDI contains three wellbeing outcome variables: years of life expectancy at birth, the adult literacy rate and the gross school enrolment rate (UNDP 2005). The last of these indicators should be seen as a quasi-wellbeing outcome variable, in that attending school does not necessarily mean the acquisition of knowledge, especially in developing countries. Note that the studies just cited looked at the HDI as a whole and not at aid’s impact on its individual component variables. 10. For six countries – Armenia, Eritrea, Kazakhstan, Kyrgyz Republic, US Virgin Islands and Uzbekistan – data on one or more of the conditioning variables in our regression equations were absent, so these countries are excluded from our analysis. See note 11 for the reasons for also excluding Turkey. 11. There is one exception to this statement: the presence of an electricity supply is not recorded in Turkey, where one might assume that all households have access to electricity. Turkish data are excluded from our analysis: Turkey is something of an outlier in the data set, being by far the richest country surveyed. 12. Temperature might affect the value of agricultural land and so factor productivity and material wealth, but we are already using natres to control for the value of natural resources in the ass equation. 13. Logistic regression equations produce results very similar to the ones reported here, but with slightly larger standard errors. 14. To be more precise: for ass, wtr and sch the figures record the predicted change in the indicator as measured on a 0–100 scale; for fer and mor they record the predicted proportional change in the indicator in percentage terms.

References Addison, T., G. Mavrotas and M. McGillivray (2005) ‘Aid, Debt Relief, and New Sources of Finance for Meeting the Millennium Development Goals’. Journal of International Affairs, 58(2):113–27. Ahlburg, D. (1996) ‘Population Growth and Poverty’. In D. Ahlburg, A. Kelley and K. Oppenheim Mason (eds), The Impact of Population Growth on Wellbeing in Developing Countries. Berlin: Springer. Becker, G. (1981) A Treatise on the Family. Cambridge, MA: Harvard University Press. Bloom, D., D. Canning and J. Sevilla (2004) ‘The Effect of Health on Economic Growth: A Production Function Approach’. World Development, 32(1):1–13. Boone, P. (1996) ‘Politics and the Effectiveness of Aid’. European Economic Review, 40(2):289–329. Burnside, C. and D. Dollar (2000) ‘Aid, Policies and Growth’. American Economic Review, 90(4):847–68.

88 David Fielding, Mark McGillivray and Sebastián Torres Chauvet, L. and P. Guillaumont (2002) ‘Aid and Growth Revisited: Policy, Economic Vulnerability and Political Instability’. Paper presented at the Annual Bank Conference on Development Economics Towards Pro-Poor Policies, June. Oslo. CIA (1997) World Factbook. Washington, DC. Clemens, M., S. Radelet and R. Bhavnani (2004) ‘Counting Chickens when They Hatch: The Short-term Effect of Aid on Growth’. Centre for Global Development Working Paper 44. Washington, DC. Collier, P. and A. Hoeffler (2004) ‘Aid, Policy and Growth in Post-Conflict Societies’. European Economic Review, 48(5):1125–45. Dalgaard C., H. Hansen and F. Tarp (2004) ‘On the Empirics of Foreign Aid and Growth’. Economic Journal, 114(496):F191–F216. Dixon, J. and K. Hamilton (1996) ‘Expanding the Measure of Wealth’. Finance and Development, 33(4):15–18. Easterly, W. and R. Levine (1997) ‘Africa’s Growth Tragedy: Policies and Ethnic Divisions’. Quarterly Journal of Economics, 112(4):1203–50. Fernandez, R. and R. Rogerson (1997) ‘The Determinants of Public Education Expenditures: Evidence from the States, 1950–1990’. National Bureau of Economic Research Working Paper 5995. Cambridge, MA. Feyzioglu, T., V. Swaroop and M. Zhu (1998) ‘A Panel Data Analysis of the Fungibility of Foreign Aid’. World Bank Economic Review, 12(1):29–58. Franco-Rodriguez, S., M. McGillivray and O. Morrissey (1998) ‘Aid and the Public Sector in Pakistan: Evidence with Endogenous Aid’. World Development, 26(7):1241–50. Gang, I. N. and H. A. Khan (1991) ‘Foreign Aid, Taxes and Public Investment’. Journal of Development Economics, 34(1–2):355–69. Gomanee, K., S. Girma and O. Morrissey (2005a) ‘Aid, Public Spending and Human Welfare: Evidence from Quantile Regressions’. Journal of International Development, 17(3):299–309. Gomanee, K., O. Morrissey, P. Mosley and A. Verschoor (2005b) ‘Aid, Government Expenditure, and Aggregate Welfare’. World Development, 33(3):355–70. Guillaumont, P. and L. Chauvet (2001) ‘Aid and Performance: A Reassessment’. Journal of Development Studies, 37(6):66–87. Hansen, H. and F. Tarp (2000) ‘Aid Effectiveness Disputed’. Journal of International Development, 12(3):375–98. Hansen, H. and F. Tarp (2001) ‘Aid and Growth Regressions’. Journal of Development Economics, 64(2):547–70. Heller, P. S. (1975) ‘A Model of Public Fiscal Behaviour in Developing Countries: Aid, Investment and Taxation’. American Economic Review, 65(3):429–45. Hoare, R. (2005) ‘World Climate’. Available at: www.worldclimate.com Hudson, J. and P. Mosley (2001) ‘Aid, Policies and Growth: In Search of the Holy Grail?’. Journal of International Development, 13(7):1023–38. Kosack, S. (2003) ‘Effective Aid: How Democracy Allows for Development Aid to Improve the Quality of Life’. World Development, 31(1):1–22. Krain, M. (1997) ‘State-Sponsored Mass Murder: The Onset and Severity of Genocides and Politicides’. Journal of Conflict Resolution, 41(3):331–60. La Porta, R., F. Lopez-de-Silanes, A. Shleifer and R. Vishny (1998) ‘Law and Finance’. Journal of Political Economy, 106(6):1113–55.

Aid and the MDGs

89

McArthur, J. and J. Sachs (2001) ‘Institutions and Geography: Comment on Acemoglu, Johnson and Robinson (2000)’. National Bureau of Economic Research Working Paper 8114. Cambridge, MA. McGillivray, M. (2005) ‘Measuring Non-economic Wellbeing Achievement’. Review of Income and Wealth, 51(2):337–64. McGillivray, M. and O. Morrissey (2004) ‘Fiscal Effects of Aid’. In T. Addison and A. Roe (eds), Fiscal Policy for Development: Poverty, Reconstruction and Growth. Basingstoke: Palgrave Macmillan for UNU-WIDER. McGillivray, M., S. Feeny, N. Hermes and R. Lensink (2006) ‘Controversy over the Macroeconomic Impact of Aid: It Works, It Doesn’t, It Might, but That Depends …’. Journal of International Development, 18(7):1031–50. Morrissey, O. (2001) ‘Does Aid Increase Growth?’. Progress in Development Studies, 1(1):37–50. Mosley, P., J. Hudson and A. Verschoor (2004) ‘Aid, Poverty Reduction and the New Conditionality’. Economic Journal, 114(496):F217–F243. OECD (Organisation for Economic Co-operation and Development) (2005) ‘International Development Statistics On-line’. Available at: www.oecd.org/ dataoecd/50/17/5037721.htm Pack, H. and J. R. Pack (1990) ‘Is Foreign Aid Fungible? The Case of Indonesia’. Economic Journal, 100(399):188–94. Pack, H. and J. R. Pack (1993) ‘Foreign Aid and the Question of Fungibility’. Review of Economics and Statistics, 75(2):258–65. Pritchett, L. and L. Summers (1993) ‘Wealthier Is Healthier’. World Bank Country Economics Department Working Paper 1150. Washington, DC. Strulik, H. and S. Siddiqui (2002) ‘Tracing the Income–Fertility Nexus: Nonparametric Estimates for a Panel of Countries’. Economics Bulletin, 15:1–9. Teulings, C. and T. van Rens (2003) ‘Education, Growth and Income Inequality’. Centre for Economic Policy Research Discussion Paper 3863. London. UNDP (United Nations Development Programme) (2003) Human Development Report, 2003. New York: Oxford University Press. UNDP (United Nations Development Programme) (2005) Human Development Report, 2005. New York: Oxford University Press. World Bank (2003) World Development Report 2004: Making Services Work for Poor People. New York: Oxford University Press. World Bank (n.d.) HNP Poverty Data. Available online.

4 Achieving the Water and Sanitation Millennium Development Goal P. B. Anand

Introduction Millennium Development Goal 7 (MDG7) focuses on ensuring environmental sustainability. There are serious concerns about whether countries that have made significant progress with regard to other MDGs have made much progress in respect of sustainability issues, especially in relation to forest cover, biodiversity and global climate change issues. Access to water and sanitation1 (MDG Target 10) is an important ingredient of quality of life and is also crucial to other MDGs, including reducing poverty and infant mortality, and improving maternal health, gender equality and educational opportunities for girls. According to the WHO-UNICEF (2004) assessment, by 2002 about 83 per cent of the global population were estimated to have access to ‘improved’ sources of water, compared with some 77 per cent in 1990,2 indicating clear progress towards achieving the target of halving the proportion without safe access by 2015. Much of this progress is attributed to increased access in South and East Asia. Coverage remains significantly low in much of sub-Saharan Africa (SSA), though there has been significant progress in a small number of countries. So, while globally the target may have been achieved, there will remain significant regional disparities. Also, the question remains whether the increased access is the same as sustainable access. For instance, in many cities in Southern Asia, a significant proportion of the population now depend on ‘unimproved sources’ such as tanker trucks for water supply. Groundwater resources are being depleted at significant rates, raising the question of whether supplies can be sustained even until 2015. Pressures on contested water resources are testing the fragile nature of water entitlements and property rights institutions, triggering interstate, intersectoral and community-level conflicts. 90

Achieving the Water and Sanitation MDG

91

It is highly unlikely that the development target of halving the proportion of people not having access to adequate sanitation will be achieved. In 2002, some 2.6 billion people worldwide did not have access to safe sanitation options. Of these, nearly 2 billion were in rural areas. While by 2002 the proportion of people having access to improved sanitation had increased in almost every country compared with 1990, still in 2002, in 27 countries, including Eritrea, Ethiopia, India, Laos, Namibia, Nepal and Yemen, 2 out of 3 people did not have access to improved sanitation. On the other hand, significant progress seems to have been made between 1990 and 2002 in countries like Pakistan, Paraguay, Senegal and Sri Lanka (United Nations 2002). There have been some previous assessments of progress with regard to MDG7 in general, and water and sanitation sectors in particular. The most notable study is that of the UN Millennium Project Task Force on Water and Sanitation (UNDP 2005). In that study, a number of issues related to water and sanitation sectors were examined and ten main recommendations made. These include the need to put water and sanitation firmly on the development agenda, the need to develop and strengthen various institutional mechanisms in the state, NGO and private sectors, the need to pursue a policy of recovering the cost of operations, maintenance and investment based on users’ willingness to pay while ensuring that the poor have access to the services, and so on. While that study emphasized the urgency to act, it did not, however, examine systematically the progress made so far in order to identify patterns in terms of whether progress where it has been made is associated with other development and governance indicators. An interim assessment by the WHO-UNICEF (2004) joint monitoring programme, which is responsible for monitoring Target 10 relating to water supply and sanitation, examined country-level data. Based on a comparison of progress made during the period 1990–2002, each country is assessed in terms of the likelihood of achieving the MDG (though this information is not included in the report). Among the main messages of this report are that:

• of the 1.1 billion people lacking access to an improved source of water in 2002, nearly two-thirds were in Asia;

• lowest drinking water coverage levels were in SSA and Oceania; • people without access to water and sanitation are among ‘the hardest to reach – families living in remote rural areas and urban slums, families displaced by war and famine, and families mired in poverty– disease traps’ (WHO-UNICEF 2004:17).

92 P. B. Anand

Other assessments such as the Global Monitoring Report (World Bank 2005) compare the progress made with regard to each MDG between 1990 and 2005 with the target year of 2015 for each region (such as SSA, South Asia and so on). Jolly (2004:273–91) also examines the progress with regard to water and sanitation targets (mainly at the regional level rather than at country level) and identifies a detailed agenda for reform. Shordt et al. (2004) focus mainly on monitoring aspects and provide a good summary of various approaches in use at the moment, including participatory methods such as quantified participatory monitoring, with examples from Karnataka and Malawi. UNDP (2003) also assesses progress with regard to each of the various MDGs but its analysis precedes the publication of WHO-UNICEF (2004) and hence the data used in it are from earlier assessments. Forecasts about achievement or shortfall are made by region in that report, though the methodology behind the forecasts is not fully explained. While the above assessments provide considerable insight into the progress made and the magnitude of task still ahead, there is limited information in the public domain on cross-country comparisons or patterns. Against this background, this chapter attempts to review and summarize the progress so far and, based on available evidence, to examine: (i) whether there are patterns or regularities in progress with regard to the two targets in relation to per capita income, human development index and aspects of governance; (ii) whether the synergy effect is significant – that is, are countries that have made significant progress with one target more likely to make a significant progress with other related targets; and (iii) whether and to what extent the achievement or lack of progress on these two targets can impinge on performance in relation to other MDGs or targets. The chapter uses country-level data (available for two years, namely 1990 and 2002) and individual country case studies to highlight specific aspects. An alternative approach is to examine individual country-specific policies and assess progress. Owing to limitations of space, that is not attempted here. However, a country-specific analysis of India is reported elsewhere (Anand 2007a, 2007b).

Hypotheses, methodology, data and limitations Why is there a significant variation across countries in the proportion of population having access to water and sanitation? A positive and significant relationship has been observed between GDP per capita and the proportion of population having access to water and sanitation in World

Achieving the Water and Sanitation MDG

93

Bank (1992) and Shafik (1994). This seems to suggest that water and sanitation are normal goods with positive income elasticities of demand. Water and sanitation are not pure public goods. Thus, while there is a significant role for the state, these are also private goods where ‘exit’ is possible. For instance, households can provide themselves with water supply by sinking a well or tube well and pumping groundwater, or draw water from rivers, streams and so on. Similarly, sanitation technologies such as improved pit latrines or septic tanks enable households to provide themselves with safe sanitation when such goods are not provided by the state. However, such self-provision is capital- and land-intensive and usually beyond the reach of the poor. Also, many households may not have such options owing to terrain, climate, incomplete property rights institutions (to land and water) and human settlement patterns. However, while water and sanitation can be private goods at the level of the individual, there are significant externalities in terms of public health impacts. For instance, when a significant proportion of the population does not have access to sanitation and as a result many individuals resort to what is euphemistically called the ‘bush latrine’, the scope for contamination of water sources and the resulting risk of infectious diseases increase significantly. Also, owing to economies of scale, it is socially efficient that water and sanitation should be provided for the whole human settlement rather than leaving it to individuals. For these reasons, as incomes begin to increase, citizens may use their voice to put pressure on the state to act and thus the proportion of the population having access to water and sanitation starts to increase. Let us refer to this as the ‘Kuznets effect’. While this is a plausible explanation, as the critics of the environmental Kuznets curve (EKC) point out there is need for caution (see Goldin and Winters 1995; Munasinghe et al. 2001; Panayotou 2003). One interpretation of such a relationship is that access to water and sanitation will improve as incomes increase and that low-income countries thus can (and ought to) focus on improving incomes. An alternative interpretation is that improving water and sanitation is essential to achieve productivity growth, which is crucial for sustaining progress and economic development. Against this background, the exploration in this chapter attempts to examine the following issues: (i) To what extent does access to water and sanitation continue to be associated with per capita GDP and other indicators such as the human development index? (ii) Is there a synergy between access to water and access to sanitation?

94 P. B. Anand

(iii) Is access to water or sanitation determined by legacy or policy? That is, whether the proportion of people having access to water or sanitation in period t is determined by the proportion of population with access to water or sanitation in period t − 1 (legacy), or whether it is also determined by other policy-relevant variables. (iv) Is there any association between access to water and sanitation and other MDG-relevant indicators? (v) What are the policy implications? Methodology The analysis in this chapter is based on national-level data and crosssection regression analysis to identify associations between variables and use the regression models to forecast progress. I develop a model where access to water or sanitation in a given year (say, 2000) is determined by access status in a previous year (say, 1990) and various explanatory variables such as per capita GDP in recent years, past values of economic and population growth rates, social expenditures in total government spending, aid received and so on. Based on this model, I then forecast the percentage of the population that will have access to water and sanitation for a future year (2015) using the current values of explanatory variables. The projected access figures are compared with the MDG targets to ascertain how far off the country is going to be. Data Much of the analysis here is based on data on access to water and sanitation at national level in terms of percentage of population having access. There is considerable subjectivity in defining access; also the data are furnished by national governments and may be difficult to verify. The most recent set of data are available for two years, 1990 and 2002, from the Millennium Indicators Dataset.3 There is a very important limitation in terms of endogeneity. Many of the variables considered in the analysis are facets of development and change. At the same time, access to water and sanitation can depend significantly on many other variables such as actions of NGOs, communities, local-level leadership, private sector activity and so on. There are limited data at aggregate level on such variables to include in the models. Also, national-level aggregates may not fully capture the considerable variation in a given country. Hence, the regressions attempted here may not capture the causality completely or adequately. Second, reliable data on access to water and sanitation are available for two points in time

Achieving the Water and Sanitation MDG

95

Table 4.1 Mean and standard deviation of percentage of population having access to water and sanitation, 1990 and 2000.

Access

Country category

N

Mean

WAT1990 Access to water: % of population 1990 WAT2000 Access to water: % of population 2000 SAN1990 Access to sanitation: % of population 1990 SAN2000 Access to sanitation: % of population 2000

Developing country 99 72.354 Not a developing 11 99.455 country Developing country 137 76.613 Not a developing 11 99.636 country Developing country 91 53.835 Not a developing 10 98.700 country Developing country 134 59.731 Not a developing 10 98.700 country

Std error Std dev. mean 21.341 1.809

2.145 0.545

19.326 1.206

1.651 0.364

29.794 4.111

3.123 1.300

26.954 4.111

2.328 1.300

Source: Author’s calculations based on WHO–UNICEF (2004) data.

and this may not sufficiently capture the dynamic changes in institutional processes and policies. Third, there is considerable variation in the quality of data. For example, while up to 165 countries are listed in the WHO-UNICEF monitoring, for some countries data are available only for one of the two data years. The number of countries in the sample decreases rapidly when one attempts to increase the number of variables of interest such as per capita GDP or per capita quantity of freshwater available and so on.

Analysis Access to water and sanitation: state of progress From the summary statistics in Tables 4.1 and 4.2 it is clear that there is considerable variation in access to water and sanitation, mainly for lowor middle-income developing countries. As of 2000, there are at least 6 countries4 where such coverage is less than 40 per cent of the population; that is 3 out of every 5 persons do not have access to water and sanitation. There are 5 countries5 which have achieved substantial progress with regard to water (over 80 per cent of the population have access to an improved source) but face serious deficit with regard to sanitation (coverage less than 40 per cent). On the other hand, Kyrgyzstan, Libya and Sri Lanka have improved sanitation to cover over 90 per cent, but

96 P. B. Anand Table 4.2 Distribution of number of countries as per access to water and sanitation, 2000 People having People having access to sanitation in 2000 (%) access to water in 2000 (%) Less than 20 20–40 40–60 60–80 80–90 90–100 Total Less than 20 20–40 40–60 60–80 80–90 90–100 Total

1 3 5

9

2 9 10 4 1 26

1 10 15 6 3 35

5 14 7 26

2 9 11

3 3 30 36

1 6 24 35 27 50 143

Source: Author’s assessment based on WHO–UNICEF (2004) data.

water supply coverage remains less than 80 per cent of the population. These peculiarities are indicators of the variations in progress. Given this considerable variation, it is necessary to examine whether there are any systematic tendencies and whether access to water and sanitation are influenced by various country-specific variables. This is attempted in the rest of this section. Is access to water or sanitation a function of per capita GDP? To examine this, a relationship is estimated using simple linear regression models. In the regressions R1 to R4 presented in Table 4.3, access to water or sanitation in a given year is considered to be a function of GDP per capita. The average of per capita GDP (expressed in constant $US in 2000 prices) for three preceding years was used as the independent variable. The results for data relating to 2000 are shown in Figures 4.1 and 4.2. These results are in line with World Bank (1992) and Shafik (1994) estimates based on 1980s data. Since both the dependent and independent variables in Table 4.3 are in logarithmic form, the parameters can be interpreted as elasticities. Access to water and access to sanitation appear to be normal but inelastic goods (necessities). Why should the elasticities decrease over time? We can conjecture that as the better-off (and hence more organized) sections of the population get access to water (or sanitation), the pressure on governments to provide these services eases and as a result government resources are diverted to providing other normal (and elastic) goods and services.

Achieving the Water and Sanitation MDG Table 4.3

97

Access to water or sanitation: is it a function of per capita GDP?

Dependent variable

Independent variable

R1 log WAT1990 log GDPCAP8789

Parameter of independent Constant variable

2.201*** (11.039) R2 log WAT2000 log GDPCAP9799 2.90*** (18.733) R3 log SAN1990 log GDPCAP8789 −0.486 (1.198) R4 log SAN2000 log GDPCAP9799 0.638** (2.062)

0.251*** (10.345) 0.175*** (9.308) 0.533*** (10.705) 0.412*** (10.866)

R2

F-value

0.543 107.023 0.421

N 92

86.648 121

0.577 114.603 86 (0.000) 0.507 118.063 117 (0.000)

Notes: Independent variable is also log-transformed. WAT2000 is the percentage of people having access to water in 2000. SAN2000 is the percentage of people having access to sanitation in 2000. GDPCAP8789 and GDPCAP9799 respectively are the average of per capita GDP for years 1987, 1988 and 1989 and the average for years 1997, 1998 and 1999, all measured in constant 2000 prices in international dollars (from World Bank 2005). Figures in parentheses are t -statistics. ***, ** and * indicate significance at less than 1 per cent; 5 per cent; and 10 per cent respectively. These notations apply to all subsequent tables.

An alternative interpretation is that in a world where every country has achieved 100 per cent access, the line becomes parallel to the horizontal axis and the slope of the line (and hence the parameter) would be zero. Thus, the parameter values in Table 4.3 suggest that access to water and sanitation has improved between 1990 and 2000. Linear interpolation6 of the changes suggests that if these trends continue, access to water will not approach 100 per cent until 2023 and access to sanitation will not approach 100 per cent until 2034. However, the regressions are only indicative of the structural relationships and it is a lot more difficult to improve access from 80 per cent to 90 per cent of the population than it is from 20 per cent to 30 per cent of population. Hence, the parameters may approach zero asymptotically rather than in linear fashion and it may take far many more years to achieve 100 per cent access than the linear interpolation suggests. An alternative indicator of the level of development is the human development index (HDI) compiled by the UNDP. The index incorporates measures of poverty, inequality and life expectancy. Since the conjecture is that access to water and sanitation can have a bearing on life expectancy, it is of interest to explore the association between HDI

log percentage of population with access to water 2000

98 P. B. Anand

4.50

4.00

3.50 R 2 linear ⫽ 0.421 3.00 6.00

7.00

8.00

9.00

10.00

11.00

log Average GDP per capita 1997–99 ($US in 2000 prices) Figure 4.1 GDP per capita (average for 1997–9 in $US, 2000 prices) and percentage of people having access to water in 2000

and access to water and sanitation. This is done in Table 4.4 with simple linear regressions where access to water or sanitation in a given year is considered to be a function of human development index in that (or a specified) year. When we compare the results in Table 4.4 with those in Table 4.3 we find that in general access to water and sanitation is associated with HDI and that there may also be a lag effect, that is, access in period t may be influenced by human development in period (t − 1). This seems to support the working of a Kuznets effect mentioned in the previous section. Human development in period (t − 1) may trigger people to put pressure on the state; however, as water and sanitation projects take time before the effects are felt, there is a lag effect. The results above also seem to suggest that such a Kuznets effect is stronger with regard to access to sanitation. This is plausible given that ‘exit’ is more difficult and negative externalities are more pronounced with sanitation than with water. Apart from the association between access to water or sanitation and human development at a given point in time, change in human development over a period of time and change in access to water or sanitation are examined in Figures 4.3 and 4.4. These indicate that while improvement

Achieving the Water and Sanitation MDG

99

log percentage of population with access to sanitation 2000

5.00

4.00

3.00

2.00

R 2 linear ⫽ 0.507

6.00

7.00

8.00

9.00

10.00

11.00

log Average GDP per capita 1997–9 ($US in 2000 prices) Figure 4.2 GDP per capita (average for 1997–9 in $US, 2000 prices) and percentage of people having access to sanitation in 2000

in access to water has no relationship with improvement in HDI during 1990–2000, with regard to access to sanitation the picture is quite remarkable. There is a fairly strong positive association between change in HDI between 1990 and 2000 and a change in access to sanitation between 1990 and 2000. Is there synergy in water supply and sanitation? If the proportion of population having access to sanitation is correlated to the proportion of population having access to water, we can conjecture that there is some synergy between these two services. In general, countries that have done well with regard to access to water supply are also highly likely to do well with sanitation. This is evident from Figure 4.5. The slope and intercept seem to suggest that there may be a slight lag between progress with regard to access to water and progress with sanitation. (This is also evident from model R11 in Table 4.5). What does this mean? Of course, access to water is essential for sanitation to be functional. However, why a lag effect? We return to this issue in the next subsection.

100 P. B. Anand Table 4.4 HDI and access to water and sanitation

Dependent variable

Independent variable

R5 log WAT2000 HDI1990 R6 log WAT2000 HDI2000 R7 log SAN2000

HDI1990

R8 log SAN2000

HDI2000

Constant

Parameter of independent variable

3.628*** (46.622) 3.591*** (38.734) 2.429*** (18.768) 2.198*** (13.765)

1.135*** (9.288) 1.119*** (8.079) 2.548*** (12.524) 2.730*** (11.427)

R2

F-value

0.461

86.276 (0.000) 0.423 65.271 (0.000) 0.611 156.856 (0.000) 0.600 130.573 (0.000)

N 103 91 101 88

Change in percentage of population with access to water 1990–2000

40.00

30.00

20.00

10.00

0.00 R 2 linear ⫽ 0.001

⫺10.00 ⫺0.20

⫺0.10

0.00 0.10 HDI2000 minus HDI1990

0.20

Figure 4.3 Change in HDI and change in percentage of population with access to water, 1990–2000

Legacy or policy? Does progress depend on the starting-point? Progress with regard to access to water and access to sanitation seems to depend much on the starting-point (or legacy). Countries that had

Achieving the Water and Sanitation MDG

101

Change in percentage of population with access to sanitation 1990–2000

30.00

20.00

10.00

0.00 R 2 linear ⫽ 0.18

⫺10.00 ⫺0.20

⫺0.10

0.00

0.10

0.20

HDI2000 minus HDI1990 Figure 4.4 Change in HDI and change in percentage of population with access to sanitation, 1990–2000

already made significant progress by 1990 had, in general, progressed in 2000 also. This is evident from Figures 4.6 and 4.7. If the proportion of population having access to water or sanitation in a given year is closely related to the proportion of population having access in previous years, we can conjecture that legacy matters. This is explored in Table 4.6. It appears that the proportion of population with access to water and sanitation in period t (say, 2000) is very significantly influenced by the proportion of people having such access in period t − 1 (1990). This suggests that improving access to water and sanitation and developing institutions to deal with this requires commitment over a considerable period of time rather than being something that can be achieved in a short duration. This means that it may be far harder and more difficult for countries such as the six identified in Table 4.2 to make significant progress. This is an important point to keep in mind while assessing regional disparities in progress. For example, it is possible to argue that many SSA countries, which started with low coverage of water and sanitation

102 P. B. Anand

Percentage of population with access to sanitation 2000

100.00

80.00

60.00

40.00

20.00 R 2 linear ⫽ 0.605 0.00 0.00

20.00

40.00 60.00 80.00 Percentage of population with access to water 2000

100.00

Figure 4.5 Countries as percentages with access to water and sanitation in 2000

Table 4.5 Synergy in providing access to water and access to sanitation

Dependent variable

Parameter of the independent Independent Constant variable variable

−3.058*** (−5.281) R10 log SAN2000 −2.252*** (−3.921) R11 log SAN2000 −1.816*** (−4.213) R9

log SAN1990

1.624*** (11.932) 1.448*** (14.617) 1.374*** (13.608)

R2

F-value

N

log WAT1990 0.602 142.375 95 (0.000) log WAT2000 0.602 213.670 142 (0.000) log WAT1990 0.643 185.170 104 (0.000)

services in 1990, continue to have low coverage in 2000. Results presented in Table 4.7 indicate that other things being the same, SSA countries tend to have made less progress with sanitation coverage than other countries.

Achieving the Water and Sanitation MDG

103

Percentage of population with access to water 2000

100.00

80.00

60.00

40.00 R 2 linear ⫽ 0.866 20.00 20.00

40.00

60.00

80.00

100.00

Percentage of population with access to water 1990 Figure 4.6 Access to water in 1990 and in 2000

However, legacy is not destiny. The above results do not mean that countries with a low level of access are stuck in a low-level equilibrium trap. In percentage terms, the biggest increase in the proportion of population having access to water during 1990–2000 was in the countries listed in Table 4.8. While the figures in Table 4.8 relate to increase in the proportion of population with access, countries achieving the biggest increase in the absolute number of people provided with such access are (in descending order): India, China, Indonesia, Pakistan, Brazil, Nigeria, Bangladesh, Mexico, Myanmar and Turkey. Similarly, the top ten countries in terms of increase in percentage of population having access to sanitation during 1990–2000 are listed in Table 4.9. While 6 out of the top 10 countries achieving improved access to water were SSA countries, 3 out of the top 10 countries achieving significant progress with sanitation are from SSA. What distinguishes the high achievers? This is explored in Table 4.10. Independent sample t-tests (not shown here) suggest that the mean values of the top ten achievers are not significantly different from the

104 P. B. Anand

100.00

Percentage of population with access to sanitation 2000

80.00

60.00

40.00

20.00 R 2 linear ⫽ 0.915 0.00 0.00

20.00

40.00

60.00

80.00

100.00

Percentage of population with access to sanitation 1990 Figure 4.7 Access to sanitation in 1990 and in 2000 Table 4.6 Proportion of population with access to water and sanitation in 2000 is a function of the proportion of population having such access in 1990

Dependent variable R12 log WAT2000

Parameter of the independent Independent Constant variable variable

1.358*** (10.821) R13 log SAN2000 1.056*** (8.665)

0.705*** (24.022) 0.769*** (24.696)

R2

F-value

N

log WAT1990 0.845 577.078 107 (0.000) log SAN1990 0.863 609.899 98 (0.000)

mean values of the rest of the countries (N = 97) or the bottom 10 countries in terms of change in the percentage of population having access to water between 1990 and 2000. If anything, figures in Table 4.10 seem to suggest that the top ten achievers had more hurdles to overcome than the bottom 10 countries (for example, lower level of GDP per capita, lower level of economic growth rate, greater inequality, higher level of child mortality rate and so on).

Achieving the Water and Sanitation MDG Table 4.7

105

Starting point effect and being in sub-Saharan Africa

Dependent variable

log Constant

log WAT1990 SAN1990

SSA dummy

Adj. R2

F-value

R14 log WAT2000

1.280*** 0.722*** – 0.019 0.842 287.100 (7.293) (18.226) (0.631) (0.000) R15 log SAN2000 1.523*** – 0.673*** −0.268*** 0.936 400.272 (10.829) (20.164) (−5.198) (0.000)

Table 4.8 Top ten countries in terms of progress with regard to access to water, 1990–2000 (% of population) People having access Change during to water (%) in: 1990–2000 in people having 1990 2000 access (%) Tanzania Myanmar Central African Republic Malawi Ghana Namibia Paraguay India Haiti Guatemala

38 48 48 41 54 58 62 68 53 77

73 80 75 67 79 80 83 86 71 95

35 32 27 26 25 22 21 18 18 18

Total aid (ODA+ OA) for water and sanitation in 1990–2000 ($m) 344.21 1.07 54.17 78.01 336.56 98.93 17.86 587.55 9.63 121.43

Source: Based on WHO–UNICEF (2004).

Thus, while legacy seems to be important, the examples in Tables 4.8 and 4.9 seem to suggest that countries can and do get out of the low-level equilibrium. We would like to explore whether this is due to policy or to providence. By providence, we refer to variables such as population, inequality and per capita freshwater resources. To examine these issues, multiple regression analysis is used. In Table 4.11 are reported alternative models in terms of combinations of different independent variables. While a number of specifications are possible, only three selected regression models are presented in the table. Model R16 includes many variables of interest including WAT1990 and GDPCAP8789. There is a trade-off between increasing the number of variables and data availability. For this model, the sample size is 37

106 Table 4.9 Top 10 movers with regard to per cent of population having access to sanitation, 1990 to 2000 People having access to water in (%):

Myanmar Cameroon Bangladesh Benin Madagascar Sri Lanka China Paraguay Nicaragua Honduras

1990

2000

Change during 1990–2000 in of people having access (%)

21 21 23 11 12 70 23 58 47 49

73 48 48 32 33 91 44 78 66 68

52 27 25 21 21 21 21 20 19 19

Source: Based on WHO–UNICEF (2004).

Table 4.10 Were high achievers different? Change in access to water between 1990–2000

Variables Public expenditure on health as % of GDP average for 1998–2000 Average GDP per capita, 1987–89 (in US$ 2000 prices) Average GDP per capita, 1997–9 (in US$ 2000 prices) Per cap GDP growth rate, 1990–2000 Gini index HDI 1990 HDI 2000 Annual pop growth rate, 1975–2000 Child mortality rate, 1990 Child mortality rate, 2000 Aid as % of GDP, 1990 Corruption perception index of TI 2004 Water resources cubic metre per capita

Top 10 countries N Mean 10

2.33

Bottom 10 countries N Mean 9

2.35

9

2567.3

6

4038.5

9

2498.1

7

3641.9

10 7 7 6 10 10 10 5 9 10

0.77 50.8 0.49 0.56 2.58 145.4 128.1 4.9 2.5 15765.6

9 6 6 6 9 10 10 8 7 7

1.98 36.9 0.59 0.64 2.47 81.9 70.3 14.9 3.1 2092.1

Table 4.11 Results of multiple regression analysis: dependent variable is access to water (% of population), 2000 Model R16

Constant Access to water as % of population, 1990 Annual population growth rate, 1975–2000 Per capita GDP – average for 1997–1999 (US$ in 2000 prices) Per capita GDP growth rate, 1990–2000 Health expenditure (public sector) as a share of GDP average for 1998–2000 Gini coefficient Aid as % of GDP, 1990 Malnourished people as % of population, 1990 Fresh water resources available cubic metres per capita (log) Adjusted R2 F-value N

WAT1990

Model R17

Model R18

Parameter

Collinearity diagnostic

Parameter

Collinearity diagnostic

12.436 (0.687) 0.579*** (6.341)

0.467

16.146*** (2.645) 0.763*** (16.197)

0.646

Parameter

Collinearity diagnostic

−8.539 (−0.499)

–

POPGROW

1.670 (0.706)

0.443

–

−1.661 (−0.936)

0.669

log GDPCAP

2.131 (1.027)

0.473

–

12.310*** (6.877)

0.530

ECONGROW HEALTH

0.676 (1.000) 1.010 (0.859)

0.579 0.453

– −0.153 (−0.261)

1.365*** (2.567) 0.119 (0.914)

0.957 0.643

−0.047 (−0.283) GINI −0.100 (−0.638) AID2GDP MALNURISH −0.043 (−0.406)

0.374 0.696 0.725

0.123* (1.535) – –

0.817

log WATRES

0.636

0.476 (0.408)

0.886

0.690 (0.783) 0.754 13.237 (0.000) 37

0.845 96.305 71

0.698

– – – −1.126 (−1.338)

0.948

0.551 22.834 90

Notes: Collinearity diagnostic, tolerance is the percentage of the variance in a given predictor that cannot be explained by the other predictors. When the tolerances are close to 0, there is high multicollinearity and the standard error of the regression coefficients will be inflated. Variance inflation factor was also estimated but not shown here. A rule of thumb is when the value of collinearity diagnostic is less than 0.5 (or VIF above 2) some degree of collinearity is present. In the above cases, all variables have VIF less than 2.

108 P. B. Anand

countries. In this model, apart from legacy indicator (WAT1990), three indicators of economy (GDPCAP, ECONGROW, AID2GDP); one indicator of demographic trend (POPGROW); two indicators of inequality (GINI, MALNURISH); one indicator of social sector spending (public sector expenditure on health) and one indicator of providence in terms of freshwater resources per capita are included. While the overall goodness of fit is very high (adjusted R2 = 0.754), the only variable statistically significant at less than the 5 per cent level is WAT1990. Model R17 is an alternative with fewer variables, sample size increased to 71 and adjusted R2 increased to 0.845. Here too, WAT1990 is still the only variable that is highly significant; the constant term is significant and GINI index is slightly significant. Finally, model R18 is a specification where WAT1990 is omitted and instead the average per capita GDP in 1997–9 is used as a predictor of the proportion of population with access to water in 2000. Though adjusted R2 decreases to 0.551, the number of countries in the sample increased to 90. The model seems to suggest that if the legacy effect is excluded, then per capita GDP and also economic growth rate are significant and this seems to be in line with the Kuznets effect discussed earlier. In Table 4.12 are presented similar results with regard to access to sanitation in 2000 (SAN2000). In this regression model R18, apart from SSA and governance quality, population size also seems to be important. This seems to be indicating the fact that apart from percentage of population, the magnitude of task in terms of the absolute number of people to be provided with services in populous countries such as India, China, Ethiopia and Nigeria can be a significant challenge. Forecasting access to water and sanitation in 2015 based on progress so far Regression equations in Tables 4.11 and 4.12 were of the form of equation 4.A1 (in Appendix) whereby the proportion of population having access in period t is estimated based on independent variables of period t − 1. Now, based on availability of data for period t (year 2000) using the regression equations, forecasts can be made for period t + 1 (year 2015). Two forecasts for access to water in 2015 based on equations R17 and R18 were made. The forecast is then compared with the MDG target. These are reported in the Appendix. Similarly, for access to sanitation, forecasts were made using equations R19 and R20. The forecasts and comparison with MDG targets for sanitation for various countries are reported in Appendix Table 4.A2.

Table 4.12 Results of multiple regression analysis: dependent variable is access to sanitation (% of population), 2000 Model R16

Constant Access to sanitation as % of population, 1990 Population in 1990 in millions Annual population growth rate, 1975–2000 Per capita GDP – average for 1987–1989 (US$ in 2000 prices) Per capita GDP growth rate, 1990–2000 Health expenditure (public sector) as a share of GDP average for 1998–2000 Gini coefficient Aid as % of GDP, 1990 Child mortality rate, 1990 Adjusted R2 F-value N

Model R17

Model R18

Parameter

Collinearity diagnostic

Parameter

Collinearity diagnostic

Parameter

SAN1990

38.573*** (4.180) 0.754*** (10.325)

0.282

35.392*** (8.972) 0.731*** (16.373)

0.282

−97.146*** (−3.819) –

POP1990 POPGROW

−0.00006 (−0.010) 1.259 (0.495)

0.615 0.534

log GDPCAP

–

– ECONGROW −1.974* (−1.947) HEALTH GINI AID2GDP CMR

−0.103 (−0.792) 0.091 (0.510) −0.103*** (−3.367) 0.913 67.324 (0.000) 45

0.528 0.678 0.651 0.361

Collinearity diagnostic

– –

−0.016 (−1.497) −0.862 (−0.752)

0.648 0.598

–

22.817*** (8.536)

0.648

– −1.543*** (2.645) – – −0.098*** (−5.698) 0.936 466.376 (0.000) 97

0.286 (0.801)

0.564

−0.409 (−2.206) −0.098 (−0.493)

0.865 0.799

0.618

0.329 0.712 22.020 (0.000) 51

110 P. B. Anand Table 4.13 Whether child mortality rate (2000) is affected by access to water and sanitation

(Constant) Public expenditure on health as % of GDP average for 1998–2000 Access to water % of population, 2000 Access to sanitation % of population, 2000 Urban pop as % of 1975 population Sub-Saharan Africa dummy Aid as % of GDP 2002 Gini index Adjusted R2 N

Model R22 parameter

Model R23 parameter

137.037*** (7.630) −4.272* (−1.806)

172.477*** (7.319) −5.76*** (−2.444)

−0.932*** (−5.045) −0.075 (−0.372) 55.53*** (6.891) 0.469 (0.994) −0.251 (−0.763) 0.758 71

−1.068*** (−4.614) −0.284 (−1.481) 56.672*** (6.709) 0.031 (0.059) −0.191 (−0.592) 0.728 73

These forecasts suggest that the MDG target with respect to water is highly likely to be met in most of the countries. The gap between MDG targets and projected figures of access is greater than 10 per cent of the population for 12 countries. The scenario is much bleaker with regard to access to sanitation. 72 countries are expected to miss the MDG target by 10 per cent or more of population without access to sanitation. Of these, the gap between the MDG target and projected access by 2015 is greater than 20 per cent for 40 countries. A regression equation with the gap between MDG and projected access to water (MDGGAP) shown in the last column of Appendix Table 4.A1 as the dependent variable suggests that such a gap is positively associated with GDP per capita (2000) and quantity of freshwater resources available per capita. This seems to suggest that complacency may be affecting progress, especially when a country is endowed with freshwater resources. Does it matter? The inclusion of access to water and sanitation as targets in MDGs signals the intrinsic importance attached by the global community to achieving them. There are also possible instrumental reasons for improving access to water and sanitation. Access to water and sanitation seems to have a very significant impact on reducing the child mortality rate. This is seen in the results7 reported in Table 4.13. Since access to water and access to sanitation are strongly correlated, both of them have not been

Achieving the Water and Sanitation MDG

111

included in the same equation. The results do indicate that each of them is highly significant when included with other policy-relevant variables. With regard to maternal mortality, similar regression equations indicated that access to water was not significant; however, access to sanitation was highly significant at the less-than-1-per-cent level. As can be expected, there is no positive relationship between access to water or sanitation and malaria incidence. However, with either of these variables when included, the improvement in adjusted R2 suggests that the models are better specified when these are included along with other variables such as the share of (public sector) health expenditure in GDP, freshwater resources per capita, the urban population’s share in the total population and the SSA dummy. The last three variables were highly significant. Thus, while access to water and sanitation does not have any relationship per se with malaria, the regression seems to suggest that their availability may be important inasmuch as they are crucial in patient’s recovery and whether recurrence of malaria is more likely if a patient is weak owing to, say, diarrhoea. These three explorations are indicative of the likely negative impacts on other MDG targets of not achieving the MDG targets related to water and sanitation.

Conclusions and further issues While an assessment of progress with water and sanitation goals has been included in the Millennium Task Force reports and the WHO–UNICEF assessment, the strength of the present chapter is that here regression models are used for forecasting the progress rather than merely comparing the increase in the percentage of population having access between 1990 and 2000 and assuming that the same level of increase will continue. Therefore, the projections made here are methodologically more robust. Much of the analysis here was limited to country-level average figures which do not capture considerable intra-country variations. Also, data are very limited in terms of observations for just two periods and these may not adequately capture the underlying institutional changes taking place over a period of time. As much of the variation is country-specific, it is plausible that the results may be slightly biased. In spite of these and other limitations, the following conclusions can be drawn from the various regression models presented here: (i) Access to water and sanitation is closely and significantly related to economic and human development.

112 P. B. Anand

(ii) There is strong evidence to suggest that legacy in terms of the starting-point matters and as such there is a bigger mountain to climb for those countries which are starting with a lower base. However, the lists of top-ten countries presented here with respect to each target seem to suggest that it is possible to make significant progress. (iii) There is some evidence also to suggest that apart from legacy, some policy variables matter. The most important ones seem to be per capita GDP, economic growth rate and social sector spending, represented in the analysis here by public sector health spending as a proportion of GDP. (iv) The forecasts indicate that while the target related to water is likely to be achieved, or missed only slightly, in a majority of countries, the target related to sanitation is going to be missed in the great majority. (v) There is some degree of synergy between access to water and access to sanitation. More importantly, access to water and sanitation has a highly significant impact on child mortality rate; access to sanitation seems to have a highly significant impact on maternal mortality rate; and there is some slight evidence that access to water and sanitation may also be negatively associated with malaria incidence (though the connections are not direct). These seem to highlight the instrumental role of access to water and sanitation in promoting health and wellbeing and other MDGs. While the analysis in this chapter has focused on national-level aggregates, it is important to supplement such analysis with microlevel analysis based on case studies and other methods. First, assessing progress through indicators may seem like a technocratic exercise (Harcourt 2005). In detailed case studies on India (Anand 2001, 2007a, 2007b) it is evident that citizens do not passively accept access to water and sanitation but engage actively in improving access through both ‘exit’ and ‘voice’ options. It is therefore important to explore and understand the role of national and local governments, private sectors, NGOs and community groups in significantly increasing access to water and sanitation in countries such as the top ten identified earlier. Second, given that water is a highly contested resource, the institutional space is not without conflicts. As most of the freshwater resources are already committed in many countries, increasing access to water not merely is a matter of money and technology but also involves conflicts between different users and legal and institutional mechanisms. Analysis presented

Achieving the Water and Sanitation MDG

113

elsewhere (Anand 2004) indicates that there may exist political incentives to increase water disputes or keep them unresolved. It is, therefore, important to identify conflict-preventing and cooperation-promoting mechanisms and good practices. Third, it is important to examine what role the so-called new public management approaches in the late 1990s have played in changing management practices in water utilities and contributed to increased access. The microlevel evidence seems to suggest that where water utilities have become more customer-focused their performance improves in terms of increasing access and also in delivering services efficiently. This needs further examination. Fourth, it is important to find microlevel and longitudinal evidence for the Kuznets effect in terms of the relationship between economic growth, development and access to water and sanitation as well as other health and quality-of-life indicators.

Appendix Methodology It is hypothesized that: ACCESSt, i = α + β1 X1t−1, i + β2 X2t−1, i + · · · + βn Xnt−1, i where the dependent variable is ACCESS (to water or sanitation) in country i in period t and the right-hand side is an appropriate specification, with X1, X2 and so on up to Xn being the various country-specific independent variables. Based on observed relationships between access to water or sanitation in period t and other relevant independent variables in period t − 1, an attempt will be made here to forecast access figures for t + 1 (year 2015): PROJACCESSt+1, i = αi + β1 X1t, i + β2 X2t, i + · · · + βn Xnt, i

(4.A1)

The projected figures of access will then be compared with the MDG. The target is to halve the proportion of population without access to water and sanitation. Suppose that the proportion of population having access to water in country i in 2000 is estimated to be pi . TARGETi = 0.5 ∗ (100 − pi )

(4.A2)

MDGi = ACCESS2000i + TARGETi

(4.A3)

Therefore,

114

Access to water (% of pop.) 2000

Additional % of pop. to be provided water to meet MDG (half of 100 minus WAT2000)

MDG % of population with access to water in 2015 = Access in 2000 + IDTarget

Projection 1 for 2015 based on model R17

Projection 2 for 2015 based on model R18

Gap (Projection 1 minus MDG)

Oman Argentina Saudi Arabia Lebanon Belize Gabon Syrian Arab Rep. Sudan Togo Haiti Angola Benin Eritrea Chad Congo Canada Chile Botswana Uruguay Costa Rica Switzerland Netherlands Finland Austria Bulgaria Russian Fed. Malaysia Sweden Guatemala Israel Japan Hungary Egypt Bosnia & Herzegovina Albania

Access to water (% of pop.) 1990

Table 4.A1 Access to water: forecasts for year 2015 by country based on regression models

77 94 90 100 – – 79

79 – – 100 91 87 79

10.5 – – 0 4.5 6.5 10.5

89.5 – – 100 95.5 93.5 89.5

– – – – – – –

97.83 97.79 97.62 89.18 83.31 83.08 82.29

– – – – – – –

64 49 53 32 60 40 20 – 100 90 93 – – 100 100 100 100 100 94 – 100 77 100 100 99 94 98

69 51 71 50 68 57 34 46 100 95 95 98 97 100 100 100 100 100 96 95 100 95 100 100 99 98 98

15.5 24.5 14.5 25 16 21.5 33 27 0 2.5 2.5 1 1.5 0 0 0 0 0 2 2.5 0 2.5 0 0 0.5 1 1

84.5 75.5 85.5 75 84 78.5 67 73 100 97.5 97.5 99 98.5 100 100 100 100 100 98 97.5 100 97.5 100 100 99.5 99 99

– – – – – – – – 100.00 100.00 100.00 100.00 99.89 99.88 99.76 99.71 99.68 99.55 99.44 99.25 99.13 98.58 98.57 98.46 98.29 98.08 97.82

74.53 70.06 69.14 69.11 65.25 64.50 60.16 54.98 100.00 95.53 93.59 93.08 92.69 100.00 – 100.00 100.00 – 85.86 93.57 100.00 81.06 100.00 – 100.00 84.05 –

– – – – – – – – 0.00 2.50 2.50 1.00 1.39 −0.12 −0.24 −0.29 −0.32 −0.45 1.44 1.75 −0.87 1.08 −1.43 −1.54 −1.21 −0.92 −1.18

97

97

1.5

98.5

97.78

87.17 −0.72 (Continued )

115

MDG % of population with access to water in 2015 = Access in 2000 + IDTarget

Projection 1 for 2015 based on model R17

92 91 93

4 4.5 3.5

96 95.5 96.5

97.55 96.73 96.35

85.95 1.55 87.79 1.23 92.72 −0.15

83 80 83 91

89 91 90 93

5.5 4.5 5 3.5

94.5 95.5 95 96.5

95.95 95.90 95.68 95.60

89.06 1.45 93.32 0.40 74.69 0.68 90.25 −0.90

81 92 – 92

93 93 92 91

3.5 3.5 4 4.5

96.5 96.5 96 95.5

95.30 95.17 94.51 94.08

89.86 81.73 – 96.65

– – 83 83 69 98 72 62 87 89 – – 81 69 58 86 95 77 69 68 74 69 67 –

92 92 87 90 86 91 85 83 85 89 83 83 85 84 80 86 87 83 81 86 81 84 82 82

4 4 6.5 5 7 4.5 7.5 8.5 7.5 5.5 8.5 8.5 7.5 8 10 7 6.5 8.5 9.5 7 9.5 8 9 9

96 96 93.5 95 93 95.5 92.5 91.5 92.5 94.5 91.5 91.5 92.5 92 90 93 93.5 91.5 90.5 93 90.5 92 91 91

94.01 – −1.99 93.38 100.00 −2.62 92.61 – −0.89 92.48 72.71 −2.52 91.84 – −1.16 91.83 85.86 −3.67 91.29 73.67 −1.21 91.20 79.55 −0.30 90.58 82.25 −1.92 90.52 69.15 −3.98 90.19 82.73 −1.31 90.18 – −1.32 90.16 92.01 −2.34 89.62 70.65 −2.38 89.50 88.58 −0.50 89.49 – −3.51 89.41 88.15 −4.09 89.21 80.58 −2.29 89.19 77.56 −1.31 89.13 81.49 −3.87 89.02 83.98 −1.48 88.90 69.08 −3.10 88.65 87.32 −2.35 87.15 – −3.85

Gap (Projection 1 minus MDG)

Additional % of pop. to be provided water to meet MDG (half of 100 minus WAT2000)

92 – 86

Projection 2 for 2015 based on model R18

Access to water (% of pop.) 2000

Colombia Panama Dominican Rep. Brazil Mexico Honduras Iran (Islamic Rep.) Turkey Jamaica Armenia Trinidad & Tobago Moldova, Rep. Korea Rep. South Africa Pakistan Ecuador Jordan Bolivia Paraguay Philippines Uzbekistan Venezuela Guyana Thailand Côte d’Ivoire Namibia Kazakhstan Algeria Zimbabwe Nicaragua India Peru Nepal El Salvador Gambia

Access to water (% of pop.) 1990

Table 4.A1 (Continued)

−1.20 −1.33 −1.49 −1.42

(Continued )

116

75

12.5

87.5

85.72

– 71 70 54 69 68 66 – – 71 72 38 66 – 58 41 69 50 62 – 45 49 – 50 – 41 – 42 39 34 40 40 – – – 25

76 78 77 79 79 78 77 76 77 75 73 73 72 71 73 67 69 63 62 57 62 60 59 55 58 56 57 51 51 48 46 45 42 37 34 22

12 11 11.5 10.5 10.5 11 11.5 12 11.5 12.5 13.5 13.5 14 14.5 13.5 16.5 15.5 18.5 19 21.5 19 20 20.5 22.5 21 22 21.5 24.5 24.5 26 27 27.5 29 31.5 33 39

88 89 88.5 89.5 89.5 89 88.5 88 88.5 87.5 86.5 86.5 86 85.5 86.5 83.5 84.5 81.5 81 78.5 81 80 79.5 77.5 79 78 78.5 75.5 75.5 74 73 72.5 71 68.5 67 61

84.30 82.21 −3.70 84.28 80.72 −4.72 83.76 93.73 −4.74 83.51 74.84 −5.99 83.42 53.09 −6.08 83.37 100.00 −5.63 83.16 – −5.34 82.99 68.83 −5.01 82.03 65.65 −6.47 81.45 72.24 −6.05 80.48 79.66 −6.02 79.91 57.36 −6.59 79.84 69.50 −6.16 78.96 72.28 −6.54 78.04 67.06 −8.46 76.48 60.59 −7.02 75.20 64.90 −9.30 74.21 70.20 −7.29 72.75 69.79 −8.25 72.03 44.38 −6.47 71.85 66.71 −9.15 71.62 61.66 −8.38 71.42 55.43 −8.08 68.48 59.70 −9.02 68.25 78.90 −10.75 67.37 71.51 −10.63 67.17 – −11.33 64.74 71.77 −10.76 64.01 66.09 −11.49 62.97 60.38 −11.03 60.94 57.78 −12.06 60.86 57.03 −11.64 57.07 67.47 −13.93 53.95 70.37 −14.55 51.84 – −15.16 39.79 62.15 −21.21

Gap (Projection 1 minus MDG)

MDG % of population with access to water in 2015 = Access in 2000 + IDTarget

48

Projection 1 for 2015 based on model R17 Projection 2 for 2015 based on model R18

Additional % of pop. to be provided water to meet MDG (half of 100 minus WAT2000)

Central African Rep. Lesotho Indonesia China Ghana Burundi Sri Lanka Azerbaijan Georgia Kyrgyzstan Bangladesh Viet Nam Tanzania-UR Senegal Turkmenistan Rwanda Malawi Yemen Cameroon Mongolia Sierra Leone Kenya Nigeria Guinea-Bissau Zambia Tajikistan Mauritania Romania Guinea Burkina Faso Mali Niger Madagascar Mozambique Lao PDR Cambodia Ethiopia

Access to water (% of pop.) 2000

(Continued)

Access to water (% of pop.) 1990

Table 4.A1

63.70 −1.78

Sources: Figures for 1990 and 2000 based on WHO–UNICEF (2004). Projections based on models discussed in the text.

117

Access to sanitation (% of pop.) 2000

Additional % of pop. to be provided sanitation to meet MDG (half of 100 minus SAN2000)

MDG target access by 2015

Projection 1 for 2015 based on model R19

Projection 2 for 2015 based on model R20

Gap (Projection 2 minus MDG)

Cyprus Mauritius Libyan Arab Jamahiriya Ukraine Barbados Finland Trinidad & Tobago Thailand Bulgaria Netherlands Austria Slovakia Switzerland Israel Japan Grenada Lebanon St Kitts & Nevis Chile Kyrgyzstan Sri Lanka Tonga Cuba Uruguay Hungary Bosnia & Herzegovina Oman Algeria Albania Jordan Saint Lucia Costa Rica Suriname Russian Fed. Armenia Georgia Iran (Islamic Rep.)

Access to sanitation (% of pop.) 1990

Table 4.A2 Access to sanitation: forecasts by country based on regression models

100 99 97 99 100 100 100 80 100 100 100 100 100 100 100 97 – 96 85 – 70 97 98 – – – 83 88 – – – – – 87 – – 83

100 99 97 99 99 100 100 99 100 100 100 100 100 100 100 97 98 96 92 100 91 97 98 94 95 93 89 92 89 93 89 92 93 87 84 83 84

0 0.5 1.5 0.5 0.5 0 0 0.5 0 0 0 0 0 0 0 1.5 1 2 4 0 4.5 1.5 1 3 2.5 3.5 5.5 4 5.5 3.5 5.5 4 3.5 6.5 8 8.5 8

100 99.5 98.5 99.5 99.5 100 100 99.5 100 100 100 100 100 100 100 98.5 99 98 96 100 95.5 98.5 99 97 97.5 96.5 94.5 96 94.5 96.5 94.5 96 96.5 93.5 92 91.5 92

– – – – – – 100.00 100.00 100.00 – – 100.00 – 100.00 – – – – – 100.00 99.12 – – 96.42 95.71 96.90 – 96.46 96.85 96.27 94.30 – – 90.20 91.37 90.38 89.51

100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 99.74 99.60 99.57 99.05 98.82 98.79 98.67 98.66 97.92 97.90 97.70 97.27 97.16 97.00 96.68 95.79 94.98 94.68 94.39 94.33 93.72 93.68 92.97 92.61 92.02 90.52 90.03 88.60

0.00 0.50 1.50 0.50 0.50 0.00 0.00 0.50 0.00 −0.26 −0.40 −0.43 −0.95 −1.18 −1.21 0.17 −0.34 −0.08 1.90 −2.30 1.77 −1.34 −2.00 −0.32 −1.71 −1.52 0.18 −1.61 −0.17 −2.78 −0.82 −3.03 −3.89 −1.48 −1.48 −1.47 −3.40

(Continued )

118

Additional % of pop. to be provided sanitation to meet MDG (half of 100 minus SAN2000)

MDG target access by 2015

Projection 1 for 2015 based on model R19

Projection 2 for 2015 based on model R20

Gap (Projection 2 minus MDG)

Serbia & Montenegro Dominica Jamaica Tunisia Syrian Arab Rep. Turkey Colombia Mexico Paraguay Ecuador Philippines Brazil Iraq Venezuela Korea Rep. Myanmar Egypt Kazakhstan Moldova, Rep. Honduras Panama Nicaragua Morocco Guyana Peru South Africa El Salvador Bhutan Guatemala Dominican Rep. Indonesia Turkmenistan Uzbekistan Azerbaijan Romania Mongolia Ghana Maldives

Access to sanitation (% of pop.) 2000

(Continued) Access to sanitation (% of pop.) 1990

Table 4.A2

87 – 75 75 76 84 82 66 58 56 54 70 81 – – 21 54 72 – 49 – 47 57 – 52 63 51 – 50 48 46 – 58 – – – 43 –

87 83 80 80 77 83 86 77 78 72 73 75 80 68 63 73 68 72 68 68 72 66 61 70 62 67 63 70 61 57 52 62 57 55 51 59 58 58

6.5 8.5 10 10 11.5 8.5 7 11.5 11 14 13.5 12.5 10 16 18.5 13.5 16 14 16 16 14 17 19.5 15 19 16.5 18.5 15 19.5 21.5 24 19 21.5 22.5 24.5 20.5 21 21

93.5 91.5 90 90 88.5 91.5 93 88.5 89 86 86.5 87.5 90 84 81.5 86.5 84 86 84 84 86 83 80.5 85 81 83.5 81.5 85 80.5 78.5 76 81 78.5 77.5 75.5 79.5 79 79

– – 87.79 87.57 – 86.12 83.52 – 84.74 83.44 83.49 79.68 – 78.26 – – 80.06 77.48 76.11 76.78 74.16 74.92 74.87 – 72.58 68.90 71.10 – 72.45 70.09 69.07 66.75 67.37 66.20 64.18 65.62 67.20 –

88.53 87.80 86.97 86.70 86.13 85.73 85.61 84.74 84.45 82.40 82.40 81.40 80.86 78.11 77.70 77.62 77.60 77.57 76.31 76.30 76.00 74.02 73.36 72.92 72.76 72.56 72.09 71.56 71.35 70.15 67.58 66.63 65.63 65.06 64.96 64.72 64.43 63.93

−4.97 −3.70 −3.03 −3.30 −2.37 −5.77 −7.39 −3.76 −4.55 −3.60 −4.10 −6.10 −9.14 −5.89 −3.80 −8.88 −6.40 −8.43 −7.69 −7.70 −10.00 −8.98 −7.14 −12.08 −8.24 −10.94 −9.41 −13.44 −9.15 −8.35 −8.42 −14.37 −12.87 −12.44 −10.54 −14.78 −14.57 −15.07 (Continued)

119

Access to sanitation (% of pop.) 2000

Additional % of pop. to be provided sanitation to meet MDG (half of 100 minus SAN2000)

MDG target access by 2015

Projection 1 for 2015 based on model R19

Projection 2 for 2015 based on model R20

Gap (Projection 2 minus MDG)

Vanuatu Tajikistan Pakistan Belize Bangladesh Fiji China Viet Nam Zimbabwe Gambia Senegal Cape Verde Equatorial Guinea Bolivia Kenya Swaziland Papua New Guinea Cameroon Djibouti Mauritania Botswana Gabon Sudan Tanzania-UR Solomon Islands Uganda Malawi India Zambia Lesotho Lao PDR Madagascar Haiti Nepal Côte d’Ivoire Yemen

Access to sanitation (% of pop.) 1990

Table 4.A2 (Continued)

– – 38 – 23 – 23 22 49 – 35 – – 33 42 – 45 21 48 28 38 – 33 47 – 43 36 12 41 37 – 12 15 12 31 21

50 53 54 47 48 43 44 41 57 53 52 42 53 45 48 52 45 48 50 42 41 36 34 46 31 41 46 30 45 37 30 33 34 27 40 30

25 23.5 23 26.5 26 28.5 28 29.5 21.5 23.5 24 29 23.5 27.5 26 24 27.5 26 25 29 29.5 32 33 27 34.5 29.5 27 35 27.5 31.5 35 33.5 33 36.5 30 35

75 76.5 77 73.5 74 71.5 72 70.5 78.5 76.5 76 71 76.5 72.5 74 76 72.5 74 75 71 70.5 68 67 73 65.5 70.5 73 65 72.5 68.5 65 66.5 67 63.5 70 65

– 64.77 65.30 – 63.32 – 59.72 61.40 53.89 60.42 58.65 – – 55.87 54.69 – 52.73 52.50 – 56.78 46.45 – – – – 51.37 – 46.63 – 40.30 47.40 45.60 – 45.81 – 47.08

63.56 62.63 62.52 61.92 61.09 60.73 60.67 60.02 58.16 57.93 57.10 56.74 55.99 55.58 55.34 53.64 53.27 52.55 52.54 51.84 51.30 49.86 49.43 49.28 48.74 48.48 46.12 46.03 45.73 45.04 44.86 44.27 44.15 44.11 43.99 43.45

−11.44 −13.87 −14.48 −11.58 −12.91 −10.77 −11.33 −10.48 −20.34 −18.57 −18.90 −14.26 −20.51 −16.92 −18.66 −22.36 −19.23 −21.45 −22.46 −19.16 −19.20 −18.14 −17.57 −23.72 −16.76 −22.02 −26.88 −18.97 −26.77 −23.46 −20.14 −22.23 −22.85 −19.39 −26.01 −21.55 (Continued)

120

Gap (Projection 2 minus MDG)

27.5 32 38.5 29.5 31 34 33 33.5 35.5 33 36.5 36.5 30.5 42 37.5 35 38 45.5 45.5 43.5 37 44 47 46 44

Projection 2 for 2015 based on model R20

45 36 23 41 38 32 34 33 29 34 27 27 39 16 25 30 24 9 9 13 26 12 6 8 12

Projection 1 for 2015 based on model R19

36 44 23 37 39 11 37 – 18 – 23 – – – – 30 – – 8 17 38 13 4 6 7

MDG target access by 2015

Additional % of pop. to be provided sanitation to meet MDG (half of 100 minus SAN2000)

Mali Burundi Comoros Rwanda Nigeria Benin Togo Timor-Leste Congo, DR of the Guinea-Bissau Central African Rep. Mozambique Sierra Leone Cambodia Somalia Angola São Tomé & Príncipe Congo Eritrea Guinea Liberia Burkina Faso Ethiopia Chad Niger

Access to sanitation (% of pop.) 2000

(Continued) Access to sanitation (% of pop.) 1990

Table 4.A2

72.5 – 43.25 −29.25 68 – 42.17 −25.83 61.5 – 41.30 −20.20 70.5 – 41.10 −29.40 69 39.75 40.78 −28.22 66 – 40.31 −25.69 67 – 39.25 −27.75 66.5 – 38.66 −27.84 64.5 – 36.19 −28.31 67 – 35.81 −31.19 63.5 – 35.41 −28.09 63.5 – 34.37 −29.13 69.5 – 32.83 −36.67 58 33.24 31.88 −26.12 62.5 – 29.75 −32.75 65 – 29.74 −35.26 62 – 29.00 −33.00 54.5 – 28.46 −26.04 54.5 – 28.04 −26.46 56.5 – 26.42 −30.08 63 – 25.58 −37.42 56 – 21.10 −34.90 53 – 18.83 −34.17 54 – 18.25 −35.75 56 – 14.56 −41.44

Sources: Figures for 1990 and 2000 based on WHO–UNICEF (2004). Projections based on models discussed in the text.

Acknowledgements The author is grateful to UNU-WIDER for supporting this research and to Tony Addison for introducing me to this project, to Mark McGillivray and all the participants at the initial project meeting, and to the anonymous reviewer for various comments on the previous draft. The usual disclaimers apply.

Achieving the Water and Sanitation MDG

121

Notes 1. In the Millennium Declaration adapted by the UN General Assembly in 2000, only the target of halving the proportion of people without sustainable access to safe drinking water was included. Based on the recommendation of the Bonn Conference in 2002, a corresponding target to halve the proportion of people without access to improved sanitation was included and adopted at the World Summit for Sustainable Development in Johannesburg, 2002. 2. The WHO–UNICEF joint monitoring programme adopted a definition of ‘improved’ source to mean water supply from household connection, public standpipe, borehole, protected well, protected spring, and rainwater collection. Sources of water such as unprotected wells, unprotected springs, vendor-provided water, bottled water, and water provided from tanker trucks are considered as ‘not improved’ sources. Access is defined as having the availability of water from improved sources to the extent of 20 litres per capita per day within a distance of 1000 m from the user’s dwelling. 3. However, there is some slight confusion. On closer inspection it is observed that the 2002 data come from the WHO–UNICEF interim assessment and are precisely the same as the data reported for year 2000. Therefore, in this analysis, I have made a presumption that the two data points are years 1990 and 2000. 4. These are: Afghanistan, Cambodia, Chad, Ethiopia, Lao PDR and Somalia. 5. Comoros, Côte d’Ivoire, Gabon, India, Nepal. 6. For example, the parameter of access to water decreased in 10 years from 0.251 (in 1990) to 0.175 (in 2000). If x is the number of years (from 1990) it takes for the parameter to become zero, from similar triangles, we have (0.251/x) = (0.175/(x − 10)). Solving for x gives us x = 33 years. 7. See Anand (2006) for further details on the results discussed in this section.

References Anand, P. B. (2001) ‘Water “Scarcity” in Chennai, India: Institutions, Entitlements and Aspects of Inequality’. WIDER Discussion Paper 2001/140. Helsinki: UNUWIDER. Anand, P. B. (2004) ‘Water and Identity: An Analysis of the Cauvery River Dispute’. Bradford Centre for International Development, University of Bradford Research Paper 3. Bradford: Bradford Centre for International Development, University of Bradford. Anand P. B. (2006) ‘The Millennium Development Goal 7: An Assessment of Progress with Respect to Water and Sanitation: Legacy, Synergy, Complacency or Policy?’, WIDER Research Paper RP1/2006, Helsinki: UNU-WIDER. Anand P. B. (2007a) ‘Semantics of Success and Pragmatics of Progress: An Assessment of India’s Progress with Drinking Water Supply’. Journal of Environment and Development, 16(1):32–57. Anand P. B. (2007b) Scarcity, Entitlements and the Economics of Water in Developing Countries. Cheltenham: Elgar. Goldin, I. and L. Winters (eds) (1995) The Economics of Sustainable Development. Cambridge University Press. Harcourt, W. (2005) ‘The Millennium Development Goals: A Missed Opportunity?’ Development, 48(1):1–4.

122 P. B. Anand Jolly, R. (2004) ‘Clean Water for All’. In R. Black and H. White (eds), Targeting Development: Critical Perspectives on the Millennium Development Goals. London: Routledge. Munasinghe, M., O. Sunkel and C. Miguel (eds) (2001) The Sustainability of Long Term Growth: Socioeconomic and Ecological Perspectives. Cheltenham: Elgar. Panayotou, T. (2003) ‘Economic Growth and the Environment’. Paper presented at Spring Seminar, United Nations Economic Commission for Europe. Geneva. Shafik, N. (1994) ‘Economic Development and Environmental Quality: An Econometric Analysis’. Oxford Economic Papers, 46:757–73. Shordt K., C. van Wijk, F. Brikke and S. Hesselbarth (2004) Monitoring Millennium Development Goals for Water and Sanitation: A Review of Experiences and Challenges. Delft: IRC International Water and Sanitation Centre. UNDP (United Nations Development Programme) (2003) Human Development Report 2003. New York: Oxford University Press for UNDP. UNDP (United Nations Development Programme) (2004) Human Development Report 2004. New York: Oxford University Press for UNDP. UNDP (United Nations Development Programme) (2005) Health, Dignity, and Development: What Will It Take? – UN Millennium Project Task Force on Water and Sanitation. London: Earthscan for UNDP. United Nations (2002) Millennium Development Goals: Data and Trends: Report of the Inter-Agency and Expert Group on MDGs (ST/ESA/STAT/120). New York. WHO–UNICEF (2004) Meeting the MDG Drinking Water and Sanitation Target: A Mid-Term Assessment. Geneva: WHO–UNICEF JMP. Available at: www.unicef.org/publications/index_23559.html World Bank (1992) World Development Report. New York: Oxford University Press. World Bank (2005) Global Monitoring Report: Millennium Development Goals: From Consensus to Momentum. Washington, DC.

5 Measuring Pro-Poor Progress Towards the Non-Income Millennium Development Goals Melanie Grosse, Kenneth Harttgen and Stephan Klasen

Introduction More than half of the time period to reach the Millennium Development Goals (MDGs) has passed and much effort has been undertaken to achieve the goals by 2015. However, the latest progress report towards the MDGs shows that progress is often very slow and, more worrisome, in some countries there have been reverses, particularly among the poor (UN 2005). This leads to a crucial question concerning the progress made so far: how is the progress towards the MDGs distributed within a country? Have the very poor benefited disproportionately compared with the less-poor or even the non-poor from improvements in reducing both the income and non-income dimension of poverty? To reach the MDGs, it will be critical that both the development path and the policies to accomplish the goals are pro-poor (UN 2003). Thus these questions have to be considered when monitoring progress towards the MDGs. In order to track progress in MDG1 and explicitly link growth, inequality, and poverty reduction, several measures of ‘pro-poor growth’ have been proposed in the literature and used in applied academic and policy work. These measures, particularly the ones derived from the growth incidence curve (GIC) by Ravallion and Chen (2003), allow a much more detailed assessment of the distributional impact of growth and its link to poverty reduction. At the same time, this toolbox has been developed and to date applied only to tracking progress in reducing the income dimension of poverty along the entire income distribution. But what about improvements in non-income indicators? There are no corresponding measures for tracking the distribution of progress in nonincome dimensions of poverty, and thus the distribution of progress in MDGs 2–7 (see Table 1.1). But not only should poverty reduction and, thus, progress towards the non-income MDGs be seen as a by-product 123

124 Melanie Grosse, Kenneth Harttgen and Stephan Klasen

of economic growth, even if it is pro-poor; pro-poor policies should also focus on the non-income dimension of poverty and ensure pro-poor progress in non-income dimensions of wellbeing. The aim of this chapter is to extend pro-poor growth measurement to assess the distribution of progress towards the non-income MDGs. As discussed in Grosse et al. (2008), this could be done by assessing the distribution of progress along the distribution of the non-income MDG in question (for example, relate progress in schooling to the initial distribution of schooling), which we refer to as an unconditional assessment. It can also be done by relating progress in non-income dimensions to the initial distribution of incomes (for example, relate progress in schooling to the initial distribution of income), which we call a conditional assessment because it is conditional on the position of households in the income distribution. In this chapter we focus on the distribution of progress towards the non-income MDGs along the income distribution, that is, the conditional assessment which we refer to as the conditional non-income growth incidence curve (NIGIC).1 The focus on the conditional NIGIC is for three reasons. First, since the income-poor are already suffering under the deprivation associated with low incomes, improvements in non-income dimensions of wellbeing are particularly important to them, particularly if one sees poverty as a multidimensional concept. Second, the income-poor particularly need non-income improvements (especially in health and education) to improve their earning opportunities to escape income poverty. Third, the conditional NIGIC is particularly important and relevant for policy-making. It provides us with an additional tool to investigate how the progress in non-income dimensions of the MDGs is distributed over the income distribution. This is very useful supplemental information to standard expenditure incidence analysis (Van de Walle and Nead 1995; Lanjouw and Ravallion 1998; Roberts 2003), which investigates how public resources and interventions are reaching various income groups. For example it provides an instrument to assess not only if public social spending programmes have reached the targeted income-poorest population groups but also if the public resources have generated the intended outcomes for different income groups. In this respect the conditional NIGIC might be a useful tool in pro-poor spending analysis or, more generally, in the evaluation of pro-poor policies. We illustrate our approach by using household survey data from Bolivia for 1989 and 1998. We order Bolivian households by their per capita income and investigate based on this income ranking the changes of non-income indicators with respect to the position of the household

Pro-Poor Progress and Non-Income Targets

125

in the income distribution. In addition to investigating relative changes (growth or progress), we also investigate absolute changes of the nonincome MDGs. As shown below, we find pro-poor progress towards the non-income MDGs in Bolivia, particularly when relative changes are used; the picture is much less clear when absolute changes are used. The rest of the chapter is organized as follows. First, we give a brief overview of how to measure pro-poor growth and pro-poor progress in a multidimensional way. Second, we explain how we apply the GIC to non-income MDGs. Third, we describe the data and present the results of the GIC and the NIGIC, conditional on the position in the income distribution, for selected MDGs. Last, we summarize and give an outlook for future research.

The concept of pro-poor progress The most glaring shortcoming of all attempts to define and measure propoor growth is that they rely exclusively on income as the only indicator of wellbeing or poverty.2 This means that they are focused on only MDG1 and leave out the multidimensionality of poverty which is at the heart of the MDGs that are based on income poverty reduction as only one of several goals.3 Measuring pro-poor growth only in the income dimension assumes implicitly that income growth is accompanied by non-income progress, or, to put it differently, that improvements of non-income indicators are a natural by-product of (pro-poor) economic growth. As it is well known in both the theoretical and the empirical literature, this need not be the case.4 While non-income indicators have recently received more and more attention in the concept and measurement of poverty,5 they have not been incorporated in the analysis of pro-poor growth, and no attempts have been made to date to measure pro-poor growth on the basis of non-income indicators. In line with the MDGs (UN 2000a), but subject to some data constraints, we have selected several non-income MDGs and use the conditional NIGIC to measure progress and its distribution of these nonincome MDGs along the income distribution. Regarding the definition of pro-poor growth, we follow Klasen (2008) and use three different definitions of pro-poor growth: weak absolute pro-poor growth, relative pro-poor growth, and strong absolute pro-poor growth. Pro-poor growth in the weak absolute sense means that growth rates are above zero for the poor. Pro-poor growth in the relative sense means that growth rates of the poor are higher than average growth rates, and thus that relative

126 Melanie Grosse, Kenneth Harttgen and Stephan Klasen

inequality falls (that is, some measure in which the relative gap between the rich and the poor is reduced). Pro-poor growth in the strong absolute sense requires that absolute income increases of the poor are stronger than average increases, thus that absolute inequality falls (that is, some measure in which the absolute gap between the rich and the poor falls). For a numerical illustration of these different definitions, see Klasen (2008).6 The latter definition can be seen as the strictest definition of pro-poor growth and hardest to meet, and so it is often ignored. But this neglects the fact that that decreases in relative inequality might be – and often are – accompanied by increases in absolute inequality which many people judge as unfair (for example, Atkinson and Brandolini 2004; Duclos and Wodon 2004; Klasen 2004). Increasing absolute inequality is arguably a particularly severe problem in considering non-income dimensions of wellbeing, as discussed in more detail below. One final note concerns terminology. In order to distinguish improvements in income and non-income dimensions, we refer to improvements in the income dimension as income growth, while we refer to improvements in non-income dimensions as non-income progress. This way we recognize that the term growth is, in most minds, inextricably bound up with incomes and thus we coined the term ‘progress’ for improvements in non-income dimensions.

Methodology For the calculation of the NIGIC we follow the general approach outlined in Grosse et al. (2008), which we describe briefly now. A measure of pro-poor growth often used is the GIC by Ravallion and Chen (2003), which plots the mean growth rate in income at each percentile of the distribution between two points in time. It shows how different population groups, that is, income groups, have benefited from income growth. In line with the definitions used above, if the GIC is above zero it indicates weak absolute pro-poor growth. If the GIC is negatively sloped it indicates relative pro-poor growth. To see weak absolute and relative pro-poor growth in a single number, one can also compute the propoor growth rate (PPGR), which is the area under the GIC up to the headcount, and compare it with the growth rate in the mean (GRIM). Accordingly, if the PPGR is above zero, we call this growth weak absolute pro-poor growth, and if the PPGR is higher than the GRIM, we can consider growth as being pro-poor in the relative sense. The calculation of the NIGIC7 follows broadly the concept of the GIC. Instead of income we use selected non-income indicators to measure pro-poor progress directly

Pro-Poor Progress and Non-Income Targets

127

via outcome-based welfare indicators. Thus, the NIGIC measures propoor progress not in an income sense but in a non-income sense – for example, the improvement of the health status or the educational level between two points in time for each centile of the income distribution. To study pro-poor progress in the strong absolute definition, we define the absolute NIGIC which shows the absolute changes in the nonincome MDGs for each centile of the income distribution. If the absolute NIGIC is negatively sloped, it indicates strong absolute pro-poor progress. Analogously to the PPGR and the GRIM, we define the ‘pro poor change’ (PPCH) and the ‘change in mean’ (CHIM) for the absolute NIGIC. If the PPCH is greater than the CHIM then we denote this progress as being strong absolute pro-poor progress. Because of the different dimensions of the income and non-income indicators, the fact that many of the non-income indicators are bounded above (for example, there is an effective upper limit to survival prospects or to educational achievements), and different levels of tolerance of inequality in different dimensions, it may well be plausible that different definitions of ‘pro-poor growth’ would be appropriate for different indicators. While one may be willing to accept the relative definition of propoor growth (the poor have higher income growth rates than average) in the income dimension, one may be willing to call progress in educational achievements or vaccination rates pro-poor only if the poor have higher absolute increments than the average, as only then would the absolute gap in achievements between the poor and the non-poor fall.8 When extending the concept of pro-poor growth to non-income dimensions, a range of conceptual issues regarding the nature of the non-income indicators including their scale, boundedness, ordinality versus cardinality, and other issues need to be considered. These issues are taken up in detail in Grosse et al. (2008) and Klasen (2008), who show that the approach can well be extended but that special care is needed when interpreting the results of the NIGICs. One last important point to note is that the data used here are not a panel data set and thus the percentiles of the income distribution in the initial and final period contain different households, so that we do not consider mobility of households but rather the development of income groups.9

Data For the empirical illustration we use demographic and health survey (DHS) data from Bolivia for the years 1989 and 1998. These data sets include households with at least one woman of reproductive age, that

128 Melanie Grosse, Kenneth Harttgen and Stephan Klasen

is, aged between 15 and 49, who serve as respondents. The data sets contain information on several non-income MDGs for 6053 and 8444 households for 1989 and 1998, respectively, for which we apply the pro-poor progress analysis. As not all variables are available for all households (for example, infant and child health and nutrition variables are available only for households which contain children), the sample size varies across the variables we use.10 For the purpose of this chapter, we have constructed the variables in such a way that they are as similar as possible to the formulation in the MDG targets. The results we present are valid only for the indicators themselves, but results might have differed had we used other indicators and there might sometimes be better ways to track MDGs which was, however, not possible with the data at hand. For example, literacy rates (details below) are a very crude measure of education as compared with years of schooling, for example. When interpreting the results, one should be aware that there is limited comparability across the variables since they are measured in different scales. And depending on the scales (on the variation of the indicators), the relative and absolute improvements can be quite similar if, for example, the values of the indicators are close to 100 per cent. For MDGs 1–7 we investigate several targets directly. Analysing MDG1, we have information on the poverty headcount ratio (Target 1a), on the poverty gap (Target 2), and on the prevalence of stunted children aged between 1 and 5 years as an indicator of the nutritional status, and, thus, an indicator of Target 4 (prevalence of underweight children). Here, we use the stunting z-score that measures chronic undernutrition for children and we consider children as stunted if the z-score is below −2 standard deviations from the median of the reference category (WHO 1995).11 For MDG2 we have information on primary completion rates (Target 7b) of adults12 and on literacy rates for female adults aged between 15 and 24 years (Target 8).13 For MDG3 the data sets contain information about the ratio of years of education of women to those of their partners (Target 10)14 and about the share of women in wage employment in the non-agricultural sector (Target 11). For MDG4 we can analyse both under-5 and under-1 mortality rates (Targets 13 and 14).15 MDG5 is analysed using the proportion of births in the household attended by skilled health personnel (Target 17). We use another variable for health concerning child immunization and the battle against diseases. Here, we take the household average number of vaccinations of children aged between 1 and 5 years, with a maximum of 8 possible vaccinations for each child.16 The average number of vaccinations per household represents access to healthcare and preventive medicines and can therefore

Pro-Poor Progress and Non-Income Targets

129

be seen as an indirect measure of MDG4 and of MDG6. A similar variable has, for example, been used in monitoring the health sector reform project in Bolivia in 1999 (Montes 2003). For MDG7, we have information on both access to clean water and access to sanitation (Targets 30 and 31). Unfortunately, the DHS data sets do not contain information on income or consumption owing to its focus on demographics, health and fertility. To overcome this problem, in our DHS data set we use simulated incomes based on a dynamic cross-survey microsimulation methodology (Grosse et al. 2005).17 The basic idea of this simulation methodology is to estimate incomes by combining information from two surveys: first, the DHS (of 1989 and 1998) and, second, the Bolivian household surveys (LSMS, that is, the second EIH of 1989 and the ECH of 1999). With this they estimate an income correlation in the household survey, apply the coefficients to the DHS, and predict – that is, simulate – incomes in the DHS (including a stochastic error term).18 As shown in Klasen et al. (2007), such an approach generates more plausible estimates of incomes than asset indices, which are often used in place of incomes when DHS data are analysed (for example, Filmer and Prichett 1998; Sahn and Stifel 2003).

Results Descriptive statistics Compared with other regions such as some countries in South Asia and most of sub-Saharan Africa, countries in Latin America have better chances of reaching the MDGs, particularly the ones where some absolute achievement is the target (UN 2005). For instance, many Latin American countries already have near-universal primary schooling rates (MDG2), have low gender gaps in education (MDG3), and have comparatively good access to reproductive health services, water and sanitation access (individual targets for MDGs 5 and 7). In addition, Bolivia experienced relatively high income growth rates in the 1990s (which also were pro-poor in both urban and rural areas). However, Bolivia was and is one of the poorest countries of the region, and the positive economic trend has reversed since 1999, combined with some episodes of social and political turmoil. In addition, Bolivia is one of the countries in Latin America with a very unequal income distribution. And in social indicators such as life expectancy or literacy, despite improvements since the 1980s, Bolivia still has quite poor outcomes compared with other countries in Latin America (see, for example, Klasen

130 Melanie Grosse, Kenneth Harttgen and Stephan Klasen

et al. 2007). The high income inequality and the persisting poverty motivate us to investigate the distributional pattern of the progress towards the MDGs, that is, to investigate if the improvements towards the MDGs made so far are higher for the poor than for other population subgroups. Table 5.1 shows the aggregate progress for selected MDGs in Bolivia between 1989 and 1998. Comparing the two points in time we see an overall sizeable progress towards reaching the MDGs during this decade. The very high headcount ratio for moderate poverty has fallen from 77 per cent to 60 per cent and that for extreme poverty from 56 per cent to 35 per cent between 1989 and 1998. However, Bolivia is still far from reaching the goal of headcount ratios for moderate and extreme poverty of 38 per cent and 28 per cent, respectively. More progress has been made in halving the poverty gap, especially for those suffering from extreme poverty.19 Much progress is also observed in reducing the prevalence of stunted children. In 1989 the prevalence of stunted children was 38 per cent. This means that the goal for the year 2015 is to reach 19 per cent. In 1998 we see that the prevalence rate of stunted children has already decreased to 23 per cent, which can be seen as a large step towards achieving the target. Concerning the achievement of universal primary education, the literacy rate has increased from 71 per cent to 83 per cent, and primary completion (of adults) has increased from 57 per cent to 71 per cent. For the goal to promote gender equality, we also see much progress. The ratio of female to male years of schooling has risen considerably, and the share of women in waged employment outside the agricultural sector has also increased considerably. Turning to the health indicators, we find relatively high but well-decreasing rates of child and infant mortality; for example, child mortality rates decreased from 135/1000 to 97/1000. The vaccination level increased only slightly from 5.30 to 5.48 of 8 possible vaccinations. Looking at the maternal healthcare indicator, we find also considerable progress: the birth attendance rate in Bolivia increased by around 40 per cent (from 41 per cent to 57 per cent). Last, water and sanitation access in Bolivia has been notably improved. In particular, the proportion of those who have access to clean water has increased from 47 per cent to 70 per cent. First evidence of the distribution of this progress can be seen when looking at Table 5.2, which shows the decile means (conditional on the income situation) of the selected MDGs and their 10:1 ratio. Overall the table reflects the positive and encouraging picture. Progress has been made for nearly all deciles for all indicators between the two points in time. For instance, the primary completion rate of the income-poorest

131 Table 5.1 Selected MDGs in Bolivia (1989 and 1998) 1989

1998

Eradicate extreme poverty and hunger Poverty headcount ratio∗ Poverty gap∗ Prevalence of stunted children

0.77 (0.56) 0.45 (0.28) 0.38

0.60 (0.35) 0.30 (0.14) 0.23

Achieve universal primary education Primary completion rate Literacy rate (females, aged 15–24)

0.57 0.71

0.71 0.83

0.71 0.24

1.05 0.45

Reduce child mortality U-5 mortality rate (∗ 100) U-1 mortality rate (∗ 100)

13.50 7.60

9.73 6.51

Improve maternal health Vaccinations Birth attendance rate

5.30 0.41

5.48 0.57

Water and sanitation Access to water Access to sanitation

0.47 0.50

0.70 0.68

Promote gender equality Ratio of education of women to men Share of women in wage employment in non-agricultural sector

Notes: The explanation of the variables for the tables and figures in this chapter is the following: Prevalence of stunted children: Stunting z-score of the last born child aged 1–5 of each respondent (averaged over the household) with a child defined to be stunted if the z-score is below −2. Primary completion rate: Primary completion can only be measured for respondents and their partners due to missing information on children in the DHS 1989. Literacy rate: It can only be computed for women (restricted to age 15–24 as in the MDG target) due to missing information for males. We use the answer ’reads easily’ to the question on literacy, thus those who answer ’cannot read’ or ’reads with difficulty’ are concerned as illiterate. Ratio of education of women to men: We use years of schooling of respondents (only those who have a partner and know their partners’ education) to their partners, averaged over the household. Share of women in wage employment in non-agricultural sector: We use the occupation codes and take the share of respondents in the household who are working outside the agricultural sector. Under-5 (1) mortality: We use life table estimates and take the sample of children born 10 (5) years prior to the sample. Here, we do not average over the household or over the mother, but treat each child separately. Vaccinations: Average vaccinations of the children in the household older than 1, where the possible vaccinations are three against polio, three against DPT, one against measles, and one against tuberculosis. Birth attendance rate: The share of children born in the household in the past 12 month for which a doctor attended during delivery. Access to water: Share of households with access to piped water. Access to sanitation: Share of households that have a toilet. ∗ Two poverty lines are used. The first number refers to the moderate poverty line, and the number in parentheses refers to the extreme poverty line.

132 Melanie Grosse, Kenneth Harttgen and Stephan Klasen

decile increased between 1989 and 1998 from 30 per cent to 40 per cent. The ratio of female to male years of schooling (that is, of respondents and their partners) increased considerably and in several deciles even exceeds one.20 Very strong improvements can also be seen for access to infrastructure and to assistance at giving birth where access for the poorest decile strongly increased. The mean number of vaccinations did not change much for all deciles but child survival increased considerably. When we look at inequality, the 10:1 ratio in the last column of Table 5.2 shows evidence of high inequality in Bolivia both for income and for non-income indicators of poverty along the income distribution. Most apparently, as expected, we find very high income inequality, with a 10:1 ratio of almost 40 that has decreased only slightly in the 1990s. The same holds for the share of women in wage employment. While Table 5.1 shows high overall progress, Table 5.2 shows high inequality both in 1989 and 1998, which indicates that the very poor have not benefited disproportionately from the progress. The 10:1 ratio has decreased only from 10.80 to 9.67. A slightly different picture shows the birth attendance rate. Although the level of inequality was also quite high in 1989, we can see that in 1998 the 10:1 ratio decreased from 12.86 to 3.21, indicating a reduction in relative inequality. The same holds also, but to a lesser extent, for the infrastructure indicators of access to water and sanitation. Here also the relative inequality decreased. Table 5.2 already gives an overview of how the progress is distributed across the income deciles and how this distribution has changed over time. The income gradient becomes obvious for most of the indicators, and it continues to be relevant over time. In the next section we visualize the changes of the distribution in a more detailed way by using the NIGIC. The NIGIC is basically the graphical implementation of how the distribution has changed along the centiles of Table 5.2, absolutely and relatively, using percentiles instead of deciles. Using the NIGIC and the PPGR–GRIM and PPCH–CHIM comparisons, we asses the pro-poor progress towards the MDGs. We investigate the questions, first, how the progress of non-income indicators in Tables 5.1 and 5.2 is distributed over the entire income distribution and, second, to what extent this progress can be considered as pro-poor using the various definitions. Pro-poor progress towards the non-income MDGs Starting our analysis with the income growth assessment, Figure 5.1 shows the GIC for Bolivia from 1989 to 1998. Over the whole distribution the GIC is above zero, indicating weak absolute pro-poor growth. Furthermore, it is negatively sloped, especially for the very poor centiles

Table 5.2

Non-income achievements by income decile (conditional on income, Bolivia, 1989 and 1998) 1

2

3

4

5

6

7

8

Mean of the deciles (conditional on income), 1989 21.88 40.27 57.50 77.33 100.61 132.39 177.08 246.12 Income∗ Prevalence of stunted children 0.43 0.45 0.53 0.41 0.38 0.28 0.33 0.29 Stunting z-scores (∗ 100) −180.19 −174.40 −183.05 −156.47 −151.49 −127.47 −128.45 −130.67 Primary completion rate 0.30 0.34 0.41 0.49 0.58 0.63 0.70 0.75 Literacy rate (females, aged 15–24) 0.43 0.57 0.64 0.65 0.75 0.78 0.81 0.83 Ratio of education of women to men 0.53 0.65 0.61 0.70 0.71 0.85 0.80 0.81 Share of women in wage employment 0.05 0.08 0.13 0.17 0.22 0.27 0.30 0.33 in non-agricultural sector U-5 survival rate 0.85 0.87 0.87 0.85 0.86 0.88 0.89 0.89 U-1 survival rate 0.91 0.91 0.92 0.93 0.92 0.94 0.94 0.95 Vaccinations 5.01 4.94 5.36 5.10 5.43 5.82 5.73 6.04 Birth attendance rate 0.07 0.20 0.29 0.38 0.47 0.58 0.73 0.80 Access to water 0.16 0.27 0.32 0.40 0.47 0.52 0.59 0.68 Access to sanitation 0.16 0.21 0.28 0.37 0.49 0.58 0.70 0.76

9

368.36 0.16 −89.68 0.82 0.84 0.91 0.42 0.91 0.94 6.32 0.86 0.70 0.86

10

10:1

863.39 39.46 0.15 0.34∗∗ −74.66 0.41∗∗ 0.88 2.93 0.86 2.00 0.90 1.71 0.54 10.80 0.91 1.07 0.93 1.01 6.45 1.29 0.90 12.86 0.82 5.13 0.92 5.75 (Continued)

133

134

Table 5.2

(Continued) 1

2

3

4

5

6

Mean of the deciles (conditional on income), 1998 36.37 63.60 89.26 119.22 155.89 203.15 Income∗ Prevalence of stunted children 0.38 0.34 0.32 0.27 0.23 0.22 Stunting z-scores (∗ 100) −157.37 −136.99 −133.95 −124.03 −115.69 −106.45 Primary completion rate 0.40 0.50 0.57 0.63 0.69 0.72 Literacy rate (females, aged 15–24) 0.60 0.67 0.72 0.79 0.81 0.88 Ratio of education of women to men 0.83 0.90 0.97 1.00 1.09 1.17 Share of women in wage employment 0.16 0.20 0.29 0.31 0.39 0.48 in non-agricultural sector U-5 survival rate 0.87 0.91 0.90 0.89 0.91 0.90 U-1 survival rate 0.92 0.94 0.94 0.93 0.92 0.94 Vaccinations 5.20 5.18 5.03 5.39 5.40 5.73 Birth attendance rate 0.28 0.36 0.40 0.53 0.64 0.74 Access to water 0.37 0.43 0.52 0.57 0.65 0.72 Access to sanitation 0.27 0.40 0.45 0.53 0.63 0.71

7

8

9

269.64 0.19 −89.53 0.80 0.88 1.26 0.50

369.20 0.14 −83.75 0.83 0.88 1.10 0.62

555.27 0.13 −71.39 0.86 0.93 1.13 0.64

0.90 0.93 5.94 0.74 0.81 0.78

0.93 0.95 5.97 0.86 0.85 0.84

0.93 0.95 6.09 0.90 0.87 0.88

10

10:1

1242.66 34.17 0.07 0.18∗∗ −39.75 0.25∗∗ 0.95 2.38 0.97 1.62 1.01 1.21 0.73 9.67 0.95 0.96 6.61 0.90 0.95 0.97

1.09 1.04 1.27 3.21 2.57 3.59

Notes: For the explanation of the variables, see Table 5.1. ∗ Real household income per capita in Bolivianos per month household. ∗∗ In the case of the prevalence of stunted children the 10:1 ratio indicates higher inequality for a low value and lesser inequality for higher values.

Pro-Poor Progress and Non-Income Targets

135

500

6

300 4 200 3

Absolute change

Annual growth rate, %

400 5

100

0

2 0

10

20

30

40

GIC (lhs)

50 60 Percentile

70

80

90

100

GIC absolute (rhs)

Figure 5.1 Growth incidence curve and absolute change for income

of the distribution, which fulfils the requirement of relative pro-poor growth. This finding of relative pro-poor growth and the relatively high growth rates of the 1990s is also reflected in Table 5.3, which shows the growth and progress rates and absolute changes of the selected MDGs. Table 5.3 shows a GRIM of 3.88 per cent for income. The PPGR of 4.53 per cent exceeds the GRIM, reinforcing the finding of relative pro-poor growth. Using the extreme poverty line with a headcount of 56 per cent, the growth process is even more pro-poor. Investigating strong absolute pro-poor growth, one can look at the absolute NIGIC in Figure 5.1 which shows anti-poor growth using the strong absolute definition as the absolute NIGIC is positively sloped. This finding is also reflected in that the CHIM is higher than the PPCH for both poverty lines, where the PPCH of the extremely poor is even lower than of the moderately poor. This finding of anti-poor income growth in the strong absolute sense is not surprising since the criterion of absolute pro-poor growth is very hard to meet in reality, as shown empirically by White and Anderson (2000). However, especially for the non-income dimensions of poverty, it is even more important for the poor to be able to catch up, and it might also be met more easily owing to the bounded-above character of most of the variables.

136 Melanie Grosse, Kenneth Harttgen and Stephan Klasen Table 5.3 Pro-poor growth and pro-poor progress in Bolivia (between 1989 and 1998)

Indicator Eradicate extreme poverty and hunger Income∗ Stunting z-scores Achieve universal primary education Primary completion rate Literacy rate (females, aged 15–24) Promote gender equality Ratio of education of women to men Share of women in wage employment in non-agricultural sector Reduce child mortality U-5 survival rate U-1 survival rate Improve maternal health Vaccinations Birth attendance rate Water and sanitation Access to water Access to sanitation

Pro-poor growth (progress) rate** (PPGR)

Growth (progress) rate in mean (GRIM)

4.53 (4.62) 0.91 (0.81)

3.88 0.78

54.66 (37.63) 38.98 (33.95)

97.71 36.15

2.43 (2.87) 1.61 (1.79)

1.67 1.28

0.12 (0.13) 0.10 (0.10)

0.11 0.10

3.96 (4.02) 7.94 (8.69)

3.40 5.57

0.33 (0.32) 0.17 (0.15)

0.30 0.18

0.34 (0.38) 0.10 (0.15)

0.33 0.13

3.01 (3.35) 0.92 (1.33)

2.98 1.20

0.13 (0.10) 4.17 (5.43)

0.08 1.81

0.06 (0.04) 0.13 (0.15)

0.04 0.10

4.60 (5.10) 3.70 (4.55)

3.22 1.96

0.19 (0.18) 0.14 (0.15)

0.18 0.12

Pro-poor change** (PPCH)

Change in mean (CHIM)

Notes: For the explanation of the variables, see Table 5.1. ∗ Real household income per capita in Bolivianos per month. ∗∗ Two poverty lines are used. The moderate poverty line is based on a headcount of 77 per cent. The numbers in parentheses refer to an extreme poverty line with a headcount of 56 per cent.

From Figure 5.2 onwards, we show the conditional NIGIC and the absolute conditional NIGIC for the selected MDG variables. As the conditional curves are very volatile, we include additionally the smoothed curves in the figures to show better the trend of the variables.21 Figure 5.2 shows the results for the prevalence of being stunted, that is, chronically malnourished children. As the NIGIC is above zero for nearly all centiles, we find weak absolute pro-poor progress. However, there is no clear trend in the curve, although it seems to be slightly downward-sloping on average. This is also reflected in the fact that the PPGR is slightly larger than the GRIM, indicating slightly pro-poor progress using the relative definition. When considering absolute improvements, there is less evidence of a downward-sloping absolute NIGIC, and the comparison of

Pro-Poor Progress and Non-Income Targets

250

4 0 2 ⫺250

Absolute change

Annual growth rate, %

6

137

0

⫺500

⫺2 0

10

20

30

40

50 60 Percentile

NIGIC smoothed (lhs) NIGIC absolute smoothed (rhs)

70

80

90

100

NIGIC (lhs) NIGIC absolute (rhs)

Figure 5.2 Conditional NIGIC and absolute change for stunting (z-score ∗ 100)

the PPCH and the CHIM shows that absolute progress was evenly spread across the distribution. Figures 5.3 and 5.4 monitor the distribution of progress towards MDG2. The primary completion rate of poor households increases, and this increase is higher for the income-poor than for the income-nonpoor. Thus, with the NIGIC being above zero and strongly negatively sloped, we find weak absolute and relative pro-poor progress. We also find strong absolute pro-poor progress, although of lower magnitude. The absolute increases in primary completion are higher for the poor than for the non-poor, which is also reflected in the (slightly) higher PPCH compared with the CHIM. Nearly the same findings hold for the literacy rate (of females aged 15–24), but the relative pro-poor progress and the strong absolute pro-poor progress are less pronounced. Figures 5.5 and 5.6 show the trends for the selected targets of MDG3. For the ratio of educated women to men we find that progress is high for all income groups, with both high proportional increases and absolute changes. Especially in the income-poorest groups of, say, the first three deciles, gender equality increased a lot, as shown in Table 5.2. The share of women in non-agricultural wage employment has increased strongly across the income distribution, but the increases have not been

138

0.5

6

0.25

4 0 2

Absolute change

Annual growth rate, %

8

⫺0.25

0

⫺0.5

⫺2 0

10

20

30

40

50 60 Percentile

NIGIC smoothed (lhs) NIGIC absolute smoothed (rhs)

70

80

90

100

NIGIC (lhs) NIGIC absolute (rhs)

12

0.5

8

0.25

4

0

0

⫺0.25

⫺4

⫺0.5 0

10

20

30

40

50

60

70

80

90

100

Percentile NIGIC smoothed (lhs) NIGIC absolute smoothed (rhs)

NIGIC (lhs) NIGIC absolute (rhs)

Figure 5.4 Conditional NIGIC and absolute change for literacy rate

Absolute change

Annual growth rate, %

Figure 5.3 Conditional NIGIC and absolute change for primary completion rate

139

1

0.5

8

Absolute change

Annual growth rate, %

12

0 4

⫺0.5 0 ⫺1 0

10

20

30

40

50 60 Percentile

NIGIC smoothed (lhs) NIGIC absolute smoothed (rhs)

70

80

90

100

NIGIC (lhs) NIGIC absolute (rhs)

Figure 5.5 Conditional NIGIC and absolute change for ratio of education of women to men

24

0.5

0.25 18 0

12 8

Absolute change

Annual growth rate, %

20

⫺0.25 4 ⫺0.5

0 0

10

20

30

40

50 60 Percentile

NIGIC smoothed (lhs) NIGIC absolute smoothed (rhs)

70

80

90

100

NIGIC (lhs) NIGIC absolute (rhs)

Figure 5.6 Conditional NIGIC and absolute change for share of women in wage employment

140 Melanie Grosse, Kenneth Harttgen and Stephan Klasen

2

0.2

1

0

0

⫺0.2

⫺1 ⫺0.4 0

10

20

30

40

50

60

70

80

90

100

Percentile NIGIC smoothed (lhs) NIGIC absolute smoothed (rhs)

NIGIC (lhs) NIGIC absolute (rhs)

Figure 5.7 Conditional NIGIC and absolute change for under-5 survival rate

Absolute change

Annual growth rate, %

the same everywhere. We find very pronounced weak absolute and relative pro-poor progress, especially for the very poor. This is reflected in the overall high progress, with a GRIM of 5.57 per cent, and the PPGR for the moderately poor of 7.94 per cent, which is exceeded by the PPGR of 8.69 per cent of the extremely poor. However, as concerns absolute changes, the increases are higher for the non-poor than for the poor, reflected in the positively sloped absolute NIGIC and the higher CHIM as compared with the PPCH. Clearly the expansion of female employment has been higher in absolute (that is, percentage point) terms among the rich than among the poor, suggesting that better-off women have benefited more from expanded female employment opportunities, which should be of interest for those concerned about gender equality in the labour market as well as those worried about further progress in poverty reduction, which will depend greatly on improving female employment opportunities. Figures 5.7 and 5.8 report pro-poor progress in under-5 survival and under-1 survival. The NIGICs are very volatile owing to the relatively small number of deaths per percentile and thus have to be interpreted with caution. The results in Table 5.3 show that there was weak absolute pro-poor progress, but that the income-poor did not benefit more than the rich, either in absolute or in relative terms. For the other variables

Pro-Poor Progress and Non-Income Targets

141

measuring health (in the sense of child and maternal health and access to healthcare) we find mixed evidence for pro-poor progress. The number of vaccinations children have received has increased only slightly and for some centiles it has even decreased, and there is no clear sign for pro-poor progress (see Table 5.3 and Figure 5.9). For the proportion of births attended by a doctor, we find quite impressive weak absolute and relative pro-poor progress, so the income-poor had better chances to deliver attended by skilled health personnel. Also the strong absolute criterion is fulfilled, in both the aggregate numbers (the PPCH exceeds the CHIM) and the curves (the absolute NIGIC) in Figure 5.10 is negatively sloped on average. Figures 5.11 and 5.12 shows progress in access to piped water and to basic sanitation by income group. Both variables show tremendous increases in access across the income distribution. For the income-poor population groups, progress was higher in relative terms and, in the case of sanitation, the increases are furthermore in line with the strong absolute criterion of pro-poor progress, whereas the absolute changes (measured in percentage points) in access to water are more equally scattered over the income distribution.

0.2

1.5

0

0

⫺0.2

⫺1.5

⫺0.4 0

10

20

30

40

50

60

70

80

90

100

Percentile NIGIC smoothed (lhs) NIGIC absolute smoothed (rhs)

NIGIC (lhs) NIGIC absolute (rhs)

Figure 5.8 Conditional NIGIC and absolute change for under-1 survival rate

Absolute change

Annual growth rate, %

3

142

1.5

6

3

0

1.5 ⫺1.5

0

Absolute change

Annual growth rate, %

4.5

⫺1.5 ⫺3

⫺3 0

10

20

30

40

50 60 Percentile

NIGIC smoothed (lhs) NIGIC absolute smoothed (rhs)

70

80

90

100

NIGIC (lhs) NIGIC absolute (rhs)

Figure 5.9 Conditional NIGIC and absolute change for vaccinations

30

0.5

0.25

20 15

0

10 5

⫺0.25

0 ⫺0.5

⫺5 0

10

20

30

40

50 60 Percentile

NIGIC smoothed (lhs) NIGIC absolute smoothed (rhs)

70

80

90

100

NIGIC (lhs) NIGIC absolute (rhs)

Figure 5.10 Conditional NIGIC and absolute change for birth attendance rate

Absolute change

Annual growth rate, %

25

143

0.5

16 0.25 12 0 8

Absolute change

Annual growth rate, %

20

4

⫺0.25

0

⫺0.5 0

10

20

30

40

50

60

70

80

90

100

Percentile NIGIC smoothed (lhs) NIGIC absolute smoothed (rhs)

NIGIC (lhs) NIGIC absolute (rhs)

Figure 5.11 Conditional NIGIC and absolute change for access to water

0.5

14

0.25

10 8

0

6 4

Absolute change

Annual growth rate, %

12

⫺0.25

2 0

⫺0.5

⫺2 0

10

20

30

40

50

60

70

80

90

100

Percentile NIGIC smoothed (lhs) NIGIC absolute smoothed (rhs)

NIGIC (lhs) NIGIC absolute (rhs)

Figure 5.12 Conditional NIGIC and absolute change for access to sanitation

144 Melanie Grosse, Kenneth Harttgen and Stephan Klasen

Conclusion In this chapter we have extended the methodology of pro-poor growth measurement to the assessment of pro-poor progress along the income distribution. This way we are able to see whether the income-poor have been able to benefit disproportionately from progress in nonincome dimensions of wellbeing. As stated above, this is important from a wellbeing perspective, will influence income poverty reduction in future, and allows us to assess the success or otherwise of public policies in reducing disparities in non-income dimensions of wellbeing. The illustration from Bolivia has demonstrated that such an approach is feasible and yields a range of important and interesting new insights about the distribution of progress in non-income dimensions of wellbeing which are critical for a more comprehensive assessment of progress towards the non-income MDGs. The overall picture of income growth and non-income progress in Bolivia is quite positive for the period from 1989 to 1998. For nearly all targets of the MDGs we find an improvement when looking at mean improvements. For most of the targets we also find that this progress was pro-poor, at least in the weak absolute and in the relative sense, suggesting that the poor participated in this progress at higher rates than the non-poor. This is particularly the case for the expansion of education, health attendance, female employment, and access to water and sanitation. It is much less the case in the income dimension, in stunting, and in infant and child mortality rates where progress was similar in relative terms across the income spectrum. Regarding absolute improvements, the record is much more mixed. While in education, sanitation access, and birth attendance absolute improvements were at least slightly pro-poor, they were of equal magnitude in most other indicators, with the exception of income where they were strongly anti-poor. How are we to interpret these findings from Bolivia? A few observations are of importance. First, the poor have benefited from progress in nonincome dimensions of wellbeing. This suggests that government policies in the 1990s to expand health and education services have reached the poor and led to significant improvements in their wellbeing and human capital.22 Bolivia in the 1990s is likely to be a case where pro-poor progress will have been particularly large. It would be interesting to replicate this analysis in sub-Saharan Africa where overall progress and the distribution of progress are likely to have been much more unfavourable. Second, the achievement is much less impressive if one considers that

Pro-Poor Progress and Non-Income Targets

145

many of the indicators are variables that are effectively bounded above and where the rich were already close to the upper bound and further progress for them was thus much harder to achieve. This is particularly the case for the education, mortality, birth attendance, and water and sanitation access indicators. In those situations, one would expect that any average progress would be pro-poor in the strong absolute dimension, as only the poor have a long way to go in these dimensions. Thus the findings of only slight pro-poor progress in the strong absolute definition using some of these indicators, and the lack of such pro-poor growth in others, are actually a disappointment. Third, this also suggests that public policies aimed at improving health and education were effective in reaching the poor but not successful in significantly narrowing the absolute gaps between the poor and the rich in quite a few of these indicators. This suggests in turn that more effort should be placed on ensuring that the poor are the disproportionate beneficiary (in an absolute sense) of public policy interventions. Lastly, the analysis shows that strong absolute pro-poor progress in non-income dimensions is feasible, but is very difficult to achieve in the income dimension where the absolute gap between the poor and the non-poor has continued to widen substantially. To what extent this is the natural state of affairs or alternatively is something that should be tackled by policy-makers is an interesting question for further research.

Acknowledgements The authors would like to thank Amartya Sen and Ravi Kanbur and participants at workshops in Göttingen, the Spring Meeting for Young Economists in Geneva, the Society for the Study of Economic Inequality in Palma, the Research Committee on Development Economics in Kiel, and Sonia Bhalotra, Mark McGillivray and participants of the UNU-WIDER project meeting of Assessing and Forecasting Progress in Millennium Development Goals in Helsinki for helpful comments and discussion.

Notes 1. See Grosse et al. (2008) and Klasen (2008) for related analyses also considering an unconditional assessment.

146 Melanie Grosse, Kenneth Harttgen and Stephan Klasen 2. In this chapter, we mention income only as the money-metric measure of living standard and do not distinguish between income and consumption. For a detailed discussion on the debate of income versus consumption as a measure, see for example Deaton (1997). 3. See Sen (1988, 1998) for a discussion on the multidimensionality of poverty. 4. See for example Anand and Ravallion (1993), Sen (1998), Klasen (2000), Ravallion (2001), Grimm et al. (2002) or World Bank (2006). 5. Examples of recent studies examining the multidimensional casual relationship between economic growth and poverty reduction are Mukherjee (2001), Bourguignon and Chakravarty (2003 and Summer (2003). Also, international organizations point to the importance of the direct outcomes of poverty eduction such as health and education (see for example UN 2000a, 2000b; World Bank 2000). 6. Most inequality measures, including the Atkinson, Gini and Theil measures as well as decile or quintile ratios, are relative inequality measures; for a discussion of the merits of also considering absolute inequality measures, see Atkinson and Brandolini (2004). 7. One might also name them ‘non-income progress incidence curves’. 8. A different way to deal with this problem would be to rescale the non-income variables, for example, by transforming the education indicator into a percentage shortfall from a maximum level, of say eighteen years of education, and then define progress as the percentage reduction in that shortfall. With such an indicator one may well decide to choose the relative definition as sufficient to define pro-poor progress. As discussed in Grosse et al. (2008), this issue will also arise when comparing the Gini coefficients of incomes with Gini coefficients in non-income indicators. 9. For an extension of these tools to panel data see Grimm (2007). 10. For example for the variable that measures undernutrition, the sample size drops to 1599 and 3287 households for 1989 and 1998, respectively. 11. We have transformed the z-score so that all numbers are strictly positive (by adding the lowest absolute value of the given z-scores (of about –6) to all numbers, so that the lowest z-score is 0) for the calculations. In the tables and figures, absolute values are given as z-score ∗ 100 for better visualization. The reason we use the height-for-age (stunting) z-score rather than the weightfor-age (underweight) z-score included in the MDGs is that data on weightfor-age are distorted by the presence of overweight children. It might be problematic that the z-score contains a lot of ‘genetic noise’, in the sense that for example a low z-score interpreted as being undernourished might simply appear because the parents are genetically short so that the child is small but well nourished. We only use the z-score for children above 1 to reduce the noise that would arise from the fact that, typically, stunting develops only during the first year of life. 12. This indicator cannot be calculated for children owing to missing information in the DHS of 1989. ‘Male adults’ here are only partners of the respondents in the households, but not single men, since there is no male module in the DHS of 1989. Thus, ‘partners’ are not a random sample of males. 13. We have used the question whether the respondent is able to read easily. This indicator cannot be calculated for males because information on partners’ literacy is missing from the DHS of 1989.

Pro-Poor Progress and Non-Income Targets

147

14. We cannot use information on literacy because this information is not given for men, so we use instead the ratio of average years of schooling of all respondents with partners in the household relative to their partners’ (it is not possible to calculate it just for the age group of 15–24 years owing to missing age information on partners). We think that this indicator might be even better than literacy itself to measure educational status because years of schooling better reflect the level of education than mere literacy. 15. In our calculation, we use child survival rates instead of child mortality rates. An improvement in child mortality comes out as a lower value, but this lower value is mathematically interpreted as a deterioration. The linear transformation used is: survival rate = (mortality rate − 1)∗ (−1). This means for example that a reduction of child mortality from 20 per cent to 10 per cent is transformed into an increase in child survival from 80 per cent to 90 per cent. Unlike with the other indicators, we use the average individual survival rates by centile (and do not produce average household survival rates first) and we compute them using life table estimations to be consistent with official estimates (thus the indicator is the probability that a child reaches its first or fifth birthday, respectively). For under-5 mortality we restrict the sample to those children born 10 years prior to the survey, and for under-1 mortality to those born 5 years prior to the survey. 16. The possible vaccinations are 3 against polio, 3 against DPT, 1 against measles, and 1 against tuberculosis (BCG). We do not include children under a year old as they cannot be expected to have completed the 8 recommended vaccinations. 17. For the calculation of the PPGR in the next section, we use the headcount of 77 per cent as found in Klasen et al. (2007) for the moderate poverty line and of 56 per cent for the extreme poverty line. We use the same headcounts for the calculation of the PPGR for all non-income indicators. 18. The authors estimate an income/consumption expenditure model in the 1999 LSMS data, restricting the set of covariates to those also available in the 1998 DHS data and interacting all variables with a rural/urban dummy. They then use the regression to predict incomes in the DHS and add a randomly distributed error term. They then repeat the procedure for the 1989 LSMS, which is available only in urban areas. When imputing incomes in rural areas, they use the model for urban areas in 1989 and add the results of the rural interaction terms from 1999, assuming that the difference in the impact of income correlates between 1989 and 1999 did not change over time. While the results work well in several validation tests, there is a tendency for the simulated income growth to be higher than the observed one, a subject currently under investigation. This overprediction should not bias the results in this chapter, but it might be useful to test the results generated here with a survey that contains detailed information on both income and non-income variables. 19. Please note that the progress in poverty reduction is larger than suggested in Klasen et al. (2007), where actual incomes were used in the final period. The discrepancy is currently under investigation. 20. This result should be treated with caution for two reasons. First, it reflects only the relative education of female respondents and their partners, where

148 Melanie Grosse, Kenneth Harttgen and Stephan Klasen such partners actually existed. This is a rather narrow indicator as it neglects women without partners as well as single men. Also, the results are due not only to educational investments along the income distribution, but also to marriage market effects which impact on how females and their partners match and form households. 21. As mentioned before, we consider improvements in non-income indicators as progress rather than growth. However, we keep the abbreviations of PPGR and GRIM in order not to confuse the reader. 22. For a discussion of these policies, see Klasen et al. (2007).

References Anand, S. and M. Ravallion (1993) ‘Human Development in Poor Countries: On the Role of Private Incomes and Public Services’. Journal of Economic Perspectives, 7(1):133–50. Atkinson, A. and A. Brandolini (2004) ‘Global World Inequality: Absolute, Relative or Intermediate’. Paper presented at the 28th General Conference of the International Association in Income and Wealth, 22–28 August, Cork. Bourguignon, F. and S. Chakravarty (2003) ‘The Measurement of Multidimensional Poverty’. Journal of Economic Inequality, 1(1):25–49. Deaton, A. (1997) The Analysis of Household Surveys. A Microeconomic Approach to Development Policy. Baltimore MD: Johns Hopkins University Press for the World Bank. Duclos, J. and Q. Wodon (2004) ‘What Is Pro Poor?’ Centre Interuniversitaire sur le Risque, les Politiques Économiques et l’Emploi (CIRPEE) Working Paper 04–25. Montreal. Filmer, D. and L. Prichett (1998) ‘Estimating Wealth Effects without Expenditure Data – or Tears: An Application to Educational Enrollments in States of India’. Demography, 38(1):115–32. Grimm, M. (2007) ‘Removing the Anonymity Axiom in Assessing Pro-Poor Growth’. Journal of Economic Inequality, 5(2):179–97. Grimm, M., C. Guénard and S. Mesplé-Somps (2002) ‘What Has Happened to the Urban Population in Côte d’Ivoire since the Eighties? An Analysis of Monetary Poverty and Deprivation over 15 years of Household Data’. World Development, 30(6):1073–95. Grosse, M., K. Harttgen and S. Klasen (2008) ‘Measuring Pro-Poor Growth in NonIncome Dimensions’. World Development, 36(6):1021–47. Grosse, M., S. Klasen and J. Spatz (2005) ‘Creating National Poverty Profiles and Growth Incidence Curves With Incomplete Income or Consumption Expenditure Data’. Background paper for the study: Operationalizing Pro-Poor Growth – Country Case Study Bolivia. Ibero America Institute for Economic Research, University of Göttingen, Discussion Paper 129. Klasen, S. (2000) ‘Measuring Poverty and Deprivation in South Africa’. Review of Income and Wealth, 46(1):33–58. Klasen, S. (2004) ‘In Search of the Holy Grail. How to Achieve Pro-Poor Growth?’. In B. Tungodden and N. Stern (eds), Towards Pro-Poor Policies: Proceedings from the ABCDE Europe. Washington, DC: World Bank.

Pro-Poor Progress and Non-Income Targets

149

Klasen, S. (2008) ‘Economic Growth and Poverty Reduction: Measurement Issues in Income and Non-Income Dimensions’. World Development (forthcoming). Klasen, S., R. Thiele, M. Grosse, J. Lay, J. Spatz and M. Wiebelt (2007) ‘Analysing Pro-poor Growth in Bolivia: Addressing Data Gaps and Modelling Policy Choices’. In M. Grimm, S. Klasen and A. McKay (eds), Determinants of Pro-Poor Growth. Basingstoke: Palgrave Macmillan, pp. 191–218. Lanjouw, P. and M. Ravallion (1998) ‘Benefit Incidence and the Timing of Program Capture’. Washington DC: World Bank. Montes, C. (2003) ‘Results-Based Public Management in Bolivia’. Overseas Development Institute Working Paper 202. London. Mukherjee, D. (2001) ‘Measuring Multidimensional Deprivation’. Mathematical Social Sciences, 42(3):233–51. Ravallion, M. (2001) ‘Growth, Inequality and Poverty: Looking Beyond Averages’. World Development, 29(11):1803–15. Ravallion, M. and S. Chen (2003) ‘Measuring Pro-Poor Growth’. Economics Letters, 78(1):93–9. Roberts, J. (2003) ‘Poverty Reduction Outcomes in Education and Health: Public Expenditure and Aid’. Overseas Development Institute Working Paper 210. London. Sahn, D. and D. Stifel (2003) ‘Exploring Alternative Measures of Welfare in the Absence of Expenditure Data’. Review of Income and Wealth, 49(4):463–89. Sen, A. (1988) ‘The Concept of Development’. In H. Chenery and T. Srinivasan (eds), Handbook of Development Economics, vol 1. Amsterdam: North Holland. Sen, A. (1998) Development as Freedom. New York: Knopf. Summer, A. (2003) ‘Economic and Non-Economic Wellbeing: A Review of Progress on the Meaning and Measurement of Poverty’. Paper prepared for WIDER Conference: Inequality, Poverty and Human Wellbeing, 30–31 May. Helsinki. UN (United Nations) (2000a) A Better World for All. New York. UN (United Nations) (2000b) ‘United Nations Millennium Declaration’. Document A/RES55/2. New York. UN (United Nations) (2003) ‘Getting Serious about Meeting the Millennium Development Goals: A Comprehensive Development Framework Progress Report’. New York. UN (United Nations) (2005) ‘The Millennium Development Goals Report 2005’. New York. Van de Walle, D. and K. Nead (eds) (1995) Public Spending and the Poor – Theory and Evidence. Baltimore and London: Johns Hopkins University Press for the World Bank. White, H. and E. Anderson (2000) ‘Growth versus Distribution: Does the Pattern of Growth Matter?’ Development Policy Review, 19(3):267–89. WHO (World Health Organization) (1995) ‘Physical Status: The Use and Interpretation of Anthropometry’. Technical Report Series 854. Geneva. World Bank (2000) World Development Report 2000/01 – Attacking Poverty. New York: Oxford University Press. World Bank (2006) World Development Report 2006 – Equity and Development. New York: Oxford University Press.

6 Links between Childhood Mortality and Economic Growth and Their Implications for Achieving the Millennium Development Goals in India Sonia Bhalotra

Introduction A set of time-bound targets for human development were agreed by 189 countries at the Millennium Summit held in New York in September 2000, and these are referred to as the Millennium Development Goals (henceforth MDGs). They represent an unprecedented commitment on the part of both rich and poor countries. One of the eight targets is to reduce under-5 mortality by two-thirds by the year 2015, relative to its level in 1990. This requires an annual rate of decline of about 4.3 per cent per annum.1 This chapter is motivated to assess the feasibility of meeting this target in India. India offers an appropriate setting for the analysis as it has 1 in 6 of the world’s people, 1 in 4 of under-5 deaths, and 1 in 3 of the world’s poor. The chapter documents trends in under-5 mortality and economic growth in India over the period 1970 to 1998. It estimates a model of under-5 mortality that includes a rich set of demographic and economic variables, including aggregate state income. The model is estimated on panel data so that the effect of income on mortality is identified by looking at deviations from trend and controlling for state-specific rainfall shocks that might otherwise drive a spurious correlation between shocks to state income and changes in child mortality. The estimated income elasticity is subject to a range of specification tests, including allowance for dynamics, endogeneity and measurement error. I investigate variation in the income elasticity across the Indian states and over time, considering especially whether it was greater or smaller before the onset of economic reform in the early 1980s. 150

Childhood Mortality–Economic Growth: India 151

Assuming that the income elasticity in the period 1998 to 2015 remains as estimated for the recent post-reform era (1981–98), I calculate the rate of growth that would be necessary to achieve a mortality rate in 2015 that is consistent with the MDG target. This required growth rate is compared with the actual growth rate in the post-reform period to assess the feasibility of the target being met. The main results are as follows. The unconditional income elasticity of under-5 mortality in India is about −0.7, which means that a 10 per cent increase in aggregate income is associated with a 7 per cent reduction in mortality. Including state fixed effects pushes the elasticity up to −1.0. This suggests that between-state heterogeneity tends to obscure the underlying relationship. Once I control for year effects, the income elasticity falls to −0.6. This is consistent with the year effects capturing trend improvements in health technology, the effects of which will tend to be projected upon a trended variable like income in a model that does not control for time effects. Although higher state income in India is associated with lower poverty and higher public expenditure on health, the direct effect of poverty and public expenditure on mortality is neither large nor well-determined once the model includes time dummies. Moreover, conditioning upon these variables does not wipe out the effect of income. This contradicts the results of Anand and Ravallion (1993) who show, using cross-country data for 22 developing countries, that income has no effect on health indicators once poverty and public expenditure are held constant. Estimates of state-specific elasticities show that childhood mortality is responsive to aggregate income changes in only 8 of the 15 major states. Estimates on subsamples split at 1981 indicate that growth was less effective in reducing mortality after 1981, which is when we might date the start of the reform process.

Why growth? The gap in life expectancy between rich and poor countries today stands at about thirty years, which is about the gain in life expectancy in the last century. So, crudely speaking, the differential is worth a century of progress and this differential is largely explained by excess childhood mortality (Cutler et al. 2006). It is therefore natural to ask how important aggregate income growth is in causing reductions in childhood mortality. The literature is inconclusive on this question (see Fuchs 2004; Cutler et al. 2006; Deaton 2006). The common problems that arise are potential feedback from health to income through productivity effects,

152 Sonia Bhalotra

the presence of unobservables such as medical technical progress or climate that influence both health and income, and endogenous selection into birth or up to a certain age. This chapter attempts to address these problems by identifying the impact of aggregate income shocks on child mortality risk using state-level panel data. We may expect income to have a larger effect on mortality in poorer countries where the level of income is low, the level of mortality is high and neither markets nor the state provide adequate insurance. In contrast to the situation in richer countries, where injuries and accidents are the main cause of childhood death, in poorer countries like India, the main causes of childhood death are poor maternal health, undernutrition and the prevalence of infectious diseases like malaria, diarrhoea and respiratory infections.2 So, most childhood deaths in developing countries are avoidable, occurring for want of household resources, public services and information. Improvements in aggregate income may raise private incomes and public spending and be associated with increases in education. Increases in private incomes among the poor will tend to improve maternal and child nutrition. Increases in public spending may avert deaths by improving sanitation, so that less infection is bred, or by increasing the prevalence of skilled midwives and of hospital facilities that might take care of delivery complications. There is a considerable role for information in the production of health by both prevention and cure, and it seems that education makes parents more efficient at acquiring and applying relevant information. The estimated effect of growth on mortality is expected to capture all of these relationships. Being a reduced form type of effect, it will also capture any interactions between these variables. For instance, we may expect the extent to which private or public health spending increases health (or survival chances) to depend upon the level of education of the parent. Also, household and public expenditures on health may be complementary. For example, Jalan and Ravallion (2003) find that the favourable effect of piped water (which depends on public spending) on diarrhoea is lower in poorer households, especially those with less-educated mothers. Overall, growth in aggregate income provides resources that facilitate the interventions necessary to reduce mortality. However, the effectiveness of growth in delivering reductions in mortality depends upon other factors, including the inclusiveness of growth, fiscal policy, education and information, health infrastructure and the political economy of public service delivery. The size of the income elasticity of mortality is therefore an empirical question, and one on which there is as yet limited evidence (see the next section).

Childhood Mortality–Economic Growth: India 153

The question of how income growth affects welfare is of wide academic and policy interest. Evidence on the distributional impact of growth has generated concern that the poor do not share equally in the benefits of growth (Bhalla 2002; Dollar and Kraay 2002; Wade 2002). Previous discussions of pro-poor growth have been dominated by investigation of income poverty.3 However, poverty and mortality reductions may not go hand in hand. Looking at world data, it seems that inequalities in income have not narrowed as much as inequalities in health, and although growth appears to have contributed to poverty reduction, its effects on health are more uncertain (Deaton 2006). Research on the effects of growth on mortality is limited and, as discussed below, the few available studies provide what appear to be conflicting results. There is, therefore, a clear niche for further research on this subject. The income elasticity is a natural parameter of interest if the question is, ‘How much would mortality decline, on average, if there were no specific intervention to aid this?’ A focus on the role of growth implies no favour for growth as the instrument for mortality reduction. Indeed, this chapter concludes that growth cannot be relied upon to reach a level of mortality consistent with the MDG target.4

Related research Previous research on child mortality has focused on its microdeterminants (for example Wolpin 1997). This section reviews evidence of the effects of economic growth. The impact of economic growth on childhood mortality Using US and UK data on age-specific mortality matched to cohort income, Deaton and Paxson (2001, 2004) find that the effect of income is small and unstable and that education is more effective in improving longevity. Analysis of a state-level panel of US data in Ruhm (2000) shows that recessions are associated with improvements in health across age groups. Dehejia and Lleras-Muney (2004) confirm and explore this result for infant health. They argue that this unexpected result arises, at least partly, because growth tends to raise the opportunity cost of time, and time is an important input into health. They show, for the state of California, that women are less likely to seek antenatal care in economic upturns, presumably because this is when they increase their market labour supply. Using Indian data I find that the opposite is the case: that is, infant mortality rises in recessions. I investigate potential mechanisms,

154 Sonia Bhalotra

including fiscal policy, health-seeking behaviours and women’s labour supply. I find that antenatal care, deliveries outside the home, child immunizations and treatment of the main childhood diseases are all lower in recessions. This appears related to pro-cyclicality in state health and development expenditure, and to women having to work harder outside the home in recessions (in direct contrast to the result for American women), presumably to compensate a fall in the main earner’s income (see Bhalotra 2007a). Van den Berg et al. (2006) analyse longitudinal data on cohorts born across almost two centuries in Utrecht. Since the Netherlands was agricultural and had living standards comparable to today’s developing countries through a lot of this period, their results are pertinent to today’s poor countries. They find that aggregate income changes were associated with reductions in mortality risk and that survivors enjoyed a further gain in life expectancy from having been exposed to a high-income environment in childhood. Many other studies of the link between growth and mortality in poor countries analyse large, one-off economic crises, and the results of these studies are mixed (see Paxson and Schady 2005). Using data for 36 Asian countries for the year 2000, Tandon (2005) estimates an unconditional elasticity (that is, a correlation, absent of any controls for time and country effects and for other covariates) of −0.7. In their landmark paper, Pritchett and Summers (1996) show, using data for 58 developing countries in 1960–85, that growth in GDP is associated with proportional declines in mortality. Since international data on under-5 mortality are collected only at five-yearly intervals, they estimate the model in fifth-differences.5 This yields an elasticity of mortality with respect to growth of −0.15, significant at 5 per cent, after controlling for time effects. This falls to −0.12 when education is included in the model. A higher elasticity, of −0.31, is obtained when differencing is replaced by inclusion of country fixed effects. The elasticity is also larger when a single long-difference is taken. Using annual panel data for the Indian states in 1980–99, Deolalikar (2005: ch. 2) finds that there is no significant relationship between income and infant mortality once a linear trend is included in the model (also see World Bank 2004).6 Using microdata for 1994–8, the same study finds a positive association of infant mortality with state income. The microdata are from the same source as used in this study. However, their mortality data are for the five years preceding 1998/9 whereas the results here are for the thirty years preceding this date, and so our results are not directly comparable. What might contribute to the ineffectiveness of income in the Deolalikar study is that it is a partial effect, obtained

Childhood Mortality–Economic Growth: India 155

conditional upon public health spending and infrastructure, which are themselves functions of income.7 Some other researchers have similarly found positive associations between GDP growth and infant or adult mortality in Argentina, Brazil and Chile (Ortega and Reher 1997; Rios and Carvalho 1997; Abdala et al. 2000). Palloni and Hill (1997) show that short-term mortality responses to recessions in the second half of the twentieth century in nine Latin American countries were erratic and statistically insignificant. These contrasting results establish the pertinence of the current analysis. Overall, evidence of the effects of aggregate income changes on child mortality is surprisingly scarce and the available evidence is inconclusive. While analysis of a cross-country data set delivers negative effects, analyses of Indian and Latin American data have produced ambiguous effects, sometimes positive and sometimes insignificant. The feasibility of attaining the MDG for mortality Two previous studies have attempted to assess whether India will achieve the MDG in health. These are described here. The first is more pessimistic than the second, but they are not comparable because they use different approaches, and make different assumptions. Tandon (2005) documents the annual rate of change in under-5 mortality between 1990 and 2000 in 36 Asian countries. India ranks 19 of 36, with an annual rate of decline of less than 3 per cent per annum. This is well below the MDG-driven target of 4.3 per cent per annum indicated at the beginning of this chapter. In looking at India’s performance, it is useful to note that Bangladesh has done much better despite having slower economic growth than India over this period. It exhibits a rate of decline of under-5 mortality close to 5 per cent per annum, and ranks 6 in 36. Using an unconditional between-country estimate of the income elasticity of −0.7, Tandon estimates that the average Asian economy in the sample will have to grow at a rate of 6 per cent per annum in order to achieve the target reduction in under-5 mortality of 4.3 per cent per annum. He acknowledges that, for countries like India that have had mortality declining at less than 4.3 per cent per annum so far, required growth needs to be even faster in order to catch up. Using an income elasticity that is more likely to have a causal interpretation, this study quantifies the growth rate required. The World Bank (2004) report discussed earlier simulates the rate of infant mortality in 2015 under a set of assumptions concerning the rate at which seven significant and policy-amenable predictors will evolve between 1998/9 (when the survey data were gathered) and 2015. These

156 Sonia Bhalotra

predictors are years of maternal schooling, per capita government expenditure on health and family welfare, population coverage of each of electricity supply, tetanus toxoid immunization for pregnant women, antenatal care and access to toilets, and village-level access to pucca (decent-quality) roads. Applying to these rates of change the parameters estimated in a multivariate probit model, the study concludes that the infant mortality goal, and hence the under-5 mortality goal, is achievable in principle. Since this conclusion depends upon the assumed rates at which the named education, health spending and infrastructure or service variables develop, the study performs two further simulations. It isolates the high mortality (and poor) states of Bihar, Madhya Pradesh, Orissa, Rajasthan and Uttar Pradesh from the other (non-poor) states on the grounds that they account for more than half of all childhood mortality. In the first simulation, it takes the levels of the named predictors in these states up to the national average and then, in the second simulation, it takes them up to the average for the non-poor states. The latter procedure yields a rate of decline in the same ballpark as the original simulation, underlining its potential feasibility. The study is careful to point out that actually achieving the target depends, for a given quantity of public expenditure, on its composition and delivery.

Data The mortality data used here are derived from the second round of the national family health survey (NFHS) conducted in 1998/9; see IIPS and ORC Macro (2000) for details of the survey and sampling strategy. The survey interviewed ever-married women aged 15–49 at the time of the survey. Every mother reported a complete retrospective history of the incidence and timing of live births and any child deaths. As births in the sample occurred between 1961 and 1999, these data have (unexploited) potential to shed light on trends in fertility, mortality and related demographic change. Issues of possible sample selection in these data are discussed in Bhalotra (2007a). The analysis is restricted to the 15 major states of India, which account for more than 95 per cent of the country’s population since these are the states for which state-level time series of aggregate income are readily available. Over the sample period, 1970–98, the data contain 163,907 children of 50,379 mothers. The focus in this study is on under-5 mortality. So I define an indicator variable for child j in family i that is 1 if the child is reported to have died before the age of 60 months and 0 otherwise. To allow five years of exposure to under-5 mortality risk for all children in the sample, children who

Childhood Mortality–Economic Growth: India 157

were less than 60 months old at the time of the survey (roughly, children born after 1994/5) are excluded from the analysis. I aggregate the microdata to the state level using sample weights to produce annual mortality rates by state. I similarly construct state-level indicators of education and demographics using the NFHS microdata, selecting characteristics that have been shown to be significant predictors of mortality risk in a number of previous studies, and also on these microdata (see Arulampalam and Bhalotra 2006). Indicators are, in aggregation to the state level, transformed into proportions – for example, gender and birth-order of the child and the religion, ethnicity and sectoral (rural/urban) location of the household. The educational level of mother and father and the age of the mother at birth of the child are entered as a set of indicators. Some of these characteristics are potentially correlated with economic growth, for example the educational level of parents. For this reason, I present estimates of the effect of income on mortality obtained before and after introducing the control variables. The state-level demographic data are merged with a panel of data on real net state domestic product per capita (income) and other relevant statistics for the fifteen Indian states, over the chosen period. These data were assembled by Ozler et al. (1996) and then extended by Besley and Burgess (2002, 2004). The merge is done by state and time, calendar time in the panel being matched to the year of birth of the child in the microdata (henceforth t). So for children born in 1980 and exposed to the risk of under-5 death during 1980–5, I have matched information on income in 1980. In the estimated model, I regress the under-5 mortality rate for children born in year t on income averaged over the period (t, t + 5), which is the relevant exposure period.

Descriptive statistics There are vast differences in the level of mortality across the Indian states, which demonstrate the scope for reduction in the overall level. Averaging over the period, the incidence of mortality ranged from 4.8 per cent in Kerala to 17.3 per cent in Uttar Pradesh. The average linear rate of decline in mortality during the period analysed here, 1970–94, is estimated at 2.83 per cent per annum. The average linear rate of growth of income during 1970–94 is estimated at 2.61 per cent per annum. Statespecific rates of growth vary considerably, ranging between 4.4 per cent in Maharashtra (the industrial capital) and 1.8 per cent in Bihar (one of the very poor states) (see Table 6.1).

158 Sonia Bhalotra Table 6.1 1970–98

Level and change of under-5 mortality and income: all-India and states,

State Andhra Assam Bihar Gujarat Haryana Karnataka Kerala Madhya Maharashtra Orissa Punjab Rajasthan Tamil Nadu Uttar Pradesh West Bengal All India

Linear rate of Linear rate of Mean Std dev. change in change in mortality (%) (income) mortality p.a. (%) income p.a. (%) 10.9 7.8 12.1 12.0 9.3 10.8 4.8 17.5 8.7 14.2 6.9 15.2 9.9 17.3 9.8 11.1

0.36 0.23 0.20 0.38 0.41 0.31 0.32 0.30 0.41 0.30 0.35 0.25 0.41 0.21 0.27 0.31

−3.4 −1.7 −3.0 −3.8 −2.5 −4.1 −7.1 −3.6 −4.9 −3.3 −2.1 −3.2 −5.3 −4.4 −5.1 −3.9

3.8 2.6 1.8 3.6 3.0 3.3 3.0 3.1 4.4 3.1 2.8 2.2 4.2 2.0 2.7 3.0

Notes: Income is the logarithm of real per capita net state domestic product. The means and standard deviations (std dev.) are for the period 1970–98 for each region. The growth rates in the last two columns are obtained as coefficients in a regression of the mortality rate or log income, as the case may be, on a linear trend term. The rank correlation coefficient between mortality and log income is −0.50.

Figure 6.1 shows trends in under-5 mortality for each of the 15 major states of India. Mortality has declined fairly steadily in every state, with some convergence over time. This said, some states with initially low mortality (like Kerala) have achieved further declines at a rapid pace, while other states (like Assam or Punjab) that started out with relatively low levels of mortality have progressed at a more sluggish pace (see also Table 6.2). Figure 6.2 shows trends in the logarithm of real per capita net state domestic product (income) for every state. There was a fairly steady increase, with some acceleration in most states starting in the 1980s. There was limited, if any, convergence across the states. The fact that there is convergence in mortality rates across states but not in income may be explained by improvements in health technology that, independent of income, have diffused across the states.

Childhood Mortality–Economic Growth: India 159

Under-5 mortality rate

0.4

AP AS BI GU HA KA KE MP MT OR PU RA TN WB UP

0.3

0.2

0.1

0.0 1970

1980 1990 Year of birth of child

2000

Figure 6.1 Trends in under-5 mortality by state Notes: AP = Andhra Pradesh, AS = Assam, BI = Bihar, GU = Gujarat, HA = Haryana, KA = Karnataka, KE = Kerala, MP = Madhya Pradesh, MT = Maharashtra, OR = Orissa, PU = Punjab, RA = Rajasthan, TN = Tamil Nadu, WB = West Bengal, UP = Uttar Pradesh. Colour-coded versions of these figures are available at http://www.efm.bris.ac.uk/www/ecsrb/bhalotra.htm

Figure 6.3 plots the unconditional relationship between under-5 mortality and income trends by state. The relationship is clearly negative in every state, although the slopes vary. In Figure 6.4a, I construct weighted averages of the state data and plot all-India trends in mortality and income. As mortality has been declining and income increasing over time, any correlation between these series will be spurious to the extent that it picks up common trends. For this reason, Figure 6.4b plots the two series after de-trending both. So what we have in Figure 6.4b is the relationship that we are really interested in identifying: the relation of growth and mortality after taking out any other trended variables that might otherwise confound the relation. A casual glance at it suggests that, after controlling comprehensively for omitted trended variables like advances in health technology and services and declines in fertility, there is no evident relation of growth and mortality. The following section explores this more carefully, using a state-level panel and conditioning upon covariates other than income, including state-specific rainfall shocks.

160 Sonia Bhalotra Table 6.2 Changes in under-5 mortality and income: all-India and states, 1970–81 and 1982–94

State Andhra Pradesh Assam Bihar Gujarat Haryana Karnataka Kerala Madhya Pradesh Maharashtra Orissa Punjab Rajasthan Tamil Nadu Uttar Pradesh West Bengal All India

Under-5 mortality

Aggregate income

1970–81

1982–94

1970–81

1982–94

−1.79 −4.28 −4.77 −6.77 −1.98 −2.78 −9.46 −2.03 −6.80 −4.49 −0.59 −3.39 −6.14 −5.50 −4.31 −4.40

−0.94 −1.58 −0.66 −3.05 −2.57 −3.56 −9.86 −2.36 −2.63 −2.82 −0.66 −1.28 −4.03 −2.69 −4.58 −2.88

1.39 2.49 1.30 3.14 2.30 1.58 3.07 1.53 4.66 1.68 2.25 −0.73 1.74 1.25 1.62 1.95

4.86 1.30 1.36 3.02 3.11 3.74 3.00 2.80 4.58 2.93 3.39 2.75 5.46 1.41 2.01 3.05

Notes: These are linear rates of growth obtained by a simple regression of the log of mortality or income, as the case may be, on a trend. The all-India regression includes a set of state dummies. All figures are percentages.

The econometric model Let M denote the under-5 mortality rate, let Y denote aggregate income, and let subscripts s and t denote state and year respectively. Then the estimated model may be written as follows: ln Mst = αs + αt + β ln Yst + λk ln Zkst + qr ln Xrst + ust

(6.1)

The parameter of interest is β, the elasticity of mortality with respect to income.8 The equation includes year and state fixed effects, denoted αt and αs , respectively. There are 15 states (s) observed over the course of 25 years (t), 1970–94, giving us a long and thin panel data set. Children born in 1993/4 are exposed to the risk of under-5 death till 1998/9, so the income data used extend up until 1998/9. Equation (6.1) represents the simplest baseline model, but I also investigated dynamics, which previous research in this area appears not to have done.

Childhood Mortality–Economic Growth: India 161

AP AS BI GU HA KA KE MP MT OR PU RA TN WB UP

8.0

log State income

7.5

7.0

6.5

6.0 1970

1980

1990

2000

Year Figure 6.2 Trends in real log income per capita by state Note: See Figure 6.1.

log Under-5 mortality

⫺1

AP AS BI GU HA KA KE MP MT OR PU RA TN WB UP

⫺2

⫺3

⫺4

⫺5 6.0

Figure 6.3 by state

6.5

7.0 log State income

7.5

8.0

The relationship of under-5 mortality and state income: quadratic fit

Note: See Figure 6.1.

I included the first and second lags of both mortality and income as additional regressors in the model. As these were insignificant, they were not retained. The equation is estimated by the least-squares dummy variables method (within-groups). I also investigated the systems estimator,

162 Sonia Bhalotra

GDP(5) & under-5 mortality: All-India

0.6

iunder5

0.4

0.2

0

⫺0.2 70

75

Detrended GDP(5) & under-5 mortality: All-India

(a)

80

85

90

95

Year of birth of child

0.06

Detrended under-5 mortality

Detrended GDP

0.04

0.02

0

⫺0.02 ⫺0.04 70

(b)

igdpe5

75

80 85 Year of birth of child

90

95

Figure 6.4 All-India trends in under-5 mortality and state income: populationweighted averages. (a) Actual series. (b) Detrended series

using lags of income to instrument its current level and testing overidentifying restrictions associated with rainfall shocks and education. I was unable to reject exogeneity of income, and the instrumental variables estimate of the income coefficient was not significantly different from

Childhood Mortality–Economic Growth: India 163

the OLS (within groups) coefficient. The means and standard deviations of all variables used in the analysis are in Bhalotra (2007a), which estimates a model similar to the model estimated here but using microdata for infant mortality rather than state-level data on under-5 mortality. Bhalotra (2007a) presents a more careful investigation of the mechanisms underlying the identified effect of aggregate income changes on mortality. State fixed effects control for initial differences in mortality and income, and for persistent elements of history, climate, culture (for example, the status of women) and other institutions (including public service delivery, corruption). The year fixed effects control comprehensively for aggregate time-variation associated with common improvements in health technology, rainfall variation, terms of trade shocks and so on. The Indian sub-national panel has an important advantage over cross-country panels which is that, within a country, the common trends specified in a within-groups model will capture more of the relevant variation. Trends in, for example, technology or fertility, are much less likely to be common in a panel of developing countries from the three different continents. It is not uncommon in the broader literature to exclude time effects and so to report inflated effects of income on human development outcomes. For instance, Ravallion and Datt (1996) appear not to include time effects in their panel data regressions concerning the effect of income on poverty. Deaton and Paxson (2004), using US and UK data, show that omission of time effects in the model tends to inflate the contribution of income to mortality reduction. The variables Xr are mostly economic variables and, like income, they are defined at the state level. They include inequality, poverty, public spending, relative growth of the agricultural sector, relative prices (rural/urban) and price inflation. The covariates Zk are demographic variables, obtained from the NFHS at the child or family level and then aggregated up to the state level. They include gender, religion, ethnicity, educational level of mother and father, and age of the mother at birth of the index child. Education and the other demographic variables contribute to controlling for household living standards. The NFHS does not have information on income or consumption at the household level. It has information on housing conditions and ownership of durables which can be used to construct a wealth index (for example, Filmer and Pritchett 2001). I do not use the wealth information because it pertains to the time of the survey, whereas the births and deaths of children that we are interested in

164 Sonia Bhalotra

occurred over a long (retrospective) period. To investigate the extent to which education proxies wealth in these data, I regressed the household wealth index on the educational levels of mothers and fathers of children born in the three years before the survey. The R2 of this regression is 0.37, which suggests that parental education is a fairly good proxy for the socioeconomic status of the household. While I cannot rule out the possibility that income effects in these data are partly proxying omitted household income, and I partly investigate this by including the poverty rate in the model, inclusion of the microdata controls suggests a supplyside (macro) interpretation of the income effect. Below, I specifically investigate the role of public expenditure on health.

Results The unconditional elasticity of mortality with respect to aggregate income is −0.71, significant at the 1 per cent level.9 Once time and state dummies are included in the model, this falls to −0.59, and remains significant (Table 6.3). The other rows of Table 6.3 show that this elasticity is robust to inclusion of other covariates, including public health expenditure and poverty. The conditional elasticity of −0.59 that I estimate is similar to that in Kakwani (1993), who also uses cross-country data. Conditional upon education, Pritchett and Summers (1996) estimate an elasticity that varies between about −0.12 and −0.31, depending upon the estimator. Their survey of previous estimates of the effect of growth on child mortality suggests that estimates cluster around the figure of −0.20 (for example Flegg 1982; Hill and King 1992; Subbarao and Raney 1995). However, these lower estimates are all partial elasticities, emerging from models that condition on variables like infrastructure or health expenditure that are themselves a function of the level of income. The state and time dummies are each jointly significant at the 1 per cent level. Conditional on state and time effects, within- and betweensector inequality, poverty, relative prices (agriculture relative to industry) and inflation are all insignificant. If I estimate the model using individual data on mortality rather than the state-level average, then I find that inflation has a significant mortality-increasing effect (Bhalotra 2007a). Using a quadratic in state expenditure on health and family welfare, I find that this has a significant mortality-reducing effect only at high levels of expenditure. However, in Bhalotra (2007b), I show that allowing for lagged effects produces a significant coefficient for rural households with a linear specification. A disaggregate analysis indicates that the

Childhood Mortality–Economic Growth: India 165 Table 6.3 The income elasticity of under-5 mortality: alternative sets of control variables

Covariates other than log income 1 2 3 4 5 6 7 8 9

None State dummies Year dummies but no state dummies State & year dummies + Inequality (Gini) + Health expenditure + Poverty gap index + Sectoral composition of income, relative prices, price inflation, income shocks, rainfall shocks + Maternal age at birth, maternal and paternal education, gender, ethnicity, religion

Income elasticity

t-statistic

−0.71 −0.97 −0.51 −0.59 −0.55 −0.51 −0.56 −0.64

13.8 12.4 6.6 3.5 3.5 3.1 3.2 3.0

−0.50

2.2

Notes: These are estimates from a within-groups model estimated on state-level panel data with N = 15 and T = 25 (1970–94). The additional regressors shown are cumulative. In other words, unless otherwise indicated, row j has all of the regressors shown in row j − 1 and also those named in row j. Precise definitions of the covariates are in Bhalotra (2007a). The absolute t-statistics reported in the last column are based on Newey-West standard errors that allow for heteroscedasticity and autocorrelation.

state health expenditure effect is significant in only 5 of the 15 states, these being Karnataka, Maharashtra, Tamil Nadu, Uttar Pradesh and West Bengal. The only significant compositional effects in the current model are secondary-level education among fathers, which reduces child mortality, and belonging to a scheduled tribe, which increases mortality. The microdata equation estimated in Bhalotra (2007a) produces significant coefficients on most of the demographic variables. The difference in the results I find here with a state-level panel suggests that the evolution of the other demographic variables can be represented by the time and state effects in the panel model. Differences in the income elasticity across the states I allow the coefficient on income to be state-specific by interacting income with state dummies. I find that income has a mortality-reducing effect in only 8 of the 15 states (see Table 6.4). In these eight states, the elasticity varies between −0.5 and −0.9, with the exception of Kerala, where the elasticity is a remarkable −1.7. Comparing the estimated effect of log income on mortality (β) with the coefficients on the state fixed effects (αs ), I find that the states that were

166 Sonia Bhalotra Table 6.4 Growth elasticities and fixed effects by state

Andhra Pradesh Assam Bihar Gujarat Haryana Karnataka Kerala Madhya Pradesh Maharashtra Orissa Punjab Rajasthan Tamil Nadu Uttar Pradesh West Bengal

Elasticity

Fixed effect

−0.20 [1.03] −0.05 [0.15] −0.36 [0.76] −0.53* [1.99] −0.18 [0.66] −0.57* [2.14] −1.69** [4.03] −0.24 [1.01] −0.71** [3.43] −0.56* [2.35] 0.77 [0.91] −0.27 [0.91] −0.72** [3.01] −0.94* [2.51] −0.89** [2.94]

0.00 n.a. −0.44** [7.18] −0.08 [0.40] 0.24* [2.41] −0.13 [1.02] 0.04 [1.07] −0.97** [11.71] 0.43** [9.24] 0.035 [0.45] 0.20** [4.57] −1.11 [1.69] 0.23** [2.67] −0.048 [1.31] 0.40** [7.95] 0.12 [1.69]

Notes: The reported figures are estimates from a model of under-5 mortality that includes additive state and year fixed effects, and interacts income with each of the 15 state dummies. The elasticity is significant in only 8 of the 15 states. The fixed effects coefficients are relative to Andhra Pradesh (normalized to zero), and 7 states are estimated to have significantly different fixed effects as compared with Andhra. Absolute t -statistics are in brackets and significance is indicated by asterisks, * denoting significance at the 5-per-cent level and ** denoting significance at the 1-per-cent level.

relatively ineffective in translating growth into lower mortality (that is, states with a small absolute income elasticity) were not those with an inherently high mortality risk (that is, states with large fixed effects).10 This is encouraging for policy because it suggests that the states in which growth does not significantly reduce mortality can more easily make their growth ‘pro-poor’ (that is, mortality-reducing) than would be the case if their observed inefficacy were tied to the sort of persistent historical or institutional factors that state fixed effects tend to capture. Was the income elasticity larger in the post-reform era? The average growth rate of real income per capita was barely 1 per cent per annum in the 1960s and 1970s but, since the early 1980s and especially since about 1993, it has been distinctly higher, averaging 4.8 per cent per annum between 1993/4 and 1999/2000. The upturn in the growth rate coincided with the onset of economic liberalization in India. A gradual process of reform was set in motion in the early to mid 1980s

Childhood Mortality–Economic Growth: India 167

and this accelerated in the 1990s. Whether the reforms caused higher growth, and how, is debatable (see Bhalotra 1998; Clark and Wolcott 2002; DeLong 2002; Virmani 2004) and, in any case, is not the subject of this chapter. However, it is interesting to investigate whether the additional growth and the structural change associated with reform altered the growth elasticity of mortality. For a review of concerns about the impact of structural adjustment on mortality, see Hill and Pebley (1989). Their discussion underlines that the effect can go either way, making this an important question to investigate empirically. To do this, I split the sample at 1981. A break-point in 1980/1 or 1981/2 is indicated by the analysis in Virmani (2004), who tests for structural breaks in income growth in India over the period 1950–2002. Possibly relevant is that the Congress Party returned to power in 1980/1, initiating a new approach to economic management in view of growing awareness of the growth-inhibiting constraints of its earlier regime. Table 6.2 summarizes rates of growth of income and rates of decline in mortality for the two periods created by a break in 1981. It is clear that, even as income accelerated, mortality decelerated. Refer to Table 6.5, where row 1 reports the benchmark estimate of −0.60 from row 4 in Table 6.3. Rows 2 and 3 show the ‘pre-reform’ and ‘post-reform’ elasticities to be −0.82 and −0.44, respectively, and I am able to reject the null that these are equal at the 10 per cent significance level.11 This result is consistent with the argument that the Indian reforms were anti-poor (childhood mortality is concentrated among the poor: see Victora et al. 2003, for example). However, since the mortality rate is bounded from below at zero, we may expect the elasticity to decrease as the level of mortality decreases even in the absence of any structural change. This is especially the case since, as the incidence of under-5 mortality declines, the fraction of neonatal deaths in all under-5 deaths tends to rise, and neonatal deaths are less closely tied to fluctuations in income (Bhalotra and van Soest 2008 contains the relevant references). All that can be safely concluded is that the post-1981 period was not associated with growth becoming evidently more pro-poor than before. Simulation to the MDG target Given that the MDG for 2015 is benchmarked to the level of mortality in 1990, the annual rate of decline in mortality needs to be 4.3 per cent per annum over this 25-year period. I estimate the average linear rate of decline of under-5 mortality per annum in India between 1970 and 1995 to have been 2.83 per cent per annum,12 which implies that from

168 Sonia Bhalotra Table 6.5 Was there a ‘structural break’ in the income elasticity? Sample 1 2 3

1970–1994 (entire period) 1970–1981 (‘pre-reform’) 1982–1994 (‘post-reform’)

Elasticity

t-statistic

–0.59 –0.82 –0.44

3.5 2.8 1.9

Notes: The dependent variable is the log of under-5 mortality, as in Tables 6.3 and 6.4. The equations include state and year fixed effects. Standard errors are Newey–West. The elasticities −0.81 and −0.44 are significantly different (F (1,167) = 2.7, p > F = 0.103).

1995 onwards a rate of decline faster than 4.3 per cent per annum will be necessary. We can recalculate the rate of decline that will be required, benchmarking to a more recent year than 1990. The under-5 mortality rate in 1998 was 9.5 per cent and the target for 2015 is 3.2 per cent (see World Bank 2004:2). Over the 17-year period between, mortality needs to decline at 6.2 per cent per annum in order for the target to be met. Were mortality to decline at 2.83 per cent per annum between 1998 and 2015, the level in 2015 would be 5.83 per cent, which exceeds the target by 2.63 percentage points. These are simple extrapolations, which assume that the predictors of mortality (including income) evolve at a constant rate, and that the parameters of the mortality equation are constant over long periods of time. Table 6.1 shows that income growth rates have varied over time and the previous section shows that the income elasticity of mortality has not remained constant over time. The exercise is therefore only illustrative. The more specific question posed at the start of this chapter was: if we were to rely upon income growth alone, how far would we be from the MDG target? Suppose, as discussed, that the rate of decline of mortality that is necessary between 1998 and 2015 is 6.2 per cent per annum. Using the estimated elasticity of mortality with respect to income for the period 1981–94 of −0.44 (Table 6.5, row 3), we can see that this rate of decline will flow from a rate of income growth of 14.1 per cent per annum. But actual income growth in the period 1981–94 was 3.1 per cent per annum, and the required growth rate is too high to be feasible. Another way of presenting these data is to say that were income to continue to grow at 3.1 per cent per annum then growth alone would generate an annual rate of decline of mortality of 1.36 per cent per annum, other things being equal. This would result in an under-5 mortality rate of 7.52 per cent in 2015, which is 4.32 percentage points above the target level.

Childhood Mortality–Economic Growth: India 169

Conclusions Growth does help reduce mortality. The average effect in India is fairly large and quite robust. Yet growth alone will not deliver mortality reduction at the rate necessary to reach the MDG target. Appropriate policy responses need to recognize that a given level of growth is consistent with different rates of mortality reduction, indicating the importance of other factors such as time-varying unobservables that most likely reflect improvements in public health and medical knowhow which appear to have contributed significantly to mortality decline, and to the convergence of mortality rates across the Indian states. Although year-to-year variation in electoral competition does not appear to influence the quantity or effectiveness of state health expenditure, there is evidence that the state fixed effects in a model of child mortality are strongly correlated with the levels of variables like turnout, party competition and media activity. In other words, political economy is a part of the explanation for health differences across the states (Bhalotra and Schmid 2007). Five Indian states account for more than half of all childhood mortality (World Bank 2004). Interventions need to be concentrated in these states. Although this was not specifically investigated in this study, the data show that under-5 death probabilities are higher among girls, first-born children and children of low-caste or uneducated mothers. Targeting these relatively vulnerable groups will bring down average mortality incidence.

Acknowledgements I am grateful to Arthur van Soest for many helpful comments, and for inviting me to RAND where this study was written. I have benefited from collaboration with Wiji Arulampalam on related topics. I would like to thank Mark McGillivray for encouraging me to write this chapter and Tony Addison for introducing me to UNU-WIDER.

Notes 1. Let M1990 be the under-5 mortality rate in 1990 and let M2015 be the target rate to be achieved by 2015. The total reduction over the 25-year period is (2/3)M1990. So per annum, it is 1 − (1/3)1/25 = 0.0429. 2. The main proximate causes of death, as summarized in Black et al. (2003), are birth asphyxia, diarrhoea, HIV/AIDS, malaria, measles, neonatal tetanus

170 Sonia Bhalotra

3. 4.

5.

6.

7.

8.

9.

10. 11.

and neonatal sepsis, pneumonia and preterm delivery. WHO (1992:table 1) estimates that infectious and parasitic diseases (mainly diarrhoea, respiratory diseases like pneumonia and tuberculosis) accounted for 71 per cent of all under-5 deaths in the developing world. Vulnerability to disease is a function of maternal health and child nutritional status – these factors do not appear in classifications such as that of the WHO because they are ‘ultimate’ or underlying rather than proximate causes of death. See, for example, Ravallion (2004). India-specific studies of the impact of growth on poverty include Besley et al. (2005) and Ravallion and Datt (2002). A pragmatic reason why the literature often looks at growth effects rather than at the effects of ‘intermediate’ variables like public expenditure is that it is usually easier to find long and consistent regional time series data on income. Differencing the data induces autocorrelation in the error term, to address which a GLS or GMM estimator is appropriate. This issue is typically not discussed or addressed. Also, the quinquennial data will miss out on higherfrequency fluctuations in mortality such as are associated with economic upswings and downswings. What exactly the addition of a trend does to the income elasticity cannot be read off annex table 2.1 in the cited report because, in the specification that includes a trend, there is a further change, namely, that income is interacted with public health expenditure. The trend and the interaction term are negative and significant but each of income and health expenditure become insignificant. As the Deolalikar/World Bank study does not control for trended unobservables, the effects of these are potentially loaded onto the income coefficient. However, this cannot explain the failure to find a negative income effect. This is because, if the unobserved trends are dominated, as we expect, by improvements in technology and declines in fertility, both of which have a mortality-reducing effect, then their omission will tend to inflate the (negative) effect of income on mortality. The estimated relationship is in log-levels. So the level (incidence) of mortality is associated with a level of aggregate income and, differencing, economic growth is associated with mortality reduction Since the model includes time dummies, it is similar to a differenced model. This happens to be almost exactly the same as the unconditional elasticity reported for the UK and the USA in Deaton and Paxson (2004) and for a cross-section of Asian countries in Tandon (2005). A similar result is reported in World Bank (2004). The regression for 1982–94, like the regression for the full period, 1970–94, allows every child in the sample full exposure to the risk of under-5 mortality, and the income variable matched to births in 1994 is the average of income over 1994–9. To similarly allow for full exposure for every child in the period 1970–81, I re-estimate this model on data for births in 1970–7, with death rates for births in 1977 being modelled as a function of income averaged over 1977–81. The pre-reform elasticity is now −1.37 rather than −0.81, and its difference from the post-reform elasticity of −0.44 is significant at the 5 per cent level.

Childhood Mortality–Economic Growth: India 171 12. I have confirmed that this rate of change is the same for all India as it is for the aggregate of the fifteen major Indian states used here (and listed in Table 6.1).

References Abdala, F., R. Geldstein and S. Mychaszula (2000) ‘Economic Restructuring and Mortality Changes in Argentina: Is there any Connection?’ In G. A. Cornia and R. Paniccià (eds), The Mortality Crisis in Transitional Economies. Oxford University Press for UNU-WIDER. Anand, S. and M. Ravallion (1993) ‘Human Development in Poor Countries: On the Role of Private Incomes and Public Services’. Journal of Economic Perspectives, 7(1):133–50. Arulampalam, W. and S. Bhalotra (2006) ‘Sibling Death Clustering in India: Scarring vs. Unobserved Heterogeneity’. Journal of the Royal Statistical Society, Ser. A, 169(4):829–48. Besley, T. and R. Burgess (2002) ‘The Political Economy of Government Responsiveness: Theory and Evidence from India’. Quarterly Journal of Economics, 117(4):1415–52. Besley, T. and R. Burgess (2004) ‘Can Labor Regulation Hinder Economic Performance? Evidence from India’. Quarterly Journal of Economics, 19(1): 91–134. Besley, T., R. Burgess and B. Esteve-Volart (2005) ‘Operationalising Pro-poor Growth: India Case Study’. Mimeo, Department of Economics, London School of Economics and Political Science. Bhalla, S. (2002) Imagine There’s No Country: Poverty, Inequality and Growth in the Era of Globalization. Washington, DC: Institute for International Economics. Bhalotra, S. (1998) ‘Changes in Utilization and Productivity in a Deregulating Economy’. Journal of Development Economics, 57(2):391–420. Bhalotra, S. (2007a) ‘Fatal Fluctuations? Cyclicality in Infant Mortality in India’. IZA Discussion Paper 3086, Bonn: Institute for the Study of Labour. Bhalotra, S. (2007b) ‘Spending to Save? Health Expenditure and Infant Mortality in India’. Health Economics, 16(9):911–28. Bhalotra, S. and A. van Soest (2008) ‘Birth-Spacing, Fertility and Neonatal Mortality in India: Dynamics, Frailty and Fecundity’. Journal of Econometrics, 143(2):274–90. Bhalotra, S. and J-P. Schmid (2007) ‘The Political Economy of Health Expenditure in India’. Mimeo, University of Bristol. Black, R., S. Morris and J. Bryce (2003) ‘Where and Why are 10 Million Children Dying Every Year?’ Lancet, 361:2226–34. Clark, J. and S. Wolcott (2002) ‘One Polity, Many Countries: Economic Growth in India, 1873–2000’. In D. Rodrik (ed.), Economic Growth: Analytical Country Narratives. Princeton University Press. Cutler, D., A. Deaton and A. Lleras-Muney (2006) ‘The Determinants of Mortality’. Journal of Economic Perspectives, 20(3):97–120. Deaton, A. (2006) ‘Global Patterns of Income and Health: Facts, Interpretations and Policies’. WIDER Annual Lecture 10. Helsinki: UNU-WIDER.

172 Sonia Bhalotra Deaton, A. and C. Paxson (2001) ‘Mortality, Education, Income and Inequality among American Cohorts’. In D. Wise (ed.), Themes in the Economics of Aging. University of Chicago Press. Deaton, A. and C. Paxson (2004) ‘Mortality, Income and Income Inequality over Time in Britain and the US’. In D. Wise (ed.), Perspectives in the Economics of Aging. University of Chicago Press. Dehejia, R. and A. Lleras-Muney (2004) ‘Booms, Busts, and Babies’ Health’. Quarterly Journal of Economics, 119(3):1091–30. DeLong, J. B. (2002) ‘India since Independence: An Analytical Growth Narrative’. In D. Rodrik (ed.), Economic Growth: Analytical Country Narratives. Princeton University Press. Deolalikar, A. B. (2005) Attaining the Millennium Development Goals in India: Reducing Infant Mortality, Child Malnutrition, Gender Disparities and Hunger-Poverty and Increasing School Enrollment and Completion. Oxford University Press for the World Bank. Dollar, D. and A. Kraay (2002) ‘Growth Is Good for the Poor’. Journal of Economic Growth, 7(3):195–225. Reprinted (2004) in A. Shorrocks and R. van der Hoeven (eds), Growth, Inequality and Poverty: Prospects for Pro-Poor Economic Development. Oxford University Press for UNU-WIDER. Filmer, D. and L. Pritchett (2001) ‘Estimating Wealth Effects without Expenditure Data – or Tears: An Application to Educational Enrolments in States of India’. Demography, 38(1):115–32. Flegg, A. (1982) ‘Inequality of Income, Illiteracy and Medical Care as Determinants of Infant Mortality in Underdeveloped Countries’. Population Studies, 36(3): 441–58. Fuchs, V. R. (2004) ‘Reflections on the Socio-economic Correlates of Health’, Journal of Health Economics, 23(4):653–61. Hill, A. and E. King (1992) ‘Women’s Education in the Third World: An Overview’. In E. King and A. Hill (eds), Women’s Education in Developing Countries: Barriers, Benefits and Policy. Baltimore: Johns Hopkins University Press for the World Bank. Hill, K. and A. Pebley (1989) ‘Child Mortality in the Developing World’. Population and Development Review, 15(4):657–87. IIPS and ORC Macro (2000) National Family Health Survey (NFHS-2) 1998–9:India. Mumbai: International Institute for Population Sciences. Jalan, J. and M. Ravallion (2003) ‘Does Piped Water Reduce Diarrhoea for Children in Rural India?’ Journal of Econometrics, 112(1):153–73. Kakwani, N. (1993) ‘Performance in Living Standards: An International Comparison’. Journal of Development Economics, 41(2):307–36. Ortega, O. and D. Reher (1997) ‘Short-term Economic Fluctuations and Demographic Behaviour: Some Examples from 20th-century South America’. In G. Tapinos, A. Mason and J. Bravo (eds), Demographic Responses to Economic Adjustment in Latin America. Oxford: Clarendon. Ozler, B., G. Datt and M. Ravallion (1996) ‘A Database on Poverty and Growth in India’. Mimeo, World Bank. Washington, DC. Palloni, A. and K. Hill (1997) ‘The Effects of Economic Changes on Mortality by Age and Cause: Latin America, 1950–1990’. In G. Tapinos, A. Mason and J. Bravo (eds), Demographic Responses to Economic Adjustment in Latin America, Oxford: Clarendon.

Childhood Mortality–Economic Growth: India 173 Paxson, C. and N. Schady (2005) ‘Child Health and Economic Crisis in Peru’. World Bank Economic Review, 19(2):203–23. Pritchett, L. and L. H. Summers (1996) ‘Wealthier Is Healthier’. Journal of Human Resources, 31(4):841–68. Ravallion, M. (2004) ‘Pro-Poor Growth: A Primer’. Mimeo, World Bank Development Research Group. Washington DC. Ravallion, M. and G. Datt (2002) ‘Why Has Economic Growth Been More ProPoor in Some States of India than Others?’ Journal of Development Economics, 68(2):381–400. Rios, N. and J. Carvalho (1997) ‘Demographic Consequences of Structural Adjustment: The Case of Brazil’. In G. Tapinos, A. Mason and J. Bravo (eds), Demographic Responses to Economic Adjustment in Latin America, Oxford: Clarendon. Ruhm, C. J. (2000) ‘Are Recessions Good for Your Health?’, Quarterly Journal of Economics, 115(2):617–50. Subbarao, K. and L. Raney (1995) ‘Social Gains from Female Education: A CrossNational Study’. Economic Development and Cultural Change, 44(1):105–28. Tandon, A. (2005) ‘Attaining Millennium Development Goals in Health: Isn’t Economic Growth Enough?’ ERD Policy Brief, Series 35. Manila: Asian Development Bank. van den Berg, G., M. Lindeboom and F. Portrait (2006) ‘Economic Conditions Early in Life and Individual Mortality’. American Economic Review, 96(1):290–302. Victora, C. G. et al. (2003) ‘Applying an Equity Lens to Child Health and Mortality: More of the Same Is Not Enough, Child Survival IV’. Lancet, 362:233–41. Virmani, A. (2004) ‘India’s Economic Growth: From Socialist Rate of Growth to Bharatiya Rate of Growth’. Indian Council for Research on International Economic Relations Working Paper 122. New Delhi. Wade, R. (2002) ‘Are Global Poverty and Inequality Getting Worse?’, Prospect Magazine (72):16–21. WHO (World Health Organization) (1992) Global Health Situation and Projections: Estimates. Division of Epidemiological Surveillance and Health Situation and Trend Assessment. Geneva. Wolpin, K. (1997) ‘Determinants and Consequences of the Mortality and Health of Infants and Children’. In M. Rosenzweig and O. Stark (eds), Handbook of Population and Family Economics, vol. 1A. Amsterdam: North Holland. World Bank (2004) Attaining the Millennium Development Goals in India: How Likely and What Will It Take to Reduce Infant Mortality, Child Malnutrition, Gender Disparities and Hunger-Poverty and to Increase School Enrolment and Completion? Washington, DC.

7 Achieving the Millennium Development Goal for Primary Schooling in India Sonia Bhalotra and Bernarda Zamora

Introduction Education is now widely valued not only for its intrinsic value in enriching the lives of individuals but also for its functional value in the development of the human capital of a nation. Educational investments in children have been shown to have high private and social returns. The private returns are associated with increased productivity and earnings in adulthood, and with further non-pecuniary gains arising from the greater efficiency with which educated individuals are able to acquire and process information (see for example Rosenzweig 1995). The social premium of education over and above the private value includes further productivity increases arising from knowledge spillovers, gains in health for one generation that flow from gains in education of the previous, and the improved functioning of civic society and democracy. These examples illustrate that widespread education not only helps growth through productivity effects, but is also crucial to distribution of the gains from growth. Growth in a society in which most people have a basic education is likely to be more pro-poor than growth in a society in which the educated are the elite few. Also, there is widespread evidence of an inter-generational correlation in educational attainment (see for example Becker and Tomes 1986), at least some of which is thought to be causal (for example, Chevalier 2004; Lleras-Muney 2005). To the extent that the impact of parental education on child education is causal, there are significant knock-on effects of public investment in education. In other words, their payoff to policy goes up immediately because investments in education at any one time have a multiplier effect, yielding additional benefits in the future. In summary, education is a powerful tool for reducing poverty, unemployment and inequality, improving 174

Primary Schooling in India 175

health and nutrition, and promoting sustained human-development-led growth (World Bank 2004:69). One of the Millennium Development Goals (MDGs) agreed in September 2000 at a UN summit of world leaders is the achievement of universal primary school attendance for boys and girls. This, of course, implies a complete closing of the gender gap. It also requires a 100 per cent primary school completion rate, that is, that all students entering grade 1 are retained until grade 5. The MDG couched in these terms reflects recognition of the importance of basic (primary) education. This is particularly pertinent in India where primary education has historically been neglected by the state, with educational expenditures being concentrated on the tertiary sector (see for example Drèze and Sen 1995). As a result, there are vast inequalities in educational attainment in India, a remarkable degree of illiteracy coexisting with frontier research in science and technology. India is also marked for being one of the group of countries in South Asia and Northern Africa where outcomes tend systematically to be better for boys than for girls, suggesting gender discrimination or at least undesirable gender differentiation. A further reason why India offers an interesting case study is that it exhibits striking diversity in educational indicators across its states that, in further work, we will exploit to consider more carefully the sorts of policy interventions likely to be effective.1 With India being such a large country, sample sizes available for statistical analysis are large, allowing more general pursuit of heterogeneity in the data, for example by religion (Muslims have lower educational attainment than Hindus) or by caste (scheduled castes and tribes exhibit lower educational attainment than the higher castes). These social divisions in education are analysed in Bhalotra and Zamora (2008), and the intergeneration transmission of education by social group is documented in Bhalotra et al. (2008). The National Family Health Survey of India (NFHS) data we describe below show that, in India in 1998/9, the school attendance rate was 82.5 per cent and the primary school completion rate was 61.7 per cent. We argue here that it is challenging, a priori, to expect both of these rates to rise to 100 per cent by 2015.

Data and definitions The data used in the analysis are from the first two rounds of the NFHS, conducted in 1992/3 and in 1998/9, respectively. Although this survey was concerned primarily with reproductive and child health, the household questionnaire of the survey contains information on schooling for

176 Sonia Bhalotra and Bernarda Zamora

every individual in the surveyed household. The survey covered the 26 main states of India, interviewing 88,563 households in 1992 and 92,486 in 1998. In 1992/3, 69 per cent and, in 1998/9, 66 per cent of households reported living in rural areas. For rural households, we have merged information on relevant infrastructure indicators that is available from a village questionnaire. In 1992, 485 villages were surveyed and, in 1998, 622. An advantage of household survey data over administrative data is that the latter often exaggerate school enrolment, possibly because this reflects well on school administrators and district officials, and because public expenditure allocations to schools and districts are often based upon the number of enrolled students (for example, World Bank 2004). As education is on the concurrent list of the constitution, it is partly a state subject. As a result, there are some differences in school structure and in definitions of progression across states. We will not concern ourselves with these here as we are interested in applying a uniform scale across all states, with a view to assessing the likelihood of India as a whole attaining the MDG for education. In this chapter, we look at two indicators, primary school attendance and the primary school completion rate. For each of these, the analysis is conducted both for all children and for boys and girls separately. Primary school age is defined as 6–11 years. This corresponds to grades 1–5, and is sometimes referred to as lower primary.2 The primary school attendance or enrolment rate is the ratio of the number of children aged 6–11 attending school to the total number of children aged 6–11. We are further interested in the completion rate since, in many developing countries, including India, it is common that children enrol in school but then fail to progress, or drop out. This may reflect the quality of schools or the volatility of parental incomes, children being taken out of school in response to unanticipated income shocks (see for example Jacoby and Skoufias 1997; Bhalotra and Deolalikar 2008). The completion rate is defined as the ratio of the number of children aged 12 at the time of the survey who report having completed primary school to the number of children aged 12 who report having enrolled in primary school. Ideally, we would use longitudinal data that allow us to follow a child through school, to completion. In the absence of such data, retrospective information such as available in the NFHS for level of schooling, completed at the time of the interview, is a second-best alternative. There are approximately 70,000 children aged 6–11 in each year and approximately 11,000 aged 12 (exact sample sizes are in the tables). Construction of the estimation samples is described in Table 7.1. Comparing (weighted) averages from the two rounds of the survey shows that

Primary Schooling in India 177 Table 7.1

Selection of the samples for analysis NFHS92/93 # obs

Sample size and means for school attendance Children aged 6–11 in the de facto population* Number of observations dropped on account of missing data in (1) whether attending school (2) years of education of most-educated adult (3) distance to nearest town (4) distance to pucca road Full attendance sample Boys attending school Boys not attending school Girls attending school Girls not attending school Sample size and means for primary school completion Children aged 12 in the de facto population Number of observations dropped on account of missing data in (1) whether attending school (2) years of education of most-educated adult (3) distance to nearest town (4) distance to pucca road Full primary completion sample Boys who have completed primary Boys who have not completed primary Girls who have completed primary Girls who have not completed primary

%

74,510 100

NFHS98/99 # obs

%

71,479 100

367 0.49 283 0.40 69 0.09 31 0.04 166 0.22 236 0.33 1,148 1.54 658 0.92 72,841 100 70,392 100 30,322 41.63 31,958 45.40 7,613 10.45 4,549 6.46 23,876 32.78 27,335 38.83 11,030 15.14 6,550 9.31

14,204 100

14,086 100

3,212 22.61 2,108 14.97 21 0.15 12 0.09 36 0.25 42 0.30 207 1.46 130 0.92 10,834 100 11,659 100 3,984 36.77 4,050 34.74 2,191 20.22 2,473 21.21 3,118 28.78 3,270 28.05 1,541 14.22 1,946 16.69

Note: ∗ De facto population estimated as population who slept the night previous to the survey in the household.

attendance among 6–11-year-olds increased from 69.5 per cent in 1992/3 to 82.5 per cent in 1998/9, and that growth in attendance was more rapid for girls than for boys (see the bottom of Table 7.4). In contrast, the primary school completion rate declined, from 65.3 per cent to 61.7 per cent. The decline was larger for girls than for boys, suggesting that the gender gap in completion widened even as the gap in attendance shrank (see the bottom of Table 7.5). Means and standard deviations of all microdata variables used in the analysis are in Table 7.2, where we also present t-tests of the significance of the difference of the means in the two years (this is defined

178 Table 7.2 Microdata sample statistics (weighted by all-India sample weight) Attendance sample Mean (92/93) Age 7 0.167 Age 8 0.189 Age 9 0.144 Age 10 0.197 Age 11 0.126 Female 0.479 Pucca house 0.212 Own flush toilet 0.141 Electricity 0.476 Potable water into the house 0.392 Separate room for cooking 0.530 Landowner 0.558 Livestock owner 0.603 Durables index −0.332 Rural∗ pucca house 0.080 Rural∗ own flush toilet 0.038 0.276 Rural∗ electricity Rural∗ potable water into the house 0.239 Rural∗ separate room for cooking 0.378 Rural∗ land owner 0.510 0.557 Rural∗ livestock owner −0.256 Rural∗ durables index School years of highest5.948 education adult Highest-educated adult is female 0.12048 Household size 7.639 Proportion of females under 5 0.066 Proportion of males under 5 0.073 Proportion of females aged 6–16 0.194 Proportion of males aged 6–16 0.208 Proportion of females aged 17–30 0.106 Proportion of males aged 17–30 0.070 Proportion of females aged 50+ 0.040 0.037 Proportion of males aged 50+ Household head female 0.063 Child of head 0.746 0.260 Principal female working1 Rural resident 0.754 Scheduled caste 0.128 Scheduled tribe 0.094 Muslim 0.148 Christian 0.020 Other religion 0.032 Rural∗ distance to nearest town 13.803 Rural∗ distance pucca road 1.863

Mean (98/99)

Completion sample t-ratio2 Mean Mean t-ratio2 (92/93) (98/99)

0.163 0.191 0.141 0.201 0.127 0.482 0.277 0.153 0.551 0.385 0.468 0.536 0.546 −0.321 0.128 0.049 0.338 0.238 0.327 0.488 0.506 −0.431 6.316

1.63 – – −0.77 – – 0.69 – – −1.16 – – 0,000 – – −0.78 0.410 0.438 −23.8∗∗ 0.270 0.315 −4.73∗∗ 0.192 0.178 −19.5∗∗ 0.573 0.613 4.64∗∗ 0.455 0.400 24.1∗∗ 0.609 0.526 −0.12 0.567 0.550 0.578 0.555 11.8∗∗ 6.4∗∗ 0.084 −0.069 −25.5∗∗ 0.094 0.143 −7.2∗∗ 0.052 0.057 −16.7∗∗ 0.320 0.369 0.14 0.260 0.237 16.1∗∗ 0.409 0.358 1.92 0.508 0.494 9.7∗∗ 0.525 0.509 31.8∗∗ −0.104 −0.276 −13.8∗∗ 7.369 6.983

0.446 7.514 0.062 0.068 0.199 0.210 0.110 0.065 0.039 0.034 0.072 0.742 0.314 0.765 0.193 0.096 0.157 0.020 0.029 11.183 3.368

3.4∗∗ 6.4∗∗ 4.8∗∗ 4.0∗∗ −5.3∗∗ 2.19∗ −4.5∗∗ 8.6∗∗ −0.24 4.1∗∗ −0.15 5.6∗∗ −13.9∗∗ −5.7∗∗ −28.6∗∗ −5.8∗∗ −12.4∗∗ 3.5∗∗ 3.5∗∗ 12.7∗∗ −45.7∗∗

– – – – – −2.19∗ −6.8∗∗ 1.26 −2.53∗ 7.0∗∗ 12.9∗∗ 0.34 2.12∗ 8.4∗∗ −8.7∗∗ 0.09 −2.9∗∗ 3.3∗∗ 7.6∗∗ 0.87 0.92 13.7∗∗ 2.51∗

0.371 0.398 −1.62 7.495 7.260 4.9∗∗ 0.042 0.037 2.7∗∗ 0.044 0.041 1.36 0.193 0.201 −2.30∗ 0.233 0.236 0.16 0.085 0.088 −1.64 0.075 0.068 5.5∗∗ 0.040 0.039 −0.61 0.041 0.038 2.6∗∗ 0.079 0.080 2.22∗ 0.783 0.789 1.1 0.254 0.300 −4.1∗∗ 0.707 0.736 −3.3∗∗ 0.114 0.184 −11.6∗∗ 0.072 0.084 −2.9∗∗ 0.128 0.146 −7.0∗∗ 0.026 0.025 3.6∗∗ 0.038 0.033 2.20∗ 12.541 10.571 2.55∗ 1.520 3.194 −20.8∗∗ (Continued)

Primary Schooling in India 179 Table 7.2

(Continued) Attendance sample

Completion sample

Mean Mean t-ratio2 Mean Mean (92/93) (98/99) (92/93) (98/99) Distance to pucca road∗ primary school Distance to pucca road∗ girl Rural∗ village electrified Rural∗ primary school in village Rural∗ middle school in village Rural∗ secondary school in village Rural∗ bank in village Rural∗ post office in village Rural∗ no. of TV sets per 1000 hab. Rural∗ missing number of TV sets

1.665 0.896 0.556 0.666 0.379 0.207 0.179 0.344 5.230 0.007

t-ratio2

3.012 −44.0∗∗

1.392

2.846

−19.3∗∗

−31.4∗∗ −20.8∗∗ −12.9∗∗ −9.0∗∗ −1.92 7.7∗∗ −3.7∗∗ −10.1∗∗ −18.9∗∗

0.533 0.550 0.635 0.386 0.213 0.188 0.347 5.724 0.007

1.385 0.603 0.665 0.396 0.221 0.171 0.350 18.453 0.030

−14.9∗∗ −8.7∗∗ −6.5∗∗ −4.0∗∗ −0.5 3.8∗∗ −0.42 −3.6∗∗ −7.6∗∗

1.641 0.608 0.685 0.381 0.214 0.163 0.354 16.956 0.031

Notes: 1. Female head or spouse of head. 2. t -statistic for test of the null hypothesis mean(92/93) − mean(98/99) = 0. ∗ Significant at 5 per cent. ∗∗ Significant at 1 per cent. 92/93 refers to NFHS1 conducted in 1992/93 and 98/99 refers to NFHS2 conducted in 1998/99.

in the notes to the table). Summary statistics for the state-level data used in the analysis are in Table 7.3.

Related literature and contributions Closely related to the current study is a recent analysis of education and health conducted for the World Bank by Anil Deolalikar (World Bank 2004; also see Deolalikar 2005). Motivated in the same spirit as the current analysis, to assess the likelihood of India attaining the MDG in education, this study provides a comprehensive analysis of primary schooling is India. It uses the 55th round of the national sample survey (NSS), conducted in 1999/2000. Multivariate probits are estimated for primary school attendance, school attendance and primary completion rates. The study finds that the largest marginal effects are associated with household living standards, access to electricity and expenditure on elementary schooling. The parameter estimates obtained for 1999/2000 are used to simulate indicators of school achievement in 2015 under three alternative scenarios. All of these involve an assumed change in each of nine predictor variables that were significant in the estimated model. Consider the attendance equation. Here, the explanatory variables are

180

Table 7.3

Annual growth rates of state level variables (per cent p.a.), 1982–99

State Andhra Pradesh Assam Bihar Gujarat Haryana Jammu Karnataka Kerala Madhya Pradesh Maharashtra Orissa Punjab Rajasthan Tamil Nadu Uttar Pradesh West Bengal All India

Real Devexp/ Eduexp/ Rural income p.c. income income poverty 5.00 3.12 4.10 6.39 − 1.59 5.39 5.63 4.77 5.14 5.30 5.12 5.72 6.09 3.80 4.71 4.79

1.43 3.05 2.68 1.66 0.54 7.66 0.80 0.93 1.82 −0.08 1.88 3.03 2.29 0.37 1.01 1.63 1.92

0.98 5.66 4.18 2.45 2.29 6.16 1.40 −0.15 2.79 2.69 4.04 1.43 2.98 1.25 3.02 3.28 2.78

−12.1 −8.63 −2.87 −6.91 6.23 34.5 −10.5 −7.99 −6.29 −11.7 −1.93 3.41 −4.45 −10.1 −5.12 −5.71 −3.13

Urban poverty −8.19 16.8 −10.7 −7.46 −12.5 21.9 −11.7 −11.2 −0.79 −1.29 −8.31 −15.4 −2.27 −4.72 −12.1 −14.5 −5.14

Rural/ Female F/M F/M teachers urban cons. illiteracy rate illiteracy primary 1.90 −0.85 −1.68 0.23 −4.70 −9.22 0.99 −2.10 −0.57 1.32 6.61 −4.02 −1.34 5.64 −1.96 0.45 −0.58

−0.83 −1.28 −0.64 −1.40 −1.36 −0.87 −1.39 −6.65 −0.87 −1.84 −1.19 −1.77 −0.58 −1.57 −0.86 −1.64 −1.55

0.31 −2.71 0.41 0.60 1.18 1.16 0.40 1.12 0.91 0.58 0.32 0.69 1.35 −0.28 0.74 0.36 0.45

6.22 64.5 0.11 1.52 12.5 −0.12 7.82 3.51 3.32 2.49 −1.21 −0.79 72.6 0.68 1.31 115.5 18.1

Notes: These are the growth rates of the variables that are significant in the regression shown in Table 7.6. In Table 7.6, we use the average of the variable over the 5 years preceding the date of the survey. To describe the overall trend in these variables, this table presents data for 1982–99, that is, starting 10 years before the date of the first survey (1992/93). Income is state net domestic product, p.c. is per capita, devexp is development expenditure, eduexp is education expenditure, cons. is consumption, F and M are female and male. Rural and urban poverty are measured by the poverty gap index. Rural/urban cons is the ratio of the mean per capita consumption in rural and urban areas.

Primary Schooling in India 181

the level of education attained by adult men and women, household consumption, annual public expenditure on elementary education per 6–14-year-old child, and the following indicators of district-level conditions: village access to pucca roads and electricity, the number of primary schools per thousand children, the pupil–teacher ratio at the primary level, and crimes against (kidnappings of) women and girls. Since the education deficit is concentrated in the poor states,1 the simulations conducted in this study group the Indian states as poor and non-poor. In the first scenario, the specified characteristics in the poor states are brought up to the national average. In the second scenario, they are brought up to the average for the non-poor states. In the third scenario, they are increased at a specified rate per annum between 1999/2000 and 2015. The specified rate is set for each of the nine variables in an ad-hoc way for illustrative purposes. The predicted outcomes for 2015 get progressively more encouraging as one moves from the first scenario to the third. The overall conclusion in this study is that attaining the MDG for education is extremely unlikely in the poor states and, as a result, in India as a whole. The current study employs definitions of school outcomes similar to those used by World Bank (2004) and it estimates multivariate probits that are similarly specified. However, we use the NFHS data rather than the NSS, which is useful in that it provides an opportunity to crosscheck the results of one study against the other. A contribution of the current study is that it uses repeated cross-sectional data (two rounds of the NFHS) to investigate the growth in schooling indictors. It then assesses the extent to which (i) the predictor variables actually change and (ii) the parameters are stable over time. We find that the predictor variables change much less than is hypothesized in the third scenario of the simulations conducted in World Bank (2004).3 For example, in scenario 3, the assumed annual change in male and female years of schooling is 0.25 and 0.3, respectively. Between 1992/3 and 1998/9, these variables increased (for the sample of 6–11-year-olds) by only 0.055 and 0.066 years per year respectively. We also find that the parameters are not stable over time, which makes it very difficult to extrapolate to the future. Indeed, we find that almost all of the growth in schooling can be attributed to changes in the elasticities. We caution against the common practice of making predictions on the assumption of stable parameters, while recognizing that there may be no better alternative. We conclude that the prospect of India attaining universal primary attendance is good, but that the prospect of attaining universal completion rates in primary school is bleak unless a major intervention is undertaken.

182 Sonia Bhalotra and Bernarda Zamora

Analytical approach Educational enrolment depends upon supply and demand factors. In a competitive markets framework, education is an investment in human capital and the extent of this investment will depend only upon its relative rate of return. When credit markets are imperfect, or when parents value the education of their children as a consumption or a status good (see for example Banerjee 2004) then parental wealth also affects the level of education demanded. A role for religion, gender and ethnicity in influencing the demand for education may be argued to arise from differences in tastes, opportunity costs (wages) or perceived returns along these dimensions. Empirically, the demand for education can be modelled like the demand for any other good, as a function of total resources (parental wealth), relative prices (rates of return), demographics and taste-shifters. Supply variables like access to school are included in the model to allow for disequilibria: not everyone who demands education can have it; see Ham (1986) for similar reasoning for the inclusion of the regional unemployment rate in models of labour demand. Our measure of access to school is an indicator of whether or not there is a school in the village – this is a discretization of distance to school, which may be thought of as positively related to the price of attending school. Unfortunately, we have no measures of school quality. The estimated equations are similar to those in numerous previous studies of educational enrolment and progression (see for example Behrman and Knowles 1999). Our approach to developing projections to 2015 is as follows. We estimate equations for selected educational indicators for each of the years 1992/3 and 1998/9 for all children and also separately for boys and girls. We analyse changes in schooling outcomes between 1992/3 and 1998/9, decomposing them into changes in characteristics (regressors, X) and changes in model parameters (β). We then assume that the contribution of characteristics over time is the same between 1998/9 and 2015 as it was between 1992/3 and 1998/9.

Empirical model As both attendance and completion of primary school are binary variables (0/1), they are modelled as probits and the parameters are estimated by maximum likelihood. The estimated model can be written as Sis = αs + β Xis + εis

(7.1)

Primary Schooling in India 183

where S is an indicator of the school outcome for index child i in state s and X is a vector of child-, household- and, in the case of rural children, village-level covariates. εis captures residual variation in schooling outcomes. Time-invariant (or sluggishly evolving) state-level unobservables are captured by the fixed effects, αs . These will include political-economic variables, historically determined attitudes to education and initial conditions. The model is estimated for each of the two rounds of the survey, so there is an implicit t-subscript everywhere, where t = 1992/3, 1998/9. The socioeconomic status of the household is captured by wealth indicators, adult education and demographics. For rural areas, we include indicators of the supply of schooling at the village level. Since no similar information is available for urban regions, these variables appear in interaction with a dummy for whether the household lives in a rural area. If a variable has a sufficiently large number of missing values, then, rather than discard all observations with any missing data, we create a dummy to indicate missing values and include this in the model as an additional regressor; this is the case for caste and religion. Estimates of marginal effects for the attendance equations are in Table 7.4 and, for the completion equations, in Table 7.5. Estimates are presented by year and gender. The independent variables in the attendance and completion models are the same, with one exception. Since the attendance equation is for 6–11-year-olds, it includes a set of age dummies and, in the sample that pools genders, this is interacted with a dummy for whether the child is a girl (1) or a boy (0). The completion equation, which is for 12-year-olds, simply includes a gender dummy. To take account of the survey design, the regressions are weighted using sample weights available in the data file. Reported standard errors are robust to arbitrary forms of heteroscedasticity, which is likely given clustering in the sample design. Tests of the joint significance of subsets of variables (for example, village infrastructure, state fixed effects) are reported in the tables. The contribution of the state fixed effects to the total explained variation as measured by the pseudo-R2 is also reported in the tables. As this is large, the explanatory power of some policyamenable variables at the state level is examined as explained in the next section. Explaining the state fixed effects In a second stage of the analysis, we investigate which state-level variables might explain the state fixed effects in the estimated microlevel equations. The state dummy coefficients are saved for each year and then pooled to generate a panel. The panel of 26 state fixed effects coefficients

184 Sonia Bhalotra and Bernarda Zamora

for each of two years is merged with a panel of data on state-level income, inequality, state education expenditure and other relevant predictors. As time series data on these predictor variables is readily available only for the 15 major states, the panel is reduced to N = 15, T = 2.4 Since there is no time variation in the data that the probits are estimated upon, the fixed-effects estimates from these data are regressed upon averages of the predictors over the five years preceding the data of each survey (that is, 1987/8–1992/3 for NFHS1 and 1993/4–1998/9 for NFHS2). The estimated model can be written as αs = ϕ Zs + us

(7.2)

where αs are estimates of the state effects derived from equation (7.1) and Z are time-averaged variables such as state income, education expenditure and female illiteracy, and us is a residual.

Results Probit estimates of models for school attendance and completion are reported in Tables 7.4 and 7.5 respectively. Significant predictors of attendance rates include the presence of primary and middle schools in the village (in the case of rural India) and, at least in the first survey-year, the prevalence of television sets. Wealth and living conditions at the household level are also relevant, significant indicators being an index of household possessions, whether the household has access to electricity, and whether it has a separate room for cooking. The higher the educational level of the most educated adult in the household, the greater the likelihood that a child in that household is attending school. It does not seem to matter to attendance rates whether this person is a man or a woman. However, where the head of the household is a woman, children are more likely to be in school. Children are less likely to attend school when the principal female (head or head’s wife) in the household is working. As the proportion of women in work is expected to rise, this factor will constrain increases in attendance unless the parameters of the model change to nullify this effect (the latter is a real possibility since increases in women’s labour force participation have, historically, been associated with wider structural change in the organization of both markets and households). Children in larger households are less likely to attend school and, for a given household size, children of primary school age are less likely to attend if children under the age of five are present in the household. These results suggest that, if India experiences further reductions in

185 Table 7.4 Probit estimates of school attendance among 6–11-year-old children Marginal effects (regression weighted all India sample weight) NFHS92/93

NFHS98/99

All

Boys

Girls

All

Boys

0.1208 (15.70)∗∗ 0.1626 (23.23)∗∗ 0.1640 (21.03)∗∗ 0.1575 (18.42)∗∗ 0.1519 (15.19)∗∗ −0.1042 (7.72)∗∗ −0.1589 (11.11)∗∗ −0.2176 (16.29)∗∗ −0.2108 (13.52)∗∗ −0.2329 (17.10)∗∗ −0.2684 (15.93)∗∗

0.0939 (15.90)∗∗ 0.1271 (23.25)∗∗ 0.1279 (21.13)∗∗ 0.1236 (18.12)∗∗ 0.1212 (15.35)∗∗ – – – – – – – – – – – –

0.1086 (10.70)∗∗ 0.1207 (12.22)∗∗ 0.1297 (11.90)∗∗ 0.0942 (8.39)∗∗ 0.0620 (4.60)∗∗ – – – – – – – – – – – –

0.0640 (13.70)∗∗ 0.0734 (16.01)∗∗ 0.0758 (14.25)∗∗ 0.0621 (10.28)∗∗ 0.0544 (6.98)∗∗ –0.0426 (3.98)∗∗ −0.0689 (5.61)∗∗ −0.0868 (7.50)∗∗ −0.0895 (6.30)∗∗ −0.1101 (9.74)∗∗ −0.1219 (7.89)∗∗

0.0523 (13.77)∗∗ 0.0601 (15.91)∗∗ 0.0631 (14.64)∗∗ 0.0524 (10.33)∗∗ 0.0473 (7.45)∗∗ – – – – – – – – – – – –

0.0614 (10.24)∗∗ 0.0602 (10.09)∗∗ 0.0637 (9.27)∗∗ 0.0251 (3.32)∗∗ 0.0055 (0.57) – – – – – – – – – – – –

0.0365 (2.34)∗ Own flush toilet 0.0116 (0.65) Electricity 0.0704 (4.04)∗∗ Potable water into 0.0118 the house (0.72) Separate room for 0.0356 cooking (2.32)∗ Landowner −0.0168 (0.88) −0.0009 Livestock owner (0.05) Durables index 0.0401 (9.56)∗∗ Rural∗ pucca house 0.0049 (0.23) Rural∗ own flush toilet 0.0228 (0.75)

0.0214 (1.06) 0.0287 (1.31) 0.0557 (2.51)∗ 0.0004 (0.02) 0.0328 (1.68) 0.0141 (0.55) 0.0107 (0.45) 0.0295 (6.33)∗∗ −0.0012 (0.04) 0.0002 (0.00)

0.0607 (2.62)∗∗ −0.0192 (0.68) 0.0746 (2.94)∗∗ 0.0319 (1.34) 0.0394 (1.81) −0.0626 (2.29)∗ −0.0137 (0.54) 0.0502 (7.63)∗∗ 0.0095 (0.32) 0.0423 (0.97)

0.0048 (0.44) 0.0112 (0.96) 0.0437 (3.05)∗∗ 0.0114 (0.98) 0.0298 (2.65)∗∗ 0.0215 (1.31) 0.0127 (0.75) 0.0204 (7.12)∗∗ 0.0153 (0.97) 0.0061 (0.27)

−0.0004 (0.03) 0.0176 (1.19) 0.0304 (1.52) 0.0061 (0.40) 0.0208 (1.41) 0.0296 (1.31) 0.0150 (0.63) 0.0170 (5.08)∗∗ 0.0010 (0.04) −0.0080 (0.24)

0.0118 (0.74) 0.0013 (0.07) 0.0592 (2.79)∗∗ 0.0188 (1.09) 0.0408 (2.38)∗ 0.0135 (0.57) 0.0099 (0.42) 0.0239 (5.55)∗∗ 0.0307 (1.39) 0.0204 (0.64)

Child age and gender Age 7 Age 8 Age 9 Age 10 Age 11 Female∗ age 6 Female∗ age 7 Female∗ age 8 Female∗ age 9 Female∗ age 10 Female∗ age 11 Assets Pucca house

Girls

(Continued)

186 Table 7.4 (Continued) Marginal effects (regression weighted all India sample weight) NFHS92/93

Rural∗ electricity Rural∗ potable water into the house Rural∗ separate room for cooking Rural∗ landowner Rural∗ livestock owner Rural∗ durables index Education and demographics Schooling years of most-educated adult Most-educated adult is female log Household size Proportion of females under 5 Proportion of males under 5 Proportion of females aged 6–16 Proportion of males aged 6–16 Proportion of females aged 17–30 Proportion of males aged 17–30 Proportion of females aged 50+ Proportion of males aged 50+ Household head female Child of head Principal female working Scheduled caste Scheduled tribe

NFHS98/99

All

Boys

Girls

All

Boys

Girls

−0.0137 (0.69) 0.0159 (0.90) 0.0079 (0.47) 0.0585 (2.90)∗∗ −0.0098 (0.49) 0.0080 (1.11)

−0.0186 (0.70) 0.0243 (1.05) 0.0033 (0.15) 0.0332 (1.19) −0.0144 (0.55) 0.0101 (1.24)

0.0092 (0.34) 0.0013 (0.05) 0.0118 (0.51) 0.0932 (3.22)∗∗ −0.0074 (0.26) 0.0084 (0.75)

−0.0250 (1.30) 0.0015 (0.10) −0.0124 (0.88) −0.0025 (0.14) −0.0077 (0.45) 0.0114 (3.05)∗∗

−0.0198 (0.74) 0.0067 (0.35) −0.0041 (0.22) −0.0006 (0.02) −0.0144 (0.58) 0.0103 (2.37)∗

−0.0294 (1.07) −0.0062 (0.29) −0.0228 (1.10) −0.0065 (0.26) 0.0007 (0.03) 0.0133 (2.36)∗

0.0260 (29.17)∗∗ 0.0019 (0.20) −0.0465 (8.08)∗∗ −0.2296 (6.64)∗∗ −0.2236 (6.42)∗∗ −0.0750 (2.23)∗ −0.1324 (3.90)∗∗ 0.1076 (3.67)∗∗ −0.2666 (8.61)∗∗ 0.0793 (1.86) 0.0191 (0.42) 0.0416 (3.20)∗∗ 0.0306 (3.42)∗∗ −0.0489 (6.78)∗∗ −0.0247 (2.81)∗∗ −0.0814

0.0189 (19.40)∗∗ −0.0094 (0.75) −0.0400 (6.04)∗∗ −0.1424 (3.64)∗∗ −0.1328 (3.38)∗∗ −0.0164 (0.45) −0.1458 (3.78)∗∗ 0.0877 (2.71)∗∗ −0.2100 (6.16)∗∗ 0.0302 (0.64) 0.0165 (0.33) 0.0284 (1.69) 0.0269 (2.22)∗ −0.0504 (5.49)∗∗ −0.0130 (1.10) −0.0769

0.0344 (24.57)∗∗ 0.0139 (1.11) −0.0633 (7.07)∗∗ −0.3435 (6.21)∗∗ −0.3421 (6.16)∗∗ −0.1773 (3.15)∗∗ −0.0944 (1.76) 0.1288 (2.74)∗∗ −0.3428 (6.82)∗∗ 0.1587 (2.30)∗ 0.0189 (0.26) 0.0624 (3.34)∗∗ 0.0354 (2.77)∗∗ −0.0437 (4.34)∗∗ −0.0409 (3.25)∗∗ −0.0835

0.0138 (24.27)∗∗ 0.0116 (1.40) −0.0403 (10.47)∗∗ −0.1850 (8.16)∗∗ −0.1842 (7.98)∗∗ −0.0728 (3.29)∗∗ −0.1385 (6.18)∗∗ 0.1135 (5.85)∗∗ −0.1744 (8.36)∗∗ 0.0134 (0.45) −0.0372 (1.19) 0.0196 (2.69)∗∗ 0.0182 (2.99)∗∗ −0.0421 (32.85)∗∗ −0.0206 (3.10)∗∗ −0.0707

0.0105 (15.95)∗∗ 0.0040 (0.35) −0.0393 (8.75)∗∗ −0.1016 (3.78)∗∗ −0.1356 (4.99)∗∗ −0.0405 (1.64) −0.1318 (4.95)∗∗ 0.1010 (4.63)∗∗ −0.1165 (4.87)∗∗ 0.0057 (0.16) −0.0445 (1.24) 0.0047 (0.46) 0.0238 (3.31)∗∗ −0.0316 (10.47)∗∗ −0.0101 (1.05) −0.0636

0.0180 (20.76)∗∗ 0.0205 (1.72) −0.0450 (7.62)∗∗ −0.2856 (8.04)∗∗ −0.2413 (6.68)∗∗ −0.1153 (3.19)∗∗ −0.1471 (4.26)∗∗ 0.1275 (4.13)∗∗ −0.2522 (7.64)∗∗ 0.0140 (0.29) −0.0240 (0.48) 0.0414 (4.38)∗∗ 0.0126 (1.32) −0.0555 (16.68)∗∗ −0.0346 (3.78)∗∗ −0.0801

(Continued)

187 Table 7.4 (Continued) Marginal effects (regression weighted all India sample weight) NFHS92/93 All

NFHS98/99 Boys

Girls

All

Boys

(7.60)∗∗

(5.61)∗∗

(5.44)∗∗

(16.05)∗∗

(12.16)∗∗

−0.0936 (10.07)∗∗ −0.0170 (0.73) −0.0013 (0.07)

−0.0991 (8.64)∗∗ −0.0224 (0.76) −0.0025 (0.10)

−0.0815 (5.95)∗∗ −0.0130 (0.37) 0.0024 (0.09)

−0.0507 (11.92)∗∗ 0.0404 (3.04)∗∗ −0.0011 (0.07)

−0.0586 (8.19)∗∗ 0.0372 (2.11)∗ −0.0144 (0.63)

(8.85)∗∗ −0.0429 (4.72)∗∗ 0.0433 (2.07)∗ 0.0189 (0.93)

Rural∗ no. of TVs in village per 1000 habs Rural∗ missing number of TV sets

−0.0732 (3.49)∗∗ −0.0002 (1.27) −0.0013 (0.97) −0.0004 (0.29) −0.0008 (0.94) 0.0037 (0.43) 0.0314 (2.79)∗∗ 0.0204 (2.42)∗ 0.0053 (0.55) 0.0050 (0.51) −0.0147 (1.71) 0.0007 (3.06)∗∗ 0.0553 (1.73)

−0.0403 (1.46) −0.0001 (0.47) −0.0024 (1.79) 0.0004 (0.26) – – 0.0096 (0.84) 0.0373 (2.50)∗ 0.0045 (0.39) −0.0030 (0.23) 0.0025 (0.19) −0.0139 (1.20) 0.0002 (1.17) 0.0322 (0.69)

−0.1182 (3.99)∗∗ −0.0004 (1.49) −0.0007 (0.37) −0.0010 (0.50) – – 0.0014 (0.11) 0.0200 (1.21) 0.0374 (3.19)∗∗ 0.0158 (1.18) 0.0088 (0.61) −0.0111 (0.91) 0.0012 (3.13)∗∗ 0.1021 (2.50)∗

0.0140 (0.76) −0.0003 (2.49)∗ 0.0016 (2.41)∗ −0.0012 (1.73) −0.0003 (0.73) 0.0098 (1.63) 0.0244 (2.93)∗∗ −0.0010 (0.16) 0.0097 (1.40) −0.0010 (0.13) −0.0039 (0.62) 0.0000 (0.52) −0.0191 (1.16)

0.0193 (0.78) −0.0003 (2.21)∗ 0.0007 (1.04) −0.0006 (0.88) – – 0.0058 (0.70) 0.0209 (1.95) 0.0009 (0.10) 0.0040 (0.40) 0.0011 (0.11) −0.0034 (0.40) −0.0000 (0.83) −0.0008 (0.04)

0.0059 (0.22) −0.0003 (1.49) 0.0025 (2.47)∗ −0.0019 (1.80) – – 0.0132 (1.50) 0.0305 (2.47)∗ −0.0029 (0.32) 0.0167 (1.70) −0.0037 (0.31) −0.0044 (0.49) 0.0001 (1.36) −0.0366 (1.52)

State fixed effects

yes

yes

yes

yes

yes

yes

– – – – – – 0.7646

– – – – – – 0.6242

−0.0076 (1.07) −0.0232 (1.76) −0.0092 (0.15) 0.8252

0.0029 (0.31) −0.0210 (1.26) 0.0512 (1.05) 0.8603

−0.0213 (2.11)∗ −0.0238 (1.27) −0.0905 (0.80) 0.7872

Muslim Christian Other religion Rural infrastructure Rural resident Rural∗ distance

to nearest town Rural∗ distance pucca road Rural∗ distance to pucca road∗ primary school Rural∗ distance to pucca road∗ girl Rural∗ village electrified Rural∗ primary school in village Rural∗ middle school in village Rural∗ secondary school in village Rural∗ bank in village Rural∗ post office in village

Variables available only for 1998/9 Other ‘backward’ caste – – Missing caste – – Missing religion – – Mean of dependent variable 0.6971

Girls

(Continued )

188 Sonia Bhalotra and Bernarda Zamora Table 7.4 (Continued) Marginal effects (regression weighted all India sample weight) NFHS92/93

Observations Pseudo R2 Pseudo R2 due to state fixed effects log Pseudolikelihood Wald Chi2 F -age F -gender∗ age F -assets F -rural∗ assets F -village variables F -household demographics F -state fixed effects

NFHS98/99

All

Boys

Girls

All

Boys

Girls

72841 0.2787 0.0280

37935 0.2343 0.0195

34906 0.3079 0.0426

69456 0.2285 0.0201

36027 0.2071 0.0146

33429 0.2457 0.0298

−32221 8634 593∗∗ 1028∗∗ 309∗∗ 79∗ 71∗∗ 2108∗∗ 1571∗∗

−15850 3862 606∗∗ – 153∗∗ 43∗∗ 42∗∗ 1110∗∗ 622∗∗

−15991 5372 197∗∗ – 191∗∗ 68∗∗ 55∗∗ 1355∗∗ 1325∗∗

−24828 5634 290∗∗ 420∗∗ 188∗∗ 42∗∗ 31∗∗ 1595∗∗ 753∗∗

−11554 2641 306∗∗ – 91∗∗ 22∗ 15 827∗∗ 286∗∗

−13051 36429 186∗∗ – 122∗∗ 34∗∗ 28∗∗ 1051∗∗ 705∗∗

Notes: 1. Head or spouse of head. Interaction of rural residence with assets variables is included to capture the difference of the impact of assets variables in rural as opposed to urban households. Interaction of rural residence with infrastructure variables is necessary as there is no corresponding urban information. The F -tests at the end of the table are tests of joint significance of the named groups of variables. Absolute value of robust z-statistics in parentheses. ∗ significant at 5% level. ∗∗ significant at 1% level.

fertility, it will see further rises in school attendance. There are some significant compositional effects on attendance, indicating that scheduled castes, scheduled tribes and Muslims are less likely to have children attend school (see Bhalotra and Zamora 2008). Age dummies in the model are consistent with late entry, especially among boys. At every age, girls are less likely than boys to be attending school, the gender gap tending to increase with age. This indicates higher dropout rates among girls – consistent with their lower completion rates. The determinants of completion are different from the determinants of attendance. The presence of a middle or a secondary school in the village (in rural India) has a positive impact on completion probabilities, consistent with the notion that children will be less likely to complete primary school if there is nowhere to go after. Note that attendance was affected instead by the presence of primary and middle schools. Wealthier households, as indicated by the durables index (only), are more likely to have children complete primary school. The educational level of the

189 Table 7.5 Probit estimates of school completion among 12-year-old children Marginal effects (regression weighted by all India sample weights) NFHS92/93

NFHS98/99

All Gender Female

Boys

Girls

All

Boys

Girls

−0.0484 (2.98)∗∗

– –

– –

−0.0620 (4.19)∗∗

– –

– –

0.0373 (1.37) −0.0152 (0.54) 0.0827 (2.20)∗ 0.0190 (0.65) 0.0482 (1.76) 0.0125 (0.36) −0.0088 (0.27) 0.0523 (6.25)∗∗ 0.0013 (0.03) 0.0101 (0.22) −0.0153 (0.36) −0.0022 (0.06) −0.0221 (0.70) 0.0617 (1.57) −0.0212 (0.55) −0.0207 (1.59)

0.0501 (1.36) −0.0370 (0.92) 0.1080 (2.19)∗ 0.0346 (0.84) 0.0570 (1.48) 0.0025 (0.05) 0.0006 (0.01) 0.0431 (3.79)∗∗ −0.0056 (0.11) 0.0081 (0.12) −0.0458 (0.81) 0.0003 (0.01) −0.0437 (1.01) 0.0724 (1.36) −0.0004 (0.01) −0.0123 (0.70)

0.0128 (0.29) 0.0086 (0.21) 0.0475 (0.78) 0.0007 (0.02) 0.0438 (1.10) 0.0314 (0.63) −0.0250 (0.53) 0.0623 (5.07)∗∗ 0.0266 (0.47) 0.0047 (0.07) 0.0349 (0.52) −0.0153 (0.30) 0.0115 (0.24) 0.0411 (0.70) −0.0451 (0.78) −0.0298 (1.52)

0.0371 (1.43) 0.0842 (3.43)∗∗ 0.1113 (2.44)∗ 0.0324 (1.18) 0.0384 (1.48) −0.0430 (1.29) 0.0125 (0.39) 0.0190 (2.27)∗ −0.0206 (0.59) −0.0408 (0.96) −0.0769 (1.55) −0.0314 (1.00) −0.0199 (0.68) 0.0406 (1.13) 0.0046 (0.13) 0.0227 (1.96)

0.0654 (1.91) 0.0846 (2.46)∗ 0.0777 (1.36) 0.0374 (1.00) 0.0540 (1.53) −0.0301 (0.65) 0.0513 (1.17) 0.0083 (0.76) −0.0298 (0.64) −0.1000 (1.75) −0.0671 (1.07) −0.0393 (0.91) −0.0335 (0.83) 0.0421 (0.84) −0.0312 (0.63) 0.0362 (2.37)∗

0.0089 (0.21) 0.0825 (2.23)∗ 0.1594 (2.10)∗ 0.0185 (0.45) 0.0108 (0.28) −0.0707 (1.48) −0.0231 (0.48) 0.0343 (2.69)∗∗ −0.0200 (0.36) 0.0302 (0.49) −0.0961 (1.19) −0.0133 (0.28) 0.0037 (0.09) 0.0630 (1.22) 0.0380 (0.70) 0.0047 (0.27)

Education and demographics Schooling years of 0.0200 most-educated adult (10.33)∗∗ Most-educated adult 0.0676 is female (3.79)∗∗ log household size −0.0251 (1.72)

0.0186 (7.23)∗∗ 0.0634 (2.76)∗∗ −0.0131 (0.67)

0.0233 (8.02)∗∗ 0.0671 (2.33)∗ −0.0466 (2.08)∗

0.0239 (13.52)∗∗ 0.0538 (3.40)∗∗ −0.0315 (2.29)∗

0.0235 (9.81)∗∗ 0.0497 (2.34)∗ −0.0279 (1.48)

0.0253 (9.69)∗∗ 0.0600 (2.52)∗ −0.0495 (2.45)∗

Assets Pucca house Own flush toilet Electricity Potable water into the house Separate room for cooking Landowner Livestock owner Durables index Rural∗ pucca house Rural∗ own flush toilet Rural∗ electricity Rural∗ potable water into the house Rural∗ separate room for cooking Rural∗ landowner Rural∗ livestock owner Rural∗ durables index

(Continued)

190 Table 7.5 (Continued) Marginal effects (regression weighted by all India sample weights) NFHS92/ 93

Proportion of females under-5 Proportion of males under-5 Proportion of females aged 6−16 Proportion of males aged 6–16 Proportion of females aged 17–30 Proportion of males aged 17–30 Proportion of females aged 50+ Proportion of males aged 50+ Household head female Child of head Principal female working Scheduled caste Scheduled tribe Muslim Christian Other religion Rural infrastructure Rural resident Rural∗ distance to nearest town Rural∗ distance pucca road Rural∗ distance to pucca road∗ primary school Rural∗ distance to pucca road∗ girl Rural∗ village electrified

NFHS98/ 99

All

Boys

Girls

All

Boys

Girls

−0.1230 (1.26) −0.3802 (3.85)∗∗ −0.0499 (0.61) −0.1037 (1.27) −0.0730 (0.99) −0.3890 (4.84)∗∗ 0.0141 (0.14) 0.1561 (1.54) −0.0003 (0.01) −0.0167 (0.88) −0.0177 0.0057 (0.29) −0.0937 (3.34)∗∗ −0.0847 (3.89)∗∗ −0.0386 (0.85) −0.0098 (0.28)

−0.0646 (0.49) −0.3242 (2.52)∗ −0.0246 (0.23) −0.0623 (0.58) −0.0564 (0.59) −0.4216 (4.05)∗∗ −0.0148 (0.11) 0.1715 (1.30) 0.0005 (0.01) −0.0444 (1.78) −0.0409 0.0139 (0.54) −0.0621 (1.74) −0.1068 (3.70)∗∗ 0.0603 (0.99) −0.0500 (1.04)

−0.1974 (1.35) −0.4255 (2.70)∗∗ −0.1183 (0.94) −0.1218 (0.96) −0.0743 (0.66) −0.3795 (2.99)∗∗ 0.0556 (0.34) 0.1551 (0.98) 0.0068 (0.17) 0.0218 (0.73) 0.0163 −0.0097 (0.29) −0.1412 (3.09)∗∗ −0.0530 (1.53) −0.1471 (2.24)∗ 0.0303 (0.59)

−0.5574 (5.81)∗∗ −0.5408 (5.72)∗∗ −0.0825 (1.10) −0.3571 (4.60)∗∗ −0.1412 (2.07)∗ −0.5598 (7.13)∗∗ −0.1187 (1.19) 0.0105 (0.11) 0.0726 (3.43)∗∗ 0.0337 (1.96) −0.0536 −0.0087 (0.47) −0.0595 (2.44)∗ −0.1273 (6.06)∗∗ −0.0893 (1.72) −0.0866 (2.21)∗

−0.3909 (3.03)∗∗ −0.3345 (2.65)∗∗ 0.0690 (0.70) −0.2642 (2.58)∗∗ −0.0911 (1.02) −0.5110 (5.03)∗∗ 0.0543 (0.41) 0.1805 (1.42) 0.0421 (1.31) 0.0521 (2.16)∗ −0.0346 −0.0130 (0.52) −0.0905 (2.82)∗∗ −0.1627 (5.82)∗∗ −0.1805 (2.95)∗∗ −0.1043 (2.14)∗

−0.7775 (5.37)∗∗ −0.7959 (5.50)∗∗ −0.3000 (2.53)∗ −0.4625 (3.75)∗∗ −0.2175 (2.04)∗ −0.6471 (5.19)∗∗ −0.3655 (2.42)∗ −0.2043 (1.35) 0.1170 (5.71)∗∗ 0.0105 (0.44) −0.0776 −0.0003 (0.01) −0.0309 (0.86) −0.0926 (3.20)∗∗ 0.0276 (0.36) −0.0666 (1.06)

−0.0151 (0.30) −0.0006 (1.40) −0.0111 (2.06)∗ 0.0127 (2.32)∗ −0.0030 (0.95) 0.0120 (0.55)

−0.0183 (0.28) −0.0007 (1.37) −0.0140 (2.07)∗ 0.0144 (2.07)∗ – – 0.0069 (0.25)

−0.0316 (0.38) −0.0004 (0.53) −0.0075 (0.96) 0.0091 (1.11) – – 0.0253 (0.68)

0.1007 (1.76) −0.0004 (0.79) 0.0003 (0.12) 0.0005 (0.23) –0.0004 (0.32) 0.0199 (0.98)

0.0716 (0.96) 0.0006 (0.79) 0.0014 (0.52) −0.0014 (0.48) – – 0.0272 (1.00)

0.1019 (1.17) −0.0018 (2.23)∗ −0.0004 (0.12) 0.0019 (0.54) – – 0.0083 (0.27)

(Continued)

Primary Schooling in India 191 Table 7.5

(Continued) Marginal effects (regression weighted by all India sample weights) NFHS92/93

Rural∗ primary

school in village Rural∗ middle school in village Rural∗ secondary school in village Rural∗ bank in village Rural∗ post office in village Rural∗ no. of TVs in village per 1000 habitants Rural∗ missing no. of TV sets State fixed effects

NFHS98/99

All

Boys

Girls

All

Boys

0.0094 (0.32) 0.0014 (0.07) 0.0525 (2.58)∗∗ 0.0251 (1.16) −0.0442 (2.21)∗ 0.0011 (2.82)∗∗ −0.0799 (1.04) yes

0.0402 (1.07) 0.0131 (0.51) 0.0629 (2.30)∗ 0.0132 (0.44) −0.0611 (2.31)∗ 0.0009 (1.83) −0.0488 (0.49) yes

−0.0404 (0.80) −0.0084 (0.27) 0.0315 (0.97) 0.0463 (1.45) −0.0179 (0.57) 0.0016 (2.18)∗ −0.1264 (1.03) yes

−0.0332 (1.23) −0.0028 (0.16) 0.0321 (1.70) −0.0216 (1.02) 0.0284 (1.68) 0.0000 (0.33) −0.0336 (0.82) yes

−0.0198 (0.54) −0.0250 (1.08) 0.0333 (1.27) −0.0412 (1.39) 0.0290 (1.25) 0.0006 (1.90) 0.0069 (0.13) yes

−0.0376 (0.97) 0.0289 (1.24) 0.0288 (1.10) −0.0063 (0.22) 0.0282 (1.18) 0.0000 (0.11) −0.0739 (1.15) yes

– – – – – – 0.6473 6175 0.1468 0.0243

– – – – – – 0.6612 4659 0.2000 0.0346

−0.0086 (0.52) −0.0751 (2.32)∗ −0.0711 (0.42) 0.6167 11659 0.2093 0.0519

−0.0135 (0.61) −0.0700 (1.67) 0.2358 (0.89) 0.6156 6476 0.1951 0.0473

−0.0059 (0.24) −0.0815 (1.64) −0.0742 (0.42) 0.6181 5183 0.2465 0.0628

−3419 765∗∗ 58∗∗ 28∗∗ 28∗∗ 131∗∗ 200∗∗

−2386 653∗∗ 53∗∗ 14 13 123∗∗ 186∗∗

−6136 1914∗∗ 75∗∗ 17 13 435∗∗ 647∗∗

−3472 1102∗∗ 40∗∗ 14 12 255∗∗ 363∗∗

−2596 921∗∗ 41∗∗ 16 15 221∗∗ 326∗∗

Variables available only for 1998/99 Other backward caste – – Missing caste – – Missing religion – – Mean of dependent variable 0.6530 Observations 10834 0.1608 Pseudo-R-squared 0.0273 Pseudo-R-squared due to state fixed effects −5868 log Pseudolikelihood Wald Chi2 1363∗∗ F -assets 107∗∗ F -rural∗ assets 37∗∗ F -village variables 37∗∗ F -household demographics 218∗∗ F -state fixed effects 365∗∗

Girls

Notes: See notes to Table 7.4.

most educated adult in the household is significant and, in contrast to the attendance results, completion rates are favoured by the most educated adult being a woman. Completion is also more likely when the household head is a woman and, as in the case of attendance, less likely when the principal female is working. Large families and, further,

192 Sonia Bhalotra and Bernarda Zamora

families with small children appear to find it harder to support their children through primary school. Although scheduled-tribe children are less likely to complete, in contrast to the attendance results, children from scheduled castes and from Muslim families are no less likely to complete. An important finding is that, holding constant a rich set of household and state-level covariates, girls are less likely than boys to complete primary school, the probability differential being 0.05 in 1992/3 and rising to 0.07 in 1998/9. Overall, the results suggest that a first step towards improving completion rates would be to close the gender gap in completion. The state dummies are jointly highly significant in each equation and in both years and most are individually significant. They explain around 2 to 4 (2 to 6) per cent of the total variance in attendance (completion) after controlling for the remaining variables. The second-stage results which take the state fixed effects from the probits for each year, pool them and regress them on state-level variables are in Table 7.6. The statelevel variables displayed in Table 7.6 explain 89 per cent (71 per cent) of the state variation in attendance (completion) that remains after conditioning on household wealth, education and demographics and, for rural households, on some infrastructure indicators. The rest of this section describes the state-level effects in some detail as these are particularly relevant to the design of interventions aimed at improving schooling outcomes. The female illiteracy rate significantly reduces attendance and completion for both genders. For attendance, this effect is four times as large for girls as for boys but, for completion, it is only marginally larger for girls. The male illiteracy rate has no effect (not shown), nor does the ratio of the female to the male illiteracy rate (shown). Real per capita income (net state domestic product) has a significantly positive effect on attendance, but no effect on completion. The effect of income on attendance is larger for girls than for boys. At a given level of income, the share of education expenditure has a positive effect on attendance and, again, there is no effect on completion. We also included the fraction of education expenditure that goes towards primary education, but this was insignificant in every specification. The ratio of development expenditure to state income has a significant effect on both attendance and completion (although significance is marginal for girls’ completion), even after controlling for education expenditure. Development expenditure includes, in addition to expenditure on education, expenditure on health, famine relief and food subsidies. We included the share of health expenditure in the equation, expecting that it may have

Primary Schooling in India 193 Table 7.6 Stage-2 results: regression of the state fixed effects on state-level variables Attendance All log real income p.c.

1.448∗∗

[4.70] log (development 0.667∗ expend/income) [2.32] log (education 1.450∗ expend/income) [2.71] Rural poverty gap −0.009∗∗ index [2.99] Urban poverty gap 0.014 index [1.69] Rural/urban mean 0.346∗∗ p.c. consumption [3.50] Female illiteracy −1.245∗∗ rate [4.33] Female/male illiteracy 0.084 rate [0.44] Female/male teachers- 0.004∗∗ primary school [3.69] Number of 0.000∗∗ elementary schools [3.82] Dummy for year −0.590∗∗ 1998/9 [5.00] Constant 2.475∗∗ [4.37] Observations 28 R-squared 0.89

Completion Boys

Girls

1.240∗∗

1.637∗∗

[3.65] 0.677∗ [2.27] 1.257 [2.12] −0.010∗∗ [3.55] 0.016∗ [2.27] 0.231∗ [2.19] −0.495 [1.38] 0.285 [1.40] 0.005∗∗ [3.97] 0.000∗ [2.39] −0.455∗∗ [3.67] 1.774∗∗ [3.46] 28 0.84

[5.21] 0.694∗ [2.21] 1.598∗∗ [3.11] −0.008 [1.82] 0.012 [0.97] 0.433∗∗ [4.59] −1.969∗∗ [6.31] −0.091 [0.42] 0.004∗ [2.38] 0.000∗∗ [4.44] −0.705∗∗ [5.74] 3.137∗∗ [4.26] 28 0.91

All

Boys

Girls

−0.457 [0.64] 1.274 [1.89] −1.038 [1.07] −0.004 [0.54] 0.001 [0.07] 0.251 [1.09] −2.368∗∗ [3.36] −0.232 [0.68] −0.010∗∗ [3.03] −0.000 [0.20] 0.345 [1.68] 2.486∗ [2.56] 28 0.71

−0.659 [0.93] 1.445∗ [2.14] −1.376 [1.45] −0.004 [0.49] −0.005 [0.29] 0.178 [0.75] −2.303∗∗ [3.02] −0.164 [0.47] −0.009∗ [2.69] −0.000 [0.43] 0.457∗ [2.36] 2.219∗ [2.12] 28 0.72

−0.249 [0.33] 1.141 [1.63] −0.733 [0.71] −0.004 [0.50] 0.006 [0.32] 0.332 [1.57] −2.701∗∗ [4.11] −0.399 [1.15] −0.012∗∗ [3.56] 0.000 [0.11] 0.205 [0.89] 3.053∗∗ [3.13] 28 0.71

Notes: The state fixed effects are estimates from Tables 7.4 and 7.5 respectively. All reported standard errors are robust to heteroscedasticity. The regressors are, in each round, averages over the 5 years preceding the date of the survey. These data are from the Indian state panel assembled by Besley and Burgess (2002, 2004).

a positive impact, especially on completion, given that there is considerable evidence that children with poor health join school late or drop out early (see for example Alderman et al. 2003). However, this variable had no effect in any of the equations, and so it was not retained. The significance of development expenditure conditional upon education and health expenditure suggests that poverty-alleviation policies like famine relief and food subsidies, which inter alia protect nutrition, significantly affect school attendance and completion. The ratio of rural to urban consumption (mean real consumption per capita) is an (inverse) indicator of between-sector inequality. We find

194 Sonia Bhalotra and Bernarda Zamora

that it has a positive effect on attendance for both genders, the effect being almost twice as large for girls as for boys. We see also a hint of a positive effect on girls’ completion rates. This result is plausible given that schooling outcomes are worse in rural areas. It suggests that improving average living standards in rural as compared with urban areas will benefit children and especially girls. The Gini coefficient was included to capture within-sector inequality. As both the rural and the urban Gini were insignificant, they were dropped. An alternative indicator of income distribution within each sector is poverty incidence, which is measured here by the poverty gap index. We find that higher rural poverty is associated with lower attendance, but there is no effect on completion. Urban poverty is insignificant in every specification. The number of elementary schools (that is, lower and upper primary; see note 2) has a positive effect on attendance but no effect on completion. The ratio of female to male teachers in primary schools encourages attendance, though unexpectedly this effect is no larger for girls than for boys. Also possibly unexpected is the result that the feminization of the teacher workforce adversely affects completion, this effect being greater for girls than for boys! This variable deserves further investigation. The year dummy indicates that, other things equal, state-level unobservables specific to the year 1998/9 pushed attendance rates down and completion rates up. We have no ready interpretation of this result.

Decomposition and simulation This section reports estimates of the extent to which the change in school attendance and completion rates between 1992/3 and 1998/9 can be attributed to changes in characteristics. It then simulates the change from 1998/9 to 2015, applying alternative weighting schemes corresponding to the estimated elasticities for 1992/3 and 1998/9 respectively. For the purpose of developing projections of the school indicators to 2015, based on the evolution of predictor variables between 1992/3 and 1998/9, it is convenient to have a one-step model that shows the impact of state- and household-level variables on education all at once. We therefore re-estimate the probits, replacing the state fixed effects with the state-level variables that appeared as regressors in stage 2 above: Sis = µZs + λ Xis + νis

(7.3)

Having confirmed that the linear probability model gives results similar to the probit, we use the linear estimator for this one-step model.

Primary Schooling in India 195

The estimates are reported in Appendix Tables 7.A1 and 7.A2. These are the equations that the decomposition described below is based upon. Since data on the state-level variables is available for only 15 of the 26 states, the sample is reduced and the estimates in Appendix Tables 7.A1 and 7.A2 are therefore not strictly comparable with those in Tables 7.4–7.6. The effects of the state-level variables are, in many cases, quite different in the Appendix tables as compared with Table 7.6. This may be because of omitted state-level variables in the one-step (Appendix) equation or because of correlations in this equation between the state-level variables and other regressors in the model. A standard if ad-hoc and questionable way of making extrapolations or predictions is to assume that the parameters of a model are stable and to predict changes in the outcome from changes in the predictor variables. When only a single cross-section of data is available, there may be no choice but to assume parameter stability. However, there is no reason a priori to believe that the relation of interest is time-invariant. The probability of attending or completing school is bounded between zero and one, and we may expect the marginal effect of ‘inputs’ to get smaller as educational outcomes improve. Alternatively, it may be argued that there are diminishing returns to some inputs. For instance, the positive effect of access to television on schooling outcomes may decline as TV sets become common enough for the relevant information to diffuse through the community. This may explain our finding that the effect of increasing the number of TV sets per capita in a village raises attendance and completion probabilities in 1992/3, but not in 1998/9 (see Tables 7.4 and 7.5). In the current study, we estimate the same model on each of two rounds of data. As a result, we can investigate the assumption that the elasticities are constant over time (that is, the same in the 1998/9 survey as in the 1992/3 survey). In particular, we assess the extent to which the observed change in schooling between the two periods can be attributed to changes in the predictor variables over that period versus changes in the parameters. Decomposition of differences in outcomes has a long history, following the pioneering work of Oaxaca (1973) and Blinder (1973). Most applications perform decompositions in order to compare two groups of people, such as men and women (as in Oaxaca 1973 and Blinder 1973), or Hindus and Muslims (as in Bhalotra and van Soest 2006; Bhalotra and Zamora 2008). In this case, the change attributable to differences in sample characteristics is regarded as ‘explained’ while that attributable to differences in parameters between the two groups is thought of as

196 Sonia Bhalotra and Bernarda Zamora

‘unexplained’ and, therefore, potentially related to discrimination or norms. The same ideas apply to the decomposition of changes over time that we undertake here. We can assess the extent to which changes in schooling outcomes are created by growth in characteristics (the predictor variables), the residual being the changes in (behavioural) responses to given characteristics. For example, we may find that the increase in school attendance observed can be attributed, in part, to an increase in the fraction of villages with a primary school. But had attendance in 1998/9 responded less to an increase in the supply of schooling than it did six years before, we would find a smaller increase in attendance than if the elasticity were constant over time. Most applications of the decomposition methodology have been to linear models. The procedure can be extended in a fairly straightforward manner to nonlinear models such as the probit (see Yun 2004). However, the detailed decomposition (that is, decomposition by individual variable or by specified subgroups of the regressor set) is harder to interpret in the nonlinear model. For this reason, we report decompositions based upon the linear probability model. The coefficients used to weigh changes in the predictor variables are those obtained from a pooled model.5 Consequently, our predictions do not implicitly assume stability of parameters for a given year, but we allow for some behavioural dynamics between both waves. Refer to Table 7.7 and consider attendance first. The increase in attendance rates of boys and girls over the six-year period 1992/3–1998/9 is predicted to be 9.8 and 16.7 percentage points, respectively, close to the actual increases of 9.5 and 16.3 per cent. Most of this is explained by the regressors: 78 per cent in the case of boys and 67 per cent in the case of girls. Decomposition by subgroup shows that most of the explained variation, in turn, is on account of the state-level variables. Assuming that the change per annum in attendance rates that is attributable to the regressors remains the same between 1998/9 and 2015 as it was between 1992/3 and 1998/9, and weighting by the coefficients estimated on a model that pools the 1992/3 and 1998/9 data, we predict primary school attendance rates to be 100 per cent in 2015 for both boys and girls. Now consider completion rates, for which the predicted change between 1992/3 and 1998/9 is 3.4 per cent for boys and 5 per cent for girls. These predicted changes are much more positive than the actual changes, which were −3.1 per cent for boys and −4.3 per cent for girls. As in the case of attendance, most of the predicted change is accounted for by growth in the regressors: 65 per cent in the case of boys and 90 per cent in the case of girls. Our predictions for the year 2015 are that

Table 7.7

Decomposition and prediction based on the linear probability model Attendance All

Completion Boys

Girls

All

Boys

Girls

Predicted change between 1992/93 and 1998/99

13.10% (0.0036)

9.79% (0.0042)

16.66% (0.0051)

4.05% (0.0084)

3.40% (0.0111)

4.98% (0.0129)

1. Predicted change due to variables

9.32% (0.0030)

7.56% (0.0034)

11.17% (0.0042)

3.06% (0.0067)

2.16% (0.0423)

4.45% (0.0104)

−0.004% (0.0005) 0.285% (0.0009) 0.682% (0.0011) 0.424% (0.0005) −0.186% (0.0003) −0.143% (0.0004) −0.043% (0.0002) 0.467% (0.0007)

0.011% (0.0004) 0.041% (0.0011) 0.560% (0.0010) 0.317% (0.0005) −0.181% (0.0003) −0.084% (0.0005) −0.060% (0.0003) 0.384% (0.0009)

−0.048% (0.0003) 0.556% (0.0013) 0.805% (0.0016) 0.536% (0.0007) −0.185% (0.0004) −0.202% (0.0005) −0.024% (0.0002) 0.580% (0.0011)

0.132% (0.0005) 0.540% (0.0024) 0.732% (0.0017) −0.481% (0.0011) 0.069% (0.0005) 0.179% (0.0010) 0.120% (0.0007) −0.195% (0.0020)

– – 0.279% (0.0030) 0.146% (0.0020) −0.356% (0.0014) 0.065% (0.0006) 0.128% (0.0013) 0.183% (0.0010) −0.396% (0.0029)

– – 1.177% (0.0042) 1.764% (0.0029) −0.326% (0.0019) 0.063% (0.0008) 0.247% (0.0015) 0.051% (0.0008) −0.225% (0.0036)

Decomposition of the change due to variables 1.1 Child demographics (age and gender) 1.2 Assets 1.3 Household adult education 1.4 Household demographics 1.5 Principal female working 1.6 Ethnicity 1.7 Religion 1.8 Rural villages supply

197

(Continued)

(Continued)

198

Table 7.7

Attendance

Completion

All

Boys

Girls

All

Boys

Girls

1.9 State variables

7.840% (0.0023)

6.573% (0.0029)

9.147% (0.0033)

1.961% (0.0057)

2.108% (0.0075)

1.701% (0.0086)

2. Predicted change due to coefficients

3.77% (0.0017)

2.23% (0.0021)

5.49% (0.0024)

0.99% (0.0044)

1.24% (0.0423)

0.53% (0.0066)

Predicted annual change due to variables

1.55%

1.26%

1.86%

0.51%

0.36%

0.74%

Schooling level in 1998/99

82.52%

86.03%

78.72%

61.67%

61.56%

61.81%

100%

69.82%

67.31%

73.68%

100%

66.89%

65.20%

69.57%

Predicted schooling level to 2015 Predicted schooling level to 2015

Linear prediction 100% 100% Compound rate prediction 100% 100%

Notes: The changes are in percentages and the standard deviation in parenthesis. The linear probability models upon which the decomposition is based are reported in Appendix Tables 7.A1 and 7.A2. Row 1 shows the predicted total change. This total change is decomposed into the change attributable to included regressors or variables (1) and the change associated with changes in coefficients (2). The change due to variables is estimated as


 k

 X98,k βk∗ −






 X92,k βk∗

k

where βk∗ are the coefficients estimated by pooling the data for the two NFHS rounds, 1992/3 and 1998/9. This is further decomposed into the contribution of nine groups of variables. The ‘predicted annual change due to variables’ is simply the ‘predicted change due to variables’ divided by 6, this being the number of years between 1992/3 and 1998/9. Linear predictions to the year 2015 (strictly, to 2014/15) are made by adding to the 1998/99 schooling level the predicted change to 2015, which is the result of multiplying the annual changes by 16, the number of years between 1998/9 and 2014/15. Compound rate predictions to 2015 are made by multiplying the 1998/99 schooling level by (1 + annual change)16 .

Primary Schooling in India 199

completion rates will rise from 61.6 per cent in 1998/9 to 65.2 per cent for boys and from 61.8 per cent in 1998/9 to 70 per cent for girls. A positive feature of these results is that the gender gap appears set to reverse. However, for neither boys nor girls are completion rates set to rise to anywhere near 100 per cent.

Conclusions Comparing educational data for children in the 1992/3 and 1998/9 surveys, we find that primary school attendance grew for both boys and girls in the age range 6–11, indeed, more rapidly for girls. However, reported completion rates for 12-year-old children deteriorated over this period. The parameters of models of the probability of attendance and completion change between 1992/3 and 1998/9, as a consequence of which any projections we make are sensitive to which parameters we use to weight the contribution of the change of variables over time. We use elasticities from a model that pools the 1992/3 and 1998/9 data. Assuming that the predicted change due to the regressors is the same between 1998/9 and 2015 as it was between 1992/3 and 98/9, we project that all girls and boys aged 6–11 will be attending primary school by the year 2015, but that, conditional upon enrolling, only 65 per cent of boys and 70 per cent of girls at age 12 will have completed primary school. Putting the encouraging results for attendance together with the pessimistic results for completion rates highlights the importance of late entry and dropout. This problem is particularly severe among girls. It is notable that we find that completion probabilities are increasing with the adult female literacy rate in the state and with the share of state income that goes towards development expenditure other than that on health and education. While attendance is sensitive to the level of state income and its distribution, completion is not. Further research motivated to identify policy interventions that will improve completion rates is merited. Factors such as poor health may delay enrolment and weaken cognitive ability and therefore progression. At the same time, school curricula that are uninteresting to the children or irrelevant to their future earnings prospects, or timetables that conflict with peak agricultural seasons, may be important constraints on completion. A further possibility is that children enrol but then fail to complete because the household is subject to an income or health shock that makes the opportunity cost of schooling too high for the family to afford at the time. Once a child has dropped out, she or he may not enrol again. A dynamic model estimated on longitudinal data is likely to provide some important insights.

200

Appendix Table 7.A1 Linear probability model of school attendance for children aged 6–11 NFHS1992/93

Age 7 Age 8 Age 9 Age 10 Age 11 Female∗ age 6 Female∗ age 7 Female∗ age 8 Female∗ age 9 Female∗ age 10 Female∗ age 11 Pucca house Own flush toilet Electricity Potable water into the house Separate room for cooking Land owner Livestock owner Durables index Rural∗ pucca house Rural∗ own flush toilet Rural∗ electricity

NFHS1998/99

All

Boys

Girls

All

Boys

0.1193 (12.54)∗∗ 0.1660 (18.92)∗∗ 0.1664 (17.80)∗∗ 0.1598 (17.82)∗∗ 0.1537 (15.98)∗∗ −0.0874 (8.49)∗∗ −0.1195 (12.01)∗∗ −0.1617 (18.05)∗∗ −0.1410 (14.28)∗∗ −0.1696 (18.71)∗∗ −0.1788 (16.89)∗∗ 0.0152 (1.65) −0.0196 (2.24)∗ 0.0834 (5.84)∗∗ 0.0008 (0.08) 0.0052 (0.54) −0.0213 (2.29)∗ 0.0041 (0.35) 0.0173 (6.44)∗∗ 0.0084 (0.69) −0.0038 (0.29) −0.0211 (1.38)

0.1201 (12.60)∗∗ 0.1672 (19.04)∗∗ 0.1717 (18.37)∗∗ 0.1618 (18.01)∗∗ 0.1656 (17.17)∗∗ – – – – – – – – – – – – 0.0053 (0.48) −0.0055 (0.54) 0.0776 (4.46)∗∗ −0.0054 (0.47) 0.0118 (1.03) −0.0013 (0.11) 0.0095 (0.70) 0.0134 (4.16)∗∗ 0.0014 (0.10) −0.0112 (0.77) −0.0336 (1.79)

0.0878 (9.12)∗∗ 0.0922 (10.15)∗∗ 0.1048 (10.67)∗∗ 0.0729 (7.82)∗∗ 0.0494 (4.66)∗∗ – – – – – – – – – – – – 0.0307 (2.34)∗ −0.0359 (2.80)∗∗ 0.0874 (4.40)∗∗ 0.0072 (0.53) 0.0001 (0.01) −0.0455 (3.28)∗∗ −0.0008 (0.05) 0.0217 (5.60)∗∗ 0.0110 (0.64) −0.0039 (0.20) −0.0024 (0.11)

0.0889 (11.40)∗∗ 0.0946 (13.14)∗∗ 0.1010 (13.03)∗∗ 0.0830 (11.02)∗∗ 0.0725 (8.70)∗∗ −0.0509 (5.61)∗∗ −0.0671 (8.41)∗∗ −0.0771 (10.40)∗∗ −0.0743 (9.05)∗∗ −0.1073 (13.76)∗∗ −0.1048 (11.03)∗∗ 0.0071 (0.90) −0.0110 (1.59) 0.0849 (5.09)∗∗ 0.0098 (1.29) 0.0190 (2.59)∗∗ 0.0105 (1.39) 0.0214 (2.20)∗ 0.0030 (1.31) 0.0044 (0.46) −0.0111 (1.13) −0.0622 (3.58)∗∗

0.0904 (11.64)∗∗ 0.0959 (13.36)∗∗ 0.1074 (13.89)∗∗ 0.0865 (11.42)∗∗ 0.0823 (9.84)∗∗ – – – – – – – – – – – – 0.0051 (0.53) −0.0021 (0.26) 0.0705 (3.38)∗∗ 0.0040 (0.43) 0.0181 (2.12)∗ 0.0142 (1.58) 0.0238 (2.09)∗ 0.0035 (1.18) −0.0079 (0.69) −0.0194 (1.74) −0.0593 (2.74)∗∗

Girls 0.0724 (8.50)∗∗ 0.0681 (8.48)∗∗ 0.0724 (8.26)∗∗ 0.0251 (2.95)∗∗ 0.0094 (0.97) – – – – – – – – – – – – 0.0076 (0.69) −0.0226 (2.27)∗ 0.1007 (4.37)∗∗ 0.0177 (1.67) 0.0205 (1.92) 0.0096 (0.89) 0.0184 (1.35) 0.0026 (0.81) 0.0194 (1.45) 0.0017 (0.12) −0.0634 (2.63)∗∗ (Continued)

201 Table 7.A1 (Continued) NFHS1992/93 All Rural∗ potable

water into the house Rural∗ separate room for cooking Rural∗ landowner

0.0248 (2.18)∗ 0.0478 (4.33)∗∗ 0.0687 (6.00)∗∗ −0.0191 Rural∗ livestock owner (1.43) ∗ 0.0119 Rural durables index (2.85)∗∗ Schooling years of highest- 0.0215 educated adult (31.26)∗∗ Highest-educated adult −0.0440 is female (7.98)∗∗ log household size −0.0352 (6.63)∗∗ Proportion of females −0.1611 under-5 (4.92)∗∗ Proportion of males −0.1540 under-5 (4.68)∗∗ Proportion of females −0.0088 aged 6–16 (0.29) −0.0502 Proportion of males aged 6–16 (1.62) Proportion of females 0.1595 aged 17–30 (6.22)∗∗ Proportion of males −0.1734 aged 17–30 (6.11)∗∗ Proportion of females 0.0798 (2.13)∗ aged 50+ Proportion of males 0.0407 (0.99) aged 50+ Household head female 0.0416 (4.12)∗∗ Child of head 0.0155 (2.47)∗ −0.0400 Principal female working (7.61)∗∗ Rural resident −0.1132 (6.45)∗∗ −0.0252 Scheduled caste (3.41)∗∗ Scheduled tribe −0.0878 (9.47)∗∗ Muslim −0.0668 (8.81)∗∗

NFHS1998/99

Boys

Girls

All

Boys

Girls

0.0311 (2.25)∗ 0.0375 (2.79)∗∗ 0.0662 (4.72)∗∗ −0.0198 (1.22) 0.0078 (1.58) 0.0172 (21.49)∗∗ −0.0522 (7.78)∗∗ −0.0322 (4.62)∗∗ −0.1186 (2.75)∗∗ −0.1093 (2.54)∗ 0.0188 (0.50) −0.1070 (2.62)∗∗ 0.1446 (4.47)∗∗ −0.1628 (4.51)∗∗ 0.0296 (0.63) 0.0159 (0.30) 0.0351 (2.76)∗∗ 0.0153 (1.89) −0.0462 (6.90)∗∗ −0.0904 (4.04)∗∗ −0.0176 (1.85) −0.0950 (7.97)∗∗ −0.0851 (8.72)∗∗

0.0210 (1.31) 0.0567 (3.65)∗∗ 0.0722 (4.38)∗∗ −0.0216 (1.14) 0.0161 (2.61)∗∗ 0.0262 (26.20)∗∗ −0.0339 (4.38)∗∗ −0.0454 (6.35)∗∗ −0.2115 (4.72)∗∗ −0.2164 (4.81)∗∗ −0.0596 (1.35) 0.0204 (0.48) 0.1721 (4.78)∗∗ −0.1963 (4.91)∗∗ 0.1422 (2.63)∗∗ 0.0527 (0.91) 0.0491 (3.57)∗∗ 0.0161 (1.82) −0.0336 (4.72)∗∗ −0.1370 (5.71)∗∗ −0.0337 (3.33)∗∗ −0.0779 (6.35)∗∗ −0.0479 (4.75)∗∗

−0.0006 (0.07) 0.0110 (1.25) 0.0267 (2.82)∗∗ −0.0114 (1.02) 0.0263 (7.73)∗∗ 0.0139 (25.37)∗∗ −0.0237 (5.39)∗∗ −0.0428 (8.79)∗∗ −0.1779 (6.26)∗∗ −0.1845 (6.50)∗∗ −0.0283 (1.12) −0.1088 (4.11)∗∗ 0.1902 (8.93)∗∗ −0.1425 (5.87)∗∗ 0.0193 (0.61) −0.0488 (1.33) 0.0301 (3.69)∗∗ 0.0104 (1.81) −0.0517 (10.60)∗∗ −0.0030 (0.15) −0.0153 (2.80)∗∗ −0.0863 (10.37)∗∗ −0.0425 (6.36)∗∗

0.0048 (0.43) 0.0099 (0.96) 0.0382 (3.37)∗∗ −0.0201 (1.52) 0.0224 (5.28)∗∗ 0.0111 (16.86)∗∗ −0.0236 (4.48)∗∗ −0.0455 (7.37)∗∗ −0.0978 (2.70)∗∗ −0.1410 (3.92)∗∗ −0.0158 (0.52) −0.1335 (3.88)∗∗ 0.1703 (6.65)∗∗ −0.0952 (3.21)∗∗ 0.0105 (0.26) −0.0704 (1.54) 0.0108 (1.04) 0.0208 (2.89)∗∗ −0.0427 (7.13)∗∗ 0.0077 (0.30) −0.0106 (1.58) −0.0933 (8.91)∗∗ −0.0625 (7.51)∗∗

−0.0090 (0.69) 0.0113 (0.88) 0.0113 (0.84) −0.0018 (0.12) 0.0302 (6.37)∗∗ 0.0172 (21.83)∗∗ −0.0232 (3.62)∗∗ −0.0431 (6.38)∗∗ −0.2572 (6.36)∗∗ −0.2332 (5.77)∗∗ −0.0419 (1.07) −0.0864 (2.32)∗ 0.2138 (6.80)∗∗ −0.2088 (5.88)∗∗ 0.0127 (0.27) −0.0257 (0.48) 0.0528 (4.57)∗∗ 0.0009 (0.11) −0.0609 (8.80)∗∗ −0.0165 (0.59) −0.0196 (2.50)∗ −0.0777 (6.60)∗∗ −0.0226 (2.51)∗

(Continued)

202 Table 7.A1

(Continued) NFHS1992/93 All

−0.0181 (1.36) −0.0280 Other religion (2.50)∗ Rural∗ distance to nearest town −0.0002 (1.35) −0.0012 Rural∗ distance pucca road (0.61) ∗ ∗ Rural distance to pucca road 0.0008 primary school (0.39) Rural∗ distance to pucca −0.0036 (3.94)∗∗ road∗ girl 0.0154 Rural∗ village electrified (1.96) Rural∗ primary school 0.0292 in village (2.77)∗∗ Rural∗ middle school in village 0.0234 (3.21)∗∗ 0.0070 Rural∗ secondary school in village (0.91) 0.0064 Rural∗ bank in village (0.82) −0.0110 Rural∗ post office in village (1.56) Rural∗ no. of TV sets in village 0.0002 per 1000 habitants (2.47)∗ Rural∗ missing number of 0.0445 TV sets (1.53) State-level variables log real income p.c. 0.3546 (13.88)∗∗ log (development expenditure/ 0.0460 income) (1.78) log (education expenditure/ 0.3416 income) (9.01)∗∗ Rural poverty gap index −0.0080 (8.99)∗∗ Urban poverty gap index 0.0049 (2.89)∗∗ Rural/urban mean p.c. −0.2004 consumption (4.64)∗∗ Female illiteracy rate 0.1214 (2.08)∗ Female/male illiteracy rate 0.2040 (5.61)∗∗ Female/male teachers in −0.0161 primary school (4.61)∗∗

Christian

NFHS1998/99

Boys

Girls

All

Boys

Girls

−0.0222 (1.35) −0.0293 (2.10)∗ −0.0001 (0.48) −0.0047 (1.86) 0.0021 (0.81) – – 0.0208 (2.07)∗ 0.0422 (3.10)∗∗ 0.0082 (0.88) 0.0010 (0.10) 0.0045 (0.47) −0.0130 (1.45) −0.0000 (0.29) 0.0231 (0.61)

−0.0255 (1.33) −0.0219 (1.37) −0.0004 (1.66) −0.0014 (0.60) 0.0001 (0.03) – – 0.0128 (1.22) 0.0140 (0.99) 0.0365 (3.68)∗∗ 0.0142 (1.32) 0.0087 (0.77) −0.0058 (0.61) 0.0005 (3.41)∗∗ 0.0837 (2.09)∗

0.0306 (2.28)∗ −0.0418 (3.91)∗∗ −0.0006 (3.29)∗∗ 0.0032 (3.27)∗∗ −0.0023 (2.32)∗ −0.0008 (1.91) 0.0397 (5.23)∗∗ 0.0246 (2.51)∗ −0.0012 (0.19) 0.0081 (1.23) −0.0013 (0.19) −0.0106 (1.78) 0.0000 (1.72) −0.0012 (0.09)

0.0430 (2.80)∗∗ −0.0508 (3.80)∗∗ −0.0007 (3.02)∗∗ 0.0016 (1.29) −0.0013 (1.07) – – 0.0323 (3.47)∗∗ 0.0226 (1.89) −0.0014 (0.19) 0.0055 (0.68) 0.0035 (0.43) −0.0084 (1.14) −0.0000 (0.13) 0.0201 (1.34)

0.0177 (0.85) −0.0244 (1.68) −0.0005 (1.79) 0.0041 (3.21)∗∗ −0.0034 (2.56)∗ – – 0.0454 (4.28)∗∗ 0.0301 (2.20)∗ 0.0007 (0.08) 0.0092 (0.95) −0.0061 (0.61) −0.0121 (1.43) 0.0000 (7.72)∗∗ −0.0231 (1.15)

0.3739 (11.81)∗∗ 0.0528 (1.65) 0.3837 (8.20)∗∗ −0.0051 (4.49)∗∗ 0.0039 (1.87) −0.0035 (0.06) 0.2566 (3.53)∗∗ 0.2022 (4.45)∗∗ −0.0070 (1.61)

0.3321 (9.29)∗∗ 0.0357 (0.99) 0.2984 (5.65)∗∗ −0.0112 (9.05)∗∗ 0.0063 (2.65)∗∗ −0.4175 (6.84)∗∗ −0.0367 (0.46) 0.1964 (3.93)∗∗ −0.0269 (5.50)∗∗

0.0162 (1.11) 0.0032 (0.15) −0.0916 (4.85)∗∗ 0.0003 (1.14) 0.0032 (5.44)∗∗ 0.0596 (9.77)∗∗ −0.2342 (9.13)∗∗ −0.0512 (4.22)∗∗ 0.0004 (3.01)∗∗

0.0841 (4.84)∗∗ 0.0497 (1.93) −0.0223 (0.98) −0.0004 (1.14) 0.0014 (1.96)∗ 0.0288 (3.78)∗∗ −0.0231 (0.77) 0.0164 (1.12) 0.0004 (2.01)∗

−0.0578 (2.72)∗∗ −0.0482 (1.57) −0.1684 (6.17)∗∗ 0.0011 (2.80)∗∗ 0.0051 (5.94)∗∗ 0.0925 (10.53)∗∗ −0.4667 (12.27)∗∗ −0.1229 (7.00)∗∗ 0.0005 (2.30)∗

(Continued)

203 Table 7.A1 (Continued) NFHS1992/93

Constant Observations R2

NFHS1998/99

All

Boys

Girls

All

Boys

Girls

0.6754 (8.80)∗∗ 57740 0.28

0.5345 (5.57)∗∗ 30159 0.22

0.8010 (7.50)∗∗ 27581 0.33

0.6001 (11.31)∗∗ 56201 0.18

0.5327 (8.16)∗∗ 29212 0.15

0.6231 (8.38)∗∗ 26989 0.21

Notes: Robust t -statistics in parentheses. ∗ significant at 5%. ∗∗ significant at 1%

Table 7.A2 12 years

Linear probability model of school completion for children aged

NFHS1992/93 All −0.0388 (2.62)∗∗ Pucca house 0.0266 (1.16) −0.0271 Own flush toilet (1.27) Electricity 0.1061 (2.80)∗∗ Potable water into the house 0.0030 (0.13) Separate room for cooking 0.0425 (1.77) Landowner 0.0125 (0.52) −0.0127 Livestock owner (0.47) Durables index 0.0306 (4.64)∗∗ −0.0067 Rural∗ pucca house (0.22) 0.0162 Rural∗ own flush toilet (0.50) −0.0305 Rural∗ electricity (0.76) 0.0088 Rural∗ potable water into the house (0.31) −0.0084 Rural∗ separate room for cooking (0.30) 0.0543 Rural∗ landowner (1.84)

Female

NFHS1998/99

Boys

Girls

All

Boys

– – 0.0385 (1.25) −0.0442 (1.51) 0.1299 (2.64)∗∗ 0.0139 (0.42) 0.0369 (1.11) 0.0115 (0.34) −0.0116 (0.32) 0.0265 (2.91)∗∗ −0.0107 (0.26) 0.0157 (0.35) −0.0550 (1.05) 0.0140 (0.36) −0.0179 (0.47) 0.0596 (1.47)

– – 0.0031 (0.09) −0.0116 (0.36) 0.0697 (1.17) −0.0074 (0.22) 0.0535 (1.53) 0.0184 (0.55) −0.0172 (0.44) 0.0355 (3.72)∗∗ 0.0115 (0.25) 0.0173 (0.37) 0.0136 (0.21) −0.0057 (0.14) 0.0144 (0.34) 0.0397 (0.93)

−0.0514 (4.07)∗∗ 0.0231 (1.08) 0.0475 (2.37)∗ 0.1214 (2.90)∗∗ 0.0144 (0.70) 0.0399 (1.89) −0.0150 (0.64) 0.0087 (0.34) 0.0061 (0.94) −0.0007 (0.03) −0.0231 (0.77) −0.0692 (1.58) −0.0273 (1.13) −0.0201 (0.82) 0.0271 (1.00)

– – 0.0418 (1.45) 0.0517 (1.88) 0.0868 (1.64) 0.0143 (0.49) 0.0485 (1.64) −0.0097 (0.29) 0.0381 (1.12) 0.0007 (0.08) −0.0027 (0.08) −0.0662 (1.61) −0.0634 (1.13) −0.0322 (0.96) −0.0278 (0.82) 0.0370 (0.97)

Girls – – 0.0067 (0.21) 0.0382 (1.32) 0.1674 (2.53)∗ 0.0113 (0.38) 0.0265 (0.89) −0.0248 (0.77) −0.0248 (0.66) 0.0139 (1.42) −0.0077 (0.19) 0.0291 (0.66) −0.0845 (1.22) −0.0136 (0.39) −0.0080 (0.23) 0.0304 (0.79) (Continued)

204 Table 7.A2

(Continued) NFHS1992/93 All

Rural∗ livestock

−0.0356 (1.11) −0.0108 Rural∗ durables index (1.05) Schooling years of highest0.0175 educated adult (10.28)∗∗ Highest-educated adult 0.0462 is female (3.43)∗∗ −0.0220 log Household size (1.60) Proportion of females −0.0907 under-5 (0.95) Proportion of males −0.3044 under-5 (3.12)∗∗ Proportion of females −0.0063 aged 6–16 (0.08) Proportion of males aged 6–16 −0.0573 (0.76) Proportion of females aged −0.0178 17–30 (0.26) Proportion of males aged 17–30 −0.2906 (3.85)∗∗ Proportion of females aged 50+ 0.0389 (0.41) Proportion of males aged 50+ 0.1568 (1.68) −0.0051 Household head female (0.22) Child of head −0.0196 (1.22) −0.0104 Principal female working (0.80) Rural resident −0.0106 (0.23) −0.0010 Scheduled caste (0.06) −0.1090 Scheduled tribe (4.26)∗∗ −0.0714 Muslim (3.81)∗∗ Christian −0.0158 (0.49) −0.0679 Other religion (2.50)∗ Rural∗ distance to nearest town −0.0007 (1.45) Rural∗ distance pucca road −0.0155 (2.64)∗∗

owner

NFHS1998/99

Boys

Girls

All

Boys

Girls

−0.0096 (0.22) −0.0099 (0.70) 0.0166 (7.33)∗∗ 0.0412 (2.35)∗ −0.0122 (0.65) −0.0178 (0.14) −0.2244 (1.75) 0.0185 (0.18) −0.0105 (0.10) −0.0099 (0.11) −0.3358 (3.39)∗∗ 0.0279 (0.23) 0.1938 (1.60) 0.0046 (0.14) −0.0445 (2.13)∗ −0.0309 (1.81) −0.0256 (0.42) 0.0121 (0.50) −0.0701 (2.13)∗ −0.0883 (3.49)∗∗ 0.0606 (1.47) −0.1046 (2.78)∗∗ −0.0010 (1.67) −0.0150 (2.37)∗

−0.0700 (1.47) −0.0138 (0.90) 0.0199 (7.72)∗∗ 0.0493 (2.35)∗ −0.0363 (1.73) −0.1776 (1.28) −0.4076 (2.68)∗∗ −0.0738 (0.65) −0.0882 (0.77) −0.0140 (0.14) −0.2570 (2.22)∗ 0.0613 (0.41) 0.1225 (0.84) −0.0117 (0.35) 0.0161 (0.65) 0.0180 (0.90) −0.0174 (0.23) −0.0250 (0.80) −0.1631 (4.08)∗∗ −0.0478 (1.71) −0.1023 (2.23)∗ −0.0257 (0.66) −0.0002 (0.32) −0.0158 (1.24)

−0.0068 (0.23) 0.0239 (2.59)∗∗ 0.0201 (13.58)∗∗ 0.0347 (2.92)∗∗ −0.0220 (1.80) −0.4848 (5.59)∗∗ −0.4751 (5.60)∗∗ −0.0569 (0.90) −0.2664 (4.03)∗∗ −0.0646 (1.12) −0.4379 (6.57)∗∗ −0.0751 (0.89) 0.0049 (0.06) 0.0565 (2.79)∗∗ 0.0184 (1.25) −0.0196 (1.62) 0.0690 (1.37) −0.0078 (0.56) −0.0674 (3.43)∗∗ −0.1054 (6.72)∗∗ −0.0794 (2.23)∗ −0.1075 (3.46)∗∗ −0.0005 (0.97) −0.0002 (0.09)

−0.0361 (0.92) 0.0331 (2.68)∗∗ 0.0202 (9.99)∗∗ 0.0337 (2.09)∗ −0.0178 (1.08) −0.3181 (2.68)∗∗ −0.2775 (2.41)∗ 0.0903 (1.06) −0.1686 (1.87) −0.0318 (0.41) −0.3885 (4.39)∗∗ 0.0801 (0.71) 0.1637 (1.51) 0.0290 (1.04) 0.0355 (1.81) −0.0008 (0.05) 0.0308 (0.48) −0.0064 (0.34) −0.0891 (3.41)∗∗ −0.1338 (6.25)∗∗ −0.1255 (2.67)∗∗ −0.1225 (2.97)∗∗ 0.0005 (0.81) 0.0006 (0.18)

0.0253 (0.59) 0.0117 (0.85) 0.0202 (9.35)∗∗ 0.0336 (1.92) −0.0357 (1.97)∗ −0.6738 (5.36)∗∗ −0.6948 (5.54)∗∗ −0.2514 (2.64)∗∗ −0.3546 (3.63)∗∗ −0.1040 (1.19) −0.4980 (4.94)∗∗ −0.2744 (2.21)∗ −0.1763 (1.41) 0.0913 (3.05)∗∗ 0.0002 (0.01) −0.0429 (2.38)∗ 0.0853 (1.08) −0.0051 (0.24) −0.0420 (1.40) −0.0760 (3.36)∗∗ −0.0186 (0.34) −0.0968 (2.11)∗ −0.0017 (2.43)∗ −0.0003 (0.09)

(Continued)

205 Table 7.A2 (Continued) NFHS1992/93 All road∗ primary

Distance to pucca school Distance to pucca road∗ girl

0.0166 (2.77)∗∗ −0.0035 (1.04) 0.0183 Rural∗ village electrified (0.87) Rural∗ primary school in village 0.0020 (0.07) Rural∗ middle school in village 0.0062 (0.33) 0.0611 Rural∗ secondary school in village (3.09)∗∗ 0.0193 Rural∗ bank in village (0.96) −0.0313 Rural∗ post office in village (1.70) 0.0002 Rural∗ no. of TV sets in village per 1000 habitants (1.29) −0.1020 Rural∗ missing number of TV sets (1.23) State-level variables log real income p.c. 0.1825 (2.82)∗∗ log (development expenditure/ 0.2734 income) (4.13)∗∗ −0.0001 log (education expenditure/ income) (0.00) Rural poverty gap index 0.0083 (3.50)∗∗ Urban poverty gap index 0.0111 (2.64)∗∗ Rural/urban mean p.c. 0.2741 consumption (2.48)∗ Female illiteracy rate −0.5735 (4.03)∗∗ Female/male illiteracy rate −0.2981 (3.40)∗∗ Female/male teachers: primary 0.0027 school (0.31) Constant 1.1718 (6.13)∗∗ Observations 8225 R2 0.17 Notes: Robust t -statistics in parentheses. ∗ Significant at 5%. ∗∗ Significant at 1%.

NFHS1998/99

Boys

Girls

All

Boys

Girls

0.0150 (2.27)∗ – – 0.0091 (0.34) 0.0390 (1.08) 0.0120 (0.50) 0.0704 (2.73)∗∗ 0.0092 (0.34) −0.0403 (1.68) 0.0002 (0.62) −0.0861 (0.85)

0.0169 (1.30) – – 0.0378 (1.08) −0.0538 (1.08) 0.0035 (0.12) 0.0410 (1.36) 0.0399 (1.31) −0.0174 (0.61) 0.0002 (1.61) −0.1158 (0.78)

0.0007 (0.29) −0.0004 (0.31) 0.0295 (1.63) −0.0405 (1.66) 0.0114 (0.75) 0.0346 (2.05)∗ −0.0126 (0.69) 0.0194 (1.27) 0.0000 (0.11) −0.0393 (1.21)

−0.0006 (0.19) – – 0.0317 (1.33) −0.0245 (0.76) −0.0096 (0.48) 0.0309 (1.36) −0.0278 (1.12) 0.0235 (1.16) 0.0007 (2.05)∗ −0.0063 (0.15)

0.0012 (0.34) – – 0.0251 (0.89) −0.0475 (1.29) 0.0410 (1.80) 0.0369 (1.47) −0.0004 (0.02) 0.0143 (0.63) −0.0000 (0.07) −0.0659 (1.29)

0.1719 (2.05)∗ 0.3044 (3.45)∗∗ −0.0460 (0.37) 0.0088 (2.81)∗∗ 0.0081 (1.44) 0.3766 (2.52)∗ −0.4999 (2.65)∗∗ −0.2611 (2.24)∗ 0.0004 (0.03) 0.9254 (3.60)∗∗ 4807 0.16

0.1873 (1.87) 0.2428 (2.44)∗ 0.0303 (0.21) 0.0071 (1.98)∗ 0.0139 (2.19)∗ 0.1207 (0.73) −0.6729 (3.09)∗∗ −0.3436 (2.56)∗ 0.0017 (0.13) 1.4308 (5.01)∗∗ 3418 0.21

0.1314 (3.41)∗∗ 0.2648 (4.60)∗∗ −0.0225 (0.46) −0.0012 (1.71) −0.0013 (0.85) 0.0974 (5.92)∗∗ −0.3691 (5.50)∗∗ −0.0046 (0.15) −0.0035 (9.64)∗∗ 0.9031 (6.77)∗∗ 9240 0.23

0.1246 (2.45)∗ 0.3724 (4.76)∗∗ −0.0393 (0.60) −0.0023 (2.36)∗ −0.0021 (0.98) 0.0782 (3.53)∗∗ −0.3380 (3.84)∗∗ 0.0357 (0.86) −0.0029 (5.97)∗∗ 0.9425 (5.19)∗∗ 5262 0.22

0.1172 (1.98)∗ 0.1564 (1.83) −0.0402 (0.54) 0.0000 (0.04) −0.0010 (0.41) 0.1159 (4.66)∗∗ −0.4780 (4.60)∗∗ −0.0663 (1.43) −0.0043 (7.98)∗∗ 0.8795 (4.43)∗∗ 3978 0.27

206 Sonia Bhalotra and Bernarda Zamora

Acknowledgements We are grateful to Mark McGillivray for inviting us to write this chapter and to Tony Addison for introducing us to the UNU-WIDER community.

Notes 1. For instance, nearly half of all children aged 6–11 who were not in school in 1999/2000, according to the NSS data, were in Uttar Pradesh and Bihar, with a further 11 per cent being in Madhya Pradesh. There is further geographic concentration at the village level. Just 10 per cent of villages in India account for nearly 50 per cent of all out-of-school children aged 6–11 while 20 per cent of all villages account for 75 per cent of all out-of-school children (World Bank 2004:77). 2. The next three years (age 12–14) are then referred to as upper primary (for example World Bank 2004: ch. IV). Upper primary may alternatively be referred to as middle school. And lower and upper primary together are sometimes referred to as elementary education. 3. Anil Deolalikar clarifies that the figures he uses in Deolalikar (2005) and World Bank (2004) are only illustrative. Also, despite more optimistic assumptions on the rate of growth of variables that improve schooling probabilities, his conclusions are, like ours, pessimistic. 4. These data were compiled by Besley and Burgess (2002, 2004). 5. Stata provides an excellent summary in the help-file for the command ‘Oaxaca’.

References Alderman, H., J. Hoddinott and B. Kinsey (2003) ‘Long Term Consequences of Early Childhood Malnutrition’. FCND Discussion Paper 168. Washington, DC: International Food Policy Research Institute. Banerjee, A. (2004) ‘Educational Policy and the Economics of the Family’. Journal of Development Economics, 74(1):3–32. Becker, G. and N. Tomes (1986) ‘Human Capital and the Rise and Fall of Families’. Journal of Labor Economics, 4(3): S1–S39. Behrman, J. and J. Knowles (1999) ‘Household Income and Child Schooling in Vietnam’. World Bank Economic Review, 13(2):211–56. Besley, T. and R. Burgess (2002) ‘The Political Economy of Government Responsiveness: Theory and Evidence from India’. Quarterly Journal of Economics, 117(4):1415–52. Besley, T. and R. Burgess (2004) ‘Can Labour Regulation Hinder Economic Performance? Evidence from India’. Quarterly Journal of Economics, 19(1):91–134. Bhalotra, S. and A. B. Deolalikar (2008) ‘Childhood Shocks and Completed School Years’. Mimeo, University of Bristol.

Primary Schooling in India 207 Bhalotra, S. and A. van Soest (2006) ‘A Decomposition of Religion Differences in Childhood Mortality in India’. Presented at the Annual European Society of Population Economics meetings, Verona, June. Bhalotra, S. and B. Zamora (2008) ‘Social Divisions in Education in India’. In A. Sharif and R. Basant (eds), Handbook of Muslims in India. Delhi: Oxford University Press. Bhalotra, S., A. Langer, F. Stewart and B. Zamora (2008) ‘Persistent Muslim/Hindu Inequalities in India’. Mimeo, CRISE, University of Oxford. Blinder, A. (1973) ‘Wage Discrimination: Reduced Form and Structural Variables’. Journal of Human Resources, 8(4):436–55. Chevalier, A. (2004) ‘Parental Education and Child’s Education: A Natural Experiment’. IZA Discussion Paper 1153. Bonn: Institute for the Study of Labour. Deolalikar, A. (2005) ‘The Health Millennium Development Goals in India: How Attainable?’ Paper presented at a workshop on Child Health in Developing Countries, 14–15 June, Department of Economics, University of Bristol. Drèze, J. and A. Sen (1995) India: Economic Development and Social Opportunity. New Delhi: Oxford University Press. Ham, J. (1986) ‘On the Interpretation of Unemployment in Empirical Labour Supply Analysis’. In R. Blundell and I. Walker (eds), Unemployment, Search and Labour Supply. Cambridge University Press, pp. 121–42. Jacoby, H. and E. Skoufias (1997) ‘Risk, Financial Markets and Human Capital in a Developing Country’, Review of Economic Studies, 64(3):311–35. Lleras-Muney, A. (2005) ‘The Relationship between Education and Adult Mortality in the US’. Review of Economic Studies, 72(1):189–221. Oaxaca, R. (1973) ‘Male-Female Differentials in Urban Labour Markets’. International Economic Review, 14(3):693–709. Rosenzweig, M. (1995) ‘Why Are there Returns to Schooling?’ American Economic Review, 85(2):153–8. World Bank (2004) Attaining the Millennium Development Goals in India: Role of Public Policy and Service Delivery. Washington, DC. Yun, M.-S. (2004) ‘Decomposing Differences in the First Moment’. Economics Letters, 82(2):275–80.

8 The Burden of Government Debt in the Indian States: Implications for the MDG Poverty Target Indranil Dutta

Introduction The purpose of this chapter is to explore what impact, if any, government debts have on achieving the poverty target of the Millennium Development Goals (MDGs) for the Indian states. The MDGs specify the target levels to be achieved for a set of specific indicators by 2015. By addressing a broad range of indicators such as income poverty, health, literacy, gender and environment, with strong interlinkages between them, the UN General Assembly which ratified the MDGs hoped to bring about a reduction in the overall level of deprivation in the world (UN 2000). The goals are ambitious, and include the halving of poverty, illiteracy and hunger by 2015. This also means however that to fulfil the goals, national governments, especially in the developing world, have to undertake major investments in the social sector. But how much they will really be able to do so will depend on the conditions of their finances, which indirectly determines therefore the success of the MDGs. In this chapter we take government debt as one of the indicators of financial health of governments and their ability to increase and sustain expenditure in the social sectors. Typically one would presume that large government debts are incurred in subsidizing health or education programmes or direct poverty eradication programmes. Therefore, an increased government debt would reflect an increased involvement of the government in such programmes. Thus higher debt will alleviate poverty. This, however, is not at all obvious. If interest payments on debts are high, a country may easily slip into a debt trap, where it is incurring larger debts just to be able to pay its previous ones. Higher debt may persuade governments to reduce some of their social programmes that may have directly benefited the poor. In such circumstances debt 208

Debt and the Poverty MDG in India

209

will increase poverty. In this chapter we explore this issue of how debt effects poverty in greater detail. Given the large concentration of poor and deprived in South Asia, the performance of this region becomes crucial to the achievements of the MDGs (World Bank 2006). In India, which is the largest country in the region, due to the federal nature of the political system, the constitution separates the responsibilities of the centre and the states. The centre and the states each have a list of areas which are under their direct control and there is also a concurrent list for which both the centre and the states are responsible.1 Most of the MDGs fall under the concurrent list or the state list. Hence, for our study we have focused on the Indian states. Moreover, given the differences between the states in India, in terms of both economic growth and quality of life indicators, such statelevel analysis provides a more realistic base to study the progress towards the MDGs. In studying poverty in India, therefore, it is important to assess the state governments’ role and capabilities. For the Indian states, in a series of papers Besley and Burgess (2000), Besley and Burgess (2004) and Burgess and Pande (2005) have discussed how institutional environments, business climate and access to finance impact poverty; the role of government finances in poverty reduction has not however been studied so far. This chapter will assess both the direction and the magnitude of the effects of debt on poverty; it will also place it in the context of having sustainable poverty reduction in the long run and, thus, of achieving the MDG with respect to poverty. According to news reports, in Orissa, which is one of the poorer states, ‘the government debt was 63 per cent of the state’s gross production and 329 per cent of its total revenue in 2003–4. Salary bills, pension and interest payments on loans are a whopping 77 per cent of the state’s annual expenditure’ (The Telegraph [Calcutta], 19 November 2004). This is not just the case for the poorer states; many other states in India face similar situations. West Bengal, relatively a medium-level state in terms of its achievements, spends around 46 per cent of its total receipts including tax, non-tax and loan receipts, to service debts. Its expenditure on salary, pensions and loan repayments is more than 100 per cent of its total revenue.2 Obviously, this does not leave much room for developmentrelated expenditures. In a more rigorous study of the public sector debts in India, Kochar (2004) notes that ‘India has among the most largest and most intractable fiscal imbalances in the world’. Rangarajan and Srivastava (2003) recommend a reduction in the level of the primary deficits so that overall the debt can be sustainable. In fact, taking account of hidden

210 Indranil Dutta

subsidies and future commitments by the various state governments, the debt burden takes on a serious magnitude notwithstanding the assets of the governments. In their study comparing Indian government finances with other emerging markets, Roubini and Hemming (2004) find that India faces a higher risk of a debt crisis owing to its huge debt burden. Recognizing the gravity of the issue and its potential to create severe macroeconomic imbalances, the Twelfth Finance Commission of India has recommended a radical restructuring of the state-level debts to reduce the overall debt burden (Bagchi 2005; Kurian 2005). Although there are several dimensions of the MDGs, we have chosen to study income poverty in particular. Apart from its importance within the MDGs, it is also one of the most studied indicators for the Indian states. Further, detailed data on poverty for each state has been collected for all states in India for several years. However, we should point out that the methodology used in this chapter can be applied equally to study the impact of government debt on any other MDG indicators. The plan of our chapter is as follows. In the next section we explore in some detail the interlinkages between debt and the MDGs, followed by a description of the data and the methodology we use. The section that follows estimates several econometric models and presents the predicted values of poverty in 2007 and 2015 along with an analysis of the results. The penultimate section discusses the results from simulation exercises that check for robustness of the predictions and the final section highlights the main implications of the results.

Debt and MDGs The literature describing how debt and the MDGs are related is limited. Not all of the MDGs will be affected by government fiscal policy. For instance, government debt may not have any bearing on the goal of achieving gender equality in both primary and secondary education, but it will certainly affect the goals of halving poverty and hunger, achieving universal education and reducing child mortality by two-thirds. Any goal that may require a government to pour in resources will be affected by the conditions of the government’s finances. Given the interest on the goal of halving poverty, this section explores how debt affects poverty reduction both directly and indirectly through its impact on economic growth. There are several channels through which debt can impact economic growth. First, higher debt increases the possibility of higher taxes in the future, which in turn dampens long-term investments. Investors may divert resources to short-term investment and may hold back on

Debt and the Poverty MDG in India

211

any current investment. This can lead to a case of reduced efficiency along with a lower level of investment (Bräuninger, 2002). All these may cause ‘debt overhang’, where the state’s ability to honour its future debt commitments may be lower than its actual debt. In turn, this may create an environment of economic uncertainty, and the possibility of capital flight increases substantially, leading to a decrease in growth and hence in poverty alleviation. The empirical evidence on debt overhang, however, remains inconclusive.3 On the other hand, under a Keynesian approach, debt can have a positive impact on growth by generating demand and creating employment. This is particularly apt for developed countries under depression. How much this theory is applicable to developing countries, where the problem is not just the lack of demand, is arguable. Although the causal direction between debt and economic growth may be difficult to establish, economic theory also predicts that higher public debt lowers savings and thus increases interest rates. The increased interest rates then reduce growth through a reduction in investments. Kochar (2004) shows that public debt in India has been financed through private savings. This has allowed India to avoid significant external imbalances and inflationary pressures but has forced the government to offer an interest rate much higher than the market, thus making the public debt even more unsustainable. A higher debt also leads to reduction in the availability of credit for private investments and given that private investments are more efficient, this reduces the overall level of growth (Easterly 2004). Apart from the indirect impact of debt on poverty through economic growth, there is also the direct effect when governments with high debt curtail their social expenditures. For instance, IMF (2000) shows that for many of the highly indebted poor countries, a reduction in their debts has led to an increase in social expenditure that in addition to health and education includes spending on basic sanitary infrastructure, water supply and rural development. The direct impact of debt on social expenditure crucially affects the MDGs since most of the goals rely implicitly on government investments. For instance, to ensure universal primary education, the government needs to expand schools, hire more teachers and provide teaching tools; all these require substantial investment in education. Similarly, to reduce child mortality and achieve improvement in maternal health, governments in developing countries have to undertake more investment in the healthcare sector. If higher debt reduces such investments, then clearly it affects the achievements of the goals. In India with increased debt, the social expenditure decreased from 6.7 per cent in 1990–1 to 5.2 per cent in 2004–5 (Ghosh 2005). Typically,

212 Indranil Dutta

many of the government social expenditures are availed by the poor who lose out most when such expenditures are curtailed. Reduction in government involvement in these areas may prompt more private sector investment but the poor may be priced out of availing such services. Further, as Kochar (2004) argues, the increased public debt in India has led to a change in the composition of revenue expenditures. A higher proportion of government revenue is going towards financing the debt. Governments’ investment in infrastructure has reduced and in turn has led to a slowdown in economic growth. Lahiri (2000) shows that the level of debt in India is high compared with international standards and discusses the reasons behind the persistence of debt and how it impedes fiscal reforms. Kochar (2004) goes on to summarize that such increased levels of public debt have lead to a reduction in growth potential ‘through deterioration in the quality of public expenditure, limitations on the room for macroeconomic policy manoeuvre and on the scope for further structural reforms and liberalization’.

Methodology Our aim here is to understand whether debt actually helps or hinders the achievement of the MDGs’ poverty target. In modelling the causal direction from debt to poverty, we have used evidence from the literature (Kochar 2004). We proceed in two steps. First we estimate empirically the impact of government debt on poverty. The estimated equation may also involve other variables that matter for poverty reduction, such as GDP or health expenditure. Then we derive the trend values of those variables, along with debt, for 2007 and 2015. Using the estimated equation, and the derived trend values, we predict the levels of poverty for different states for 2007 and 2015. For the first step, since we have a panel data set, we run both the fixed-effects and the random-effects regressions. The fixed-effect regression that we estimate is ln pit = αi + Dt + β ln Xit + uit

(8.1)

where αi captures the state-specific effects, pit is the poverty headcount ratio for state i in year t, Xit is a vector of explanatory variables such as government debt, per capita health expenditure, per capita income and per capita electricity consumption and β indicates a vector which has the standard interpretation of being the coefficient of the associated control variables. Dt is a year dummy which takes into account year-specific

Debt and the Poverty MDG in India

213

effects. uit is the error term. Similarly the random-effects regression is as follows: ln pit = a + Dt + β ln Xit + εi + uit (8.2) where εi ∼ N(0, σε2 ) represents the state-specific random effects. Although we are interested in the effect of government debt on poverty, one can however claim that debt itself is influenced by the level of poverty. This will indeed bias the estimates from (8.1) and (8.2). We shall deal with this issue later in the chapter. The next step is to use the estimated equation to derive the impact of debt on poverty. We use the following equation: ln piT = α∗i + β∗ ln XiT where pi is the predicted level of poverty in time T , α∗i β∗ are estimated coefficients (derived from equation (8.1) or (8.2)), and XiT represents the trend levels of the explanatory variables at T . For our purposes we consider T = 2007 and T = 2015. Data The main data we use to estimate equations (8.1) and (8.2) are for 25 states in India for 1993 and 1999.4 We describe the data below. For poverty we have the headcount ratio for each of the 32 states and union territories in India from 1973–4 to 1999–2000, for, on average, every five years. These are based on the National Sample Surveys; our particular data comes from the Economic Survey of Delhi 2001–2. For 1999–2000 the data was collected using both a 30-day recall period and a 7-day recall period. We have used the 30-day recall period for our case, because it is closer to most of the adjusted estimates that various studies have pointed out.5 For calculating the trend of poverty for different states we have considered the whole data set from 1973 onwards, but we have used only the data for 1993–4 and 1999–2000 for estimating equations (8.1) and (8.2). The main reason for doing so is the limited data we have with regard to government debt, health expenditure and other variables of interest. As an indicator of government debts, we consider the ratio of debt to gross state domestic product (GSDP) in each state. Simply considering the level of debt is not sufficient, since it does not give an indication of the paying capability of the government. By taking the ratio of debt to GSDP, we get a fair idea of the burden of the debt on the government. We have this information from the report of the Twelfth Finance Commission for each state from 1993–4 to 2002–3 for every year. The debt includes

214 Indranil Dutta

internal debts, loans, advances from the central government, provident funds and insurance funds. Since our intention here is to investigate how government debt affects poverty reduction, we also need to control for government expenditure in the social sector. We take government expenditure on health as a close indicator of the government’s expenditure in the social sector. For 25 states we have data from 1950–1 to 2001–2, on per capita state government expenditure on health, on average, for every five years. Not all states have information on all years. Based on previous studies (Datt and Ravallion 1998) we also take into account other variables of interest which may help explain poverty, such as real GSDP per capita and literacy rates. While GSDP per capita has a direct impact on poverty, variables such as literacy account for the level of human capital in the state. For the 25 states we have data on GSDP per capita for 1993–4 and 1999–2000. For one state, Mizoram, real GSDP per capita or net domestic product for 1999–2000 is unavailable. Datt and Ravallion (1998) show that literacy plays an important role in explaining why some states have been more successful at reducing poverty. From the Department of Education, Government of India, we have data for 1991, 1997 and 2001. We derive the literacy rates for 1993–4 and 1999–2000 through linear interpolation. We would also use electricity consumption per capita to test endogeneity issues in our estimations, which we discuss later. For electricity consumption per capita we have data for different states for 1990–1, 1994–5 to 1999–2000. Since we are interested in the year 1993–4, we derive the values for 1993–4 through linear interpolation using the available data points.

Results and analysis In order to estimate the factors that effect poverty, we consider several possible models, each with different control variables. The results here are based on a panel data for 25 Indian states for 1993–4 and 1999–2000. Table 8.1 shows the results, estimated using random-effect models. We also calculate the Breusch–Pagan test to check for the validity of the models. We will consider the fixed-effect estimation later. The first column in Table 8.1 shows the regression of the log of the headcount ratio on the log of debt ratio. The negative and significant time dummy implies that there is a decreasing trend in poverty – that is, over time poverty is decreasing in the Indian states. Also, the coefficient of the log of the debt ratio is significant and positive, which implies that increased debt will increase poverty. This result is not very obvious. Higher debt can also mean lower poverty, through higher employment

Debt and the Poverty MDG in India

215

Table 8.1 Random-effect models on log of the head count ratio Model 1 log Debt ratio

0.472 (0.224)

log Per capita health expenditure log Per capita GSDP Time dummy Constant Number of observations R2 Wald test p-value Breusch–Pagan

Model 2

Model 3

∗

Model 4 ∗

−0.349∗ (0.166)

−0.094∗ (0.016) 1.980∗ (0.738)

−0.038∗ (0.017) 5.119∗ (0.733)

−0.647∗ (0.208) −0.057∗ (0.017) 9.269∗ (1.841)

50 0.626 2674.19∗ 0.001

50 0.569 3154.72∗ 0.001

50 0.531 4745.43∗ 0.003

Model 5

−0.034∗ (0.014) 3.533∗ (0.719)

0.731∗ (0.241) −0.500∗ (0.187) −0.345 (0.244) −0.029∗ (0.015) 6.474∗ (2.218)

50 0.730 2124.70∗ 0.0004

50 0.727 2816.56∗ 0.001

0.810 (0.213) −0.584∗ (0.147)

Notes: The values in parenthesis are the robust standard errors. ∗ Indicates significance at 5 per cent.

from increased government expenditure. However, the poor are clearly not benefiting from any increased government debt. One explanation for such an occurrence may be that for many of the states, expenditure on salaries, pensions and loan payments is already close to 100 per cent of revenue. Further increase in debt is resulting from expenditure that is not necessarily targeted at the poor. This trend decrease in poverty holds true for all the models in Table 8.1. Compared with other single explanatory variable models, such as Models 2 and 3, Model 1 has a higher R2 . The second column in Table 8.1 shows the regression of the log of the headcount ratio on the log of health expenditure per capita. The coefficient of the log of health expenditure per capita is highly significant and negative, indicating that as health expenditure is increased poverty will be reduced. It provides an argument for continuing and increased government investment in the social sector. In column 3 we run the same regression but with per capita real GSDP as the control variable. The coefficient is negative and significant. In fact, if the regression is run without the time dummy, the elasticity is close to unity. Note also that the reduction in poverty through income growth is almost twice that from increased government expenditure in the social sector. The next column controls for both log of health expenditure and log of the debt ratio. The coefficient of both the log of the debt ratio and the log of the health expenditure is significant. However, the coefficient of log of health expenditure is negative and the coefficient of the log

216 Indranil Dutta Table 8.2 Fixed-effect model on log of the headcount ratio Model 6 log Debt ratio log Per capita health expenditure log Per capita GSDP Time dummy Constant Number of observations Adjusted R2 F-test

Model 7

Model 6

∗

1.579 (0.475)

Model 9 ∗

−0.055 (0.028) 1.214 (1.773)

1.594∗ (0.474) −0.496∗ (0.206) −0.622 (0.432) −0.046 (0.024) 6.088 (3.134)

50 0.846 16.47∗

50 0.848 13.57∗

1.519 (0.510) −0.584∗ (0.209)

−0.677∗ (0.323)

−0.127∗ (0.021) −1.652 (1.540)

−0.001 (0.033) 6.618∗ (1.507)

−0.515 (0.571) −0.061∗ (0.024) 8.095 (5.081)

50 0.823 21.96∗

50 0.737 14.92∗

50 0.705 12.84∗

Model 10

Notes: The values in parenthesis are the robust standard errors. ∗ Indicates significance at 5 per cent.

of the debt ratio is positive. This implies that after controlling for social expenditure, as the debt burden increases, poverty also goes up. But an interesting difference between Model 1 and Model 4 is that the elasticity of debt ratio on poverty is higher in Model 1, which implies that once the level of health expenditure is controlled, increase in debt just increases poverty at a higher rate. Column 5 takes into account GSDP per capita in addition to log of health expenditure and log of the debt ratio. Both the coefficient of the log of the debt ratio and the coefficient of the log of the health expenditure are significant, with a positive and negative sign respectively. But unlike other studies we find that coefficient of the per capita GSDP, though positive, is insignificant. It shows that at least for the Indian states, after controlling for health expenditure, increase in income does not make a significant dent in poverty. This brings to the fore the role of government expenditure in tackling poverty. Table 8.2 shows the fixed-effect estimation for the same regressions as in Table 8.1. It is clear from Table 8.2 that most of the results are similar to the random-effects model in Table 8.1. In the fixed-effects case, health expenditure reduces poverty, while higher debt increases poverty. Also, we see (Model 10) that log of per capita GSDP is insignificant when we control for both log of debt ratio and log of per capita health expenditure.

Debt and the Poverty MDG in India

217

There are, however, several notable differences between Tables 8.2 and 8.1. Interestingly, in Model 8, in contrast to the random-effect models, the per capita GSDP is positive but insignificant, indicating that GSDP per capita may have a limited role in reducing poverty. Another difference lies in the higher debt elasticity of poverty under the fixedeffect than the random-effect model. Within the fixed-effect models, the debt elasticity of poverty is more than twice that of other variables such as GDP per capita or health expenditure. Further, the debt elasticity of poverty is greater than one, which shows that an increase in debt by 1 per cent increases poverty by more than 1 per cent. Clearly, debt is not being incurred to undertake programmes to combat poverty; instead it is being used in a manner that exacerbates poverty. Hence, debt will be a dominating factor affecting poverty. Interestingly, for the fixed-effect models the time trend is not always significant, which shows that once we take the state-specific effects into account, the time effects may not be that important. Thus interstate differences matter more than differences over time. To compare the fixed-effect and the random-effect models, in Table 8.1 we calibrate the p-value of the Breusch-Pagan test, which indicates that the random-effect models may be more appropriate. Hence for further analysis we use the random-effect models. Although we have not reported the results here, unlike in other studies, we have found that literacy does not have a significant impact on poverty, especially in the presence of log per capita health expenditure. MDG:2007 and 2015 We choose the random-effects estimation of Model 1, Model 2 and Model 4 to deduce the impact of debt on achieving the MDG with respect to poverty. Model 4 is chosen because it is the most parsimonious model with a good fit. Models 1 and 2 on the other hand will give us good comparative scenarios, by showing the effects of debt and health expenditure, respectively, on poverty. Broadly, we can then discuss two cases: one, the impact of government investment in the social sector on poverty and two, the impact on poverty as such when we take into account government debt. Model 1 will be useful to compare the effect of debt on poverty, when we do not control for social expenditures. Is there a simultaneity bias? Before we use the three different specifications to predict the poverty in 2007 and 2015, it is imperative that we address the issue of simultaneity that may arise in this context.

218 Indranil Dutta

As mentioned earlier, while our interest is in measuring the impact of government debt on poverty, the argument runs that debt may itself be incurred while trying to reduce poverty. This two-way relationship between debt and poverty would in our estimation give rise to a ‘simultaneity bias’. In other words, debt in our specifications may be endogenous and hence correlated with the error term. We would like to address this issue in two fronts. First, it is important here to distinguish between fiscal deficit and debt. If government is indeed using its resources to reduce poverty, then higher poverty in the current period may thus be reflected in an increased budget deficit for the government in the current period. However, this increased deficit will lead to an increased government debt only in the future. Therefore, the current-period poverty and current-period debt are not directly related and hence issues of simultaneity do not arise. Second, for Model 1 and 4 we undertake 2SLS estimation where we explicitly take into account the impact of poverty on debt.6 We use log per capita electricity consumption (ln e) as an identifying variable. Thus, for instance, the revised estimation for Model 4 is based on the following: ln pit = α + Dt + β1 ln dit + β2 ln hit + εi + uit ln dit = a + Dt + β3 ln pit + β4 ln hit + β5 ln e + εi + vit

(8.3)

where dit represents the levels of debt for state i in time t, hit represents the level of health expenditure and the rest of the terms have the same interpretation as in equation (8.2). Our choice of log per capita electricity consumption as an identifying variable is based on the intuition that electricity consumption is reflective of a state’s level of development and industrialization, and if indeed debt is being used for income generation then states with more electricity consumption should have less need to incur debts. On the other hand, a direct link from per capita electricity consumption to poverty is not very obvious, especially since it seems such a link will have to work through per capita income because without income the level of electricity consumption cannot be very high. A similar twostage estimation is also undertaken for Model 1. Ideally we would also have liked to take in to account past poverty levels in our regressions, but owing to lack of data we cannot. The results comparing the 2SLS random-effect estimation and the standard random-effect estimation are provided in Table 8.3. From Table 8.3 two broad results stand out. First, for both models the estimated coefficients for debt under 2SLS are much higher than

Debt and the Poverty MDG in India

219

Table 8.3 2SLS and random-effect models on log of the headcount ratio Model 1 2SLS Random Effect log Debt ratio log Per capita health expenditure Time dummy Constant Number of observations R2 Wald test Hausman test p-value

Model 4

Random Effect

1.660∗ (0.678)

0.472 (0.249)

−0.0129∗ (0.028) −1.916 (1.232) 50 0.719 21.50∗ 0.170

−0.094∗ (0.016) 1.980∗ (0.827) 50 0.625 38.33∗

2SLS Random Effect 2.237∗ (0.768) −0.993∗ (0.294) −0.027 (0.031) 0.722 (1.740) 50 0.732 26.63∗ 0.280

Random Effect 0.810∗ (0.238) −0.584∗ (0.157) −0.034 (0.021) 3.533∗ (0.719) 50 0.730 62.85∗

Note: ∗ Indicates significance at 5 per cent.

under the standard framework. The coefficients are also positive and highly significant. It implies that under 2SLS estimation, debt will have an even greater negative impact on poverty, that is increased debt will result in more poverty. Thus our standard random-effects estimation underestimates the effect of debt on poverty and can be interpreted as a lower bound of the impact of debt on poverty. Second, more importantly, the Hausman test, which compares the coefficients from the two different estimation procedures, indicates a lack of any systematic differences between the coefficients.7 Thus for both Model 1 and Model 4 the standard random-effects estimation is equally valid. We believe that the 2SLS estimation together with the theoretical arguments should assuage concerns of an simultaneity bias in our estimation. Hence, for rest of the chapter our estimation is based on the standard random-effects model. Predicted values Tables 8.4 and 8.5 give the details of the predicted poverty for 2007 and 2015 for a smaller set of 16 states. These 16 major states comprise of 95 per cent of India’s population. Note, however, that our estimated equation is based on a larger number of states. First we discuss Table 8.4.

220 Indranil Dutta Table 8.4 Predicted values of poverty in 2007

State

MDG Poverty target 1990 2007

Poverty Projected poverty 2007 trend 2007 Model 1 Model 2 Model 4

Andhra Pradesh Assam Bihar Gujarat Haryana Himachal Pradesh Karnataka Kerala Madhya Pradesh Maharashtra Orissa Punjab Rajasthan Tamil Nadu Uttar Pradesh West Bengal

23.950 40.422 52.546 25.093 17.570 16.379 32.221 26.390 44.658 36.699 54.427 11.645 26.828 35.937 40.287 39.956

13.999 28.680 34.981 13.359 8.761 7.891 18.206 12.942 30.383 21.907 37.846 6.002 14.409 19.649 26.014 23.734

12.324 31.291 42.394 12.473 8.809 8.945 18.510 10.704 31.929 23.167 41.723 5.104 15.030 20.323 28.994 23.198

9.497 9.714 12.847 10.719 9.385 13.922 8.626 10.349 10.211 8.637 14.761 9.569 12.272 8.809 11.097 12.259

12.555 14.484 17.289 12.142 13.075 8.138 9.784 10.949 14.199 12.173 13.392 11.041 12.685 11.770 14.649 13.135

12.543 16.564 35.983 14.599 13.155 11.705 7.007 11.562 17.452 10.112 29.784 10.247 19.811 9.897 21.210 20.964

Unweighted average

32.813

19.923

20.932

10.792

12.591

16.412

Note: Model 1 includes only debt, Model 2 includes only health expenditure and Model 4 includes both debt and health expenditure.

Column 1 reports the level of poverty in 1990, the base year for the MDGs. Using a linear trend, column 2 reports the level of poverty that has to be attained by 2007 to be in line with achieving the MDG with respect to poverty by 2015. In column 3, using the poverty data from 1973–4 to 1999–2000 and fitting a linear trend, we derive the trend values of the headcount ratios for the 16 states in 2007. Columns 4, 5 and 6 show the predicted values of poverty in 2007 using Model 1, Model 2 and Model 4 respectively. The values for the log of health expenditure and log of the debt ratio are the trend values of those variables for 2007. There are several features that stand out. The first is that the unweighted average for the 2007 MDG poverty target is around 20 per cent, and judging by the trend poverty level India on average will not be able to meet the goal. On the other hand when we take the predictions from the estimated models, we find that India is on track for satisfying the MDG poverty target. In fact according to Model 1, which tracks the effect of debt, India will be well within the MDG target for 2007, indicating that in the medium term state government debt may not have much of a negative consequence on poverty. Further, if we take

Debt and the Poverty MDG in India

221

into account just the impact of government investment in the social sector (Model 2), most of the major states in India will be in line with the 2007 MDG, although average poverty now is higher than in the case when debt only is taken in to account. It is important to note, comparing Model 1 and 2, that the health expenditure elasticity of poverty is less than that of debt. In other words, from a same-percentage decline in debt the decline in poverty will be sharper than the increase in health expenditure. However, there are variations within states. States such as Haryana, Himachal Pradesh and Punjab fail to meet the targets under both Model 1 and Model 2. Surprisingly, a large part of the reduction in poverty is coming from poorer states like Assam and Orissa. But when government debt and health expenditure are both taken into account (Model 4) the number of states that will not be able to meet the 2007 MDG doubles. States such as Bihar, Gujarat and Rajasthan, which by Models 1 and 2 were within the MDG target, are now way above it. If we considered just the health expenditure, Gujarat would have reduced its poverty from around 25 per cent in 1990 to 12 per cent by 2007 – below its 2007 MDG target of 13.3 per cent. But when we take the debt into account, Gujarat’s poverty increases to 15 per cent. In the case of Bihar the jump in poverty is the largest, from 17 per cent when just health expenditure is considered to 35 per cent when debt also is taken into account. What is interesting here is that on their own, debt and health expenditure seem each to be able to reduce poverty significantly. But when we look at the effect of debt while controlling for the level of health expenditure, poverty increases dramatically. This applies to both Orissa and West Bengal too, though their increase in poverty is not high enough to overshoot their MDG target for 2007. One of the surprises in our empirics is the failure of Punjab and Gujarat, which are generally deemed to be the richer states, to meet the MDG targets. In a broader sense one may question why some of the ‘better’ states such as Gujarat and Punjab are not able to meet their MDG targets whereas the poorer states such as Assam and Orissa are able to do so. The answer to some extent lies in our modelling structure. Since we are using log-linear models, it implies that states with already low levels of poverty will need to put in more in terms of their investing in health and lowering of debt to reduce poverty than will states with high levels of poverty. Hence we see a dramatic decline in poverty for the poorer states. However, this also means that over time as the level of poverty comes down it will become difficult to achieve further reductions in poverty. This is highlighted in Table 8.5, which provides the same information as Table 8.4 but for 2015.

222 Indranil Dutta Table 8.5 Predicted values of poverty in 2015

State Andhra Pradesh Assam Bihar Gujarat Haryana Himachal Pradesh Karnataka Kerala Madhya Pradesh Maharashtra Orissa Punjab Rajasthan Tamil Nadu Uttar Pradesh West Bengal

Poverty Projected poverty 2015 Poverty MDG target trend 1990 2015 2015 Model 1 Model 2 Model 4 23.950 40.422 52.546 25.093 17.570 16.379 32.221 26.390 44.658 36.699 54.427 11.645 26.828 35.937 40.287 39.956

11.975 20.211 26.273 12.547 8.785 8.190 16.110 13.195 22.329 18.349 27.213 5.822 13.414 17.968 20.143 19.978

9.015 27.739 38.320 8.976 6.365 6.729 14.260 7.001 27.266 18.658 36.818 3.462 11.443 15.542 24.837 17.961

11.510 10.765 16.064 14.652 11.354 17.143 10.308 12.155 13.287 11.344 19.864 9.774 16.257 10.856 13.480 17.045

9.164 11.040 13.138 8.946 9.907 5.670 6.559 7.978 10.539 9.029 9.839 8.178 9.495 8.670 10.757 9.952

10.309 12.552 33.377 14.985 11.475 9.146 4.875 8.978 16.668 9.809 29.623 6.437 19.786 8.499 17.680 23.218

Unweighted average 32.813

16.406

17.149

13.491

9.304

14.839

Note: Model 1 includes only debt, Model 2 includes only health expenditure and Model 4 includes both debt and health expenditure.

Considering Model 4 (column 6), 8 out of the 16 Indian states will clearly not be able to meet the MDGs, although on average India will meet the MDG targets. While in 2007 the predicted poverty was lower than the targeted poverty by 3 percentage points, by 2015 the difference reduces to less than 1.5 percentage points. Interestingly, the average poverty predicted by just the debt ratio (Model 1) is greater than that predicted by just taking health expenditure (Model 2). Since the health expenditure elasticity of poverty is lower than that of debt, and despite the fact that we see that poverty is lower under health, it implies that by 2015 the trend decrease in debt ratio will be lower than the trend increase in health expenditure. In other words, failure to reduce government debt is impeding the reduction of poverty, although for all the three models the predicted poverty (Model 2 in column 5) will be within the MDG targets by 2015. However, the experience between the states is not uniform. As expected, states such as Maharastra and Karnataka are showing the greatest decrease in poverty. On the other hand, in addition to states such as Bihar, Gujarat, Haryana, Himachal Pradesh and Punjab, which

Debt and the Poverty MDG in India

223

fail to meet the MDG targets for 2015, two more states, Orissa and West Bengal, join this group. For West Bengal, poverty will increase in 2015 to 23 per cent from 21 per cent in 2007. The poverty in Orissa barely budges between 2007 and 2015. This is because, although from the trend levels of health expenditure poverty should decline, this is being countered by the increase in government debts. The average poverty predictions from our models differ form the trend average poverty rates for both 2007 and 2015. In particular while all our models predict that India would satisfy the MDG target for both years, the trend predictions imply the opposite. Thus, for states with low levels of poverty to begin with in 1990, such as Himachal Pradesh and Punjab, while the trend predictions for 2015 indicate that these states will meet the MDG targets, our predictions show that they will not do so. Punjab, with both high trend levels of debt ratio and low trend levels of health expenditure, may thus end up with higher poverty in 2015 than in 2000. There are also states like Assam and Madhya Pradesh where the trend predictions from column 3 show that they will not be able to meet their MDGs for poverty, but in our calculations they will be able to fulfil the targets. It is interesting to note that since the trend poverty rates are predicted using a linear model, the differences between the MDG targets and the poverty targets carry through for both 2007 and 2015.

Simulation Our predicted levels of poverty depended on the forecasted levels of debt and health expenditure. The forecasts were done by fitting a linear trend on a longer time series of these variables. However, it is quite probable that the forecasts will not match with the realized values, especially when the forecast period gets longer. Therefore in this section we discuss the predicted levels of poverty for 2015 based on Model 4, under different scenarios of debt and health expenditure. In particular we consider four cases each for debt ratio and health expenditure levels. In Table 8.6 we consider the cases where the debt ratio increases (and decreases) by 10 per cent and 25 per cent from the trend values, with health expenditure remaining unchanged at the trend levels. As is obvious, an increase in debt ratio will take the Indian states further away from achieving the MDG poverty targets. Note that in 2015, given the trend levels of debt ratio and health expenditure, the Indian states on average will be only just able to reduce poverty by half. Hence, increasing the debt ratio will make that task even harder. But more interestingly, a reduction of the debt ratio by 10 per cent from the trend values still does

224 Indranil Dutta Table 8.6 Simulated values of poverty in 2005, with varied levels of debt ratio Predicted poverty in 2015

State Andhra Pradesh Assam Bihar Gujarat Haryana Himachal Pradesh Karnataka Kerala Madhya Pradesh Maharashtra Orissa Punjab Rajasthan Tamil Nadu Uttar Pradesh West Bengal Unweighted average

10% 25% MDG increase increase Poverty target in debt in debt 1990 2015 ratio ratio 23.950 40.422 52.546 25.093 17.570 16.379 32.221 26.390 44.658 36.699 54.427 11.645 26.828 35.937 40.287 39.956 32.813

11.975 20.211 26.273 12.547 8.785 8.190 16.110 13.195 22.329 18.349 27.213 5.822 13.414 17.968 20.143 19.978 16.406

11.137 13.559 36.056 16.187 12.396 9.881 5.267 9.698 18.006 10.596 32.000 6.953 21.374 9.181 19.099 25.081 16.030

12.351 15.038 39.989 17.953 13.748 10.958 5.841 10.756 19.970 11.752 35.491 7.712 23.706 10.183 21.183 27.817 17.778

10% decrease in debt ratio

25% decrease in debt ratio

9.466 11.525 30.647 13.759 10.536 8.398 4.477 8.243 15.305 9.007 27.200 5.910 18.168 7.804 16.234 21.319 13.625

8.166 9.943 26.440 11.870 9.090 7.245 3.862 7.112 13.204 7.770 23.466 5.099 15.674 6.733 14.006 18.392 11.755

Note: Model 4, which included both debt and health expenditure, is used to predict the poverty under the different scenarios.

not reduce poverty to within the MDG target for the states which fail to meet the 2015 target. On the other hand with a 25 per cent decrease in the debt ratio, these states will on average come close to achieving the MDG targets, although the goal remains unattainable for Bihar, Haryana and Rajasthan. Therefore, a significant reduction in debt will help more states realize the MDG goals. Next we perform the same exercise for health expenditure levels. Using Model 4, we predict the level of poverty in 2015, when health expenditure is changed (increased and decreased) by 10 and 25 per cent. The results are reported in Table 8.7. As expected, higher health expenditure reduces poverty. But even with a 25 per cent increase in the health expenditure levels, 5 out of the 16 Indian states fail to meet the goals. On the other hand a 25 per cent decrease in health expenditure will stop 10 states from fulfilling the MDGs, the highest in terms of the number of failed states under the

Debt and the Poverty MDG in India

225

Table 8.7 Simulated values of poverty in 2015, with varied levels of health expenditure per capita Predicted poverty in 2015

State

10% MDG increase Poverty target in health 1990 2015 expenditure

25% increase in health expenditure

10% decrease in health expenditure

25% decrease in health expenditure

Andhra Pradesh Assam Bihar Gujarat Haryana Himachal Pradesh Karnataka Kerala Madhya Pradesh Maharashtra Orissa Punjab Rajasthan Tamil Nadu Uttar Pradesh West Bengal

23.950 40.422 52.546 25.093 17.570 16.379 32.221 26.390 44.658 36.699 54.427 11.645 26.828 35.937 40.287 39.956

11.975 20.211 26.273 12.547 8.785 8.190 16.110 13.195 22.329 18.349 27.213 5.822 13.414 17.968 20.143 19.978

9.751 11.872 31.571 14.174 10.854 8.651 4.612 8.492 15.766 9.278 28.019 6.088 18.715 8.039 16.723 21.961

9.050 11.018 29.300 13.154 10.073 8.029 4.280 7.881 14.632 8.611 26.004 5.650 17.369 7.461 15.520 20.381

10.963 13.348 35.495 15.936 12.203 9.727 5.185 9.547 17.726 10.432 31.503 6.845 21.042 9.039 18.802 24.691

12.195 14.848 39.483 17.726 13.574 10.820 5.767 10.620 19.718 11.604 35.042 7.614 23.406 10.054 20.915 27.465

Unweighted average

32.813 16.406

14.035

13.026

15.780

17.553

Note: Model 4, which included both debt and health expenditure, is used to predict the poverty under the different scenarios.

different scenarios. However, fewer states will be able to meet the MDG target with a 25 per cent increase in health expenditure (as in Table 8.7) compared with the number of states that fulfil the goals when the debt ratio is decreased by 25 per cent (Table 8.6). The differences in the numbers are not large, with Gujarat and West Bengal being the only states which are switching under the two conditions, that is, they fulfil the goals under a 25 per cent decrease in debt ratio but not under a 25 per cent increase in health expenditure. Further comparing Table 8.6 and Table 8.7, the impact of debt on achieving MDG is clear. A 10 per cent fall in the debt ratio decreases poverty by more than that achieved by a 10 per cent increase in health expenditure. A 25 per cent decrease in debt ratio does more to lower poverty than an increase in health expenditure by a similar percentage. Although there are states which do not achieve the MDG targets under

226 Indranil Dutta

any of the scenarios we have discussed, there is a considerable variation in the number of states that satisfy the MDG target by 2015. Out of the 16 states the number of states that achieve the target varies between 13 and 6. But still some broad patterns emerge. In general, all the southern states in India – Andhra Pradesh, Karnataka, Kerala and Tamil Nadu – will definitely achieve the MDG targets. On the flip side, the northern states such as Bihar, Haryana and Rajasthan would consistently fail to satisfy the targets. Orissa and West Bengal are the two states where deterioration in the debt ratio over time makes it difficult for them to achieve the MDG target by 2015.

Conclusion Our objective in this chapter has been to investigate whether government debts in India impact the ability to achieve the MDGs. The results show that debt is a hindrance to the achievement of the MDG poverty targets. We find strong evidence that government investment in the social sector is extremely important in reducing poverty, but government debt burden is actually stopping several states from attaining the MDGs. Increasing both debt and health expenditure by similar percentage points will lead to an increase in overall poverty, since debt’s marginal impact on increasing poverty is more than health’s impact on reducing poverty. Clearly then, a strategy of increasing debt to fund health and other social expenditures may not be a sensible policy from the point of view of reducing poverty. Also, as we had observed in Table 8.6 and 8.7, a 25 per cent decrease in debt ratio will help more states achieve the goal than will a 25 per cent increase in health expenditure. Therefore for policy purposes reduction of debt should be of crucial importance. We should point out that our model is based on a panel data of 25 states over just two years. A richer data set may yield different results. We took health expenditures as the main indicator for social expenditures by the government, but a more comprehensive measure may be a better predictor of poverty. Also, our health expenditure data are nominal values and there has been a significant increase in nominal health expenditure in the recent years. This may be driving some of results where some states are able to substantially reduce their poverty. If real expenditure on health is considered, it is quite probable that predicted levels of poverty may be even higher, since the increase in real expenditure on health is going to be lower than the increases in nominal expenditures on health. Further, for most part we find a remarkable consistency in the states that

Debt and the Poverty MDG in India

227

are able to achieve the goals and those that do not. What the reasons behind this remarkable consistency are is an issue for future research.

Acknowledgements I am grateful to participants of the UNU-WIDER project meeting on Millennium Development Goals: Assessing and Forecasting Progress, and to Adam Swallow for comments. I am indebted to an anonymous referee, to Sonia Bhalotra and to Mark McGillivray for extensive comments. The usual disclaimer applies.

Notes 1. The Seventh Schedule of the Constitution of India contains the lists of activities that come under the centre or the state. For more details refer to the Government of India website: http://indiacode.nic.in/coiweb/welcome.html 2. ‘Bengal on the Verge of Debt Trap’. The Telegraph [Calcutta], 8 February 2005. 3. See the discussions in Clements et al. (2003) on how debt can effect growth. Although these authors focused mainly on external debt, the analysis will also be valid for total debt that includes both domestic and external debt. 4. All the data used in this paper are available from www.indiastat.com 5. For a discussion of the issues in this context refer to Popli et al. (2005). 6. Government debt is not one of the regressors in Model 2 and hence no estimation of 2SLS model was undertaken. 7. For details on the Hausman test for 2SLS model refer to Baum et al. (2003).

References Bagchi, A. (2005) ‘Symposium on Report of Twelfth Finance Commission: Introduction and Overview’. Economic and Political Weekly, 30 July:3388–95. Baum, C. F., M. Schaffer and S. Stillman (2003) ‘Instrumental Variables and GMM: Estimation and Testing’. Boston College Working Paper 545. Chestnut Hill, MA. Besley, T. and R. Burgess (2000) ‘Land Reform, Poverty and Growth: Evidence from India’. Quarterly Journal of Economics, 115(2): 389–430. Besley, T. and R. Burgess (2004) ‘Can Labor Regulation Hinder Economic Performance? Evidence from India’. Quarterly Journal of Economics, 119(1):91–134. Bräuninger, M. (2002) ‘The Budget Deficit, Public Debt and Endogenous Growth’. Mimeo, University of Hamburg. Burgess, R. and R. Pande (2005) ‘Do Rural Banks Matter? Evidence from the Indian Social Banking Experiment’. American Economic Review, 95(3):780–95. Clements, B., R. Bhattacharya and T.Q. Nguyen (2003) ‘External Debt, Public Investments and Growth in Low Income Countries’. IMF Working Paper, WP/03/249. Datt, G. and M. Ravallion (1998) ‘Why Have Some Indian States Done Better than Others at Reducing Rural Poverty?’ Economica, 65, 17–35.

228 Indranil Dutta Easterly, W. (2004) ‘The Widening Gyre: The Dynamics of Rising Public Debt and Falling Growth’. Mimeo, New York University. Ghosh, J. (2005) ‘Twelfth Finance Commission and Restructuring of State Government Debt: A Note’. Economic and Political Weekly, 30 July:3435–9. IMF (International Monetary Fund) (2000) The Impact of Debt Reduction under the HIPC Initiative on External Debt Service and Social Expenditure. Washington, DC. Available at: http://www.imf.org/external/np/hipc/2000/112900.htm Kochar, K. (2004) ‘Macroeconomic Implications of the Fiscal Imbalances’. Mimeo, International Monetary Fund. Washington, DC. Kurian, N. J. (2005) ‘Debt Relief for States’. Economic and Political Weekly, 30 July:3429–34. Lahiri, A. (2000) ‘Budget Deficits and Reforms’. Economic and Political Weekly, 11 November:4048–54. Popli, G., A. Parikh and R. Palmer-Jones (2005) ‘Are the 2000 Poverty Estimates for India a Myth, Artefact or Real?’ Economic and Political Weekly, 22 October: 4619–24. Rangarajan, C. and D. K. Srivastava (2003) ‘Dynamics of Debt Accumulation in India: Impact of Primary Deficit, Growth and Interest Rate’. Economic and Political Weekly, 15 November:4851–8. Roubini, N. and R. Hemming (2004) ‘A Balance Sheet Crisis in India?’ Mimeo, International Monetary Fund, Washington, DC. UN (United Nations) (2000) ‘United Nations Millennium Declaration. Resolution adopted by the General Assembly’. Washington, DC. Available at: http://www.un.org/millenniumgoals/background.html World Bank (2006) Global Monitoring Report 2006: Strengthening Mutual Accountability – Aid, Trade and Convergence. Washington, DC.

Index Key: bold=extended discussion; f=figure; n=note; t=table.

correlations 66t ’critical [most important] finding’ 85 cross-country asset measure 62–4, 84 data deficiencies 60, 73, 74, 87(n10–11) data definition and measurement 60–5, 87(n10–11) data set 55, 56 data sources for conditioning variables 86t descriptive statistics 65–7, 72–3t, 77 determinants of cross-country wellbeing achievement and aid effectiveness 55 econometric results 72–3t empirical analysis 55, 56, 57, 58, 65–84, 87(n12–14) endogenous well-being variables 61–5 exogenous conditioning variables 64, 67 further research required 75, 85 impact of aid 78–84, 85, 87(n14) interdependencies 76 literature review 57–60, 65, 85, 86–7(n1–9) model structure 67–9, 87(n12) policy implications 55 quantitative importance 56 quintiles 56, 61–9, 72–3t, 75–84 regression R2 statistics 72t, 75 regression results 69–76, 87(n13) regression results: implications 76–8 residual correlations 72–3t search for pivotal variable 55–6, 58, 84–5 shocks 76t, 76–8 simultaneous equation system 84 standard deviation/standard errors 62t, 62, 64, 66t, 68, 69, 70–2t, 75, 76t, 77, 81, 82–3t, 87(n13) structural model 56

Addison, T. xiii, 120, 169, 206 Addison, T., et al. (2005) 86(n4), 87 Mavrotas, G. 87 McGillivray, M. 87 adults 128, 130, 131t, 146(n12) literacy rate 87(n9) underweight 39t, 39–40 Afghanistan 121(n4) Africa xv, 8, 25, 28, 29, 32f, 33–5, 40–1, 44 ’africa’ (variable) 64, 68t, 69, 70–1t, 74, 75 HIV/AIDS 42f, 42–3 poverty 44 age 9, 22, 131n, 133t, 136t, 137, 147(n14), 156–7, 163, 176–7, 188t, 197t, 206(n1–2) 6–11 years 183, 185t, 199, 200–3t 7–11 years 178t, 181, 200t 12 years 189–91t, 199, 203–5t ageing population 34, 45 agencies 86(n1) agriculture 41t, 41, 49, 154, 163, 199 Ahlburg, D. 57 aid 105t efficient intra-country allocation 55 impact on categories of government expenditure 59, 86–7(n6–8) ’OA’ (other aid) 105t as percentage of GDP 106–7t, 108, 109–10t targeting 16 well-being outcomes 59–60, 87(n8–9) see also ODA ’aid’ (variable) 64, 65, 67, 68t, 69, 70–1t, 72t aid effectiveness xv, 16, 55, 58–60, 86–7(n4–9), 85 aid and growth literature 58–9, 87(n8) aid impact 78–84, 87(n14) aid and MDGs: achieving health, wealth, wisdom 15, 16, 55–89 assumptions 67, 85 229

230 Index aid and MDGs: achieving health, wealth, wisdom – continued summary statistics 66t theory 86(n2) t-ratio 69, 70–1t variables 56 aid received (explanatory variable) 94 Albania 114t, 117t Alderman, H., et al. (2003) 193, 206 Hoddinott, J. 206 Kinsey, B. 206 Algeria 117t Anand, P. B. xiv, 92, 113, 121(n7), 121, 146(n4), 151 Anderson, E. 53(n23), 135 Andhra Pradesh 158t, 159f, 160t, 161f, 166t, 180t, 220t, 222t, 224–5t, 226 Angola 114t, 120t antenatal care 153, 154, 156 anthropometric measurement 39 Argentina 114t Armenia 33, 87(n10), 117t arms 18(n2) Arulampalam, W. 157, 169 Asia xv, 25, 29, 32f, 33–4, 41, 43t, 51(n7), 91, 154, 170(n9) HIV/AIDS 42–3 Assam 158t, 158, 159f, 160t, 161f, 166t, 180t, 188t, 220t, 221, 222t, 223, 224–5t Atkinson, A. 126, 146(n6) Atkinson inequality measure 146(n6) Australia xv, 30t, 36t Austria 114t, 117t auto-correlation 165t Azerbaijan 116t, 118t Bagchi, A. 210 Banerjee, A. 182 Bangladesh 21, 25, 51(n3), 61t, 79t, 103, 106t, 116t, 119t Barbados 117t ’basic needs multiplier’ (White) 53(n25) BCG [bacillus of Calmette and Guérin] 147(n16) Becker, G. 174 Behrman, J. 182 Belize 114t, 119t Benin 61t, 79t, 106t, 114t, 120t Berg, G. van den, et al. (2006) 154, 173

Besley, T. 157, 193n, 206(n4), 209 Bhalotra, S. xiv 145, 153–4, 156–7, 159, 163–5, 167, 169, 170(n11), 171(n12), 171, 175–6, 188, 195, 206–7, 227 Bhattacharya, R. 227 Bhutan 118t Bihar 156, 158t, 159f, 160t, 161f, 166t, 180t, 206(n1), 220t, 221–6 birth attendance rate 130–2, 133–4t, 136t, 141, 142f, 144–5 Black, R., et al. (2003) 169–70(n2), 171 Bryce, J. 171 Morris, S. 171 Blinder, A. 195 Blöndal, N. xiv see also White and Blöndal Bloom, D., et al. (2004) 57, 87 Canning, D. 87 Sevilla, J. 87 Bloom, G. 51–2(n10) Blossner, M. 53 Bolivia 17, 61t, 79t, 119t DHS data 127–8, 146–7(n12–14, n18) distribution of progress towards MDGs 130–2 GIC (1989, 1998) 132, 135f health sector reform 129 household surveys (LSMS) 129, 147(n18) income growth rates (pro-poor, 1990s) 129 non-income achievements by income decile (1989, 1998) 133–4t pro-poor growth towards MDGs 123–49 pro-poor growth and pro-poor progress (1989–98) 136t progress towards MDGs 130, 131t Bonn Conference (2002) 121(n1) Boone, P. 60 Borghi, E. 53 Bosnia & Herzegovina 114t, 117t Botswana 114t, 119t Bourguignon, F. 146(n5) boys 17, 38t, 51(n3); 175–7, 177–8t, 181–3, 185–8t, 188–94, 196–9, 200–5t see also ’children/not in school’ Brandolini, A. 126, 146(n6) Braun, J. von, et al. (2005) 39, 49, 54

Index Brazil 61t, 79t, 103, 118t Breusch–Pagan test 214, 215t, 215 Brikke, F. 122 Bryce, J. 171 Bulgaria 114t, 117t Burgess, R. 157, 193n, 206(n4), 209 Burkina Faso 61t, 79t, 116t, 120t Burnside, C. 58 Burundi 116t, 120t bush latrine 61–3, 93 business climate 209 California 153 calories 41t Cambodia 61t, 79t, 81, 116t, 120t, 121(n4) Cameroon 61t, 79t, 106t, 116t, 119t Canada 114t Canning, D. 87 capacity-building 65 Cape Verde Islands 119t capital flight 211 capital to labour ratio 86(n2) capital markets 57 Caribbean 26t, 30t, 36t case studies 112 caste 169, 175, 183, 187t, 191t centiles 132, 136, 141 Central Africa 25, 26t, 29, 30t, 34, 36t Central African Republic (CAR) 61t, 79t, 105t, 115t, 120t Chad 61t, 79t, 114t, 120t, 121(n4) Chakravarty, S. 146(n5) change in mean (CHIM) 127, 135–6, 136t, 137, 140–1 Chauvet, L. 58 Chen, S. 123, 126 Chevalier, A., 174 child demographics 197t child mortality (under-five years old) 1, 6–7f, 8–10, 22, 28–34, 43, 44, 46, 49, 51–2(n9–10), 57, 63–85, 87(n14), 104, 106t, 109–10t, 110, 112, 130, 131t, 136t, 144, 210, 211 ’central variable’ 16 de-trended 159, 162f definition 51(n9) different estimates 28, 29t, 51(n9) income elasticity 152 individual data versus state-level average 164 MDG: annual rate of decline required to meet 150, 169(n1)

231

MDG: feasibility of achieving 155–6 predictors 155–6, 157 regression-based estimates 28 rural versus urban 32–3, 51–2(n10) subregional and country data 29, 30t, 31f trends 29, 32 unconditional income elasticity 151 vulnerable group 169 water supply 33, 34t White and Blöndal’s projections 29t see also infant mortality child mortality: links with economic growth (implications for achieving MDG in India) 15, 17, 150–73 alternative sets of control variables 165t assumptions 151, 156 data deficiencies 154, 156, 170(n4–5) descriptive statistics 157–60, 161–2f differences in income elasticity across states 165–6, 170(n10) dynamics 160 econometric model 150, 160–4, 170(n8) economic growth: distributional impact 153 economic growth: effect on child mortality 151–3, 169–70(n2–4) empiricism 167 further research required 153 growth elasticities and fixed effects by state 166t income elasticity (post-reform era) 166–7, 168t, 170(n11) key relationship 159 literature 153–6, 163, 170(n3–7) ’no evident relation of growth and mortality’ 159 panel data 150, 152, 154, 156–7, 159, 165 policy implications 169 research: feasibility of attaining MDG for mortality 155–6 research: impact of economic growth on childhood mortality 153–5, 170(n5–7) results 164–8, 170(n9–12)

232 Index child mortality: links with economic growth – continued simulation to MDG target 167–8, 171(n12) standard deviations 158t state fixed effects 151 systems estimator 160–2 trend variation (captured) 163 website 159n year effects 151 child mortality rate 128, 147(n15) child nutrition 170(n2) child survival rates 132, 133–4t, 140, 140f, 147(n15) childbirth deliveries outside the home 154 delivery complications 152, 169–70(n2) children aged 12 in de facto population (household size) 177t birth-order 157 first-born 169 health versus educational attainment 193 not in school 35, 37f, 177t, 206(n1) not in school: White and Blöndal’s projections 36t not in school: projections (2015) 38t premature death 44 rural 183 ’street children’ 22, 45 stunted 128, 130, 131n, 133t, 134n, 136t, 136, 137f, 144, 146(n11) underweight 9, 39t, 39–40 Chile 114t, 117t China 50, 51(n10), 103, 106t, 116t, 119t predicted to meet MDG poverty target 25 western provinces versus eastern seaboard 51–2(n10) Christians 178t, 187t, 190t, 202t, 204t ’chrs’ (variable) 64, 68t, 70–1t, 74, 86t CIA 86t civil society xiv, 2 Clark, J. 167 Clemens, M., et al. (2004) 86(n4), 88 Guillaumont, P. 88 Radelet, S. 88

Clements, B., et al. (2003) 227(n3), 227 Bhattacharya, R. 227 Nguyen, T. Q. 227 climate 58, 152, 163 Collier, P. 58 Colombia 61t, 79t, 118t Comoros 61t, 79t, 81, 120t, 121(n5) computable general equilibrium (CGE) models 21, 49, 50t conflict/conflict zones 25, 45 Congo Democratic Republic (Congo-Kinshasa) 120t Congo Republic (Congo-Brazzaville) 114t, 120t Congress Party 167 constant time trend projections 46–7, 52(n18–20) consumption 129, 145–6(n2), 180t, 181 rural versus urban 193t, 193–4 consumption goods 86(n2) contraception 69 corruption 106t, 163 Costa Rica 114t, 117t Côte d’Ivoire 61t, 79t, 119t, 121(n5) countries 28, 45, 50, 59 number expected to reach MDGs 25, 26t, 29, 30t, 32 covariance matrix 67 credit markets 182 cross-country regressions 16, 23, 47, 90–122 Cuba 117t culture 60, 64, 163 Cutler, D., et al. (2006) 151, 171 Deaton, A. 171 Lleras-Muney, A. 171 Cyprus 117t Czech Republic 32 Dalgaard, C., et al. (2004) 58, 88 Hansen, H. 88 Tarp, F. 88 data administrative 176 cross-country 16, 151, 164 cross-sectional 181 historical 46 national-level (drawbacks) 16 sub-national 56, 59 time-series 170(n4), 184 see also household surveys data sets 57–8

Index Datt, G. 163, 172, 214 death/’proximate’ versus ’underlying’ causes, 169–70(n2) Deaton, A. 146(n2), 151, 153, 163, 170(n9), 171 debt 10, 18, 18(n2) domestic and external 227(n3) debt burden (Indian states) 208–28 impact on economic growth 210–22 debt elasticity of poverty 217 debt forgiveness 13 debt overhang 210–11, 227(n3) debt ratio 214–17, 219t, 222–6 debt relief 4t debt trap 208, 227(n2) deciles 132, 133–4t, 146(n6) decreasing returns to labour 86(n2) decreasing returns to scale 57 deforestation 10, 11f Dehejia, R. 153 DeLong, J. B. 167 demand 211 Demery, L. 46, 48, 49, 51(n5) demographic accounting identities 46 demographic change 156 Demographic and Health Survey (DHS) data 22, 32, 51(n10) Bolivia 127–9, 131n demographic models 46 demographic transition 34, 45 demographic variables 163, 165 demographics 129, 157, 183, 186t, 189–91t, 192 Denmark: Royal Ministry of Foreign Affairs xiii De Onis, M., et al. (2004) 39, 52(n18) Blossner, M. 53 Borghi, E. 53 Frongillo, E. A. 53 Morris, R. 53 nutrition outcomes 39t, 39–40, 52(n18), 53 Deolalikar, A. B. 154, 170(n6–7), 176, 179, 206(n3) developed countries 3, 11f, 38t, 43, 18(n2) ’high-income countries’ 38t times of depression 211 see also OECD developing countries xviii, 5, 8, 9, 10, 11f, 15, 18, 18(n2), 25, 34, 39t, 41t, 43, 56, 58, 85, 87(n9), 95t, 151–2, 154, 163, 176, 208

233

’highly-indebted poor countries (HIPCs)’ 4t, 211 ’low-income countries’ 45, 93 ’off track’ 20 relevance of Keynesianism 211 small islands 3 development 3, 113 global challenges 2 socioeconomic 44 diarrhoea 111, 152, 169–70(n2) disabilities 45 discrimination 196 distribution 48, 51 distributional issues 48 Dixon, J. 86t Djibouti 119t Dollar, D. 52–3(n23), 58 Dominica 118t Dominican Republic 61t, 79t, 118t donor agencies/reports 86(n1) diphtheria, pertussis and tetanus (DPT) 131n, 147(n16) Drèze, J. 175 drugs (medicines) 4t Duclos, J. 126 durables index 178t, 185–6t, 188, 189t, 200–1t, 203–4t Dutta, I. xiv East Africa 25, 26t, 30t, 36t East Asia 39t, 40f, 41t, 41f, 90 ’East Asia and Pacific’ 24t, 24f, 26t, 29t, 33t, 33, 34–5t, 38t ’Eastern Asia’ 26t, 28, 30t, 36t Easterly, W. 64, 71 Eastern Europe 26t, 30t, 36t ’Eastern Europe and Central Asia’ 38t ’Eastern Europe and CIS’ 24t econometrics xv, 57, 67 economic cycle 170(n5) economic development 111 economic growth 16, 17, 20–1, 22, 28, 49, 52–3(n23), 113, 123, 146(n5), 175, 209, 211 cyclical nature 51(n7) distribution-neutral 48 distributional impact 123, 152 effects on health ’more uncertain’ 153 hampered by debt burden 210–12 impact on childhood mortality 153–5, 170(n5–7)

234 Index economic growth – continued links with childhood mortality (implications for achieving MDGs in India) 150–73 literature 58–9, 87(n8) per capita 58 performance 28 recent performance 51(n7) economic growth rate 94, 104, 112 economic liberalization (India) 150, 166–7 economic performance 35 Economic Survey of Delhi 2001–2 213 economic theory 211 economics terminology 52(n19) economies of scale 74, 93 Ecuador 118t education 14, 17, 20, 28, 34–8, 52(n11–12), 53(n25), 57–61, 65, 86(n2), 87(n7), 124, 127, 130, 131t, 144–5, 146(n5, n8), 152, 157, 164, 211 access to school 182 attitudes (historically determined) 183 curricula 199 effect on longevity 153 ’elementary’ (lower and upper primary) 206(n2) female 46, 48, 69 gender dimension 35, 38t gender equality 210 gender gap 129, 133t, 175, 199 India 17–18, 174–207 literacy 36, 38t, 52(n12) ’lower primary’ 176, 206(n2) net enrolment rates 2015 (naïve projections) 37f numbers out of school (2015 projection) 36t, 38t ’partly a state subject’ (India) 176 progress versus initial distribution 124 ratio of women to men 136t, 137, 139f secondary 18(n1) subregional primary enrolment rates (NER, 2015 projection) 36t summary 36 ’upper primary’ or ’middle school’ 206(n2) education programmes 208

educational attainment 56, 62–84 adult 181, 183, 197t female adult 178t, 184, 186t, 189t, 191, 201t, 204t higher castes versus Muslims 175 highest-educated adult in household 177–8t, 184, 186t, 188, 189t, 191, 201t, 204t inequalities (India) 175 inter-generational correlation 174 maternal 156, 169 parental 157, 163, 164, 174 paternal 165 upper limit 127 see also years of schooling educational status 55, 147(n14) efficiency 210 Egypt 32, 61t, 79t, 114t, 118t El Salvador 118t elasticities 22, 23, 25, 27f, 47, 49–50, 53(n24), 96, 150–3, 164–7, 168t, 170(n6), 181, 196, 199 ’elasticity of debt ratio on poverty’ 216 elasticity of mortality with respect to income 154, 160, 165t, 170(n6, n8) income elasticity of demand 93 outcome-income elasticity 20–1 electoral competition 169 electricity 178t, 179, 184, 185–6t, 189t, 200t, 203t electricity consumption 212, 214, 218 electricity supply 61, 87(n11), 156 electrification 179t, 181, 187t, 190t, 205t elementary schools 179, 193t, 194 emerging markets 210 employment 130, 131t, 211, 214 decent and productive 4t women 137, 139f, 140, 144, 178t, 184 environmental catastrophe 45 environmental Kuznets curve (EKC) 93 environmental sustainability xviii, 1 equations reduced form 45, 49, 52(n16) Equatorial Guinea 119t Eritrea 87(n10), 91, 114t, 120t

Index Ethiopia 49, 61t, 79t, 91, 116t, 120t, 121(n4) ethnic minorities 44 ethnicity 157, 163, 182, 197t ethno-linguistic fractionalization index (variable, ’ethno’) 64, 68t, 69, 70–1t, 86t Europe and Central Asia 24f, 26t, 29t, 33–5t, 38t ‘exit’ option 112 export revenues 18(n2) export access 4t factor productivity 87(n12) family 63, 91, 199 famine 91, 193 Feeny, S. 89 Fernandez, R. 57 fertility (human) 21, 56–7, 86(n2), 129, 156, 159, 163, 188t variable (’fer’) 62–84, 87(n14) Feyzioglu, T., et al. (1998) 59, 88 Swaroop, V. 88 Zhu, M. 88 Fielding, D. xiv Fiji 119t Filmer, D. 129, 163 Finland 114t, 117t Finland: Ministry for Foreign Affairs xiii fiscal deficit distinguished from ’debt’ 218 fiscal reform 152, 154, 210, 212 Flegg, A. 164 food availability 41, 42 food consumption 41 per capita (1964–2015) 41t food subsidies 193 France as colonial power (variable) 64, 68t, 70–1t, 73, 86t Franco-Rodriguez, S., et al. (1998) 59, 88 McGillivray, M. 88 Morrissey, O. 88 Frongillo, E. A. 53 Fuchs, V. R. 151 Gabon 61t, 79t, 82, 114t, 119t, 121(n5) Gambia 119t Gang, I. N. 59 gender 157, 163, 182–3, 188–9t, 193t, 197t

235

primary education 35, 38t see also women gender equality xviii, 1, 130, 131t, 136t, 137, 140, 210 gender gap education 175, 199 primary school completion 192 ’tends to increase with age’ (education) 188 generalized least squares (GLS) xvi, 170(n5) generalized method of moment (GMM) xvi, 170(n5) Geneva 145 geography 64 Georgia 116–17t Gini coefficients 51(n6), 146(n6, n8), 194 Gini index 106t, 107t, 108, 109t, 110t, 165t girls 17, 38t, 45, 51(n3), 169, 175–7, 178t, 182–3, 185–8t, 188–94, 196–9, 200–5t crimes (kidnappings) against 181 not in school 177t Girma, S. 88 Global Economic Prospects (World Bank) 21, 23, 50t Global Monitoring Report (World Bank, 2005) 23, 28, 46, 54, 92, 122 Goldin, I. 93 Gomanee, K., et al. (2005a) 60, 88 Girma, S. 88 Morrissey, O. 88 Gomanee, K., et al. (2005b) 60, 88 Morrissey, O. 88 Mosley, P. 88 Verschoor, A. 88 good governance 4t, 16 government debt distinguished from ’fiscal deficit’ 218 implications (Indian states) 208–28 ratio to GSDP 213 government expenditure see public expenditure Government of India: Department of Education 214 Government of India: website 227(n1) Grenada 117t Grimm, M. 146(n9) Grimm, M., et al. (2002) 146(n4), 148 Guénard, C. 148 Mesplé-Somps, S. 148

236 Index gross domestic product (GDP) 154, 212 de-trended 159, 162f explanatory variable 92–5, 96–8, 99f, 100t, 100–1f, 104, 106–7t, 108, 109t, 110, 112, 121(n6) growth 52(n22) link with access to water and sanitation 16, 93 per capita 16 PPP 47 gross national income (GNI) 5 gross state domestic product (GSDP) per capita 214–17 ratio of debt to 213 see also net state domestic product Grosse, M. xiv, 147(n18), 149 Grosse, M., et al. (2005) 129, 148 Klasen, S. 148 Spatz, J. 148 Grosse, M., et al. (2008) 124, 126, 127, 145(n1), 146(n8) Harttgen, K. 148 Klasen, S. 148 growth incidence curve (GIC) (Ravallion and Chen) 123, 126, 132, 135f growth rate in mean (GRIM) 126, 127, 135–6, 136t, 140, 148(n21) growth-based estimates 22 GTZ (Deutsche Gesellschaft für Technische Zusammenarbeit) xv Guatemala 61t, 79t, 105t, 114t, 118t Guénard, C. 148 Guillaumont, P. 58, 88 Guinea 61t, 79t, 116t, 120t Guinea-Bissau 116t, 120t Gujarat 158t, 159f, 160t, 161f, 166t. 180t, 220t, 221–5 Guyana 118t Haddad, L., nutrition outcomes 39t, 39–40, 49 Haiti 61t, 79t, 105t, 114t, 119t Ham, J. 182 Hamilton, K. (1996) 86t Hanmer, L. 24t, 28, 29t, 47, 49, 51(n4) Hanmer, L., et al. (1997a, 1997b) 23, 53 Jong, N. de 53 Kurian, R. 53 Mooij, J. 53

Hanmer, L., et al. (2000) 53(n24), 53 Healey, J. 53 Naschold, F. 53 Hanmer, L., et al. (2003) 22, 53 Lensink, R. 53 White, H. 53 Hansen, H. 58, 86(n4), 88 Harttgen, K. xv, 147(n18), 148 Haryana, 158t, 159f, 160t, 161f, 166t, 180t, 220t, 221–6 Hausman test 219t, 219, 227(n7) headcount poverty 136n, 213 Healey, J. 53 health 14, 56–65, 69, 86(n2), 87(n7), 106–7t, 112–13, 124, 129–30, 131t, 141, 146(n5), 174–5, 179, 199, 211–12 feedback to income through productivity effects 151–2 India 17, 150–73 upper limit 127 see also public expenditure health programmes 208 health services 28 health technology 158–9, 163, 169 health, wealth, wisdom 16, 55–89 healthcare 128–9 Heller, P. S. 59 Hemming, R. 210 Hermes, N. 89 Hesslebarth, S. 122 heteroscedasticity 165t, 183, 193n Hill, A. 164 Hill, K. 167 Himachal Pradesh 220t, 221–3, 224–5t Hindus 195 history 49, 64, 163 HIV/AIDS xviii, 1, 5, 6–7f, 9, 20, 28, 42–3, 46, 49, 52(n14–15) HNP (health, nutrition, poverty) World Bank data set 60, 63 Hoare, R. 86t Hoddinott, J. 206 Hoeffler, A. 58 Honduras 106t, 118t house (pucca) 178–9t, 185t, 189t, 200t, 203t house flooring 61–2 62t household assets 60, 189t, 191t, 192, 197t household consumption 180t, 181 household data 23 household demographics 188t, 197t

Index household living standards 163–4, 179 household mobility 127 household size 73, 184, 189t, 191–2 females/males 178t, 186t, 190t, 201t, 204t household surveys 47, 52(n22), 56, 60–4, 87(n10), 127–8, 146(n10), 175–6 households 93, 131n, 148(n20), 152, 183 Bolivian 124–5 child survival rates 147(n15) children aged under five 192 female head 178t, 184, 186t, 190t, 191, 201t, 204t income per capita (real) 134n, 136n principal female working, 178t, 184, 186t, 190t, 191, 197t, 201t, 204t rural 164 socio-economic status 183 variables 194 Hudson, J. 58, 89 human capital 57, 86(n2), 144, 174, 182, 214 human development 16, 111, 175 Human Development and Capability Association xv human development index (HDI) 60, 87(n9), 97–9, 100t, 100–1f, 106t access to water and sanitation 97–8, 100t Human Development Report: China (1997) 51(n10) Human Development Report (UNDP) 51(n2), 52(n21) human rights xiv, 14 Hungary 114t hunger xviii, 4t, 6–7f, 8, 9, 10f, 14, 15, 47, 136t, 210 ICT 4t IMF 211 IMF Economic Forum: Health, Wealth, Welfare (2004) 86(n3) IMF website 86(n3) immunization 34, 43, 154 income 28, 46, 50, 58–9, 93, 129, 145–6(n2), 147(n18), 147–8(n20), 151–2, 170(n6) absolute change 135f aggregate 152, 160 initial distribution 124

237

per capita 48, 57, 86(n2), 87(n8), 180t, 192, 193t, 202t, 205t, 212, 218 private 152 ’income’ (’real net state domestic product per capita’ variable) 157–68 income distribution 58, 132, 144 income elasticity of demand 93 income growth 21, 126 income inequality 22, 130 income poverty xviii, 8, 9, 10f, 20, 22, 23–8, 51(n4–8), 123, 124, 154, 170(n3) calculation 52(n17) estimates 47, 52(n22) forecasts 24t, 51(n7) percentage of population 26t, 27f projections 47 reduction 44 rural versus urban 25, 26t, 51(n8) sub-regional estimates 25, 26t, 27f income shocks 176, 199 income-based models 50t income-based projections 47–8, 52–3(n21–4) India viii, 9, 35, 44, 61t, 79t, 91, 103, 105t, 119t, 121(n5) childhood mortality and economic growth (implications for achievement of MDGs) 17, 150–73 country-specific analysis (water) 92, 121 difficulty of achieving MDGs 15 effect of recessions on health 153–4 focus on 15 impact of growth on poverty (literature) 170(n3) population, poverty, child mortality (share of world total) 150 post-reform era (1981–98) 150–1 predicted to meet MDG poverty target 25 primary schooling 17–18, 174–207 rural 184 state government debt versus MDG income poverty target 18, 208–28 trends in under-5 mortality and state-income (population-weighted averages) 159, 162f

238 Index Indian constitution 209, 227(n1) concurrent list 176 Indian National Family Health Survey (INFHS) xvi, 17 Indian States 160, 163–7, 169, 171(n12), 182–3, 195, 198t, 199 changes in under-5 mortality and income (1970–94) 158, 160t debt and poverty MDGs 212–13 dummy variable (education) 192 income 150, 183, 184, 199 level and change of under-5 mortality and income (1970–98) 157, 158t poor 181 poor versus non-poor 156 real income per capita 166–7 relationship of under-5 mortality and state income (quadratic fit) 159, 161f trends in real log income (1970–98) 156, 161f trends in under-5 mortality 158, 159f variables 194, 202t, 205t Indian States: government debt (implications for achieving poverty MDG) 15, 18, 208–28 assumptions 208 data 213–14, 227(n4–5) data deficiencies 213, 214, 226 debt and MDGs 210–12, 227(n3) empiricism 210, 212, 221, 227(n3) fixed-effect regression 212, 216t, 216–17 further research required 226–7 impact of debt on achieving poverty MDG 225–6 justification for state-level analysis 209 linear interpolation 214 literature 209–10, 210–12, 213, 227(n5) log-linear models 221, 223 MDG: 2007 and 2015 217–23, 227(n6–7) methodology 212–13 panel data 212, 214 policy implications 215, 226 predicted values 219–23 p-value Breusch–Pagan 215t, 215 R2 215t, 215, 219t random-effect regression 212–13, 214–16, 217, 218–19

results and analysis 214–23, 227(n6–7) robust standard errors 215t, 216t simulation 223–6 ’simultaneity bias’ issue 217–19, 227(n6–7) state-specific effects 217 website 227(n4) Indonesia xiv, 25, 32, 61t, 79t, 81, 103, 116t, 118t industrialization 218 inequality 28, 41–2, 47–8, 53(n24), 97, 104, 123, 126–7, 163, 174, 184 explanatory variable 105 initial 23, 51(n6) 90:10 ratio 130, 132, 133–4t, 134n inequality measures 146(n6) infant mortality 28–9, 30t, 32f, 32–4, 44, 51(n9), 58–9, 130, 131t, 136t, 144, 147(n15), 154, 156, 167 definition 51(n9) microdata 163 see also child mortality infant mortality rate 60, 63, 128 infant survival rate 133–4t, 140, 141f infectious disease 1, 93, 152, 170(n2) inflation 163, 164, 211 infrastructure 211–12 Institute of Social Studies (ISS) 23, 51(n4–5) institutional mechanisms 91, 112–13 institutions 58, 101, 163, 209 ’inter-generational multiplier’ 53(n25) interdependencies xviii, 2, 16–17, 55, 76, 90, 92, 94, 110–11, 124–5 interest rates 211 International Development Targets 15, 20 international dollars 97n International Food Policy Research Institute (IFPRI, USA) xvi, 39, 40, 49, 52(n13) International Institute for Population Sciences (IIPS, India) xvi, 172 and ORC Macro 156 internet users 12f, 18(n2) interviews 156, 176 investment 210, 211–12 public (education) 174 Iran 117t Iraq 118t Israel 114t, 117t

Index Jacoby, H. 176 Jalan, J. 152 Jamaica 118t Jamison, D. 86(n3) Jammu 180t Japan 114t, 117t Jolly, R. 92 Jong, N. de 53 Jordan 61t, 79t, 117t Kakwani, N. 164 Kanbur, R. 145 Karnataka 92, 158t, 159f, 160t, 161f, 165, 166t, 180t, 220t, 222t, 224–5t, 226 Kazakhstan 87(n10), 118t Kenya 61t, 79t, 116t, 119t Kerala 157, 158t, 159f, 158, 159f, 160t, 161f, 165, 166t, 180t, 220t, 222t, 224–5t, 226 Keynesian approach 211 Khan, H. A. 59 Kiel 145 King, E. 164 Kinsey, B. 206 Klasen, S. xv, 125, 127, 145(n1), 146(n4), 147(n18), 148–9 pro-poor growth typology 125–6 Klasen, S., et al. (2007) 129–30, 147(n17, n19), 148(n22), 149 Grosse, M. 149 Lay, J. 149 Spatz, J. 149 Thiele, R. 149 Wiebelt, M. 149 Knowles, J. 182 Kochar, K. 209, 212 Korea (South) 118t Kosack, S. 60 Kraay, A. 52–3(n23) Krain, M. 86t Kurian, N. J. 210 Kurian, R. 53 Kuznets effect 93, 98, 108, 113 Kyrgyz Republic/Kyrgyzstan 87(n10), 95, 116–17t La Porta, R., et al. (1998) 86t, 88 Lopez-de-Silanes, F. 88 Shleifer, A. 88 Vishny, R. 88 labour (opportunity cost) 86(n2) labour demand 182

239

labour productivity 57 labour supply (women) 154 Lahiri, A. 212 land (agricultural) 87(n12) landlocked countries 3 ’coast’ (variable) 64, 68t, 69, 70–1t, 75 landowners 178t, 185–6t, 189t, 200–1t, 203t Lao PDR 91, 116t, 119t, 121(n4) Latin America xiv–xv, 28, 129 ’Central America’ 25, 26t, 30t, 36t ’South America’ 25, 26t, 28, 30t, 36t Latin America and Caribbean (LAC) 24, 26t, 29t, 29, 32–5, 38–41, 43t, 48 Lay, J. 149 least-developed countries (LDCs) 3, 10, 12f least-squares dummy variables method (within-group) 161, 163 Lebanon 114t, 117t Lensink, R. 53, 89 Lesotho 115t, 119t Levine, R. 64, 71 Liberia 120t Libyan Arab Jamahiriya 95, 117t life expectancy 43, 59, 87(n9), 97, 129, 154 rich countries versus poor countries 151–2 literacy 22–3, 36, 38t, 49, 52(n12), 59, 87(n9), 129, 130, 131t, 137, 138f, 146–7(n13–14), 199, 217 female versus male 180t, 202t, 205t ’illiteracy’ 175, 180t, 184, 192, 193t, 202t, 205t literacy rate 128, 131n, 133t, 136t, 214 livestock-owner (variable) 178t, 185–6t, 189t, 200–1t, 203–4t living standards 5, 14, 85, 145(n2), 154, 163 gaps 3 rural versus urban 194 Lleras-Muney, A. 153, 171, 174 loans 5 local governments 112 log-linear equations 69 log-linear form 64 logistic regression equations 87(n13) Lopez-de-Silanes, F. 88

240 Index macroeconomic imbalances 210 macroeconomic policy 212 Madagascar 61t, 80t, 106t, 116t, 119t Madhya Pradesh 156, 158t, 159f, 160t, 161f, 166t, 180t, 206(n1), 220t, 222t, 223, 224–5t Maharashtra 157, 158t, 159f, 160t, 161f, 165, 166t, 180t, 220t, 222t, 222, 224–5t malaria 111, 112, 152, 169(n2) Malawi 61t, 80t, 92, 105t, 116t, 119t Malaysia 114t Maldives 118t ’malfal’ (fraction of population at risk from malaria) (variable) 64, 68t, 69, 71t, 75, 86t Mali 61t, 80t, 116t, 120t malnourishment 107t, 108 malnutrition 41–2, 52(n18) Maquette for MDG Simulations (MAMS, World Bank) 21, 49, 52(n16) market access 12f, 18(n2) marriage 148(n22) maternal health xviii, 136t, 141, 152, 170(n2), 211 maternal healthcare 130, 131t maternal mortality 4t, 22, 49, 111, 112 Mauritania 61t, 80t, 116t, 119t Mauritius 117t Mavrotas, G. 87 maximum likelihood 182 McArthur, J. 86t McGillivray, M. xv, 86(n6), 87, 88, 120, 169, 206, 227 McGillivray, M., et al. (2006) 86(n4), 89 Feeny, S. 89 Hermes, N. 89 Lensink, R. 89 measles 131n, 147(n16), 169(n2) measuring pro-poor growth in progress towards non-income MDGs 15, 17, 123–49 assumptions 125, 147(n18) comparability issue 128 concept of pro-poor progress 125–6, 145–6(n2–6) ’conditional’ versus ’unconditional’ assessments 124 ’crucial question’ 123 data 124, 127–9, 146(n9–18)

data deficiencies 125, 128, 129, 131n, 146(n10, n12–13) descriptive statistics 129–32, 133–4t, 147–8(n19–20) empiricism 125, 127, 135 further research required 145, 147(n18–19) government policies (Bolivia) 144–5, 148(n22) inequality (90:10 ratio) 130, 132, 133–4t, 134n interdependencies 124, 125 literature 123, 125, 146(n4–5) methodology 126–7, 129, 144, 146(n7–9) ’new insights’ 144 pro-poor progress towards non-income MDGs 132, 135–43, 148(n21) results 129–43, 147–8(n19–21) stochastic error term 129, 148(n18) terminology 126 theory 125 medical progress 152 Melanesia 30t, 36t men 146–7(n12–14), 147–8(n20), 195 see also gender Mensbrugghe, D. van der 53 Mesplé-Somps, S. 148 Mexico 118t microdata 154, 157, 163–5 microlevel analysis 112, 113 Micronesia 30t, 36t Middle East and North Africa 24t, 24f, 26t, 29t, 33t, 34t, 35t, 35, 36, 38t, 41t Miguel, C. 122 Millennium Declaration (2000) 2–3, 14, 121(n1) Millennium Development Goals (MDGs) aid 55–89 childhood mortality and economic growth (India) 150–73 China versus rest-of-world 50 criticisms 2 debt and 210–12, 227(n3) ’dire predictions’ (assessed) 15–16, 20–54 empiricism xviii failure to achieve targets (projected) 44 hardest to achieve in SSA 9 India 150–73

Index interdependencies 55 lack of commitment to quantitative ODA targets 5 non-income 15, 17, 123–49 poverty (implications of debt burden of Indian states) 208–28 primary schooling (India) 174–207 target variables: core determinants xviii targets 3–5, 14, 16, 90, 128–9 water and sanitation 90–122 Millennium Development Goals: MDG1 to MDG8 MDG1 (poverty and hunger) 8, 9, 10f, 15, 17, 123, 125, 128; progress towards 6f, 7f; targets 3, 4t MDGs1–7 3, 128 MDG2 (primary education) 3, 4t, 6, 7f, 8, 10f, 18(n1), 128–9, 137, 138f MDGs2–7 123 MDG3 (gender equality) 4t, 128–9, 137, 139f MDG4 (child mortality) 4t, 6, 7f, 8–9, 10f, 10, 15, 16, 17, 128, 129 MDG5 (maternal health) 4t, 128, 129 MDG6 (diseases) 4t, 6, 7f, 8, 129 MDG7 (environment) 3, 4t, 10, 11f, 17, 90, 129 MDG8 (global partnership for development) xviii, 1, 3, 4t, 5,10, 12f see also individual subject headings Millennium Development Goals: overview, progress, prospects 1–19 assumptions 8, 16 data deficiencies 6–7n, 8, 11–12n, 16 empiricism 2, 14, 17 further research required 15 interdependencies 2, 16–17 literature 15 outline 2–5 progress towards 5–14, 18(n1–2) Millennium Development Goals: projecting progress towards 15–16, 20–54 approaches to making projections 20–3, 51(n1–3)

241

assumptions 20, 21, 46, 48, 49, 50, 51, 51(n2), 52(n12, n20) base year (1990) 23, 28 baseline data 22 caveat 21 data deficiencies 21–2, 23, 25, 26n, 35, 43, 44, 47, 49, 51(n2), 52(n14–15, n21) determinants 20, 25, 45, 47, 50 ’dire projections’ assessed 20 education 34–8, 52(n11–12) HIV/AIDS 42–3, 52(n14–15) income poverty 23–8, 51(n4–8) level of disaggregation 21 mortality 28–34, 51–2(n9–10) nutrition 39–42, 52(n13) parameterization 20, 46, 50t, 50 policy simulations 50t projection methods 45–51, 51(n1), 52–3(n16–25) ’right policies’ 21 Millennium Indicators Database (UN) 22, 94, 121(n3) Millennium Project Task Force (MPTF) 111 gender 38t primary education 38t Millennium Project Task Force on Education (MPTFE) 34, 35t, 35 Millennium Summit (New York, 2000) xviii, 150 Mizoram 214 Moldova 32, 118t Mongolia 116t, 118t monitoring 92, 123 Monterrey Conference on Financing for Development (2002) 1 Montserrat 32 Mooij, J. 53 Morocco 61t, 80t, 118t Morris, R. 53 Morris, S. 171 Morrissey, O. 86(n4, n6), 88 mortality 16, 20–1, 145 Mosley, P. 58, 88 Mosley, P., et al. (2004) 60, 89 Hudson, J. 89 Verschoor, A. 89 Mozambique 61t, 80t, 116t, 120t Mukherjee, D. 146(n5) multi-equation models 21, 49, 50t, 52(n16), 53(n25) multivariate probit analysis 156, 179, 181

242 Index Munasinghe, M., et al. (2001) 93, 122 Miguel, C. 122 Sunkel, O. 122 Muslims 175, 178t, 187t, 188, 190t, 195, 201t, 204t ’mus/musl’ (variable) 64, 68t, 70–1t, 74, 86t mutlicollinearity 107n Myanmar 103, 105–6t, 118t ’naïve projections’ 20–2, 28, 29t, 34n, 35–9, 50t constant time trend 46–7, 52(n19–20) Namibia 61t, 80t, 91, 105t Naschold, F. 24t, 28, 29t, 47, 49, 53(n24), 53 national family health survey (NFHS, India 1998/9) 156–7, 163, 175–6, 181, 184, 185–91t, 198n, 200–5t national governments 18, 94, 112, 208 National Sample Survey (NSS), India 179, 181, 206(n1), 213 natural resource capital value (’natres’) (variable) 64, 68t, 70–1t, 87(n12), 86t Near East and North Africa 39t, 40f neonatal sepsis 170(n2) neonatal tetanus 169(n2) Nepal 61t, 80t, 119t, 121(n5) net enrolment rate (NER) 22, 34, 35, 36t, 37f, 51(n3), 52(n11), 54 net state domestic product 192, 193t per capita 180t, 202t, 205t real per capita (’income’ variable) 157–68 see also GSDP Netherlands 114t, 117t, 154 new public management 113 New Zealand 30t, 36t Newey–West standard errors 165t, 168n newspapers 63 Nguyen, T. Q. 227 Nicaragua 61t, 80t, 106t, 118t Niger 61t, 80t, 81, 116t, 120t Nigeria 61t, 80t, 103, 116t, 120t nomads 22, 44, 45 non-agricultural sector 128 non-governmental organizations (NGOs) 91, 94, 112 non-income growth incidence curve (NIGIC)

absolute 135f, 135, 136, 137–43f absolute smoothed 137–43f calculation 126, 146(n7) conditional 124–7, 132, 136, 137–43f, 148(n21) smoothed 137–43f unconditional 145(n1) non-income indicators xviii, 146(n8) ’non-income progress’ 126 ’non-income progress incidence curves’ 146(n7) normal cumulative density function 67 North Africa 30t, 36t, 175 Northern America 30t, 36t Northern Europe 26t, 30t, 36t Norway: Royal Ministry of Foreign Affairs xiii nutrition 20, 39–42, 49, 52(n13), 153, 170(n2), 175, 193 alternative estimates of under-nourishment 39t inequality 41–2 trends 40–1 see also HNP nutritional status 128 ’genetic noise’ 146(n11) Oaxaca, R. 195 obesity 45 Oceania 91 OECD xv, 2, 5, 13 OECD Development Assistance Committee (DAC) 10, 13f, 13 online database 64, 65 official development assistance (ODA) xvi, 1–2, 4t, 10, 12f, 13f, 13, 105t disbursements (ratio to GDP) 64 ’discom’ (disbursement-commitment) ratio 65, 67, 71t lack of commitment to quantitative targets 5 oil 18(n2) Oman 114t, 117t ordinary least squares two-stage (2SLS) estimation 218–19, 227(n6–7) three-stage (3SLS) 67 Orissa 158t, 159f, 160t, 161f, 166t, 180t, 220t, 221–6 government debt 209 Ortega, O. 155

Index Overseas Development Institute (ODI) 23, 51(n4–5) Ozler, B., et al. (1996) 157, 172 Datt, G. 172 Ravallion, M. 172 p-value 215t Pack, H. 59 Pack, J. R. 59 Pakistan 61t, 80t, 91, 103, 119t Palma 145 Palmer-Jones, R. 228 Panama 118t Panayotou, T. 93 Pande, R. 209 Papua New Guinea 119t Paraguay 61t, 80t, 91, 105–6t, 118t parents 45, 53(n25), 182 Parikh, A. 228 Paxson, C. 153, 154, 163, 170(n9) Pebley, A. 167 Peru 33, 51(n10), 61t, 80t, 118t Philippines 61t, 80t, 82, 118t physicians 49 polio 131n, 147(n16) political economy 183 ’part of explanation’ (health differences, India) 169 public service delivery 152 political stability 58 Polynesia 30t, 36t poorest people 55, 58–9, 60, 124, 132 ’not benefiting from increased government debt’ 214–15 ’not principal beneficiaries of aid’ 57 Popli, G. et al. (2005) 227(n5), 228 Palmer-Jones, R. 228 Parikh, A. 228 population (explanatory variable) 105, 109t population growth 49, 57, 86(n2) population growth rate 106t, 107t, 108, 109t explanatory variable 94 population size 15 population sub-groups 16, 55, 56, 60 Portugal 32 possessions (radio, television, refrigerator, car) 61, 62t poverty 58, 61, 81, 97, 129, 163–4 changing character 44–5 debt elasticity 217 elasticity 23

243

extreme 14, 130, 131t, 135, 136t, 140 ’extreme’ versus ’moderate’ 131n global (regional share, 1990–2015) 24f health expenditure elasticity of 221, 222 MDG: implications of government debt (Indian states) 208–28 moderate 136n, 140 multidimensional concept 124, 125, 146(n3, n5) non-income indicators 132 rural versus urban 180t, 194 see also HNP; income poverty poverty alleviation 193, 208 poverty disease trap 91 poverty elasticity 23, 51(n6) poverty eradication 3 poverty gap 128, 130, 131t, 193t poverty gap index 165t, 180t, 194, 202t, 205t poverty headcount 128, 130, 131t, 135, 212, 214, 215t, 215, 216t, 219–20t, 220, 222t poverty lines 47, 50, 131n, 135, 136n one dollar a day 3, 4t, 8, 15, 24t, 24f, 24, 28, 44 two dollars a day 15, 24, 28 poverty reduction xv, xviii, 5, 6–7f, 16, 18, 22, 123, 140, 146(n5), 151, 174, 194 Bolivia 147(n19) MDG targets 3, 4t progress 43–4 preventive medicines 128–9 primary education 1, 3, 10f, 16, 17, 18(n1), 46 attendance rate (India, 1998/9) 175 net enrolment rate (NER) 22, 34, 35, 36t, 37f, 51(n3) number of schools per thousand children 181 progress towards 6f, 7f universal (UPE) xviii, 36t, 43–4, 129, 130, 131t, 136t, 181, 210, 211 primary school age: definition 176 primary school attendance 185–8t, 200–3t determinants 188 predictors 184 primary school completion 189–91t, 203–5t

244 Index primary school completion– continued determinants 188 gender gap 192 primary school completion rates Bolivia 128, 130–2, 133t, 136t, 137, 138f India 175, 176, 177t, 177, 178–9t, 181–4, 188–99, 203–5t primary school enrolment female 28 gross rate 87(n9) rate (India) 176, 177t, 178–9t, 179, 181–8, 191–9, 200–3t ’last 5 per cent’ 35, 36, 45 primary school teachers 180t, 202t, 205t primary schooling (India): achieving the MDG 15, 17–18, 174–207 analytical approach 182 assumptions 181, 182, 196, 199, 206(n3) data and definitions 175–9, 180t, 181, 193n, 206(n2) data deficiencies 176, 177t, 182, 183, 184, 195 decomposition and simulation 194–9, 206(n5) empirical model 182–4, 185–91t, 210(n4) F-tests 188t, 188n, 191t further research required 199 initial conditions 183 linear probability model 194–9, 200–5t linear probability model of school attendance for children aged 6–11 200–3t linear probability model of school completion for children aged 12 years 203–5t literature 179–81, 206(n1, n3) means and standard deviations 177, 178–9t parameters 181, 184, 195–6, 199 policy implications 183, 192, 193 pooled model 196, 206(n5) probit estimates of school attendance (children 6–11) 185–8t probit estimates of school completion (children aged 12) 189–91t probit model 182–4, 185–91t, 194, 196

pseudo-R2 183, 188t, 191t puzzle 194 R2 203t, 205t reasons for Indian case-study 175 regression of state fixed assets on state-level variables 192–3, 193t results 184–94 sample size 176, 177t scenarios 179, 181 standard errors 183, 193t, 198n state fixed effects 183–4, 210(n4) t-tests 177, 178–9t, 203n Wald Chi2 188t, 191t z-statistics 188n Pritchett, L. 57, 129, 154, 163–4, 170(n5) private goods 93 private sector 4t, 91, 94, 112 pro-poor change (PPCH) 127, 135–7, 140–1 PPCH-CHIM 132 pro-poor growth xiv, xviii, 17, 48, 52–3(n23), 123–49, 153, 166, 167, 170(n3), 174 definition 125 Klasen’s typology 125–6 relative 125–6, 127, 137, 140, 141, 144, 146(n8) strong absolute 125, 126, 127, 135, 137, 141, 145 weak absolute 125, 126, 132, 136, 137, 140, 141, 144 pro-poor growth rate (PPGR) 126, 127, 135–6, 136t, 140, 147(n17), 148(n21) PPGR-GRIM 132 pro-poor progress 148(n21) concept 125–6, 145–6(n2–6) Probit form 64, 69 ’problem groups’ 35, 36 productivity 57, 86(n2), 93, 151–2, 174 projection methods 45–51, 51(n1), 52–3(n16–25) addition of explanatory variables 48–9 constant time trend: naïve projections 46–7, 50t, 52(n18–20) equations 45–7 income-based projections 47–8, 50t, 52–3(n21–4) level of analysis 49–50 model specification 45, 50t

Index multi-equation models 49, 51, 52(n16), 53(n25) parameterization 46, 50t, 50 projections of independent variables 45 summary 50t, 50–1 property rights 93 public expenditure/public spending 85, 152, 163, 170(n4), 215 aid-augmented 59 determinants 59 development 180t, 192, 193t, 193, 202t, 205t education 60, 87(n7), 176, 179, 180t, 181, 184, 192, 193t, 202t, 205t ’government expenditure’ 64–5, 214–17 health 60, 87(n7), 106t, 107t, 108–12, 151–2, 156, 164–5. 165t, 169, 170(n6), 192–3, 212, 214, 218–26 health (per capita) 215t, 215, 216t, 216–17 health, education, water 64–5 health, famine relief, food subsidies 192 Orissa 209 quality 212 social sector 112 targeted 124 twofold disaggregation 87(n7) public health 44 public sector 59, 91 public service delivery 163 Punjab (India) 158t, 158, 159f, 160t, 161f, 166t, 180t, 220t, 221–3, 224–5t pupil–teacher ratio 181 pure public goods 93 quality of life 113, 209 questionnaires 175–6 quintiles 146(n6) poorest 65, 75, 81, 85 richest 65, 75 R2 48, 97t, 98–104f, 105t, 107t, 108–11, 215t, 215, 219t Radelet, S. 88 rainfall 163 rainfall shocks 150, 159, 162 Rajasthan 156, 158t, 159f, 160t, 161f, 166t, 180t, 220t, 221, 222t, 224–5t, 226

245

Ramadas, K., et al. (2002) 52(n18), 53 Mensbrugghe, D. van der 53 Wodon, Q. 53 RAND 169 Raney, L. 164 Rangarajan, C. 209 Ravallion, M. 123, 126, 146(n4), 151–2, 163, 170(n3), 172, 214 recessions effect on health 153–4 recipient country policies 58 regional tabulations 51(n5) regions 45 Reher, D. 155 religion 157, 163, 175, 182–3, 187t, 190–1t, 197t, 202t, 204t Rens, T. van 57 reproductive health services 129 Research Committee on Development Economics 145 respiratory infections 152 rich groups 58 developing countries 60 rich–poor gap 56, 85, 127, 145 roads (pucca) 156, 177–9t, 181, 187t, 190t, 202t, 204–5t Rockefeller Foundation 51 Rogerson, R 57 Romania 116t, 118t Rosenzweig, M. 174 Roubini, N. 210 Ruhm, C. J. 153 Ruohonen, A. xiii rural areas distance to nearest town 177–8t, 187t, 190t, 202t, 204t India 176, 178t, 183–8, 188–91t, 193–6, 197t, 200–5t water and sanitation 91 rural development 211 rural infrastructure 187t, 190–1t rural–urban differences 22–3, 147(n18) Bolivia 129 child mortality 32–3, 51–2(n10), 157 consumption 180t, 193t, 193–4, 202t, 205t incidence of HIV/AIDS 43, 52(n15) income poverty 25, 26t, 51(n8), 89 living standards 194 poverty 44, 194 prices 163 water supply 33, 34t, 34

246 Index Russian Federation 43n, 114t, 117t Rwanda 61t, 80t, 116t, 120t s-curve 34 Sachs, J. 86t Sahn, D. 129 St Kitts & Nevis 117t Saint Lucia 117t sample selection 156 sanitation 4t 10, 11f, 16–17, 55, 61–84, 129, 152, 156, 178–9t, 185t, 189t, 200t, 203t, 211 MDG after-thought 121(n1) see also water and sanitation São Tomé and Príncipe 120t Saudi Arabia 114t savings 211 Schady, N. 154 scheduled castes/tribes (India) 165, 178t, 186t, 188, 190t, 192, 201t, 204t science and technology 175 Sen, A. 145, 146(n3–4), 175 Senegal 91, 116t, 119t separate room for cooking 178t, 184, 185–6t, 189t, 200–1t, 203t Serbia and Montenegro 118t Sevilla, J. 87 Shafik, N. 93, 96 Shleifer, A. 88 shocks 16, 45, 150, 159, 162–3, 176, 199 Shordt, K., et al. (2004) 92, 122 Brikke, F. 122 Hesslebarth, S. 122 Wijk, C. van 122 Shorrocks, A. xiii, xviii Sida 2015 project 23, 24t Siddiqui, S. 57 Sierra Leone 116t, 120t Simulations for Social Indicators and Poverty (SimSIP) xvii, 48–9 simultaneity bias 217–19 ’size’ (country surface area) (variable) 64, 68t, 69, 70–1t, 86t Skoufias, E. 176 Slovakia 117t slum-dwellers 3, 4t, 10, 11f Smith, L. C. nutrition outcomes 39t, 39–40, 49 smoking 45 social expenditures in total government spending (explanatory variable) 94

social indicators 48, 49 social sector government investment 18 Society for Study of Economic Inequality 145 Soest, A. van 167, 169, 171, 195 Solomon Islands 119t Somalia 120t, 121(n4) South Africa 61t, 80t, 118t South Asia 15, 17, 24t, 24f, 24, 26t, 28, 29t, 33–5t, 35, 36, 38–9t, 40f, 40, 41t, 41f, 90, 92, 129, 175, 209 difficulty in reaching MDG targets 9 lack of progress towards MDGs 5 South-Central Asia 26t, 30t, 36t South-East Asia 26t, 30t, 36t Southern Africa 26t, 30t, 35, 36t, 36 Southern Europe 26t, 30t, 36t Spatz, J. 148, 149 Sri Lanka 91, 95, 106t, 116t, 117t Srivastava, D. K. 209 standard deviation/standard errors 62t, 62, 64, 66t, 68, 69, 70–2t, 75, 76t, 77, 81, 82–3t, 87(n13), 158t, 177, 178–9t standard incidence analysis 124 Standing, H. 51(n10) Stata 206(n5) state, the 98 state fixed effects 183–4, 187–8t, 191t, 192–4 state role 93 statistical methodologies 57–8 Stifel, D. 129 stipend schemes 51(n3) structural adjustment 167, 168t, 170(n11) structural vulnerability 58 Strulik, H. 57 stunting (z-score) 128, 130, 131n, 133t, 134n, 136t, 136, 137f, 144, 146(n11) sub-Saharan Africa (SSA) xv, 15, 23–4, 26t, 28, 29t, 33t, 33–6, 38t, 39t, 40f, 40–2, 43t, 90–2, 110t, 129, 144 HIV 9 MDG4 9 MDGs ’hardest to achieve’ in 9 poverty 8 progress towards MDGs 7f ’unlikely to achieve MDGs’ 5, 14

Index water and sanitation 101–2, 103, 105t, 111 Subbarao, K. 164 subsidies 209 Sudan 114t, 119t Summer, A. 146(n5) Summers, L. H. 57, 154, 164, 170(n5) Sunkel, O. 122 supply and demand 182 Suriname 117t Swallow, A. xiii, 227 Swaroop, V. 88 Swaziland 119t Sweden 114t Swedish International Development Cooperation Agency (Sida) xiii Switzerland 114t, 117t synergies 16, 93, 99, 102t, 112 Syrian Arab Republic 32, 114t, 118t t-statistics 69, 70–1t, 97n, 165–6t, 168t Tajikistan 116t, 119t Tamil Nadu 158t, 159f, 160t, 161f, 165, 166t, 180t, 220t, 222t, 224–5t, 226 Tandon, A. 154, 155, 170(n9) Tanzania (United Republic) 61t, 80t, 105t, 116t, 119t Tarp, F. 58, 86(n4), 88 taxation 210 teachers/teaching staff 211 feminization (effect on primary completion rates) 194 male versus female (primary education) 193t, 194 primary school 193t technical assistance 65 technocracy 112 technology 10, 12f television sets (rural areas) 179t, 184, 187t, 191t, 195, 202t, 205t ’temp’ (temperature) (variable) 64, 68t, 69, 71t, 86t, 87(n12) terms of trade 163 tetanus toxoid immunization 156 Teulings, C. 57 Thailand 117t Theil inequality measure 146(n6) Thiele, R. 149 time 3, 17, 34, 45, 46–7, 50, 52(n18–20), 94–6, 98, 100–1f, 101, 132, 147(n18), 150–1, 154, 157–8, 160, 163–5, 166n, 168–9,

247

170(n8), 181–3, 193t, 194–6, 199, 212–18, 219t, 221 forecasts based on historical trends 20 opportunity cost 153 time-series analysis 47, 52(n21), 170(n4), 184 time trend 49, 50t Timor-Leste 120t Togo 61t, 80t, 81, 114t, 120t Tomes, N. 174 Tonga 117t Torres, S. xv trade 49 trading and financial system 4t Trinidad & Tobago 117t tropical diseases 29 tuberculosis 5, 6f, 9, 131n, 147(n16), 170(n2) Tunisia 32, 118t Turkey 87(n10–11), 103, 118t Turkmenistan 116t, 118t Twelfth Finance Commission of India 210, 213 Uganda 61t, 80t, 119t Ukraine 117t unconditional assessments 124, 145(n1) undernourishment 41 definition 39 undernutrition 146(n10), 152 United Nations system 2, 8, 13, 18(n1), 46, 146(n5) UN Charter 14 UN DESA: Population Division 52(n14) UN ’Education for All’ campaign (1978–) 1, 2, 3 UN FAO 39–42, 49, 54 UN General Assembly 208 UN ’Health for All’ campaign (1990-) 1, 2 UN Millennium Project Task Force on Water and Sanitation 91 UN Millennium Summit (2000) xviii, 1, 2, 175 UN Population Database (2004 Revision) 22 UN Second Millennium Summit (2005) 2 UN World Population Prospects (2004 revision) 28–32 UNAIDS 5, 42, 54

248 Index United Nations system – continued UNDP xiv-xv, 2, 51(n2), 92, 97 UNESCO xiv-xv UNFPA 42 UNICEF xiv UNIDO xv see also IMF; WHO; World Bank unemployment 174, 182 United Kingdom 153, 163, 170(n9) ’britain’ (UK as colonial power) (variable) 64, 68t, 70–1t, 73, 74, 75, 86t United Kingdom: Department for International Development xiii United Nations University: World Institute for Development Economics Research (UNU-WIDER, Helsinki) xiv-xv, 120, 169, 206 ’MDGs: Assessing and Forecasting Progress’ (research project) xiii, xviii project meeting (Helsinki, 19–20 August 2005) 120, 145, 227 United States of America 43n, 153, 163, 170(n9) universal human values 14 universities xiv–xv upper middle income countries 38t urban areas 23, 44 water and sanitation 91 urban population 110t, 111 urbanization 23, 45, 49, 51(n8) Uruguay 114t, 117t US Census Bureau: HIV/AIDS Surveillance Database 52(n15) US Virgin Islands 87(n10) Utrecht 154 Uttar Pradesh 156, 157, 158t, 159f, 160t, 161f, 165, 166t, 180t, 206(n1), 220t, 222t, 224–5t Uzbekistan 87(n10), 118t vaccinations 127, 131t, 132, 133–4t, 136t, 141, 142f, 147(n16) Vandemoortele, J. 14 Vanuatu 32, 119t variance inflation factor (VIVE) 107n Venezuela 118t Verschoor, A. 88, 89 Victora, C. G., et al. (2003) 167, 173 Vietnam/Viet Nam 25, 32, 61t, 80t, 116t, 119t village/s 176, 183, 190t, 195, 197t

bank 179t, 187t, 191t, 202t, 205t children not in school 206(n1) infrastructure 183 post office 179t, 187t, 191t, 202t, 205t school 179t, 182, 184, 187t, 188, 191t, 196, 202t, 205t Virmani, A. 167 Vishny, R. 88 ’voice’ option 112 Wald test 215t, 219t Walton, M. 46, 48, 49, 51(n5) war 45, 91 water 4t, 33, 34t, 34, 43, 55–6, 65, 87(n14), 129, 152, 178t, 185–6t, 189t, 200–1t, 203t, 211 conflict 90 contested resource 112–13 lack of access ’largely rural phenomenon by 2015’ 33 sources 90, 121(n2) water and sanitation xviii, 130, 131t, 132, 133–4t, 136t, 141, 143f, 144–5 see also sanitation water and sanitation: achieving the MDG 15, 16–17, 90–122 access to sanitation (forecasts for 2015) 117–20t access to water (forecasts for 2015) 114–16t access to water and sanitation: state of progress 95–6, 121(n4–5) analysis 95–111, 121(n4–7) collinearity diagnostic 107t, 109t cross-country comparisons 92, 94, 111, 113 data 92, 94–5, 121(n3) data deficiencies 94–5, 105 development agenda 91 endogeneity 94 equations 113 forecasting 108–10, 114–20t further issues 111–13 further research required 112–13 F-value 97t, 100t, 102t, 104–5t, 107t, 109t hypotheses/questions 92–4, 113 interdependencies 90, 92, 94, 110–11 lag effect 99 legacy 94, 100–8, 109t, 112

Index likelihood of achieving MDG 91–2, 110, 112, 140–20t mean and standard deviation 95t methodology 94, 108, 111, 113 multiple regression analysis 105, 107t parameters 86, 97t, 100t, 102t, 104t, 107t, 109t, 110t per capita GDP 96–8, 99f, 100t, 100–1f, 121(n6) policy 94, 100–8, 109t, 112 policy implications 94 progress (dependence on starting point) 100–8, 112 quintiles 96t R2 97t 102t, 104t R2 adjusted 105t, 107t, 108, 109t, 110t, 111 R2 linear 98–104f regional disparities 101–2 regional level 92 significance 110–11, 121(n7) simple linear regression models 96, 97t, 98, 100t standard error 107n summary statistics 95–6t synergy between water supply and sanitation 93, 99, 102t, 112 t-statistics 97t, 103–4 wellbeing xv, xviii, 3, 16, 56–7, 85, 112, 144 cross-country determinants 55 cross-country measure (’ass’) 61–83, 87(n12) dimensions 58–9, 86(n5) material 60 outcomes 59 West Africa 25, 26t, 29, 30t, 34, 36t West Bengal 158t, 159f, 160t, 161f, 165, 166t, 180t, 220t, 221–6 government debt 209, 227(n2) Western Asia 26t, 30t, 36t Western Europe 30t, 36t White, H. xv, 53(n23, n25), 53, 135 White and Blöndal income poverty estimates 25, 26t income poverty forecasts (for 2015) 24t, 25 projections 38t projections (primary enrolment rate, 2015) 35t, 35 WHO 8, 170(n2)

249

WHO-UNICEF (author) 90, 91, 92, 95, 111, 116n, 120n, 121(n2–3) Wiebelt, M. 149 Wijk, C. van 122 Winters, L. 93 Wodon, Q. 53, 126 Wolcott, S. 167 women 63, 86(n2), 127, 147–8(n20), 153, 163, 193t, 195, 199 ’better-off benefit more’ 140 crimes (kidnappings) against 181 empowerment xviii, 1 labour supply 154 literacy 128, 131n, 133t, 136t, 137 married 156 pregnant 156 wage employment 128, 130–2, 133–4t, 136t, 137, 139f, 140, 178t, 184 see also gender World Bank xiv, xv, 2, 49, 51(n5) author/data source 8, 21–3, 28, 46, 50t, 59, 60, 92–3, 96, 146(n4–5), 154, 155–6, 168, 170(n7, n10), 176, 179, 181, 206(n3) income poverty forecasts (for 2015) 24t poverty headcount threshold 3 World Bank SimSIP project 52(n18), 53 World Development Indicators (World Bank) 22, 52(n21) World Summit for Sustainable Development (Johannesburg, 2002) 121(n1) years of schooling 128, 181 gender equality 130, 131t, 132, 147–8(n20) see also educational attainment Yemen 61t, 80t, 81, 91, 119t youth 4t youth unemployment 10, 12f Yun, M.-S. 196 z-score (stunting) 128, 130, 131n, 133t, 134n, 136t, 136, 137f, 144, 146(n11) Zambia 61t, 80t, 116t, 119t Zamora, B. xv, 175, 188, 195, 207 Zhu, M. 88 Zimbabwe 61t, 80t, 119t