A FR IC A DEV EL OPM ENT FORUM All Hands on Deck Reducing Stunting through Multisectoral Efforts in Sub-Saharan Africa Emmanuel Skoufias, Katja Vinha, and Ryoko Sato All Hands on Deck All Hands on Deck Reducing Stunting through Multisectoral Efforts in Sub-Saharan Africa Emmanuel Skoufias, Katja Vinha, and Ryoko Sato A copublication of the Agence française de développement and the World Bank © 2019 International Bank for Reconstruction and Development / The World Bank 1818 H Street NW, Washington, DC 20433 Telephone: 202-473-1000; Internet: www.worldbank.org Some rights reserved 1 2 3 4 22 21 20 19 This work is a product of the staff of The World Bank with external contributions. The findings, interpretations, and conclusions expressed in this work do not necessarily reflect the views of The World Bank, its Board of Executive Directors, or the governments they represent, or the Agence française de développement. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Nothing herein shall constitute or be considered to be a limitation upon or waiver of the privileges and immunities of The World Bank, all of which are specifically reserved. Rights and Permissions This work is available under the Creative Commons Attribution 3.0 IGO license (CC BY 3.0 IGO) http://creative​ commons.org/licenses/by/3.0/igo. Under the Creative Commons Attribution license, you are free to copy, distrib- ute, transmit, and adapt this work, including for commercial purposes, under the following conditions: Attribution—Please cite the work as follows: Skoufias, Emmanuel, Katja Vinha, and Ryoko Sato. 2019. All Hands on Deck: Reducing Stunting through Multisectoral Efforts in Sub-Saharan Africa. Africa Development Forum series. Washington, DC: World Bank. doi:10.1596/978-1-4648-1396-2. License: Creative Commons Attribution CC BY 3.0 IGO Translations—If you create a translation of this work, please add the following disclaimer along with the ­attribution: This translation was not created by The World Bank and should not be considered an official World Bank translation. The World Bank shall not be liable for any content or error in this translation. Adaptations—If you create an adaptation of this work, please add the following disclaimer along with the attribu­ tion: This is an adaptation of an original work by The World Bank. Views and opinions expressed in the adaptation are the sole responsibility of the author or authors of the adaptation and are not endorsed by The World Bank. Third-party content—The World Bank does not necessarily own each component of the content contained within the work. The World Bank therefore does not warrant that the use of any third-party-owned individual component or part contained in the work will not infringe on the rights of those third parties. The risk of claims resulting from such infringement rests solely with you. If you wish to reuse a component of the work, it is your responsibility to determine whether permission is needed for that reuse and to obtain permission from the copyright owner. Examples of components can include, but are not limited to, tables, figures, or images. All queries on rights and licenses should be addressed to World Bank Publications, The World Bank Group, 1818 H Street NW, Washington, DC 20433, USA; e-mail: pubrights@worldbank.org. ISBN (paper): 978-1-4648-1396-2 ISBN (electronic): 978-1-4648-1397-9 DOI: 10.1596/978-1-4648-1396-2 SKU: 211396 Cover photo: ©Vincent Tremeau / World Bank. Beni, North Kivu region, Democratic Republic of Congo. Used with permission; further permission required for reuse. Cover design: Bill Pragluski, Critical Stages, LLC. Library of Congress Cataloging-in-Publication Data has been requested. Africa Development Forum Series The Africa Development Forum Series was created in 2009 to focus on issues of significant relevance to Sub-Saharan Africa’s social and economic develop- ment. Its aim is both to record the state of the art on a specific topic and to contribute to ongoing local, regional, and global policy debates. It is designed specifically to provide practitioners, scholars, and students with the most up-to- date research results while highlighting the promise, challenges, and opportuni- ties that exist on the continent. The series is sponsored by Agence française de développement and the World Bank. The manuscripts chosen for publication represent the highest quality in each institution and have been selected for their relevance to the development agenda. Working together with a shared sense of mission and interdisciplinary purpose, the two institutions are committed to a common search for new insights and new ways of analyzing the development realities of the Sub-Saharan Africa region. Advisory Committee Members Agence française de développement Gaël Giraud, Chief Economist Thomas Melonio, Executive Director, Innovations, Research and Knowledge Directorate Pierre Icard, Director, Head of Knowledge Department on Sustainable Development Sophie Chauvin, Head, Edition and Publication Division Hélène Djoufelkit, Deputy Head, Research Department World Bank Albert G. Zeufack, Chief Economist, Africa Region Markus P. Goldstein, Lead Economist, Africa Region v Sub-Saharan Africa CABO MAURITANIA VERDE NIGER MALI SUDAN ERITREA SENEGAL CHAD THE GAMBIA BURKINA FASO GUINEA BENIN GUINEA-BISSAU NIGERIA CÔTE GHANA ETHIOPIA SIERRA LEONE CENTRAL SOUTH D’IVOIRE SUDAN AFRICAN REPUBLIC SOMALIA LIBERIA TOGO CAMEROON EQUATORIAL GUINEA UGANDA KENYA REP. OF SÃO TOMÉ AND PRÍNCIPE GABON CONGO RWANDA DEMOCRATIC REPUBLIC BURUNDI OF CONGO TANZANIA SEYCHELLES COMOROS ANGOLA MALAWI Mayotte ZAMBIA (Fr.) ZIMBABWE MOZAMBIQUE MAURITIUS MADAGASCAR NAMIBIA BOTSWANA La Réunion (Fr.) ESWATINI LESOTHO SOUTH AFRICA IBRD 39088 | MAY 2019 Titles in the Africa Development Forum Series Africa’s Infrastructure: A Time for Transformation (2010) edited by Vivien Foster and Cecilia Briceño-Garmendia Gender Disparities in Africa’s Labor Market (2010) edited by Jorge Saba Arbache, Alexandre Kolev, and Ewa Filipiak Challenges for African Agriculture (2010) edited by Jean-Claude Deveze Contemporary Migration to South Africa: A Regional Development Issue (2011) edited by Aurelia Segatti and Loren Landau * Light Manufacturing in Africa: Targeted Policies to Enhance Private Investment and Create Jobs «L’industrie légère en Afrique: Politiques ciblées pour susciter l’investissement privé et créer des emplois» (2012) by Hinh T. Dinh, Vincent Palmade, Vandana Chandra, and Frances Cossar * Informal Sector in Francophone Africa: Firm Size, Productivity, and Institutions «Les entreprises informelles de l’Afrique de l’ouest francophone: Taille, produc- tivité et institutions» (2012) by Nancy Benjamin and Ahmadou Aly Mbaye * Financing Africa’s Cities: The Imperative of Local Investment «Financer les villes d’Afrique: L’enjeu de l’investissement local» (2012) by Thierry Paulais * Structural Transformation and Rural Change Revisited: Challenges for Late Developing Countries in a Globalizing World «Transformations rurales et dével­ oppement: Les défis du changement structurel dans un monde globalisé» (2012) by Bruno Losch, Sandrine Fréguin-Gresh, and Eric Thomas White The Political Economy of Decentralization in Sub-Saharan Africa: A New Imple- mentation Model (2013) edited by Bernard Dafflon and Thierry Madiès Empowering Women: Legal Rights and Economic Opportunities in Africa (2013) by Mary Hallward-Driemeier and Tazeen Hasan Enterprising Women: Expanding Economic Opportunities in Africa (2013) by Mary Hallward-Driemeier Urban Labor Markets in Sub-Saharan Africa (2013) edited by Philippe De Vreyer and François Roubaud Securing Africa’s Land for Shared Prosperity: A Program to Scale Up Reforms and Investments (2013) by Frank F. K. Byamugisha * Youth Employment in Sub-Saharan Africa «L’emploi des jeunes en Afrique subsaharienne» (2014) by Deon Filmer and Louise Fox vii viii   Titles in the Africa Development Forum Series Tourism in Africa: Harnessing Tourism for Growth and Improved Livelihoods (2014) by Iain Christie, Eneida Fernandes, Hannah Messerli, and Louise Twining-Ward * Safety Nets in Africa: Effective Mechanisms to Reach the Poor and Most Vulnerable «Les filets sociaux en Afrique: Méthodes efficaces pour cibler les populations pau- vres et vulnérables en Afrique» (2015) edited by Carlo del Ninno and Bradford Mills * Land Delivery Systems in West African Cities: The Example of Bamako, Mali «Le système d’approvisionnement en terres dans les villes d’Afrique de l’Ouest: L’exemple de Bamako» (2015) by Alain Durand-Lasserve, Maÿlis Durand- Lasserve, and Harris Selod Enhancing the Climate Resilience of Africa’s Infrastructure: The Power and Water Sectors (2015) edited by Raffaello Cervigni, Rikard Liden, James E. Neumann, and Kenneth M. Strzepek * Africa’s Demographic Transition: Dividend or Disaster? «La transition demographique de l’Afrique» (2015) edited by David Canning, Sangeeta Raja, and Abdo S. Yazbeck The Challenge of Fragility and Security in West Africa (2015) by Alexandre Marc, Neelam Verjee, and Stephen Mogaka Highways to Success or Byways to Waste: Estimating the Economic Benefits of Roads in Africa (2015) by Ali A. Rubaba, Federico Barra, Claudia Berg, Richard Damania, John Nash, and Jason Russ Confronting Drought in Africa’s Drylands: Opportunities for Enhancing Resilience (2016) edited by Raffaello Cervigni and Michael Morris * Reaping Richer Returns: Public Spending Priorities for African Agriculture Productivity Growth (2017) by Aparajita Goyal and John Nash Mining in Africa: Are Local Communities Better Off? (2017) by Punam Chuhan- Pole, Andrew L. Dabalen, and Bryan Christopher Land Realizing the Full Potential of Social Safety Nets in Africa (2018) edited by Kathleen Beegle, Aline Coudouel, and Emma Monsalve Facing Forward: Schooling for Learning in Africa (2018) by Sajitha Bashir, Marlaine Lockheed, Elizabeth Ninan, and Jee-Peng Tan The Skills Balancing Act in Sub-Saharan Africa: Investing in Skills for Productivity, Inclusivity, and Adaptability (2019) by Omar Arias, David K. Evans, and Indhira Santos All Hands on Deck: Reducing Stunting through Multisectoral Efforts in Sub- Saharan Africa (2019) by Emmanuel Skoufias, Katja Vinha, and Ryoko Sato * Available in French All books in the Africa Development Forum series are available for free at https://openknowledge.worldbank.org/handle/10986/2150. Contents Foreword xvii Acknowledgments xix About the Authors xxi Abbreviations xxiii Overview 1 Motivation 1 Design and Methods 5 Results 6 Policy Considerations 15 Attention to the Incentive Structure 19 Notes 20 References 22 1 Introduction and Motivation 25 Notes 31 References 31 2 Methodology and Data 33 The United Nations Children’s Fund Conceptual Framework 33 Caveats and the Contributions of This Study 37 Notes 42 References 43 3 Stunting in Sub-Saharan African Countries 45 Note 54 References 55 ix x  Contents 4 The Underlying Determinants of Nutrition 57 Measures of Food Security and Child Care 57 Measures of Water, Sanitation, and Hygiene 62 Measures of Health 64 Empirical Measures of the Underlying Determinants of Nutrition 66 Access to Food and Care, Health, and WASH 68 Simultaneous Access to the Underlying Determinants of Nutrition 76 Notes 81 References 81 5 Stunting and Access to the Underlying Determinants of Nutrition 83 Stunting and Children’s Simultaneous Access to the Drivers of Nutrition 84 Stunting and the Underlying Determinants of Nutrition 98 A Brief Summary of the Country-Specific Estimates 102 Testing the Sensitivity of the Findings 102 Notes 105 References 106 6 An Application at the Country Level: The Case of Tanzania 107 Stunting and the Underlying Determinants of Nutrition in Tanzania 108 Correlation between Stunting and Access to the Drivers of Nutrition 112 Monitoring Progress in Nutrition and Access to the Drivers of Nutrition 119 Notes 123 References 123 7 Income Growth Is an Essential Element of a More Effective Multisectoral Approach 125 Average Income 126 Variability of Income 135 Notes 138 References 140 Contents  xi 8 How Can a Multisectoral Strategy to Reducing Stunting Produce the Desired Results? 143 The Joint Targeting of Interventions by Different Sectors 144 Attention to the Incentive Structure 147 Notes 148 References 149 Appendix A Exploring the Relationship between Height- for-Age and the Three Drivers of Nutrition 151 Appendix B  In Search of Synergies 153 Appendix C Stunting and the Three Drivers of Nutrition: Country-Specific Estimates 161 Appendix D Stunting and the Three Drivers of Nutrition: Subpopulation-Specific Estimates 165 Boxes 1.1 Nutrition-Specific and Nutrition-Sensitive Interventions 27 4.1 Additional Questions in Surveys for Better Adequacy Measures 58 4.2 Access to and Utilization of Health Facilities 65 5.1 Tariffs, Trade, and Child Stunting 96 7.1 Agricultural Chemicals and Child Nutrition 127 Figures O.1 Differences in Access to Adequate Food and Care, Sub-Saharan Africa 7 O.2 Differences in Access to Adequate WASH, Sub-Saharan Africa 8 O.3 Differences in Access to Adequate Health, Sub-Saharan Africa 9 O.4 Children’s Access to the Three Determinants of Nutrition in Sub-Saharan Africa 10 O.5 Prevalence of Stunting in Sub-Saharan Africa, Based on Children’s Access to the Three Determinants of Nutrition 11 O.6 Prevalence of Stunting in Sub-Saharan Africa, Based on Children’s Access to Zero, One, Two, or Three Determinants of Nutrition 12 xii  Contents O.7 Marginal Effects on the Probability of Stunting, Based on Children’s Access to Combinations of the Three Determinants of Nutrition 13 O.8 The Relationship between Height-for-Age Z-scores and Wealth by Percentile of Household Wealth among Older and Younger Children 14 1.1 Causes, Correlates, and Consequences of Stunting 26 2.1 Determinants of Child Malnutrition 34 2.2 Framework for Actions to Achieve Optimum Fetal and Child Nutrition and Development 35 3.1 Stunting Prevalence among Younger (0–23 Months) and Older (24–59 Months) Children 46 3.2 Growth Faltering (Children 0–59 Months) 47 3.3 Growth Faltering, by Socioeconomic Characteristics (Children 0–59 Months) 48 3.4 Distribution of Height-for-Age for Different Populations 49 3.5 Stunting Prevalence, by Country (Children 0–23 Months) 50 3.6 Stunting Prevalence in Urban and Rural Areas within Each Country (Children 0–23 Months) 51 3.7 Stunting Prevalence among Boys and Girls within Each Country (Children 0–23 Months) 52 3.8 Stunting Prevalence, by Wealth Quintile within Each Country (Children 0–23 Months) 53 3.9 Stunting Prevalence, by Mother’s Education within Each Country (Children 0–23 Months) 54 B4.2.1 The Nonrandom Allocation of High- and Low-Quality Health Facilities, Nigeria 65 4.1 Percentage of Children with Access to Adequate Food and Care within Each Country (Children 0–23 Months) 68 4.2 Inequities in Access to Adequate Food and Care within Each Country (Children 0–23 Months) 69 4.3 Percentage of Children with Access to Adequate WASH within Each Country (Children 0–23 Months) 71 4.4 Inequities in Access to Adequate WASH within Each Country (Children 0–23 Months) 72 4.5 Percentage of Children with Access to Adequate Health within Each Country (Children 0–23 Months) 74 4.6 Inequities in Access to Adequate Health within Each Country (Children 0–23 Months) 75 Contents  xiii 4.7 Inequities in Joint Access to the Underlying Drivers of Nutrition within Each Country (Children 0–23 Months) 77 4.8 Inequities in Joint Access to the Underlying Drivers of Nutrition within Each Country, Urban versus Rural Areas (Children 0–23 Months) 78 4.9 Inequities in Simultaneous Access to the Underlying Drivers of Nutrition within Each Country, B20 versus T20 (Children 0–23 Months) 80 5.1 The Cumulative Distribution of Height-for-Age Z-Scores (Children 0–23 Months) 86 5.2 Access to No Drivers versus Access to One Driver of Nutrition and the Prevalence of Stunting (Children 0–23 Months) 87 5.3 Access to One versus Access to Two Drivers of Nutrition and the Prevalence of Stunting (Children 0–23 Months) 88 5.4 Access to Two versus Access to Three Drivers of Nutrition and the Prevalence of Stunting (Children 0–23 Months) 88 5.5 Access to No Drivers versus Access to Three Drivers of Nutrition and the Prevalence of Stunting (Children 0–23 Months) 89 5.6 Access to No Drivers versus Access to One or More Drivers of Nutrition and the Prevalence of Stunting (Children 0–23 Months) 90 5.7 The Effect of Simultaneous Access to One or More of the Drivers of Nutrition at the Lower Quantiles of the HAZ Distribution 104 5.8 The Effects of Access to Combinations of the Three Drivers of Nutrition at the Lower Quantiles of the HAZ Distribution 105 6.1 Growth Faltering in Tanzania, by Socioeconomic Characteristics (Children 0–59 Months) 108 6.2 Stunting Prevalence among Children under 24 Months 109 6.3 Access to the Different Components of Food/Care, WASH, and Health (Children under 24 Months) 111 6.4 Access to No Drivers versus Access to One or More Drivers of Nutrition and the Prevalence of Stunting (Children 0–23 Months) 115 6.5 Fraction of Increase in Mean HAZ between 2005 and 2016 Attributable to Explained Factors, Tanzania (Children 0–23 Months) 120 6.6 Decomposing the Contribution of Access to the Drivers of Nutrition to the Increase in Mean HAZ between 2005 and 2016, Tanzania (Children 0–23 Months) 121 xiv  Contents 6.7 Decomposing the Contribution of Access to the Drivers of Nutrition to the Increase in Mean HAZ (with Restrictions Imposed), Tanzania (Children 0–23 Months) 122 7.1 The Relationship between Height-for-Age Z-Scores and Wealth Percentiles by Wealth Deciles among Older and Younger Children 130 7.2 Differences in the Relationship between Height-for-Age Z-Scores and Wealth Percentiles by Wealth Deciles between Some Key Basic Determinants of Nutrition 132 7.3 Differences in the Relationship between Height-for-Age Z-Scores and Wealth Percentiles by Wealth Deciles between Groups of Countries 133 7.4 Country-Specific Estimates of the Relationship between Median Height-for-Age Z-Scores and Household Wealth Index Percentile 134 Map 6.1 Tanzania: Regional Stunting Rates and Access to Health and WASH 113 Tables 2.1 Characterization of Countries Included in the Study 41 4.1 Components of Food Security Measured 59 4.2 Components of Care 61 4.3 Components of WASH 63 4.4 Components of Health 64 4.5 Inclusion of Select Health and WASH Components 67 5.1 Percentage of Children with Adequate Access in Each Group/Category 84 5.2 Changes in the Stunting Rate for Groups of Children with Simultaneous Access to One or More of the Drivers of Nutrition, National-Level Estimates 92 5.3 Marginal Effects on the Probability of Stunting of Simultaneous Access to One or More of the Drivers of Nutrition, Child-Level Estimates 93 5.4 Marginal Effects on the Probability of Stunting of Simultaneous Access to One or More of the Drivers of Nutrition in Different Groups of Countries 95 5.5 Marginal Effects on the Probability of Stunting of Simultaneous Access to One or More of the Drivers of Nutrition for Different Areas and Socioeconomic Groups within Countries 97 Contents  xv 5.6 The Marginal Effect on the Probability of Stunting of Access to Adequate Levels in Combinations of the Underlying Drivers of Nutrition 99 5.7 The Marginal Effects on Stunting Prevalence of Access to the Components of the Underlying Drivers of Nutrition 101 6.1 Percentage of Children (under 24 Months) in Tanzania with Access to the Three Drivers of Nutrition, Using Two Different Definitions of “Adequate” 112 6.2 Changes in the Stunting Rate for Groups of Children with Simultaneous Access to One or More of the Drivers of Nutrition, Tanzania 116 6.3 The Marginal Effects on the Probability of Stunting of Access to Adequate Level in Combinations of the Underlying Drivers of Nutrition, Tanzania 117 7.1 The Relationship between Rainfall Variability (Coefficient of Variation) on the Prevalence of Stunting among Children 0–23 Months of Age in Rural Areas of Sub-Saharan Africa 136 7.2 The Relationship between Rainfall Shortfalls during the Growing Season and the Prevalence of Stunting among Children under 60 Months of Age in Selected Countries in Sub-Saharan Africa 137 B.1 Synergies among the Three Drivers of Nutrition in the Pooled Sample of Countries 156 B.2 Synergies in the Pooled Sample of Countries Based on the Components of the Three Drivers of Nutrition 157 C.1 Stunting and the Three Drivers of Nutrition within Countries in Sub-Saharan Africa 161 D.1 Stunting and the Three Drivers of Nutrition for Subpopulations within Sub-Saharan Africa 165 Foreword The prevalence of stunting in Sub-Saharan Africa has been declining signifi- cantly. Yet, because of high fertility rates, this is the only region in the world with a growing number of children under the age of five who have stunted growth, meaning they are too short for their age. This trend, if not arrested and reversed, can have grim long-term implications for the region’s human capital and economic growth. It has long been recognized that improving nutrition requires enlisting not only the health sector but also many others—agriculture; education; social pro- tection; and water, sanitation, and hygiene (WASH). Unfortunately, simply involving multiple sectors has not necessarily produced the desired outcomes. This book lays the groundwork for more effective multisectoral action on reducing stunting by analyzing and generating empirical evidence useful for informing the joint targeting and, if necessary, the sequencing of sector-specific interventions in countries in Sub-Saharan Africa. The analysis offers fresh insights on how empirical evidence and analysis can be used to inform policy-driven approaches that can more effectively combat stunting. These insights are derived from widely used surveys, such as the Demographic and Health Surveys (DHS), and provide a more holistic view of the multiple deprivations experienced by children with respect to the underly- ing drivers of nutrition and the prevalence of stunting. The DHS contain rather limited information on the components of food security and child care. Although serious, this shortcoming is compensated for by the availability of child anthropometric measures in all countries covered and the availability of information on the joint distribution of access—or lack of access—to adequate levels for some of the other underlying determinants of nutrition, such as improved water, improved WASH, and health services, according to interna- tionally accepted standards. The findings emerging from the analysis are worth highlighting here. One key ingredient of a successful strategy to reducing undernutrition is taking stock of the areas where stunting is more prevalent and where different sectors xvii xviii  Foreword operate or not. It is only then that target areas or target population groups can be identified. The analysis suggests that with limited budgetary resources, the greatest decline in stunting can be accomplished by targeting the scarce resources to children who do not have adequate access to any of the three nutri- tion drivers. Children with adequate access to one of the nutrition drivers are less likely to be stunted than children with access to none of the drivers. Moreover, chil- dren with simultaneous access to adequate levels of two or more nutrition driv- ers are even less likely to be stunted. The latter result validates the fundamental importance of scaling up the relevant interventions in ways that jointly target geographic areas (or populations within these areas) with a high prevalence of stunting. This also implies that the absence of some key sector from the targeted areas—be it agriculture, WASH, or social protection—may deter the sector or sectors already operating in the target areas from being the “first mover” in adopting nutrition-sensitive interventions. Thus, for the individual sectors to contribute jointly in the reduction of stunting, much must be done to coordi- nate their targeting of service delivery. The estimates reported in the book also provide policy guidance useful for the sequencing of sector-specific interventions in target areas or target popula- tions. First, if budgetary or other considerations allow for interventions cover- ing deprived children by only one sector, this sector should be health. Second, if a target area is already covered by the health sector, the decision of whether to cover the same target area by sectors such as WASH or agriculture should be based mainly on costs rather than on benefits. The book also provides evidence that increased access to adequate levels of the underlying drivers of nutrition coordinated across different sectors should not be considered in isolation from programs increasing incomes and minimiz- ing income variability in rural areas, both of which are important determinants of household demand for better nutrition. It is my hope that the findings of this regional report stimulate further analytic work that is operationally useful for the design of more effective multisectoral interventions for stunting at the country level in Sub-Saharan Africa. Albert Zeufack Chief Economist Africa Region Acknowledgments The Task Team Leader was Emmanuel Skoufias (Lead Economist, Poverty and Equity Global Practice). The team included Dr. Katja Vinha (International Consultant) and Dr. Ryoko Sato (Research Associate at the Harvard T. H. Chan School of Public Health). Boya Liu (Consultant) and Lucas Finamore (Consultant) provided excellent research support. The team is grateful to Albert Zeufack (Chief Economist for Africa) for his support throughout the process of this report. Patrick Hoang-Vu Eozenou (Senior Economist, Health, Nutrition, and Population Global Practice) and Nkosinathi Vusizihlobo Mbuya (Senior Nutrition Specialist, Health, Nutrition, and Population Global Practice) served as technical advisers on the nutrition-related aspects of the task, and the team is deeply thankful for their advice. Financial support for this work was provided by the Regional Studies Program of the Africa Chief Economist Office of the Africa Regional Vice Presidency of the World Bank and by the Government of Japan through the Japan Trust Fund for Scaling Up Nutrition. The peer reviewers for this report (including an anonymous reviewer) were Claire Chase (Economist, Water Global Practice); Carlo del Ninno (Lead Economist, Social Protection and Labor Global Practice, and Program Manager of the Sahel Adaptive Social Protection Program); Tom Bundervoet (Economist, Poverty and Equity Global Practice); Patrick Huang-Vu Eozenou; and Meera Shekar (Lead Health Specialist and Global Lead for Nutrition, Health, Nutrition, and Population Global Practice). Additional comments were received at various stages of this report from members of the Agriculture Global Practice: Emanuela di Gropello (Program Leader, on behalf of Elizabeth Huybens and the Country Management Unit); Ahmadou Moustapha Ndiaye (Country Director, Democratic Republic of Congo) and team; Ana Maria Munoz-Boudet (Senior Social Scientist, Poverty Global Practice); and Menno Mulder-Sibanda (Senior Nutrition Specialist, Health, Nutrition, and Population Global Practice). xix About the Authors Emmanuel Skoufias is a lead economist in the Poverty and Equity Global Practice of the World Bank Group and a global lead of the Global Solutions Group on Managing Risks for Poverty Reduction and Shared Prosperity. He joined the World Bank in 2004 after an academic career at the University of Colorado in Boulder and Pennsylvania State University; assignments as a senior  researcher in the Research Department of the Inter-American Development Bank; and time as a senior research fellow at the International Food Policy Research Institute, where he led the impact evaluation of the PROGRESA program of Mexico. He has published numerous papers on the targeting and the impacts of social transfer programs, child undernutrition, welfare disparities across regions, risk management among poor households, and the impacts of climate change and climatic variability on different dimen- sions of welfare. He holds a PhD in economics from the University of Minnesota (1988) and a BA from the University of California, Berkeley (1981). Katja Vinha holds a PhD in agricultural and resource economics from the University of Maryland, College Park (2005) and a BA from Macalester College (1996). She is a consultant for the World Bank; her current research focuses on child undernutrition, household vulnerability analysis, and the effects of climate change on welfare. She previously taught in the environmental economics mas- ter’s program at the Universidad de los Andes in Bogotá, Colombia. Ryoko Sato is a research associate at the Harvard T. H. Chan School of Public Health.  Before joining Harvard, she was a postdoctoral fellow at National University of Singapore and a junior professional officer at the World Bank. Her research focuses on evaluating barriers to and incentives for health behaviors among vulnerable populations through field experiments in developing and developed countries, as well as through secondary data analysis. She holds a PhD in economics from the University of Michigan at Ann Arbor (2015). xxi Abbreviations B20 bottom wealth quintile BO Blinder–Oaxaca CDF cumulative density function CRECER (To Grow) National Strategy for Combating Poverty and Chronic Child Malnutrition in Peru DDS dietary diversity score DHS Demographic and Health Surveys FAO Food and Agriculture Organization GDP gross domestic product GNI gross national income HAZ height-for-age Z-score HFIAS Household Food Insecurity Access Scale IDA International Development Association JMP Joint Monitoring Program LPM linear probability model LSMS-ISA Living Standards Measurement Study–Integrated Surveys on Agriculture MAD median absolute deviation/minimum acceptable diet (context-specific) MPA Multiphase Programmatic Approach PCE per capita expenditures PDF probability density function PPP purchasing power parity s. d. standard deviation SPI Standardized Precipitation Index xxiii xxiv  Abbreviations SUN Scaling Up Nutrition T20 top wealth quintile UNICEF United Nations Children’s Fund UQR unconditional quantile regression WASH water, sanitation, and hygiene WFP World Food Programme WHO World Health Organization Overview Motivation In 2014, 171 million children under the age of five had stunted growth, meaning that they were excessively short for their age.1 In Sub-Saharan Africa, specifi- cally, the scale of undernutrition is staggering; 58 million children under the age of five are too short for their age (stunted) and 13.9 million weigh too little for their height (wasted) (IFPRI 2016). Poor diets in terms of diversity, quality, and quantity, together with illness, are linked with micronutrient deficiencies (that is, a lack of important vitamins and minerals—such as iodine, vitamin A, and iron—linked to growth, development, and immune function) and contribute to stunting, as well as poor health and developmental outcomes. Income poverty and inequalities in access to basic services, such as health, water, and sanitation, and the use of improper care and feeding practices in the initial stages of children’s lives are associated with delayed child growth and serious costs that are eventually borne by the rest of society. Inequities in access to the underlying determinants of good nutrition and long-term well-being are associated with immediate costs in child welfare: no access or access to inade- quate levels of the drivers of nutrition is associated with an increase in the inci- dence of undernutrition and diarrheal disease. There are also important consequences in the long term—both to the individual and society—associated with the chronic undernutrition of children: a high risk of stunting, impaired cognitive development, lower school attendance rates, reduced human capital attainment, and a higher risk of chronic disease and health problems in adult- hood (Black et al. 2013; Hoddinott et al. 2013; Victora et al. 2008). Thus, inequi- ties in access to services early in the life of a child also contribute to the intergenerational transmission of poverty. The negative outcomes later in life are numerous, and some can even be quantified in economic terms. Recent World Bank estimates suggest that the income penalty a country incurs for not having eliminated stunting when today’s workers were children is about 7 percent of gross domestic product (GDP) per capita, on average. In Sub-Saharan Africa and South Asia, these figures rise to about 9–10 percent of GDP per capita (Galasso et al. 2017). The conceptual framework developed by the United Nations Children’s Fund (UNICEF) highlights the basic and underlying determinants of undernutrition that include environmental, economic, and sociopolitical factors, with income poverty and inequalities in access to services having a central role. An emphasis 1 2   ALL HANDS ON DECK on policies aimed at accelerating economic development and shared income growth is generally effective at reducing undernutrition, but the speed at which undernutrition declines with economic growth varies from country to country and from region to region. Thus, although growth is a necessary condition for the reduction of stunting, it is not a sufficient condition, especially in the region of Sub-Saharan Africa, where cross-country time series data suggest that the strength of the relationship between growth in real GDP and stunting is weaker in comparison to other regions in the world (World Bank 2017). Much of the effort to date has focused on the cost, financing, and impact of nutrition-specific interventions delivered mainly through the health sector, for the purpose of reaching the global nutrition targets for stunting, anemia, and breastfeeding and interventions for treating wasting (Horton et al. 2010; Shekar et al. 2016). However, an acceleration of the progress to reduce stunting in Sub- Saharan Africa requires enlisting more sectors in addition to the health sec- tor—such as agriculture; education; social protection; and water, sanitation, and hygiene (WASH)—in the effort to improve nutrition. Large scale nutrition-­ sensitive interventions in these sectors will have to be able not only to address the key underlying determinants of nutrition effectively but also to intensify the role of nutrition-specific interventions (Black et al. 2013). In recent years, there has been a significant increase in the number of initia- tives at the international as well as at the country level, aiming to scale up nutri- tion-sensitive interventions. One prominent initiative is the Scaling Up Nutrition (SUN) movement, whose framework is by now endorsed by 59 developing coun- tries, most of which are in Sub-Saharan Africa. Also, the Investing in the Early Years initiative, started in 2016, takes a holistic approach to the nature of inter- ventions needed for the healthy physical and cognitive development of children. The initiative aims to help children reach their full potential through an increased emphasis on the good nutritional status of mothers, expecting mothers, and chil- dren, especially in their first 1,000 days of life (including nine months in utero); exclusive and continued breastfeeding; health care; immunization; and proper feeding accompanied by good hygiene practices, early stimulation and age appropriate learning opportunities, nurturing care, and social protection that provides buffers from poverty and stress.2 All these initiatives are based on the premise that the determinants of undernutrition are multisectoral and that the solution to undernutrition requires multisectoral approaches. The effectiveness and ultimate success of multisectoral approaches to reduce stunting depend on having a more holistic view of the inequities and gaps in access to adequate levels of the underlying determinants (drivers) of nutrition, that is, care, food security, health, and WASH. The complex interdependence among the underlying determinants of nutrition is usually beyond the consid- eration of any given sector. On the one hand, the targeting of communities or the integration of nutritional considerations in existing interventions of the agricultural sector, for example, is unlikely to take into consideration the status Overview  3 of WASH services and facilities in the communities where these agricultural interventions are being considered. Consequently, the extent to which such nutrition-sensitive interventions can ultimately accelerate or enhance the impact of nutrition-specific interventions on key nutrition outcomes can be impeded considerably by the absence of adequate WASH facilities. On the other hand, the impact of agricultural interventions on child nutrition could be enhanced considerably if those interventions were to be accompanied by simul- taneous improvements in the water and sanitation facilities in the same com- munities. Thus, a more holistic approach to the targeting and nutrition sensitivity of interventions is likely to be better able to address the key underly- ing determinants of nutrition effectively, as well as reinforce the impacts of nutrition-specific interventions through the health sector. This report lays the groundwork for more effective multisectoral action on reducing stunting by analyzing and generating empirical evidence useful for informing the joint targeting and, if necessary, the sequencing of sector-specific interventions in countries in Sub-Saharan Africa. The analysis in the report offers new insights on how data can be used to inform allocation decisions of different sectors that can strengthen multisectoral efforts aiming to reduce stunting. These insights are derived from widely used surveys such as the Demographic and Health Surveys (DHS) and provide a more holistic view of the multiple deprivations experienced by children with respect to the underly- ing drivers of nutrition and the prevalence of stunting. The analysis is based on 33 recent DHS from the Sub-Saharan African region. Except for Angola, all 32 countries are part of the global SUN initiative.3 The main selection criterion for inclusion in the group of countries analyzed is that the survey had to be quite recent, that is, collected in 2010 or later. DHS contain rather limited informa- tion on the components of food security and childcare. This shortcoming is offset, however, by the availability of child anthropometric measures in all coun- tries covered and the information on the joint distribution of access (or lack of access) to adequate levels for some of the other underlying determinants of nutrition such as access to improved water, improved sanitation (WASH), and health services, according to internationally accepted standards.4 Information about the joint distribution of the underlying drivers of nutrition is essential for identifying particularly important gaps in access to those drivers that, if addressed through joint targeting by the different sectors, can serve to strengthen the impacts of nutrition initiatives. A univariate analysis carried out independently by any given sector, such as agriculture, for specific determinants of undernutrition (for example, food security and dietary diversity) without tak- ing into consideration other determinants of undernutrition typically in the pur- view of another sector (for example, water and sanitation), is unable to provide much guidance on the geographic areas where the interventions in the agricul- ture sector are likely to be less or more effective in terms of their effect on under- nutrition. In contrast, a multivariate analysis of the simultaneous (or the lack of) 4   ALL HANDS ON DECK access to adequate food security and water and sanitation services is able to pro- vide a more holistic view and pinpoint better the groups of children and the geographic areas where these inadequacies are prevalent. This improved under- standing enables the joint prioritization of operations and improved cost effi- ciency of interventions aimed at contributing to the reduction of undernutrition in the agricultural and water and sanitation sectors. The report also provides a fresh perspective on the relationship between household wealth and income and child undernutrition. Income is a basic determinant of nutrition underpinning much of the demand for the underlying drivers of nutrition. Specifically, two different dimensions of income are inves- tigated: income level and income variability. A multisectoral approach aiming to increase access to the underlying determinants of nutrition needs to be cog- nizant of the fact that household constraints faced at different times may inter- act with use and behavior. The implication worthy of serious policy consideration is that multisectoral interventions against undernutrition may be more effective if accompanied by broader development policies that mitigate the impacts of weather-related risks affecting income generation. Specifically, this report provides data-driven answers to the following questions: 1. What is the extent to which children have inadequate access to the underlying determinants of nutrition? Information available at the child level from the DHS allows one to get a better sense about the joint distribution of the various determinants (or drivers) of undernutrition. Information about the joint distri- bution of inadequate access to the drivers of nutrition is essential for better targeting and more effective programs in the context of budgetary constraints. 2. What is the association of stunting (or low height-for-age), at any given time, with inadequate food and care practices, inadequate WASH, and inadequate health? Is simultaneous access to an adequate level of one or more of the underlying determinants of nutrition associated with lower stunting? Empirical evidence confirming that simultaneous access to adequate levels of two or more nutrition drivers is associated with lower stunting validates the importance of the joint targeting of interventions by different sectors in the geographic areas and populations therein where stunting is prevalent. 3. If a multisectoral approach is not feasible, what is the sequencing of sector- specific interventions that could have the greatest impact on stunting? For  example, if budgetary constraints prevent the joint coverage of geo- graphic areas (or populations within these areas) with high prevalence of stunting by sectors such as agriculture, health care, and WASH, in which sector should the limited resources be allocated? 4. What is the role of income growth and income variability on child stunting, and how does income interact with the underlying drivers of nutrition? Overview  5 The findings of this regional report are intended to stimulate and provide a blueprint for further analytic work that is operationally useful for the design of more effective multisectoral interventions on stunting at the country level in Sub-Saharan Africa. The main findings of the report are based on data on chil- dren pooled across 33 countries, thus reflecting relationships that prevail on average among stunting and access to the drivers of nutrition by children resid- ing in different countries. Pooling data on children across many countries using common definitions and thresholds offers an advantage not only in terms of total sample size but also in that pooling increases the number of observations on children with access to adequate levels to different combinations of nutrition drivers. This advantage, however, comes at some costs. Applying definitions and using thresholds that are common across all countries tend to minimize the role of country-specific factors. As a consequence, the relationship between stunting and access to nutrition drivers derived from the sample of children pooled across countries may not necessarily reflect the relationship that may prevail within any given country. In view of these considerations, chapter 6 of the report provides an example applying the same methodology using data from only one country, Tanzania. In addition, appendix C of this report includes a brief for each of the 33 countries used in the report, summarizing access to the determinants of nutrition and their components as well as the simple correla- tion between stunting and the number of determinants.5 Appendix C also pres- ents the country-specific estimates summarizing the relationship between stunting and access to different combinations of nutrition drivers. Design and Methods The analysis in the report is guided by the insights provided by the UNICEF conceptual framework. The UNICEF framework, first proposed in 1990 (UNICEF 1990), was one of the first attempts at emphasizing food security, environment, health, and childcare practices as the main underlying determi- nants of child undernutrition in developing countries. A fundamental premise of this conceptual framework is that increases in access to adequate services in any one of the drivers of undernutrition—say, for example, food security— alone cannot make up for inadequate levels of access to the other drivers. Despite widespread acknowledgment of the role of the underlying determinants of nutrition, there is limited quantitative information on the size and direction of the interdependence among adequate (or inadequate) access to food security and care practices, environment, and health. For each child, an indicator for each of the subcomponents of food security, care, WASH, and health is constructed using available data in the DHS in the 33 Sub-Saharan African countries analyzed in this report. Next, for each 6   ALL HANDS ON DECK indicator a binary variable identifying “adequacy” is defined using thresholds based on accepted international standards. In consideration of the complexity of the links between the underlying determinants of undernutrition and the economic situation of the family, the analysis is also carried out separately for urban and rural households and for wealthier (top 20 percent of the asset dis- tribution) and poorer households (bottom 20 percent of the asset distribution). Thus, a more holistic view is provided of the extent to which adequate levels of the drivers of nutrition— adequate food and care, access to health services, and a safe and hygienic environment at the household and community levels—on their own as well as in combination are associated with better nutrition as mea- sured by height-for-age Z-scores (HAZs) and stunting rates. Results Substantial inequalities within countries in access between rural and urban areas, and between poorer and wealthier households, are prevalent in Sub-Saharan Africa (figures O.1, O.2, and O.3). The differences in access to adequate food/care6 are in general small across rural and urban households (figure O.1). In Liberia, children in rural areas are more than 5 percentage points more likely to have access to adequate food/care than children in urban areas. In Burundi, Ethiopia, Kenya, Malawi, Namibia, Niger, and Nigeria, urban children are more than 5 per- centage points more likely to have access to adequate food/care than rural chil- dren. There are greater differences, however, in access to adequate food and care by wealth category. Over one-third of the sample has more than a 5-percentage- point difference between the two wealth quintiles (figure O.1, panel b). Regarding access to adequate WASH,7 in nine countries the difference in access to WASH between urban and rural children is more than 30 percentage points (figure O.2, panel a). In 24 countries, children from the top wealth quintile (T20) are more than 30 percentage points more likely to have access to WASH than children from the bottom wealth quintile (B20) (figure O.2, panel  b). In Namibia, Senegal, and Niger the difference is more than 70 percentage points. Urban children are always more likely to have access to adequate health8 than rural children, and children from richer households are more likely to have access than children from poorer households. In 11 of the countries the difference in access to adequate health between urban and rural children is more than 30 ­percentage points (figure O.3, panel a). In 14 countries children from the richest households are more than 40 percentage points more likely to have access to adequate health than children from the poorest households (­ ­ figure O.3, panel b). Very few children have access to adequate levels of all three drivers of nutrition at the same time. Figure O.4 gives a sense of the inequities in joint/simultaneous access to the drivers of nutrition. Many children do not have access to any of the three determinants of nutrition. In three countries—Chad, Ethiopia, and Overview  7 Figure O.1  Differences in Access to Adequate Food and Care, Sub-Saharan Africa a. Children’s access to food and care, b. Children’s access to food and care, urban versus rural top quintile versus bottom quintile Burundi Rwanda Ethiopia Burundi Malawi Kenya Niger Niger Nigeria Malawi Namibia Ethiopia Kenya Comoros Madagascar Nigeria Ghana Madagascar Rwanda Namibia Burkina Faso Benin Mali Tanzania Guinea Côte d’Ivoire Cameroon Angola Zimbabwe Burkina Faso Gabon Sierra Leone Benin Zimbabwe Zambia Chad Mozambique Cameroon Gambia, The Zambia Chad Lesotho Côte d’Ivoire Senegal Uganda Mozambique Angola Gabon Tanzania Guinea Congo, Dem. Rep. Mali Sierra Leone Ghana Comoros Congo, Dem. Rep. Togo Gambia, The Senegal Liberia Congo, Rep. Uganda Lesotho Togo Liberia Congo, Rep. −5 0 5 10 15 –10 0 10 20 Percentage difference, Percentage difference, urban to rural top quintile to bottom quintile Source: Estimates based on pooled Demographic and Health Surveys from 33 countries in Sub-Saharan Africa. Note: Wealth quintiles calculated by World Bank staff. For this graph, children are considered 24 months or younger. Niger—more than 50 percent of children do not have access to even one of the determinants. The countries with the highest share of children with simultaneous access to all three determinants are Rwanda, Malawi, and Burundi—where more than 10 percent of the children have access to all three determinants. The prevalence of stunting among children with simultaneous access to ade- quate levels of all three nutrition drivers is significantly lower than children who do not have access to an adequate level in any of the nutrition drivers. The preva- lence of stunting among children who do not have access to an adequate level in 8   ALL HANDS ON DECK Figure O.2  Differences in Access to Adequate WASH, Sub-Saharan Africa a. Children’s access to WASH, b. Children’s access to WASH, urban versus rural top quintile versus bottom quintile Niger Namibia Namibia Senegal Burkina Faso Niger Madagascar Mozambique Mali Madagascar Angola Angola Cameroon Zimbabwe Côte d’Ivoire Cameroon Senegal Côte d’Ivoire Mozambique Gabon Guinea Tanzania Gabon Zambia Chad Gambia, The Tanzania Mali Togo Guinea Zimbabwe Benin Zambia Burkina Faso Benin Lesotho Ethiopia Nigeria Ghana Kenya Sierra Leone Togo Kenya Uganda Gambia, The Ethiopia Nigeria Sierra Leone Uganda Liberia Comoros Burundi Liberia Chad Congo, Rep. Ghana Congo, Dem. Rep. Rwanda Lesotho Malawi Burundi Congo, Rep. Malawi Comoros Rwanda Congo, Dem. Rep. −20 0 20 40 60 80 0 20 40 60 80 Percentage difference, Percentage difference, urban to rural top quintile to bottom quintile Source: Estimates based on pooled Demographic and Health Surveys from 33 countries in Sub-Saharan Africa. Note: Wealth quintiles calculated by World Bank staff. For this graph, children are considered 24 months or younger. WASH = water, sanitation, and hygiene. any of the three drivers of nutrition (access to 0) is 34 percent (figure O.5). In con- trast, the prevalence of stunting among children with access to adequate levels of all three of the drivers of nutrition (access to 3) is 18 percent, much lower than the prevalence among children with access to none. In addition, as the cumulative distribution function of the HAZs for children less than 24 months of age high- lights, significant reductions in severe stunting rates—children more than 3 stan- dard deviations (s. d.) below the median height-for-age—are also prevalent. Overview  9 Figure O.3  Differences in Access to Adequate Health, Sub-Saharan Africa a. Children’s access to health, b. Children’s access to health, urban versus rural top quintile versus bottom quintile Niger Angola Ethiopia Cameroon Guinea Nigeria Angola Madagascar Nigeria Ethiopia Mali Kenya Chad Niger Tanzania Mali Madagascar Tanzania Cameroon Senegal Uganda Guinea Mozambique Côte d’Ivoire Kenya Mozambique Côte d’Ivoire Togo Togo Lesotho Senegal Uganda Zambia Benin Congo, Dem. Rep. Gabon Congo, Rep. Zambia Gabon Comoros Gambia, The Chad Benin Liberia Burundi Congo, Rep. Liberia Rwanda Rwanda Burundi Ghana Burkina Faso Zimbabwe Ghana Lesotho Malawi Burkina Faso Congo, Dem. Rep. Malawi Zimbabwe Comoros Namibia Namibia Gambia, The Sierra Leone Sierra Leone 0 10 20 30 40 50 0 20 40 60 Percentage difference, Percentage difference, urban to rural top quintile to bottom quintile Source: Estimates based on pooled Demographic and Health Surveys from 33 countries in Sub-Saharan Africa. Notes: Wealth quintiles calculated by World Bank staff. For this graph, children are considered 24 months or younger. The greatest reductions in stunting are associated with increases in access from none (0) to any one nutrition driver and from any one driver to simultane- ous access to any two drivers.9 For example, the stunting rate among children with access to an adequate level in any one nutrition driver is 26.7 percent com- pared to the stunting rate of 34.1 percent prevailing among children with access to none (figure O.6, panel a). Lower stunting rates are also prevalent among children with simultaneous access to any two of the drivers (20.4 percent) 10   ALL HANDS ON DECK Figure O.4  Children’s Access to the Three Determinants of Nutrition in Sub-Saharan Africa Chad Ethiopia Niger Guinea Madagascar Nigeria Mozambique Angola Côte d’Ivoire Cameroon Mali Congo, Dem. Rep. Tanzania Kenya Zambia Togo Liberia Gabon Benin Senegal Gambia, The Comoros Uganda Burkina Faso Burundi Sierra Leone Namibia Malawi Zimbabwe Lesotho Congo, Rep. Rwanda Ghana 0 10 20 30 40 50 60 70 80 Share of children (%) Access to zero determinants of nutrition Access to all 3 determinants of nutrition Source: Estimates based on pooled Demographic and Health Surveys from 33 countries in Sub-Saharan Africa. Note: For this graph, children are considered 24 months or younger. The three determinants of nutrition are access to food and care; water, sanitation, and hygiene (WASH); and health. compared to the stunting rates prevalent among children with access to any one of the drivers (26.7 percent) (figure O.6, panel b). The marginal decline (gain) in the stunting rate is, however, smaller than that associated with a change in access from none (0) to any one of the drivers. Finally, even lower stunting rates are prevalent among children with simultaneous access to all three of the drivers (18.3 percent) compared to the stunting rates prevalent among children with access to any two of the drivers (20.4 percent) (figure O.6, panel c). Overview  11 Figure O.5  Prevalence of Stunting in Sub-Saharan Africa, Based on Children’s Access to the Three Determinants of Nutrition 0.8 0.6 Cumulative share of children 0.4 0.341 0.2 0.183 0 –6 –5 –4 –3 –2 –1 0 Height-for-age Z-score Access to zero determinants of nutrition Access to all 3 determinants of nutrition Source: Estimates based on pooled Demographic and Health Surveys from 33 countries in Sub-Saharan Africa. Note: For this graph, children are considered 24 months or younger. The three determinants of nutrition are access to food and care; water, sanitation, and hygiene (WASH); and health. The marginal effect on the probability of a child being stunted from access to adequate health only is greater than the marginal effect from access to adequate food and care only or access to adequate WASH only. Controlling for child, paren- tal, and household characteristics as well as for the geographic location of the household within a country, the probability of stunting associated with having access to adequate health only decreases by 3.0 percentage points (figure O.7). The estimates also reveal similar marginal effects on stunting of complementing health interventions with simultaneous access to adequate food/care or with simul- taneous access to adequate WASH. Controlling for child, parental, and household characteristics, as well as for the geographic location of the household within the country of residence, the decrease in the probability of stunting associated with having simultaneous access to adequate health and adequate WASH or adequate health and adequate food/care ranges from 4.3 to 5.0 percentage points (figure O.7). In the context of budgetary constraints, these results have important implica- tions for the targeting and the sequencing of sector-specific interventions in target areas (or target populations). These are elaborated in this overview as well as in the body of the report. The report also provides new evidence on (a) the contribution of wealth in reducing stunting by using household assets/wealth as a measure of living 12   ALL HANDS ON DECK Figure O.6  Prevalence of Stunting in Sub-Saharan Africa, Based on Children’s Access to Zero, One, Two, or Three Determinants of Nutrition a. Stunting and access to zero versus 1 determinant b. Stunting and access to 1 versus 2 determinants 0.8 Cumulative share of children 0.6 Cumulative share of children 0.6 0.4 0.341 0.3 0.267 0.2 0.267 0.204 0 0 −6 −5 −4 −3 −2 −1 0 −6 −5 −4 −3 −2 −1 0 Height-for-age Z-score Height-for-age Z-score Access to zero determinants of nutrition Access to 1 determinant of nutrition Access to 1 determinant of nutrition Access to 2 determinants of nutrition c. Stunting and access to 2 versus 3 determinants d. Stunting and determinant access 0.8 0.6 Cumulative share of children Cumulative share of children 0.6 0.4 0.34 0.3 0.204 0.2 0.183 0.18 0 0 −6 −5 −4 −3 −2 −1 0 −6 −5 −4 −3 −2 −1 0 Height-for-age Z-score Height-for-age Z-score Access to 2 determinants of nutrition Access to zero determinants of nutrition Access to 3 determinants of nutrition Access to 1 determinant of nutrition Access to 2 determinants of nutrition Access to 3 determinants of nutrition Source: Estimates based on pooled Demographic and Health Surveys from 33 countries in Sub-Saharan Africa. Note: For this graph, children are considered 24 months or younger. The three determinants of nutrition are access to food and care; water, sanitation, and hygiene (WASH); and health. HAZ = height-for-age Z-score. standards (in place of income) and (b) the effects of weather-related shocks on stunting. Building on recent evidence supporting the use of the wealth (or asset) index as a valid predictor of child nutrition outcomes (Krishna et al. 2015; Sahn and Stifel 2003), an analysis examines the relationship between child HAZs and the percentile ranking of households in the national wealth index distribu- tion within each of the 33 countries in the report.10 To unpack the role of wealth Overview  13 Figure O.7  Marginal Effects on the Probability of Stunting, Based on Children’s Access to Combinations of the Three Determinants of Nutrition 0.04 0.02 0 Marginal effects on probability –0.02 –0.04 –0.06 –0.08 –0.10 –0.003 –0.019 –0.03 –0.032 –0.05 –0.043 –0.054 Food and WASH Health Food and Food and WASH and Food and care care care health care, health, and WASH and health and WASH Coefficient estimate 95% Confidence Interval Source: Estimates based on 33 recent Demographic and Health Surveys from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: Marginal effects are based on the coefficient estimates obtained from the logit model in equation (2) in chapter 5 controlling for country-strata fixed effects and child, parental, and household characteristics, consisting of the following variables: indicator variables for age (in months), gender, birth order, and multiple birth; mother’s education in years; indicator for mother’s marital status; mother’s height; mother’s age; wealth quintile; number of children under five years of age; number of household members; and whether household is in a rural or urban area. Robust standard errors corrected for correlation at the cluster level. WASH = water, sanitation, and hygiene. as a determinant of growth faltering among children, separate estimates are pre- sented for younger (0–23 months) and older children (24–59 months). Income or wealth may have what appears to be a small positive impact on the HAZ of children in their first two years of life, but these small positive effects compound as the child ages, resulting in larger differences in HAZs that are more apparent later in life. The analysis demonstrates that increased wealth has a significant effect on child HAZs among both younger children (0–23 months) and older children (24–59 months) (­figure O.8). For example, an increase in the ranking of the wealth index value of the household by 10 percentage points is associated 14   ALL HANDS ON DECK Figure O.8  The Relationship between Height-for-Age Z-scores and Wealth by Percentile of Household Wealth among Older and Younger Children a. Excluding components of the underlying drivers of nutrition 0.009 0.008 0.007 Standard deviations 0.006 0.005 0.004 0.003 0.002 0.001 0 0–23 months 24–59 months b. Including components of the underlying drivers of nutrition* 0.009 0.008 0.007 Standard deviations 0.006 0.005 0.004 0.003 0.002 0.001 0 0–23 months 24–59 months bottom 10% bottom 20% bottom 30% bottom 40% bottom 50% Source: Estimates based on data for children 0–23 and 24–59 months old from 33 country Demographic and Health Surveys from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: Unconditional quantile regression estimates including country fixed effects. Detailed estimates available upon request. Both sets of regressions include child, parental, and household characteristics containing the following variables: indicator variables for age (in months), gender, birth order, and multiple birth; mother’s education in years; indicator for mother’s marital status; mother’s height; mother’s age; wealth quintile; number of children under five years of age; number of household members; and whether household is in a rural or urban area. HAZ = height-for-age Z-score; s. d. = standard deviation. * The components of the underlying drivers of nutrition include the following: whether a child had immediate skin-to-skin contact at birth and age-appropriate breastfeeding; child’s mother had four or more prenatal care visits; child delivery was assisted by a trained professional; child has been to at least one postnatal care visit; child received all vaccinations; child sleeps under mosquito net; the household has access to improved water, improved toilet, and handwashing facilities; the household disposes of stools; and the majority of households in the community (75 percent) use some sort of a sanitation facility. For 24–59-month-old children, the components of the underlying drivers of nutrition do not include the following variables (not collected for older children): whether a child had immediate skin-to-skin contact at birth and age-appropriate breastfeeding, the  child’s mother had 4 or more prenatal care visits, the child delivery was assisted by the trained professional, and child has been to at least one postnatal care visit. Overview  15 with a 0.03 s. d. increase in HAZ for younger children at the median (50 percent) quantile of the HAZ distribution.11 For older children at the median of the HAZ distribution, the same 10-percentage-point increase in the ranking of the wealth index of the household is associated with an increase in the HAZ by 0.06 s. d. (figure O.8, panel b).12 In addition, income appears to have a significant direct effect on child nutrition (HAZ) aside from that captured by the components of food and care, WASH, and health (compare panels a and b in figure O.8)—that is, even after controlling for the components of the underlying drivers of nutrition. Country-specific estimates of the effects of rainfall shortfalls during the growing season on the probability of stunting underline the importance of pro- grams that decrease the vulnerability of household income to weather-related shocks. A shortage of rainfall by 1 s. d. from the normal mean precipitation during the most recently completed growing season is associated with a 4.08-percentage-point increase in the probability of being stunted across all children age 0–60 months in Benin, a 3.59-percentage-point increase in the Democratic Republic of Congo, a 1.67-percentage-point increase in Mozambique, and 2.71-percentage-point and 2.50-percentage-point increases in Nigeria and Rwanda, respectively. These estimates confirm that policies and programs that decrease the vulnerability of household income to weather- related shocks can significantly contribute to the decline of stunting and should be considered as components of any multisectoral effort. Policy Considerations The analysis in the report offers new insights on how data can be used to inform allocation decisions of different sectors that can strengthen multisectoral efforts aiming to reduce stunting. In practice, however, there is little assurance that involving multiple sectors in the effort to reduce undernutrition will produce the desired outcomes. The history of multisectoral initiatives on nutrition contains many nonper- forming projects (IEG 2009; World Bank 2014). Multisectoral nutrition plan- ning was favored by the international development community in the 1970s, but it quickly became apparent that it was overly ambitious and too dependent on other sectors that were reluctant to be coordinated (Levinson and Balarajan 2013). A more recent example is the case of multisectoral AIDS projects in Africa, where sectors such as health, WASH, and education were involved. When budgets, governance, and accountability structures are driven mainly by sector-specific considerations within country ministries as well as within inter- national organizations, “multisectoral” initiatives tend to reduce clarity and specificity on the role and responsibility of each sector (IEG 2009). 16   ALL HANDS ON DECK In recent years, however, a few countries such as Brazil, Peru, and Senegal have been quite successful in significantly reducing undernutrition through multisectoral approaches tailored to their needs and circumstances. This section synthesizes the main findings of this report with the lessons offered from these successful country cases to identify the key ingredients of a multisectoral strat- egy in the reduction of undernutrition that maximizes the potential to produce the desired outcomes in practice. The Joint Targeting of Interventions by Different Sectors One fundamental ingredient of a successful strategy is the scale-up of interven- tions by agriculture (food security), health, care, and WASH that are jointly targeted to geographic areas (or populations within these areas) with high prev- alence of stunting.13 The primary purpose of these jointly targeted operations is to increase access to the underlying determinants of nutrition as envisioned by the UNICEF conceptual framework underpinning this report. For sector-­ specific investments to contribute to the reduction of stunting and to speed up progress toward the Sustainable Development Goals, the prevalence of stunting needs to be considered as an additional criterion when prioritizing and allocat- ing scarce resources at the country and subnational levels. The very high pro- portions of children with inadequate access to all three drivers of nutrition (figure O.4) suggest that the strategy of joint targeting would apply to almost all countries in Sub-Saharan Africa. This strategy requires taking stock of the sectors operating in the target areas (or target population groups) and redirecting operations of the missing sectors to the target areas. The absence of some key sector from the target areas, such as agriculture, WASH, or social protection, may also act as a deterrent for the sector (or sectors) already operating in the target areas to be the “first mover” in terms of adopting nutrition-sensitive interventions. The descriptive statistics on access to adequate WASH in the 33 countries covered by the report chapter  3), in combination with the recently completed WASH Poverty (­ Diagnostic of the Water and Poverty and Equity Global Practices (World Bank 2017b), confirm the limited coverage by the WASH sector in the rural areas where stunting is prevalent. The same also applies to the fraction of the popula- tion covered by social protection schemes (Del Ninno, Coll-Black, and Fallavier 2016). For example, the Productive Safety Net program of Ethiopia, one of the largest in Sub-Saharan Africa, covers only 10 percent of the population (World Bank 2012). This report provides country authorities with a holistic picture of the gaps in access to the drivers of nutrition within countries; this picture is critical for the formulation of a more informed, evidence-based, and balanced multisectoral strategy against undernutrition in their countries. Much work remains to be done in terms of coordinating the targeting of service delivery in areas with Overview  17 stunting if all key sectors are to contribute jointly in the reduction of stunting. Despite a broad correlation between monetary poverty and children’s health at the country level, the targeting of stunted children is not as simple as distin- guishing between urban and rural areas or using “poverty maps” identifying the poor and nonpoor regions. Not all children in poor households or in rural areas are undernourished, and, in many countries, not all children in nonpoor house- holds or in urban areas are well nourished. With limited budgetary resources, the greatest decline in stunting can be accomplished by targeting the scarce resources to children who do not have ­adequate access to any of the three nutrition drivers (chapter 5). If the same resources were to be allocated toward increasing access to an additional nutri- tion driver among children who already have access to one driver, the conse- quent decline in stunting is likely to be smaller. The estimates can also provide policy guidance useful for the sequencing of sector-specific interventions in target areas (or target populations). • First, if budgetary or other considerations allow for interventions covering deprived children by only one sector, that sector should be health. Thus, the biggest bang for the buck in reducing stunting is through expanded cover- age by the health sector addressing the immediate causes of undernutrition figure O.7). The findings also empirically validate the sequencing of inte- (­ grated approaches to improving nutrition outcomes such as the one currently in process for Madagascar and in preparation for other first wave countries. As soon as other sectors succeed in redirecting their operations to the target areas, an acceleration in the reductions in stunting is very likely to follow. • Second, if a target area is already covered by the health sector, the decision of whether to cover the same target area by sectors, such as WASH or agriculture, should be based mainly on costs rather than benefits. This basis is recom- mended because the benefits in terms of reductions in stunting through simultaneous coverage by WASH or agricultural (food/care) operations appear to be similar; however, country-specific analyses would need to be carried out to confirm that the regional relationship holds nationally. Increased access to adequate levels of the underlying drivers of nutrition coordinated across different sectors should not be considered in isolation from programs increasing incomes and minimizing income variability in rural areas, both important determinants of household demand for better nutrition (chap- ter 7). Instead, programs and interventions aimed at increasing the level and stability of income among populations where stunting is prevalent should be considered indispensable components of a multisectoral approach to reducing undernutrition in Sub-Saharan Africa. Household constraints faced at different times may interact with the use of services. Thus, multisectoral interventions 18   ALL HANDS ON DECK against undernutrition are likely to be more effective when accompanied by broader development policies and programs that mitigate the impacts of risks. There is an increasing policy emphasis on adaptive social protection and climate-smart agriculture programs that provide the basis for increasing the level and decreasing the variability of incomes in rural areas. (Del Ninno, Coll- Black, and Fallavier 2016; Del Ninno and Mills 2015; Lipper et al. 2014; Tirado et al. 2013; Wheeler and von Braun 2013). These programs not only serve as useful instruments to respond ex post to the incidence of droughts, floods, and other natural disasters; they also help households build their resilience before shocks hit. Drought-resistant seeds for maize, for example, have the potential to increase both the level and the stability of income from agricultural activities in Sub-Saharan African countries. Cash transfer programs can redistribute income to the poorer segments of the population and allow households to invest in human capital and in child nutrition, to build assets and diversify their livelihood strategies. Public works programs can help households and com- munities reduce their vulnerability to shocks while improving community infrastructure and the opportunities for new and improved livelihoods. In par- allel or in combination, increased access to insurance products and credit mar- kets can ensure better and more efficient use of resources by eliminating the incentive to adopt low-risk/low-return crops and production methods and alle- viate intertemporal distortions on human and productive capital investment such as cutting down on food consumption and health services or withdrawing children from school. The experience of Peru highlights the contribution of income growth in reducing stunting. Between 2007 and 2012, the prevalence of stunting declined by 21.4 percentage points, from 54.7 percent to 33.3 percent, in the districts targeted by the National Strategy for Combating Poverty and Chronic Child Malnutrition in Peru (CRECER) strategy (Levinson and Balarajan 2013; World Bank 2012). The contemporaneous income growth and poverty reduction in Peru during the same period had a critical facilitating role in the success of the multisectoral efforts at decreasing undernutrition. Specifically, the poverty rate between 2004 and 2011 decreased by 31 percentage points (from 58.8 percent to 27.8 percent), and extreme poverty decreased from 16.7 percent to 6.3 percent. Although the declines in stunting cannot be exclusively attributed to economic growth and poverty reduction in Peru, it is important for policy makers in Sub-Saharan African countries to consider income growth and reduced income variability as necessary but not sufficient conditions for the reduction of child stunting. More research is needed on the extent to which sector-specific nutrition- sensitive interventions have any measurable impact on stunting over and above the impact from access and use (or the targeting) of the services provided by the  normal operations of sector-specific programs. Nutrition-sensitive Overview  19 interventions are believed to be essential for achieving adequate access to the underlying determinants of nutrition by improving or redirecting or adding marginal changes to normal sector operations to enhance the coverage and effectiveness of nutrition-specific interventions through the health sector. Recent reviews of the evidence available on the nutritional effects of nutrition- sensitive interventions (Galasso et al. 2017; Ruel, Alderman, and the Maternal and Child Nutrition Study Group 2013) confirm the strong potential of such interventions but offer very little in terms of solid evidence. For example, the evidence of the nutritional effect of agricultural programs is inconclusive, mainly because of poor-quality evaluations. Evidence is also scarce on the nutri- tional effect of social safety net programs, although combined early child devel- opment and nutrition interventions show promising synergies in child development and in some cases nutrition. Moreover, future evaluations of nutrition-sensitive interventions should take into serious consideration the pos- sibility that the impacts of these interventions are likely to depend on the scale of coverage by other sectors as well as on the extent to which the other sectors implement nutrition-sensitive interventions.14 Attention to the Incentive Structure The adoption of a governance and accountability structure that provides the right incentives to all actors involved is the other fundamental ingredient of a successful multisectoral strategy. A common feature of all countries with suc- cessful multisectoral projects against stunting is the importance of a coordina- tion system that is supported by high levels of government. The Nutrition Enhancement Program of Senegal, for example, involving the Ministries of Health and Education, was coordinated by a unit attached to the Prime Minister’s office (Agency in Charge of the Fight against Malnutrition) (IEG 2016). Along similar lines, the CRECER strategy in Peru was placed not in a line ministry but directly under the prime minister’s office. In Brazil, a new ministry, the Ministry of Social Development and Fight against Hunger, was created to provide a platform for coordination with other ministries, such as the Ministry of Education administering the National School Food Program and the Ministry of Agricultural Development involved in the Food Acquisition Program. In the development community, aside from the impetus generated by the Sustainable Development Goals, there is a confluence of factors that contribute to a more solid foundation for multisectoral projects aiming to reduce under- nutrition. This foundation allows for operations that are better structured, bet- ter performing, and potentially much more effective than in the past. Of the 33 countries analyzed in this report, all but Angola are members of the SUN 20   ALL HANDS ON DECK movement whose framework is now endorsed by 59 developing countries and over 100 partners and nearly 3,000 civil society organizations that are members of SUN. All members of SUN prioritize nutrition as an investment in their growth and recognize nutrition as an investment in economic and social devel- opment to strengthen their nations. The renewed emphasis on eliminating extreme poverty and boosting shared prosperity has also provided the incentive for different sectors to reevaluate their country engagement strategies. The recent WASH Poverty Diagnostic Initiative, for example, generated new insights on how data can be used to inform allocation decisions to reduce inequalities and prioritize investment in WASH to boost human capital.15 Also, new models of operational engagement with client countries such as the Multiphase Programmatic Approach (MPA) offer the opportunity to improve coherence across interventions and strengthen the strategic focus of operations within those countries. The MPA approach is ideally suited for long-term engagement with multiple sectors as in the case of nutrition. In addition, the MPA separates engagements into phases, which facil- itates greater learning and adaptation. Combined with performance-based financing with disbursement-linked targets, the MPA program has the potential to give the proper incentives to the local authorities implementing the program. In the case of Brazil and Peru, for example, the use of specific targets has been associated with highly positive results in terms of generating proactive initia- tives and encouraging local ownership and accountability at subnational levels. This report contributes to the analytical foundations of a more effective mul- tisectoral approach and as such constitutes only one component of a much broader effort aimed at improving the results of multisectoral projects toward the reduction of undernutrition. It is imperative that the renewed efforts yield the desired results—especially in Sub-Saharan African countries. Notes 1. This 2014 figure is agreed upon by the United Nations Children’s Fund (UNICEF), the World Health Organization (WHO), and the World Bank. In statistical terms, a child is stunted if his/her height-for-age Z-score is more than 2 standard ­deviations (s. d.) below the median height of a healthy reference population (that is, HAZ < –2). 2. Along parallel lines, initiatives aim to foster knowledge exchange and cross-sectoral collaboration and coordination at the project level for improving nutrition (Secure Nutrition Knowledge Platform, World Bank). 3. More information on the SUN initiative and the countries participating in the SUN initiative can be found at http://scalingupnutrition.org/sun-countries/about​ -sun-countries/. Overview  21 4. In contrast, other surveys such as the Living Standards Measurement Surveys (LSMS) contain more information on food and nonfood consumption and other dimensions of food security, and sometimes repeated observations (panel data) on the same children and households, but with the result of more limited country cov- erage, lack of information on anthropometric measurements, or insufficient infor- mation for the construction of measures of WASH and health based on internationally accepted standards. 5. The 33 country-specific notes can be accessed at http://documents.worldbank.org​ /curated/en/467711529406437446. 6. A child is considered adequate in food and care if he or she meets the age-appropri- ate criterion for minimum acceptable diet and at least one of the following care giving measures: (a) early initiation of breastfeeding and (b) age-appropriate breast- feeding (chapter 4). 7. A child/household is defined to have access to adequate WASH if at least three of the following components are satisfied: (a) the household has improved water facilities; (b) the household has improved sanitation (based on the WHO/UNICEF Joint Monitoring Program); (c) less than 25 percent of the households in the community revert to open defecation; (d) the household has adequate handwashing facilities with water and soap or detergent, as observed by the interviewer; and (e) the child’s stools are disposed of into an improved sanitation facility. For all countries in the analysis, information is available to construct at least the first three components; except for the Republic of Congo and Gabon, at least one of the other two compo- nents is also available. 8. A child is defined to have access to adequate health if at least three of the following components are satisfied: (a) mother had four prenatal visits; (b) birth was assisted by a health care professional; (c) the child was seen in a postnatal checkup visit; (d) child is immunized according to schedule; and (e) child sleeps under a mosquito net (when applicable). For all countries, information is available on prenatal visits, assisted birth, and immunizations. For four countries (Cameroon, The Gambia, Mozambique, and Rwanda), no information is available on postnatal checkup, and for two countries (Ethiopia and Lesotho) no information is available on mosquito nets; for these countries the index is constructed without considering the unavail- able data. 9. The analysis in chapter 5 of this report confirms that the same patterns hold for the marginal effect of access to one or more drivers on the probability of a child being stunted (after controlling for country-level effects and a variety of child, parental, and household characteristics). The same pattern also holds in different groups of countries, in urban and rural areas, and for children in the B20 and T20 of the dis- tribution of wealth in each country. 10. The standard of living of households is usually measured by household per capita expenditures (PCE) or income. Household surveys with detailed household PCE and child anthropometric measures (such as height and weight) in the same survey are scarce. For example, using the nine LSMS-Integrated Surveys on Agriculture (LSMS-ISA) currently available and most likely to collect PCE and child height, it was possible to confirm that only five of the nine surveys had information on both PCE and measures of child height. 22   ALL HANDS ON DECK 11. The estimates in the figures reflect the effect of a 1-percentage-point increase in the ranking of the wealth index value of the household in the HAZ. 12. The same patterns also prevail among children in urban and rural areas, by mother’s educational attainment, and by mother’s autonomy in decision making at home. Estimates are also provided for countries in the first wave of the Investing in Early Years initiative and in countries in the Sahel region and compared to countries not in the first wave or not in the Sahel region, respectively, as well as within each of the 33 countries. In Tanzania, for example, the estimates at the median (50 percent) quantile of HAZ suggest that a 10-percentage-point increase in the wealth index, which is equivalent to a 17 percent increase in mean consumption per capita, leads to a 0.06 s. d. increase in the HAZ of 24–59-month-old children, an effect compa- rable to that found by Alderman, Hoogeveen, and Rossi (2006) based on PCE (instead of the percentile of the asset index used here). 13. For conceptual clarity, nutrition sensitivity in this report is considered as an add-on component to the normal operations and activities of a sector program, distinct from the targeting of the program or project. Nutrition-sensitive interventions are generally identified with efforts to redirect, or improve, or add marginal changes to normal sector operations to enhance the coverage and effectiveness of nutrition- specific interventions through the health sector (for example, see Ruel, Alderman, and the Maternal and Child Nutrition Study Group 2013). 14. This conclusion also has important implications for the design of randomized con- trol trials of nutrition-sensitive interventions, if the findings of the impact evaluation are to have any external validity for scaling up. 15. The synthesis report and the country-specific studies of the WASH Poverty Diagnostic can be accessed at http://www.worldbank.org/en/topic/water/publication/wash​ -poverty-diagnostic. References Alderman, H., H. Hoogeveen, and M. Rossi. 2006. “Reducing Child Malnutrition in Tanzania: Combined Effects of Income Growth and Program Interventions.” Economics and Human Biology 4 (1): 1–23. Black, R. E., C. G. Victora, S. P. Walker, Z. A. Bhutta, P. Christian, M. de Onis, M. Ezzati, S. Grantham-McGegor, J. Katz, R. Martorell, R. Uauy, and the Maternal and Child Nutrition Study Group. 2013. “Maternal and Child Undernutrition and Overweight in Low-Income and Middle-Income Countries.” Lancet 382 (9890): 427–51. http:// dx.doi.org/10.1016/S0140-6736(13)60937-X. Del Ninno, C., S. Coll-Black, and P. Fallavier. 2016. Social Protection Programs for Africa’s Drylands. Washington, DC: World Bank. Del Ninno, C., and B. Mills, eds. 2015. Safety Nets in Africa: Effective Mechanisms to Reach the Poor and Most Vulnerable. Washington, DC: World Bank. Galasso, E., A. Wagstaff, S. Naudeau, and M. Shekar. 2017. “The Economic Costs of Stunting and How to Reduce Them.” World Bank Policy Research Note, PRN/17/05, World Bank, Washington, DC. Overview  23 Hoddinott, J., H. Alderman, J. R. Behrman, L. Haddad, and S. Horton. 2013. “The Economic Rationale for Investing in Stunting Reduction.” Maternal and Child Nutrition 9 (Suppl. 2): 69–82. Horton, S., M. Shekar, C. McDonald, A. Mahal, and B. J. Krystene. 2010. Scaling Up Nutrition: What Will It Cost? Washington, DC: World Bank. http://hdl.handle. net/10986/2685. IEG (Independent Evaluation Group). 2009. Improving Effectiveness and Outcomes for the Poor in Health, Nutrition and Population: An Evaluation of World Bank Group Support since 1997. Washington, DC: World Bank. https://openknowledge.worldbank. org/handle/10986/2645. ———. 2016. A Decade of World Bank Support to Senegal’s Nutrition Program. Program Performance Assessment Report No. 110290. Washington, DC: World Bank. IFPRI (International Food Policy Research Institute). 2016. Global Nutrition Report 2016: From Promise to Impact: Ending Undernutrition by 2030. Washington, DC: International Food Policy Research Institute. Krishna, A., J. Oh, J. Lee, H.-Y. Lee, J. M. Perkins, J. Heo, Y. Sun Ro, and S. V. Subramanian. 2015. “Short-Term and Long-Term Associations between Household Wealth and Physical Growth: A Cross-Comparative Analysis of Children from Four Low- and Middle-Income Countries.” Global Health Action 8 (1). doi:10.3402/gha.v8.26523. Levinson, J., and Y. Balarajan. 2013. “Addressing Malnutrition Multisectorally: What Have We Learned from Recent International Experience?” UNICEF Nutrition Working Paper, UNICEF and MDG Achievement Fund, New York. Lipper, Leslie, Louise Jackson, Andrew Jarvis, Fred Kossam, Wendy Mann, Nancy McCarthy, Alexandre Meybeck, Henry Neufeldt, Tom Remington, Pham Thi Sen, Reuben Sessa, Reynolds Shula, Austin Tibu, and Emmanuel F. Torquebiau. 2014 “Climate-Smart Agriculture for Food Security.” Nature Climate Change 4 (December): 1068–72. Ruel, M. T., H. Alderman, and the Maternal and Child Nutrition Study Group. 2013. “Nutrition-Sensitive Interventions and Programmes: How Can They Help to Accelerate Progress in Improving Maternal and Child Nutrition?” Lancet 382 (9891): 536–51. http://dx.doi.org/10.1016/S0140-6736(13)60843-0. 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. Shekar, M., J. Kakietek, J. D. Eberwein, and D. Walters. 2016. An Investment Framework for Nutrition: Reaching the Global Targets for Stunting, Anemia, Breastfeeding, and Wasting. Washington, DC: World Bank. Tirado, M. C., P. Crahy, L. Mahy, M. Neira, S. Msangi, R. Brown, C. Scaramella, D. Costa Coitinho, and A. Müller. 2013. “Climate Change and Nutrition: Creating a Climate for Nutrition Security.” Food and Nutrition Bulletin 34 (4): 533–47. UNICEF (United Nations International Children’s Emergency Fund). 1990. Strategy for Improved Nutrition of Women and Children in Developing Countries. A UNICEF Policy Review. New York: UNICEF. Victora, C. G., L. Adair, C. Fall, and others, for the Maternal and Child Undernutrition Group 2008. “Maternal and Child Undernutrition: Consequences for Adult Health and Human Capital.” Lancet 371: 340–57. 24   ALL HANDS ON DECK Wheeler, T., and J. von Braun. 2013. “Climate Change Impacts on Global Food Security.” Science 341 (6145): 508–13. doi:10.1126/science.1239402. World Bank. 2012. Managing Risk, Promoting Growth: Developing Systems for Social Protection in Africa. The World Bank’s Africa Social Protection Strategy 2012–2022. Washington, DC: World Bank. ———. 2014. “Learning from World Bank History: Agriculture and Food-Based Approaches for Addressing Malnutrition.” Agriculture and Environmental Services Discussion Paper 10, World Bank, Washington, DC. ———. 2017a. “Stunting Reduction in Sub-Saharan Africa.” HNP Discussion Paper, World Bank, Washington, DC. ———. 2017b. Reducing Inequalities in Water Supply, Sanitation, and Hygiene in the Era of the Sustainable Development Goals: Synthesis Report of the WASH Poverty Diagnostic Initiative. Washington, DC: World Bank. https://openknowledge.worldbank.org​ /bitstream/handle/10986/27831/W17075.pdf?sequence=5&isAllowed=y. Chapter 1 Introduction and Motivation In 2014, 171 million children under the age of five had stunted growth, meaning that they were excessively short for their age.1 In Sub-Saharan Africa, specifi- cally, the scale of undernutrition is staggering—with 58 million children under age five being too short for their age (stunted) and 13.9 million weighing too little for their height (wasted) (IFPRI 2016). Poor diets in terms of diversity, quality, and quantity, together with illness, are linked with micronutrient defi- ciencies (that is, a lack of important vitamins and minerals linked to growth, development, and immune function, such as iodine, Vitamin A, and iron) and contribute to stunting, poor health, and poor development outcomes. Income poverty and inequalities in access to basic services, such as health, water, sanitation, and proper care and feeding practices, in the initial stages of children’s lives are associated with delayed child growth and serious costs that are eventually borne by the rest of society. Inequities in access to the underlying determinants of good nutrition and long-term well-being are associated with immediate costs in child welfare: access to inadequate levels of drivers of nutri- tion, or no access to those drivers, is associated with an increase in the incidence of undernutrition and diarrheal disease. There are also important consequences in the long term—both to the individual and society—associated with the chronic undernutrition of children: a high risk of stunting, impaired cognitive development, lower school attendance rates, reduced human capital attainment, and a higher risk of chronic disease and health problems in adulthood (Black et al. 2013; Hoddinott et al. 2013; Victora et al. 2008). Thus, inequities in access to services early in the life of a child also contribute to the intergenerational transmission of poverty. The negative outcomes later in life are numerous, and some can even be quan- tified in economic terms (figure 1.1). Recent World Bank estimates suggest that the per capita income penalty a country incurs for not having eliminated stunt- ing when today’s workers were children is about 7 percent of gross domestic 25 26   ALL HANDS ON DECK Figure 1.1  Causes, Correlates, and Consequences of Stunting Cognition in early Cognition in childhood adulthood Parental knowledge, Nutrients, Socioeconomic Socioeconomic family disease, skills in early skills in Wages in resources, stimulation, childhood adulthood adulthood environment, nurturing care food cost, and availability Height in early Height in childhood adulthood Health in early Health in childhood adulthood Source: Galasso et al. 2017. product (GDP) per capita, on average. In Sub-Saharan Africa and South Asia, these figures rise to about 9–10 percent of GDP per capita (Galasso et al. 2017). The conceptual framework developed by the United Nations Children’s Fund (UNICEF) highlights the basic and underlying drivers of undernutrition that include environmental, economic, and sociopolitical factors with income pov- erty and inequalities in access to services having a central role. A focus on poli- cies aimed at accelerating economic development and shared income growth is generally effective at reducing undernutrition, but the speed at which undernu- trition declines with economic growth varies from country to country and from region to region. Thus, although growth is a necessary condition for the reduc- tion of stunting, it is not a sufficient condition, especially in Sub-Saharan Africa, where cross-country time series data suggest that the strength of the relation- ship between growth in real GDP and stunting is weaker in comparison to other regions in the world (World Bank 2017). Much of the effort to date has focused on the costing, financing, and effect of nutrition-specific interventions delivered mainly through the health sector to reach the global nutrition targets for stunting, anemia, and breastfeeding and interventions for treating wasting (Horton et al. 2010; Shekar et al. 2016). However, an acceleration of the progress toward reducing stunting in the Sub- Saharan African region requires enlisting more sectors in addition to the health sector—such as agriculture; education; social protection; and water, sanitation, and hygiene (WASH)—in the effort to improve nutrition. Large-scale nutrition- sensitive interventions in these sectors will have to be able not only to effectively Introduction and Motivation   27 address the key underlying determinants of nutrition but also to intensify the role of nutrition-specific interventions (Black et al. 2013) (box 1.1). In recent years, there has been a significant increase in the number of initia- tives at the international as well as at the county level to scale up nutrition-­sensitive interventions. One prominent example is the Scaling Up Nutrition (SUN) BOX 1.1 Nutrition-Specific and Nutrition-Sensitive Interventions Nutrition-specific interventions address the immediate determinants of fetal and child nutrition and development—adequate food and nutrient intake, feeding, caregiv- ing and parenting practices, and low burden of infectious diseases. Examples include adolescent, preconception, and maternal health and nutrition; maternal dietary or micronutrient supplementation; promotion of optimum breastfeeding; complementary feeding and responsive feeding practices and stimulation; dietary supplementation; diversification and micronutrient supplementation or fortification for children; treat- ment of severe acute undernutrition; disease prevention and management; and nutri- tion in emergencies. Nutrition-sensitive interventions address the underlying determinants of fetal and child nutrition and development—food security; adequate caregiving resources at the maternal, household, and community levels; and access to health services and a safe and hygienic environment—and incorporate specific nutrition goals and actions. Examples include agriculture and food security; social safety nets; early child develop- ment; maternal mental health; women’s empowerment; child protection; schooling; water, sanitation, and hygiene; and health and family planning services. Nutrition-sensitive interventions are believed to be essential for achieving adequate access to the underlying determinants of nutrition by redirecting, improving, or adding marginal changes to normal sector operations to enhance the coverage and effective- ness of nutrition-specific interventions through the health sector (Ruel, Alderman, and the Maternal and Child Nutrition Study Group 2013). It is important to point out that this definition conflates coverage of normal operations (or targeting) with the improve- ment or the addition of marginal changes enhancing the effectiveness of normal oper- ations. Thus, the nutrition sensitivity of a sector-led project can be enhanced simply by improving its targeting to geographic areas and the populations therein with high stunting prevalence. For conceptual clarity, nutrition sensitivity in this report is considered an add-on component to the normal operations and activities of a sector program, distinct from the targeting of the program. On the basis of this distinction, nutrition-sensitive opera- tions by a sector in a specific geographic area can be implemented only if a sector car- ries out normal operations in the targeted areas. Sources: Adapted from Ruel, Alderman, and the Maternal and Child Nutrition Study Group 2013 and World Bank 2013. 28   ALL HANDS ON DECK movement, launched in April 2010, whose framework is now endorsed by 59 developing countries, over 100 partners, and nearly 3,000 civil society organiza- tions that are members of SUN. Many countries are prioritizing nutrition as an investment in their growth and are recognizing nutrition as an investment in economic and social development to strengthen their nations. Along parallel lines, initiatives within the World Bank and other development agencies and research institutions aim to foster knowledge exchange and cross-sectoral col- laboration and coordination at the project level for improving nutrition (Secure Nutrition Knowledge Platform, World Bank). Also, the Investing in the Early Years initiative, started in 2016, takes a holistic approach to the nature of interven- tions needed for the healthy physical and cognitive development of children. The initiative aims to help children reach their full potential through an increased emphasis on the good nutritional status of mothers, expecting mothers, and chil- dren especially in their first 1,000 days of life (including nine months in utero); exclusive and continued breastfeeding; health care; immunization; and proper feeding, accompanied by good hygiene practices, early stimulation and age- appropriate learning opportunities, nurturing care, and social protection that provides buffers from poverty and stress.2 All of these initiatives are based on the premise that the determinants of undernutrition are multisectoral and that the solution to undernutrition requires multisectoral approaches. The effectiveness and ultimate success of such multisectoral approaches toward reducing stunting, and in furthering progress toward the Sustainable Development Goals, depend on having a more holistic view of the inequities and gaps in access to adequate levels of the underlying determinants of nutri- tion, that is, care, food security, health, and environment (including WASH). The complex interdependence among the underlying determinants of nutri- tion is usually beyond the consideration of any given sector. On the one hand, the targeting of communities or the integration of nutritional considerations in existing interventions of the agricultural sector, for example, is unlikely to take into consideration the status of WASH services and facilities in the com- munities where these agricultural interventions are being considered. Consequently, the extent to which such nutrition-sensitive interventions can ultimately accelerate or enhance the impact of nutrition-specific interventions on key nutrition outcomes can be impeded considerably by the absence of adequate WASH facilities. On the other hand, the impact of agricultural inter- ventions on child nutrition could be enhanced considerably if those interven- tions were to be accompanied by simultaneous improvements in the water and sanitation facilities in the same communities. Thus, a more holistic approach to the targeting and nutrition sensitivity of interventions is likely to be better able to address the key underlying determinants of nutrition effectively, as well as reinforce the impacts of nutrition-specific interventions through the health sector. Introduction and Motivation   29 This report lays the groundwork for more effective multisectoral action on reducing stunting by analyzing and generating empirical evidence useful for informing the joint targeting of nutrition-sensitive interventions in countries in Sub-Saharan Africa. The report takes stock of the available data that are useful for applying some of the fundamental concepts of the UNICEF conceptual model for nutrition. These concepts include the critical role of the underlying drivers of nutrition, such as care, food security, health, and WASH, under- pinned by the basic determinants of nutrition such as income and the education level of the mother, among others. In the process, some important data limita- tions are identified, especially in relation to the availability of information on the different dimensions of food security and care in surveys that contain child anthropometric measures such as height and weight. Despite these data limita- tions, the analysis in the report serves to highlight the fact that much useful information can be extracted from the existing surveys, such as the Demographic and Health Surveys, in terms of getting a more holistic view of the multiple deprivations experienced by children and the prevalence of stunting. Information about the joint distribution of the underlying drivers of nutrition is essential for identifying particularly important gaps (or binding constraints) that, if addressed through joint targeting, can serve to strengthen the impacts on nutrition. With this background in mind, this report addresses four key questions: 1. What is the extent to which children have inadequate access to the underlying determinants of nutrition? Information available at the child level from the Demographic Health Surveys allows one to get a better sense about the joint distribution of the various determinants (or drivers) of undernutrition. Information about the joint distribution of inadequate access to the drivers of nutrition is essential for better targeting and more effective programs in the context of budgetary constraints. 2. What is the association of stunting (or low height-for-age), at any given time, with inadequate food and care practices, inadequate WASH, and inad- equate health? Is simultaneous access to an adequate level of one or more of the underlying determinants of nutrition associated with lower s ­ tunting? Empirical evidence confirming that simultaneous access to adequate lev- els of two or more nutrition drivers is associated with lower stunting vali- dates the importance of the joint targeting of interventions by different sectors in the geographic areas and populations therein where stunting is ­ prevalent. 3. If a multisectoral approach is not feasible, what is the sequencing of sector- specific interventions that could have the greatest impact on stunting? For example, if budgetary constraints prevent the joint coverage of geographic areas (or of populations within these areas) with high prevalence of stunting 30   ALL HANDS ON DECK by sectors such as agriculture, health, care, and WASH, in which sector should the limited resources be allocated? 4. What is the role of income growth and income variability on child stunting, and how does income interact with the underlying drivers of nutrition? The findings of this regional report are intended to stimulate and provide a blueprint for further analytic work that is operationally useful for the design of more effective multisectoral interventions on stunting at the coun- try level in Sub-Saharan Africa. The main findings of the report are based on data on children pooled across 33 countries, thereby reflecting relation- ships that prevail on average among stunting and access to the drivers of nutrition by children residing in different countries. Pooling data on chil- dren across many countries using common definitions and thresholds offers an advantage not only in terms of total sample size but also in that pooling increases the number of observations on children with access to adequate levels to different combinations of nutrition drivers. This advantage, how- ever, comes at some costs. Applying definitions and using thresholds that are common across all countries tend to minimize the role of country-specific factors. As a consequence, the relationship between stunting and access to nutrition drivers derived from the sample of children pooled across coun- tries may not necessarily reflect the relationship that prevails within any given country. In view of these considerations, chapter 6 of the report pro- vides an example applying the same methodology using data from only one country, Tanzania. In addition, appendix C of this report includes a brief for each of the 33 countries used in the report, summarizing access to the determinants of nutri- tion and their components as well as the simple correlation between stunting and the number of determinants. Appendix C also presents the country-specific estimates summarizing the relationship between stunting and access to different combinations of nutrition drivers. The report is structured as follows. • Chapter 2 describes the methodology and the data used in the analysis, along with the caveats and the value added of this study. • Chapter 3 provides descriptive statistics on the prevalence of stunting in Sub- Saharan African countries, as well as differences in the prevalence of stunting among children in urban and rural areas and children living in richer and poorer households within the countries studied. • Chapter 4 describes the empirical measures of the components of the under- lying drivers of nutrition and presents descriptive statistics on the extent to which children have access to adequate levels of one or more of the underly- ing determinants of nutrition one (question 1 above). Introduction and Motivation   31 • Chapter 5 contains a detailed investigation of the relationship between stunt- ing and access to one or more of the underlying determinants of nutrition, and the effects of access to different combinations of the drivers of nutrition (questions 2 and 3). • Chapter 6 provides an application of the same methodology used in chapters 2–5 of the report, using data from only one country, Tanzania. • Chapter 7 focuses on the complex relationship of the basic determinants of nutrition such as income growth, income stability, and parental education with child stunting (question 4). • Chapter 8 synthesizes the main findings of this report, with the lessons offered from successful country-level multisectoral initiatives on nutrition, to iden- tify the key ingredients of a multisectoral strategy to reduce undernutrition that maximizes the potential to produce the desired outcomes in practice. Notes 1. This 2014 figure is agreed upon by the United Nations Children’s Fund (UNICEF), the World Health Organization, and the World Bank. In statistical terms, a child is stunted if his/her height-for-age Z-score (HAZ) is less than or equal to 2 standard deviations below the median height of a healthy reference population (that is, HAZ ≤ –2). 2. Along parallel lines, initiatives aim to foster knowledge exchange and cross-sectoral collaboration and coordination at the project level for improving nutrition (Secure Nutrition Knowledge Platform, World Bank). References Black, R. E., C. G. Victora, S. P. Walker, Z. A. Bhutta, P. Christian, M. de Onis, M. Ezzati, S. Grantham-McGegor, J. Katz, R. Martorell, R. Uauy, and the Maternal and Child Nutrition Study Group. 2013. “Maternal and Child Undernutrition and Overweight in Low-Income and Middle-Income Countries.” Lancet 382 (9890): 427–51. http:// dx.doi.org/10.1016/S0140-6736(13)60937-X. Galasso, E., A. Wagstaff, S. Naudeau, and M. Shekar. 2017. “The Economic Costs of Stunting and How to Reduce Them.” World Bank Policy Research Note, PRN/17/05, World Bank, Washington, DC. Hoddinott, J., H. Alderman, J. R. Behrman, L. Haddad, and S. Horton. 2013. “The Economic Rationale for Investing in Stunting Reduction.” Maternal and Child Nutrition 9 (Suppl. 2): 69–82. Horton, S., M. Shekar, C. McDonald, A. Mahal, and B. J. Krystene. 2010. Scaling Up Nutrition: What Will It Cost? Washington, DC: World Bank. http://hdl.handle​ .net/10986/2685. 32   ALL HANDS ON DECK IFPRI (International Food Policy Research Institute). 2016. Global Nutrition Report 2016: From Promise to Impact: Ending Undernutrition by 2030. Washington, DC: International Food Policy Research Institute. Ruel, M. T., H. Alderman, and the Maternal and Child Nutrition Study Group. 2013. “Nutrition-Sensitive Interventions and Programmes: How Can They Help to Accelerate Progress in Improving Maternal and Child Nutrition?” Lancet 382 (9891): 536551. http://dx.doi.org/10.1016/S0140-6736(13)60843-0. Shekar, M., J. Kakietek, J. D. Eberwein, and D. Walters. 2016. An Investment Framework for Nutrition: Reaching the Global Targets for Stunting, Anemia, Breastfeeding, and Wasting. Washington DC: World Bank. Victora C. G., L. Adair, C. Fall, P. C. Hallal, R. Martorell, L. Richter, and H. S. Sachdev, for the Maternal and Child Undernutrition Group 2008. “Maternal and Child Undernutrition: Consequences for Adult Health and Human Capital.” Lancet 371 (9609): 340–57. World Bank. 2013. Improving Nutrition through Multisectoral Approaches. Washington, DC: World Bank. http://documents.worldbank.org/curated/en/625661468329649726​ /Improving-nutrition-through-multisectoral-approaches. ———. 2017. “Stunting Reduction in Sub-Saharan Africa.” HNP Discussion Paper, World Bank, Washington, DC. Chapter 2 Methodology and Data The United Nations Children’s Fund Conceptual Framework The United Nations Children’s Fund (UNICEF) conceptual framework sum- marized in figure 2.1 models undernutrition as the consequence of a variety of interlinked and interrelated factors. The causes of undernutrition are classified into three hierarchical categories: • The immediate causes • The underlying causes • The basic causes of undernutrition. In any given context, identification of the immediate causes of ­undernutrition (disease or inadequate dietary intake) is useful for guiding policy actions, espe- cially in situations of crisis. Disease and inadequate dietary intake, however, are typically consequences of a variety of underlying drivers that are interrelated. For conceptual simplicity, the underlying causes of undernutrition are them- selves grouped into the four clusters that manifest themselves mainly at the household level: (a) inadequate household food security, (b) inadequate care and feeding practices, (c) unhealthy household environment, and (d) inade- quate health services. The basic causes of undernutrition summarize the social, cultural, economic, and political context and the prevailing inequalities in the distribution of resources in the society. In combination, these contextual or structural factors play a fundamental role in the extent to which inequalities exist among households and their members in having adequate food security, care, and feeding practices; a healthy environment; and adequate health services (that is, the underlying causes of undernutrition). Since its conception, this conceptual framework has been revised and extended in various dimensions. Various international organizations have 33 34   ALL HANDS ON DECK Figure 2.1  Determinants of Child Malnutrition Long-term consequences: Adult height, cognitive ability, economic Short-term consequences: productivity, reproductive performance, Mortality, morbidity, disability overweight and obesity, metabolic and cardiovascular diseases Maternal and child undernutrition Immediate Inadequate dietary intake Diseases causes Underlying Household food Inadequate care and Unhealthy household Inadequate health causes insecurity feeding practices environment services Household access to adequate quantity and quality of resources: Land, education, employment, income, and technology Basic Inadequate financial, human, physical, and social capital causes Social, cultural, economic, and political context Source: Adapted from UNICEF 1990. adopted this framework as well. For example, the Food and Agriculture Organization of the United Nations (FAO) (2011) discusses adaptation of this framework for nutrition analysis. The U.S. Agency for International Development–Food and Nutrition Technical Assistance also adapted this framework (Riely et al. 1999). The World Food Programme (WFP) refers to the model as the Food and Nutrition Security Conceptual Framework in its Emergency Food Security Assessment Handbook (WFP 2009, 25). The latest reincarnation of the original UNICEF framework is the frame- work for actions to achieve optimum fetal and child nutrition and develop- ment from the Lancet Maternal and Child Nutrition Series (Black et al. 2013). This updated framework distinguishes between parental care practices (or  behaviors) and parental care resources . The former are considered an important component of the nutrition-specific interventions delivered pri- marily through the health system to address the immediate determinants of fetal and child nutrition and development; the latter are considered part of the nutrition-sensitive interventions delivered through sectors other than health) (see figure 2.2). In contrast, in the original version of the UNICEF model, Methodology and Data   35 Figure 2.2  Framework for Actions to Achieve Optimum Fetal and Child Nutrition and Development Benefits during the life course Morbidity and Cognitive, motor, and School performance Work capacity Adult stature mortality in childhood socioemotional development and learning capacity and productivity Obesity and noncommunicable diseases Nutrition-specific Optimum fetal and child nutrition and development Nutrition-sensitive interventions and programs and approaches programs • Agriculture and food • Adolescent health security and preconception Breastfeeding, Feeding and • Social safety nets nutrition Low burden nutrient-rich caregiving • Early child development • Maternal dietary of infectious foods, and practices, parenting, • Maternal mental health supplementation diseases eating routine and stimulation • Women’s empowerment • Micronutrient • Child protection supplementation or • Classroom education fortification • Water and sanitation Food security, Feeding and Access to, • Breastfeeding and • Health and family planning including caregiving and use of, complementary services availability, resources health services feeding economic at maternal, and a safe • Dietary access, and household, and and hygenic supplementation use of food community levels environment for children Building and enabling • Dietary environment diversification • Rigorous evaluations • Feeding behaviors Knowledge and evidence • Advocacy strategies and stimulation Politics and governance • Horizontal and vertical • Treatment of severe Leadership, capacity and financial resources coordination acute malnutrition Social, economic, political, and environmental • Accountability, incentives • Disease prevention context (national and global) regulation, and legislation and management • Leadership program • Nutrition • Capacity investments interventions • Domestics resource in emergencies mobilization Source: Black et al. 2013. the  distinction between care practices/behavior and care resources is not prominent, with care practices and care resources considered jointly as under- lying determinants of child undernutrition. Regardless of the adaptations and the extensions to the original framework, the underlying determinants of nutrition, consisting of food security, environment and health, and care, have remained at the core. With that background in mind, this report focuses primarily on the underly- ing determinants or drivers of nutrition as an entry point to a more systematic analysis of how different sectors can contribute to the decline in ­undernutrition— not only in theory but also in practice. 36   ALL HANDS ON DECK Food Security at the Household Level ­ ousehold The first underlying driver of nutrition is adequate food security at the h level. The “ideal” measure of food security consists of four broad dimensions: availability, access, utilization, and stability (over time) (Barrett 2009). Availability is associated with the supply side of food at the national (or regional, village, or local market) level, and it is measured on the basis of agricultural production relative to the size of consumption. Access, in contrast, is associated with the demand side. Conditional on what is available in the local market and the price at which it is available, what is the range of food choices open to house- holds given their incomes? Conceptually, it is this dimension of food security that has the strongest resonance with poverty and vulnerability—not only because of its direct relationship with income but also because of its links to broader issues of social and political enfranchisement. Food security of indi- vidual household members, for example, hinges on their social standing within the household almost as much as it does on the household’s overall ability to procure enough food (vulnerable groups within the household may include chil- dren, daughters, daughters-in-law, and the elderly). The utilization dimension brings to bear the quality dimension of the accessed food. Do households make good use of the food they are able to access? Are diets diverse enough to provide all the micro- and macronutrients necessary for healthy physiological and cogni- tive growth? Are cooking methods sanitary and healthy enough to preserve the nutritional attributes of the eaten food? Finally, stability in some ways cuts through the other three themes and captures how robust availability, access, and utilization are to seasonal patterns prevailing through the agricultural produc- tion cycle (planting versus harvesting season) or in the event of unexpected shocks such as political, economic, or climate or weather-related events. Adequate Care The second driver of nutrition is access to adequate care. This driver summa- rizes the ability of the primary caregiver to provide a safe and appropriate envi- ronment for the child to grow and develop. Ideally, it should be measured on the basis of the child’s caregivers’ (a) knowledge, practices, and beliefs regarding childcare; (b) health and nutritional status; (c) mental health, stress level, and self-confidence; (d) autonomy and control of resources; (e) workload and time constraints; and (f) social support received from family members and the com- munity (Engle, Menon, and Haddad 1999). Healthy Environment The third driver is access to a healthy environment, summarized by the water, sanitation, and hygiene (WASH) services and conditions prevailing in the household or in the community. This driver summarizes children’s exposure to pathogens in the physical environment where they live and is measured using Methodology and Data   37 the definitions adopted by the World Health Organization (WHO)/UNICEF Joint Monitoring Program and as part of monitoring the Sustainable Development Goals. Its components include (a) access to improved drinking water, (b) access to improved sanitation, (c) adequate handwashing practices, and (d) adequate disposal of child feces. Given that it is not only the child’s immediate environment (that is, the facilities at the dwelling unit) but also the child’s surrounding area (that is, the facilities in the immediate neighborhood) that affect the degree of exposure to pathogens, community-wide access to improved sanitation is a fundamental component. Adequate Health Care The fourth driver is access to adequate health care. This driver summarizes the child’s access to skilled medical care to minimize the effects of illness and pre- ventively address health issues, especially those linked with undernutrition, such as diarrheal diseases. The measure encompasses the availability and use of health care services for prenatal, birth, and postnatal care. Caveats and the Contributions of This Study Although the conceptual framework outlined above represents a useful way to think about the variety of determinants of child nutrition, it is also important to bear in mind the constraints encountered in quantifying and measuring the framework’s components. Smith and Haddad (2015) provide one of the first econometric applications of the UNICEF conceptual model of the underlying and basic determinants of nutrition using data from 116 developing countries collected from 1970 to 2012. The analysis in this report builds on that earlier study, with the difference that the focus of this study is at the child level in Sub-Saharan African countries rather than at the country level. The discussion in the following paragraphs outlines the value added of carrying out the analysis at the child level. The analysis in this report relies on recent Demographic and Health Surveys (DHS). The next section provides a more detailed description of the specific components of care, health, environment, and food security that can be con- structed using the data available in typical DHS. Although the data available in these surveys are far from being able to cover the full spectrum of factors con- sidered as important determinants of nutrition, for the few factors that are avail- able, such as the components of care or food security, it is not possible to distinguish between practice (or utilization) and availability or accessibility of the resource available. This is because the data collected in these types of sur- veys reflect only utilization (or behavior), which is conditional on the availabil- ity of the relevant resource. For example, the information that a child was 38   ALL HANDS ON DECK immunized reveals that the mother took the child to the available health center for the shots. If the child was not immunized, it does not imply that there is no health center in the community. In fact, there are two possibilities: first, it is possible that there is a health center in the community (that is, mother had access but faced constraints that prevented her from immunizing her child); second, it is possible that there is no health center within reasonable distance from the community (no access). Thus, coverage of stunted communities or stunted children by a given sector, and the extent to which sector-specific inter- ventions are targeted to stunted children, can be identified only if there was reported utilization of the sector service or facility by anyone in the community.1 Considering all the preceding qualifications and caveats, one question con- cerns the value added of this report in terms of its additions to the stock of knowledge and its implications for operations. First, the primary benefit derived from the proposed analysis is the fact that, in spite of the shortcom- ings associated with data sources such as the DHS in relation to measuring or categorizing the various components of the determinants of undernutrition, the information available at the child level in these surveys allows one to get a better sense about the joint distribution of the variety of determinants of undernutrition. Information about the joint distribution of these determi- nants is essential for identifying particularly important gaps (for example in water or sanitation) that may affect the impact of other nutrition-related interventions. As emphasized in the literature of multidimensional poverty, the “joint dis- tribution” of deprivations (or the lack of access to an adequate level of the vari- ous determinants of undernutrition) contains more information than the “marginal distribution” of deprivations (Alkire and Foster 2011a, 2011b; Atkinson 2003; Bourguignon and Chakravarty 2003; Duclos, Sahn, and Younger 2006). For example, a univariate analysis carried out independently by any given sector, such as agriculture, for specific determinants of undernutrition (for example, food security and dietary diversity), without taking into consideration other determinants of undernutrition typically in the purview of another sector (for example, water and sanitation), is unable to provide much guidance on the geographic areas where the interventions in the agriculture sector are likely to be less or more effective in terms of their effect on undernutrition. In contrast, a multivariate analysis of the simultaneous (or the lack of) access to adequate food security and water and sanitation services is able to provide a more holistic view and pinpoint better the groups of children and the geographic areas where these inadequacies are prevalent. This understanding enables the joint prioriti- zation of operations and improved cost efficiency of interventions aimed at con- tributing to the reduction of undernutrition in the agricultural and water and sanitation sectors. In addition, one of the potential contributions of this study Methodology and Data   39 is to highlight the shortcomings in the availability of data necessary for tracking the components of adequate food and care, health, and WASH. Second, the availability of child anthropometric measures together with information on the joint distribution of access (or the lack of access) to adequate levels for some of the key determinants of nutrition can shed light on the extent to which the underlying components are substitutes or complements for each other. Using a strict interpretation of the original UNICEF conceptual frame- work of the factors considered important for child nutrition (figure 2.1) or of the actions necessary to achieve optimal fetal and child nutrition and develop- ment (see figure 2.2), all of the components of the underlying determinants of nutrition, or the actions needed, appear to be strict complements. However, the extent to which specific components are complements or substitutes is mainly an empirical question that seems to have received very little attention in the field of nutrition. Third, the report also reevaluates the role of household income in child nutrition. Income is a basic determinant of nutrition that underpins much of the demand for the underlying drivers of nutrition. Specifically, two different dimensions of income are investigated: income level and income variability. A multisectoral approach aiming to increase access to the underlying determi- nants of nutrition needs to be cognizant of the fact that household constraints faced at different times may interact with utilization and behavior. The implica- tion worthy of serious policy consideration is that multisectoral interventions against undernutrition may be more effective if accompanied by broader devel- opment policies that mitigate the impacts of weather-related risks, such as crop or weather insurance policies. The analysis is based on 33 recent DHS from Sub-Saharan Africa. Of the 33 countries, 32 are part of the global Scaling Up Nutrition initiative.2 The main selection criterion for inclusion into the group of countries analyzed is that the survey had to be quite recent, that is collected in 2010 or later.3 DHS contain rather limited information on the components of food security and childcare. This shortcoming is balanced, however, by the availability of child anthropo- metric measures in all countries covered and the information on the joint dis- tribution of access (or lack of access) to adequate levels for some of the other underlying determinants of nutrition, such as access to improved WASH and to health services, according to internationally accepted standards. In contrast, other surveys such as the Living Standards Measurement Surveys contain more information on food and nonfood consumption and other dimensions of food security. Those other surveys also sometimes contain repeated observations (panel data) on the same children and households but with the result of more limited country coverage, lack of information on anthropometric measure- ments, or insufficient information for the construction of measures of WASH and health based on internationally accepted standards. 40   ALL HANDS ON DECK The 33 countries were also categorized into different groups for the pur- pose of making some useful and interesting comparisons. Each country is categorized as having high or low gross national income (GNI) per capita, high or low income inequality, and high or low average tariff rates on primary products. In addition, first wave countries, landlocked countries, and coun- tries or subnational regions in the Sahel region are identified. Within the set of countries that are designated as fragile, the countries are characterized by their degree of fragility. Table 2.1 lists the 33 countries included in the report along with their differ- ent income groupings: high and low GNI per capita countries based on the most recent GNI per capita and inequality index (Gini) for the country in the World Development Indicators. 4 The 33 countries in the study are divided into two groups, with 16 countries categorized as high GNI per capita or high Gini coun- tries and 17 countries as low GNI per capita or low Gini countries. The measure of income used is the purchasing power parity (PPP) GNI per capita. The high GNI per capita countries have per capita GNI ranging from 1,950 PPPs in Chad to 16,720 PPPs in Gabon. The low GNI per capita countries have GNI per capita ranging from 700 PPPs in Liberia to 1,920 PPPs in Zimbabwe. The Gini coeffi- cients in the high income inequality countries range from 43 (Nigeria) to 61  (Namibia). The Gini coefficients in the low income inequality countries range from 31.5 (Niger) to 42.8 (Ghana). Landlocked countries are those within the sample that do not border either the Atlantic Ocean or the Indian Ocean. The table includes 12 landlocked coun- tries and 21 nonlandlocked countries. The landlocked countries are Burkina Faso, Burundi, Chad, Ethiopia, Lesotho, Malawi, Mali, Niger, Rwanda, Uganda, Zambia, and Zimbabwe. High- and low-tariff countries are designated using the weighted average tariff on primary products. Countries with higher tariffs are presumed to have higher food prices on average, which could contribute to child undernutrition, especially in poorer countries. The tariffs in the high-tariff countries range from the Democratic Republic of Congo’s 11 percent to Chad’s 21 percent. In the low-tariff countries, tariff rates range from Namibia’s 1.1 percent to Togo’s 10.3 percent. The share of the diet from cereals and starchy roots in the national diet is used to classify countries as having a high-starch diet or a low-starch diet. If the share of starches is above 66 percent, then a country is classified as a high-starch diet country.5 The Sahel region includes the following subnational regions: Centre-Nord, Est, Nord, and Sahel in Burkina Faso; Extrême-Nord, and Nord in Cameroon; Batha, Guéra, Hadjer-Lamis, Kanem, Lac, Ouaddaï, Wadi Fira, Bahr el Gazel, and Sila in Chad; Affar, Amhara, and Tigray in Ethiopia; Kayes, Koulikoro, Sikasso, Segou, Mopti, and Bamako in Mali; all regions except Agadez in Niger; North regions in Nigeria; and all regions except Ziguinchor, Kolda, Kedougou, and Sedhiou in Senegal. The first wave countries are those countries designated Methodology and Data   41 Table 2.1  Characterization of Countries Included in the Study DHS survey GNI Gini Tariff Starch 1st Country year (latest) (latest) (latest) in diet Landlocked Sahela Wave Fragility Angola 2015–16 High Low High Low Benin 2011–12 High High High High Burkina Faso 2010 Low Low High Low √ √ √ Burundi 2016–17 Low Low Low √ High Cameroon 2011 High High Low Low √ √ Chad 2014–15 High High High Low √ √ Low Comoros 2012 Low High Low Low Congo, Dem. Rep. 2013 Low Low High High Congo, Rep. 2011–12 High High High Low Côte d’Ivoire 2011–12 High High High High √ High Ethiopia 2016 Low Low Low High √ √ √ Gabon 2012 High Low High Low Gambia, The 2013 Low High High Low Low Ghana 2014 High Low High Low Guinea 2012 Low Low High Low Kenya 2014 High High Low Low Lesotho 2014 High High High High √ Liberia 2013 Low Low Low High High Madagascar 2008–09 Low Low Low High √ High Malawi 2015–16 Low High Low High √ √ Mali 2012–13 High Low High High √ √ √ High Mozambique 2011 Low High Low High √ Namibia 2013 High High Low Low Niger 2012 Low Low High Low √ √ √ Nigeria 2013 High High Low Low √ √ Rwanda 2014–15 Low High Low Low √ √ Senegal 2015 High Low High Low √ Sierra Leone 2013 Low Low High Low High Tanzania 2015–16 High Low Low Low √ Togo 2013–14 Low High Low High Low Uganda 2011 Low Low Low Low √ Zambia 2013–14 High High Low High √ Zimbabwe 2015 Low High Low Low √ Low Source: All data, except for diet composition, were downloaded from http://wdi.worldbank.org/tables, accessed January 2018. The GNI measure (in PPP) is from Table 1.1, the Gini measure is from Table 2.9, and tariff rates (weighted) from Table 6.6. The degree of fragility is found in Table 5.8. The share of diet from cereals and starchy roots in the national diet come from the Food and Agriculture Organization of the United Nations (http://www.fao.org/3/a-i4175e.pdf). Note: Above 66 percent is considered high. a. The Sahel region includes the following subnational regions: Centre-Nord, Est, Nord, and Sahel from Burkina Faso; Extrême-Nord, and Nord from Cameroon; Batha, Guéra, Hadjer-Lamis, Kanem, Lac, Ouaddaï, Wadi Fira, Bahr el Gazel, and Sila from Chad; Affar, Amhara, and Tigray from Ethiopia; Kayes, Koulikoro, Sikasso, Segou, Mopti, and Bamako from Mali; all regions except Agadez from Niger; North regions from Nigeria; and all regions except Ziguinchor, Kolda, Kedougou, and Sedhiou from Senegal. 42   ALL HANDS ON DECK by the World Bank. Fragile country designations are based on the World Bank list of fragile countries in 2015. For this report, the fragile countries are catego- rized as high or low fragility on the basis of their resource allocation index by the International Development Association (IDA).6 The asset or wealth index that is publicly available with the DHS data includes WASH-related variables as part of its components. Given the focus of the report on WASH as an underlying determinant of nutrition, a new asset index was estimated excluding WASH from the components of the index, and this original index is used throughout the report. All households in each country survey are used (not just households with children) for the construction of this index. Thus, the “new” asset or wealth index is estimated on the basis of the first principal component of ownership of radio, refrigerator, bike, motorcycle or ­ scooter, car or truck, and mobile phone; whether the dwelling has electricity; the number of rooms per household member; the type of floor material; the type of wall material; and the type of roof material. In most of the regressions, the wealth of the household is represented by five binary variables indicating the quintile of the value of the wealth index of the household in the national distri- bution of the asset index. In contrast, in chapter 6, where a more detailed analy- sis is carried out on the association between household income as proxied by the value of household asset index and height-for-age Z-scores (HAZ) the per- centile of the value of the asset index is used as a continuous variable. The wealth distribution is determined for each country separately. In some of the figures, the bottom and top quintiles are used to explore differences in stunting and access to the underlying drivers of nutrition by wealth. Notes 1. Also, there is no information available in the DHS on whether there are ongoing nutrition-specific or nutrition-sensitive interventions that children are exposed to. 2. More information on the SUN initiative and the countries participating in the SUN initiative can be found at http://scalingupnutrition.org/sun-countries/about​ -sun-countries/. 3. An execption is Madagascar, which was included in the study even though the DHS data were collected in 2008–09. 4. The data were downloaded from http://wdi.worldbank.org/tables, in January 2018. The GNI measure is from Table 1.1, the Gini measure is from Table 2.9, and tariff rates from Table 6.6. The degree of fragility is found in Table 5.8. 5. Starch shares in the country diet were accessed from the Food and Agriculture Organization of the United Nations, http://www.fao.org/3/a-i4175e.pdf. 6. The index considers 16 criteria in peace building and peacekeeping, battle-related deaths, intentional homicides, military expenditure, crime, and informality. Information on the construction of this index can be found at http://pubdocs.world​ bank.org/en/600961531149299007​/CPIA-Criteria-2017.pdf. Methodology and Data   43 References Alkire, Sabina, and James Foster. 2011a. “Counting and Multidimensional Poverty Measurement.” Journal of Public Economics 95 (7): 476–87. ———. 2011b. “Understandings and Misunderstandings of Multidimensional Poverty Measurement.” Journal of Economic Inequality 9 (2): 289–314. Atkinson, A. B. 2003. “Multidimensional Deprivation: Contrasting Social Welfare and Counting Approaches.” Journal of Economic Inequality 1 (1): 51–65. Barrett, C. B. 2009. “Measuring Food Insecurity.” Science 327 (5967): 825–28. Black, R. E., C. G. Victora, S. P. Walker, Z. A. Bhutta, P. Christian, M. de Onis, M. Ezzati, S. Grantham-McGegor, J. Katz, R. Martorell, R. Uauy, and the Maternal and Child Nutrition Study Group. 2013. “Maternal and Child Undernutrition and Overweight in Low-Income and Middle-Income Countries.” Lancet 382 (9890): 427–51. http:// dx.doi.org/10.1016/S0140-6736(13)60937-X. Bourguignon, F., and S. R. Chakravarty. 2003. “The Measurement of Multidimensional Poverty.” Journal of Economic Inequality 1 (1): 25–49. Duclos, J. Y., D. E. Sahn, and S. D. Younger. 2006. “Robust Multidimensional Poverty Comparisons.” Economic Journal 116 (514): 943–68. Engle, P. L., P. Menon, and L. Haddad. 1999. “Care and Nutrition: Concepts and Measurement.” World Development 27 (8): 1309–37. FAO (Food and Agriculture Organization of the United Nations). 2011. “Developing a Response Analysis Framework for Food Security Emergencies.” Discussion Papers, FAO, Rome. http://www.fao.org/fileadmin/templates/tc/tce/pdf/Response_Analysis​ _Framework_Discussion_Papers.pdf. Riely, F., N. Mock, B. Cogill, L. Bailey, and E. Kenefick. 1999. “Food Security Indicators and Framework for Use in the Monitoring and Evaluation of Food Aid Programs.” Food Security and Nutrition Monitoring (IMPACT) Project for the U.S. Agency for International Development, Arlington, VA. https://pdf.usaid.gov/pdf_docs/Pnacg170.pdf. Smith, L. C., and L. Haddad. 2015. “Reducing Child Undernutrition: Past Drivers and Priorities for the Post-MDG Era.” World Development 68 (April): 180–204. UNICEF (United Nations International Children’s Emergency Fund). 1990. Strategy for Improved Nutrition of Women and Children in Developing Countries. A UNICEF Policy Review. New York: UNICEF. WFP (World Food Programme). 2009. Emergency Food Security Assessment Handbook. Rome: WFP. http://documents.wfp.org/stellent/groups/public/documents/manual​ _guide_proced/wfp203244.pdf. Chapter 3 Stunting in Sub-Saharan African Countries Figure 3.1 presents the stunting rates of younger (0–24 months) and older chil- dren (24–59 months) in each of the Sub-Saharan Africa countries in the study. This figure shows quite clearly that the stunting rates are higher for older chil- dren (24–59 months) than for younger children (0–24 months). This pattern is a consequence of the growth faltering that occurs after birth. Growth faltering is the rapid decline in height- and weight-for-age of children in the first two years of life and is common in many developing countries. Growth faltering among children was first documented in a study by Shrimpton et al. (2001); it is to a large extent in response to their findings that several global health policy and information campaigns with emphasis on the first 1,000 days window have been initiated (Prentice et al. 2013). Figure 3.2 presents the cross-sectional age profile of height-for-age Z-scores (HAZs) for children between 0 and 59 months of age in the 33 countries in Sub- Saharan Africa analyzed in this report. As can be seen clearly, HAZs decrease rapidly in the first 21 months of life; continue decreasing at a slower pace after 21 months, bottom out at about 33 months of age; and increase at a very slow rate afterward, with only minor fluctuations.1 Figure 3.3 presents the growth faltering curves for selected groups of countries (first wave versus others and Sahel versus others) as well as urban versus rural areas within countries, boys versus girls within countries, and different socioeco- nomic groups within countries. Growth faltering is slightly faster in first wave countries and in countries in the Sahel region, which highlights the urgency for policy action in these groups of countries. Also, the differences in the pattern of child growth faltering with maternal characteristics such as the level of education of the mother or the level of wealth of the household (for example, bottom [B20] or top 20 [T20] percent of the wealth distribution) attest to the important role of the basic determinants in child stunting and underline the need to also improve maternal and income conditions of the households with stunted children. 45 46   ALL HANDS ON DECK Figure 3.1  Stunting Prevalence among Younger (0–23 Months) and Older (24–59 Months) Children Burundi Mozambique Madagascar Zambia Benin Angola Niger Sierra Leone Rwanda Malawi Nigeria Congo, Dem. Rep. Comoros Ethiopia Mali Tanzania Uganda Chad Lesotho Burkina Faso Zimbabwe Côte d’Ivoire Cameroon Kenya Liberia Congo, Rep. Gambia, The Guinea Togo Senegal Namibia Gabon Ghana 0 10 20 30 40 50 60 70 Stunting rate per age group (%) Children ages 0–23 months Children ages 24–59 months Source: Estimates based on data for children under 60 months of age from 33 country Demographic Health Surveys (DHS) from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: Sampling weights were standardized across the 33 DHS following the procedure outlined by DHS. The sampling weight of the mother was multiplied by the ratio of the total number of women 15 to 49 years of age in the surveyed country to the number of women (15 to 49 years of age) surveyed. The total number of women was obtained from UN DESA 2015. Given the critical importance of intervening earlier in a child’s life to prevent growth faltering, most of the analysis and discussion in this report will focus on the prevalence of stunting and the underlying determinants of stunting among children between 0 and 23 months of age. Stunting in Sub-Saharan African Countries   47 Figure 3.2  Growth Faltering (Children 0–59 Months) 0 −0.5 Average height-for-age Z-score −1.0 −1.5 −2.0 0 6 12 18 24 30 36 42 48 54 60 Age (months) Source: Estimates based on children on data for children under 60 months of age from 33 country Demographic Health Surveys (DHS) from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: Sampling weights were standardized across the 33 DHS following the procedure outlined by DHS. The sampling weight of the mother was multiplied by the ratio of the total number of women 15 to 49 years of age in the surveyed country to the number of women (15 to 49 years of age) surveyed. The total number of women was obtained from UN DESA 2015. The stunting prevalence among children under 24 months of age in the pooled sample of countries is 28 percent; however, differences exist in the dis- tribution of height-for-age and stunting for different population groups. Figure 3.4 presents the probability density functions of HAZs for rural and urban areas in the sample of 33 countries and for different socioeconomic groups. The area under the probability density function to the left of the red vertical line at the HAZ of −2 provides a visual representation of the fraction of the child population that is stunted. Rural children are more likely to be stunted than urban children (see panel a in figure 3.4). Nearly 31 percent of rural chil- dren are stunted, compared with 22 percent of urban children. Both boys and children of mothers with fewer than seven years of education are more likely to be stunted than girls or children of mothers with at least seven years of educa- tion. The largest difference in the distribution of HAZs analyzed is by household wealth. About 34 percent of children from the B20 families are stunted, whereas only 17 percent of children from the T20 families are stunted. 48   ALL HANDS ON DECK Figure 3.3  Growth Faltering, by Socioeconomic Characteristics (Children 0–59 Months) a. First wave countries versus others b. Countries in the Sahel region versus others 0 0 Average height-for-age Average height-for-age –0.5 –0.5 –1.0 Z-score Z-score –1.0 –1.5 –1.5 –2.0 –2.0 –2.5 0 6 12 18 24 30 36 42 48 54 60 0 6 12 18 24 30 36 42 48 54 60 Age (months) Age (months) First wave Not in first wave Sahel region Not in Sahel region c. Urban versus rural d. Boys versus girls 0 0 Average height-for-age Average height-for-age –0.5 –0.5 Z-score Z-score –1.0 –1.0 –1.5 –1.5 –2.0 –2.0 0 6 12 18 24 30 36 42 48 54 60 0 6 12 18 24 30 36 42 48 54 60 Age (months) Age (months) Rural Urban Boy Girl e. Top quintile versus bottom quintile f. Educated mother versus less educated mother 0 0 Average height-for-age Average height-for-age –0.5 –0.5 –1.0 Z-score Z-score –1.0 –1.5 –1.5 –2.0 –2.0 –2.5 0 6 12 18 24 30 36 42 48 54 60 0 6 12 18 24 30 36 42 48 54 60 Age (months) Age (months) Bottom quintile Mother’s education < 7 years Top quintile Mother’s education > 7 years Source: Estimates based on data for children under 60 months of age from 33 country Demographic Health Surveys (DHS) from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: Sampling weights were standardized across the 33 DHS following the procedure outlined by DHS. The sampling weight of the mother was multiplied by the ratio of the total number of women 15 to 49 years of age in the surveyed country to the number of women (15 to 49 years of age) surveyed. The total number of women was obtained from UN DESA 2015. Stunting in Sub-Saharan African Countries   49 Figure 3.4  Distribution of Height-for-Age for Different Populations a. Urban versus rural b. Boys versus girls 0.25 0.25 0.20 0.20 0.15 0.15 Density Density 0.10 0.10 0.05 0.05 0 0 −6 −4 −2 0 2 4 6 −6 −4 −2 0 2 4 6 Average height-for-age Z-score Average height-for-age Z-score Urban Rural Boys Girls c. Educated mother versus less educated mother d. Top quintile versus bottom quintile 0.25 0.30 0.20 0.20 0.15 Density Density 0.10 0.10 0.05 0 0 −6 −4 −2 0 2 4 6 −6 −4 −2 0 2 4 6 Average height-for-age Z-score Average height-for-age Z-score Mother’s education > 7 years Bottom quintile Mother’s education < 7 years Top quintile Source: Estimates based on children under 24 months of age from 33 country Demographic Health Surveys (DHS) from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: Sampling weights were standardized across the 33 DHS, following the procedure outlined by DHS. The sampling weight of the mother was multiplied by the ratio of the total number of women 15 to 49 years of age in the surveyed country to the number of women (15 to 49 years of age) surveyed. The total number of women was obtained from UN DESA 2015. The main patterns summarized above also hold within each of the countries. Country-specific estimates of stunting prevalence, however, highlight not only the variability in stunting rates between countries but also the large differences within countries, particularly in relation to location (rural or urban) and the level of house- hold wealth (B20 versus T20). Stunting among younger children (0–23 months) is high across the region; only Ghana (12.7 percent) and Gabon (14.8 percent) have stunting rates below 15 percent. Nearly one-half of the children under 24 months of age are stunted in Burundi (46.3 percent) (see figure 3.5). Also, across the countries studied, urban children tend to be less likely to be stunted than rural children. Liberia and Benin are exceptions; the point 50   ALL HANDS ON DECK Figure 3.5  Stunting Prevalence, by Country (Children 0–23 Months) Burundi Madagascar Mozambique Zambia Benin Angola Niger Sierra Leone Rwanda Malawi Nigeria Congo, Dem. Rep. Comoros Ethiopia Mali Tanzania Uganda Chad Lesotho Burkina Faso Zimbabwe Côte d’Ivoire Cameroon Kenya Liberia Congo, Rep. Gambia, The Guinea Togo Senegal Namibia Gabon Ghana 0 10 20 30 40 50 Stunting rate (%) Source: Estimates based on children under 24 months of age from 33 country Demographic Health Surveys (DHS) from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: Sampling weights were standardized across the 33 DHS, following the procedure outlined by DHS. The sampling weight of the mother was multiplied by the ratio of the total number of women 15 to 49 years of age in the surveyed country to the number of women (15 to 49 years of age) surveyed. The total number of women was obtained from UN DESA 2015. For this graph, children are considered 23 months or younger. estimates for stunting are higher for urban children than for rural children (see figure 3.6). The prevalence of stunting is higher among boys than among girls (see figure 3.7). This difference is generally attributed to genetic factors; parental dis- ­ crimination in the allocation of household resources in favor of boys over girls may still be practiced. The prevalence of stunting is also higher among children Stunting in Sub-Saharan African Countries   51 Figure 3.6  Stunting Prevalence in Urban and Rural Areas within Each Country (Children 0–23 Months) a. Stunting rate, urban versus rural b. Difference in stunting rate, urban versus rural Burundi Burundi Madagascar Angola Angola Mali Mozambique Rwanda Zambia Guinea Benin Gabon Nigeria Nigeria Niger Congo, Dem. Rep. Rwanda Malawi Sierra Leone Togo Malawi Niger Congo, Dem. Rep. Cameroon Comoros Ethiopia Mali Burkina Faso Tanzania Tanzania Ethiopia Comoros Uganda Côte d’Ivoire Cameroon Namibia Côte d’Ivoire Gambia, The Lesotho Madagascar Burkina Faso Mozambique Chad Lesotho Zimbabwe Zambia Gabon Kenya Kenya Chad Congo, Rep. Uganda Gambia, The Sierra Leone Togo Senegal Guinea Congo, Rep. Senegal Ghana Liberia Zimbabwe Namibia Benin Ghana Liberia 0 10 20 30 40 50 −10 0 10 20 30 Stunting rate (%) Percentage difference, Urban Rural urban to rural Source: Estimates based on children under 24 months of age from 33 country Demographic Health Surveys (DHS) from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: Sampling weights were standardized across the 33 DHS, following the procedure outlined by DHS. The sampling weight of the mother was multiplied by the ratio of the total number of women 15 to 49 years of age in the surveyed country to the number of women (15 to 49 years of age) surveyed. The total number of women was obtained from UN DESA 2015. For this graph, children are considered 23 months or younger. in households at the B20 of the distribution of assets within countries compared to children in households at the T20 of the distribution (see figure 3.8). Finally, children whose mothers have fewer than seven years of education are  more likely to be stunted than children with mothers who have at least seven years of education (see figure 3.9). Liberia is again an exception; children 52   ALL HANDS ON DECK Figure 3.7  Stunting Prevalence among Boys and Girls within Each Country (Children 0–23 Months) a. Stunting rate, boys b. Difference in stunting rate, versus girls boys versus girls Burundi Rwanda Madagascar Côte d’Ivoire Mozambique Comoros Zambia Lesotho Benin Angola Rwanda Liberia Angola Zambia Sierra Leone Gambia, The Malawi Benin Comoros Malawi Niger Madagascar Congo, Dem. Rep. Burundi Nigeria Kenya Lesotho Mozambique Côte d’Ivoire Sierra Leone Tanzania Gabon Ethiopia Cameroon Uganda Burkina Faso Mali Congo, Dem. Rep. Burkina Faso Guinea Zimbabwe Tanzania Cameroon Senegal Chad Zimbabwe Kenya Nigeria Liberia Ghana Gambia, The Uganda Guinea Namibia Congo, Rep. Togo Togo Mali Senegal Niger Gabon Congo, Rep. Namibia Ethiopia Ghana Chad 0 10 20 30 40 50 0 5 10 15 Stunting rate (%) Percentage difference, boys to girls Boys Girls Source: Estimates based on children under 24 months of age from 33 country Demographic Health Surveys (DHS) from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: Sampling weights were standardized across the 33 DHS, following the procedure outlined by DHS. The sampling weight of the mother was multiplied by the ratio of the total number of women 15 to 49 years of age in the surveyed country to the number of women (15 to 49 years of age) surveyed. The total number of women was obtained from UN DESA 2015. For this graph, children are considered 23 months or younger. there with more educated mothers are, on average, more likely to be stunted than children of mothers who are less educated. Nigeria and Burundi have the largest differences in the probability of being stunted depending on the mother’s educational attainment. Stunting in Sub-Saharan African Countries   53 Figure 3.8  Stunting Prevalence, by Wealth Quintile within Each Country (Children 0–23 Months) a. Stunting rate, top quintile b. Difference in stunting rate, top quintile versus bottom quintile versus bottom quintile Burundi Burundi Madagascar Angola Angola Gabon Zambia Rwanda Mozambique Mozambique Rwanda Nigeria Nigeria Ethiopia Benin Côte d’Ivoire Malawi Cameroon Comoros Lesotho Sierra Leone Malawi Ethiopia Guinea Niger Congo, Rep. Côte d’Ivoire Comoros Gabon Tanzania Tanzania Mali Cameroon Kenya Mali Niger Burkina Faso Zambia Chad Ghana Congo, Rep. Burkina Faso Kenya Togo Congo, Dem. Rep. Senegal Lesotho Zimbabwe Zimbabwe Gambia, The Gambia, The Congo, Dem. Rep. Senegal Namibia Uganda Sierra Leone Guinea Madagascar Liberia Chad Togo Liberia Ghana Uganda Namibia Benin 0 10 20 30 40 50 60 0 10 20 30 Stunting rate (%) Percentage difference, top quintile to bottom quintile Top quintile Bottom quintile Source: Estimates based on children under 24 months of age from 33 country Demographic Health Surveys (DHS) from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: Sampling weights were standardized across the 33 DHS, following the procedure outlined by DHS. The sampling weight of the mother was multiplied by the ratio of the total number of women 15 to 49 years of age in the surveyed country to the number of women (15 to 49 years of age) surveyed. The total number of women was obtained from UN DESA 2015. For this graph, children are considered 23 months or younger. 54   ALL HANDS ON DECK Figure 3.9  Stunting Prevalence, by Mother’s Education within Each Country (Children 0–23 Months) a. Stunting rate, more educated mother b. Difference in stunting rate, more educated versus less educated mother mother versus less educated mother Burundi Nigeria Madagascar Burundi Mozambique Niger Zambia Burkina Faso Angola Angola Nigeria Mozambique Benin Lesotho Lesotho Cameroon Malawi Chad Rwanda Rwanda Niger Namibia Congo, Dem. Rep. Congo, Dem. Rep. Sierra Leone Ghana Tanzania Ethiopia Comoros Congo, Rep. Zimbabwe Comoros Ethiopia Kenya Uganda Gabon Mali Zimbabwe Kenya Togo Cameroon Gambia, The Chad Côte d’Ivoire Congo, Rep. Malawi Burkina Faso Tanzania Côte d’Ivoire Guinea Namibia Zambia Gambia, The Senegal Togo Mali Liberia Uganda Guinea Sierra Leone Gabon Madagascar Senegal Benin Ghana Liberia 0 10 20 30 40 50 –5 0 5 10 15 20 Stunting rate (%) Percentage difference, more educated mother versus Mother’s education > 7 years less educated mother Mother’s education < 7 years Source: Estimates based on children under 24 months of age from 33 country Demographic Health Surveys (DHS) from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: Sampling weights were standardized across the 33 DHS, following the procedure outlined by DHS. The sampling weight of the mother was multiplied by the ratio of the total number of women 15 to 49 years of age in the surveyed country to the number of women (15 to 49 years of age) surveyed. The total number of women was obtained from UN DESA 2015. For this graph, children are considered 23 months or younger. Note 1. For a recent cross-country study on the determinants of growth faltering, see Rieger and Trommlevora (2016). The growth faltering curves are derived using a local poly- nomial smooth of HAZ with the Stata command “lpoly”. Stunting in Sub-Saharan African Countries   55 References Prentice, A. M., K. A. Ward, G. R. Goldberg, L. M. Jarjou, S. E. Moore, A. J. Fulford, and A. Prentice. 2013. “Critical Windows for Nutritional Interventions against Stunting.” American Journal of Clinical Nutrition 97 (5): 911–18. Rieger, M., and S. K. Trommlerova. 2016. “Age-Specific Correlates of Child Growth.” Demography 53 (1): 241–67. Shrimpton, R., C. G. Victora, M. de Onis, R. C. Lima, M. Blossner, and G. Clugston. 2001. “Worldwide Timing of Growth Faltering: Implications for Nutritional Interventions.” Pediatrics 107 (5): E75. UN DESA (United Nations Department of Economic and Social Affairs). 2015. World Population Prospects 2015. New York: United Nations. Chapter 4 The Underlying Determinants of Nutrition Measurement is very important for a proper diagnosis of the longer-run con- straints to reducing chronic undernutrition in any given context. Each of the three drivers of nutrition is inherently multidimensional, making measurement difficult and costly. This section briefly surveys the ideal measures of the drivers of nutrition and the actual measures used on the basis of the information avail- able in the Demographic Health Surveys (DHS). The large gaps between the ideal measures and what is actually available in national surveys underline the urgent need to include in nutrition surveys more and better measures of the different dimensions of food security at the household and ideally the individual level as well as information on childcare practices (see box 4.1). In contrast, relatively more data seem to be available for the measurement of some of the components of environment and health, with the latter being measured almost as well as one could hope for. This level of availability is probably a reflection of the trade-offs involved between more complete or better measurement and costs. Given con- strained budgets for surveys collecting information on the different dimensions of nutrition, greater emphasis may be placed on collecting information on envi- ronment and health and, to some degree, on childcare: information on these dimensions can be collected at lower cost, and perhaps even more reliably, com- pared to the cost of a detailed survey collecting information on food availability, access, and utilization at the household or individual level. Measures of Food Security and Child Care Table 4.1 summarizes the various indicators of food security, each capturing different dimensions of the ideal food security measure and the indicators that can be constructed from the DHS in the countries studied. A detailed descrip- tion of the different indicators of food security is contained in Tiwari, Skoufias, and Sherpa (2013). 57 58   ALL HANDS ON DECK BOX 4.1 Additional Questions in Surveys for Better Adequacy Measures Although the Demographic and Health Surveys (DHS) collect sufficient information to capture some aspects of all four determinants of nutrition—food security, child care practices, access to health care, and access to water and sanitation—additional information to build more robust measures would be useful. Given that chronic malnu- ­ trition, such as stunting, is the product of a poor nutritional environment during a period, the conditions at the time of the survey may not reflect the historical nutritional experience. Longitudinal data, collecting information from various times, would give a richer picture of not only the current nutritional conditions but also those that have led up to the current state. Furthermore, longitudinal data on a specific child would also increase the tools available for causal inferences. Additional specific measures could also complement those already collected for a more robust description of the child’s access to nutrition determinants. For food security, the currently available data measure the utilization aspect of food security but do not measure availability or access. Whereas availability measures the national, or regional, conditions of how much food is available for consumption, access is a household-specific measure that could be incorporated into surveys. One measure of access is the Household Food Insecurity Access Scale (HFIAS), which aims to capture the changes in food consumption patterns and reflect the severity of food insecurity faced by households because they lack or have limited resources to access food (Coates, Swindale, and Bilinksy 2007). Some surveys, for example the 2009 Guatemala DHS, already incorporate questions allowing for the partial assessment of households’ food insecurity. Additionally, questionnaires with expanded food categories reflecting the local cuisine may better capture the true dietary diversity of the child. In Uganda, for example, the questionnaire specifically asked if the child ate a variety of meats, such as pigeon or duck, not usually included in the description of meat or a variety of starches, such as specific types of bananas. The resulting average dietary diversity measure was much higher than in countries where the surveys were not as specific, even though in any survey pigeon or banana could be included in the given categories. It may be that Ugandan chil- dren truly have much better access to food diversity, or it could be that the addi- tional prompting elicited different information. Collecting information, on a large scale, of many of the child care practices—such as caregiver responsiveness during feedings, caregiver and child interactions, observed time in child care, and child appearance in a public place—would be difficult to imple- ment. However, questions could elicit information on the mother’s knowledge and beliefs of proper feeding practices and milestones, the mother’s access and source of such information, recalled time in child care, and the characteristics of alternative care- givers who are with the child for extended periods. Access to health care and access to sanitation and drinking water are relatively well covered in the surveys used. Apart from collecting information on the location of the (continued next page) The Underlying Determinants of Nutrition   59 Box 4.1 (continued) closest health care facility to measure the availability of health care or the levels of vari- ous pathogens in the household environment to measure the actual exposure risk to illnesses, the information collected in a typical DHS is sufficient for relatively robust measures. For health care, the only potentially easily added measure would be the fre- quency of growth checkup visits that the child has been to and the quality of feedback from these visits to the mother. Most surveys collect information only on the first post- natal visit but not on any subsequent follow-up visits specifically focusing on the child’s growth and development. Table 4.1  Components of Food Security Measured Indicators Data available Children’s Dietary Diversity Score (CDDS) Yes Mom’s Dietary Diversity Score (MDDS) No Minimum Acceptable Diet (for children 6–24 months) Yes Household Food Insecurity Access Scale (HFIAS) No Food Consumption Score (FCS) No Relative prices of different food groups No Starchy staple ratio or the proportion of household calories derived from starchy staples No The DHS datasets used in this study contain sufficient information to con- struct a dietary diversity score for the child and the minimum acceptable diet measure, but they lack information relevant for the other dimensions of food security. The minimum acceptable diet combines dietary diversity, breastfeed- ing, and meal frequencies, whereas the dietary diversity score depends on the age of the child. Considering the absence of more data on different dimensions of food security, it is encouraging that the literature suggests that dietary diver- sity holds promise as a means of measuring household food security (defined as household energy availability) (Ruel 2002). Hoddinott and Yohannes (2002), for example, find strong associations between dietary diversity and per capita expenditures on food, as well as between dietary diversity and number of calo- ries consumed from nonstaples. Adequate care measures the capacity of the child’s caregiver to provide a healthy environment for the child to grow up in. Ideally, the measure is based on information on (a) the caregiver’s education, knowledge, and beliefs; (b) the health and nutritional status of the caregiver; (c) the mental health, lack of stress, and self-confidence of the caregiver; (d) the caregiver’s auton- omy and control of resources; (e) the workload and time constraints of the caregiver; and (f) the social support received by the caregiver from family 60   ALL HANDS ON DECK members and the community. The ideal components of an adequate care mea- sure include information on the following: • Workload and time availability of caregiver –– Observed time in childcare (observed in sample of time or continuously) –– Recalled time in childcare (24-hour recall) –– Quality of care during work time (characteristics of alternate caregivers, for example, age and gender) • Social support for caregiver –– Availability of alternate caregivers –– Community support (assessment of community institutions for childcare feeding programs, childcare programs) • Psychosocial care –– Caregiver/child interactions—naturalistic observation of caregiver and child for a short period (code variables such as delay to respond, type of response, and level of vocalization by caregiver and child) –– Child appearance (rating of appearance either in a public place or over a period of visits) –– Caregiver’s understanding of motor milestones • Caring behavior: Breastfeeding and complementary feeding, health seek- ing, and hygiene –– Caregiver feeding behavior (observation of one or more eating episodes) –– Caregiver responsiveness during feeding episodes –– Frequency of behavior such as feeding, number of spoonfuls, and number of touches –– Breastfeeding practices (for example, exclusive breastfeeding up to six months, early initiation of breastfeeding, breastfeeding at two years) –– Introduction of solid/semisolid/soft foods at six to eight months –– Child feeding index (constructed from DHS data using the following yes/ no questions: current breastfeeding, use of bottles, dietary diversity, and feeding/meal frequency) –– Taking a child to a health clinic for treatment of illness –– Maternal handwashing with soap • Maternal caregiver’s education, knowledge, and beliefs –– Years of schooling –– Literate/illiterate (self-reported, simple test, or existing data) The Underlying Determinants of Nutrition   61 –– Beliefs and knowledge about initiation of breastfeeding –– Beliefs about complementary feeding—timing, types, and control of intake The DHS contains information on only a few of the caring behaviors, namely, some information on breastfeeding and complementary feeding, as well as pro- tective steps taken by the family to reduce susceptibility to mosquito-borne illnesses. Table 4.2 lists the ideal and the available measures. The measure of adequate care used and initial breastfeeding for immediate skin-to-skin contact must have occurred within the first hour after birth. For children under six months of age, adequate care consists of exclusive breastfeeding. For children six to eight months of age, complementary feedings are required. All children under 24 months of age are required to be breastfed. Although the surveys have information on the educational level of the mother, the presumed caregiver, there is no consensus on how to translate that information to a measure of the mother’s caregiving abilities, so that information is not included in the measure. Because of the paucity of available data, the care components are combined into food and health. Although access to adequate food and access to adequate care measure important but different aspects of the nutritional environment to which the child is exposed, the two measures are combined into one for the purposes of this report. The choice is driven by the fact that the care-based measures available from the DHS are food based. Any recommendations for adequate food (as possible to measure here) would also promote the caring behaviors that can be measured from the surveys. The additional measure of a having a child sleep under a mosquito net is also considered a caring behavior by the caregiver to protect the child from malaria. This measure is included as a component of the health determinant, given its nexus with illness prevention. These three components do not capture the breadth of the care determinant, Table 4.2  Components of Care Determinant Data available Caregiver’s workload and time availability No Social support for caregiver No Psychosocial care No Caring behaviors  Breastfeeding Yes  Health seeking No  Complementary feeding Yes  Hygiene No  Child feeding index No Note: Another important determinant of care is maternal education, although there is no consensus on the threshold (or level of education) for adequate care behaviors. 62   ALL HANDS ON DECK so the three care-based components are distributed into food or health, depending on which sector policies could potentially affect these caregiving behaviors. Three components are therefore considered in the food and care determi- nant. The first is a measure of the food quality—minimum acceptable diet. For children under six months of age, the only acceptable diet considered is exclu- sive breastfeeding. For children 6 to 23  months of age, the acceptable diet depends on their Dietary Diversity Score (DDS) and meal frequency. The DDS is a measure of the nutritional quality of the food consumed. The World Health Organization (WHO) defines the DDS measure for children under 24 months of age on the basis of consumption of seven food groups consumed during the past 24 hours (WHO 2008). The seven food groups considered are grains, roots, and tubers; legumes and nuts;  dairy products; flesh foods including organ meats; eggs; Vitamin A–rich fruits and vegetables including orange and yellow vegetables; and other fruits. To have a minimum acceptable diet, a child needs to have consumed from at least four of the seven categories. Furthermore, breastfed children 6 to 8 months of age need to have been fed at least twice in the past 24 hours and children 9 to 23 months at least thrice. Nonbreastfed children from 6 to 23 months of age need to have been fed four times in the past 24 hours. The other two measures are measures of care practices: early initiation of breastfeeding and age-appropriate breastfeeding. • Early initiation of breastfeeding is measured as breastfeeding initiated within one hour of birth (WHO 2008). • Age-appropriate breastfeeding is defined as children under 6 months of age being exclusively breastfed, all children 6 to 24 months of age being breast- fed, and children 6 to 8 months of age receiving complementary foods (WHO 2008). For children to be considered adequate in the food/care determinant, they must meet the age-appropriate criteria for minimum acceptable diet and at least one of the two care practice measures. Information for the DDS is relatively consistent across the surveys with one notable exception: in Uganda, the survey used much more detailed food groupings and, potentially with the extra prob- ing, gathered additional valid data. Measures of Water, Sanitation, and Hygiene The environmental indicators measure the sanitary and hygienic conditions in the dwelling where the child lives. Components that are considered relate to access to drinking water, access to sanitation, and the disposal of a child’s The Underlying Determinants of Nutrition   63 Table 4.3  Components of WASH Component Data available Access to safe water Yes Access to improved sanitation Yes Community-level sanitation Yes Handwashing facilities Yes Feces disposal Yes Note: WASH = water, sanitation, and hygiene. stools (table 4.3). As per the WHO/United Nations Children’s Fund Joint Monitoring Program classification, an improved drinking water source is con- sidered to be one that “protects drinking water from outside contamination, especially from fecal matter” (WHO/UNICEF JMP 2015, 21). An improved water source is one that is piped into the dwelling, yard, or plot; comes from a public tap or standpipe; comes from a tube well or a bore well; comes from a protected well or spring; or is rainwater.1 For most countries, bottled water is considered an unimproved source. The JMP definition for improved sanitation is that it “effectively separates excreta from human contact, and ensure[s] that excreta do not re-enter the immediate household environment” (WHO/UNICEF JMP 2015, 20). Accordingly, households having access to a flush toilet, a ventilated improved pit latrine, a pit latrine with slab, or a composting toilet are considered to have improved sanitation. Unimproved sanitation facilities include pit latrine with- out slab, bucket, hanging toilet, latrine, and no facilities. The sanitation is con- sidered improved only if it is not shared with other households.2 To capture the general sanitary conditions in the child’s immediate sur- roundings, a measure of community-level sanitation is included. The measure is based on the percentage of households in the child’s locality (that is, the primary sampling unit in which the child lives) that do not have access to any type of facility but resort to open defecation.3 We use a threshold of 75 percent in these analyses; for the community not to be an open defecation community, less than 25 percent of the households in the community use open defecation. Two additional components are considered: handwashing facilities and disposal of stools. The household is classified as having appropriate hand- washing facilities if such facilities, along with availability of water and soap or detergent, are observed by the interviewer. Child’s stool disposal is con- sidered adequate if the child’s stools are disposed of into an improved sani- tation facility, either because the child actually used the toilet or improved latrine or because the child’s stools were disposed of into the toilet or improved latrine. 64   ALL HANDS ON DECK A child is considered adequate in water, sanitation, and hygiene (WASH) if the child meets at least three of the five components. When it is not possible to construct all five components, the possible set of components is reduced to four or three; nonetheless, the child still needs to meet at least three of the conditions for adequacy. For example, in the Republic of Congo where there is no informa- tion on handwashing facilities or feces disposal, a child must meet the remain- ing three conditions: access to improved water, improved sanitation, and community sanitation is required for the child to be considered adequate in WASH. Measures of Health The final driver is access to adequate health care. This driver summarizes the child’s access to and use of skilled medical care to minimize the effects of illness and preventively address health issues, especially those linked with undernutri- tion, such as diarrheal diseases. The measure encompasses the availability and use of health care services for prenatal, birth, and postnatal care (table 4.4; see box 4.2 for a discussion of issues related to measuring access to and use of health care facilities). The first measure considered is the use of prenatal services. The WHO (2007) recommends at least four prenatal visits by a pregnant woman, and the ade- quacy measure uses four visits as the threshold. The second measure we explore is birth assisted by a health care professional—including birth assisted by a doc- tor, nurse, birth attendant, or midwife. The third measure considered is postna- tal growth control visit because it is an indicator of the child’s health assessment by medical professionals after birth. The fourth component considered is vac- cination status. The measure is based on the WHO’s compilation of recom- mended vaccination schedules for each country.4 The immunizations considered are BCG (bacille Calmett-Guerin), DPT (diphtheria, pertussis, and tetanus), Table 4.4  Components of Health Component Data available Use of prenatal services Yes Birth assisted by a health care professional Yes Postnatal checkups Yes Age-appropriate immunization status Yes Mosquito nets Yes Note: ORS = oral rehydration solution. The use of ORS for diarrhea and antibiotics for pneumonia are measures often associated with adequate access to health services. However, given that information on such behaviors is available only for the subset of children with symptoms in the prior two weeks, these measures are not included. The Underlying Determinants of Nutrition   65 BOX 4.2 Access to and Utilization of Health Facilities In the Demographic and Health Surveys, it is not possible to distinguish access from use of a service. One might argue that utilization is a choice variable whereas access is not under the control of the household. A household may have access to a health facility but not use the services for reasons beyond its control, such as high costs or other con- straints. At the same time, one can easily conceive of reasons that access would be an endogenous variable. Households, for example, may choose to migrate to be able to access a service. Irrespective of whether utilization is a choice variable, the p ­ lacement of clinics is also likely to be nonrandomly allocated but rather based on unobserved com- munity characteristics. In that case, the coefficient of access to the service may be biased. For example, as first pointed out by Rosenzweig and Wolpin (1982, 1986), it is possible that clinics are placed in areas with higher incidence of illness. In this case, a simple regression of the incidence of illness on a variable identifying the presence of a clinic in the community with other controls that do not typically include community- level fixed effects is likely to have a biased and negative coefficient, leading to the incor- rect inference that access to clinics decreases health. Pitt, Rosenzweig, and Gibbons (1993) demonstrate, with a panel dataset in Indonesia, that controlling for community- level fixed effects removes the bias in the coefficient of interest, thus allowing for the correct inference about the positive effect of access to a clinic on child health. Figure B4.1.1, using data from the 2014 Health Census of Nigeria, demonstrates that the placement of health facilities in Nigeria is not random: low-quality facilities are rather dispro- portionately concentrated in the north region, which is the poorest region of the country, and high-quality facilities are concentrated in the relatively more well-off southern region. Figure B4.2.1  The Nonrandom Allocation of High- and Low-Quality Health Facilities, Nigeria a. Low-quality health facility b. High-quality health facility 14 14 12 12 10 10 Latitude Latitude 8 8 6 6 4 4 0 5 10 15 0 5 10 15 Longitude Longitude Source: 2014 Health Census, Ministry Health, Nigeria. The census contains the location of all the health facilities in existence as of 2014 (n = 34,218). Note: The quality of a health facility is determined by the following characteristics: the clinic has (a) improved water supply, (b) improved sanitation, (c) refrigerator/freezer for vaccines, (d) antenatal care service, (e) ­ family planning service, (f) malaria treatment service, (g) emergency transport service, (h) skilled birth attendant, (i) electricity, (j) caesarian-section service, and (k) measles vaccination service. A health facility is classified as low quality if it has 4 or fewer out of the total number of 11 services available. A clinic is categorized as high quality if the number of characteristics is more than 8. 66   ALL HANDS ON DECK polio, measles, and yellow fever. In general, the vaccination schedule is as follows: BCG at birth; DPT/pentavalent at two, three, and four months; oral ­ polio at two, three, and four months; measles at nine months; and yellow fever, when part of the schedule, at nine months. We allow for a three-month leeway in immunization compliance. For example, all infants zero to two months of age are considered in compliance with BCG regardless of actual vaccination status. Only at three months of age will a child be considered to be noncompliant with the BCG vaccination. Similarly, we apply the same three-month window to all vaccinations considered. The fifth health component considered is whether the child sleeps under a mosquito net.5 For a child to be considered as having access to adequate health, the child must meet at least three of the five requirements. As with adequate WASH, if the survey did not collect information on all five components, then this analysis considers three or four components. For adequacy in health, however, the child must meet three of the requirements. Empirical Measures of the Underlying Determinants of Nutrition In previous work, access to an adequate level in any one of the nutrition drivers was defined according to a strict rule requiring that the child must have access to an adequate level in each and every one of the components of the particular determinant (Skoufias and Vinha 2017). Here a slightly different, less stringent, approach is followed. A child is considered as having access to an adequate level in an underlying driver if the child meets a certain number of the components. This approach, which is the intermediate case of the union and the intersection approaches to defining multidimensional poverty discussed in detail by Alkire and Foster (2011) and Atkinson (2003), allows some substitutability among the components of any given nutrition determinant. Table 4.5 summarizes some of the components that are not consistent across the different countries. Regarding the health determinant, the table identifies those countries where yellow fever vaccination is part of the vacci- nation schedule, surveys with information on postnatal visits, and the use of mosquito nets. Regarding the WASH determinant, the table identifies those countries where bottled water is classified as improved because of additional work by the World Bank’s WASH sector in determining the common second- ary source for those using bottled water as their drinking water source or if the survey had information on the secondary source. It also identifies whether information was collected on handwashing facilities and feces disposal and those components could thus be included as components of the WASH determinant. The Underlying Determinants of Nutrition   67 Table 4.5  Inclusion of Select Health and WASH Components Yellow Bottled Fever Postnatal Mosquito water Handwashing Feces Other Country vaccine visit nets (improved) facilities disposal Notes Angola √ √ √ √a √ √ Benin √ √ √ √ √ Burkina Faso √ √ √ √ √ Burundi √ √ √ √ √ Cameroon √ √ √ √ √ Chad √ √ √ √ √ Comoros √ √ √ √ Congo, Dem. Rep. √ √ √ √ √ √ Congo, Rep. √ √ √ Côte d’Ivoire √ √ √ √ √ Ethiopia √ √a √ √ Gabon √ √ √ Gambia, The √ √ √ √ Ghana √ √ √ √ √ Guinea √ √ √ √ √ Kenya √ √ √ √ Lesotho √ √ Liberia √ √ √ √ √ Madagascar √ √ Malawi √ √ √a √ √ Mali √ √ √ √ √ Mozambique √ √ √ Namibia √ √ √ √ Niger √ √ √ √ √ Nigeria √ √ √ √ √ Rwanda √ √ √ Senegal √ √ √ √ √ √ Sierra Leone √ √ √ √ √ Tanzania √ √ √a √ Togo √ √ √ √ √ Uganda √ √ √ √ Zambia √ √ √ √ Zimbabweb √ √ √a √ √ No information on polio at birth Source: Latest Demographic and Health Surveys. See table 2.1 in chapter 2 for specific years. Note: WASH = water, sanitation, and hygiene. a. Based on whether secondary source is improved. b. No information on polio vaccination at birth available. 68   ALL HANDS ON DECK Access to Food and Care, Health, and WASH In all countries studied, the percentage of children with access to adequate food and care ranges between 7 percent and 49 percent (see figure 4.1). In Uganda, the percentage reaches 49 percent—probably in part because the Uganda DHS collects more detailed information on food. The differences in access to ade- quate food/care are in general small across rural and urban households (see figure 4.2). In Liberia, children in rural areas are more than 5 percentage points more likely to have access to adequate food/care than children in urban areas. Figure 4.1  Percentage of Children with Access to Adequate Food and Care within Each Country (Children 0–23 Months) Uganda Rwanda Burundi Ghana Togo Zambia Malawi Lesotho Congo, Dem. Rep. Kenya Tanzania Gambia, The Mozambique Ethiopia Zimbabwe Liberia Benin Angola Namibia Senegal Sierra Leone Madagascar Mali Nigeria Niger Cameroon Congo, Rep. Burkina Faso Côte d’Ivoire Guinea Comoros Gabon Chad 0 10 20 30 40 50 Share of children (%) Source: Estimates based on data for children under 24 months of age from 33 country Demographic and Health Surveys from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: For this graph, children are considered younger than 24 months. The Underlying Determinants of Nutrition   69 In Burundi, Ethiopia, Kenya, Malawi, Namibia, Niger, and Nigeria, urban chil- dren are more than 5 percentage points more likely to have access to adequate food/care than rural children. There are greater differences in access to adequate food and care by wealth category. One-third of the sample had differences of more than 5 percentage points between the two wealth quintiles; however, not all of the differences favor children from the wealthier families. In Democratic Republic of Congo, Figure 4.2  Inequities in Access to Adequate Food and Care within Each Country (Children 0–23 Months) a. Children’s access to food and b. Difference in children’s access to food care, urban versus rural and care, urban versus rural Uganda Burundi Rwanda Ethiopia Burundi Malawi Togo Niger Ghana Nigeri Zambia Namibia Lesotho Kenya Liberia Madagascar Congo, Dem. Rep. Ghana Tanzania Rwanda Malawi Burkina Faso Gambia, The Mali Mozambique Guinea Kenya Cameroon Zimbabwe Zimbabwe Ethiopia Gabon Benin Benin Angola Zambia Senegal Mozambique Namibia Gambia, The Sierra Leone Chad Congo, Rep. Côte d’Ivoire Mali Uganda Madagascar Angola Nigeria Tanzania Niger Congo, Dem. Rep. Cameroon Sierra Leone Burkina Faso Comoros Comoros Togo Côte d’Ivoire Senegal Guinea Congo, Rep. Gabon Lesotho Chad Liberia 0 10 20 30 40 50 −5 0 5 10 15 Share of children (%) Percentage difference in access, urban to rural Urban Rural (continued next page) 70   ALL HANDS ON DECK Figure 4.2 (continued) c. Children’s access to food and care, d. Difference in children’s access to by wealth quintile food and care, by wealth quintile Uganda Rwanda Togo Burundi Rwanda Kenya Ghana Niger Burundi Malawi Zambia Ethiopia Congo, Dem. Rep. Comoros Lesotho Nigeria Mozambique Madagascar Liberia Namibia Gambia, The Benin Tanzania Tanzania Zimbabwe Côte d’Ivoire Ethiopia Angola Malawi Burkina Faso Benin Sierra Leone Angola Zimbabwe Mali Chad Congo, Rep. Cameroon Kenya Zambia Senegal Lesotho Namibia Senegal Madagascar Mozambique Sierra Leone Gabon Cameroon Guinea Nigeria Mali Burkina Faso Ghana Niger Congo, Dem. Rep. Guinea Gambia, The Côte d’Ivoire Liberia Gabon Uganda Comoros Togo Chad Congo, Rep. 0 10 20 30 40 50 −10 0 10 20 Share of children (%) Percentage difference in access, Top quintile Bottom quintile top quintile to bottom quintile Source: Estimates based on data for children under 24 months of age from 33 country Demographic and Health Surveys from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Republic of Congo, The Gambia, Ghana, Guinea, Liberia, Mali, Togo, and Uganda, children from poorer households are more likely have access to ade- quate food/care than children from the richest households. In The Gambia, for example, these differences are driven by the higher prevalence of the care com- ponents, whereas in Liberia they are driven by the higher prevalence of access to minimum acceptable diet and age-appropriate breastfeeding in the lower wealth quintile households. The Underlying Determinants of Nutrition   71 There are large inequities in access to adequate WASH services across countries. Except in Malawi and Rwanda, less than one-half of children have access to adequate WASH in the study countries (see figure 4.3). Also, Malawi and Rwanda are the only countries where a significantly larger share of rural children than urban children has access to adequate WASH Figure 4.3  Percentage of Children with Access to Adequate WASH within Each Country (Children 0–23 Months) Rwanda Malawi Lesotho Burundi Senegal Gambia, The Comoros Gabon Angola Namibia Ghana Zimbabwe Nigeria Cameroon Zambia Mali Mozambique Guinea Côte d’Ivoire Kenya Congo, Dem. Rep. Niger Uganda Tanzania Benin Madagascar Togo Sierra Leone Burkina Faso Congo, Rep. Liberia Chad Ethiopia 0 10 20 30 40 50 Share of children (%) Source: Estimates based on data for children under 24 months of age from 33 country Demographic and Health Surveys from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: For this graph, children are considered younger than 24 months. WASH = water, sanitation, and hygiene. 72   ALL HANDS ON DECK (figure 4.4, panel a). In nine countries, the difference in access to WASH between urban and rural children is more than 30 percentage points (see figure 4.4, panel b). In 24 countries, children from the top wealth quintile (T20) are more than 30 percentage points more likely to have access to WASH than children from the bottom wealth quintile (B20) (see figure O.2 Figure 4.4  Inequities in Access to Adequate WASH within Each Country (Children 0–23 Months) a. Children’s access to WASH, b. Difference in children’s access urban versus rural to WASH, urban versus rural Rwanda Niger Malawi Namibia Lesotho Burkina Faso Burundi Madagascar Senegal Mali Comoros Angola Gambia, The Cameroon Nigeria Côte d’Ivoire Zimbabwe Senegal Ghana Mozambique Mali Guinea Zambia Gabon Congo, Dem. Rep. Chad Uganda Tanzania Angola Togo Kenya Zimbabwe Cameroon Zambia Guinea Benin Namibia Ethiopia Mozambique Ghana Madagascar Sierra Leone Gabon Kenya Tanzania Gambia, The Sierra Leone Nigeria Côte d’Ivoire Uganda Niger Comoros Benin Liberia Ethiopia Congo, Rep. Burkina Faso Congo, Dem. Rep. Congo, Rep. Lesotho Togo Burundi Liberia Malawi Chad Rwanda 0 20 40 60 80 100 –20 0 20 40 60 80 Share of children (%) Percentage difference in access, Urban Rural urban to rural (continued next page) The Underlying Determinants of Nutrition   73 Figure 4.4 (continued) c. Children’s access to WASH, d. Difference in children’s access by wealth quintile to WASH, by wealth quintile Malawi Namibia Rwanda Senegal Burundi Niger Comoros Mozambique Lesotho Madagascar Gambia, The Angola Ghana Zimbabwe Congo, Dem. Rep. Cameroon Nigeria Côte d’Ivoire Zambia Gabon Mali Tanzania Zimbabwe Zambia Angola Gambia, The Cameroon Mali Kenya Guinea Uganda Benin Mozambique Burkina Faso Sierra Leone Lesotho Guinea Nigeria Senegal Kenya Côte d’Ivoire Togo Ethiopia Uganda Benin Ethiopia Madagascar Sierra Leone Tanzania Liberia Niger Burundi Namibia Chad Gabon Ghana Burkina Faso Rwanda Liberia Malawi Chad Congo, Rep. Congo, Rep. Comoros Togo Congo, Dem. Rep. 0 20 40 60 80 100 0 20 40 60 80 Share of children (%) Percentage difference in access, top quintile to bottom quintile Top quintile Bottom quintile Source: Estimates based on data for children under 24 months of age from 33 country Demographic and Health Surveys from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: For this graph, children are considered younger than 24 months. B20 = bottom wealth quintile; T20 = top wealth quintile; WASH = water, sanitation, and hygiene. in the overview). In Namibia, Niger, and Senegal, the difference is more than 70 percentage points. In most of the countries, at least 50 percent of the children have access to adequate health. The notable exception is Ethiopia, where only 14 percent of the children under 24 months of age have access to adequate health (figure 4.5). 74   ALL HANDS ON DECK Figure 4.5  Percentage of Children with Access to Adequate Health within Each Country (Children 0–23 Months) Congo, Rep. Ghana Zimbabwe Burkina Faso Namibia Sierra Leone Lesotho Benin Comoros Malawi Gabon Liberia Senegal Togo Rwanda Burundi Gambia, The Tanzania Zambia Côte d’Ivoire Kenya Mali Congo, Dem. Rep. Uganda Cameroon Madagascar Guinea Mozambique Angola Niger Nigeria Chad Ethiopia 0 20 40 60 80 Share of children (%) Source: Estimates based on data for children under 24 months of age from 33 country Demographic and Health Surveys from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: For this graph, children are considered younger than 24 months. However, despite the relatively high access to adequate health, there are large disparities in access to adequate health services by household location as well as by wealth (figure 4.6). Urban children are always more likely to have access to adequate health than rural children, and children from richer households are more likely to have access than children from poorer households. In 11 of the The Underlying Determinants of Nutrition   75 countries, the difference in access to adequate health between urban and rural children is more than 30 percentage points (figure 4.6, panel b). In 14 countries, children from the richest households are more than 40 percentage points more likely to have access to adequate health than children from the poorest house- holds (figure 4.6, panel d). Figure 4.6  Inequities in Access to Adequate Health within Each Country (Children 0–23 Months) a. Children’s access to health services, b. Difference in children’s access urban versus rural to health services, urban versus rural Zimbabwe Niger Ghana Ethiopia Sierra Leone Guinea Burkina Faso Angola Namibia Nigeria Lesotho Mali Congo, Rep. Chad Comoros Tanzania Malawi Madagascar Benin Cameroon Rwanda Uganda Burundi Mozambique Liberia Kenya Senegal Côte d’Ivoire Togo Togo Gambia, The Senegal Gabon Zambia Zambia Congo, Dem. Rep. Tanzania Congo, Rep. Kenya Gabon Côte d’Ivoire Gambia, The Mali Benin Congo, Dem. Rep. Burundi Uganda Liberia Madagascar Rwanda Mozambique Ghana Guinea Zimbabwe Cameroon Lesotho Niger Burkina Faso Nigeria Malawi Angola Comoros Chad Namibia Ethiopia Sierra Leone 0 20 40 60 80 100 0 10 20 30 40 50 Share of children (%) Percentage difference in access, urban to rural Urban Rural (continued next page) 76   ALL HANDS ON DECK Figure 4.6 (continued) c. Children’s access to health d. Difference in children’s access to services, by wealth quintile health services, by wealth quintile Sierra Leone Angola Zimbabwe Cameroon Ghana Nigeria Namibia Madagascar Malawi Ethiopia Burkina Faso Kenya Congo, Rep. Niger Gambia, The Mali Benin Tanzania Lesotho Senegal Burundi Guinea Rwanda Côte d’Ivoire Congo, Dem. Rep. Mozambique Liberia Togo Comoros Lesotho Zambia Uganda Togo Benin Uganda Gabon Gabon Zambia Senegal Comoros Tanzania Chad Côte d’Ivoire Liberia Mali Congo, Rep. Madagascar Rwanda Kenya Burundi Guinea Burkina Faso Mozambique Ghana Niger Malawi Angola Congo, Dem. Rep. Chad Zimbabwe Cameroon Namibia Nigeria Gambia, The Ethiopia Sierra Leone 0 20 40 60 80 100 0 20 40 60 Share of children (%) Percentage difference in access, top quintile to bottom quintile Top quintile Bottom quintile Source: Estimates based on data for children under 24 months of age from 33 country Demographic and Health Surveys from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: For this graph, children are considered younger than 24 months. Simultaneous Access to the Underlying Determinants of Nutrition One major advantage offered by using information at the child level from the DHS is the opportunity to get a better sense of the joint distribution of determinants of undernutrition. Information on the joint distribution of the ­ the  determinants is essential for identifying particularly important gaps The Underlying Determinants of Nutrition   77 (for example, in water or sanitation) which may affect the impact of other nutri- tion-related interventions. Figure 4.7 gives a sense of the inequities in joint/simultaneous access to the drivers of nutrition. Many children do not have access to any of the three deter- minants of nutrition. In three countries—Chad, Ethiopia, and Niger—most Figure 4.7  Inequities in Joint Access to the Underlying Drivers of Nutrition within Each Country (Children 0–23 Months) Chad Ethiopia Niger Guinea Madagascar Nigeria Mozambique Angola Côte d’Ivoire Cameroon Mali Congo, Dem. Rep. Tanzania Kenya Zambia Togo Liberia Gabon Benin Senegal Gambia, The Comoros Uganda Burkina Faso Burundi Sierra Leone Namibia Malawi Zimbabwe Lesotho Congo, Rep. Rwanda Ghana 0 10 20 30 40 50 60 70 80 Share of children (%) Access to zero determinants of nutrition Access to all 3 determinants of nutrition Source: Estimates based on data for children under 24 months of age from 33 country Demographic and Health Surveys from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: For this graph, children are considered younger than 24 months. 78   ALL HANDS ON DECK children do not have access to even one of the determinants. Simultaneous access to all three determinants is under 20 percent across the region. Figure 4.8 reveals that simultaneous access to the drivers of nutrition for rural and urban children differs significantly. Rural children are more likely to have access to none (0) of the drivers of nutrition than urban children and, in general, less likely to have access to all three drivers. Rwanda and Lesotho are Figure 4.8  Inequities in Joint Access to the Underlying Drivers of Nutrition within Each Country, Urban versus Rural Areas (Children 0–23 Months) a. Rural children’s access to the b. Urban children’s access to the determinants of nutrition determinants of nutrition Chad Chad Ethiopia Ethiopia Niger Angola Angola Nigeria Guinea Congo, Dem. Rep. Nigeria Cameroon Cameroon Mozambique Madagascar Gabon Côte d’Ivoire Liberia Mozambique Guinea Mali Côte d’Ivoire Gabon Comoros Tanzania Madagascar Kenya Zambia Congo, Dem. Rep. Gambia, The Zambia Benin Liberia Kenya Togo Sierra Leone Senegal Malawi Benin Burundi Gambia, The Togo Congo, Rep. Mali Uganda Tanzania Namibia Lesotho Burkina Faso Niger Comoros Rwanda Sierra Leone Namibia Burundi Burkina Faso Ghana Senegal Zimbabwe Uganda Lesotho Congo, Rep. Malawi Zimbabwe Rwanda Ghana 0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 Share of children (%) Share of children (%) Access to zero determinants of nutrition Access to all 3 determinants of nutrition (continued next page) The Underlying Determinants of Nutrition   79 Figure 4.8 (continued) c. Children with no access d. Children with access to to the determinants of all three determinants of nutrition, urban versus rural nutrition, urban versus rural Chad Rwanda Ethiopia Malawi Niger Burundi Angola Lesotho Guinea Senegal Nigeria Zimbabwe Cameroon Ghana Madagascar Gambia, The Côte d’Ivoire Kenya Mozambique Zambia Mali Comoros Gabon Uganda Tanzania Namibia Kenya Congo, Dem. Rep. Congo, Dem. Rep. Mali Zambia Mozambique Liberia Tanzania Togo Cameroon Senegal Benin Benin Madagascar Gambia, The Sierra Leone Congo, Rep. Angola Uganda Nigeria Namibia Niger Burkina Faso Liberia Comoros Gabon Sierra Leone Burkina Faso Burundi Congo, Rep. Ghana Guinea Zimbabwe Ethiopia Lesotho Côte d’Ivoire Malawi Togo Rwanda Chad 0 10 20 30 40 50 60 70 80 0 10 20 Share of children (%) Share of children (%) Urban Rural Source: Estimates based on data for children under 24 months of age from 33 country Demographic and Health Surveys from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. exceptions because more rural children have access to all three than urban chil- dren (figure 4.8, panel d). The differences are even larger by wealth (see figure 4.9). In all countries stud- ied, children from the B20 are more likely to have access to zero (or access to inadequate levels of all three drivers) and less likely to have access to adequate levels of all three drivers than children from the T20. In 12 countries, more than 50 percent of the children in the poorest households do not have access to any 80   ALL HANDS ON DECK Figure 4.9  Inequities in Simultaneous Access to the Underlying Drivers of Nutrition within Each Country, B20 versus T20 (Children 0–23 Months) a. Children with access to zero b. Children with access to all three determinants of nutrition, determinants of nutrition, by wealth quintile by wealth quintile Chad Rwanda Ethiopia Burundi Nigeria Zimbabwe Niger Zambia Cameroon Niger Angola Malawi Guinea Namibia Madagascar Senegal Côte d’Ivoire Angola Kenya Lesotho Gabon Uganda Mozambique Kenya Senegal Ethiopia Mali Mozambique Tanzania Tanzania Liberia Ghana Benin Mali Zambia Benin Togo Gambia, The Burkina Faso Madagascar Comoros Burkina Faso Congo, Dem. Rep. Nigeria Congo, Rep. Cameroon Lesotho Togo Burundi Comoros Gambia, The Côte d’Ivoire Namibia Congo, Dem. Rep. Uganda Gabon Rwanda Guinea Sierra Leone Congo, Rep. Zimbabwe Liberia Ghana Sierra Leone Malawi Chad 0 10 20 30 40 50 60 70 80 0 5 10 15 20 Share of children (%) Share of children (%) Top quintile Bottom quintile Source: Estimates based on data for children under 24 months of age from 33 country Demographic and Health Surveys from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: For this graph, children are considered younger than 24 months. B20 = bottom wealth quintile; T20 = top wealth quintile. determinant, but in none of the countries studied is this the case for the wealthiest households. Similarly, in none of the countries studied do at least 10 percent of children from the poorest households have access to all three determinants, whereas in 17 countries more than 10 percent of the children from the wealthiest households have access to all three. Even so, in no country do more than 23 percent of the children from the wealthiest households have access to all three. The Underlying Determinants of Nutrition   81 Notes 1. The stricter definitions of basic water—an improved source within a 30-minute round trip from the dwelling—or safely managed water—piped to premises with a minimum quality—are not considered here. 2. We do not have information on shared facilities for Ethiopia so cannot adjust the access to improved sanitation for Ethiopia. Thus, access to improved sanitation is less stringently defined for Ethiopia than for the other countries. 3. The measure is based on all surveyed households in the primary sampling unit, not only those with children under 24 months of age. 4. Information on vaccination schedules comes from http://apps.who.int/immunization​ _monitoring/globalsummary/schedules. 5. Although sleeping under a mosquito net may be better considered as a caregiving behavior, given the scarcity of components to construct an index of care, it is intro- duced as a component of health because of its nexus to preventing illness along the lines of use of oral rehydration solution and antibiotics to treat illness. References Alkire, Sabina, and James Foster. 2011. “Counting and Multidimensional Poverty Measurement.” Journal of Public Economics 95 (7): 476–87. Atkinson, A. B. 2003. “Multidimensional Deprivation. Contrasting Social Welfare and Counting Approaches.” Journal of Economic Inequality 1 (1): 51–65. Coates, J., A. Swindale, and P. Bilinsky. 2007. Household Food Insecurity Access Scale (HFIAS) for Measurement of Household Food Access: Indicator Guide (v.3). Washington, DC: FANTA, Academy for Educational Development. Hoddinott, J., and Y. Yohannes. 2002. “Dietary Diversity as a Household Food Security Indicator.” Food and Nutrition Technical Assistance Project, Academy for Educational Development, Washington, DC. Pitt, M. M., M. R. Rosenzweig, and D. Gibbons. 1993. “The Determinants and Consequences of the Placement of Government Programs in Indonesia.” World Bank Economic Review 7 (3): 319–48. Rosenzweig, M. R., and K. I. Wolpin. 1982. “Governmental Interventions and Household Behavior in a Developing Country: Anticipating the Unanticipated Consequences of Social Programs.” Journal of Development Economics 10 (2): 209–25. ———. 1986. “Evaluating the Effects of Optimally Distributed Public Programs: Child Health and Family Planning Interventions.” American Economic Review 76 (3): 470–82. Ruel, M. 2002. “Is Dietary Diversity an Indicator of Food Security or Diet Quality? A Review of Measurement Issues and Research Needs.” FCND Discussion Paper 140, International Food Policy Research Institute, Washington, DC. Skoufias, E., and K. Vinha. 2017. WASH as Part of a Multi-Sectoral Effort to Reducing Undernutrition: A Synthesis of the Findings in WASH Poverty Diagnostic . Washington, DC: World Bank. 82   ALL HANDS ON DECK Tiwari, S., E. Skoufias, and M. Sherpa. 2013. “Shorter, Cheaper, Quicker, Better: Linking Measures of Food Security to Nutritional Outcomes in Bangladesh, Nepal, Pakistan, Uganda, and Tanzania.” Policy Research Working Paper 6584, World Bank, Washington, DC. http://documents.worldbank.org/curated/en/986391468207262071​ /pdf/WPS6584.pdf. WHO (World Health Organization). 2007. “Standards for Maternal and Neonatal Care.” Department of Making Pregnancy Safer, World Health Organization, Geneva. http:// apps.who.int/iris/bitstream/10665/69735/1/a91272.pdf. ———. 2008. “Indicators for Assessing Infant and Young Child Feeding Practices—Part I: Definitions.” Conclusions of a consensus meeting, Washington, DC, November 6–8, 2007. World Health Organization, Geneva. http://www.who.int/maternal_child​ _adolescent/documents/9789241596664/en/. WHO/UNICEF JMP (World Health Organization/United Nations Children’s Fund Joint Monitoring Program for Water Supply and Sanitation). 2017. Progress on Drinking Water, Sanitation and Hygiene: 2017 Update and SDG Baselines. Geneva: WHO and UNICEF. http://www.unwater.org/publications/whounicef-joint-monitoring​ -program-water-supply-sanitation-hygiene-jmp-2017-update-sdg-baselines/. Chapter 5 Stunting and Access to the Underlying Determinants of Nutrition Following the description of the major patterns of access to the three under- lying drivers of nutrition within countries in Sub-Saharan Africa, the analy- sis in this chapter proceeds with a closer investigation of the relationship between stunting prevalence (or the probability of stunting at the child level) and access to adequate levels of the underlying drivers on nutrition. The relationship between stunting prevalence and access is examined in a variety of ways. The correlation of stunting with the number of the drivers accessed at an adequate level at the same time is investigated first. This investigation sheds more light on the general question of whether access to more nutrition determinants is better for nutrition outcomes; it also helps address policy concerns about the marginal impact on stunting of providing access to one more nutrition driver in different target groups. Next, the analysis moves to a finer level of detail by focusing on the relationship between stunting and access to the three specific nutrition drivers (food and care; health; and water, sanitation, and hygiene [WASH]) and all the differ- ent combinations of the three drivers, such as access to WASH and health at the same time. The advantage of the finer level of analysis is that it allows for identification of the marginal effects on stunting associated with access to specific sectors that can help address which underlying determinants of nutrition to focus on. The shortcoming associated with a finer level of anal- ysis relates to the constraints and limitations imposed by the large inequali- ties in access to some of the drivers of nutrition, such as access to adequate WASH. These limitations become even more apparent when simultaneous access to two or more drivers is correlated with stunting. As documented in this chapter, only a very small proportion of children have simultaneous access to all three nutrition drivers or even simultaneous access to two nutrition drivers, such as access to food/care and WASH (see table 5.1). 83 84   ALL HANDS ON DECK Table 5.1  Percentage of Children with Adequate Access in Each Group/Category Full Mother Mother not sample Rural Urban Boys Girls educated educated Stunted 28 31 22 31 26 21 32 Adequate in: None 39 47 16 38 39 19 48 Any 1 40 38 44 41 39 45 37 Food/care 9 11 4 9 9 6 10 WASH 6 6 8 7 6 4 7 Health 25 21 33 25 24 34 20 Any 2 19 13 33 18 19 30 13 Food/care and WASH 1 1 2 1 2 1 2 Food/care and health 8 7 11 8 9 13 6 WASH and health 9 5 20 9 9 16 5 All 3 3 2 7 3 3 6 2 Observations 74,781 52,672 22,109 37,377 37,404 23,381 51,373 Low High Low High Low High Low High B20 T20 GDP GDP Gini Gini tariff tariff starch starch Stunted 34 17 30 27 27 29 29 25 27 31 Adequate in: None 57 11 43 35 41 37 41 32 35 49 Any 1 34 41 37 41 39 41 38 44 42 35 Food/care 12 3 12 7 11 7 10 6 7 13 WASH 4 8 4 8 4 9 7 5 8 3 Health 18 30 22 27 24 25 22 33 27 19 Any 2 9 39 17 20 18 19 17 21 20 14 Food/care and WASH 1 2 1 2 1 2 2 1 2 1 Food/care and health 6 10 8 8 9 8 8 9 8 7 WASH and health 2 27 7 10 8 9 8 11 10 6 All 3 1 10 3 3 3 3 3 3 4 2 Observations 16,916 11,752 32,033 42,748 34,201 40,580 44,239 30,542 37,096 37,685 Source: Estimates based on data for children under 24 months of age from 33 country Demographic and Health Surveys from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: B20 = bottom wealth quintile; GDP = gross domestic product; T20 = top wealth quintile; WASH = water, sanitation, and hygiene. Stunting and Children’s Simultaneous Access to the Drivers of Nutrition The analysis proceeds gradually, beginning with a general investigation of the relationship between stunting or height-for-age Z-scores (HAZs) and the num- ber of nutrition drivers accessed at the same time. The patterns of correlation Stunting and Access to the Underlying Determinants of Nutrition   85 between stunting and access are best summarized by the probability density function (PDF) of HAZ and the cumulative density function (CDF) of HAZ for groups of children with access to different numbers of nutrition drivers, using pooled Demographic and Health Surveys (DHS) across 33 countries for chil- dren 0–23 months of age. The CDF is particularly useful because it allows one to make inferences easily about the general relationship between having access to one or more drivers of nutrition not only for the prevalence of stunting (HAZ ≤ 2) but also for the prevalence of severe stunting (HAZ ≤ 3). It is important to bear in mind that the patterns emerging from a comparison of the CDF of HAZ of children with access to one, two, or all three drivers of nutrition are simple correlations and do not imply causation. In the remainder of this chapter, more concerted efforts are made to control for country effects (or even country-strata effects), as well as for child, parental, and household characteristics in an effort to get closer to a causal inference. In addition, unconditional quantile regres- sions are estimated for different quantiles of the bottom half of the distribution of HAZs to explore differences in the relationship based on the child’s nutri- tional status. Figure 5.1 presents the CDF of HAZs of children 0–23 months of age to sum- marize the information contained in the CDF. The horizontal axis contains the range of values of HAZs in the sample of children whereas the vertical axis is the cumulative fraction of children with HAZ less than a given value of HAZ. The gold vertical line at −2 denotes the value of the threshold used to define stunting. A child is considered stunted if the child’s HAZ is less than 2 standard deviations (s. d.) from the reference population. The gold horizontal line denotes the point of intersection of the CDF with the −2 threshold for stunting, and the blue horizontal line denotes the point of intersection of the CDF with the −3 threshold used for severe stunting. Thus, 28.2 percent of the children under 24 months of age are stunted or have a HAZ that is less than −2, and 12.4 percent of the children under 24 months of age are severely stunted or have a HAZ that is less than −3. Three binary variables are constructed by considering whether the child has access to adequate levels in the other drivers. Specifically, the binary variable Aany1 is equal to 1 if the household has an adequate level in one and only one of any of the three drivers of nutrition care. For example, Aany1 = 1 if the child has adequate food and care only (and inadequate health and environment), or when the household is adequate in WASH ser- vices only (and inadequate in food and care and health), or when the house- hold is adequate in health services only (and inadequate in food and care and WASH). In the same fashion, Aany2 takes the value of 1 if the household has adequate access to any two of the drivers at the same time, whereas Aall3 identifies the few children that have simultaneous access to adequate level of all three nutrition drivers. 86   ALL HANDS ON DECK Figure 5.1  The Cumulative Distribution of Height-for-Age Z-Scores (Children 0–23 Months) 0.8 Cumulative share of children 0.6 0.4 0.28 0.2 0.12 0 –6 −5 −4 −3 −2 −1 0 Height-for-age Z-score Source: Estimates based on data for children under 24 months of age from 33 country Demographic and Health Surveys (DHS) from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: Sampling weights were standardized across the 33 DHS, following the procedure outlined by DHS. The sampling weight of the mother was multiplied by the ratio of the total number of women 15 to 49 years of age in the surveyed country to the number of women (15 to 49 years of age) surveyed. The total number of women was obtained from UN DESA 2015. For this graph, children are considered less than 24 months old. HAZ = height-for-age Z-score. Figure 5.2 displays the PDF (panel a) and CDF (panel b) of HAZ for children who do not have access to an adequate level of any of the three drivers of nutri- tion (access to 0) together with the distribution functions of the children who have access to only one (access to 1) of the three nutrition drivers. It is easily seen from panel a that children with access to one driver have a higher mean HAZ (the PDF shifts right) and that most of the differences in the PDF between children with access to none and access to one are in the lower left tail of the bell-shaped curve. In the figure with the PDF, stunting is represented by the area under the curve and to the left of the −2 threshold. In this case, it is easy to tell that stunting is lower among children with access to one driver compared to the stunting among children with access to none (area under the dark blue PDF and to the left of the −2 threshold is smaller than the area under the light blue PDF and to the left of the −2 threshold.) The CDF curve in panel b summarizes the same information but in a different way. The lower stunting rate associated with access to one nutrition driver can be inferred by the fact the dark blue CDF intersects the −2 threshold at a lower point than the light blue CDF. The preva- lence of stunting among children who do not have access to an adequate level Stunting and Access to the Underlying Determinants of Nutrition   87 Figure 5.2  Access to No Drivers versus Access to One Driver of Nutrition and the Prevalence of Stunting (Children 0–23 Months) a. Probability of stunting b. Prevalence of stunting 0.25 0.8 Cumulative share of children 0.20 0.6 0.15 Density 0.4 0.34 0.10 0.2 0.27 0.05 0 0 −6 −4 −2 0 2 4 6 −6 −5 −4 −3 −2 −1 0 Height-for-age Z-score Height-for-age Z-score Access to zero determinants of nutrition Access to 1 determinant of nutrition Source: Estimates based on data for children under 24 months of age from 33 country Demographic and Health Surveys (DHS) from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: Sampling weights were standardized across the 33 DHS, following the procedure outlined by DHS. The sampling weight of the mother was multiplied by the ratio of the total number of women 15 to 49 years of age in the surveyed country to the number of women (15 to 49 years of age) surveyed. The total number of women was obtained from UN DESA 2015. For this graph, children are considered less than 24 months old. in any of the three drivers of nutrition (access to 0) is 34 percent (figure 5.2, panel b). In contrast, the prevalence of stunting among children with access to an adequate level of only one (any one) of the three of the drivers of nutrition (access to 1) is 26 percent, 8 percentage points lower than the prevalence among children with access to none. Figures 5.3 through 5.5 allow similar comparisons between children with access to one driver and access to two drivers (figure 5.3), children with access to two drivers and access to all three drivers (figure 5.4), and children with access to no drivers and access to all three drivers (figure 5.5). Figure 5.6 dis- plays the PDF and CDF of the different groups of children in the same graph. Figure 5.5 reveals that the prevalence of stunting among children with simul- taneous access to adequate levels of all three nutrition drivers is significantly lower than among children who do not have access to an adequate level in any of the nutrition drivers. The prevalence of stunting among children who do not have access to an adequate level in any of the three drivers of nutrition (access to 0) is 34 percent (figure 5.5). In contrast, the prevalence of stunting among children with access to adequate levels of all three of the drivers of nutrition (access to 3) is 18 percent, 16 percentage points lower than the prevalence 88   ALL HANDS ON DECK Figure 5.3  Access to One versus Access to Two Drivers of Nutrition and the Prevalence of Stunting (Children 0–23 Months) a. Probability of stunting b. Prevalence of stunting 0.8 0.25 Cumulative share of children 0.20 0.6 0.15 Density 0.4 0.10 0.27 0.2 0.2 0.05 0 0 −6 −4 −2 0 2 4 6 −6 −5 −4 −3 −2 −1 0 Height-for-age Z-score Height-for-age Z-score Access to 1 driver of nutrition Access to 2 drivers of nutrition Source: Estimates based on data for children under 24 months of age from 33 country Demographic and Health Surveys (DHS) from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: Sampling weights were standardized across the 33 DHS, following the procedure outlined by DHS. The sampling weight of the mother was multiplied by the ratio of the total number of women 15 to 49 years of age in the surveyed country to the number of women (15 to 49 years of age) surveyed. The total number of women was obtained from UN DESA 2015. Figure 5.4  Access to Two versus Access to Three Drivers of Nutrition and the Prevalence of Stunting (Children 0–23 Months) a. Probability of stunting b. Prevalence of stunting 0.25 0.8 Cumulative share of children 0.20 0.6 0.15 Density 0.4 0.10 0.2 0.2 0.05 0.18 0 0 −6 −4 −2 0 2 4 6 −6 −5 −4 −3 −2 −1 0 Height-for-age Z-score Height-for-age Z-score Access to 2 drivers of nutrition Access to 3 drivers of nutrition Source: Estimates based on data for children under 24 months of age from 33 country Demographic and Health Surveys (DHS) from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: Sampling weights were standardized across the 33 DHS, following the procedure outlined by DHS. The sampling weight of the mother was multiplied by the ratio of the total number of women 15 to 49 years of age in the surveyed country to the number of women (15 to 49 years of age) surveyed. The total number of women was obtained from UN DESA 2015. For this graph, children are considered less than 24 months old. Stunting and Access to the Underlying Determinants of Nutrition   89 Figure 5.5  Access to No Drivers versus Access to Three Drivers of Nutrition and the Prevalence of Stunting (Children 0–23 Months) a. Probability of stunting b. Prevalence of stunting 0.25 0.8 Cumulative share of children 0.20 0.6 0.15 Density 0.4 0.34 0.10 0.18 0.2 0.05 0 0 −6 −4 −2 0 2 4 6 −6 −5 −4 −3 −2 −1 0 Height-for-age Z-score Height-for-age Z-score Access to zero determinants of nutrition Access to 3 determinants of nutrition Source: Estimates based on data for children under 24 months of age from 33 country Demographic and Health Surveys (DHS) from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: Sampling weights were standardized across the 33 DHS, following the procedure outlined by DHS. The sampling weight of the mother was multiplied by the ratio of the total number of women 15 to 49 years of age in the surveyed country to the number of women (15 to 49 years of age) surveyed. The total number of women was obtained from UN DESA 2015. For this graph, children are considered less than 24 months old. among children with access to none. In addition, as the cumulative distribution function of the HAZs for children under 24 months of age highlights, significant reductions in severe stunting rates—children more than 3 s. d. below the median height-for-age—are also prevalent. Finally, figure 5.6 reveals that the greatest reductions in stunting are associ- ated with increases in access from none (0) to access to any one nutrition driver and from access to any one driver to simultaneous access to any two drivers. For example, the stunting rate among children with access to an adequate level in any one nutrition driver is 26.7 percent compared to the stunting rate of 34.1 percent prevailing among children with access to none (see figure 5.2). Lower stunting rates are also prevalent among children with simultaneous access to any two of the drivers (20.4 percent) compared to the stunting rates prevalent among children with access to any one of the drivers (26.7 percent) figure 5.3). However, the marginal decline (gain) in the stunting rate is (see ­ smaller than that associated with a change in access from none (0) to any one of the drivers. Finally, even lower stunting rates are prevalent among children with simultaneous access to all three of the drivers (18.3 percent) compared to the stunting rates prevalent among children with access to any two of the drivers (20.4 percent) (see figure 5.4). In the context of budgetary constraints, these 90   ALL HANDS ON DECK Figure 5.6  Access to No Drivers versus Access to One or More Drivers of Nutrition and the Prevalence of Stunting (Children 0–23 Months) a. Probability of stunting b. Prevalence of stunting 0.25 0.8 Cumulative share of children 0.20 0.6 0.15 Density 0.4 0.34 0.10 0.2 0.18 0.05 0 0 −6 −4 −2 0 2 4 6 −6 −5 −4 −3 −2 −1 0 Height-for-age Z-score Height-for-age Z-score Access to zero determinants of nutrition Access to 1 determinant of nutrition Access to 2 determinants of nutrition Access to all 3 determinants of nutrition Source: Estimates based on data for children under 24 months of age from 33 country Demographic and Health Surveys (DHS) from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: Sampling weights were standardized across the 33 DHS, following the procedure outlined by DHS. The sampling weight of the mother was multiplied by the ratio of the total number of women 15 to 49 years of age in the surveyed country to the number of women (15 to 49 years of age) surveyed. The total number of women was obtained from UN DESA 2015. For this graph, children are considered less than 24 months old. patterns have important implications for the targeting and potential impact of nutrition-sensitive operations in agriculture, WASH, and social protection, which are elaborated on in the next section. The preceding figures are useful in summarizing the broad relationships in Sub-Saharan African countries between child stunting and access to the drivers of nutrition. The patterns are summarized by these figures but ignore the poten- tial influence that individual child and parental characteristics or the country/ region of residence may have on the relationship between stunting and access to one or more nutrition driver. To increase the credibility in these observed patterns, it is necessary to examine the robustness of the observed relationship in more detail. For this purpose, the following model is estimated using the logit model for binary dependent variables: Stuntedic = α + α1Aany1ic + α2Aany2ic + α3Aall3ic + Xic +μc + εic (5.1) In equation (5.1), the dependent variable Stunted is a binary variable taking the value of 1 if child i in country c is stunted (that is, the child’s HAZ is less than Stunting and Access to the Underlying Determinants of Nutrition   91 −2 s. d. for the reference population, HAZi ≤ –2) and = 0 otherwise. The variable μc denotes country fixed effect (or country-specific binary variables), whereas Xic denotes a set of control variables summarizing parental, child, and house- hold characteristics. These control variables include indicator variables for age, gender, multiple births, and birth order; the age of the mother (in years); the marital status of the mother; the mother’s education level (in years); the moth- er’s height (in cm); the number of household members; the number of children under five years of age; and whether the household lives in an urban or rural area. In this specification, the constant term α provides an estimate of the stunting rate (or the probability of being stunted) in the reference group of children, whereas the coefficients α1, α2, and α3 yield estimates of the “marginal decline” in stunting associated with a child having access to adequate levels in one, two, or all three of the nutrition drivers.1 Specifically, the coefficient α1 yields an estimate of the decline in the stunting rate of children with access to any one of the three drivers of nutrition relative to the stunting rate in the reference group summarized by the constant term α that has inadequate access to all three nutri- tion drivers. A similar interpretation holds for the coefficients α2 and α3. It is also important to bear in mind that the regression equation employed above does not allow for causal inferences on the impacts of having access to adequate levels in the various clusters’ adequacy components on nutrition nor does it provide a formal test of the United Nations Children’s Fund (UNICEF) conceptual framework. Most of the components of each of the three drivers of nutrition are household choice variables (such as childcare variables, immuni- zations, and visits for prenatal care), meaning that they are likely to be corre- lated with the error term of the regression εic, and it is practically infeasible to devise a credible identification strategy or find a set of instrumental variables that would allow reliable causal inferences. With these considerations in mind, the inclusion of the country fixed effects (denoted by μc) and additional control variables pertaining to parental, child, and household characteristics (denoted by Xic) may be considered as an effort at minimizing the influence of other contextual variables (omitted variable bias) on the relationship between stunting and access to the drivers of nutrition sum- marized by the coefficients α1, α2, and α3. Before presenting the estimates based on regression equation (5.1), it is use- ful to look at table 5.1, which presents the percentage of children with access to different combinations of the drivers of nutrition. The feature that stands out in this table is the very low percentage of children with access to adequate levels of all three nutrition drivers. The very low number of observations in this specific cell/group implies that standard errors of the coefficient estimate of simultane- ous access to all three drivers are likely to be very high, making the coefficient α3 statistically insignificant from 0. 92   ALL HANDS ON DECK The estimates in table 5.2 reveal that the preliminary inference drawn from the simple cumulative distribution functions of stunting among children with differential access to the drivers of nutrition is very robust.2 At the national level, stunting rates of children are lower for groups with access to one or more nutri- tion drivers than the stunting rates among children with inadequate access to all three determinants. This result holds irrespective of the specification used (for example, including country-level fixed effects or fixed effects at a finer level such as the stratum that typically identifies regions and urban/rural areas within regions of the country). The estimates in table 5.3 present the marginal effects on the probability of a child being stunted associated with simultaneous access to one or more of the drivers of nutrition. These child-level estimates reveal the same story as that obtained with the differences between the stunting rates of groups of children with access to one or more nutrition drivers. The probability that an individual child is stunted, controlling for child, parental, and household characteristics, is lower when a child has access to one or more nutrition drivers than the prob- ability of stunting for a child with inadequate access to all three determinants. As in table 5.2, this result holds irrespective of the specification used, for exam- ple, including country-level fixed effects or country-strata-level fixed effects. Also, the coefficients in table 5.3 do not differ substantially between country fixed effects and country-strata fixed effects. Controlling for child, parental, and household characteristics leads to a sub- stantial decline in the coefficient estimates (compare estimates in table 5.2 Table 5.2  Changes in the Stunting Rate for Groups of Children with Simultaneous Access to One or More of the Drivers of Nutrition, National-Level Estimates Country fixed effects Country-strata fixed effects Variables No covariates No covariates −0.078*** −0.057*** Aany1: Adequate in any 1 driver (0.008) (0.008) −0.143*** −0.110*** Aany2: Adequate in any 2 drivers (0.008) (0.008) −0.167*** −0.134*** Aall3: Adequate in all 3 drivers (0.013) (0.013) Observations 74,781 74,634 Source: Estimates based on data for children under 24 months of age from 33 country Demographic and Health Surveys (DHS) from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Notes: Marginal effects are based on the coefficient estimates obtained from the logit model in equation (5.1) excluding child, parental, and household characteristics. Robust standard errors in parentheses corrected for correlation at the cluster level. Stratum is the lowest level of statistical representation of the DHS within a country, typically identifying regions and urban/rural areas within regions in a country. ***p < 0.01, **p < 0.05, *p < 0.1. Stunting and Access to the Underlying Determinants of Nutrition   93 Table 5.3  Marginal Effects on the Probability of Stunting of Simultaneous Access to One or More of the Drivers of Nutrition, Child-Level Estimates Country fixed effects Country-strata fixed effects Including child, parental, Including child, parental, and Variables and household covariates household covariates −0.022*** −0.020*** Aany1: Adequate in any 1 driver (0.007) (0.007) −0.042*** −0.043*** Aany2: Adequate in any 2 drivers (0.009) (0.009) −0.038** −0.051*** Aall3: Adequate in all 3 drivers (0.016) (0.015) Observations 68,533 68,391 Source: Estimates based on data for children under 24 months of age from 31 country Demographic and Health Surveys (DHS) from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. No information on mother’s height is available for Angola or Senegal, and therefore they are not part of the analyses with covariates. Note: Marginal effects are based on the coefficient estimates obtained from the logit model in equation (5.1). Stratum is the lowest level of statistical representation of the DHS within a country, typically identifying regions and urban/rural areas within regions in a country. Detailed estimates available upon request. All regressions include year fixed effects as well as child, parental, and household characteristics that consist of the following variables: indicator variables for age (in months), gender, birth order, and multiple birth; mother’s education in years; indicator for mother’s marital status; mother’s height; mother’s age; wealth quintile; number of children under five years of age; number of household members; and whether household is in a rural or urban area. ***p < 0.01, **p < 0.05, *p < 0.1. against estimates in table 5.3). Irrespective of the specification used, these esti- mates are also useful at identifying target groups of children who are deprived in all or some of the drivers of nutrition. Assuming that the policy objective is to decrease the prevalence of stunting, these estimates can be used to shed some light on the extent to which the allocation of scarce resources to different groups of children with differential access to the drivers of nutrition can have potential benefits in terms of reducing stunting. Consider, for example, the choice between allocating the same resources to two groups of children: group A composed of children who have inadequate access to all three nutrition drivers, and group B composed of children who have adequate access to only one of the nutrition drivers. The estimated mar- ginal effect of −0.022 of Aany1 (first column in table 5.3) implies that the prob- ability of stunting for a child with access to any one driver is 2.2 percentage points lower than the probability of stunting in the reference group of children with inadequate access to all three nutrition drivers (group A). The coefficient −0.042 of Aany2 (first column in table 5.3) implies that the probability of stunt- ing among children with access to any two of the drivers of nutrition is 4.2 ­ percentage points lower than the probability of stunting among children with inadequate access to all three nutrition drivers. This, however, implies that 94   ALL HANDS ON DECK there is only a 2.0 percentage point decline in the probability of stunting associ- ated with having access to two drivers compared to one driver (−0.042 − (−0.022) = −0.020). Thus, allocating the same resources to group B instead of group A for the purpose of increasing their access from one to two nutrition drivers will decrease the stunting rate by only 2.0 percentage points. The 0.2-percentage-point-higher marginal effect on stunting between pro- viding access to one nutrition driver to children in group A composed of chil- dren that have inadequate access to all three nutrition drivers compared to allocating the resources to group B composed of children that have adequate access to only one of the nutrition drivers is not sufficiently robust to serve as a basis for policy advice. This finding becomes even more apparent from the esti- mates obtained after controlling for country-strata fixed effects (see second col- umn of table 5.3). In this case, the marginal effect on stunting of investing in group B is −2.3 percentage points compared to the −2-percentage-point mar- ginal effect of investing in group A. The small differences in the marginal effects on stunting associated with investing in group A versus group B suggest that the final decision will have to be based primarily on cost considerations. Table 5.4 presents estimates of equation (5.1) for different country groupings, such as countries with relatively higher or lower values (among the group of 33 countries analyzed in this report) of gross national income (GNI), Gini index of inequality, tariff rates, first wave countries as categorized by the World Bank versus all other countries, high- versus low-fragility countries, countries whose consumption is low or high in starch, landlocked versus nonlandlocked coun- tries, and countries in the Sahel regions versus those in non-Sahel regions. The list of countries composing each group may be found in table 2.1 in chapter 2. The estimates in table 5.4 reveal that the same general patterns identified in table 5.2 also hold in most of these groups. Notable exceptions are the areas of countries in the Sahel region where access to one or more nutrition drivers does not appear to create a statistically significant decrease in the probability of stunting relative to the probability in the group of children with inadequate access to all three nutrition drivers. Also, in countries with relatively higher tariffs, typically associated with higher food prices, and in countries with low fragility, the marginal effects of access to a higher number of nutrition drivers on the probability of a child being stunted are typically higher. Thus, access to one or more nutrition drivers is associated with greater decreases in stunting in countries where food prices are higher because of higher import tariffs, which suggests that increased access to the nutrition drivers may be a more effective policy instrument against undernutrition in economies where foods prices are higher (see box 5.1). Along similar lines, table 5.5 presents estimates of equation (5.1) for different socioeconomic groups within countries, such as urban and rural areas, boys and girls, children residing in households in the bottom 20 percent (B20) and top Table 5.4  Marginal Effects on the Probability of Stunting of Simultaneous Access to One or More of the Drivers of Nutrition in Different Groups of Countries Low GNI High GNI Low Gini High Gini Low Tariff High Tariff Low Fragility High Fragility −0.034*** −0.013 −0.029*** −0.016* −0.016* −0.042*** −0.074*** −0.019 Adequate in any 1 driver (0.012) (0.008) (0.011) (0.008) (0.008) (0.011) (0.015) (0.016) −0.057*** −0.032*** −0.066*** −0.023** −0.034*** −0.066*** −0.106*** −0.017 Adequate in any 2 drivers (0.014) (0.011) (0.013) (0.011) (0.011) (0.015) (0.020) (0.021) −0.072*** −0.015 −0.091*** 0.003 −0.036* −0.044 −0.080** −0.059* Adequate in all 3 drivers (0.021) (0.024) (0.022) (0.022) (0.019) (0.030) (0.034) (0.031) Low Starch High Starch 1st Wave Non 1st Wave Sahel regions Non-Sahel regions Landlocked Non-Landlocked −0.016** −0.020 −0.011 −0.052*** −0.005 −0.031*** −0.025* −0.021*** Adequate in any 1 driver (0.008) (0.014) (0.009) (0.010) (0.013) (0.008) (0.015) (0.008) −0.038*** −0.021 −0.027** −0.074*** −0.022 −0.061*** −0.041** −0.043*** Adequate in any 2 drivers (0.011) (0.018) (0.012) (0.012) (0.020) (0.010) (0.017) (0.010) −0.016 −0.062** −0.019 −0.080*** 0.061 −0.082*** −0.076*** −0.016 Adequate in all 3 drivers (0.022) (0.031) (0.023) (0.019) (0.044) (0.016) (0.021) (0.022) Source: Estimates based on data for children under 24 months of age from 31 country Demographic and Health Surveys from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. No information on mother’s height is available for Angola or Senegal and therefore they are not part of the analyses with covariates. Note: Marginal effects are based on the coefficient estimates obtained from the logit model in equation (5.1). Robust standard errors in parentheses corrected for correlation at the cluster level. Detailed estimates available upon request. All regressions include country and year fixed effects as well as child, parental, and household characteristics that consist of the following variables: indicator variables for age (in months), gender, birth order, and multiple birth; mother’s education in years; indicator for mother’s marital status; mother’s height; mother’s age; wealth quintile; number of children under five years of age; number of household members; and whether household is in a rural or urban area. High- and low-fragility countries are categorized using IDA’s resource allocation index. The 12 countries in the study are divided into two groups, such that 6 are “low fragility” and 6 are designated as “high fragility.” See table 2.1 for specific countries in each group. First wave country designation based on the World Bank classification. The first wave countries included in the study are Burkina Faso, Cameroon, Côté D’Ivoire, Ethiopia, Madagascar, Malawi, Mali, Mozambique, Niger, Nigeria, Rwanda, and Tanzania. GNI = gross national income. ***p < 0.01, **p < 0.05, *p < 0.1. 95 96   ALL HANDS ON DECK BOX 5.1 Tariffs, Trade, and Child Stunting In this report, the sample of 33 countries is partitioned into two groups of countries with relatively higher and lower tariffs to explore any potential differences in links and interactions between access to the underlying drivers of nutrition and stunting. It is important to bear in mind that import tariff rates are an imperfect measure of the actual ease of trade given many non-tariff-based barriers, such as licensing schemes, quotas, and customs rules. The removal of trade barriers, such as import tariffs, changes the national availabil- ity of food—in terms of both the types of foods available and their price—through changes in both imports and national production. At the household level, food avail- ability is affected by these changes not only directly but also indirectly through any changes to the household’s income from the shifts in the country’s production profile. Whereas the impact of trade liberalization on economic growth and poverty has been extensively studied (for example, Dollar and Kraay 2004), very few studies link trade barriers to malnutrition measures per se. In a review of the effects of agricultural devel- opment policies, including trade liberalization, on nutrition, Dangour et al. (2013) after exhaustive search criteria find only four studies that explore the relationship empirically, of which only one measures undernutrition rates in children (in Andhra Pradesh in India). Although trade liberalization accelerates growth, and poor households on aver- age proportionately benefit from growth as much as nonpoor households (Dollar and Kraay 2004), it is not evident that the welfare of a household, especially of those that rely on agriculture for their livelihoods, is improved (Calì, Hollweg, and Bulmer 2015). Furthermore, knowing that welfare is affected is not sufficient to draw conclusions on how the nutritional status of young children would be affected. 20 percent (T20) of the asset index distribution with each country, and children with more educated mothers (more than seven years of school). For each of these groups, the same general patterns prevail in the relationship between the probabil- ity of stunting and access to one or more nutrition drivers. What is striking in these estimates, however, is the difference in the coefficients α1, α2, and α3 between some groups. For example, the marginal effect of access to one or more nutrition drivers on the probability of stunting is higher for boys than for girls and for children with more educated mothers than with less educated mothers (see table 5.5). The heterogeneity of these marginal effects provides useful information on the extent to which programs increasing access to one or more nutrition drivers may benefit some children more than others, depending on some key charac- teristics of the child (boys) or the household (educated mother). The results also underline the fact that, given the low prevalence of access to two or three deter- minants for children from the poorest wealth quintile, only the estimates for Table 5.5  Marginal Effects on the Probability of Stunting of Simultaneous Access to One or More of the Drivers of Nutrition for Different Areas and Socioeconomic Groups within Countries Mother more Mother less Urban Rural Boys Girls B20 T20 educated educated −0.026** −0.020** −0.037*** −0.005 −0.035** −0.034* −0.023* −0.021** Adequate in any 1 driver (0.012) (0.008) (0.010) (0.010) (0.014) (0.018) (0.012) (0.009) −0.040*** −0.043*** −0.045*** −0.039*** −0.041** −0.043** −0.046*** −0.039*** Adequate in any 2 drivers (0.015) (0.011) (0.013) (0.012) (0.021) (0.020) (0.013) (0.012) −0.026 −0.057*** −0.043* −0.033 −0.065 −0.045* −0.053*** −0.022 Adequate in all 3 drivers (0.023) (0.022) (0.023) (0.023) (0.054) (0.025) (0.019) (0.025) Source: Estimates based on data for children under 24 months of age from 31 country Demographic and Health Surveys from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. No information on mother’s height is available for Angola or Senegal and therefore they are not part of the analyses with covariates. Note: Marginal effects are based on the coefficient estimates obtained from the logit model in equation (5.1). Robust standard errors in parentheses corrected for correlation at the cluster level. Detailed estimates available upon request. All regressions include country and year fixed effects as well as child, parental, and household characteristics that consist of the following variables: indicator variables for age (in months), gender, birth order, and multiple birth; mother’s education in years; indicator for mother’s marital status; mother’s height; mother’s age; wealth quintile; number of children under five years of age; number of household members; and whether household is in a rural or urban area. B20 = lowest wealth quintile; T20 = top wealth quintile. ***p < 0.01, **p < 0.05, *p < 0.1. 97 98   ALL HANDS ON DECK access to one or two determinants are statistically significant. The fact that the coefficient estimate for α3 is not statistically significant does not imply that access to three determinants is not associated with better nutritional outcomes for these children; rather, it reflects the fact that there is insufficient information to confidently estimate the difference. Stunting and the Underlying Determinants of Nutrition The analysis in the previous section raises an important policy question: If access to one or more of the nutrition drivers is associated with lower stunting, which one of the three drivers is it? In the context of budgetary constraints, efforts to provide adequate access to at least one driver need to prioritize inter- ventions among different sectors. To this end, children are classified by their specific set of determinants (access to adequate food and care, access to ade- quate WASH, and access to adequate health services) to determine which com- binations have relatively stronger associations with the lower stunting rates. A variant of the methodology of the previous section is adopted to shed light on this issue. Seven binary variables are constructed by considering whether the child has access to adequate levels in the other drivers. Specifically, the binary variable FC is equal to 1 if the household has adequate care and food security only (and inadequate health and environment) and 0 otherwise. Similarly, W is 1 when the household is adequate in WASH services (and inadequate in food and care and health), and H is 1 when the household is adequate in health ser- vices only (and inadequate in care and food and WASH) and is 0 otherwise. In the same fashion, FC_H takes the value of 1 if the household has adequate food and care practices and adequate health services at the same time (and inade- quate WASH), and FC_W takes the value of 1 if the household has adequate food and care and adequate WASH (and inadequate health). The model estimated using the logit model for binary dependent variables is as follows:3  + βFCFC + βWW + βHH + βFC_WFC_W + βFCHFC_H Stunted = α + βW_HW_H + βFC_W_HAll3 + X + μ + ε(5.2) In equation (5.2), the subscripts for child i in country c are dropped to mini- mize clutter, and the dependent variable Stunted along with the control vari- ables X and μ are identical to those in equation (5.1). As in equation (5.1), the constant term α provides an estimate of the stunting rate (or the probability of being stunted) among children in the reference group (that is, with inadequate access to all three drivers of nutrition: care and food security (FC = 0), WASH/ environment (W = 0), and health (H = 0). The coefficients β yield estimates of Stunting and Access to the Underlying Determinants of Nutrition   99 the marginal effect on the probability of a child being stunted when a child has access to adequate levels in only one specific driver or a specific combination of two of the three drivers (holding child, parental, and household characteristics, X, constant). Specifically, the coefficient βFC yields an estimate of the decline in the probability of stunting associated with access to adequate food and care only (FC = 1) but inadequate access to health (H = 0), and inadequate access to WASH (W = 0), in comparison to the probability of a child being stunted in the reference group summarized by the constant term α. The coefficients βH and βW have analogous interpretations for health and WASH, respectively. Table 5.6 presents the estimates of equation (5.2) including country-level fixed effects and strata-level fixed effects with the controls X excluded (left col- umn) and then included (right column). When the control variables X are excluded, the coefficients can be interpreted as the difference in the stunting rate Table 5.6  The Marginal Effect on the Probability of Stunting of Access to Adequate Levels in Combinations of the Underlying Drivers of Nutrition Variables Country fixed effects Country-strata fixed effects −0.009 −0.003 Food/care determinant (FC ) (0.013) (0.012) 0.032** −0.019 WASH determinant (W ) (0.012) (0.011) −0.048*** −0.030*** Health determinant (H) (0.008) (0.008) −0.002 −0.032 Food/care and WASH (FC_W ) (0.021) (0.020) −0.066*** −0.050*** Food/care and health (FC_H) (0.012) (0.011) −0.041*** −0.043*** WASH and health (W_H) (0.012) (0.011) −0.044*** −0.054*** All 3 (0.017) (0.016) Observations 67,125 66,992 Source: Estimates based on 31 recent Demographic and Health Surveys (DHS) from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. No information on mother’s height is available for Angola or Senegal, and therefore they are not part of the analyses with covariates. Note: Marginal effects are based on the coefficient estimates obtained from the logit model in equation (5.2). Robust standard errors in parentheses corrected for correlation at the cluster level. Stratum is the lowest level of statistical representation of the DHS within a country, typically identifying regions and urban/rural areas within regions in a country. All regressions include year fixed effects as well as child, parental, and household characteristics that consist of the following variables: indicator variables for age (in months), gender, birth order, and multiple birth; mother’s education in years; indicator for mother’s marital status; mother’s height; mother’s age; wealth quintile; number of children under five years of age; number of household members; and whether household is in a rural or urban area. ***p < 0.01, **p < 0.05, *p < 0.1. 100   ALL HANDS ON DECK among children in different groups relative to the group of children with inad- equate access to all three nutrition drivers. When the control variables X are included, the coefficients can be interpreted as the marginal effects in the prob- ability of stunting relative to the probability of stunting in the reference group included in the constant term. The estimates in table 5.6 confirm the leading role of the nutrition-specific interventions carried out primarily by the health sector. This result is also robust to the inclusion of country-strata fixed effects instead of country fixed effects. Controlling for child, parental, and household characteristics as well as for the country of residence, the probability of stunting associated with having access to adequate health only decreases by 4.8 percentage points when the geographic loca- tion of the household within a country is not taken into consideration, and by 3.0 percentage points after taking into account potential differences associated with living in an urban or rural area and a specific region of the country. In both cases the marginal effect of health on the probability of stunting is greater than the mar- ginal effect of having access to adequate food and care only,4 or access to adequate WASH only. Thus, if budgetary or other considerations allow for interventions covering deprived children by only one sector, this sector should be health. As discussed (see table 5.3), the estimates in table 5.6 could also provide policy guidance on which sector could be scaled up in a targeted group that is already covered by the health sector. Controlling for child, parental, and house- hold characteristics as well as for the geographic location of the household within the country of residence (that is, for country-strata fixed effects), the probability of stunting associated with having simultaneous access to ade- quate health and adequate WASH (W_H), decreases by 4.3 percentage points relative to the reference group (that is, children with inadequate access to all three nutrition drivers). This result implies that the marginal decline in stunting associated with adding access to adequate WASH services to the group of chil- dren who already have access to adequate health is 1.3 percentage points (−0.043 − [−0.030] = −0.013). The marginal decline in stunting associated with adding access to WASH is similar to the 2.0-percentage-point marginal decline in the probability of stunting associated with providing simultaneous access to adequate health and adequate food and care. As earlier (see discussion of table 5.3), the estimates of the differences in the marginal effects are not suffi- ciently robust to serve as a basis for policy advice,5 suggesting that the final decision will have to be based primarily on cost considerations. In any case, it is important to bear in mind that the relationship prevailing in the pooled sample of children may not necessarily reflect the relationship that may prevail within any given country and serves as a warning against policies being guided by evidence that is not country specific. Table 5.7 conducts a more in-depth investigation, by replacing in the regres- sions the three drivers of nutrition with the individual components of the three Stunting and Access to the Underlying Determinants of Nutrition   101 Table 5.7  The Marginal Effects on Stunting Prevalence of Access to the Components of the Underlying Drivers of Nutrition Country fixed effects Country-strata fixed effects Including child, Including child, parental, and parental, and No household No household Variables covariates controls covariates controls −0.060*** −0.035*** −0.052*** −0.028*** F: Child has adequate food intake (0.008) (0.009) (0.008) (0.008) −0.008 −0.005 −0.008 −0.007 C1: Immediate skin-to-skin contact (0.007) (0.007) (0.007) (0.007) 0.053*** 0.053*** 0.037*** 0.040*** C2: Age-appropriate breastfeeding (0.007) (0.009) (0.007) (0.009) W1: Improved non-shared sanitary −0.057*** −0.032** −0.056*** −0.029** facilities (0.013) (0.014) (0.013) (0.013) −0.006 0.007 0.001 0.002 W2: Improved water source (0.007) (0.008) (0.007) (0.007) −0.036*** −0.012 −0.026** −0.017 W3: Handwashing facility on site (0.012) (0.013) (0.012) (0.013) W4: Percentage of households using −0.051*** −0.076*** 0.015 0.011 open defecation in cluster (0.013) (0.014) (0.014) (0.014) 0.073*** 0.049*** 0.048*** 0.023 W5: Disposal of feces (0.015) (0.016) (0.014) (0.015) −0.048*** −0.039*** −0.021*** −0.024*** H1: Had 4 or more prenatal checks (0.006) (0.007) (0.006) (0.006) −0.056*** −0.044*** −0.008 −0.015** H2: Birth assisted by trained professional (0.007) (0.007) (0.007) (0.007) H3: Child has been to at least one postnatal 0.011 0.004 0.021*** 0.006 checkup (0.007) (0.007) (0.007) (0.007) −0.111*** −0.027*** −0.104*** −0.022*** H4: Vaccinations up to date (0.007) (0.007) (0.006) (0.007) −0.031*** −0.023*** −0.017*** −0.013* H5: Child sleeps under a mosquito bed net (0.007) (0.007) (0.007) (0.007) Observations 51,724 46,685 51,491 46,456 Source: Estimates based on 26 recent Demographic and Health Surveys (DHS) from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Cameroon, Republic of Congo, Gabon, and Niger have no information on handwashing facilities; Republic of Congo and Gabon have no information on disposal of feces; Ethiopia and Lesotho have no information on mosquito nets; and Mozambique has no information on postnatal checkups. These countries are therefore not part of the pooled components analyses. No information on mother’s height is available for Angola or Senegal, and therefore they are not part of the analyses with covariates. Note: Marginal effects are based on the coefficient estimates obtained from the logit model in equation (5.2). Robust standard errors in parentheses corrected for correlation at the cluster level. Stratum is the lowest level of statistical representation of the DHS within a country, typically identifying regions and urban/rural areas within regions in a country. Child, parental, and household characteristics consist of the following variables: indicator variables for age (in months), gender, birth order, and multiple birth; mother’s education in years; indicator for mother’s marital status; mother’s height; mother’s age; wealth quintile; number of children under five years of age; number of household members; and whether household is in a rural or urban area. ***p < 0.01, **p < 0.05, *p < 0.1. 102   ALL HANDS ON DECK drivers of nutrition. As in table 5.6, estimates are presented by excluding and then including child, parental, and household characteristics (X) with country fixed effects and then with strata-level fixed effects. Except for access to improved sanitation, all the variables related to water (W1–W5) turn statisti- cally insignificant when strata-level fixed effects are included along with other controls X. This is because there is very little or no variation among households in access to the different components of WASH within strata. The coefficient of age-appropriate breastfeeding, however, appears to have a significant positive effect on stunting irrespective of the specification estimated. One possible explanation for this rather unexpected sign is the possibility of endogeneity bias, arising from age-appropriate breastfeeding behavior being followed closely mainly for children who are stunted. A related explanation is selection bias. Part of the positive correlation between breastfeeding (both exclusive and complementary) and stunting is due to the fact that children who are breastfed are less likely to die (Victora et al. 2016). The children who would have died had they not received breastmilk are most likely the weaker ones, so in the pool of breastfed children are those who would have died otherwise and who do not make it in the nonbreastfed group.6 A Brief Summary of the Country-Specific Estimates Detailed estimates of regression equation (5.2) (with strata fixed effects) for each of the 33 countries analyzed in this report can be found in appendix C. It is important to point out, however, that the determinant that is most fre- quently correlated with lower stunting is access to health care. Access to WASH is associated with lower probabilities of stunting in Gabon, Guinea, Lesotho, and Uganda. Access to health is associated with lower probabilities of stunting in Burundi, Cameroon, Chad, Democratic Republic of Congo, Kenya, Lesotho, Niger, Namibia, Niger, Togo, and Zimbabwe. Access to food/care is associated with lower probabilities of stunting in Angola, Benin, Chad, Democratic Republic of Congo, and The Gambia. Also, the combination of WASH and health is associated with lower probabil- ity of stunting in nine countries, WASH and food/care in two countries, health and food/care in seven countries, and access to all three in seven countries. Testing the Sensitivity of the Findings The analysis so far has focused on the extent to which the access to adequate levels of the underlying determinants is associated with stunting, which is a binary variable. In this section, the sensitivity of the findings is explored by Stunting and Access to the Underlying Determinants of Nutrition   103 applying the unconditional quantile regressions (UQR) method on different quantiles of the distribution of HAZs. The UQR method (Firpo, Fortin, and Lemieux 2009) allows for the estimation of the effect of a variable of interest, such as increased access to an adequate level in one or more of the drivers of nutrition, at specific points (for example, lower 20th percentile) of the uncon- ditional or marginal distribution of HAZ. Thus, the UQR method allows for the possibility that the effects of increased access to the drivers of nutrition may differ between children who have lower HAZs and children who have higher HAZs. In fact, the method allows for the possibility of separate effects at any quantile of interest. Given that stunting prevalence in the sample of 33 countries analyzed in this report ranges between 10 and 50 percent, we estimate equations (5.3a and 5.3b) using the UQR method at the bottom 10 percent, 20 percent, 30 percent, and 40 percent of the distribution of HAZs. HAZ = α + α1Aany1 + α2Aany2 + α3Aall3 + X + μ + ε (5.3a) α + βFCFC + βWW + βHH + βFC_WFC_W + βFC_HFC_H + HAZ =  βW_HW_H + βFC_W_HAll3 + X +μ + ε(5.3b) The subscripts for child i in country c are dropped to minimize clutter. Figure 5.7 presents the differential effect of simultaneous access to one or more drivers of nutrition on height-for-age by quantiles of HAZs—that is, equa- tion (5.3a)—for children 0–23 months of age. (Only coefficients that are statisti- cally significant are included in the figure.) Among children at the bottom 10 percent of the distribution of HAZ, the only two variables that have a positive and statistically significant effect on height-for-age are whether the child has access to one or two of the three nutrition drivers (Aany1 and Aany2). The absence of any significant effect of the All3 variable is because very few of the children at the bottom decile of the HAZ distribution have simultaneous access to all three of the nutrition drivers. The same pattern holds for children at the bottom 20 percent of the HAZ distribution; both Aany1 and Aany2 are signifi- cantly positive, but access to any two appears to have a slightly stronger effect on the HAZ. At the 30th quantile, access to all three nutrition drivers is associ- ated with an increase in HAZ of 0.15 s. d., whereas at the 40th quantile, access to one of the three nutrition drivers is associated with an increase in HAZ of 0.19 s. d. In sum, these estimates suggest that access to one or two or all three of the drivers of nutrition is associated with a heterogeneous effect on HAZ—an effect that varies according to the level of the HAZ. Access to one nutrition driver has the highest effect on HAZ among children at the bottom of the distribution of HAZ (bottom 10 percent) whereas the effect size of access to two nutrition 104   ALL HANDS ON DECK Figure 5.7  The Effect of Simultaneous Access to One or More of the Drivers of Nutrition at the Lower Quantiles of the HAZ Distribution 0.25 0.20 Standard deviations 0.15 0.10 0.05 0 Bottom 10% Bottom 20% Bottom 30% Bottom 40% Access to 1 determinant Access to 2 determinants Access to all 3 determinants Source: Estimates based on 33 recent Demographic and Health Surveys from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: Unconditional quantile regression estimates obtained including country fixed effects. Detailed estimates available upon request. The regressions include child, parental, and household characteristics including the following variables: indicator variables for age (in months), gender, birth order, and multiple birth; mother’s education in years; indicator for mother’s marital status; mother’s height; mother’s age; wealth quintile; number of children under five years of age; number of household members; and whether household is in a rural or urban area. drivers increases between the 10th and 30th quantiles of the HAZ distribution reaching a maximum at the 30th quantile. Also, the marginal effect of increased access to nutrition drivers seems to vary according to the level of the HAZ. The marginal effect on HAZ of access to two nutrition drivers at the 20th quantile (or the difference between the red and blue bars at the 20th quantile) is consid- erably smaller than the marginal effect of access to two nutrition drivers at the 30th or 40th quantile. Figure 5.8 presents the differential effect of access to combinations of the three specific drivers of nutrition (food and care, WASH, and health) on height- for-age by quantiles of HAZs—that is, equation (5.3b). Access to health only has a positive effect on height-for-age, regardless of the quantile evaluated. The importance (or marginal effect on HAZ) of access to only health is higher at lower quantiles of HAZ. Access to WASH alone has a negative correlation with height-for-age. This is an unexpected result, although, in general, the group of children with access to adequate WASH services only and inadequate access to the other drivers of Stunting and Access to the Underlying Determinants of Nutrition   105 Figure 5.8  The Effects of Access to Combinations of the Three Drivers of Nutrition at the Lower Quantiles of the HAZ Distribution All 3 determinants WASH and health Food and care and health Food and care and WASH Health WASH Food and care –0.5 –0.4 –0.3 –0.2 –0.1 0 0.1 0.2 0.3 0.4 0.5 Standard deviations Bottom 40% Bottom 30% Bottom 20% Bottom 10% Source: Estimates based on 33 recent Demographic and Health Surveys from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Note: Unconditional quantile regression estimates obtained including country fixed effects. Detailed estimates available upon request. The regressions include child, parental, and household characteristics including the following variables: indicator variables for age (in months), gender, birth order, and multiple birth; mother’s education in years; indicator for mother’s marital status; mother’s height; mother’s age; wealth quintile; number of children under five years of age; number of household members; and whether household is in a rural or urban area. nutrition such as health and food and care is most likely a unique group in itself. Access to both food and care and health is positively correlated with height-for- age across the four quantile regressions. Access to both WASH and health and access to all three are also positively correlated with height-for-age. However, WASH and health are not correlated with HAZ at the bottom quintile, and all three are correlated with HAZ at only the 30th and 40th quantiles. Notes 1. Strictly speaking, because the control vector Xic and the country fixed effects μc are included in the regression, the constant term a is the stunting rate among ­children in a certain country (the one associated with the omitted country indicator) with inad- equate access to all three drivers of nutrition and a specific combination of child, parental, and household characteristics that depends on the way the set of control variables in Xic is constructed. 2. Sampling weights were standardized across the various DHS, following the proce- dure outlined by DHS. The sampling weight of the mother was multiplied by the ratio of the total number of women 15 to 49 years of age in the surveyed country to 106   ALL HANDS ON DECK the number of women (15 to 49 years of age) surveyed. The total number of women was obtained from UN DESA (2015). 3. A more detailed discussion of the regression model and the interpretation of the coefficients is provided in appendix A, and appendix B provides a discussion of a model directly analyzing synergies among the determinants. 4. The lack of significance of the F_W (food/care and WASH) variable is due to the very small number of children in this group. As table 5.1 indicates, only 1 percent of the total sample of children has simultaneous access to adequate food/care and WASH. In the country fixed effects model, the positive and significant marginal effect of WASH on stunting is mainly driven by the component on the presence of proper facilities for the disposal of feces (W5). As can be seen in table 5.7, this is the only component of WASH that has a positive significant effect in three of the four speci- fications estimated. 5. It is also important to point out that, unless the geographic location of the household within the country of residence is taken into consideration, different policy conclu- sions emerge. The estimates in table 5.6, controlling for country fixed effects, imply that the marginal effect of simultaneous access to health and WASH on stunting is smaller than the marginal effect of having access to health only (−0.048), whereas the marginal effect of simultaneous access to food/care and health is −0.066, imply- ing a 1.8-percentage-point decline in stunting above what is accomplished by access to health only. 6. In an exhaustive review of the literature, Horta and Victora (2013) and Victora et al. (2016) did not find evidence that breastfeeding affects child growth (based on studies in mainly middle-income countries). References Calì, M., C. H. Hollweg, and E. R. Bulmer. 2015. “Seeking Shared Prosperity through Trade.” Policy Research Working Paper 7314, World Bank, Washington, DC. Dangour, A. D., S. Hawkesworth, B. Shankar, L. Watson, C. S. Srinivasan, E. H. Morgan, L. Haddad, and J. Waage. 2013. “Can Nutrition Be Promoted through Agriculture-Led Food Price Policies? A Systematic Review.” BMJ Open 3 (6): e002937. doi:10.1136​ /bmjopen-2013-002937. Dollar, D., and A. Kraay. 2004. “Trade, Growth, and Poverty.” Economic Journal 114 (493): F22–49. Firpo, S., N. Fortin, and T. Lemieux. 2009. “Unconditional Quantile Regressions.” Econometrica 77 (3): 953–73. Horta, B. L., and C. G. Victora. 2013. Long-Term Effects of Breastfeeding: A Systematic Review. Geneva: World Health Organization. UN DESA (United Nations Department of Economic and Social Affairs). 2015. World Population Prospects 2015. New York: United Nations. Victora, C. G., R. Bahl, A. J. Barros, G. V. França, S. Horton, J. Krasevec, S. Murch, M. J. Sankar, N. Walker, N. C. Rollins, and The Lancet Breastfeeding Series Group. 2016. “Breastfeeding in the 21st Century: Epidemiology, Mechanisms, and Lifelong Effect.” Lancet 387 (10017): 475–90. Chapter 6 An Application at the Country Level: The Case of Tanzania The analysis so far has investigated the relationship between access to nutrition drivers and stunting prevalence using cross-sectional data on ­ ­ children pooled across 33 countries in Sub-Saharan Africa. As such, the patterns of correlation emerging from the analysis reflect relationships that prevail, on average, among stunting and access to the drivers of nutrition by children residing in different countries. This approach requires that uni- form definitions and thresholds based on international standards be applied across all countries. In addition, the threshold levels used to define adequate access to health; food and care; and water, sanitation, and hygiene (WASH) have to be sufficiently lax to not preclude representation of the children in countries that are relatively more disadvantaged. Pooling data on children across many countries, using common definitions and thresholds, offers an advantage not only in terms of increased sample size overall but also in that pooling increases the number of observations on children with access to adequate levels of different combinations of nutrition drivers. This advan- tage, however, comes at a cost, because the relationship prevailing in the pooled sample of children may not necessarily reflect the relationship that may prevail within any given country. Bearing these considerations in mind, this chapter applies the same methodology using data from only one country, Tanzania. Appendix C accompanying this report includes a brief for each of the 33 countries included in the report summarizing access to the determinants of nutrition and their components as well as the simple correlation between stunting and the number of determinants. Appendix C presents the country-specific regression (logit) estimates of stunting and access to combinations of drivers of nutrition. 107 108   ALL HANDS ON DECK Stunting and the Underlying Determinants of Nutrition in Tanzania With a stunting rate of 28 percent among children under two years of age, Tanzania ranks around the middle among the 33 countries included in this study (for example, 16th in rank from the country with the highest stunting rate, Burundi, and 18th from the country with the lowest stunting rate, Ghana; see figure 3.5 in chapter 3). Tanzania’s ranking provides the opportunity of determining the extent to which relationships among sample averages based on the sample of children pooled across the 33 different countries also hold for a specific country, such as Tanzania, with an average stunting rate. Figure 6.1 highlights the fact that growth faltering is prevalent among chil- dren in Tanzania, with growth faltering being faster in rural areas relative to Figure 6.1  Growth Faltering in Tanzania, by Socioeconomic Characteristics (Children 0–59 Months) a. Urban versus rural b. Boys versus girls 0 0 –0.5 –0.5 Average HAZ Average HAZ –1.0 –1.0 –1.5 –1.5 –2.0 –2.0 –2.5 –2.5 0 6 12 18 24 30 36 42 48 54 60 0 6 12 18 24 30 36 42 48 54 60 Age (months) Age (months) Urban Rural Boy Girl c. Top wealth quintile versus bottom d. Educated mother versus less wealth quintile educated mother 0 0 –0.5 –0.5 Average HAZ Average HAZ –1.0 –1.0 –1.5 –1.5 –2.0 –2.0 –2.5 –2.5 0 6 12 18 24 30 36 42 48 54 60 0 6 12 18 24 30 36 42 48 54 60 Age (months) Age (months) Top quintile 7 years of school or more Bottom quintile Less than 7 years of school Source: Estimates based on Tanzania 2016 Demographic and Health Survey. Note: For this graph, children are considered younger than 60 months. HAZ = height-for-age Z-score. An Application at the Country Level: The Case of Tanzania   109 urban areas, for boys relative to girls, for children in less wealthy households (for example, the bottom 20 percent of the wealth distribution), and for children whose mother has a relatively lower level of education. Thus, an emphasis on the first 1,000 days window in Tanzania is necessary if policies and specific programs are to have an ultimate effect on the prevalence of chronic undernutrition. Figure 6.2 presents the prevalence of stunting among children under two years of age in Tanzania, as well as for different geographic areas and different socioeco- nomic groups. Clearly, Tanzania is characterized by substantial differences in the prevalence of stunting between rural and urban areas, between less and more wealthy households, and between children with less or more educated mothers. There do not appear to be any significant differences in the prevalence of stunting among children with more empowered or less empowered mothers. Mothers with more autonomy in decision making—empowered—are defined using the top half (> 50th percentile) of the values of the first principal component derived from the following variables: whether married mothers are employed all year or in seasonal work and whether they usually make the deci- sions on how to spend their earnings, on their health care, on large household purchases, on visits to family or relatives, and on what to do with the money their husbands earn. Figure 6.2  Stunting Prevalence among Children under 24 Months 35 33 32 30 30 28 Stunting rate (%) 27 26 25 21 20 18 15 National Location Income level Mother’s Mother’s education empowerment Subpoplation groups National Rural Bottom wealth quintile Less Educated Less Empowered Urban Top wealth quintile Educated Empowered Source: Estimates based on Tanzania 2016 Demographic and Health Survey. 110   ALL HANDS ON DECK The panels in figure 6.3 summarize the access to the components of food and care, environment (WASH), and health, among children 0–23 months of age, at the national level, for rural and urban areas in the country, and for children in the wealthiest (the top wealth quintile, or T20) and in the poorest (the bottom wealth quintile, or B20) households. Specifically, food and care consists of the following components, each summarized by a corresponding binary variable (0 = no, 1 = yes): if the child, depending on its age, consumes a minimum acceptable diet (based on types of foods consumed and feeding frequency); whether breastfeeding was initi- ated within an hour of birth; and whether the child is age-appropriately breastfed at the time of the survey. WASH or environment consists of the following binary com- ponents: (a) access to an improved source of water for drinking, (b) access to basic sanitation in the dwelling, (c) access to proper children’s feces disposal, (d) access to a handwashing station with soap, and (e) living in a community where less than 25 percent of the households openly defecate. Last, health consists of the following binary components: (a) mother used prenatal services at least four times while preg- nant, (b) child was delivered by a skilled professional, (c) child received a postnatal check within two months of birth, (d) child is compliant with national vaccination schedule, and (e) child sleeps under a mosquito net. Figure 6.3 reveals that access to the component of food and care is consider- ably less heterogeneous between urban and rural areas, and between the least wealthy (B20) and wealthiest (T20) households, than access to the determinants of WASH and health. For example, the percentage of children from wealthier households (T20) with access to improved water, basic sanitation, proper dis- posal of feces, handwashing facilities, more than four prenatal visits, postnatal care, vaccinations, and sleeping under mosquito nets is considerably higher than the percentage of children from less wealthy households (B20) with access to the same determinants of nutrition. Following the methodological approach applied to the pooled country data, the various underlying determinants of nutrition may be categorized into three groups: (a) adequate household food security and care practices, (b) adequate household environment (WASH), and (c) adequate health services. By necessity, the aggregation of 13 different determinants of nutrition into three aggregate groups (food/care, WASH, and health) involves a number of decisions that can be questionable. The criteria adopted in the report were based on practical consider- ations. First, the three aggregate groups (food/care, WASH, and health) arguably correspond to the activities and operations of different sectors. Second, the thresholds used to identify whether a child has access to “adequate” food/care, “adequate” WASH, or “adequate” health were rather lax in order to allow a suffi- cient number of children in the different groups or combinations of these groups. For the purpose of comparison, table 6.1 presents the percentage of children with access to adequate food/care, adequate WASH, and adequate health in different combinations using the less strict definition for adequate used in this ­ report, and a stricter definition that considers access as adequate only if the An Application at the Country Level: The Case of Tanzania   111 Figure 6.3  Access to the Different Components of Food/Care, WASH, and Health (Children under 24 Months) a. Children’s access to b. Children’s access to adequate water, adequate food and care sanitation, and hygiene 56 26 Improved 45 83 Minimum 26 water 40 26 87 acceptable diet 24 28 15 Basic 8 35 sanitation 1 52 68 13 Immediate 67 Feces 6 69 30 breastfeeding 67 disposal 1 71 43 81 OD <25% 77 90 in community 68 73 97 Appropriate 75 68 4 breastfeeding 76 Handwashing 3 70 8 facilities 2 13 0 20 40 60 80 100 0 20 40 60 80 100 Share of children who meet criteria (%) Share of children who meet criteria (%) c. Children’s access to adequate health 48 42 Prenatal visits 64 39 64 64 55 Assisted birth 88 46 94 36 34 Postnatal visit 41 32 43 69 63 Vaccinations 84 58 86 64 58 Mosquito nets 81 49 84 0 20 40 60 80 100 Share of children who meet criteria (%) National Rural Urban Bottom quintile Top quintile Source: Estimates based on Tanzania 2016 Demographic and Health Survey. Note: In panel a, “Appropriate breastfeeding” is exclusive breastfeeding for the first 6 months and complementary breastfeeding from 6 to 23 months. Panel b estimates are based on all households in the child’s primary sampling unit. OD = open defecation; WASH = water, sanitation, and hygiene. 112   ALL HANDS ON DECK Table 6.1  Percentage of Children (under 24 Months) in Tanzania with Access to the Three Drivers of Nutrition, Using Two Different Definitions of “Adequate” Less strict definition used in the report Strict definition Percentage of children adequate in Percentage of children adequate in   National Rural Urban National Rural Urban None 29 37 9 75 79 67 Food/care 9 11 3 15 16 12 WASH 1 1 2 1 0 2 Health 34 32 41 7 4 15 Food/care and WASH 0 0 0 0 0 0 Food/care and health 13 13 13 2 1 3 WASH and health 10 4 24 0 0 1 All 3 3 1 8 0 0 1 Source: Estimates based on Tanzania 2016 Demographic and Health Survey. Note: WASH = water, sanitation, and hygiene. child has access to all of the components in the determinant (all three compo- nents of food/care, all five components of WASH, and all five components of health). It should also be noted that the categories reported in table 6.1 are mutually exclusive, meaning that children in any of these categories do not have access to any of the other drivers of nutrition. For example, at the national level, 34 percent of children have access to adequate health only and no access to adequate food/care or adequate WASH, or only 10 percent of the children have access to adequate WASH and health at the same time. Table 6.1 reveals that the strict definition of adequate access results in a very small percentage of children being categorized as having access to any of the determinants. With the strict definition, the percentage of children with access to adequate health only, at the national level, decreases to 7 percent (from 34 percent using the less strict definition). There is also a large decrease in the percentage of children with simultaneous access to WASH and health (from 10 percent down to 0 with the strict definition). Clearly, applying the strict criterion of adequate access renders this methodological approach useless because 75 percent of the children end up being classified as not having access to an adequate level in any of the three determinants of nutrition. For this reason, the Tanzania case study has been conducted using the less strict definition for adequate used in this report.1 Correlation between Stunting and Access to the Drivers of Nutrition Map 6.1 presents maps with rates of stunting and access to adequate WASH and health by region of Tanzania.2 Such maps are not only helpful for highlighting An Application at the Country Level: The Case of Tanzania   113 Map 6.1  Tanzania: Regional Stunting Rates and Access to Health and WASH a. Regional stunting rates for children under 2 years old Regional Stunting Rates Kagera Mara (children 0–23 months) 15%–21% Mwanza Geita Simiyu Arusha 22%–30% Shinyanga 31%–48% Kilimanjaro Kigoma Manyara Pemba North Tabora Tanga Pemba South Singida Zanzibar North Dodoma Zanzibar West Katavi Zanzibar South and Central Dar es Salaam Pwani Iringa Rukwa Mbeya Morogoro Njombe Lindi Ruvuma Mtwara IBRD 44401 | MAY 2019 This map was produced by the Cartography Unit of the World Bank Group. The boundaries, colors, denominations and any other information shown on this map do not imply, on the part of the World Bank Group, any judgment on the legal status of any territory, or any endorsement or acceptance of such boundaries. b. Share of children under 2 years old with access to adequate health by region Percent of children (0–23 months) Kagera Mara with adequate health by region 15%–21% Mwanza Geita Simiyu Arusha 22%–30% Shinyanga 31%–48% Kilimanjaro Kigoma Manyara Pemba North Tabora Tanga Pemba South Singida Zanzibar North Dodoma Zanzibar West Katavi Zanzibar South and Central Dar es Salaam Pwani Iringa Rukwa Mbeya Morogoro Njombe Lindi Ruvuma Mtwara IBRD 44402 | MAY 2019 This map was produced by the Cartography Unit of the World Bank Group. The boundaries, colors, denominations and any other information shown on this map do not imply, on the part of the World Bank Group, any judgment on the legal status of any territory, or any endorsement or acceptance of such boundaries. (continued next page) 114   ALL HANDS ON DECK Map 6.1 (continued) c. Share of children under 2 years old with access to adequate water, sanitation, and hygiene Percent of children Kagera Mara (0–23 months) with adequate environment by region Mwanza Simiyu Arusha 1.8%–12% Geita 13%–26% Shinyanga Kilimanjaro 27%–44% Kigoma Manyara Pemba North Tabora Tanga Pemba South Singida Zanzibar North Dodoma Zanzibar West Katavi Zanzibar South and Central Dar es Salaam Pwani Iringa Rukwa Mbeya Morogoro Njombe Lindi Ruvuma Mtwara IBRD 44400 | MAY 2019 Source: Estimates based on Tanzania 2016 Demographic and Health Survey. Note: Data for blank areas in the maps are not included in the Demographic and Health Survey. WASH = water, sanitation, and hygiene. the heterogeneity of stunting and access to the drivers of nutrition across space; they are also useful for the targeting of nutrition-specific and nutrition-sensitive interventions. The maps of access to adequate WASH and adequate health provide informa- tion on the joint distribution of WASH and health that is essential for identifying important gaps in access that potentially affect the impact of other nutrition- related interventions. For example, the scaling up of nutrition-­ sensitive agricul- tural operations aiming to increase food security and dietary diversity is unlikely to take into consideration the extent to which there is access to improved water and sanitation or to adequate health services in the geographic areas under con- sideration. The maps of access to adequate WASH and health provide a more holistic view and pinpoint better the geographic areas where inadequacies in WASH or health (or in both) may be more prevalent, thus enabling the joint pri- oritization of operations and improved cost efficiency of interventions. The maps highlight the regional differences in access. Access to adequate health is concentrated in the eastern part of the country, whereas access to ade- quate environment (WASH) is concentrated in the south and northeast of the country. Even in these regions, however, stunting rates can be among the An Application at the Country Level: The Case of Tanzania   115 highest in the country. The north-central and western regions have relatively low prevalence of access to both health and WASH. Figure 6.4 provides a visual estimate of the extent to which access to only one, or simultaneous access to two, or simultaneous access to all three of the underlying drivers of nutrition is associated with lower prevalence of stunting in Tanzania.3 Figure 6.4 reveals that children with access to all three nutrition drivers have the lowest rate of stunting compared to children with access to only two drivers or one driver or access to none at all (0 drivers). Moreover, figure 6.4 suggests that the greatest reductions in stunting are associated with increases in access from any one driver to simultaneous access to any two drivers or with increases from simultaneous access to any two drivers to access to all three driv- ers of nutrition. This result contrasts with the finding from the pooled country data (see figure 5.6 in chapter 5), where, the marginal decline (gain) in the stunt- ing rate with increases in access from none (0) to any one driver is larger than that associated with a change in access from one to two drivers or from two to Figure 6.4  Access to No Drivers versus Access to One or More Drivers of Nutrition and the Prevalence of Stunting (Children 0–23 Months) 0.8 Cumulative share of children’s HAZ 0.6 0.4 0.2 0 −6 −5 −4 −3 −2 −1 0 Height-for-age Z-score Access to Zero determinants 1 determinant 2 determinants 3 determinants Source: Estimates based on Tanzania 2016 Demographic and Health Survey. Note: HAZ = height-for-age Z-score. 116   ALL HANDS ON DECK all three drivers. This finding also reinforces the point made earlier that the relationship prevailing in the pooled sample of children may not necessarily reflect the relationship that may prevail within any given country and serves as a warning against policies being guided by evidence that is not country specific. Table 6.2 provides quantitative estimates of the marginal decrease in the stunting rate relative to the stunting rate prevailing among children with access to an adequate level in none of the three drivers of nutrition).4 These estimates reinforce the preliminary findings from map 6.1. Consider, for example, the choice between allocating the same resources between two groups of children: group A composed of children who have inadequate access to all three nutrition drivers and group B composed of children who have adequate access to only one of the nutrition drivers. The estimated marginal effect of −0.043 of Aany1 (second column in table 6.2) implies that the probability of stunting for a child with access to any one driver is 4.6 percentage points lower than the probability of stunting in the reference group of children with inadequate access to all three nutrition drivers (group A). The coefficient −0.097 of Aany2 (second column in table 6.2) implies that the probability of stunting among children with access to any two of the drivers of nutrition is 9.5 percentage points lower than the prob- ability of stunting among children with inadequate access to all three nutrition drivers. This, implies that there is a 5.4-percentage-point decline in the proba- bility of stunting associated with having access to two drivers compared to one driver (−0.097−[−0.043] = −0.054). Thus, assuming the costs associated with providing access to a nutrition driver to group B are equal to the costs of provid- ing access to one driver to group A, a slightly greater decline in stunting can be Table 6.2  Changes in the Stunting Rate for Groups of Children with Simultaneous Access to One or More of the Drivers of Nutrition, Tanzania Including child, parental, and Variables No covariates household covariates −0.070*** −0.043** Aany1: Adequate in any 1 driver (0.023) (0.021) −0.168*** −0.097*** Aany2: Adequate in any 2 drivers (0.024) (0.026) −0.241*** −0.155*** Aall3: Adequate in all 3 drivers (0.037) (0.047) Observations 3,726 3,719 Source: Estimates based on data for children under 24 months of age from Tanzania Demographic and Health Survey 2016. Note: Marginal effects are based on the coefficient estimates obtained from the logit model in equation (5.1) excluding child, parental, and household characteristics. Robust standard errors in parentheses corrected for correlation at the cluster level. ***p < 0.01, **p < 0.05, *p < 0.1. An Application at the Country Level: The Case of Tanzania   117 accomplished by targeting the limited resources to provide access to a second driver to group B. It is important to point out that the preceding discussion is based on the implicit assumption that the objective of policy is to reduce the stunting (or head count) rate. One key implication of this objective is that it tends to favor the allocation of resources to the group of children who are already closer to the stunting threshold (that is, already have access to an adequate level in one of the nutrition drivers). An alternative policy objective, such as reducing the preva- lence of severe stunting among children (as opposed to the head count stunting rate assumed) is likely to favor the targeting of these resources to children in group A instead of children in group B. The estimates presented in table 6.3 are useful for addressing the question of whether access to specific nutrition drivers or to specific combinations of drivers is associated with lower stunting.5 These estimates confirm the earlier results obtained using the pooled data across Sub-Saharan African countries. In Tanzania, as is the Table 6.3  The Marginal Effects on the Probability of Stunting of Access to Adequate Level in Combinations of the Underlying Drivers of Nutrition, Tanzania Including child, parental, and Variables No covariates household covariates −0.110*** −0.032 Food/care determinant (FC) (0.031) (0.030) 0.001 −0.018 WASH determinant (W) (0.062) (0.068) −0.054** −0.043** Health determinant (H) (0.022) (0.022) −0.385*** −0.239* Food/care and WASH (FC_W) (0.135) (0.125) −0.195*** −0.094*** Food/care and health (FC_H) (0.028) (0.029) −0.136*** −0.098** WASH and health (W_H) (0.039) (0.040) −0.286*** −0.169*** All 3 (0.064) (0.062) Observations 3,726 3,719 Source: Estimates based on data for children under 24 months of age from the Tanzania Demographic and Health Survey 2016. Note: Marginal effects are based on the coefficient estimates obtained from the logit model in equation (5.2). Robust standard errors in parentheses corrected for correlation at the cluster level. Access to food/care and WASH (water, sanitation, and health) simultaneously predicts stunting perfectly. All regressions include child, parental, and household characteristics that consist of the following variables: indicator variables for age (in months), gender, birth order, and multiple births; mother's education in years; indicator for mother's marital status; mother's height; mother's age; wealth quintile; number of children under five years of age; number of household members; and whether household is in a rural or urban area. ***p < 0.01, **p < 0.05, *p < 0.1. 118   ALL HANDS ON DECK case on average in countries in the region, access to adequate health only is associ- ated with a significantly lower stunting rate among children compared to the rate among children with inadequate access to all three nutrition drivers. Controlling for child, parental, and household characteristics, as well as for the locality of residence, the prevalence of stunting among children having access to adequate health only decreases by 4.3 percentage points. Thus, with other conditions being equal, if bud- getary or other considerations allow for interventions covering deprived children by only one sector in Tanzania, this sector should be health. The estimates in table 6.3 also provide policy guidance on the next sector that could be scaled up in a targeted group that is already covered by the health sector. Controlling for child, parental, household characteristics, the probability of stunting associated with having simultaneous access to ade- quate health and adequate WASH (W_H), decreases by 9.8 percentage points relative to the reference group (that is, children with inadequate access to all three nutrition d ­ rivers). This implies that the marginal decline in stunting associated with adding access to adequate WASH services to the group of children who already have access to adequate health is 5.5 percentage points (−0.098 − [−0.043] = −0.055). The marginal decline in stunting associated with adding access to WASH is slightly higher than the 5.1 percentage point marginal decline in the probability of stunting associated with providing simultaneous access to adequate health and adequate food and care (−0.094 − [−0.043] = −0.051). In line with the results based on the pooled data for Sub-Saharan African countries, simultaneous access to all three nutrition drivers is associated with a significantly lower stunting rate. Controlling for child, parental, and house- hold characteristics, the probability of stunting associated with having simul- taneous access to all three nutrition drivers decreases by 16.9 percentage points relative to the reference group (that is, children with inadequate access to all three nutrition drivers). Bearing in mind some of the caveats already discussed, this large coefficient also implies that the marginal effect associated with increasing access from two to three drivers is larger than the marginal effect associated with increasing access from one to two drivers. For example, providing access to all three nutrition drivers to a group of children with access to adequate food/care and health (FC_H) leads to a marginal decline in stunting by 7.5 percentage points, an effect that is larger than the 5.1-­percentage-point marginal decline associated with adding access to ade- quate food/care services to the group of children who already have access to adequate health. The Tanzania case also underlines the importance of the multisectoral approach. To realize the largest reductions in stunting it is necessary to address all the underly- ing determinants of nutrition. Addressing only some will improve child nutrition, but the results suggest that one determinant is not a substitute for another. An Application at the Country Level: The Case of Tanzania   119 Monitoring Progress in Nutrition and Access to the Drivers of Nutrition In this section, an example is provided to monitor changes in the mean value of height-for-age Z-scores (HAZs) associated with increased access to the drivers of nutrition using the 2005 and 2016 waves of the Demographic and Health Survey in Tanzania.6 For children ages 0 to 23 months (the sample), the stunting rate dropped from 44.7 percent to 31.9 percent, whereas the mean value of HAZ increased by 0.48 standard deviations (s. d.) (from −1.79 s. d. in 2005 to −1.31 s. d. in 2016). The method employed is the standard Blinder–Oaxaca (BO) decomposition applied to two surveys over years (Jann 2008). The BO decomposition is best described in terms of a regression equation applied to any two groups (for example, different years or gender). Using a vari- ant of equation (5.3b), the HAZ of a child i in year t can be expressed as in equation (6.1): HAZit = at + β1t Pit + β2t Xit + ε it (6.1) where Pit denotes access to policy variables of interest and Xit denotes child, parental, and household characteristics, as in equation 5.3b. The policy variables of interest include the components of the underlying drivers of nutrition: whether the child had adequate food intake, child had immediate skin-to-skin contact at birth, child had age-appropriate breastfeeding, child’s mother had four or more prenatal care visits, child’s delivery was assisted by a trained pro- fessional, child had been to at least one postnatal care checkup, child received all age-appropriate vaccinations, child slept under a mosquito net, the house- hold had access to improved water, the household had access to an improved toilet, household had access to handwashing facilities, household disposed of the child’s stools properly, and the majority of households in the community (75 percent) used some sort of a sanitation facility. Following Jann (2008), the BO decomposition is based on the insight that the difference in the mean between the two years (or groups) can be expressed as the sum of the difference in the mean values in the different years of the explanatory variables, Pit and Xit, multiplied by a weighted average of their respective coeffi- cients in the different years, t and t−k, and the difference in the values over time of the coefficients of the same variables multiplied by a weighted average of the mean values in the different years of the explanatory variables:7 ( * lnHAZt − lnHAZt − k = Pt − Pt − k β1)+ ( β1t − β1t − k ) P * (6.2) ( ) * + Xt − Xt − k β 2 + ( β 2t − β 2t − k ) X * . 120   ALL HANDS ON DECK Figure 6.5 presents the contribution of explained factors and unexplained factors to the reduction of the mean value of HAZ over the years. In Tanzania, 24 percent of the reduction in stunting over the years is explained by the change in explained factors (that is, the changes in the mean values of the policy vari- ables Pi and the changes in the mean values of the child, parental, and household characteristics), with the remaining 76 percent attributed to unexplained factors (changes in the coefficients β1 and β2 over the years). Upon closer investigation, all (or 100 percent) of the change in explained factors is due to changes in the mean values of the policy variable Pi. Figure 6.6 focuses further on the contribution of individual components of the policy variable Pi, that is, the variables proxying the underlying drivers of nutrition. Receiving all the vaccinations (42.6 percent) has the highest contribu- tion to the increase in mean HAZ in Tanzania between 2005 and 2016, followed by sleeping under mosquito nets (31.9 percent) and then by deliveries assisted by trained professionals (25.5 percent). Figure 6.5  Fraction of Increase in Mean HAZ between 2005 and 2016 Attributable to Explained Factors, Tanzania (Children 0–23 Months) 120 100 80 Percentage 60 40 20 0 Explained Unexplained Source: Estimates based on the individual child data (0–23 months of age) from the 2005 and 2016 Tanzania Demographic and Health Surveys. Note: The regression estimated included the following explanatory variables: whether the child has adequate food intake, had immediate skin-to-skin contact at birth, and had age-appropriate breastfeeding; the child’s mother had four or more prenatal care visits; the child delivery was assisted by the trained professional; child has been to at least one postnatal care checkup; child received all the vaccinations; child sleeps under mosquito net; the household has access to improved water, improved toilet, handwashing facilities, and disposal of stools; the majority of households in the community (75 percent) use some sort of a sanitation facility; indicator variables for age (in months), gender, birth order, and multiple birth; mother's education in years; indicator for mother's marital status; mother's height; mother's age; number of children under five years of age; and number of household members. HAZ = height-for-age Z-score. An Application at the Country Level: The Case of Tanzania   121 Figure 6.6  Decomposing the Contribution of Access to the Drivers of Nutrition to the Increase in Mean HAZ between 2005 and 2016, Tanzania (Children 0–23 Months) 120 100 31.9 80 Percentage 60 42.6 40 20 25.5 0 Assisted birth Complete vaccines Sleeps under bednet Source: Estimates based on the individual child data (0–23 months of age) from the 2005 and 2016 Tanzania Demographic and Health Surveys. Note: The regression estimated included the following explanatory variables: whether the child has adequate food intake, had immediate skin-to-skin contact at birth, and had age-appropriate breastfeeding; the child’s mother had four or more prenatal care visits; the child delivery was assisted by the trained professional; child has been to at least one postnatal care checkup; child received all the vaccinations; child sleeps under mosquito net; the household has access to improved water, improved toilet, handwashing facilities, and disposal of stools; the majority of households in the community (75 percent) use some sort of a sanitation facility; indicator variables for age (in months), gender, birth order, and multiple birth; mother's education in years; indicator for mother's marital status; mother's height; mother's age; number of children under five years of age; and number of household members. HAZ = height-for-age Z-score. Only those components that were statistically significant are included. Headey and Hoddinott (2015) have also applied the methodology of the BO decomposition in various South Asian countries by imposing the restric- tion that the unexplained factors (the coefficients .b1. and .b2.) do not change over the years. In this case the general decomposition expression (3d) reduces to ( * lnHAZt − lnHAZt − k = Pt − Pt − k β1 ) ( * + Xt − Xt − k β 2 . (6.3e) ) Figure 6.7 is based on the above restriction and yields results that are consistent with those of figure 6.6 with some changes in the relative size of contribution. 122   ALL HANDS ON DECK Figure 6.7  Decomposing the Contribution of Access to the Drivers of Nutrition to the Increase in Mean HAZ (with Restrictions Imposed), Tanzania (Children 0–23 Months) 120 100 25.6 80 60 Percentage 61.5 40 20 14.9 0 –2.0 –20 Adequate food Assisted birth Complete vaccines Sleeps under bednet Source: Estimates based on the individual child data (0–23 months of age) from 2005 and 2016 Tanzania Demographic and Health Surveys. Note: The regression estimated included the following explanatory variables: whether the child has adequate food intake, had immediate skin-to-skin contact at birth, and had age-appropriate breastfeeding; the child’s mother had four or more prenatal care visits; the child delivery was assisted by the trained professional; child has been to at least one postnatal care checkup; child received all the vaccinations; child sleeps under mosquito net; the household has access to improved water, improved toilet, handwashing facilities, and disposal of stools; the majority of households in the community (75 percent) use some sort of a sanitation facility; indicator variables for age (in months), gender, birth order, and multiple birth; mother's education in years; indicator for mother's marital status; mother's height; mother's age; number of children under five years of age; and number of household members. HAZ = height-for-age Z-score. Standard errors are corrected for clustering at cluster level. Only those components that were statistically significant are included Receiving all the vaccinations has the highest contribution (61.5 percent) to the increase in mean HAZ in Tanzania between 2005 and 2016, followed by sleeping under mosquito nets (25.6 percent) and then by deliveries assisted by trained professionals (14.9 percent). An Application at the Country Level: The Case of Tanzania   123 Notes 1. In fact, regression equations (5.1) and (5.2) of chapter 5 have also been estimated using the strict definition for adequacy, but for the reasons outlined, none of the relevant coefficients turned out to be statistically significant. 2. A regional map of access to adequate food/care is provided at http://documents​ .worldbank.org/curated/en/467711529406437446/. It is not presented here in order to simplify the discussion. 3. Map 6.1 for Tanzania corresponds to figure 5.6 constructed using the pooled data from 33 Sub-Saharan African countries, including Tanzania. 4. Table 6.2 is obtained by estimating regression equation (5.1) in chapter 5 using data from the Tanzania Demographic and Health Survey 2016. Note that the only differ- ence between the estimates reported in Table 6.2 and the estimates for Tanzania in appendix C (column 29) is that the estimates in appendix C are obtained including binary variables identifying regions and rural/urban areas within a country (strata fixed effects) that typically tend to absorb a lot of the variation in cross-sectional samples. 5. Table 6.3 is obtained by estimating regression equation (5.2) in chapter 5 using data from the Tanzania Demographic and Health Survey 2016 and corresponds to figure 5.6, which is based on the pooled data from 33 Sub-Saharan African countries, including Tanzania. 6. Given the change in the reference population of children by the World Health Organization in 2006, we have ensured that the measures of HAZ over time are based on its 2006 reference group. Wβ1t + (I − W )β1t −k   * Wβ2t + (I − W )β2t −k  (1 − W )Pt + W Pt −k , β2 =  * * 7. Note: β1 = , P =   ( I − W ) Xt + W Xt − k  , where W is a matrix of relative weights given to the and X * =   coefficients of Group A and I is the identity matrix. This is equivalent to using the coefficients from a pooled model over both groups as the reference coefficients (Jann 2008). References Headey, D., and J. Hoddinott. 2015. “Understanding the Rapid Reduction of Undernutrition in Nepal, 2001–2011.” PLOS ONE 10 (12): e0145738. Jann, B. 2008. “A Stata Implementation of the Blinder-Oaxaca Decomposition.” Stata Journal 8 (4): 453–79. Chapter 7 Income Growth Is an Essential Element of a More Effective Multisectoral Approach Increased access to adequate levels of the underlying determinants of nutrition, analyzed extensively in the previous chapters, is a necessary but not sufficient condition for reducing undernutrition. According to the United Nations Children’s Fund (UNICEF) conceptual framework (UNICEF 1990), the “basic causes of nutrition,” which consist of the economic, political, environmental, social, and cultural contexts, also have a critical role in the determination of the nutritional status of children. Prominent among these factors are the income of the household, the education of the mother, and the extent of autonomous deci- sion making by the mother. Policies related to the basic determinants of nutrition are typically consid- ered part of the long-term development strategy of countries, although many programs related to these long-term objectives can also have positive impacts on child undernutrition and stunting in the short term. For example, income redistribution programs, such as conditional or unconditional cash transfer programs can increase the income controlled by mothers, which can translate to improved child nutrition in the short term. Similarly, programs in the agri- cultural sector that increase agricultural productivity or yields sustainably can have impacts on undernutrition in the short term through higher agricultural profits after harvest. A complete analysis of the distinct role these factors play in the nutritional status of children is beyond the scope of this report.1 Instead, this chapter focuses on two key moments of income distribution, the mean and the variance, and the relation of each with child undernutrition. Specifically, the next section focuses on the role of household assets/wealth as a proxy for house- hold income and explores some of the interactions of this proxy with other basic determinants, such as the mother’s education and autonomy, on child nutrition; the section following provides more evidence on the effect of weather-related shocks on child stunting and undernutrition. 125 126   ALL HANDS ON DECK Average Income Household income (or consumption expenditure) measures a household’s abil- ity to purchase and consume goods and services and thus is a strong correlate with the poverty status of the household. Higher income allows households to afford the inputs that are essential for the child’s health and nutrition, such as the quantity and quality of food needed for proper nutrition, the water and sanitation facilities that minimize exposure to pathogens, and the preventive and curative health services needed. Thus, income affects child nutrition by determining the quantity and quality of the underlying determinants of nutri- tion demanded by households for their children (see box 7.1 for some of the potential downfalls associated with increased agricultural productivity and incomes in Sub-Saharan Africa. A recent analysis of the potential role of income growth in reducing under- nutrition in Africa based on cross-country time-series data suggests that the association between economic growth and the reduction of stunting prevalence is lower in Africa, where a 1 percent increase in per capita gross national income (GNI) is associated with a 0.2 percent reduction in stunting prevalence. In con- trast, in other regions, a 1 percent increase in GNI is associated with a 0.6 percent decrease in the prevalence of stunting (World Bank 2017). However, it is ques- tionable whether such aggregate estimates provide an adequate representation of the relation between income and undernutrition and stunting at the house- hold level within countries in Sub-Saharan Africa. The literature abounds with estimates of the calorie-income or calorie per capita expenditure elasticity (for example, Subramanian and Deaton 1996). Such estimates, however, provide information on only one of the important inputs for child health (that is, caloric availability) and nutrition. In contrast, the literature on the relationship between income and nutritional outcomes such as height-for-age or weight-for-age, especially in Sub-Saharan African countries, is rather thin. Estimates of the income elasticity of stunting or height-for-age Z-scores (HAZs) within countries require household-level data on income or consumption expenditures together with child anthropometric measures (such as height and weight) as part of the same survey, and household surveys with these two key variables are scarce.2 Notable exceptions are Haddad et al. (2003) with estimates for Kenya, Mozambique, and South Africa; Krishna et al. (2015) and Christiaensen and Alderman (2004) for Ethiopia; and Alderman, Hoogeveen, and Rossi (2006) for Tanzania. In consideration of the absence of any systematic evidence of the relationship between income and undernutrition in Sub-Saharan Africa, this report carries out an analysis of the relationship between the percentile of the wealth index distribution within each of the 33 countries analyzed, and child HAZs. 3 Sahn and Stifel (2003), in a study of 10 countries, including Côte d’Ivoire, Ghana, Income Growth Is an Essential Element   127 BOX 7.1 Agricultural Chemicals and Child Nutrition One necessary condition for the advancement of food security in Sub-Saharan Africa is the increase in the productivity of existing cropland. Currently, many households practice low-input subsistence farming. In Asia, for example, the use of the Green Revolution technologies that were initiated in the 1960s included high-yielding-variety seeds, irrigation, pesticides, and nitrogenous fertilizers. These practices resulted in dramatically higher yields of staple crops (for example, wheat and rice), a reduction in the incidence of hunger, an increase in food security, and substantial improvements in the living standards of rural poor farmers and agricul- tural wage workers. As the efforts to increase productivity and agricultural crop yields in Africa intensify, it is important to have a balanced view of not only the benefits but also the potential costs involved. One key lesson learned from the Green Revolution in Asia is that the dramatic increases in the productivity of land took place at some cost to the environment in these countries. Specifically, the heavy use of fertilizers and pesticides for the purpose of increasing crop yields resulted in high levels of toxicity and the contamination of surface and ground water, especially in India (Brainerd and Menon 2014). Heeren, Tyler, and Mandeya (2003) provide evidence from South Africa, one of the major users of pesticides in the African continent, of a statistically significant associa- tion between birth defects and the exposure of the mothers to certain types of agricul- tural chemicals. One of the three types of exposure found to be significantly correlated with birth defects was the use of plastic containers previously used for the storage of agricultural chemicals but later used for storing of water for household use. Aside from birth defects, background exposure to pesticides in the environment is also known to have adverse effects on the cognitive, neurological, and behavioral development of children (Liu and Schelar 2012). The adverse effect of agricultural chemicals on the nutritional status of children as measured by height-for-age and weight-for-age Z-scores, is still questionable because of the scarcity of evidence. Brainerd and Menon (2014) supply the only empirical evi- dence available on this issue. They evaluate the impact on child health of water con- tamination resulting from applications of fertilizer agrichemicals early in the growing season in India. They find that children exposed to higher concentrations of fertilizer agrichemicals during the month of conception are more likely to have a higher mortal- ity rate, lower height for age, and lower weight for age. Also, this adverse effect was even more apparent among the children of uneducated poor women living in rural India. They report that for a 10 percent increase in the level of agrichemical toxins in water, weight-for-age Z-scores of children five years of age decline by 0.014 standard deviations. Although this is a modest effect, it is useful to know that exposure in the first month of life has such long-lasting negative effects on child health. Sources: Adapted from Brainerd and Menon 2014, and Heeren, Tyler, and Mandeya 2003. 128   ALL HANDS ON DECK Madagascar, and South Africa, provide supporting evidence in favor of the use of the wealth (or assets) index as a valid predictor of child nutrition outcomes.4 A more recent longitudinal study by Krishna et al. (2015) also finds that the baseline wealth index is significantly associated with higher HAZs and lower odds of stunting and that household wealth in early life influences growth falter- ing even beyond the 1,000-day window. The advantage of the approach adopted in this report is that such an analysis can be carried out for all countries where a Demographic and Health Survey (DHS) is available because the wealth index is included as part of the publicly available survey data or can be easily calculated. The disadvantage is that com- parable estimates are not available in the literature on undernutrition, thus making it rather difficult to relate the estimate of the coefficient of the wealth percentile, to the available income or per capita expenditures (PCEs) elasticity estimates in the literature. To unpack the role of wealth as a determinant of growth faltering among children, separate estimates are presented for younger (0–23 months) and older children (24–59 months). Income or wealth may have what appears to be a small positive impact on the HAZ of children in their first two years of life, but these small positive effects compound as the child ages, resulting in larger dif- ferences in HAZs that are more apparent later in the life of children. Given that the focus of the report is on stunted children which implies children with lower HAZs, the analysis is carried out for the five bottom deciles of the HAZ distribu- tion (that is, q = 0.10 to q = 0.50) using the unconditional quantile regression method proposed by Firpo, Fortin, and Lemieux (2009).5 Specifically, two dif- ferent specifications are estimated. The specification of equation (7.1a) is essen- tially a reduced form that can be considered as being derived from an underlying behavioral model of household utility maximization subject to constraints with child health and nutrition as the main determinant of household welfare (see appendix D): HAZic =  α + γ 1 pctileWIic + β1 Xic + µc + ε ic (7.1a) where pctileWIic denotes the percentile of the wealth index (treated as a continu- ous variable); Xic denotes a set of control variables summarizing parental, child and household characteristics; and μc denotes country fixed-effect (or country- specific binary variables). The Xic control variables include indicator variables for age, gender, multiple births, and birth order; the age of the mother (in years); the marital status of the mother; mother’s education level (in years); mother’s height (in cm); the number of household members; and the number of children under five years of age. Equation (7.1b) is a variant of equation (7.1a) with the addition of the avail- able components of the composite measures of food and care; water, sanitation, and hygiene (WASH); and health (denoted by Pic). For children between 0 and Income Growth Is an Essential Element   129 23 months, these components include whether a child had a minimum accept- able diet, child had immediate skin-to-skin contact at birth, child was age- appropriately breastfed, child’s mother had four or more prenatal care visits, child delivery was assisted by the trained professional, child had been to at least one postnatal checkup, child received all the age-appropriate vaccinations, child slept under a mosquito net, the household had access to improved water, the household had access to improved toilet, household had access to handwashing facilities, household disposed of stools properly, and most households in the community (75 percent) used some sort of a sanitation facility. For children between 24 and 59 months, the components of the underlying drivers of nutri- tion do not include the following variables that are not collected for older chil- dren: whether the child had a minimum acceptable diet, child had immediate skin-to-skin contact at birth, child had age-appropriate breastfeeding, child’s mother had four or more prenatal care visits, child delivery was assisted by the trained professional, and child has been to at least one postnatal care visit. HAZic =  α + γ 1 pctileWIic + β1 Xic + β2 Pic + µc + ε ic (7.1b) Specification (7.1b) offers the advantage of testing whether including the components of the underlying drivers of nutrition has a significant impact on the size of the coefficient of the pctileWI variable, that is, γ1 in equation (7.1a). According to the assumption that is implicit in the UNICEF model, higher household income affects nutrition primarily, if not exclusively, by increasing the demand for adequate levels of the components of the underlying drivers of nutrition. Under the maintained assumption that the available components of the composite measures of food and care, WASH, and health (Pic) included in equation (7.1b) provide sufficient statistics for “access to the underlying drivers of nutrition,” the direct effect of income (proxied here by pctileWI) in equation (7.1b) should be considerably diminished (or even become insignificantly dif- ferent from 0, that is, γ1 = 0) relative to the one obtained from specification (7.1a). Evidence that the estimate of the coefficient γ1 from equation (7.1b) does not differ substantially from that in equation (7.1a) would suggest that income has a significant direct effect on child nutrition (HAZ) aside from that captured by the components of food and care, WASH, and health, denoted by Pic in equa- tion (7.1b).6 Figure 7.1 presents the effect of increased wealth on HAZs for different quan- tiles of HAZs for children 0–23 months and 24–59 months of age for the speci- fications in equations (7.1a) and (7.1b). Two important findings emerge. First, irrespective of the quantile of HAZ or the specification estimated, increased wealth has a significant effect on the HAZ of younger children and older chil- dren. For example, an increase in the ranking of the wealth index value of the household by 10 percentage points is associated with an increase in HAZ by 0.03 standard deviations (s. d.) for younger children whose HAZs are at the 130   ALL HANDS ON DECK Figure 7.1  The Relationship between Height-for-Age Z-Scores and Wealth Percentiles by Wealth Deciles among Older and Younger Children a. Excluding components of the underlying b. including components of the underlying drivers of nutrition drivers of nutritiona 0.009 0.009 0.008 0.008 Standard deviations Standard deviations 0.007 0.007 0.006 0.006 0.005 0.005 0.004 0.004 0.003 0.003 0.002 0.002 0.001 0.001 0 0 0–23 months 24–59 months 0–23 months 24–59 months bottom 10% bottom 20% bottom 30% bottom 40% bottom 50% Source: Estimates based on data for children ages 0–23 and 24–59 months from 33 country Demographic and Health Surveys from Sub-Saharan Africa. See table 2.1 in chapter 2 for list of countries and years. Note: Unconditional quantile regression estimates including country fixed effects. Detailed estimates available upon request. Both sets of regressions include child, parental, and household characteristics that consist of the following variables: indicator variables for age (in months), gender, birth order, and multiple births; mother’s education in years; mother’s marital status; mother’s height; mother’s age; wealth quintile; number of children under five years of age; number of household members; and whether household is in a rural or urban area. HAZ = height-for-age Z-score. a. The components of the underlying drivers of nutrition include whether a child had immediate skin-to-skin contact at birth, child had age-appropriate breastfeeding, child’s mother had four or more prenatal care visits, child delivery was assisted by a trained professional, child has been to at least one postnatal care visit, child received all the vaccinations, child sleeps under a mosquito net, household has access to improved water, household has access to improved toilet, household has access to handwashing facilities, household properly disposes of stools, and most households in the community (75 percent) use some sort of a sanitation facility. For children 24–59 months of age, the components of the underlying drivers of nutrition do not include the following variables (not collected for older children): whether child had immediate skin-to-skin contact at birth, child had age appropriate breastfeeding, child’s mother had four or more prenatal care visits, child delivery was assisted by a trained professional, and child has been to at least one postnatal care visit. 50th percent quantile of the HAZ distribution.7 For older children (24–50 months of age) who are at the 50th percent quantile of the HAZ distribution, the same 10-percentage-point increase in the ranking of the wealth index of the house- hold is associated with an increase in the HAZ by 0.06 s. d. (see figure 7.1, panel a).8 The higher coefficient for older children confirms that the true effect of wealth on HAZ is discernible only at later points in the life of children. The HAZ of older children is the outcome of chronic undernutrition experienced earlier in life, and the average HAZ of older children differs significantly between older children from the top quintile (T20) and bottom quintile (B20) of the wealth distribution. As the growth faltering curves reveal in panel e of figure 3.3, there is a much larger gap between the growth faltering curve of children at the B20 and the T20 of the distribution of wealth among older children (older than 24 months) than among younger children. This difference implies that, all other things being equal, any given increase in wealth is by Income Growth Is an Essential Element   131 default associated with a larger change in HAZs among older children than younger children. Second, a comparison of the coefficients γ1 from equation (7.1a) (panel a in figure 7.1) does not differ substantially from those obtained from equation (7.1b) (panel b in figure 7.1). The coefficient at the bottom 10 percent of the HAZ distribution for older children (24–59 months) remains unchanged at 0.008 s. d. (compare panels a and b in figure 7.1). Similarly, the coefficients for the 40 percent quantile for the same age group remain unaffected at the value of 0.007 s d. Irrespective of the quantiles or the age groups compared, coeffi- cients are either the same or slightly lower under specification (7.1b) (see panel b in figure 7.1), but only for the lowest quintile in the younger group does the γ1 become statistically insignificant. These results suggest that income has a significant direct effect on child nutrition (HAZ) aside from that captured by the components of food and care, WASH, and health (Pic). Figure 7.2 presents corresponding estimates of the effect of higher wealth on HAZs by quantiles of HAZs based on equation (7.1a) for children 0–23 months and 24–59 months of age, in urban and rural areas, for children whose mothers have more or less education, and for children whose mothers have more or less autonomy in decision making at home. It was also confirmed that estimation of equation (7.1b) for the same groups resulted in minor differences in the esti- mate of γ1 as in figure 7.2, based on the full sample. More educated mothers are those with more than seven years of education, whereas mothers with more autonomy in decision making are defined using the top half (> 50th percentile) of the values of the first principal component derived from the following vari- ables: whether the married mother usually makes the decisions on how to spend her earnings, on her health care, on large household purchases, on visits to family or relatives, and on what to do with the money the husband earns, and whether she is employed all year or in seasonal work. As before, the association between increases in wealth ranking and increases in child HAZ are stronger among older children irrespective of location and level of education and autonomy of the mother. Among younger children, there are some minor differences in how increases in wealth are associated with increases in HAZ, between urban and rural areas (figure 7.2, panel a) or between more and less educated mothers (figure 7.2, panel c) or more or less autono- mous mothers (figure 7.2, panel e). Among older children, the larger coeffi- cients in urban areas relative to rural areas (figure 7.2, panel b), for more educated mothers relative to less educated mothers (figure 7.2, panel d), and for children whose mothers have more autonomy in decision making than those with less autonomy (figure 7.2, panel f), imply that the smaller differences in the basic determinants of nutrition early in the life of children end up compound- ing with others and thus translating to greater differences in the HAZs as ­children age. 132   ALL HANDS ON DECK Figure 7.2  Differences in the Relationship between Height-for-Age Z-Scores and Wealth Percentiles by Wealth Deciles between Some Key Basic Determinants of Nutrition a. Increased HAZ in younger children b. Increased HAZ in older children in urban and rural areas in urban and rural areas 0.014 0.014 0.012 0.012 Standard deviations Standard deviations 0.010 0.010 0.008 0.008 0.006 0.006 0.004 0.004 0.002 0.002 0 0 Urban Rural Urban Rural c. Increased HAZ in younger children d. Increased HAZ in older children by mother's education by mother's education 0.010 0.010 Standard deviations Standard deviations 0.008 0.008 0.006 0.006 0.004 0.004 0.002 0.002 0 0 Mother less Mother more Mother less Mother more educated educated educated educated e. Increased HAZ in younger children f. Increased HAZ in older by mother's autonomy by mother's autonomy 0.010 0.010 Standard deviations Standard deviations 0.008 0.008 0.006 0.006 0.004 0.004 0.002 0.002 0 0 No autonomy Autonomy No autonomy Autonomy Bottom 10% Bottom 20% Bottom 30% Bottom 40% Bottom 50% Source: Estimates based on data for children ages 0–23 and 24–59 months from 33 country Demographic and Health Surveys from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Notes: Unconditional quantile regression estimates including country fixed effects. Detailed estimates available upon request. The regressions include child, parental, and household characteristics that consist of the following variables: indicator variables for age (in months), gender, birth order, and multiple births; mother’s education in years; indicator for mother’s marital status; mother’s height; mother’s age; wealth quintile; number of children under five years of age; number of household members; and whether household is in a rural or urban area. HAZ = height-for-age Z-score. Income Growth Is an Essential Element   133 In the same spirit, figure 7.3 presents the association between increases in wealth ranking and increases in child HAZ for younger and older children in two groups of countries: (a) where efforts to decrease child undernutrition are starting as part of the Investing in Early Years initiative (first wave versus other countries) or (b) where there are ongoing concerns regarding the potential of cash transfers relative to direct food transfers in kind in reducing food insecu- rity and child undernutrition (that is, countries in the Sahel region). In line with the patterns observed before, increases in the wealth ranking percentile in the Sahel countries and in the first wave countries have a higher effect on child HAZ for older children than for younger children. Figure 7.3  Differences in the Relationship between Height-for-Age Z-Scores and Wealth Percentiles by Wealth Deciles between Groups of Countries a. Increased HAZ in younger children b. Increased HAZ in older children by first wave status by first wave status 0.010 0.010 0.008 0.008 Standard deviations Standard deviations 0.006 0.006 0.004 0.004 0.002 0.002 0 0 Not in the1st wave 1st wave Not in the 1st wave 1st wave c. Increased HAZ in younger children d. Increased HAZ in older children by presence in Sahel region by presence in Sahel region 0.008 0.008 Standard deviations Standard deviations 0.006 0.006 0.004 0.004 0.002 0.002 0 0 Not in the Sahel Sahel region Not in the Sahel Sahel region region region Bottom 10% Bottom 20% Bottom 30% Bottom 40% Bottom 50% Source: Estimates based on data for children ages 0–23 and 24–59 months from 33 country Demographic and Health Surveys from Sub-Saharan Africa. See table 2.1 in chapter 2 for list of countries and years. Notes: Unconditional quantile regression estimates including country fixed effects. Detailed estimates available upon request. The regressions include child, parental, and household characteristics that contain the following variables: indicator variables for age (in months), gender, birth order, and multiple births; mother’s education in years; mother’s marital status; mother’s height; mother’s age; wealth quintile; number of children under five years of age; number of household members; and whether household is in a rural or urban area. HAZ = height- for-age Z-score. 134   ALL HANDS ON DECK To complete the picture, figure 7.4 presents the country-specific estimates of equation (7.1a) at the median value of HAZs for younger and older children. The  country-specific estimates suggest a significant pattern of correlation between wealth and child HAZs for both older and younger children. In Tanzania, for example, the stated estimates suggest that a 10-percentage-point Figure 7.4  Country-Specific Estimates of the Relationship between Median Height-for-Age Z-Scores and Household Wealth Index Percentile a. Younger children (0–23 months) b. Older Children (24–59 months) Angola Angola Benin Benin Burkina Faso Burkina Faso Burundi Burundi Cameroon Cameroon Chad Chad Comoros Comoros Congo, Dem. Rep. Congo, Dem. Rep. Congo, Rep. Congo, Rep. Côte d’Ivoire Côte d’Ivoire Ethiopia Ethiopia Gabon Gabon Gambia, The Gambia, The Ghana Ghana Guinea Guinea Kenya Kenya Lesotho Lesotho Liberia Liberia Madagascar Madagascar Mali Mali Mozambique Mozambique Namibia Namibia Niger Niger Nigeria Nigeria Rwanda Rwanda Senegal Senegal Sierra Leone Sierra Leone Tanzania Tanzania Togo Togo Uganda Uganda Zambia Zambia Zimbabwe Zimbabwe 0. 0 0. 0 0. 2 0. 4 0. 6 0. 8 0. 0 0. 2 4 0. 2 0. 4 0. 6 0. 8 0. 0 0. 2 4 00 00 00 00 01 01 01 00 00 00 00 01 01 01 Standard deviations Standard deviations Source: Estimates based on data for children ages 0–23 and 24–59 months from 31 country Demographic and Health Surveys from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Notes: Unconditional quantile regression estimates including country fixed effects. Detailed estimates available upon request. The regressions include child, parental, and household characteristics and contain the following variables: indicator variables for age (in months), gender, birth order, and multiple births; mother’s education in years; mother’s marital status; mother’s height; mother’s age; wealth quintile; number of children under five years of age; number of household members; and whether household is in a rural or urban area. No information on mother’s height was supplied for Angola and Senegal, and thus those countries are not in the analyses. Income Growth Is an Essential Element   135 increase in the wealth index, which is equivalent to a 17 percent increase of mean consumption, leads to an increase in HAZ of 0.04 (for children 0–23 months) and 0.06 s. d. (for children 24–59 months). The new estimates are comparable to existing estimates for Tanzania. Alderman, Hoogeveen, and Rossi (2006), for example, depending on the specification used, estimate that a doubling (or a 100 percent increase) of household PCE would result in an increase in HAZ of 0.1–0.2 s. d. among children under 60 months of age. Given that a 100 percent increase in PCE is equivalent to an increase of 58.8 percentage points in the value of the wealth index, the estimated increase in HAZ based on these new estimates is 0.35 s. d. for older children and 0.23 s. d. for younger children, only slightly higher than those estimated by Alderman, Hoogeveen, and Rossi (2006). Variability of Income In addition to the critical role of the average level of household income, the variability of income over time has a distinct effect on child nutrition. Variability in income, one of the main causes of food insecurity, is caused by a confluence of factors composed primarily of weather-related shocks such as irregularities in the amount and timing of rainfall in relation to the crop cycle, the seasonal nature of agricultural production (lean versus peak season), and the constraints in the ability of households to transfer resources across time (that is, the absence of credit and insurance markets). In rural economies based largely on rain-fed agriculture, household income is highly dependent on rainfall realizations. Credit constraints limit households’ ability to smooth consumption over time, rendering health more vulnerable to economic shocks (Behrman and Deolalikar 1987). Last, insofar as households are spatially dispersed and transport infra- structure is weak, markets in food staples may not be well integrated. Localized rainfall shocks may, consequently, influence food prices. When seasonal rains are plentiful, yields will be high, food supplies robust, and prices low. Such gen- eral equilibrium effects reinforce the positive association between rainfall and household purchasing power or real income. Given the seasonal nature of agri- cultural production and limited borrowing opportunities, the effect from income to consumption, and thereby to child nutrition, is likely to take place with some delay; higher rainfall during the current crop season can increase consumption only after harvest. Another important channel through which excess rainfall can influence child nutritional status is through the alteration of the disease environment net of any parental responses to child illness. Flooding for example, may hinder access to health facilities and damage the existing water and sanitation facilities, thereby increasing its association with higher contemporaneous incidence of 136   ALL HANDS ON DECK diarrheal disease, and even typhoid and cholera. Standing water also indirectly leads to an increase in vector-borne diseases, such as malaria and dengue, through the expansion in the number and range of vector habitats. Such ill- nesses lower the capacity to take in and retain essential nutrients from food. Insofar as parents cannot entirely prevent or perfectly ameliorate these effects of child illness, excess rainfall shocks will have a negative impact on nutritional status through the disease channel. Table 7.1 provides suggestive evidence attesting to the impact of rainfall vari- ability experienced during the 1,000-day window on the prevalence of stunting among children under 24 months of age in the rural areas of Sub-Saharan Africa.9 Grid-level data of mean and median absolute deviation (MAD) of rainfall based on historic rainfall data between 1960 and 2014 for all of Sub-Saharan Africa were merged to the clusters of the DHS of 21 countries.10 The coefficient of variation at cluster level was calculated as the ratio of MAD and mean rainfall. The estimates reveal that increases in the variability of rainfall as proxied by the coefficient of variation of rainfall at the cluster level are associated with an increase in the prob- ability of stunting by 33.4–70.2 percentage points depending on the specification. Taken at face value these large coefficients suggest that significant declines in stunting could be accomplished by policies that decrease the local variability of rainfall. However, these estimates are not particularly useful for country-specific Table 7.1  The Relationship between Rainfall Variability (Coefficient of Variation) on the Prevalence of Stunting among Children 0–23 Months of Age in Rural Areas of Sub-Saharan Africa Variables Specification 1 Specification 2 Specification 3 Coefficient of variation of 0.507*** 0.334*** 0.702*** rainfall at the DHS cluster level (0.110) (0.113) (0.175) Year of interview dummies Yes Yes Yes Child, parental, and household characteristics No Yes Yes Drivers of nutrition No No Yes Observations 44,456 37,539 23,031 R-squared 0.015 0.137 0.142 Source: Grid-level rainfall mean and median absolute deviation come from Harvest Choice, https://harvestchoice​ .org/data/pre_mean and https://harvestchoice.org/data/pre_mad. Note: Ordinary least squares estimates including country fixed effects. Detailed estimates available upon request. Child, parental, and household characteristics consists of the following variables: indicator variables for age (in months), gender, and birth order; preceding birth interval; mother’s education in years; mother’s marital status; mother’s height; number of children under five years of age; and number of household members. “Drivers of nutrition” consist of the following variables: whether a child had immediate skin-to-skin contact at birth, child had exclusive breastfeeding, child’s mother had four or more prenatal care visits, child delivery was assisted by a trained professional, child has been to at least one postnatal care checkup, child received all the vaccinations, child sleeps under mosquito net, household has access to improved water, household has access to improved toilet, household has access to handwashing facilities, household properly disposes of stools, and the fraction of households in the community practicing open defecation (have no sanitation facility). DHS = Demographic and Health Survey. Robust standard errors in parentheses ***p < 0.01, **p < 0.05, *p < 0.1. Income Growth Is an Essential Element   137 policy design, because they become statistically insignificant when country-strata fixed effects are used in place of country fixed effects.11 To shed more light on the effect of rainfall variability on child chronic under- nutrition, table 7.2 presents estimates from selected countries of the impact of rainfall shortfalls during the growing season in the DHS cluster where the child lived. As discussed, the seasonal nature of agricultural production and limited borrowing opportunities imply that the effect from income on consumption, and therefore child nutrition, is likely to take place with some delay; lower (higher) rainfall during the current crop season can decrease (increase) con- sumption only after harvest. To this end, observations on children and their anthropometric measures from the DHS are merged at the cluster level with grid-level data on the beginning and end date of the growing season12 and data on the value of the Standardized Precipitation Index (SPI) during the last com- pleted growing season before the date of interview of the household and its members. The SPI is a standardized measure analogous to a Z-score that can be interpreted in terms of standard deviation units from the normal mean precipitation.13 The estimates in table 7.2 confirm that significant declines in stunting could be accomplished by policies and programs that decrease the vulnerability of house- hold income from weather-related shocks. A shortfall in rainfall by 1 s. d. from the normal mean precipitation in the cluster during the most recently completed growing season is associated with a 3.2-percentage-point increase in the Table 7.2  The Relationship between Rainfall Shortfalls during the Growing Season and the Prevalence of Stunting among Children under 60 Months of Age in Selected Countries in Sub-Saharan Africa Variables Benin Congo, Dem. Rep. Mozambique Nigeria Rwanda Rainfall shortfall in s. d. units 0.0320* 0.0331* 0.0155 0.0242** 0.0221* during the last growing season (0.0167) (0.0178) (0.0107) (0.00945) (0.0124) Year of interview dummies Yes Yes Yes Yes Yes Other controls Yes Yes Yes Yes Yes Observations 4,142 1,845 5,021 13,073 1,970 R-squared 0.040 0.056 0.042 0.090 0.077 Source: Grid-level information on growing period come from Harvest Choice, http://harvestchoice.org/labs​ /measuring-growing-seasons. Stunting and demographic information from Demographic and Health Surveys from Benin (2011–12); Congo, Dem. Rep. (2007, 2013–14); Mozambique (2014); Nigeria (2008, 2013); and Rwanda (2014–15). Note: The coefficients reported in the table are obtained using (−1) × SPI (Standard Precipitation Index) as a right-hand-side variable in the regression. SPI for the last growing season is the SPI for the Demographic and Health Survey cluster where the child resides. The SPI dataset is accessible at http://stream.princeton.edu/AWCM​ /WEBPAGE/interface.php?locale=en. Other controls include indicator variables for age (in months), gender, and birth order; preceding birth interval; mother’s education in years; mother’s marital status; mother’s height; number of children under five years of age; and number of household members. Robust standard errors are in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1. 138   ALL HANDS ON DECK probability of being stunted across all children 0–60 months of age in Benin, a 3.31-percentage-point increase in the Democratic Republic of Congo, no signifi- cant increase in the probability of stunting in Mozambique, and 2.42-percentage- point and 2.21-percentage-point increases in Nigeria and Rwanda, respectively. In recent years, there has been an increasing policy emphasis on social pro- tection programs that provide the basis for both increasing the level and decreasing the variability of incomes (Del Ninno, Coll-Black, and Fallavier 2016; Del Ninno and Mills 2015). These programs not only serve as useful instruments to respond ex post to the incidence of droughts, floods, and other natural disasters but also help households build their resilience before shocks hit. Cash transfer programs, for example, redistribute income to the poorer seg- ments of the population and allow households to invest in human capital and child nutrition, build assets, and diversify their livelihood strategies. Public works programs help households and communities reduce their vulnerability to shocks while improving community infrastructure and the opportunities for new and improved livelihoods. In parallel or in combination, increased access to insurance products and credit markets can ensure better and more efficient use of resources by eliminat- ing the incentive to adopt low-risk/low-return crops and production methods. Increased access can also alleviate intertemporal distortions on human and pro- ductive capital investment, such as cutting down on food consumption and health services or withdrawing children from school. All in all, the arguments in this chapter highlight the importance of pursuing a balanced strategy toward the reduction of undernutrition. Increased access to adequate levels of the underlying drivers of nutrition coordinated across differ- ent sectors (analyzed in chapter 5) should not be considered in isolation from programs increasing income and minimizing income variability. Instead, pro- grams and interventions aimed at increasing the level and stability of income among populations where stunting is prevalent should be considered as indis- pensable and essential (sine qua non) ingredients of multisectoral efforts to reducing undernutrition in Sub-Saharan Africa. Notes 1. Carlson, Kordas, and Murray-Kolb (2015) provide an up-to-date review of the litera- ture on the links between women’s autonomy and child nutrition. Alderman and Headey (2017) and Desai and Alva (1998) carry out some of the most comprehen- sive evaluations of the causal relationship between mother’s education and children’s height-for-age. 2. Moreover, if such data are available, then the sample of children covered is relatively small. For example, in the recent Living Standards Measurement Surveys-Integrated Surveys on Agriculture (LSMS-ISA) covering nine countries in Sub-Saharan Africa, Income Growth Is an Essential Element   139 the emphasis on the panel nature of the sample tends to reduce the sample coverage of children 0 to 5 years of age for which anthropometric measurements are taken. 3. An alternative approach is to impute per capita expenditures (PCEs) from a house- hold survey such as the LSMS-ISA to the Demographic and Health Survey (DHS) using variables that are available in both surveys and then use the imputed PCE instead of the percentile of the household wealth index used here. Given that it can be done only in the countries with contemporaneous LSMS and DHS, this alterna- tive approach was not pursued. 4. In fact, they report that “for most of the samples included in their paper, the asset index performed as well, if not better than reported expenditures in predicting chil- dren’s height-for-age Z-scores” (Sahn and Stifel 2003). It should be noted however, that the systematic review of Howe et al. (2009) suggests that the wealth index is a poor proxy for consumption expenditure. 5. In the pooled sample of children younger than 24 months, the 30 percent and 40 percent quantiles of the HAZ distribution bound the value of HAZ used as the threshold for the definition of stunting (HAZ < –2). 6. It is also possible that the maintained and untested assumption is incorrect, meaning that the variables included in Pic do a very poor job of capturing “access to the under- lying drivers of nutrition.” However, the analysis in chapter 5 suggests the contrary. 7. The estimates in the figures reflect the effect of a 1-percentage-point increase in the ranking of the wealth index value of the household on the HAZ. 8. A 10-percentage-point increase in the ranking of the wealth index value of the household translates to a 12240.38 naira (US$76.5) increase in per capita expendi- ture in Nigeria, a 640.1 birr per year per adult equivalent increase (US$27.2) in Ethiopia, a 708,626.2 shilling per year (about US$315.6) increase in Tanzania, and a 107,643.4 shilling per month (US$29.72) increase in Uganda. 9. Additional evidence on the long-term impacts of rainfall shocks experienced during infancy in Sub-Saharan Africa is contained in the recent report by Damania et al. (2017). 10. Prior to release of the geographic dataset corresponding to the respective Demographic and Health Survey of each country, the cluster Global Positioning System coordinates for the center of the populated areas of the clusters were geo- graphically displaced. Coordinates of urban clusters were displaced up to a maximum distance of 2 km. In rural areas, the displacement distance was up to 5 km, with a further, randomly selected 1 percent of rural clusters displaced up to 10 km. 11. This is primarily because there is very small variability in the coefficient of variation of rainfall within rural areas of different regions in any given country. 12. General information and some data can be found at Harvest Choice, http://­ harvestchoice.org/labs/measuring-growing-seasons. 13. The SPI dataset is available for a grid with 0.25° resolution, which is equal to a square of 25 km by 25 km in the equator (with this square getting smaller away from the equator) with geo-coordinates for each point in the grid accessible at http://stream​ .princeton.edu/AWCM/WEBPAGE/interface.php?locale=en from the African Flood and Drought Monitor online platform developed by Princeton University and the United Nations Educational, Scientific, and Cultural Organization’s International Hydrological Program. 140   ALL HANDS ON DECK References Alderman, H., and D. Headey. 2017. “How Important Is Parental Education for Child Nutrition?” World Development 94: 448–64 Alderman, H., H. Hoogeveen, and M. Rossi. 2006. “Reducing Child Malnutrition in Tanzania: Combined Effects of Income Growth and Program Interventions.” Economics and Human Biology 4 (1): 1–23. Behrman, J., and A. Deolalikar. 1987. “Will Developing Country Nutrition Improve with Income? A Case Study for Rural South India.” Journal of Political Economy 95 (3): 492–507. Brainerd, E., and N. Menon. 2014. “Seasonal Effects of Water Quality: The Hidden Costs of the Green Revolution to Infant and Child Health in India.” Journal of Development Economics 107 (March): 49–64. Carlson, G., K. Kordas, and L. Murray-Kolb. 2015. “Associations between Women’s Autonomy and Child Nutritional Status: A Review of the Literature.” Maternal and Child Nutrition 11(4): 452–82. Christiaensen, L., and H. Alderman. 2004. “Child Undernutrition in Ethiopia: Can Maternal Knowledge Augment the Role of Income?” Economic Development and Cultural Change 52 (2): 287–312. Damania, R., S. Desbureaux, M. Hyland, A. Islam, S. Moore, A. S. Rodella, J. Russ, and E. Zaveri. 2017. Uncharted Waters: The New Economics of Water Scarcity and Variability. Washington, DC: World Bank. https://openknowledge.worldbank.org​ /handle/10986/28096. Del Ninno, C., S. Coll-Black, and P. Fallavier. 2016. Social Protection Programs for Africa’s Drylands. Washington, DC: World Bank Del Ninno, C., and B. Mills, eds. 2015. Safety Nets in Africa: Effective Mechanisms to Reach the Poor and Most Vulnerable. Washington, DC: World Bank. Desai, S., and S. Alva. 1998. “Maternal Education and Child Health: Is There a Strong Relationship?” Demography 35 (1): 71–81. Firpo, S., N. Fortin, and T. Lemieux. 2009. “Unconditional Quantile Regressions.” Econometrica 77 (3): 953–73. Haddad, L., H. Alderman, S. Appleton, L. Song, and Y. Yohannes. 2003. “Reducing Child Undernutrition: How Far Does Income Growth Take Us?” World Bank Economic Review 17 (1): 107–31. https://openknowledge.worldbank.org/bitstream/handle/109 86/17172/774030JRN020030g0Child0Malnutrition.pdf?sequence=1&isAllowed=y. Heeren, G. A., J. Tyler, and A. Mandeya. 2003. “Agricultural Chemical Exposures and Birth Defects in the Eastern Cape Province, South Africa: A Case–Control Study.” Environmental Health 2 (1): 11. Howe, L. D., J. R. Hargreaves, S. Gabrysch, and S. R. A. Huttly. 2009. “Is the Wealth Index a Proxy for Consumption Expenditure? A Systematic Review.” Journal of Epidemiology and Community Health 63 (11): 871–77. Krishna, A., J. Oh, J. Lee, H.-Y. Lee, J. M. Perkins, J. Heo, Y. Sun Ro, and S. V. Subramanian. 2015. “Short-Term and Long-Term Associations between Household Wealth and Physical Growth: A Cross-Comparative Analysis of Children from Four Low- and Middle-Income Countries.” Global Health Action 8 (1). doi:10.3402/gha.v8.26523. Income Growth Is an Essential Element   141 Liu, J., and E. Schelar. 2012. “Pesticide Exposure and Child Neurodevelopment: Summary and Implications.” Workplace Health & Safety 60 (5): 235–43. 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. Subramanian, S., and A. Deaton. 1996. “The Demand for Food and Calories.” Journal of Political Economy 104 (1): 133–62. UNICEF (United Nations International Children’s Emergency Fund). 1990. Strategy for Improved Nutrition of Women and Children in Developing Countries: A UNICEF Policy Review. New York: UNICEF. World Bank. 2017. “Stunting Reduction in Sub-Saharan Africa.” HNP Discussion Paper, World Bank, Washington, DC. Chapter 8 How Can a Multisectoral Strategy to Reducing Stunting Produce the Desired Results? The analysis in this report offers new insights on how data can be used to inform allocation decisions of different sectors that can strengthen multisectoral efforts aiming to reduce stunting. In practice, however, there is little assurance that involving multiple sectors in the effort to reduce undernutrition will produce the desired outcomes. The history of multisectoral initiatives on nutrition contains many non- performing projects (IEG 2009; World Bank 2014). Multisectoral nutrition planning was favored by the international development community in the 1970s, but it quickly became apparent that this type of planning was overly ambitious and too dependent on other sectors that are reluctant to be coor- dinated (Levinson and Balarajan 2013). A more recent example is the case of multisectoral AIDS projects in Africa, where multiple sectors—such as health; education; and water, sanitation, and hygiene (WASH)—were involved. When budgets, governance, and accountability structures are driven mainly by sector-specific considerations within country ministries, as well as within international organizations, “multisectoral” initiatives tend to reduce clarity and specificity on the role and responsibility of each sector (IEG 2009). In recent years, however, a few countries, such as Brazil, Peru, and Senegal, have been quite successful in reducing undernutrition significantly through multisectoral approaches tailored to their needs and circumstances. This section synthesizes the main findings of this report with the lessons offered from these successful country cases for the purpose of identifying the key ingredients of a multisectoral strategy in the reduction of undernu- trition that maximizes the potential to produce the desired outcomes in practice. 143 144   ALL HANDS ON DECK The Joint Targeting of Interventions by Different Sectors One fundamental ingredient of a successful strategy is the scale-up of interven- tions by agriculture (food security), health care, and WASH that are jointly targeted to geographic areas (or populations within these areas) with high prevalence of stunting.1 The primary purpose of these jointly targeted opera- tions is to increase access to the underlying determinants of nutrition as envi- sioned by the United Nations Children’s Fund conceptual framework underpinning this report (UNICEF 1990). For sector-specific investments to contribute to the reduction of stunting and to speed up progress toward the Sustainable Development Goals (UN 2015), the prevalence of stunting needs to be considered as an additional criterion when prioritizing and allocating scarce resources at the country and subnational levels. The very high propor- tions of children with inadequate access to all three drivers of nutrition figure O.4 in the overview) suggest that the strategy of joint targeting would (­ apply to almost all countries in Sub-Saharan Africa. Planning a successful jointly targeted program requires taking stock of the sectors operating in the target areas (or target population groups) and redirect- ing operations of the missing sectors to the target areas. The absence of some key sector from the target areas, such as agriculture, WASH, or social protec- tion, may also act as a deterrent for the sector or sectors already operating in the target areas to be the “first mover” in terms of adopting nutrition-sensitive interventions. The descriptive statistics on access to adequate WASH in the 33 countries covered by the report (chapter 3), in combination with the recently completed WASH Poverty Diagnostic of the Water and Poverty and Equity Global Practices (World Bank 2017), confirm the very limited coverage by the WASH sector in the rural areas where stunting is prevalent. The same also applies to the fraction of the population covered by social protection schemes (Del Ninno, Coll-Black, and Fallavier 2016). For example, the Productive Safety Net Program of Ethiopia, one of the largest in Sub-Saharan Africa, covers only 10 percent of the population (World Bank 2012). This report provides country authorities with a holistic picture of the gaps in access to the drivers of nutrition within countries; this picture is critical for the formulation of a more informed, evidence-based, and balanced multisectoral strategy against undernutrition in their countries. Much work remains to be done in terms of coordinating the targeting of service delivery in areas where stunting is prevalent if all key sectors are to contribute jointly in the reduction of stunting. Despite a broad correlation between monetary poverty and chil- dren’s health at the country level, targeting stunted children is not as simple as distinguishing between urban and rural areas or using poverty maps identifying the poor and nonpoor regions. Not all children in poor households or in rural How Can a Multisectoral Strategy Succeed?   145 areas are undernourished, and, in many countries, not all children in nonpoor households or in urban areas are well nourished. The 33 country-specific notes in appendix C of this report are a first step in this direction. For countries with limited budgetary resources, the greatest decline in stunt- ing can be accomplished by targeting the scarce resources to children who do not have adequate access to any of the three nutrition drivers (more details in chapter 5). If the same resources were to be allocated to increasing access to an additional nutrition driver among children who already have access to one driver, the consequent decline in stunting is likely to be smaller. The estimates also provide policy guidance useful for the sequencing of sector-specific interventions in target areas or target populations. ­ • First, if budgetary or other considerations allow for interventions covering deprived children by only one sector, this sector should be health. Thus, the “biggest bang for the buck” in reducing stunting is through expanded cov- erage by the health sector addressing the immediate causes of undernutri- tion (figure O.7 in the overview). The findings also empirically validate the sequencing of integrated approaches to improving nutrition outcomes, such as the one currently in process for Madagascar and in preparation for other first-wave countries. As soon as other sectors succeed in redirecting their operations to the target areas, an acceleration in the reductions in stunting is very likely to follow. • Second, if a target area is already covered by the health sector, the decision of whether to cover the same target area by sectors, such as WASH or agri- culture, should be based mainly on costs rather than benefits. The benefits in terms of accelerating reductions in stunting through simultaneous coverage by WASH or agricultural (food/care) operations appear to be similar; how- ever, country-specific analyses would need to be carried out to confirm whether the regional relationship holds nationally. Increased access to adequate levels of the underlying drivers of nutrition coordinated across different sectors should not be considered in isolation from programs increasing incomes and minimizing income variability in rural areas, both important determinants of household demand for better nutrition (see chapter 7). Instead, programs and interventions aimed at increasing the level and stability of income among populations where stunting is prevalent should be considered as indispensable components of a multisectoral approach to reducing undernutrition in Sub-Saharan Africa. Household constraints faced at different points in time may interact with the utilization of services. Thus, multisectoral interventions against undernutrition are likely to be more effec- tive when accompanied by broader development policies and programs that mitigate the impacts of weather-related risks. 146   ALL HANDS ON DECK There is an increasing policy emphasis on adaptive social protection and climate-smart agriculture programs that provide the basis for increasing the level and decreasing the variability of incomes in rural areas (Del Ninno, Coll- Black, and Fallavier 2016; Del Ninno and Mills 2015; Lipper et al. 2014; Tirado et al. 2013; Wheeler and von Braun 2013). These programs not only serve as useful instruments to respond ex post to the incidence of droughts, floods, and other natural disasters but also help households build their resilience before shocks hit. Drought-resistant seeds for maize, for example, have the potential of increasing both the level and the stability of income from agricultural activi- ties in Sub-Saharan African countries. Cash transfer programs can redistribute income to the poorer segments of the population and allow households to invest in human capital and in child nutrition, to build assets, and to diversify their livelihood strategies. Public works programs can help households and communities reduce their vulnerability to shocks while improving community infrastructure and the opportunities for new and improved livelihoods. In par- allel or in combination, increased access to insurance products and credit mar- kets can ensure better and more efficient use of resources by eliminating the incentive to adopt low-risk/low-return crops and production methods. Increased access can also alleviate intertemporal distortions on human and pro- ductive capital investment, such as cutting down on food consumption and health services or withdrawing children from school. The experience of Peru highlights the contribution of income growth in reducing stunting. Between 2007 and 2012 the prevalence of stunting declined by 21.4 percentage points from 54.7 percent to 33.3 percent in the districts tar- geted by the CRECER strategy (Levinson and Balarajan 2013; World Bank 2012). The contemporaneous income growth and poverty reduction in Peru during the same period had a critical facilitating role in the success of the mul- tisectoral efforts at decreasing undernutrition. Specifically, the poverty rate between 2004 and 2011 decreased by 31 percentage points (from 58.8 percent to 27.8 percent) and extreme poverty fell from 16.7 percent to 6.3 percent. Although the declines in stunting cannot be exclusively attributed to economic growth and poverty reduction in Peru, it is important for policy makers in Sub- Saharan African countries to consider income growth and reduced income vari- ability as necessary but not sufficient conditions for the reduction of child stunting. More research is needed on the extent to which sector-specific nutrition- sensitive interventions have any measurable impact on stunting over and above the impact from access and utilization (or the targeting) of the services provided by the normal operations of sector-specific programs. Nutrition-sensitive inter- ventions are believed to be essential for achieving adequate access to the under- lying determinants of nutrition by improving or redirecting or adding marginal changes to normal sector operations to enhance the coverage and effectiveness How Can a Multisectoral Strategy Succeed?   147 of nutrition-specific interventions through the health sector. Recent reviews of the evidence available on the nutritional effects of nutrition-sensitive interven- tions (Galasso et al. 2017; Ruel, Alderman, and the Maternal and Child Nutrition Study Group 2013), confirm that a strong potential exists, but they offer very little in terms of solid evidence. For example, the evidence of the nutritional effect of agricultural programs is inconclusive, mainly because of poor-quality evaluations. There is also a scarcity of evidence on the nutritional effect of social safety net programs, whereas combined early child development and nutrition interventions show promising synergies in child development and, in some cases, nutrition. Moreover, future evaluations of nutrition-sensitive interven- tions should take into serious consideration the possibility that the impacts of nutrition-sensitive interventions are likely to depend on the scale of coverage by other sectors as well as on the extent to which the other sectors implement nutrition-sensitive interventions.2 Attention to the Incentive Structure The adoption of a governance and accountability structure that provides the right incentives to all actors involved is the other fundamental ingredient of a successful multisectoral strategy. A common feature of all countries with suc- cessful multisectoral projects against stunting is the importance of a coordina- tion system that is supported by high levels of government. The Nutrition Enhancement Program of Senegal, for example, involving the Ministries of Health and Education, was coordinated by a unit attached to the Prime Minister’s office (Agency in Charge of the Fight against Malnutrition) (IEG 2016). Along similar lines the CRECER strategy in Peru was placed not in a line ministry but directly under the Prime Minister’s Office. In Brazil, a new minis- try, the Ministry of Social Development and Fight Against Hunger, was created to provide a platform for coordination with other ministries, such as the Ministry of Education administering the National School Food Program and the Ministry of Agricultural Development involved in the Food Acquisition Program. In the development community, aside from the impetus generated by the Sustainable Development Goals, there is a confluence of factors that contrib- ute to a more solid foundation for multisectoral projects to reduce undernu- trition. This foundation allows for operations that are better structured, better performing, and potentially much more effective than in the past. Most of the 33 countries analyzed in this report (with Angola being the only excep- tion) are members of the Scaling Up Nutrition (SUN) movement, whose framework is by now endorsed by 59 developing countries and over 100 part- ners and nearly 3,000 civil society organizations that are members of SUN. 148   ALL HANDS ON DECK All members of SUN prioritize nutrition as an investment in their growth and recognize nutrition as an investment in economic and social development to strengthen their nations. The renewed emphasis on eliminating extreme poverty and boosting shared prosperity has also provided the incentive for different sectors to reevaluate their country engagement strategy. The recent WASH Poverty Diagnostic Initiative, for example, generated new insights on how data can be used to inform allocation decisions to reduce inequalities and prioritize investment in WASH to boost human capital.3 Also, new models of opera- tional engagement with client countries, such as the Multiphase Programmatic Approach (MPA), offer the opportunity to improve coherence across inter- ventions and strengthen the strategic focus of operations within client coun- tries. The MPA is ideally suited for long-term engagement with multiple sectors as in the case of nutrition. In addition, the MPA separates engage- ments into phases, which facilitates greater learning and adaptation. Combined with performance-based financing with disbursement-linked tar- gets, the MPA program has the potential to give the proper incentives to the local authorities implementing the program. In the cases of Brazil and Peru, for example, the use of specific targets has been associated with highly positive results in terms of generating proactive initiatives and encouraging local own- ership and accountability at subnational levels. This report contributes to the analytical foundations of a more effective mul- tisectoral approach and as such constitutes only one component of a much broader effort aimed at improving the results of multisectoral projects toward the reduction of undernutrition. It is imperative that the renewed efforts yield the desired results especially in the countries of Sub-Saharan Africa. Notes 1. For conceptual clarity, nutrition sensitivity in this report is considered as an add-on component to the normal operations and activities of a sector program, distinct from the targeting of the program or project. Nutrition-sensitive interventions are generally identified with efforts to redirect, or improve, or add marginal changes to normal sector operations to enhance the coverage and effectiveness of nutrition- specific interventions through the health sector (for example, see Ruel, Alderman, and the Maternal and Child Nutrition Study Group 2013). 2. This also has important implications for the design of randomized control trials of nutrition-sensitive interventions, if the findings of the impact evaluation are to have any external validity for scaling up. 3. The synthesis report and the country-specific studies of the WASH Poverty Diagnostic can be accessed at: http://www.worldbank.org/en/topic/water/publication​ /wash-poverty-diagnostic. How Can a Multisectoral Strategy Succeed?   149 References Del Ninno, C., S. Coll-Black, and P. Fallavier. 2016. Social Protection Programs for Africa’s Drylands. Washington, DC: World Bank. Del Ninno, C., and B. Mills, eds. 2015. Safety Nets in Africa: Effective Mechanisms to Reach the Poor and Most Vulnerable. Washington, DC: World Bank. Galasso, E., A. Wagstaff, S. Naudeau, and M. Shekar. 2017. “The Economic Costs of Stunting and How to Reduce Them.” World Bank Policy Research Note, PRN/17/05, World Bank, Washington, DC. IEG (Independent Evaluation Group). 2009. Improving Effectiveness and Outcomes for the Poor in Health, Nutrition and Population: An Evaluation of World Bank Group Support since 1997. Washington, DC: World Bank. https://openknowledge.worldbank​ .org/handle/10986/2645. ———. 2016. A Decade of World Bank Support to Senegal’s Nutrition Program. Program Performance Assessment Report No. 110290. Washington, DC: World Bank. Levinson, J., and Y. Balarajan. 2013. “Addressing Malnutrition Multisectorally: What Have We Learned from Recent International Experience?” United Nations Children’s Fund (UNICEF) Nutrition Working Paper, UNICEF and MDG Achievement Fund, New York. Lipper, L., L. Jackson, A. Jarvis, F. Kossam, W. Mann, N. McCarthy, A. Meybeck, H. Neufeldt, T. Remington, P. T. Sen, R. Sessa, R. Shula, A. Tibu, and E. F. Torquebiau. 2014. “Climate-Smart Agriculture for Food Security.” Nature Climate Change 4 (December): 1068–72. Ruel, M. T., H. Alderman, and the Maternal and Child Nutrition Study Group. 2013. “Nutrition-Sensitive Interventions and Programmes: How Can They Help to Accelerate Progress in Improving Maternal and Child Nutrition?” Lancet 382 (9891): 536–551. http://dx.doi.org/10.1016/S0140-6736(13)60843-0. Tirado, M. C., P. Crahy, L. Mahy, M. Neira, S. Msangi, R. Brown, C. Scaramella, D. Costa Coitinho, and A. Müller. 2013. “Climate Change and Nutrition: Creating a Climate for Nutrition Security.” Food and Nutrition Bulletin 34 (4): 533–47. UN (United Nations). 2015. “Resolution A.70/L.1. Transforming Our World: The 2030 Agenda for Sustainable Development.” United Nations General Assembly, New York, September 18. UNICEF (United Nations Children’s Fund). 1990. Strategy for Improved Nutrition of Women and Children in Developing Countries: A UNICEF Policy Review. New York: UNICEF. Wheeler, T., and J. von Braun. 2013. “Climate Change Impacts on Global Food Security.” Science 341 (6145): 508–13. doi:10.1126/science.1239402. World Bank. 2012. Managing Risk, Promoting Growth: Developing Systems for Social Protection in Africa. The World Bank’s Africa Social Protection Strategy 2012–2022. Washington, DC: World Bank. ———. 2014. “Learning from World Bank History: Agriculture and Food-Based Approaches for Addressing Malnutrition.” Agriculture and Environmental Services Discussion Paper 10, World Bank, Washington, DC. 150   ALL HANDS ON DECK ———. 2017. Reducing Inequalities in Water Supply, Sanitation, and Hygiene in the Era of the Sustainable Development Goals: Synthesis Report of the WASH Poverty Diagnostic Initiative. Washington, DC: World Bank. https://openknowledge.worldbank.org​ /bitstream/handle/10986/27831/W17075.pdf?sequence=5&isAllowed=y. Appendix A Exploring the Relationship between Height-for-Age and the Three Drivers of Nutrition This appendix provides more details about the model used to investigate the relationship between height-for-age Z (HAZ) scores and access to the three drivers or different combinations of these three drivers. To keep the notation simple, the presentation uses the continuous HAZ score as a dependent variable instead of stunting, which is a binary variable. Each child is assigned into one of the seven exclusive groups, each group distinguished by whether the household/child has access to an adequate level of only one or only two or all three of the drivers of nutrition. For this purpose, seven binary variables are constructed by taking into consideration whether the child has access to adequate levels in the other clusters. Specifically, the binary variable BFC is equal to 1 if the household has adequate care and food security only (and inadequate health, and environment) and 0 otherwise. Similarly, BW is 1 when the household is adequate in water, sanitation, and hygiene (WASH) services only (and inadequate in food and care and health) and 0 otherwise; and BH is 1 when the household is adequate in health services only (and inadequate care and food, and WASH) and is 0 otherwise. In the same fashion, BFC_H takes the value of 1 if the household has adequate food and care practices and ade- quate health services at the same time (and inadequate WASH), and BFC_W takes the value of 1 if the household has adequate food and care and adequate WASH (and inadequate health). A comparison of the mean values of HAZ scores in each of these groups of children can shed light on the extent to which simultaneous access to adequate levels of more than one of the drivers is associated with higher HAZ scores of children. Estimates of the regression model below can provide the answer to this question: α + βFCBFC + βWBW + βHBH + βFC_WBFC_W + βFC_HBFC_H HAZi =  βW_HBW_H + βFC_W_HBFC_W_H + εi(A.1) 151 152   ALL HANDS ON DECK In this specification, the constant term α provides an estimate of the mean value of HAZ scores for children without access to an adequate level for all three underlying determinants of nutrition: care and food security (BFC= 0), WASH/ environment (BW = 0), and health (BH = 0). With E(HAZ|B = 1 or 0) denoting the expected (or mean) value of height-for-age (outcome), conditional on hav- ing adequate access (B = 1) or inadequate access (B = 0), the expected HAZ score when the child does not have adequate access to any of the three determi- nants is1 E(HAZi|BFC = 0, BH = 0, BW = 0, BFC_W = 0, …., BFC_H_W = 0) = α The coefficients βj yield estimates of the increase in the mean HAZ score of children when a child has access to adequate levels in one of the clusters only (or net of the potential gain in HAZ of having access to an adequate level in one or more of the other clusters). That is, E(HAZi|BFC = 1) = α + βFC E(HAZi|BW = 1) = α + βW E(HAZi|BH = 1) = α + βH E(HAZi|BFC_W = 1) = α + βFC_W E(HAZi|BFC_H = 1) = α + βFC_H E(HAZi|BW_H = 1) = α + βW_H Specifically, the coefficient βFC yields an estimate of the increase in the mean HAZ score of children (compared to the mean HAZ score of the reference group summarized by the constant term α), who have access to adequate food and care only (BFC = 1) but do not have access to adequate health or to adequate WASH. The coefficients βH, and βW have analogous interpretations for health and WASH, respectively. When a child has access to two adequacies, for example FC and W, the dif- ference in the mean HAZ score for these children in comparison with those who have access to none is given by βFC_W. That is, as with the case of access to one determinant, the improvement in the mean height for those with access to two determinants is given by the respective estimate of β. Note 1. It is also assumed that E(εi|BFC, BH, …, BFC_H_W) = 0. Appendix B In Search of Synergies By definition, synergies are present when various elements are combined and interact to produce a total effect that is greater than the sum of the individual elements. In this appendix, an effort is made to derive some quantitative esti- mates of the role of synergies associated with having simultaneous access to adequate levels in one or more of the clusters of food and care, health, and water, sanitation, and hygiene (WASH) in child nutrition. The estimates of the coefficients βFC, βW, βH, βFC_W, βFC_H, βW_H, and βFC_W_H, in equation (A.1) of appendix A can also provide evidence on the presence or absence of synergies among the drivers of nutrition. If synergies (or complemen- tarities) are present, then one would expect that βjk > βj + βk. For example, if there are significant synergies between health and WASH, then, all other things being equal, one would expect the decline in the stunting rate of children with simul- taneous access to adequate health and adequate WASH to be greater than reduc- tion in stunting associated with one group of children having access to adequate health only, and the reduction in stunting associated with yet another group of children having access to adequate WASH only (that is, βW_H > βW + βH). The case βjk > βj + βk, suggests that there is some substitutability between the two determinants. Access to one determinant may change the nutritional context of the child such that access to another determinant does not yield as large returns than were it to be employed in a context without access to that determi- nant. It is also plausible that the potential synergies associated with simultaneous access to two of the underlying determinants of nutrition are not realized either because of poor coordination between the local entities implementing interven- tions on j and k, for example, health and WASH practices, or because of neglect of the impediments and negative side effects associated with inadequate access to the other underlying determinants of nutrition (food and care practices). For more direct evidence on synergies, an alternative econometric specifica- tion is also estimated. The main difference in the model used to investigate more 153 154   ALL HANDS ON DECK fully the potential role of synergies is that the definition of the access variables is “inclusive,” meaning that the access variables are constructed without any consideration to whether the child has access to adequate levels in the other drivers. Specifically, the model (model B) estimated is α + γFCAFC + γWAW + γHAH + γFC_W (AFC ∗ AW) + γFC_H (AFC ∗ AH) + HAZi =  γWH (AW ∗ AH) + γFC_W_H (AFC ∗ AW ∗ AH) + εi,(B.1) where HAZi is the height-for-age Z (HAZ) scores for the child i, and AFC is equal to 1 if the household has adequate care and food and 0 otherwise. Similarly, AH is 1 when the household is adequate in health services and is 0 otherwise; AW is 1 when the household is adequate in WASH services. The key difference between these binary variables from the ones in appendix A is that they are constructed irrespective of whether the child also fulfills the conditions to other groupings of adequacies. As in equation (A.1) in the appendix A, the constant term α provides an estimate of the mean value of HAZ scores for children with access to inade- quate food and care (AFC = 0), inadequate health (AH = 0), and inadequate WASH (AW = 0). That is, with E(HAZ|A = 1 or 0) denoting the expected (or mean) value of height-for-age (outcome), conditional on having adequate access (A = 1) or inadequate access (A = 0) to driver A, the expected HAZ score for when the child does not have adequate access to any of the three drivers is1 E(HAZi|AFC = 0, AH = 0, AW = 0) = α The coefficients γ yield estimates of the increase in the mean HAZ score of children when a child has access to adequate levels in one of the drivers only (or net of the potential gain in HAZ of having access to an adequate level in one or more of the other clusters). That is, E(HAZi|AFC = 1, AH = 0, AW = 0) = α + γFC E(HAZi|AFC = 0, AW = 1, AH = 0) = α + γW E(HAZi|AFC = 0, AW = 0, AH = 1) = α + γH Specifically, the coefficient γFC yields an estimate of the increase in the mean HAZ score of children (compared to the mean HAZ score of reference group summarized by the constant term, α) that have access to adequate food and care (AFC = 1) but do not have access to adequate health, (AH = 0), and adequate WASH (AW = 0). The coefficients γH and γW have analogous interpretations for health and WASH, respectively. The γ estimates for having access to one deter- minant are the same as the β estimates for access to one determinant in appen- dix A. That is, γFc = βFC. In Search of Synergies   155 The coefficients γjk in model B above yield estimates of the synergies associ- ated with having access to adequate levels in more than one of the drivers of nutrition. Specifically, the mean HAZ score of children having simultaneous access to adequate health (AH = 1) and adequate WASH (AW = 1) is summarized by the expression E(HAZi|AFC = 0, AW = 1, AH = 1) = α + γW + γH + γW_H. The expression for the mean value of HAZ scores of children in households with access to adequate health and adequate WASH consists of the sum of three components: the first component is the increase in HAZ scores associated with children in households with adequate health only (that is, γH); the second compo- nent (that is, γW) is the increase in HAZ scores associated with children in house- holds with adequate WASH only, and the third component (that is, γW_H) is the increase in HAZ scores associated with children in households that have access to both adequate health and adequate environment. Thus, the coefficient γW_H yields information on whether there are additional (extra) gains in HAZ scores derived from having simultaneous access to adequate health and adequate WASH. A sig- nificant and positive value of the coefficient γW_H implies synergies from the simultaneous access to adequate WASH and adequate health services in the pro- duction of child nutrition (after controlling for possible synergies of having simul- taneous access to adequate food and care and health).2 The mean HAZ of children from having access to adequate levels of the other two drivers (for example, food/ care and health, or food/care and WASH) are similarly defined. Note that these estimates are different from those in appendix A and γW_H ≠ βW_H. The mean HAZ of children having simultaneous access to three components—that is, adequate food/care (A FC = 1) and adequate health ­ (AH = 1), and adequate WASH (AW = 1)—is given by the expression E(HAZi|AFC = 1, AW = 1, AH = 1) = α + γFC + γW + γH + γFC_W + γFC_H + γW_H + γFC_W_H, with the coefficient γFC_+H_W summarizing the potential synergies from simulta- neous access to the three components. Note that these are synergies in addition to any synergies from pairwise interactions. It follows that only when the complete and full specification (that is, with all the interactions) of model A and model B are estimated is there a relationship between the coefficients β in model A and the coefficients β and γ in model B as follows: βFC = γFC βW = γW βH = γH βFC_W = γFC + γW + γFC_W 156   ALL HANDS ON DECK βFC_H = γFC + γH + γFC_H βW_H = γW + γH + γW_H βFC_H_W = γFC + γW + γH + γFC_W + γFC_H + γW_H + γFC_W_H It is also important to note that, if the specification estimated is not the com- plete/full specification (that is, with all the interaction terms), then the relation- ship between the coefficients of model A and model B breaks down. For example, if in model B we construct a dummy that identifies whether the child has access to (any) two of the clusters (irrespective of which two) and run the regression with this dummy instead of the separate interaction dummies AFC ∗ AH and AFC ∗ AW and AH ∗ AW, then the relations above are not going to hold. Tables B.1 and B.2 present the results from estimating model B using a linear probability model (LPM) with stunting status as the dependent variable. Table B.1  Synergies among the Three Drivers of Nutrition in the Pooled Sample of Countries Including child, parental, Variable No covariates and household controls Food/care determinant (FC) −0.068*** −0.009 (0.013) (0.013) WASH determinant (W) 0.010 0.032** (0.012) (0.012) Health determinant (H) −0.104*** −0.048*** (0.008) (0.008) Food/care and WASH 0.030 −0.024 (0.026) (0.026) Food/care and health −0.010 −0.009 (0.018) (0.017) WASH and health −0.047*** −0.024 (0.016) (0.016) All three 0.010 0.039 (0.034) (0.033) Observations 74,781 68,533 Source: Estimates based on data from 33 recent Demographic and Health Surveys from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. No information on mother’s height is available for Angola or Senegal and therefore they are not part of the analyses with covariates. Note: Marginal effects are based on the coefficient estimates obtained from the logit model in equation (B.1). Robust standard errors in parentheses corrected for correlation at the cluster level. Stratum is the lowest level of statistical representation of the Demographic and Health Survey within a country, typically identifying regions and urban/rural areas within regions in a country. The analyses include country fixed effects as well as the following child, parental, and household characteristics: dummy variables for age (in months), gender, multiple births and birth order, the age of the mother, the marital status of the mother, mother’s education level (in years), mother’s height (in cm), the number of household members, the number of children under five years of age, the household’s wealth quintile, and whether the household lives in an urban/rural area. WASH = water, sanitation, and hygiene. ***p < 0.01, **p < 0.05, *p < 0.1. In Search of Synergies   157 Table B.2  Synergies in the Pooled Sample of Countries Based on the Components of the Three Drivers of Nutrition Including child, parental, Variables No covariates and household controls F: Child has adequate food intake −0.040*** −0.028* (0.015) (0.015) C1: Immediate skin-to-skin contact −0.008 −0.005 (0.007) (0.007) C2: Age-appropriate breastfeeding 0.055*** 0.057*** (0.007) (0.009) W1: Improved non-shared sanitary −0.034** −0.020 facilities (0.017) (0.018) W2: Improved water source −0.006 0.007 (0.007) (0.008) W3: Handwashing facility on site −0.033*** −0.010 (0.012) (0.013) W4: Percentage of households using −0.048*** −0.073*** open defecation in cluster (0.013) (0.014) W5: Disposal of feces 0.072*** 0.050*** (0.015) (0.016) H1: Had 4 or more prenatal checks −0.050*** −0.045*** (0.008) (0.008) H2: Birth assisted by trained −0.055*** −0.043*** professional (0.007) (0.007) H3: Child has been to at least one 0.010 0.004 postnatal checkup (0.007) (0.007) H4: Vaccinations up to date −0.092*** −0.020** (0.008) (0.008) H5: Child sleeps under a mosquito −0.022*** −0.015* bed net (0.008) (0.008) Food/care and health F_H1 0.024* 0.021 (0.014) (0.014) F_H4 −0.065*** −0.035** (0.015) (0.015) F_H5 −0.010 −0.003 (0.015) (0.014) Food and WASH F_W1 0.024 −0.000 (0.016) (0.016) (continued next page) 158   ALL HANDS ON DECK Table B.2 (continued) Including child, parental, Variables No covariates and household controls WASH and health W1_H1 −0.013 0.009 (0.014) (0.014) W1_H4 −0.018 −0.005 (0.014) (0.014) W1_H5 −0.031** −0.036** (0.014) (0.014) Observations 51,724 46,685 Source: Estimates based on data from 26 recent Demographic and Health Surveys from Sub-Saharan Africa. See table 2.1 in chapter 2 for the list of countries and years. Cameroon, Republic of Congo, Gabon, and Niger have no information on handwashing facilities; Republic of Congo and Gabon have no information on disposal of feces; Ethiopia and Lesotho have no information on mosquito nets; and Mozambique has no information on postnatal checkups. Thus these countries are not part of the pooled components analyses. No information on mother’s height is available for Angola or Senegal and therefore they are not part of the analyses with covariates. Note: Marginal effects are based on the coefficient estimates obtained from the logit model in equation (B.1). Robust standard errors in parentheses corrected for correlation at the cluster level. Stratum is the lowest level of statistical representation of the Demographic and Health Survey within a country, typically identifying regions and urban/rural areas within regions in a country. Child, parental, and household characteristics consist of the following variables: dummy variables for age (in months), gender, multiple births and birth order, the age of the mother, the marital status of the mother, mother’s education level (in years), mother’s height (in cm), the number of household members, the number of children under five years of age, the household’s wealth quintile, and whether the household lives in an urban/ rural area. WASH = water, sanitation, and hygiene. ***p < 0.01, **p < 0.05, *p < 0.1. The advantages and the shortcomings of the LPM in comparison to the logit or probit nonlinear model are well known in the literature. The LPM is used because the coefficients of the LPM provide direct estimates of the marginal effects of the change in the probability of stunting associated with a change in any of the independent variables. A detailed comparison of the marginal effects estimated by the LPM models confirmed that they are practically identical in sign and size to the marginal effects obtained from the logit model reported in the body of the report.3 Table B.1, based on the child data pooled across the 33 countries, reveals that the results regarding synergies are not robust (that is, they are very sensitive to the specification estimated). For example, when child, parental, and household controls are excluded from the regression, there appear to be some synergies between joint access to health and WASH. Access to WASH alone is not statisti- cally significantly different from zero and access to health alone is associated with a 10.5-percentage-point decrease in the probability of being stunted. However, access to health and WASH jointly is associated with an additional 4.5-percentage-point decrease in the probability of being stunted that is In Search of Synergies   159 statistically significant. However, when child, parental, and household controls are included as controls in the regression, there do not appear to be any signifi- cant synergies from joint access to health and WASH or from any other combi- nation of nutrition drivers. It is possible that a similar analysis based on country-level data may be able to shed more light on the presence of synergies. Rather than searching for synergies associated with access to three drivers of nutrition, each of which consists of numerous components, Table B.2 presents estimates based on the individual components of food/care, health, and WASH, along with interactions of selected pairs of components. The estimates in table B.2 provide some spotty but more robust evidence in favor of synergies between adequate food intake and vaccinations (F_H4) and between improved non- shared sanitary facilities and sleeping under a mosquito net. Notes 1. It is also assumed that E(εi|AFC, AH, AW) = 0. 2. Note that an equivalent test for the presence of statistically significant synergies in the model estimated in Annex A is a test of the null hypothesis that βjk – (βj + βk) = 0 against the one-sided alternative βjk – (βj + βk) > 0. Although this test is fully equiva- lent, the alternative test carried out in this annex B is presented since it is more direct. 3. It should be noted, however, that the constant terms obtained from the LPM in the regressions including child, parental, and household characteristics, were consistently greater than 1. This result occurs because the LPM does not constrain predicted prob- abilities (that is, the stunting rate of the reference group) to be between 0 and 1. The comparisons between the LPM and logit methods confirmed that the estimate of the constant term was the main difference between the two models, because the marginal effects obtained by the two methods were qualitatively the same. Appendix C Stunting and the Three Drivers of Nutrition: Country-Specific Estimates Table C.1  Stunting and the Three Drivers of Nutrition within Countries in Sub-Saharan Africa (3) (1) (2) Burkina (4) (5) (6) (7) Angola Benin Faso Burundi Cameroon Chad Comoros F/C −0.080* −0.135** 0.001 0.005 0.014 −0.058* 0.029 (0.042) (0.058) (0.067) (0.055) (0.049) (0.032) (0.075) W 0.016 0.091 0.021 −0.051 −0.008 −0.056 0.049 (0.047) (0.097) (0.133) (0.041) (0.030) (0.056) (0.060) H −0.011 0.029 0.014 −0.064* −0.065** −0.068** −0.017 (0.036) (0.026) (0.022) (0.034) (0.029) (0.029) (0.048) F/C & W −0.025 0.262* −0.037 −0.070 0.061 (0.120) (0.136) (0.053) (0.088) (0.094) F/C & H −0.001 0.018 −0.035 −0.095** 0.022 −0.061 −0.108 (0.044) (0.033) (0.038) (0.042) (0.050) (0.061) (0.095) W&H −0.085* −0.012 0.001 −0.077** −0.082** −0.000 −0.029 (0.050) (0.049) (0.048) (0.038) (0.036) (0.061) (0.057) All 3 0.034 0.067 0.082 −0.094** −0.018 −0.025 −0.152 (0.080) (0.071) (0.104) (0.044) (0.056) (0.152) (0.119) Observations 2,553 2,747 2,531 2,415 2,014 3,498 769 (continued next page) 161 162   ALL HANDS ON DECK Table C.1 (continued) (9) (10) (13) (8) Côte Congo, (11) (12) Gambia, (14) Congo, Rep. d’Ivoire Dem. Rep. Ethiopia Gabon The Ghana F/C −0.013 0.038 −0.075** 0.021 0.056 −0.116* −0.079 (0.086) (0.055) (0.034) (0.032) (0.091) (0.065) (0.087) W −0.065 −0.098 −0.024 −0.103 −0.190** −0.031 0.111** (0.099) (0.101) (0.044) (0.080) (0.074) (0.044) (0.055) H −0.039 −0.011 −0.056** −0.043 0.044* 0.009 0.035 (0.033) (0.032) (0.023) (0.047) (0.026) (0.037) (0.041) F/C & W −0.087 −0.105 −0.178** −0.061 (0.089) (0.137) (0.077) (0.133) F/C & H −0.041 0.052 −0.087** −0.056 −0.028 0.026 −0.015 (0.043) (0.066) (0.039) (0.061) (0.067) (0.052) (0.049) W&H −0.137 −0.041 −0.087** 0.083 −0.092 0.005 0.064 (0.097) (0.057) (0.040) (0.076) (0.057) (0.041) (0.046) All 3 −0.339*** −0.065 −0.223** −0.084 0.047 (0.120) (0.063) (0.109) (0.061) (0.055) Observations 1,760 1,288 3,163 3,440 1,288 1,293 996 (15) (16) (17) (18) (19) (20) (21) Guinea Kenya Lesotho Liberia Madagascar Malawi Mali F/C 0.053 −0.030 −0.100 −0.024 −0.084 −0.166 0.070 (0.074) (0.030) (0.123) (0.059) (0.092) (0.110) (0.056) W −0.091* −0.061 −0.185** −0.009 0.111 −0.040 −0.069 (0.052) (0.043) (0.090) (0.130) (0.103) (0.055) (0.056) H −0.005 −0.053** −0.166*** −0.033 −0.014 −0.059 −0.038 (0.027) (0.023) (0.059) (0.039) (0.034) (0.043) (0.029) F/C & W 0.234*** 0.102* −0.130 −0.223 0.036 −0.274 (0.089) (0.057) (0.160) (0.268) (0.068) (0.212) F/C & H −0.104 −0.081** −0.219*** −0.067 0.030 0.018 0.059 (0.068) (0.034) (0.072) (0.046) (0.081) (0.055) (0.065) W&H −0.034 −0.046 −0.136** −0.061 0.223*** −0.015 −0.083 (0.058) (0.037) (0.060) (0.085) (0.076) (0.046) (0.050) All 3 −0.013 −0.043 −0.211* −0.079 0.039 −0.062 0.009 (0.178) (0.048) (0.112) (0.138) (0.131) (0.056) (0.084) Observations 1,265 3,359 540 1,303 1,629 1,921 1,522 (continued next page) Stunting and the Three Drivers of Nutrition: Country-Specific Estimates   163 Table C.1 (continued) (28) (22) (23) (24) (25) (26) (27) Sierra Mozambique Namibia Niger Nigeria Rwanda Senegal Leone F/C −0.013 0.071 0.061 −0.021 −0.006 0.013 0.078 (0.033) (0.159) (0.051) (0.020) (0.056) (0.040) (0.071) W 0.001 −0.117 −0.040 −0.009 −0.027 0.004 0.225 (0.041) (0.157) (0.077) (0.015) (0.048) (0.041) (0.253) H −0.036 −0.124** −0.076*** −0.025 −0.023 −0.018 0.017 (0.022) (0.051) (0.028) (0.019) (0.045) (0.026) (0.034) F/C & W −0.103 0.385* 0.254*** −0.033 −0.032 −0.086 (0.102) (0.230) (0.098) (0.026) (0.056) (0.100) F/C & H −0.044 −0.260** 0.037 −0.035 −0.040 −0.003 −0.030 (0.041) (0.122) (0.050) (0.033) (0.052) (0.043) (0.049) W&H 0.018 −0.164* −0.032 −0.068*** −0.030 −0.019 0.063 (0.045) (0.085) (0.055) (0.025) (0.046) (0.035) (0.063) All 3 0.024 −0.059 −0.005 −0.046 −0.018 −0.067 (0.076) (0.089) (0.035) (0.050) (0.050) (0.103) Observations 3,729 386 1,736 9,252 1,355 2,410 1,480 (29) (30) (31) (32) (33) Tanzania Togo Uganda Zambia Zimbabwe F/C −0.029 −0.012 0.046 −0.020 −0.027 (0.030) (0.053) (0.061) (0.036) (0.062) W −0.024 0.183* −0.336** −0.038 −0.094 (0.067) (0.106) (0.131) (0.043) (0.072) H −0.026 −0.055* −0.132** 0.001 −0.122*** (0.022) (0.031) (0.067) (0.020) (0.033) F/C & W −0.218* 0.117 −0.093 −0.080 (0.121) (0.228) (0.069) (0.101) F/C & H −0.075** −0.040 −0.092 0.025 −0.190*** (0.030) (0.038) (0.077) (0.028) (0.047) W&H −0.083** −0.059 −0.124 0.018 −0.144*** (0.040) (0.060) (0.095) (0.036) (0.038) All 3 −0.149** 0.004 −0.127 −0.096* −0.136*** (0.066) (0.121) (0.121) (0.055) (0.050) Observations 3,719 1,275 345 4,225 1,909 Source: Estimates based on Demographic and Health Survey data. See table 2.1 in chapter 2 for details. Note: Marginal effects are based on the coefficient estimates obtained from the logit model in equation (B.1) in appendix B. Robust standard errors in parentheses corrected for correlation at the cluster level. All country -specific regressions include strata fixed effects as well as child, parental, and household characteristics. These are dummy variables for age (in months), gender, multiple births and birth order, the age of the mother, the marital status of the mother, mother’s education level (in years), mother’s height (in cm), the number of household members, the number of children under five years of age, the household’s wealth quintile, and whether the household lives in an urban/rural area. For Angola and Senegal, no information on mother’s height is available and therefore not used as a covariate. F/C = food and care; H = health; W = water, sanitation, and hygiene. ***p < 0.01, **p < 0.05, *p < 0.1. Appendix D Stunting and the Three Drivers of Nutrition: Subpopulation-Specific Estimates Table D.1  Stunting and the Three Drivers of Nutrition for Subpopulations within Sub-Saharan Africa Mother Mother more less Variables Urban Rural Boys Girls B20 T20 educated educated Food/care 0.001 −0.010 −0.030 0.011 −0.032 −0.094* −0.008 −0.007 (0.025) (0.014) (0.018) (0.017) (0.024) (0.054) (0.020) (0.015) WASH 0.009 0.040** 0.024 0.040** 0.037 0.011 0.016 0.035** (0.018) (0.016) (0.017) (0.016) (0.030) (0.023) (0.023) (0.015) Health −0.042*** −0.050*** −0.063*** −0.032*** −0.067*** −0.042** −0.030** −0.058*** (0.014) (0.009) (0.011) (0.010) (0.015) (0.019) (0.012) (0.010) Food/care −0.015 0.005 0.006 −0.008 0.055 −0.035 0.006 −0.007 and WASH (0.031) (0.027) (0.029) (0.028) (0.055) (0.036) (0.032) (0.026) Food/care −0.067*** −0.063*** −0.073*** −0.059*** −0.087*** −0.046* −0.058*** −0.069*** and health (0.020) (0.014) (0.016) (0.016) (0.023) (0.025) (0.016) (0.016) WASH and −0.032* −0.045*** −0.041** −0.040** −0.009 −0.042** −0.042*** −0.028* health (0.017) (0.015) (0.016) (0.016) (0.033) (0.020) (0.015) (0.015) All 3 −0.029 −0.061*** −0.048** −0.040* −0.071 −0.045* −0.053*** −0.026 (0.023) (0.022) (0.023) (0.023) (0.054) (0.025) (0.020) (0.026) Observations 19,751 48,782 34,273 34,260 15,496 10,900 22,059 46,474 (continued next page) 165 166   ALL HANDS ON DECK Table D.1 (continued) Low High Low High Variables Low GNI High GNI Low Gini High Gini Tariff Tariff Fragility Fragility Food/care 0.002 −0.026* −0.002 −0.024* −0.003 −0.042** −0.067*** 0.001 (0.018) (0.013) (0.019) (0.013) (0.015) (0.019) (0.024) (0.031) WASH −0.045* 0.051*** −0.056** 0.057*** 0.043*** −0.038 −0.042 −0.019 (0.025) (0.014) (0.026) (0.013) (0.014) (0.025) (0.035) (0.030) Health −0.042*** −0.054*** −0.051*** −0.046*** −0.052*** −0.040*** −0.075*** −0.022 (0.011) (0.010) (0.012) (0.009) (0.010) (0.011) (0.016) (0.017) Food/care −0.100** 0.030 −0.079** 0.031 0.006 −0.059 −0.128* −0.059 and WASH (0.048) (0.022) (0.039) (0.023) (0.023) (0.050) (0.068) (0.048) Food/care −0.077*** −0.056*** −0.079*** −0.059*** −0.064*** −0.072*** −0.106*** −0.018 and health (0.018) (0.015) (0.019) (0.014) (0.014) (0.020) (0.028) (0.028) WASH and −0.056*** −0.031** −0.039** −0.044*** −0.034** −0.060*** −0.102*** −0.011 health (0.018) (0.015) (0.017) (0.016) (0.014) (0.018) (0.023) (0.024) All 3 −0.098*** −0.008 −0.077*** −0.027 −0.045** −0.043 −0.077** −0.060* (0.026) (0.021) (0.022) (0.024) (0.019) (0.031) (0.034) (0.033) Observations 28,975 39,558 31,747 36,786 43,338 25,195 11,929 9,666 Non- Low High First Non-First Sahel Sahel Non- Variables starch starch wave wave regions regions Landlocked landlocked Food/care −0.013 0.003 −0.000 −0.040*** −0.002 −0.016 0.022 −0.033*** (0.013) (0.023) (0.016) (0.015) (0.024) (0.015) (0.023) (0.012) WASH 0.053*** −0.059* 0.049*** −0.048** 0.048*** −0.032* −0.075** 0.050*** (0.013) (0.034) (0.014) (0.020) (0.018) (0.017) (0.031) (0.013) Health −0.053*** −0.029** −0.047*** −0.053*** −0.077*** −0.036*** −0.052*** −0.048*** (0.009) (0.015) (0.010) (0.010) (0.019) (0.008) (0.016) (0.009) Food/care 0.034 −0.108* 0.004 −0.026 0.004 −0.057* −0.071 0.017 and WASH (0.021) (0.060) (0.025) (0.033) (0.030) (0.029) (0.045) (0.023) Food/care −0.073*** −0.035 −0.054*** −0.084*** −0.065* −0.072*** −0.058*** −0.072*** and health (0.014) (0.022) (0.017) (0.015) (0.035) (0.012) (0.022) (0.014) WASH and −0.052*** 0.006 −0.027* −0.067*** −0.022 −0.053*** −0.039** −0.045*** health (0.015) (0.022) (0.016) (0.014) (0.029) (0.012) (0.019) (0.014) All 3 −0.029 −0.065* −0.026 −0.079*** 0.052 −0.086*** −0.089*** −0.024 (0.021) (0.034) (0.024) (0.020) (0.042) (0.017) (0.023) (0.022) Observations 38,484 23,678 34,234 34,299 13,086 55,447 25,576 42,957 Source: Estimates based on Demographic and Health Survey data. See table 2.1 in chapter 2 for details. No information on mother’s height is available for Angola or Senegal, and therefore they are not part of the analyses with covariates. No information on dietary composition is available for Burundi, Comoros, or Democratic Republic of Congo; and therefore they are not in the analyses grouped by dietary composition. Note: Marginal effects are based on the coefficient estimates obtained from the logit model in equation (B.1) in appendix B. Robust standard errors in parentheses corrected for correlation at the cluster level. All regressions include country and year fixed effects as well as child, parental, and household characteristics. These are dummy variables for age (in months), gender, multiple births and birth order, the age of the mother, the marital status of the mother, mother’s education level (in years), mother’s height (in cm), the number of household members, the number of children under five years of age, the household’s wealth quintile, and whether the household lives in an urban/rural area. B20 = bottom wealth quintile; GNI = gross national income; T20 = top wealth quintile; WASH = water, sanitation, and hygiene. ***p < 0.01, **p < 0.05, *p < 0.1. Environmental Benefits Statement The World Bank Group is committed to reducing its environmental footprint. In support of this commitment, we leverage electronic publishing options and print- on-demand technology, which is located in regional hubs worldwide. Together, these initiatives enable print runs to be lowered and shipping distances decreased, resulting in reduced paper consumption, chemical use, greenhouse gas emissions, and waste. We follow the recommended standards for paper use set by the Green Press Initiative. The majority of our books are printed on Forest Stewardship Council (FSC)– certified paper, with nearly all containing 50–100 percent recycled content. The recy- cled fiber in our book paper is either unbleached or bleached using totally chlorine-free (TCF), processed chlorine–free (PCF), or enhanced elemental chlorine–free (EECF) processes. More information about the Bank’s environmental philosophy can be found at http://www​.worldbank.org/corporateresponsibility. In Sub-Saharan Africa, the scale of undernutrition is staggering: 58 million children under the age of five are too short for their age (stunted), and 14 million weigh too little for their height (wasted). Poor diets in terms of diversity, quality, and quantity, combined with illness and poor water and sanitation facilities, are linked with deficiencies of micronutrients—such as iodine, vitamin A, and iron—associated with growth, development, and immune function. In the short term, inequities in access to the determinants of nutrition increase the incidence of undernutrition and diarrheal disease. In the long term, the chronic undernutrition of children has important consequences for individuals and societies: a high risk of stunting, impaired cognitive development, lower school attendance rates, reduced human capital attainment, and a higher risk of chronic disease and health problems in adulthood. Inequities in access to services early in life contribute to the intergenerational transmission of poverty. Recent World Bank estimates suggest that the income penalty a country incurs for not having eliminated stunting when today’s workers were children is about 9–10 percent of gross domestic product per capita in Sub-Saharan Africa. Much of the effort to date has focused on the costing, financing, and impact of nutrition- specific interventions delivered mainly through the health sector to reach the global nutrition targets for stunting, anemia, and breastfeeding, and interventions for treating wasting. However, the determinants of undernutrition are multisectoral, and the solution to undernutrition requires multisectoral approaches. An acceleration of the progress to reduce stunting in Sub-Saharan Africa requires engaging additional sectors—such as agriculture; education; social protection; and water, sanitation, and hygiene (WASH)—to improve nutrition. This book lays the groundwork for more effective multisectoral action by analyzing and generating empirical evidence to inform the joint targeting of nutrition-sensitive interventions. Using information from 33 recent Demographic and Health Surveys (DHS), measures are constructed to capture a child’s access to food security, care practices, health care, and WASH; to identify gaps in access among different socioeconomic groups; and to relate access to these nutrition drivers to nutrition outcomes. All Hands on Deck: Reducing Stunting through Multisectoral Efforts in Sub-Saharan Africa addresses three main questions: • Do children have inadequate access to the underlying determinants of nutrition? • What is the association between stunting and inadequate food, care practices, health, and WASH access? • Can the sectors that have the greatest impact on stunting be identified? This book provides country authorities with a holistic picture of the gaps in access to the drivers of nutrition within countries to assist them in the formulation of a more informed, evidence-based, and balanced multisectoral strategy against undernutrition. ISBN 978-1-4648-1396-2 90000 9 781464 813962 SKU 211396