THE WORLD BANK W O R L D B A N K O P E R A T I O N S E V A L U A T I O N D E P A R T M E N T 34462 Maintaining Momentum to 2015? An Impact Evaluation of Interventions to Improve Maternal and Child Health and Nutrition in Bangladesh OPERATIONS EVALUATION DEPARTMENT ENHANCING DEVELOPMENT EFFECTIVENESS THROUGH EXCELLENCE AND INDEPENDENCE IN EVALUATION The Operations Evaluation Department (OED) is an independent unit within the World Bank; it reports directly to the Bank's Board of Executive Directors. OED assesses what works, and what does not; how a borrower plans to run and maintain a project; and the lasting contribution of the Bank to a country's overall development. The goals of evaluation are to learn from experience, to provide an objective basis for assessing the results of the Bank's work, and to provide accountability in the achievement of its objectives. It also improves Bank work by identifying and disseminating the lessons learned from experience and by framing recommendations drawn from evaluation findings. W O R L D B A N K O P E R A T I O N S E V A L U A T I O N D E P A R T M E N T Maintaining Momentum to 2015? An Impact Evaluation of Interventions to Improve Maternal and Child Health and Nutrition in Bangladesh 2005 The World Bank Washington, D.C. http://www.worldbank.org/oed © 2005 The International Bank for Reconstruction and Development / The World Bank 1818 H Street, NW Washington, DC 20433 Telephone: 202-473-1000 Internet: www.worldbank.org E-mail: feedback@worldbank.org All rights reserved Manufactured in the United States of America First edition August 2005 The findings, interpretations, and conclusions expressed here are those of the author(s) and do not necessarily reflect the views of the Board of Executive Directors of the World Bank or the governments they represent. The World Bank cannot guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in the work do not imply on the part of the World Bank any judgment of the legal status of any territory or the endorsement or acceptance of such boundaries. Rights and Permissions The material in this publication is copyrighted. Copying and/or transmitting portions or all of this work without permission may be a violation of applicable law. The World Bank encourages dissemination of its work and will normally grant permission promptly. For permission to photocopy or reprint any part of this work, please send a request with complete information to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA, telephone 978-750-8400, fax 978-750-4470, www.copyright.com. All other queries on rights and licenses, including subsidiary rights, should be addressed to the Office of the Publisher, World Bank, 1818 H Street NW, Washington, DC 20433, USA, fax 202-522-2422, e-mail pubrights@worldbank.org. Cover photo courtesy of the World Bank Photo Library. ISBN 0-8213-6376-X e-ISBN 0-8213-6377-8 Library of Congress Cataloging-in-Publication Data is in process World Bank InfoShop Operations Evaluation Department E-mail: pic@worldbank.org Knowledge Programs and Evaluation Capacity Telephone: 202-458-5454 Development Group (OEDKE) Facsimile: 202-522-1500 Room # 19-352 MSN # 19-903 (202) 458-0382 Facsimile: (202) 522-3125 CPrinted on Recycled Paper Contents vii Acknowledgments ix Executive Summary xiii Résumé analytique xix Resumen xxiii Acronyms and Abbreviations 1 1. Maternal and Child Health in Bangladesh: A Record of Success 3 Scope of the Study 5 Evaluation Approach 5 Overview of the Report 7 2. Health, Family Planning, and Nutrition Services in Bangladesh: An Overview 7 Family Planning Programs 10 Health Services 12 Nutrition 13 3. Trends in Under-Five Mortality, Nutrition, and Fertility 13 Patterns of Mortality Decline 13 Anthropometric Outcomes 14 What Has Been Happening to Fertility? 17 4. Impact of Specific Interventions on Child Health and Fertility 17 Income Growth Accounts for Some, But Not All, Improvement in Outcomes 19 Under-Five Mortality 26 Fertility Reduction 28 Nutrition i i i M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? 29 5. A Closer Look at Nutrition: The Bangladesh Integrated Nutrition Project 30 Overview of the Project 30 Project Coverage and Targeting 33 Acquiring Knowledge 34 Turning Knowledge into Practice: The Knowledge-Practice Gap 36 The Nutritional Impact of BINP Interventions 38 Testing the Theory in Practice--How Well Did the Causal Chain Operate? 41 6. Lessons Learned 45 Annexes 45 A. Trends in Maternal and Child Health Outcomes 49 B. Cross-Country Analysis of Child Health and Nutrition Outcomes 65 C. Neonatal, Postnatal, and Child Mortality in the 1990s 109 D. Child Malnutrition during the 1990s 125 E. Women's Agency, Household Structure, and Health Outcomes 139 F. Fertility 155 G. Analysis of BINP's Community-Based Nutrition Component 187 H. DFID and World Bank Programs in Bangladesh 193 I. Agricultural Production, Natural Disasters, Seasonality, and Nutritional Outcomes 199 J. Approach Paper 207 Endnotes 219 References Boxes 2 1.1 Measures of Welfare Outcomes 22 4.1 Which Children Get Immunized? 23 4.2 Polio Eradication in Bangladesh 36 5.1 Qualitative Perspectives of the Knowledge-Practice Gap: The PPS-BD Study Figures 3 1.1 Both Under-Five Mortality and Fertility Have Fallen Rapidly 11 2.1 Immunization Coverage of Children Aged 12­23 Months 14 3.1 Nutritional Status Improved in the 1990s 14 3.2 Data from Different Sources Present a Consistent Picture 16 3.3 Fertility Decline Continued in the 1990s according to a Range of Indirect Measures 16 3.4 Knowledge of Modern Contraceptives Is Universal and Use Continues to Rise 18 4.1 Bangladesh's Improvement in Social Outcomes Is Greater than Can Be Explained by Economic Growth Alone 25 4.2 Secondary Enrollments Have Risen Rapidly in the 1990s: Educational Attainment of Women 17 to 24 Years Old 32 5.1 Various Factors Affect Women's Participation, but Restrictions on Women's Mobility in More Conservative Areas Are the Most Important i v C O N T E N T S 34 5.2 Women Living in Project Areas Are More Likely to Have Nutritional Information, Especially if They Participate in Project Activities 35 5.3 The Knowledge-Practice Gap in Project Areas: More Women Say They Know Good Behavior than Actually Practice It 38 5.4 Rice Production and Daily Energy Supply Grew Rapidly in the Late 1990s Tables 8 2.1 World Bank Credits for Health, Population, and Nutrition 15 3.1 Fertility Decline Has Always Been Erratic Based on Direct Estimates, but Continued into the 1990s Using Indirect Ones 19 4.1 Growth in GNP per Capita Accounts for at Most One-Third of the Reduction in Mortality . . . and Less than a Fifth of Lower Fertility 20 4.2 Significant Determinants of Infant and Child Mortality 23 4.3 Neonatal Mortality Has Fallen More Rapidly for Births Attended by Trained TBAs 35 5.1 Many Factors Prevent Women from Putting Nutritional Advice into Practice, Though the Project Partially Overcomes Some of These 37 5.2 Cost of Nutrition Improvements and Mortality Reduction 39 5.3 Links in the Causal Chain 42 6.1 Cost-Effectiveness of Interventions in Reducing Under-Fiver Mortality v Acknowledgments T his study is the second in a series of three set, and Hugh Waddington, and inputs provided impact studies being conducted by the by Professor Kabir and Dr. Sayed Haider of Re- Operations Evaluation Department (OED) search Evaluation Associates for Development under a Department for International Devel- (READ) and Dr. Sharifa Begum of the Bangla- opment (DFID)­OED partnership agreement. desh Institute of Development Studies (BIDS). Financial support was also provided through In addition, a background paper was com- Danish Consultant Trust Funds. In these studies, missioned from the Participatory Practitioner's OED is exploring different ways of carrying out Society­Bangladesh (PPS-BD). Preparation of ex post impact analysis in cases that did not have the report was assisted by Alain Barbu, Martha an evaluation framework in place at the start of Ainsworth, and Denise Vaillancourt. Professor the intervention. For the first study of Ghana Nicholas Mascie-Taylor (University of Cam- basic education (OED 2004b), a survey was bridge) and Dr. Arabella Duffield (Save the commissioned to follow on from a nationally Children U.K.) acted as external reviewers. Com- representative household and school survey ments were also received from Harold Alder- conducted 15 years earlier. This second study is man, Farial Mahmud, Md. Abdul Maleque (of the based on existing data sets: the Bangladesh Implementation, Monitoring and Evaluation Demographic and Health Surveys, the House- Department of the Ministry of Planning in hold Income and Expenditure Survey, two data Bangladesh), Bina Valaydon, the Planning Com- sets related to the Bangladesh Integrated Nutri- mission of the Government of Bangladesh, and tion Project--the evaluation data set and that Bernabé Sánchez, Nick York, and Shona Wynd of collected by Save the Children--and data from the DFID Evaluation Department. Helen Keller International's (HKI) Nutritional Thanks are due to all those in the gov- Surveillance Project. Thanks are due to Meera ernment of Bangladesh, nongovernmental or- Shekar of the World Bank's Nutrition Hub, Dora ganizations (NGOs), and donor agencies who Panagides of HKI, and Anna Taylor and Dr. Ara- provided their time and support to the under- bella Duffield of Save the Children for facilitating taking of this study. In the World Bank office in access to the latter three data sets. Dhaka, Rafael Cortez, Farzana Ishrat, and Shirin This report was prepared by Howard White, Jahangeer provided valuable inputs, and excel- with assistance from Nina Blöndal, Edoardo Mas- lent logistical assistance was provided by the v i i M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? support staff of the Health Project Support Of- Azizul Hoque of Thengamara Mohila Sabuj fice, including arranging a presentation of pre- Sangha for their assistance in trips to the field. liminary findings. From DFID, Dinesh Nair and Administrative support was provided by Soon Neil Squires provided useful guidance. Thanks Woon-Pak. The report was edited by Bill are due to Kaosar Afsana of Bangladesh Rural Ad- Hurlbut, Michele Spring, and Caroline McEuen. vancement Committee (BRAC) and Mohammed Acting Director-General, Operations Evaluation: Ajay Chhibber Acting Director, Operations Evaluation Department: Kyle Peters Manager, Sector, Thematic, and Global Evaluation Unit: Alain Barbu Task Manager: Howard Nial White v i i i Executive Summary I mproving maternal and child health and nutrition is central to develop- ment goals. The importance of these objectives is reflected by their in- clusion in poverty-reduction targets such as the Millennium Development Goals (MDGs) and Bangladesh's Interim Poverty Reduction Strategy Paper, supported by major development partners, including the World Bank and the U.K. Department for International Development (DFID). This report addresses the issue of what pub- Trends in Under-Five Mortality, licly supported programs and external assistance Fertility, and Nutrition from the Bank and other agencies can do to ac- Despite an inauspicious start coming out of war celerate attainment of such targets as reducing and famine, Bangladesh has achieved spectacu- infant mortality by two-thirds. The evidence pre- lar rates of progress in the past two decades, sented here relates to Bangladesh, a country that most notably with respect to fertility decline. has made spectacular progress, but needs to Contrary to common perceptions, fertility con- maintain momentum in order to achieve its own tinued to decline during the 1990s. Under-five poverty-reduction goals. The report addresses the following issues: mortality has also been reduced at a substantial (1) What has happened to child health and nu- rate--Bangladesh is one of the few countries to trition outcomes and fertility in Bangladesh achieve a sufficient rate of reduction to achieve since 1990? Are the poor sharing in the progress the MDG of a two-thirds decline by 2015. The ex- being made? (2) What have been the main de- ception to these successes has been nutrition. terminants of maternal and child health (MCH) Physical measures of nutritional status only outcomes in Bangladesh over this period? began to show some improvement in the 1990s, (3) Given these determinants, what can be said and malnutrition remains at high levels. about the impact of publicly and externally sup- Improvements in these outcomes have been ported programs--notably those of the World spread across all Bangladeshis. Although chil- Bank and DFID--to improve health and nutri- dren of the poor are more likely to suffer pre- tion? (4) To the extent that interventions have mature death than their better-off counterparts, brought about positive impacts, have they done this gap is narrowing, with mortality rates falling so in a cost-effective manner? faster among the poor than the non-poor. Con- i x M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? traceptive use and low fertility are also common schooling is lower mortality, at a cost of among the poor. $1,080­$5,400 per life saved. · Rural electrification, supported through three Sources of Under-Five Mortality Decline World Bank programs in the 1980s and 1990s, Analysis of the determinants of mortality using reduces mortality through income effects, im- both cross-country and Demographic and Health proved health services, easier water sterili- Survey (DHS) data shows that a variety of factors zation, and improved access to health have influenced the decline in under-five deaths. information, especially from TV. Taking these Improved economic well-being is the most impor- various channels into account means that chil- dren in households receiving electrification tant reason for lower child mortality, but plays less have an under-five mortality rate of 25 per of a role for infants. Social sector interventions-- 1,000 lower than that of children in non-elec- both health and education--are also found to trified households. Based on historic costs, matter, with expanded immunization coverage this amounts to $20,000 per life saved, and and greater female enrollment in primary and sec- $40,000 based on current connection costs. ondary education both playing a substantial part in mortality reduction. The results also show a Nutrition pronounced sex bias in mortality against girls, es- In order to address the poor state of nutrition, pecially in the Sylhet and Chittagong divisions. the government implemented, with World Bank Analysis of selected interventions reveals the assistance, the pilot Bangladesh Integrated Nu- following: trition Project (BINP). The core of BINP is the Community-Based Nutrition Component (CBNC), · Immunization coverage was at less than 2 per- which promotes nutrition counseling to bring cent in the early 1980s, but grew in the latter about behavior change, complemented by sup- part of the decade (largely with the support plementary feeding for pregnant women and of the United Nations Children's Fund young children. [UNICEF], but later also with the support of Analysis of the causal chain from BINP inputs other donors, including the World Bank), so to child anthropometric outcomes shows the that by 1990, close to half of all children were following: fully vaccinated in their first 12 months. Im- munization has averted more than two million · There is a weak link in the chain, as behavioral child deaths in the last two decades, at a cost change communication has been excessively of $100­$300 per life saved. focused on mothers, who are often not the · The World Bank financed the training of ap- main decisionmakers for all nutrition-related proximately 14,000 traditional birth atten- practices. dants (TBAs) through the late 1990s, when · Program coverage is generally high in project training of TBAs was abandoned following a areas, but notably lower in more conserva- shift in international opinion toward a policy tive thanas (subdistricts), especially among of all births being attended by skilled birth at- women who live with their mothers-in-law. tendants. However, the evidence presented in · There are some deficiencies in targeting: this report shows that training TBAs saved in- (a) too strict a criterion was applied in admit- fant lives, at a cost of $220­$800 per life saved. ting malnourished children to supplementary · Female secondary schooling expanded rap- feeding, while admitting children who were idly in the 1990s, especially in rural areas, growth-faltering but probably well-nourished, partly as a result of the stipend paid to all fe- and (b) feeding of pregnant women excluded male students in grades 6­10 in rural areas. many who were eligible, and included those This stipend was supported by Norwegian aid, who were not. the Asian Development Bank, the World Bank, · A large proportion of mothers of children re- and the Bangladesh government. Among the ceiving supplementary feeding claim not to benefits of the increase in female secondary have received nutritional counseling. x E X E C U T I V E S U M M A R Y · There is a substantial knowledge-practice gap, The government's health, nutrition, and pop- whereby women do not turn the advice they ulation (HNP) Strategic Investment Plan high- receive into practice (economic resource and lights the role of increasing the age at marriage time constraints are a major reason for this). as a means of reducing fertility, and several pro- · The impact on pregnancy weight gain is too grams, including the counseling provided under small to have a substantial impact on birth BINP, promote getting married later. It is a con- weight. This situation is common in other dition of the female secondary school stipend programs; the mother's pre-pregnancy nutri- program, supported by the Bank among others, tional status is a more important factor in low that recipients remain unmarried. It is true that birth weight than pregnancy weight gain and the age at marriage in Bangladesh is low, with might therefore have been a better focus for half of all women marrying by age 14. It is also the project. true that there is a well-established international pattern whereby increasing the age at marriage The list above may be read as problems to be drives down fertility. But this pattern should not fixed in the project. To some extent they have be expected to be observed in Bangladesh for been addressed under the expanded National two reasons: (1) raising the age at marriage of Nutrition Project: the targeting criteria for chil- girls aged 13 or less has no effect on the age at dren's supplementary feeding have been revised which they have their first child (as the age at and another attempt made to reach men with marriage has risen, the gap from marriage to first nutritional counseling. But the program has not birth has fallen); and (2) if a woman plans to have been a very cost effective means of improving only three to four children, as the majority of nutritional status--which has improved gener- Bangladeshi women do, then this can be accom- ally with the acceleration in food availability as- plished whether childbearing begins at 15 or 20. sociated with the yield-driven increase in rice The direct effect of expanding secondary educa- production since the late 1990s, and consequent tion will be muted, as Bangladesh has already at- reduction in the real price of rice. Simulations tained fertility levels comparable to those in show that simply giving food to families with chil- countries with higher education levels. Hence, dren would have had a larger nutritional impact. raising the age at marriage, while desirable for The cost per life saved from the hypothetical rice both maternal and child health (children born to ration is just over $2,000, half the cost of lives young mothers have a greater chance of prema- saved by BINP. ture death), will have little impact on the number of children borne by each woman during her re- productive years--though there would be a tem- Fertility Reduction porary tempo effect on the total fertility rate and The rate of fertility reduction in Bangladesh is a second-order effect as the mortality-reducing shown to exceed that which may be expected effect of later births will reduce the desired num- from other socioeconomic developments, such ber of births. Instead, high-fertility households as income growth and expanding female educa- should be targeted, partly by an attempt to re- tion. While socioeconomic developments, in- store the use of permanent contraceptive mea- cluding the demographic transition, explain a sures to their previous levels. Efforts should also part of Bangladesh's rapid fall in fertility, a large be made to tackle son preference, which creates part is attributable to the country's family plan- a barrier to fertility decline. And continued suc- ning service, built up with substantial external cess in reducing mortality will also help reduce support in the years following liberation in 1971. fertility. The continued decline of fertility in the 1990s, driven by rising contraceptive prevalence, dem- Lessons Learned onstrates the continued effectiveness of this The following general lessons follow from the program. analysis in this report: x i M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? · Externally supported interventions have had a · Gender issues are central to health strategies notable impact on MCH-related outcomes in in Bangladesh. More attention is needed to Bangladesh. Immunization has proved particu- redressing gender biases to maintain momen- larly cost effective, and has saved the lives of up tum in mortality decline and fertility reduc- to two million children under the age of five. tion. But traditional attitudes are not the · World Bank support to sectors outside of absolute constraint on service provision, as is health has contributed to better child health sometimes suggested. outcomes. · The Bank's BINP has improved nutritional · Small amounts of money save lives . . . status, but not by as much as planned. Ser- although the amount varies significantly by ious attention needs to be given to ways intervention. of improving both the efficacy and efficiency · Although interventions from many sectors af- of the program--or, if this is not possible, fect maternal and child health outcomes, this then to consideration of alternatives to scal- fact need not imply that multisectoral inter- ing up. ventions are always needed. · Rigorous impact evaluation can show which · World Bank support for training traditional government programs and external support birth attendants has reduced neonatal are contributing the most to meeting poverty- mortality . . . but this program has now been reduction goals. abandoned following the international trend · National surveys can be used for evaluation toward support for skilled birth attendants. purposes, but some adaptations would make · Programs should be based on local evidence, them more powerful, notably a more detailed rather than on general conventional wisdom. community questionnaire. x i i Résumé analytique L 'amélioration de la nutrition et de la santé maternelle et infantile tient une place centrale dans les objectifs de développement. L'importance de ces objectifs ressort du fait qu'ils sont inclus dans les objectifs de réduction de la pauvreté, tels que les objectifs de développement pour le Millénaire (ODM) ou ceux établis dans le document intérimaire de stratégie pour la réduction de la pauvreté du Bangladesh, auxquels les principaux partenaires de développement, y compris la Banque mondiale et le DFID, apportent leur concours. Ce rapport examine ce que les programmes de l'incidence des programmes exécutés sur fi- menés sur financement public et l'aide ex- nancements publics et extérieurs--tout particu- térieure fournie par la Banque et les autres or- lièrement ceux de la Banque mondiale et du ganisations peuvent faire pour accélérer la DFID--pour ce qui est d'améliorer la santé et la réalisation des objectifs tels que la réduction de nutrition? 4) Dans la mesure où les interventions deux tiers de la mortalité infantile. Les éléments ont eu des incidences positives, cela s'est-il fait d'appréciation qu'il présente concernent le dans de bonnes conditions de coût-efficacité? Bangladesh, un pays qui a accompli des progrès spectaculaires mais qui doit maintenir l'élan ac- Tendances de la mortalité des moins de quis afin d'atteindre ses propres objectifs de ré- 5 ans, de la fécondité et de la nutrition duction de la pauvreté. En dépit d'une situation de départ peu promet- Les questions ici traitées sont les suivantes: 1) teuse, puisqu'il sortait d'une période de guerre Comment ont évolué la situation sanitaire et nu- et de famine, le Bangladesh a réalisé ces 20 der- tritionnelle des enfants et la fécondité au Ban- nières années des progrès spectaculaires, sur- gladesh depuis 1990? Les pauvres prennent-ils tout pour ce qui est de la baisse de la fécondité. part aux progrès réalisés? 2) Quels ont été les Contrairement à ce que l'on croit en général, la principaux déterminants de la situation du Ban- fécondité a continué de baisser dans les années gladesh en matière de santé maternelle et infan- 90. La mortalité des moins de 5 ans a elle aussi di- tile (SMI) durant cette période? 3) Compte tenu minué sensiblement, le Bangladesh étant l'un de ces facteurs déterminants, que peut-on dire des rares pays à connaître à un taux de diminu- x i i i M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? tion suffisant pour être en mesure de parvenir à par la suite, avec celui d'autres bailleurs de la réduction de deux tiers fixée par les ODM à fonds également, y compris la Banque mon- l'horizon 2015. Le domaine qui a fait exception à diale), si bien que près de la moitié des enfants ces résultats positifs est la nutrition. Les indicateurs en 1990 étaient totalement vaccinés en attei- matériels de l'état nutritionnel ont commencé à gnant 12 mois. La vaccination a permis d'éviter donner des signes de progrès seulement dans les plus de 2 millions de décès d'enfants au cours années 90, et les niveaux de malnutrition restent des 20 dernières années, pour un coût oscil- élevés. lant entre 100 et 300 dollars par vie sauvée. Pour ce qui est des progrès réalisés, ils ont · La Banque mondiale a financé la formation concerné l'ensemble de la population. S'il est d'environ 14 000 accoucheuses traditionnelles vrai que les risques de décès prématuré sont plus jusqu'à la fin des années 90, après quoi ce type marqués pour les enfants des milieux défavori- de formation a été abandonné à la suite d'un re- sés, le décalage qui existe à cet égard diminue virement de l'opinion internationale dans le progressivement, les taux mortalité baissant à un sens d'une politique tendant à ce que toutes les rythme plus rapide chez les pauvres que chez les naissances soient assistées par des profession- non-pauvres. L'emploi de contraceptifs et un nels qualifiés. Mais les éléments d'appréciation faible taux de fécondité sont également fré- présentés dans ce rapport montrent que la for- quents parmi les pauvres. mation des accoucheuses traditionnelles a per- mis de sauver des vies d'enfants, pour un coût Facteurs à l'origine de la baisse de la de 220 à 800 dollars par décès évité. mortalité des moins de 5 ans · La scolarisation féminine dans le secondaire a L'analyse des facteurs de mortalité basée sur les progressé rapidement dans les années 90, et données couvrant différents pays ainsi que sur ce surtout en milieu rural, du fait en partie des les données de l'enquête démographique et sa- allocations d'études accordées à toutes les nitaire (EDS) révèle qu'une diversité de fac- élèves de la classe de 6e à la 3e avec l'appui de teurs sous-tend la réduction observée du l'aide norvégienne, de la Banque asiatique de nombre de décès. L'amélioration du bien-être développement, de la Banque mondiale et du économique est la principale raison expliquant gouvernement. Parmi les avantages qui en dé- la baisse de la mortalité juvénile, mais elle joue coulent figure une réduction de la mortalité, moins dans le cas de la mortalité infantile. On pour un coût de 1 080 à 5 400 dollars par constate par ailleurs que les interventions dans décès évité. les secteurs sociaux--santé et éducation--ont · L'électrification des zones rurales, qui a reçu de l'importance, l'augmentation de la couver- l'appui de trois programmes de la Banque ture vaccinale et l'accroissement des taux de mondiale dans les années 80 et 90, réduit la scolarisation féminine dans le primaire et le se- mortalité grâce aux effets de revenu qui en dé- condaire contribuant l'un et l'autre pour une coulent, à l'amélioration des services de santé, bonne part à la baisse de la mortalité. Les don- à la possibilité de stériliser plus facilement nées recueillies font par ailleurs état d'une ten- l'eau et à la plus grande accessibilité de l'in- dance particulièrement défavorable de la formation sanitaire, du fait principalement de mortalité à l'égard des filles, surtout dans les di- la télévision. La prise en compte de ces divers visions de Sylhet et de Chittagong. L'analyse de facteurs aboutit à un taux de mortalité des certaines interventions fait ressortir les élé- moins de 5 ans inférieur de 25 pour 1 000 chez ments suivants: les enfants des ménages ayant accès à l'élec- tricité, comparé à ceux qui n'y ont pas accès. · La couverture vaccinale était de moins de 2 % Sur la base des coûts historiques, cela repré- au début des années 80, mais elle a augmenté sente 20 000 dollars par vie sauvée; aux coûts dans la seconde partie de cette décennie (en de raccordement actuels, cela équivaut à 40 grande partie avec l'appui de l'UNICEF mais, 000 dollars. x i v R É S U M É A N A L Y T I Q U E Nutrition qui se retrouve fréquemment dans des pro- Pour faire face au mauvais état nutritionnel de sa grammes analogues entrepris dans d'autres population, le pays a mis en oeuvre avec l'aide de pays; l'état nutritionnel de la mère avant la gros- la Banque mondiale une opération pilote, le Pro- sesse joue davantage dans l'insuffisance pondé- jet intégré de nutrition du Bangladesh (BINP). rale à la naissance que la prise de poids en cours Son élément central est la composante de nutri- de grossesse, et aurait donc pu être un meilleur tion communautaire, consistant à promouvoir axe de focalisation pour le projet. les services de conseils nutritionnels dans le but La liste figurant dans le paragraphe qui pré- de susciter une modification des comporte- cède peut se lire comme une liste de problèmes ments, un volet complété par la fourniture d'une à régler dans le cadre du projet, comme ils l'ont alimentation d'appoint aux femmes enceintes et été dans une certaine mesure au titre de l'opé- aux jeunes enfants. ration élargie qu'est le Projet national de nutri- Une analyse de la chaîne de causalité mettant tion: les critères de ciblage pour l'alimentation en rapport, d'un côté, les apports du BINP et, de complémentaire destinée aux enfants ont été l'autre, les données anthropométriques qui se révisés, et une autre tentative effectuée pour at- dégagent au niveau des enfants révèle les élé- teindre la population masculine dans le cadre ments suivants: 1) la chaîne présente un maillon des services de conseils nutritionnels. Mais le faible en ce sens que la communication en ma- programme n'a pas été un moyen particulière- tière de modification des comportements a trop ment rentable pour l'amélioration de l'état nu- privilégié les mères, qui ne sont pas, dans bien tritionnel de la population, lequel s'est amélioré des cas, le principal décideur pour l'ensemble d'une manière générale grâce à l'augmentation des pratiques adoptées en matière nutrition- rapide des ressources alimentaires disponibles nelle; 2) le niveau de couverture du programme qui a accompagné l'accroissement, sous l'effet est généralement élevé dans les zones visées des rendements, de la production rizicole de- mais notablement moindre dans les thanas puis la fin des années 90, et la réduction du prix (sous-districts) moins en pointe, surtout parmi réel du riz qui en a résulté. Les simulations les femmes vivant avec leur belle-mère; 3) le ci- montrent que le simple fait de distribuer de la blage présente quelques lacunes: a) un critère nourriture aux familles ayant des enfants aurait trop strict a été appliqué lorsqu'il s'est agi d'ad- eu plus d'impact au plan nutritionnel. Le coût mettre les enfants souffrant de malnutrition à bé- par vie sauvée au moyen de la ration hypothé- néficier d'une alimentation complémentaire, tique de riz s'élève à un peu plus de 2 000 dol- alors qu'on a admis des enfants qui avaient des lars seulement, soit la moitié du coût des vies problèmes de croissance mais qui étaient proba- sauvées grâce au BINP. blement bien nourris; b) le volet d'alimentation des femmes enceintes a exclu bon nombre de Réduction de la fécondité femmes qui y avaient droit, tout en admettant un Le taux de réduction de la fécondité observé au certain pourcentage de femmes qui n'y avaient Bangladesh est supérieur à celui auquel on peut pas droit; 4) parmi les mères dont les enfants re- s'attendre sous l'effet d'autres facteurs socioéco- çoivent une alimentation complémentaire, une nomiques, comme l'accroissement des revenus forte proportion a indiqué ne pas avoir reçu de ou de la scolarisation féminine. Bien que les phé- conseils nutritionnels; 5) il existe un net écart nomènes socioéconomiques, et notamment la entre connaissances et pratiques, en ce sens que transition démographique, expliquent en partie la les femmes ne mettant pas en pratique les baisse rapide de la fécondité dans ce pays, cette conseils qui leur sont dispensés, situation due baisse tient pour beaucoup au service de planning pour beaucoup aux contraintes de temps et de familial que celui-ci a établi, avec d'importants ap- ressources économiques; et 6) l'impact du pro- puis extérieurs, dans les années ayant suivi l'indé- gramme sur la prise de poids en cours de gros- pendance en 1971. Le fait que le recul de la sesse est trop limité pour avoir une réelle fécondité se soit poursuivi dans les années 90, incidence sur le poids à la naissance, état de fait sous l'effet de la prévalence accrue de la contra- x v M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? ception, démontre l'efficacité que continue taux de fécondité, en essayant notamment de ra- d'avoir ce programme. mener à ses niveaux antérieurs l'utilisation de me- Le programme d'investissement stratégique sures contraceptives permanentes. Des efforts en santé, nutrition et population établi par le gou- devraient également être entrepris à l'égard de la vernement met en avant le recul de l'âge au ma- préférence pour les fils, qui crée un obstacle au riage comme moyen de réduire la fécondité, et recul de la fécondité. Enfin, la poursuite des suc- plusieurs programmes, dont les services de cès obtenus en matière de réduction de la mor- conseils fournis dans le cadre du BINP, encoura- talité contribuera également à faire baisser la gent les Bangladais à se marier plus tard. Une fécondité. condition du programme d'allocations d'études destinées aux filles inscrites dans l'enseignement Enseignements tirés secondaire, programme appuyé notamment par On peut tirer les enseignements généraux suivants la Banque, est que les bénéficiaires de ces alloca- de l'analyse contenue dans ce rapport: tions restent célibataires. Il est vrai que l'âge au mariage est peu élevé au Bangladesh: à 14 ans, la · Les opérations entreprises avec un appui ex- moitié des femmes sont mariées. Il est également térieur ont eu un impact notable sur la situa- vrai qu'il existe une tendance bien établie au plan tion de la santé maternelle et infantile au international, selon laquelle le relèvement de Bangladesh. L'effort de vaccination s'est ré- l'âge au mariage s'accompagne d'une baisse de la vélé particulièrement rentable et a sauvé la vie fécondité. Mais on ne doit pas s'attendre à obser- à quelque deux millions d'enfants âgés de ver cette tendance au Bangladesh, et ce pour moins de cinq ans. deux raisons: 1) le relèvement de l'âge au mariage · L'appui fourni par la Banque mondiale dans les pour les filles âgées de 13 ans ou moins n'a pas secteurs autres que la santé a contribué à amé- d'effet sur l'âge auquel elles ont leur premier en- liorer la situation en termes de santé infantile. fant (ainsi, à mesure que l'âge au mariage a aug- · De faibles sommes d'argent sauvent des vies... menté, le délai entre le mariage et la première quoique le montant en jeu varie sensiblement naissance s'est réduit), et 2) si une femme selon l'opération considérée. compte avoir trois ou quatre enfants seulement, · Bien que des opérations dans plusieurs sec- ce qui est le cas de la majorité des femmes au Ban- teurs influent sur la situation de la santé ma- gladesh, elle pourra y parvenir qu'elle ait son pre- ternelle et infantile, cela ne signifie pas pour mier enfant à 15 ans ou à 20 ans. L'effet direct de autant que des opérations plurisectorielles l'augmentation de la scolarisation secondaire sera s'imposent en permanence. estompé du fait que le Bangladesh a déjà atteint · L'appui de la Banque mondiale aux activités des taux de fécondité comparables à ceux obser- de formation des accoucheuses tradition- vés dans les pays qui ont des taux de scolarisation nelles a permis une réduction de la mortalité plus élevés. Ainsi, le relèvement de l'âge au ma- néonatale... mais ce programme a désormais riage, quoique souhaitable du point de vue de la été abandonné, l'opinion internationale ayant santé maternelle et infantile (les enfants nés de évolué dans le sens d'un recours à des pro- mères jeunes sont plus exposés au risque de fessionnels qualifiés pour les accouchements. décès prématuré), n'aura guère d'impact sur le · Les programmes doivent reposer sur les réali- nombre d'enfants qu'aura chaque femme durant tés locales, plutôt que sur les idées générale- sa période de reproduction--étant entendu qu'il ment admises. y aurait un effet de tempo provisoire sur l'indice · La problématique hommes-femmes est un élé- synthétique de fécondité, et qu'il y aura un effet ment déterminant pour les stratégies adoptées de second ordre à mesure que la baisse de la mor- en matière de santé au Bangladesh. Il convient talité liée aux naissances plus tardives entraînera de veiller davantage à rectifier les partis pris une réduction du nombre de naissances sou- sexistes pour maintenir l'élan acquis sur le front haité. Au lieu de cela, il conviendrait de faire un de la baisse de la mortalité et du recul de la fé- effort ciblé sur les ménages présentant de forts condité. Mais les comportements traditionnels x v i R É S U M É A N A L Y T I Q U E ne sont pas l'entrave absolue aux prestations de · Une étude d'impact rigoureuse peut mettre services que certains suggèrent parfois. en évidence les programmes gouvernemen- · Le BINP entrepris avec l'aide de la Banque a taux et appuis extérieurs qui contribuent le amélioré l'état nutritionnel de la population, plus à la réalisation des objectifs de réduction mais bien moins que prévu. Il y a lieu de prê- de la pauvreté. ter sérieusement attention aux moyens · Les enquêtes nationales peuvent être utilisées d'améliorer ce programme en termes d'effi- pour les besoins de l'évaluation, mais leurs po- cacité et d'efficience--ou, si cela n'est pas tentialités pourraient être renforcées moyen- possible, d'envisager alors des solutions de re- nant certains ajustements, notamment un change au passage à plus grande échelle. questionnaire communautaire plus détaillé. x v i i Resumen E l mejoramiento de la salud y la nutrición maternoinfantiles es indis- pensable para las metas de desarrollo. La importancia de esos objetivos se refleja en su inclusión en metas de reducción de la pobreza como, por ejemplo, los objetivos de desarrollo del milenio (ODM) y el documento provisional de estrategia de lucha contra la pobreza de Bangladesh, que cuen- tan con el respaldo de los principales asociados para el desarrollo, incluidos el Banco Mundial y el Departamento para el Desarrollo Internacional del Reino Unido (DFID, por sus siglas en inglés). En el presente informe se examina qué pueden --especialmente del Banco Mundial y del DFID-- hacer los programas que reciben ayuda pública y para mejorar la salud y la nutrición?; 4) En la me- asistencia externa del Banco y otros organismos dida que las intervenciones han propiciado efectos para acelerar el logro de metas como la reduc- positivos, ¿lo han hecho de forma rentable? ción de la mortalidad infantil en dos terceras partes. La información facilitada en el informe se Tendencias en la mortalidad refiere a Bangladesh, un país que ha avanzado de de niños menores de cinco años, forma espectacular, pero que necesita conservar la fecundidad y la nutrición el impulso para alcanzar sus propias metas de re- A pesar de un inicio poco prometedor como con- ducción de la pobreza. secuencia de la guerra y la hambruna, Bangladesh En el informe se tratan las siguientes cuestio- ha progresado espectacularmente en las dos últi- nes: 1) ¿Cómo han evolucionado los resultados en mas décadas, sobre todo en lo concerniente a la materia de salud y nutrición infantiles y de fecun- disminución de la fecundidad. Contrariamente a didad en Bangladesh desde 1990? ¿Comparten los la idea más extendida, la fecundidad mantuvo su pobres los progresos realizados?; 2) ¿Cuáles han tendencia descendente en el decenio de 1990. La sido los principales factores determinantes de los mortalidad de los niños menores de cinco años se resultados en salud maternoinfantil en Bangladesh ha reducido también a buen ritmo, siendo Ban- a lo largo de ese período?; 3) En vista de esos fac- gladesh uno de los pocos países que ha alcanzado tores, ¿qué puede decirse de los efectos de los pro- un nivel suficiente para cumplir el ODM de una gramas que reciben apoyo público y exterior reducción en dos terceras partes para 2015. La ex- x i x M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? cepción a estos éxitos ha sido la nutrición. Las los dos últimos decenios, con un costo por medidas físicas del estado nutricional sólo empe- vida salvada situado entre US$100 y US$300. zaron a registrar ligeras mejoras en los años no- · El Banco Mundial financió la capacitación de venta y la desnutrición permanece en niveles aproximadamente 14.000 parteras tradiciona- elevados. les hasta fines de los años noventa, momento Estos resultados más favorables han benefi- en que se suspendió la capacitación a raíz de ciados a todos los ciudadanos. Aunque los niños un cambio en la opinión internacional en de las familias pobres tienen más probabilidades favor de que todos los partos fueran asistidos de morir prematuramente, la brecha se va col- por profesionales. Sin embargo, los datos pre- mando, con tasas de mortalidad que descienden sentados en este informe muestran que la for- más rápido entre los pobres que en los demás mación dispensada a parteras tradicionales ha sectores de la sociedad. El uso de anticoncepti- salvado la vida de muchos lactantes, con un vos y los bajos niveles de fecundidad son tam- coste de entre US$220 y US$800 por falleci- bién comunes entre los pobres. miento evitado. · La matrícula femenina en la enseñanza se- Causas de la disminución de la mortalidad cundaria experimentó un rápido aumento de niños menores de cinco años en el decenio de 1990, especialmente en zonas rurales, debido en parte a los subsi- El análisis de los factores determinantes de la dios para becas concedidos a todas las alum- mortalidad utilizando tanto datos de compara- nas inscritas en los niveles 6 a 10 en dichas ciones entre países como de las Encuestas de zonas con ayuda de Noruega, el Banco Asiá- Demografía y Salud (EDS) muestra que son di- tico de Desarrollo, el Banco Mundial y las au- versos los factores que han influido en el des- toridades nacionales. Una de las ventajas de censo del número de muertes. Destaca el mayor la mayor escolarización femenina en secun- bienestar económico como razón de la dismi- daria es el descenso de la mortalidad, con un nución de la mortalidad infantil, pero su in- coste de entre US$1.080 y US$5.400 por fluencia es menor en lo referente a lactantes. Se muerte evitada. ha constatado que las intervenciones en el sec- · La electrificación rural, respaldada mediante tor social --tanto en salud como en educación-- tres programas del Banco Mundial en los desempeñan también un papel importante, y la años ochenta y noventa, reduce la mortali- mayor cobertura de inmunización, junto con un dad mediante efectos de ingreso, mejorando aumento en la matrícula femenina en la ense- los servicios sanitarios, facilitando la esterili- ñanza primaria y secundaria, han repercutido zación del agua y favoreciendo el acceso a in- sustancialmente en la reducción de la mortali- formación sobre la salud, especialmente por dad. Los resultados revelan, asimismo, un pro- televisión. Si se tienen en cuenta estos diver- nunciado sesgo de orden sexual, pues la sos cauces, se observa que la tasa de mortali- mortalidad femenina es más elevada, sobre todo dad de los niños menores de cinco años en en las divisiones de Sylhet y Chittagong. El análi- los hogares con suministro eléctrico es un 25 sis de algunas intervenciones indica lo siguiente: por mil menor que en los hogares que care- · La cobertura de inmunización era inferior al cen de electricidad. Sobre la base de costos 2% a principios de los años ochenta, pero au- históricos, el total asciende a US$20.000 por mentó en la segunda mitad del decenio (en vida salvada y US$40.000 en función de los gran medida gracias al apoyo del UNICEF, costos de conexión actuales. aunque posteriormente también de otros do- nantes como el Banco Mundial), de modo que Nutrición para 1990 casi la mitad de los niños habían re- A fin de hacer frente a la deficiente situación nu- cibido todas las vacunas en sus primeros 12 tricional, las autoridades pusieron en marcha a meses de vida. La inmunización ha evitado modo experimental, con ayuda del Banco Mun- más de dos millones de muertes infantiles en dial, el Proyecto de Nutrición Integrado de Ban- x x R E S U M E N gladesh (PNIB). El eje central de dicho proyecto punto incumben al Proyecto Nacional de Nutri- es el componente de nutrición de base comu- ción ampliado; se han revisado los criterios de nitaria (CNBC), que fomenta el asesoramiento selección de niños para la alimentación comple- nutricional para instaurar un cambio de com- mentaria y se ha intentado de nuevo transmitir portamiento, completado con alimentación su- consejos sobre nutrición a la población mascu- plementaria a mujeres embarazadas y niños de lina. Ahora bien, el programa no ha sido un corta edad. medio muy rentable de mejorar el estado nutri- El análisis de la cadena causal desde los insu- cional, que ha mejorado en general con la acele- mos del PNIB hasta los resultados antropométri- ración de la disponibilidad de alimentos asociada cos infantiles muestra lo siguiente: 1) existe un al aumento del rendimiento de la producción eslabón débil en la cadena, ya que la comunica- arrocera desde finales de los años noventa y la ción en pro de un cambio de comportamiento se consiguiente reducción del precio efectivo del ha centrado excesivamente en las madres, que a arroz. Las simulaciones demuestran que repar- menudo no son quienes toman las decisiones re- tiendo simplemente alimentos a las familias con lacionadas con todas las prácticas alimentarias; niños se habría logrado un mayor impacto nu- 2) la cobertura del programa es, por lo general, tricional. El costo por vida salvada con la ración elevada en las zonas del proyecto, pero mucho de arroz hipotética supera en poco los menor en thanas (subdistritos) más conserva- US$2.000, la mitad del costo de las vidas salvadas dores, en especial entre las mujeres que viven por el PNIB. con su suegra; 3) se observan algunas deficien- cias en la definición de los grupos de beneficia- Reducción de la fecundidad rios: a) se aplicaron criterios demasiado estrictos La tasa de reducción de la fecundidad en Bangla- para la admisión de niños desnutridos en la ali- desh es superior a la que cabría esperar de otros mentación suplementaria, mientras que se ad- avances socioeconómicos, como el aumento del mitió a niños con retraso de crecimiento pero ingreso y la expansión de la educación femenina. probablemente bien alimentados; b) en la ali- Si bien los avances socioeconómicos, incluida la mentación de mujeres embarazadas se excluyó a transición demográfica, explican parte de la rá- muchas que cumplían las condiciones de admi- pida caída de la fecundidad en Bangladesh, una sión mientras que se aceptó a otras que no los parte importante es atribuible al servicio nacio- cumplían; 4) un porcentaje elevado de madres nal de planificación familiar, creado con consi- cuyos hijos recibían alimentación suplementaria derable ayuda del exterior en los años que señalaron que nadie les daba consejos sobre nu- siguieron a la liberación en 1971. La disminución trición; 5) existe una brecha considerable entre persistente de la fecundidad en los años noventa, los conocimientos y la práctica que hace que las movida por la creciente prevalencia de la anti- mujeres no apliquen los consejos que reciben: concepción, demuestra la eficacia constante de ello obedece, principalmente, a la falta de recur- este programa. sos económicos y tiempo, y 6) los efectos sobre El Plan Estratégico de Inversión en Salud, Nu- el aumento de peso durante el embarazo son de- trición y Población lanzado por las autoridades masiado reducidos para que repercutan en el nacionales destaca la utilidad de aumentar la peso al nacer, lo cual sucede frecuentemente en edad de matrimonio como medio para reducir la programas similares en otros países; el estado fecundidad, y diversos programas, incluido el nutricional de la mujer antes del embarazo incide asesoramiento brindado en el marco del PNIB, de forma más acentuada en el bajo peso al naci- promueven un matrimonio más tardío. Una de miento que el aumento de peso durante el em- las condiciones del programa de subsidios para barazo y, por ende, podría haber sido más que las niñas y jóvenes cursen estudios de se- acertado ocuparse de ese aspecto en el proyecto. cundaria, apoyado por el Banco y otras entida- La lista del párrafo anterior puede conside- des, es que las beneficiarias permanezcan rarse como una lista de problemas que deben so- solteras. Es cierto que la edad de matrimonio en lucionarse en el proyecto, pues hasta cierto Bangladesh es baja y que la mitad de las jóvenes x x i M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? se casan a los 14 años o antes. También es cierto gladesh. La vacunación ha resultado especial- que existe una pauta internacional bien estable- mente eficaz en función de los costos y sal- cida según la cual el aumento de la edad de ma- vado la vida de hasta dos millones de niños trimonio reduce la fecundidad. Ahora bien, no menores de cinco años. puede esperarse que esa pauta se siga en Ban- · El apoyo del Banco Mundial a sectores distintos gladesh por dos razones: 1) el aumento de la del sanitario ha permitido obtener mejores re- edad de matrimonio de niñas de 13 años o sultados en el ámbito infantil. menos no incide en la edad a la éstas que tienen · Con pequeñas cantidades de dinero se salvan su primer hijo (la edad en el momento del ma- vidas... aunque los montos varían significativa- trimonio aumenta, pero el lapso entre el mo- mente de una intervención a otra. mento del matrimonio y el nacimiento del · Aunque las intervenciones en muchos secto- primer hijo disminuye), y 2) si una mujer ha pre- res afectan a los resultados en materia de visto no tener más de tres ó cuatro hijos, como salud maternoinfantil, ello no implica necesa- es el caso de la mayoría de las mujeres en Ban- riamente que se requieran siempre interven- gladesh, puede lograrlo aunque los embarazos ciones multisectoriales. comiencen a la edad de 15 ó 20 años. El efecto di- · El apoyo del Banco Mundial a la capacitación recto de extender la educación secundaria será de parteras tradicionales ha reducido la mor- escaso, puesto que Bangladesh ha alcanzado ya talidad neonatal... pero el programa ya se ha niveles de fecundidad equiparables a los de paí- suspendido siguiendo la tendencia interna- ses con niveles de educación más elevados. Por cional a apoyar la formación de parteras pro- lo tanto, el aumento de la edad de matrimonio, fesionales. aunque es deseable tanto para la salud materna · Los programas deberían basarse en datos lo- como para la infantil (los niños nacidos de ma- cales, en lugar de la sabiduría popular general. dres adolescentes tienen mayor probabilidad de · Las cuestiones de género son fundamentales muerte prematura), influirá poco en el número en las estrategias de salud en Bangladesh. Es de niños nacidos de cada mujer en sus años re- necesario concentrarse más en corregir los productivos; de todos modos, habría un efecto sesgos de género para mantener el impulso temporal de ritmo en la tasa total de fecundidad cobrado en la disminución de la mortalidad y y un efecto de segundo orden, ya que la reduc- la reducción de la fecundidad. Pero las actitu- ción de la mortalidad en nacimientos más tardíos des tradicionales no son la máxima cortapisa reduciría el número deseado de hijos. Sería, en la prestación de servicios, como a veces se pues, conveniente centrarse en las familias que sugiere. exhiben una alta fertilidad para, entre otras · El PNIB del Banco ha mejorado el estado nu- cosas, intentar restablecer el uso permanente de tricional, pero no por mucho menos de lo medidas anticonceptivas en sus niveles anterio- previsto. Es preciso prestar especial aten- res. Asimismo, deberían desplegarse esfuerzos ción a las formas de mejorar tanto la eficacia para abordar la preferencia por hijos varones, la como la eficiencia del programa o, en caso cual dificulta la disminución de la fecundidad. Si de que ello no sea posible, considerar las al- prosigue la reducción de la mortalidad, dismi- ternativas a la ampliación. nuirá también la fecundidad. · Si se evaluasen las consecuencias rigurosa- mente se podría conocer qué programas pú- Enseñanzas aprendidas blicos y asistencia externa contribuyen en Del análisis efectuado en el presente informe se mayor medida a alcanzar los objetivos de re- desprenden las siguientes enseñanzas generales: ducción de la pobreza. · Es posible recurrir a encuestas nacionales · Las intervenciones que cuentan con apoyo ex- con fines de evaluación, pero algunos ajus- terior han tenido un notable impacto en los tes las harían más útiles, en particular un resultados de salud maternoinfantil en Ban- cuestionario comunitario más detallado. x x i i Acronyms and Abbreviations ANC Antenatal care BCG Bacillus Calmette-Guerin BDHS Bangladesh Health Survey BFS Bangladesh Family Survey BINP Bangladesh Integrated Nutrition Project BMI Body-mass index BRAC Bangladesh Rural Advancement Committee BRSFM Bangladesh Retrospective Survey of Fertility and Mortality CAS Country Assistance Strategy CBNC Community-Based Nutrition Component CNP Community Nutrition Promoter CPS Contraceptive Prevalence Survey DFID Department for International Development (U.K.) DHS Demographic and Health Survey DPT Diphtheria, pertussis, and tetanus EOC Emergency obstetric care EPI Expanded Program of Immunization FHV Family health visitor FSSAP Female Secondary School Stipend Program FWA Family welfare assistant FWC Family Welfare Center GoB Government of Bangladesh GRR Gross Reproduction Rate HA Health assistant HAZ Height-for-age Z score HIES Household income and expenditure survey HKI Helen Keller International HPSP Health and population sector project HPNSP Health, population, and nutrition sector project IEC Information, education, and communication x x i i i M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? IPRSP Interim Poverty Reduction Strategy Paper IUD Intra-uterine device JFS Joint Finance Scheme KP Knowledge and practice MCH Maternal and child health MDGs Millennium Development Goals MOFHW Ministry of Family, Health and Welfare NGO Nongovernmental organization NIDS National Immunization Days NNP National Nutrition Program NORAD Norwegian Agency for Development NSP Nutritional Surveillance Project OED Operations Evaluation Department (World Bank) PCR Project Completion Report PHP Population and Health Project PPS-BD Participatory Practitioner's Society­Bangladesh READ Research Evaluation Associates for Development SIDA Swedish International Development Agency TBA Traditional birth attendant TFR Total fertility rate TMSS Thengamara Mohila Sabuj Sangha TTBA Trained traditional birth attendant UHZ Upazilla Health Complex UNFPA United Nations Population Fund UNICEF United Nations Children's Fund USAID United States Agency for International Development SAR Staff Appraisal Report WAZ Weight-for-age Z score WDI World Development Indicator WHO World Health Organization WHZ Weight-for-height Z score Note: Dollars amounts cited are in U.S. dollars unless stated otherwise. x x i v 1 Maternal and Child Health in Bangladesh: A Record of Success I mproving maternal and child health is central to the development chal- lenge. Bangladesh has a remarkable record in the reduction of fertility and under-five mortality, but has less impressive achievements with respect to nutrition. This study examines the impact of interventions from various sectors--health, population, nutrition, education, and electrification--on these outcomes. Will existing interventions be adequate to maintain mo- mentum toward the achievement of poverty-reduction goals, or are changes required? Improving maternal and child health (MCH) velopment is listed as the first development pri- and nutrition is central to meeting the develop- ority, and the second of the four aims in the CAS ment challenge. This is reflected in the incorpo- is to "consolidate gains in human development, ration of MCH outcomes in development goals. addressing development challenges in educa- Two of the eight Millennium Development Goals tion, health, and nutrition" (World Bank 2002b).. (MDGs) refer to MCH outcomes--reducing Adoption of a results-based approach requires under-five and maternal mortality--and child understanding the main drivers behind changes malnutrition is an indicator for the first MDG. in target outcomes. Which publicly supported Four of the 10 goals in Bangladesh's Interim interventions can acceler- Improving maternal and Poverty Reduction Strategy Paper relate to MCH. ate the pace of improve- child health and These four are to (1) reduce infant and under- ment, and so secure the five mortality rates by 65 percent and eliminate achievement of develop- nutrition is central to gender disparity in child mortality; (2) reduce the ment goals? Have the in- meeting the development proportion of malnourished children under five terventions supported by challenge. by 50 percent and eliminate gender disparity in the Bank and the Depart- child malnutrition; (3) reduce the maternal mor- ment for International Development (DFID) tality rate by 75 percent; and (4) ensure repro- helped them meet the goals of their respective ductive health services for all (Bangladesh 2003, country strategies? This report addresses these pp. 7­8) These goals are reflected in the Bank's questions in the context of Bangladesh, a coun- Country Assistance Strategy (CAS): human de- try that has made notable progress but needs to 1 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? The decline in infant and maintain momentum in prospects for development. Emerging from a child mortality will order to achieve its own violent struggle that cost more than one million poverty-reduction goals. lives, the country was beset by a famine that was enable Bangladesh to be The following outcomes responsible for at least another quarter of a mil- one of the few countries are analyzed: infant and lion deaths. Nearly one in four children was dying to meet the MDG of a two- child mortality, child nu- before reaching their fifth birthday, reflecting the trition, nutritional status highest under-five mortality rate in the region, and thirds reduction by 2015. of pregnant women and one of the highest in the world.2 Achievements in fertility low birth weight, and fer- But by the start of the 1980s the situation had reduction are even more tility (see box 1.1). Mater- changed. The decline in infant and child mortal- nal mortality is excluded ity, which began slowly in the 1960s, accelerated remarkable. on account of lack of (figure 1.1).3 By 2000, under-five mortality had data.1 However, fertility is included as a known fallen to 82 per 1,000 live births. This rate is now correlate of maternal mortality, as well as a cor- below that of neighboring countries, and the relate to child health and nutrition. rate of decline is sufficient for Bangladesh to be The years immediately following Bangladesh's one of the few countries on track to meet the liberation in 1971 were inauspicious ones for the Millennium Development Goal of a two-thirds B O X 1 . 1 Measures of Welfare Outcomes Mortality The three mortality measures are: · Infant mortality rate (IMR): the probability of death before an infant's first birthday, usually expressed per 1,000 live births · Child mortality rate (CMR): the probability of death between a child's first and fifth birthdays, expressed per 1,000 live births · Under-five mortality rate (U5M): the probability of death before a child's fifth birthday. Anthropometric measures The nutritional status of children under five is often monitored through measurements of their height, weight, and age. These three pieces of information are used to calculate three ratios: · Height for age (stunting): a measure of long-run nutritional status · Weight for height (wasting): a measure of short-run nutritional status · Weight for age (underweight): a combination of the above two measures. Nutritional status is determined by converting a measure to a z score--that is, subtracting the mean and dividing by the standard deviation (SD) from a reference population. The corresponding statistics are called the height for age z score (HAZ), weight for height z score (WHZ), and weight for age z score (WAZ). Values of less than -2 (2 SDs below the reference mean) are taken to in- dicate malnutrition; less than -3, severe malnutrition. For adults, the appropriate measure is the body mass index (BMI), which is equal to a person's weight (in kilograms) divided by the square of their height in meters. A BMI of less than 18.5 indicates that a person is underweight. Fertility The most common measure, the total fertility rate (TFR), is the number of children a woman would have if her childbearing equaled the age-specific fertility rates of women currently in those age groups. That is, TFR is a cohort measure. In the presence of de- clining fertility, a woman entering her childbearing years is expected to have fewer children than the TFR, whereas the average number of children born to women who have completed their reproductive years will exceed the TFR. 2 M A T E R N A L A N D C H I L D H E A L T H I N B A N G L A D E S H : A R E C O R D O F S U C C E S S Both Under-Five Mortality and Fertility F I G U R E 1 . 1 Have Fallen Rapidly 300 8.0 7.0 250 Total fertility rate 6.0 200 Under-five mortality 5.0 rate mortality 150 4.0 fertility 3.0 100 Total Under-five 2.0 50 1.0 0 0.0 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 Source: Annex A. reduction between 1990 and 2015. The coun- Scope of the Study try's achievements with respect to fertility are Maternal and child health and nutrition out- even more remarkable. Fertility was very high, comes result from public policies and private ac- and possibly even rising, in the 1960s and early tions in many sectors--not only health provision 1970s, with each woman bearing an average of and individual choices, but also income genera- seven live births. A rapid decline in fertility start- tion, education, and infrastructure. This study fo- ing in the mid-1970s has more than halved the cuses on a few selected interventions supported total fertility rate.4 There was less progress re- by the World Bank, DFID, and other external garding nutrition, with improvements in anthro- agencies from both within and outside the pop- pometric indicators occurring only in the 1990s. ulation and health sector. The interventions have Maternal mortality also remains very high. been selected on the grounds of (1) their dem- This report analyzes the factors behind the onstrated importance for health and nutrition success story, focusing on the 1990s. The main outcomes and (2) the involvement of external questions addressed are: partners in supporting service provision. Health, nutrition, and population programs · What has happened to child health and nutri- are clear candidates for inclusion in the evalua- tion indicators and fertility in Bangladesh tion on grounds of both an expected impact on since 1990? Are the poor sharing in the MCH-related outcomes and the level of exter- progress being made? nally supported public provision of services. · What have been the main determinants of Bangladesh has benefited from a high degree of MCH and nutrition outcomes in Bangladesh donor coordination in the health and population over this period? sector under the auspices of five World Bank · Given these determinants, what can be said credits since 1975, with substantial levels of cofi- about the impact of publicly supported pro- nancing from bilateral agencies, including those grams to improve health and nutrition? in the United Kingdom.5 The first three of these · To the extent that interventions have brought projects focused on establishing the country's re- about positive impacts, have they done so in productive health system, but the fourth and fifth a cost-effective manner? projects expanded coverage to all aspects of 3 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Bangladesh has benefited health care delivery. This It is worth mentioning some areas that have from a high degree of study is mainly concerned been excluded from the study, despite their im- with the period covered portance to MCH outcomes: income-generating donor coordination by the Fourth Population activities, disaster prevention and relief, and eco- in the health and and Health Project (PHP4, nomic infrastructure. The many interventions population sector under 1992­98) and the Health whose main impact on health is through their and Population Sector income-generation effects have been excluded the auspices of five World Project (HPSP, 1998­2004). from the analysis on two grounds. The first is Bank credits since 1975, HPSP marked the tran- that estimating income effects of interventions is with substantial levels sition from coordinated a sizeable task in itself, and not one to which the of cofinancing from project support to sector data strategy adopted for this study is well- program financing (sector suited. The second reason is that while eco- bilateral agencies. budget support). For ex- nomic growth explains part of Bangladesh's ample, DFID health spending in this period was remarkable progress in improving social out- dominated by UK£25 million denoted as "World comes, it is not, as is shown in Chapter 4, the Bank time-slice financing." whole explanation. Preparations are under way for the follow-up A second area excluded from this study is that Health, Population and Nutrition Sector Project of disaster prevention. Bangladesh is subject to (HPNSP), with further expansion in coverage. repeated disasters, most notably flooding. A sub- While nutrition activities were partly included in stantial amount of external support has gone earlier health projects, a stand-alone pilot proj- into measures to prevent flooding and to protect ect, the Bangladesh Integrated Nutrition Project people from its consequences, as well as to fund (BINP) was initiated in 1996. BINP, and the fol- relief and rehabilitation in the wake of such low-on National Nutrition Programs (NNP), em- events.7 The success of these efforts, and similar body a behavior-change approach to improving programs funded from other sources, is shown nutritional status, an approach of growing im- by the reduced death toll from natural disasters.8 portance in various areas of health care. This These flood control measures should be impor- study re-examines the data from recent evalua- tant to preserving livelihoods (though some tions of BINP to determine the effectiveness of early Bank-supported projects had the opposite this approach. effect) and saving lives. But the number of Evidence of impact on MCH outcomes from deaths directly attributable to flooding, at less two other subsectors is also assessed: female sec- than 2,000 in each disaster year, is relatively ondary schooling and rural electrification. Fe- small compared with the approximately 300,000 male education is a well-established correlate of under-five deaths each year.9 child health and nutrition, a link confirmed by Another area of great importance, but not the empirical analysis in this study (Chapter 4 considered in this study partly for the reason and Annexes C and D). The World Bank has been income-generating activities are excluded (diffi- one of the agencies supporting the female sec- culty in tracing the causal chain with the data at ondary school stipend program (FSSAP), which hand), is economic infrastructure. Large-scale in- is shown to have made a substantial contribution frastructure, such as the DFID-supported reno- to the rapid growth in secondary enrollments. vation of the Chittagong port, have helped spur Electrification appears as a determinant of mor- the rapid expansion of the garment industry, tality in many studies, although the channels for which has brought both economic growth and this effect are not so well-documented. Together changes in the position of women. Other large with other agencies, the World Bank supported projects, such as the Jamuna Bridge, have greatly rural electrification in Bangladesh through three reduced travel time across the country, promot- projects in the 1980s and 1990s,6 and this link ing social and economic integration. Smaller- thus warrants further investigation. scale rural roads, and particularly bridges, foster 4 M A T E R N A L A N D C H I L D H E A L T H I N B A N G L A D E S H : A R E C O R D O F S U C C E S S the integration of remote areas, facilitating ac- place itself in context. Qualitative information is cess to markets and other services.10 relevant to the analysis in various ways. First is the historical political context. Bangladesh has Evaluation Approach historically had a strong commitment to basic This study draws on the analysis of a number of health, introducing an essential drugs policy in existing data sets. Demographic and Health Sur- the face of substantial opposition; but in recent vey (DHS) data are used to model the determi- years there has been a poor record of delivering nants of child health and nutrition outcomes, essential health at the field level (see Chapter 2). and of fertility. Bangladesh has had three such Second is the changing role of women, and surveys, in 1992, 1996, and 1999,11 allowing an the way in which purdah Bangladesh has analysis of the relative importance of the various (social restrictions on determinants in explaining improved outcomes women) may restrict their historically had a strong in the 1990s. This analysis informed the choice ability to access health ser- commitment to basic of sectors to be included in the evaluation. vices. A third factor is the health, introducing an The impact of the interventions is estimated importance of the pri- through a structural modeling approach. That is, vate sector, although this essential drugs policy in multivariate estimates are made of the outcomes study focuses on services the face of substantial of interest (mortality, nutrition, and fertility), that have been provided opposition; but in recent drawing on the state of the art in the literature. by government with exter- The endogeneity of behavioral factors, such as nal support. years there has been a antenatal care and immunization, is controlled The findings from this poor record of delivering for through the use of appropriate instruments, study can be related to essential health at the and the selection bias from children who have OED's evaluation criteria field level. died not being in the sample for the nutrition of relevance, efficacy, and equation allowed for by a two-part sample selec- efficiency. This chapter has already illustrated tion model. Further details on methodology can the relevance of interventions intended to im- be found in the relevant Annexes (notably C, D, prove maternal and child health and nutrition to and G) and in the Approach Paper (Annex J). Bangladesh's poverty-reduction goals. The effi- Combining the marginal impact of different in- cacy of the interventions depends on establish- terventions on welfare outcomes with cost data ing a link from supported activities to welfare facilitates the cost-effectiveness analysis. outcomes, as is done in Chapters 4 and 5. Effi- The analysis of nutrition data is a partial ex- ciency is determined using cost-effectiveness ception to the above approach. The Bangladesh analysis, which is also reported in Chapters 4 Integrated Nutrition Project (BINP) conducted a and 5. survey for evaluation purposes. This study pre- sents a reanalysis of these data, modifying the Overview of the Report control group using propensity score­matching Chapter 2 presents background information on by drawing on the nationally representative Nu- the evolution of health and family planning ser- tritional Surveillance Project carried out by vices in Bangladesh and the role of external Helen Keller International. A theory-based ap- agencies in supporting these programs. The proach was adopted so that the causal chain by record with respect to mortality, fertility, and nu- which inputs are intended to affect outcomes trition is presented in Chapter 3. Chapters 4 and could be examined in detail. 5 examine the impact of selected interventions The main thrust of this approach is heavily on health, nutrition, and fertility outcomes. quantitative. But any impact evaluation needs to Chapter 6 concludes with lessons learned. 5 2 Health, Family Planning, and Nutrition Services in Bangladesh: An Overview F ollowing independence, Bangladesh established a substantial network of health and family planning facilities. This network, including staff costs, was largely financed by donor assistance during the first decade. Al- though facilities were created for both health and family planning, the focus of service delivery until the 1990s was on family planning, with the exception of immunization, for which a successful campaign was launched in the mid-1980s. The family planning services were built on a home visit system; similarly, im- munization operates through 8,000 outreach centers around the country. Plans to move to a fixed-site system, for which construction of community clinics has been undertaken at donor expense, have met with very limited success. A pilot Integrated Nutrition Project was launched in 1996, which is now being rolled out nationwide. Family Planning Programs seven. Planned improvements in the program Promotion of family planning began with the vol- were interrupted by the liberation war in 1971. unteer Family Planning Association in what was The newly independent government recognized then East Pakistan in 1952. A national family plan- population as one of the ning program was adopted in 1965, setting up a country's most pressing The newly independent network of family planning clinics run by the Fam- problems: by that time the government recognized ily Planning Board. The program used incentives country already had the population as one of the for both personnel and clients, promoting mainly highest population density sterilization and intrauterine devices (IUDs). Suc- in the world. country's most pressing cess was limited, with fewer than 10 percent of The Bangladesh gov- problems: by that time couples using contraception in 1969, though ernment's efforts to reduce the country already had awareness of modern methods increased mark- population growth cannot edly. As a result of limited acceptance, fertility re- be separated from those of the highest population mained high, with a total fertility rate (TFR) of external agencies, since density in the world. 7 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? This project was the first these agencies were cen- national Development (USAID) over whether cofinanced social sector tral from the outset, family planning should fall under a separate or- paying the majority of ganization or be integrated with health services, project undertaken by the program costs, including and with the United Nations Population Fund World Bank anywhere in salaries, in the early years. (UNFPA) over donor leadership. Hence, USAID the world. The country's first Five- and UNFPA signed their own agreements with Year Plan proposed a the government; but six bilateral agencies (Aus- health and family planning system with a hospital tralia, Canada, Germany, Norway, Sweden, and in each district,1 which in some cases was supple- the United Kingdom) cofinanced the First Pop- mented by a maternal and child welfare center at ulation Project (1975­82, table 2.1), with the the subdistrict level. Each thana was to have an up- World Bank, with the objective of assisting the azilla health complex (UHC), and each union a development of a comprehensive fertility- family welfare center (FWC). Each FWC is headed control program. This project was the first cofi- by a medical assistant, working with a family wel- nanced social sector project undertaken by the fare visitor (FWV) and a pharmacist. The FWC's World Bank anywhere in the world, and began at staff consists of a female family welfare assistant a time when the Bank was new to population is- (FWA) and a male health assistant (HA). The FWV sues. At the outset, there was little consensus-- conducts satellite clinics, the FWA makes home for example, on the importance of tackling visits, and the HA has responsibility for malaria and supply or demand and the appropriateness of epidemic control and environmental sanitation. sterilization encouraged with cash payments. The World Bank was invited to participate in The First Population Project set about creat- implementation of the plan and first fielded a ing the infrastructure needed for service deliv- mission in May 1973. Donor coordination was ery. Sixty-nine percent (US$31.7 million) of envisaged from the outset. However, disagree- project costs were for civil works--half of that ment arose with the U.S. Agency for Inter- amount for the health complexes (UHCs), and T A B L E 2 . 1 World Bank Credits for Health, Population, and Nutrition Project costs Loan amount Project Years ($ milion)a ($ milion)a Cofinanciers First Population Project 1975­82 45.70 15.0 Australia, Canada, Germany, Norway, Sweden, United Kingdom Second Population and Family Health 1979­85 89.90 30.8 Canada, Germany, Norway, Swedenb Project Third Population and Family Health 1986­92 246.40 100.9 Australia, Canada, Germany, Netherlands, Project Norway, United Kingdom Fourth Population and Health Project 1992­98 756.30 188.4 Australia, Canada, European Union, (PHP4) Germany, Netherlands, Norway, Sweden, United Kingdomc Bangladesh Integrated Nutrition 1996­02 65.74 58.6 None Project (BINP) Health and Population Sector 1998­04 2815.90 250.0 Canada, European Union, Netherlands, Program (HPSP) Sweden, United Kingdom National Nutrition Project (NNP) 2000­05 124.50 92.0 Canada, Netherlands a. Actual amounts other than HPSP and NNP, which are appraisal estimates. b. The United Kingdom was to cofinance but decided to not participate. c. Belgium was to cofinance but decided to not participate. 8 H E A L T H , F A M I L Y P L A N N I N G , A N D N U T R I T I O N S E R V I C E S I N B A N G L A D E S H : A N O V E R V I E W the bulk of the remainder for Family Health Visi- even toward the end of the Increasing the status of tor (FHV) training schools. On the software side, 1980s, the program was women was recognized most of the funds were used for FWA training and still not widely recognized for salaries for 11,700 FWAs (project savings real- as a success. While the from the outset as ized through the depreciation of the taka were achievements in output important in reducing largely used for this purpose). By 1978, 16,700 delivery were recognized fertility. FWAs had been posted, with UNFPA paying (although with project de- salaries for most of those not supported by Bank lays due to civil works problems), fertility out- finance. Although the bulk of funds went for comes were perceived as insufficient. An OED health and family planning purposes, the First project assessment report of the first two pro- Population Project had a complex design with jects, produced in 1989, did not give the impres- several multisectoral components. The Ministry sion that there was any great success story in of Health and Population Control was responsi- Bangladesh's experience. Problems were attrib- ble for the largest part of the project, but another uted to both design (too great a focus on perma- six ministries were also involved: Local Govern- nent rather than temporary contraceptive ment, Rural Development and Cooperatives; methods) and implementation, in particular the Labor and Social Welfare; Agriculture; Education; poor quality of training. The report also noted and Information and Broadcasting. In part, these continuing problems with the integration of ministries implemented programs for spreading health and family plan- Even toward the end of contraceptive knowledge--for example, through ning, and reported that the 1980s, the population agricultural extension workers. However, they managerial capability had also carried out other activities such as income been overstretched by projects were not widely generation for women, since increasing the sta- the complex multisectoral recognized as a success-- tus of women was recognized from the outset as design. but a series of surveys important in reducing fertility. However, a series of During preparations for the second Five-Year surveys started telling a started telling a different Plan, the government invited all donors to par- different story. The Bang- story. World Bank ticipate as technical advisors in drawing up the ladesh Family Survey (BFS) documents started family planning strategy, and prepared for a fol- of 1989 showed a steep low-on population project, subsequently named decline in the TFR, from proclaiming the success. the Second Population and Family Health Proj- 6.8 in 1979 to 4.6 in 1988. The CPSs of 1989 and ect. Once again, about half of project costs went 1991, as well as the registration scheme of the to civil works, mainly UHCs and family welfare Bangladesh Bureau of Statistics, revealed a simi- centers. The third population project continued lar trend. World Bank documents started pro- this pattern. Although the projects paid the claiming the success. An OED report from 1991 salaries of field staff, government was assuming noted that there was incontrovertible evidence a larger share. Under the first two projects, gov- of fertility decline and that Bank-financed pro- ernment had paid only 10 percent of project jects had contributed to that trend.2 The project costs, but its share rose to 17 percent by the completion report (PCR) for the third popula- third project, and would be 28 percent in the tion project noted that the progress was re- fourth project. markable and the contribution of the Bank's The first three population projects established projects both substantial and undeniable. Most the planned health infrastructure in much of the notably, Cleland and others (1994) argued that country. By 1990 there were more than 2,700 up- fertility reduction had been achieved as a result azilla health centers and family welfare centers, of the program, despite an inhospitable socioe- compared with 147 in 1975, when the first project conomic setting. Others have challenged this began. There were more than 21,000 family wel- conclusion, arguing that there has been socio- fare assistants in place, and the distribution of con- economic change that can account for falling traceptives had grown dramatically. However, fertility.3 9 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Subsequent projects have supported the con- These apparent successes, and that of the fam- tinued expansion and implementation of the fam- ily planning program, stand in contrast to the fail- ily planning program, although the focus on ure of current efforts to deliver the essential family planning has declined as greater emphasis service package through fixed sites. Community has been placed on health services. This changing clinics have been constructed under HPSP, but focus was reflected in a broadening of objectives; usage rates are extremely low, and there has for example, the Third Population and Family been no rise in the proportion of people utilizing Health Project was to assist government in achiev- clinics.5 This low rate may be partly attributed to ing not only its fertility goals, but also those for re- problems of accessibility and restrictions on ducing infant and maternal mortality. Two policy women's movement, but the main reason is the issues have remained contentious between the poor quality of service offered at the clinics, with government and donors. The first is the func- drugs being in short supply. In union health cen- tional relationship of health and family planning. ters many doctors simply do not take up their The second is provision of posts.6 These problems are manifestations of the Two policy issues have fixed services versus a sys- governance issues that plague service delivery in remained contentious tem of home visits. Gov- Bangladesh.7 between the government ernment has been more As with reproductive health, other health ser- and donors: the inclined to keep health vices have been assisted by a very broad array of and family planning sep- donor interventions. Each World Bank project functional relationship of arate, and to maintain has had a large and growing number of subcom- health and family home visits, which donor ponents, reaching 66 subcomponents under planning and provision critics see as reflecting the HPSP. Similarly, DFID has simultaneously sup- power of vested interests ported a large number of health-related inter- of fixed services versus a rather than a dedication to ventions, reaching a maximum of 41 separate system of home visits. improved service delivery. activities in 1999­00.8 Hence, any evaluation can- not possibly hope to meaningfully cover the full Health Services range of funded activities; indeed, it is difficult The general view is that, given the emphasis on even to imagine how they can be supervised. family planning, other health services in Bang- This study focuses its attention on a small num- ladesh have been relatively neglected.4 While there ber of selected interventions--immunization are serious shortcomings in the delivery of public and training of traditional birth attendants. health services today, two points should be noted. First, Bangladesh led the world in establishing an Immunization essential drugs policy in the early 1980s (National The government immunization program began Drugs Policy of 1982), despite opposition from the in 1979, providing immunization services from United States and, for a time, the World Bank fixed centers. Recommended vaccinations were (Chowdhury 1996). This policy helped create the one shot of bacillus Calmette-Guérin (BCG) for large pharmaceutical sector in the country and to tuberculosis; three shots of diphtheria, pertus- make essential medicines available at low cost. At sis, and tetanus (DPT) vaccine; one shot of that time, the Bangladesh government clearly had measles vaccine; and three oral doses of polio Bangladesh led the world both the capacity and will vaccine, all to be administered by 12 months of to implement an imagina- age. Funding was provided mainly by the United in establishing an tive and difficult health pol- Nations Children's Fund (UNICEF), but it was in- essential drugs policy in icy. Second, the immuniza- sufficient and vaccines were frequently not avail- the early 1980s, despite tion program, discussed able; in consequence, very limited progress was opposition from the below, is an example of a made in raising coverage, which remained well program successfully con- below 5 percent until the mid-1980s (figure 2.1). United States and, for a ducted using government The Expanded Program of Immunization time, the World Bank. staff. (EPI) was launched in 1985 with financial sup- 1 0 H E A L T H , F A M I L Y P L A N N I N G , A N D N U T R I T I O N S E R V I C E S I N B A N G L A D E S H : A N O V E R V I E W F I G U R E 2 . 1 Immunization Coverage of Children Aged 12­23 Months 90 Fully immunized (all children) 80 Fully immunized (by 12 months) 70 60 DPT only (by 12 months) children 50 of 40 30 Percentage 20 10 r l 0 1981 1986 1991 1996 2001 Sources: Annual data from WDI; bars from DHS. port from several donors, mainly the Swedish In- clinics rather than outreach centers, but this ternational Development Agency (SIDA) and change has not taken place. UNICEF. The World Health Organization (WHO) Support for immunization has become one of provided substantial technical inputs. The pro- the largest externally funded health activities in gram was revised to include community-level the country. During the 1990s, external agencies delivery through outreach centers, bringing paid about half the costs of the immunization about very rapid increases in immunization rates program. The World Bank Immunization coverage toward the end of the decade (figure 2.1). Con- and UNICEF were the tinued improvements in coverage were made main supporters; others remained well below until 1995. During the second part of the 1990s, were USAID, WHO, Japan, 5 percent until the just over half of all children were fully vaccinated and SIDA. From 1997 to mid-1980s . . . But by by 12 months. While the percentage of children 1998, SIDA stopped sup- fully vaccinated has not increased in recent porting immunization, in- the end of the 1990s years, the percentage of those having no vacci- stead paying the funds less than 10 percent of nation at all has continued to decline, reaching into the HPSP finance children had received no less than 10 percent by the end of the 1990s. pool. Donors giving money vaccination. In rural areas, immunization continues to be to the pool, which includes provided through outreach centers, comple- the UK£25 million from DFID, are indirectly con- mented by services at district and subdistrict tributing to the immunization program, which hospitals, union-level clinics, annual campaigns, consumes just under 10 percent of Ministry of and catch-up days to reach children who have Family Health and Welfare (MOFHW) expendi- missed doses. National immunization days were ture. Despite this, the government was con- instituted during the 1990s to deliver two doses cerned that the move to pool financing would of polio vaccine a year. Immunization has been reduce the funds available for the immunization included under the essential services package program, and so the World Bank assured them developed for HPSP, according to which immu- that it would make up any shortfall. In addition, nization will be made available from community USAID and the United Kingdom capitalized the 1 1 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? rotating fund for Bangladesh to utilize UNICEF's Nutrition Vaccine Independence Initiative, which is used While considerable progress had been made by to purchase DPT vaccine.9 Finally, DFID has be- the end of the 1980s in putting in place a family come the main agency financing the polio pro- planning system, and areas of primary health gram post-eradication (see box 4.2). were being developed, nutrition continued to suffer from neglect--despite high levels of mal- Training TBAs nutrition. The Bangladesh National Nutrition During the 1990s, traditional birth attendants Council was founded in 1975 to oversee nutri- (TBAs) attended over 60 percent of all births in tion policy and coordinate the various activities Bangladesh.10 Training of TBAs was, until re- being undertaken by the different ministries. cently, a central element of community-based These activities included (1) subsidized food health programs and an important part of strate- supplements to vulnerable population groups gies for safe motherhood. The World Bank, under the Ministry of Relief and Rehabilitation; under HFP III and IV, and UNFPA trained ap- (2) homestead garden production imple- proximately 14,000 TBAs in Bangladesh. How- mented by the Ministry of Agriculture; (3) dis- ever, in its decennial year (1997), the Safe tribution of vitamin A capsules twice a year to Motherhood Initiative11 disavowed the approach. children aged six months to six years; and (4) TBAs are mostly illiterate, and it was argued that iron and folic acid tablets for pregnant women they are able to understand little of the training distributed through satellite clinics. Several they receive. Drawing on international evidence, NGOs have been active in nutrition, notably the the Initiative claimed that there was little or no Bangladesh Rural Advancement Committee evidence that training TBAs had any impact on (BRAC), which implemented a community- maternal mortality--although the possible im- based nutrition project similar to that proposed Programs to train pact on neonatal mortality by the Bank. (which is demonstrated In place of this fragmented set of programs, traditional birth below) received less at- the World Bank proposed a comprehensive ap- attendants have been tention. Hence, the Safe proach to nutrition, at the heart of which was Motherhood Initiative is community-level nutritional counseling to en- largely abandoned, with now built around the ob- courage good nutritional practices among preg- a new emphasis on jective of all births being nant women and mothers of young children. A making skilled attended by skilled birth pilot program, the Bangladesh Integrated Nu- attendants--explicitly ex- trition Project (BINP), began in six thanas in attendants available. cluding trained TBAs--an 1996, expanding to 59 thanas by the time the objective that has been included as an indicator project was completed in 2002. By that time for the Millennium Development Goals. Bang- the National Nutrition Program had begun, ladesh has followed these international recom- which is rolling the program out nationally. As mendations. Previous programs to train TBAs a pilot project, BINP was subject to an inten- have been largely abandoned, with a new em- sive evaluation. These data are reexamined in phasis on making skilled attendants available. Chapter 5. 1 2 3 Trends in Under-Five Mortality, Nutrition, and Fertility T his chapter discusses trends in under-five mortality, nutrition, and fer- tility. There is no dispute that mortality has fallen, and it is shown that the decline has benefited all population groups, although disparities remain. Nutrition has also improved, although, contrary to what is suggested by some data sources, reductions in malnutrition did not occur in the 1980s. Fertility continued to decline during the 1990s, which is contrary to the pic- ture outlined by the direct measures of the total fertility rate, which suggest a stagnation during the past decade. Patterns of Mortality Decline to four years are 33 per- The decline in under-five The decline in under-five mortality has benefited cent more likely to die mortality has benefited all groups of the population, by age, wealth, than are male children. all groups of the gender, and location. But disparities between This is one of the highest boys and girls, and different regions of the coun- female-male child mortal- population, by age, try, remain high. This section examines these ity ratios in the world.3 wealth, gender, and patterns using data from the three DHSs. Marked regional varia- location. But disparities During 1985­89 and 1995­99, under-five tions in life chances also mortality fell by close to 40 percent, from over persist. A child is twice as between boys and girls, 150 deaths per 1,000 live births to less than 100 likely to die before reach- and different regions of (figure 1.1). This decline took place for all age ing his or her fifth birth- the country, remain high. groups--neonates as well as children aged one day in Sylhet as he or to five1--and for all income groups. Indeed, dur- she--but particularly she--is in Khulna. ing the 1990s under-five mortality fell marginally faster for the poor than for the non-poor.2 Anthropometric Outcomes However, there is a sex bias in Bangladeshi Until the end of the 1980s there was scant im- mortality rates. In most countries, male children provement in nutritional status in Bangladesh: are more likely to die between their first and fifth close to 70 percent of children were both birthdays than are females. Bangladesh is an ex- stunted (HAZ) and wasted (WHZ) (see box 1.1 ception to this pattern: female children aged one for definitions). However, during the 1990s 1 3 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Malnutrition rates have there was a steady im- 1980s. But the HKI data cannot be used to con- provement, although mal- struct a trend in this way, since they are only na- fallen from extremely high nutrition rates remain tionally representative since 1998. Before that levels of 70 percent in high. date, the sample was biased toward disaster- the early 1980s to still This picture is clear prone areas. Hence, as expected, the malnutri- from data reported in tion rates in those areas were higher than the high levels of around the World Bank's World national average given by the World Bank data 50 percent, with the bulk Development Indicators, (figure 3.2).4 From 1998, HKI and World Bank of the improvement taking with little change in nu- data coincide. The conclusion is thus that mal- tritional status during nutrition rates have fallen from extremely high place during the 1990s. the 1980s in either height levels of 70 percent in the early 1980s to still high for age or weight for age (figure 3.1). Data from levels of around 50 percent, with the bulk of the the Helen Keller International (HKI) Nutritional improvement taking place during the 1990s. Surveillance Project show a slightly different pic- Malnutrition varies geographically in Bangla- ture of steady improvement since the early desh. At the regional level, it is highest in Sylhet, where stunting is 57 percent, compared to just 38 percent in Khulna. Analysis of variance con- Nutritional Status Improved firms that between-cluster variation is greater than F I G U R E 3 . 1 in the 1990s (proportion that within clusters, although this is less so in <­2SDs WAZ and HAZ) urban areas, where a single cluster may include high-income housing and slums (Annex D). Mal- 80 nutrition is slightly higher among girls than among 70 WAZ (WDI) boys, although the difference is not marked and 60 does not vary by region. Finally, the nutritional sta- 50 HAZ (WDI) tus of children of poor households is worse than 40 that of children from non-poor households, and Percent 30 this gap widened slightly in the late 1990s. 20 10 What Has Been Happening to Fertility? 0 Following the rapid decline in fertility during the 1984 1986 1988 1990 1992 1994 1996 1998 2000 1980s, it was feared that there was a fertility plateau in the 1990s. Direct estimates of the total Source: WDI (2004). fertility rate (TFR) from DHS show it to have de- clined only slightly, from 3.4 to 3.3, between Data from Different 1993 and 1996, and then remained at that level F I G U R E 3 . 2 Sources Present a until 1999.5 If there is indeed a fertility plateau, it Consistent Picture (WAZ) seems that current family planning efforts have stalled. While the strategy was sufficient to 80 achieve a certain level of fertility decline (pre- 70 HKI sumably by reaching "easy acceptors"), addi- 60 tional efforts have to be made to maintain the 50 WDI momentum toward the government's target of 40 fertility at the replacement rate. Percent30 However, three arguments can be advanced to 20 suggest that fertility decline was not halted in the 10 1990s.6 First, direct estimates of fertility are less re- 0 liable than indirect ones, and indirect estimates 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 give a different picture. Second, even if the direct Sources: WDI (2004), HKI. estimates are used, there are reasons to believe 1 4 T R E N D S I N U N D E R - F I V E M O R T A L I T Y , N U T R I T I O N , A N D F E R T I L I T Y that fertility was underestimated in 1993­94 be- were being reported as Three arguments can be cause of displacement in birth reporting. Third, if being older. The effect of advanced to suggest that indirect estimates are used, they show that the di- this birth displacement is fertility decline was not rect estimates underestimate fertility (particularly exactly the same as for the in 1993­94, reinforcing the previous point), and postponement of births. halted in the 1990s. that fertility did continue to decline in the 1990s. That is, it will reduce the reported total fertility These arguments are considered in turn. rate. But since, in this case, the reduction comes from incorrectly attributing births to a later pe- Direct Versus Indirect Estimates riod, there is a downward bias of the direct fer- Table 3.1 summarizes direct fertility estimates tility estimate for 1993­94. from various surveys, together with indirect esti- mates made using data from these surveys. Two Indirect Fertility Measures points can be noted. First, the direct estimates Figure 3.3 presents indirect estimates of fertility are systematically lower than the indirect ones. using three different approaches, using BFS Second, the trend in fertility decline is far less 1989 and the three DHSs. One set of estimates from the direct estimates than from the indirect was also made using 1991 and 2001 census data ones. For example, both CPS 1983 and 1989 (see Annex F). yielded direct TFR estimates of 4.9. But these re- The data shown here confirm that direct sults have not been used to argue that there was methods underestimate The data shown here a fertility plateau in the 1980s. To the contrary, fertility. This gap is larger confirm that direct this is seen as the period of most rapid decline. for 1993­94 than the sur- veys in the succeeding or methods underestimate Underestimation of TFR in DHS, 1993­94 preceding years, which fertility. The argument that the direct estimate of TFR in the 1993­94 DHS was an underestimate is based on an apparent displacement of births--births Fertility Decline Has reported as taking place more than three years Always Been Erratic ago were actually more recent. The evidence for T A B L E 3 . 1 Based on Direct Estimates, this phenomenon comes from comparing the but Continued into the number of births reported in the three years 1990s Using Indirect Ones preceding the survey with the number reported for the three years prior to that (that is, four to TFR estimates six years before the survey). Since they cover Year Survey Directa Indirectb similar time periods, the ratio of these two 1974 Bangladesh Retrospective Survey of 4.8 7.3 amounts will be one, except that mortality will Fertility and Mortality (BRSFM) tend to force it below one, and declining fertility or postponed births will raise it above one. 1975 Bangladesh Family Survey (BFS) 5.4 7.4 But this ratio for the 1993­94 survey is greatly 1983 Contraceptive Prevalence Survey (CPS) 4.9 7.0 in excess of both one and the value of the ratio 1985 CPS 4.6 6.5 observed in the other surveys. In the earlier 1989 CPS 4.9 5.9 survey, 26 percent more births were reported 1989 BFS 4.6 5.4 for four to six years before the survey, compared with the period within three years before 1991 CPS 4.2 n.a. the survey. This ratio compares with just 4 per- 1993/94 Bangladesh Demographic and Health 3.4 4.3 cent for 1996­97 and a 5 percent reduction Survey (BDHS) in 1999­00. Yet comparing the ratio for five- 1996/97 BDHS 3.3 4.0 year periods, there is little difference between 1999/00 BDHS 3.3 3.7 the 1993­94 and 1996­97 surveys. These data a. Direct estimates are from the various survey reports. thus show that in 1993­94, children under three b. 1974­89 from Cleland and others (1994), and 1993­99 from OED analysis. 1 5 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Fertility Decline Continued in the 1990s According to a F I G U R E 3 . 3 Range of Indirect Measures 8 7 Rele: census data rate 6 Rele: survey data Proximate determinant method fertility 5 Total 4 Regression method 3 1980 1985 1990 1995 2000 Note: See Annex F for explanation. supports the argument that birth displacement edge and use (figure 3.4). By the mid-1980s, vir- exacerbated underestimation in that year. The tually all women in Bangladesh had heard of results also show that fertility continued to de- modern contraception methods, although just cline during the 1990s, albeit at a slower rate under a third had ever used them. The propor- than had been the case from 1989 to 1993. tion who have both ever used them, and are cur- The evidence of continuing fertility decline is rently doing so, has continued to rise since the consistent with patterns of contraceptive knowl- mid-1980s. Knowledge of Modern Contraceptives Is Universal F I G U R E 3 . 4 and Use Continues to Rise 100 Knowledge women 80 Ever used 60 -married ever 40 of Currently using 20 Percent 0 1975 1980 1985 1990 1995 2000 1 6 4 Impact of Specific Interventions on Child Health and Fertility A nalysis of both cross-country and Bangladeshi household data shows the clear contribution of interventions from several sectors to im- provements in maternal and child health outcomes. Immunization has averted up to two million child deaths since the launch of the Expanded Program of Immunization, and done so in a cost-effective manner. Training traditional birth attendants, which has been largely abandoned, is also shown to have been a cost-effective way of reducing neonatal deaths. From other sec- tors, female secondary schooling and rural electrification are shown to have had a significant impact on under-five mortality and fertility, although elec- trification is a much less cost-effective means for achieving these objectives. Income Growth Accounts for Some, But from the 1970s to the current decade, so that Not All, Improvement in Outcomes there are up to four observations for each coun- There is a well-established relationship between try. Analysis using data for a specific country income and social indicators. Higher income is across time reveals a similar pattern (see, for ex- generally associated with reduced mortality, ample, Haddad and others 2003). higher educational achievement, better nutrition, The solid line in each figure is the average re- and so on. However, the rapid improvements in lationship between in- Immunization has social outcomes achieved in Bangladesh are come and the social greatly in excess of what can be explained from in- outcome; that is "the fit- averted up to two come growth alone. The four graphs in figure 4.1 ted line." In the 1980s, million child deaths show this link between income and four social Bangladesh (indicated by since the launch of the outcomes--under-five mortality, the total fertility the triangular data points, rate, and the prevalence of stunting and under- each labeled by its dec- Expanded Program of weight among under-fives--using data from a ade) lay above the average Immunization, and done cross-section of 78 developing countries. Each for under-five mortal- so in a cost-effective data point represents the decade averages of in- ity and fertility, meaning come and the social outcome shown, using values that those indicators were manner. 1 7 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Bangladesh's Improvement in Social Outcomes Is Greater than F I G U R E 4 . 1 Can Be Explained by Economic Growth Alone (a) Under-five mortality (b) Total fertility rate 400 9 ths) bir 350 8 live 7 300 1,000 rate 6 250 1980s 5 (per 200 1980s fertility 4 1990s tality 150 Income-driven trajectory 1990s 3 2000 mor Total 100 Fitted line Income-driven trajectory 2 50 2000 Fitted line 1 Under-five 0 0 0 2,000 4,000 6,000 8,000 10,000 12,000 14,000 16,000 18,000 0 2,000 4,000 6,000 8,000 10,000 12,000 14,000 16,000 18,000 GDP per capita (US$) GDP per capita (US$) (c) Stunting (HAZ), percent (d) Underweight (WAZ), percent 90 90 80 80 age) 70 1980s 70 1980s age) for 60 1990s 60 for 1990s 50 50 2000 Income-driven trajectory 2000 (weight 40 (height 40 Income-driven trajectory 30 30 20 Stunting 20 10 Fitted line Fitted line Underweight 10 0 0 ­10 0 2,000 4,000 6,000 8,000 10,000 12,000 14,000 16,000 18,000 0 2,000 4,000 6,000 8,000 10,000 12,000 14,000 16,000 18,000 GDP per capita (US$) GDP per capita (US$) Note: See discussion in the text for explanations. worse than should be expected for a country at have been important, non-income-related fac- its income level. If these indicators had im- tors behind the improvement in mortality and proved following the internationally established fertility in Bangladesh. relationship with income, then subsequent ob- In the case of the nutrition indicators, the ob- servations for Bangladesh would have lain along servations from all three decades lie above the fit- This finding suggests the dashed line.1 But ted line. Bangladesh continues to have worse these later observations anthropometric outcomes than the average for that there have been lie below the fitted line, countries at a similar income level. But this dis- important, non-income- showing that Bangladesh crepancy has narrowed over time. The improve- related factors behind the now does better than ex- ment in all four outcomes greatly exceeds that in pected for a country at its gross national product (GNP) per capita, and ex- improvement in mortality income level. This find- ceeds the expected improvement derived for and fertility in Bangladesh. ing suggests that there growth in national income. 1 8 I M P A C T O F S P E C I F I C I N T E R V E N T I O N S O N C H I L D H E A L T H A N D F E R T I L I T Y The numbers behind these graphs provide an electrification to be a The largest single factor is upper estimate of the extent to which growth in fairly robust determi- immunization, which gross domestic product (GDP) per capita has nant of mortality out- contributed to improved social outcomes in comes. Second, there accounts for about one- Bangladesh (table 4.1; see Annex B for more de- is a residual (unex- third of the reduction in tails).2 For example, under-five mortality was 205 plained) element in under-five mortality, but per 1,000 live births in the 1980s. Income growth these figures, most no- alone would have reduced it to 163 by 2000, but tably for weight for age safe water, education, and by then the actual rate was 78. Hence, at most, and the fertility rate. agricultural output have just under one-third of the improvement comes The remainder of this also played a part. from higher average income. For the other out- chapter builds on these comes shown in figure 4.1, the share of income results to further analyze the factors behind im- is even less, explaining, for example, at most, proved MCH outcomes in Bangladesh and the 16 percent of the reduction in fertility. extent to which public interventions have Further analysis (contained in Annex B) of the brought about these improvements. cross-country data examines the underlying fac- tors behind changing outcomes in Bangladesh Under-Five Mortality other than GNP per capita.3 This analysis shows The reasons for premature death vary by age. that:4 For very young children, factors related to maternal health during pregnancy and the birth · The largest single factor is immunization, itself are important, with other aspects of health which accounts for about one-third of the re- care having an impact throughout infancy. duction in under-five mortality. But for children over · Increasing female literacy is an explanatory fac- For very young children, one year, the general so- tor for all four variables, accounting for between cioeconomic environment factors related to 5 and 15 percent of the observed change. matters more. It is thus · Improved daily energy supply accounts for maternal health during common to analyze mor- 10 to 15 percent of improved nutrition. pregnancy and the birth tality determinants sep- · Safe water has also contributed to improved itself are important, with nutrition, accounting for close to one-quarter arately for the different of the reduction in stunting. age groups. Such an analy- other aspects of health sis was conducted by OED care having an impact Two general conclusions emerge from these using DHS data from the throughout infancy. But results. First, interventions from different sec- three surveys carried out tors have all contributed to improved MCH in the 1990s (see Annex C for children over one year, outcomes. Not only health interventions (im- for detailed results). the general socioeconomic munization) but also safe water, education, and Table 4.2 summarizes environment matters agricultural output have all played a part. Analy- the significant determi- sis of additional data (see Annex B) also shows nants of the mortality for more. Growth in GNP Per Capita Accounts for at Most One­Third of the T A B L E 4 . 1 Reduction in Mortality . . . and Less than a Fifth of Lower Fertility 2000 income- Percent reduction 1980 actual 2000 actual based estimate explained by income Under-five mortality 205.0 77.5 163.1 32.9 Total fertility rate 5.6 3.0 5.2 16.0 Stunting 67.6 44.7 62.3 23.1 Underweight 69.5 47.7 64.6 22.4 Source: Calculated from data used for figure 4.1 (see Annex B). 1 9 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? T A B L E 4 . 2 Significant Determinants of Infant and Child Mortality Community Child characteristics Household characteristics characteristics Neonatal Public: Birth order, preceding birth interval. Public: Electricity, water and sanitation, Antenatal visits (public). mother secondary education, maternal con- Private: Multiple birth, born in April/May traceptive knowledge, mother's age at birth. (lower mortality). Private: Mother's mobility Postnatal Public: Birth order, preceding birth interval. Public: Mother's secondary education, ma- Lower in Khulna, Rajshahi, ternal contraceptive knowledge, mother's and rural areas. Private: Higher birth order female, born in age at birth. October/November in rural areas (higher mortality). Private: Wealth. Child Public: Immunization, vitamin A Public: Mother's primary and secondary edu- Lower in Khulna, Rajshahi. supplements. cation, mother's age at birth. Private: Female, stronger for higher birth Private: Wealth, mother remarried (higher order and in Chittagong. mortality). Note: "Public" indicates a characteristic that is a function of publicly provided services; "Private" indicates otherwise. Source: Annex C. the different age groups as revealed by OED's the risk of mortality for infants declines with analysis. The main findings are: birth order until a birth order of six, after which it rises again. · Mother's education matters, though it must · Sex bias manifests itself in various ways. Fe- be to at least secondary level to affect infant male children are more likely to die than mortality. Mother's contraceptive knowledge male, especially those in Chittagong5 and is included as a proxy for maternal health those of a higher birth order. knowledge, and exerts a significant impact for · The health- and nutrition-related variables infants but not children. Other measures of have the expected positive impact.6 Immu- women's agency have some effect, notably the nization is dealt with in detail below. Antena- greater mortality risk for children of women tal visits reduce neonatal mortality, but only who have remarried. if provided by trained medical personnel, · Money matters, though not for neonatal either doctor or nurse. Training of both doc- mortality. tors and nurses has been a major area of DFID · Various factors correlated to fertility influ- support over the last decade, and so a route ence mortality: there is a nonlinear rela- through which they are contributing to re- tionship between mother's age and infant duced mortality. mortality, with the probability of premature · There are seasonal patterns in mortality, re- death decreasing until close to age 30, after lated to both mother's nutritional status dur- which it increases again. For children, there ing pregnancy and seasonal factors affecting is a simpler relationship: children born to illness. older mothers have a lesser chance of dying. A short birth interval adversely affects mor- The relative importance of these different fac- tality. The pattern for birth order is more tors can be explored in three ways. The first is to complex: lower-order births, particularly the look at the extent to which they have con- first-born child, are at greater risk, though tributed to mortality decline during the 1990s. this risk is mitigated slightly if the mother at- To do so, the explanatory variable must both be tends antenatal care. The data suggest that significant and have changed notably during the 2 0 I M P A C T O F S P E C I F I C I N T E R V E N T I O N S O N C H I L D H E A L T H A N D F E R T I L I T Y decade. The results of this analysis may be mis- More generally, there The first approach shows leading, as there are some determinants, such as are various ways in which that improvements in antenatal care or immunization, that have not the deaths averted by im- economic well-being were changed much over the decade, but which munization may be calcu- would adversely affect mortality if not provided. lated (see Annex C for the main driving force A second approach, importance analysis, over- details on the calculations): behind changes in child comes this difficulty, being based on the product mortality in the 1990s, of the coefficient and the standard error of the · The cross-country re- variable. Finally, the sources that explain the dis- gression analysis un- but this is closely crepancy between high and low mortality areas dertaken for this study followed by the role of within the country are examined. (Annex B) suggests health provision and The first approach shows that improvements that increased immu- maternal education. in economic well-being were the main driving nization coverage ac- force behind changes in child mortality in the counts for close to one-third of the reduction 1990s,7 but this is closely followed by the role of in under-five mortality over the past two health provision and maternal education (see decades. Immunization began to rise from Annex C). Expanded vaccination coverage alone the mid-1980s. If mortality had remained at its accounts for over half of the mortality reduction 1987 level, there would have been 3.28 mil- achieved through higher wealth. But for infants lion additional under-five deaths from 1987 to the roles are reversed: health and education are 2002--just under one million of these lives more important than economic growth. The fer- saved is attributable to immunization. tility-related variables have a mixed effect: sur- · A study conducted in the late 1990s consid- vival chances are higher in smaller families with ered the life-saving effects of each vaccination larger birth intervals, but the reduction in birth separately, producing an estimate of 1.15 mil- order tends to increase mortality.8 The impor- lion deaths prevented in the period 1987­98 tance analysis bears out these results, showing in (Khan and Yoder 1998). particular the role of health and education. Two · The child mortality regressions show that factors mainly explain why mortality outcomes children with no vaccinations are between are so much better in Khulna than Chittagong: 50 percent to twice as likely to die as those (1) differing rates of accessing health services, children who have some coverage.10 Applying and (2) pronounced excess female mortality in this ratio to observed mortality data suggests Chittagong, which is not present in Khulna.9 that more than two million child deaths were The above analysis of determinants is now averted by immunization between 1981 and used to quantify the impact and cost-effectiveness 2002 (see Annex C for details). of the following interventions: immunization, training TBAs, female secondary schooling, and These varying approaches give somewhat rural electrification. different estimates of the number of lives saved by the Expanded Program of Immunization (EPI), Immunization but provide a range of one to two million deaths Immunization in Bangladesh has expanded averted, which can be used as a basis for cost- greatly in the last two decades (see Chapter 2 and effectiveness analysis. There are also wide es- box 4.1). Improvements in both socioeconomic timates of the full cost of immunizing a child. A conditions and immunization reduce disease. A study on immunization in disease such as polio spreads more easily where Bangladesh in the late But for infants the roles access to safe water is limited, so that improving 1990s reported a figure of are reversed: health and water supply can reduce incidence. But eradica- close to $12 per child, a lit- education are more tion of the disease requires a coordinated public tle less than international important than economic health strategy, as has been implemented in norms.11 The study also Bangladesh with external assistance (see box 4.2). presents the direct costs growth. 2 1 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? B O X 4 . 1 Which Children Get Immunized? Bangladesh's immunization program took off in the second half of the 1980s, rapidly expanding coverage. Despite this expansion, close to half of all children do not receive the full course of recommended vaccinations, and a significant number remain unvac- cinated. Multivariate analysis of DHS data shows the following: · Several variables related to women's position (see Annex E for discussion of measures and correlates of women's agency) have a significant impact on the likelihood of a child being vaccinated: mother's age, mobility, agency, and education. Chil- dren of mothers who have remarried are less likely to be vaccinated. · The divisional disadvantage of Dhaka, Sylhet, and Chittagong remains significant even once the other factors mentioned are controlled for. And the advantage of Khulna and Rajshahi is also significant. Distance from thana headquarters (HQ) reduces the likelihood of a child being immunized. · Electricity has a significant positive impact. Two indirect channels for electrification--wealth and access to information (TV)-- are already included, as is a locational variable (distance to thana HQ), which electrification may proxy for. Hence the result most likely reflects, at least in part, the impact of electrification on preserving the cold chain (keeping vaccines at the required temperature). · Wealth has a positive but statistically insignificant impact on the probability of being immunized. These results point to the importance of information campaigns in overcoming remaining barriers to immunization. As discussed in Chapter 5, the message at the heart of behavior-change communication--that all actors in the decisionmaking process need be reached--must be borne in mind. The OED analysis thus of the EPI program and has contributed approximately 10 percent of pro- confirms the relatively low estimates of total costs-- gram costs over the last 15 years, thus saving the that is, including staff and lives of between 100,000 and 200,000 Bangladeshi cost of immunization as a equipment costs not re- children. DFID's contribution to immunization means of saving lives. flected in the EPI bud- has been mainly to the polio program since eradi- get--for a single year ($18.3 million in 1997­98). cation. Before the polio eradication campaign Applying these figures gives a range of the cost more than 10,000 children a year were developing per life saved from just under $100 to just under the disease, which gives an indication of the ben- $300.12 The OED analysis thus confirms the rela- efits from keeping the disease from reappearing. tively low cost of immunization as a means of sav- ing lives.13 These figures are likely to be an Trained TBAs underestimate since they ignore the benefits As described in Chapter 2, Bangladesh followed There have been cases of of herd immunity, which international trends in abandoning training of is yet to be realized in TBAs. But the international evidence is more nu- substantial reductions in Bangladesh.14 anced than this wholesale shift in focus suggests. maternal mortality in The number of deaths First, studies are clear that training TBAs does systems relying on trained averted can be attributed make a difference to their behavior. For exam- TBAs, when backed up by to the agencies support- ple, a study in Bangladesh found that 45 percent ing the immunization of trained TBAs practiced the "three cleans" a good referral system for program on a pro rata (hand washing with soap, clean cord care, clean emergency obstetric care. basis. The World Bank surface) compared with 19 percent of untrained 2 2 I M P A C T O F S P E C I F I C I N T E R V E N T I O N S O N C H I L D H E A L T H A N D F E R T I L I T Y B O X 4 . 2 Polio Eradication in Bangladesh Polio is a viral disease that affects the nervous system and muscles and can cause paralysis. Children under three are most at risk. The disease is transmitted by physical contact, and is highly contagious in areas with poor sanitation and high population den- sity--making areas of Bangladesh very vulnerable. But polio is easily preventable using inexpensive vaccines. In 1988 the World Health Assembly of the WHO set the goal of the global eradication of polio by year 2000. There has been con- siderable progress toward this goal. In 1988 there were 350,000 cases every year worldwide. In 2003 the number had dropped to 1,000. Only six countries still have endemic polio: Afghanistan, Egypt, India, Niger, Nigeria, and Pakistan. Bangladesh has tackled polio through the four strategies identified by WHO: routine immunization as a part of EPI, national im- munization days (NIDs), a surveillance system to identify outbreaks, and a mopping up strategy in case polio reappears in any location. This strategy in Bangladesh has been very successful. Before 1986 an estimated 11,500 children developed polio every year. This number was reduced to 2,300 in 1994, and 324 in 1999. Since 2000 no polio cases have been identified. Funding efforts have been coordinated internationally by the Global Polio Eradication Initiative, which includes WHO, UNICEF, Rotary International, USAID, and other donors. In Bangladesh nearly 50 percent of the funding has been provided by the govern- ment of Japan, which spent $40­50 million on NIDs up to 2003. Other large donors are the United States' CDC and Rotary (10 per- cent each); other minor donors follow: UNICEF (6 percent), WHO (3 percent), USAID (3 percent), DFID (2 percent), and so on. However, since eradication, DFID has expanded its support with a £7 million project, Support to Polio Eradication, under which coverage through NIDs is approaching 100 percent (97 percent in 2004). Sources: WHO/CDC/UNICEF (2004), Polio eradication Web site, http://www.polioeradication.org/ and CDC, Morbidity and Mortality Weekly Re- port (various issues). TBAs (Goodburn and others 2000, p. 396; see shows the share of births attended by trained Sibley and Sipe 2004 for results of a meta- and untrained TBAs from each DHS, and the analysis). Moreover, there have been cases of neonatal mortality rate for births so attended. substantial reductions in maternal mortality in The existence of training programs for TBAs is systems relying on trained TBAs, when backed evident from the expansion of the share of births up by a good referral system for emergency ob- attended by trained TBAs (TTBAs) from 4 to stetric care (EOC) China is the best-known 10 percent between 1993 and 1999. At the same case;15 and Bangladesh has been cited as one in time the neonatal mortality rate among births at- which there was no such system of EOC and, tended by trained TBAs has more than halved-- consequently, no impact on maternal deaths.16 Finally, while there is no evidence of an impact Neonatal Mortality Has on maternal mortality, there is evidence of an Fallen More Rapidly for T A B L E 4 . 3 impact on infant mortality. The meta-analysis of Births Attended by the impact of training TBAs reported that 14 out Trained TBAs of 17 studies found a reduction of peri-neonatal mortality from births being attended by trained, Neonatal mortality rather than untrained, TBAs (Sibley and Sipe Percent of deliveries (per 1,000 live births) 2004). Untrained Trained Untrained Trained Turning to the Bangladeshi data, training TBAs appears to have a beneficial impact on 1993 55.0 4.3 49 72 neonatal mortality, though there are complica- 1996 56.7 7.6 48 49 tions in interpreting these figures. Table 4.3 1999 53.2 9.5 38 31 2 3 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? a much larger decline than that observed in gen- US$3,300­$8,000 per life saved each year. The eral or among births attended by untrained TBA may continue to practice using this knowl- TBAs. edge for 10­15 years after training, bringing the But various factors complicate the picture.17 cost per life saved down to US$220-800. These fig- Trained TBAs are likely to attend more difficult ures show training TBAs to be a cost-effective births than untrained TBAs, as is evident from the strategy for tackling infant mortality. higher mortality rate for these births in 1993.18 As In a setting where 60 to 75 percent of births training for TBAs expands, so that trained TBAs are are attended by TBAs, it is not prudent to side- more widespread, this effect may be diluted, thus line them in any strategy for improving health driving down mortality associated with TTBA- outcomes. Doing so in Bangladesh was the re- attended birth. Second, the training encourages sult of faddism on the part of the international TBAs to refer difficult births to EOC. Partly in con- community, rather than a decision based on sequence, the share of births attended by doctors solid local evidence. Such evidence suggests that also increased during this period, though this was training these women does change practices and mainly in urban areas (that is, only for a minority that doing so makes a notable and cost-effective of births). This effect will also reduce the mortality impact on infant mortality. However, there will associated with trained TBAs. Finally, training may not be an impact on maternal mortality unless reduce mortality owing to greater skills of TBAs EOC is also available. Hence, the stress needs to who are trained, which is what we hope to ob- be on increasing availability of EOC services, and serve, but which is difficult to disentangle from the ensuring that training of TBAs results in speedy other factors described above. However, it can be referrals where appropriate. Any renewed focus observed that neonatal mortality among babies on training TBAs, which the above analysis sug- whose birth was attended by a trained TBA was, in gests should certainly be considered, needs to 1999, lower than that for untrained TBAs, despite pay attention to the need for a good supervision the continued presence of the greater likelihood system for these workers and making periodic of difficult births for trained TBAs, which refresher courses available. strongly suggests that training TBAs had a bene- ficial impact on neonatal mortality. Female Secondary Schooling An estimate of the advantage of training is Female secondary enrollments grew rapidly in given by the single difference impact of 7 deaths the 1990s. DHS data show that in 1993, 20 per- averted per 1,000 live births.19 The coefficients cent of 17­24 year olds had secondary educa- from the multivariate analysis suggest a double tion, compared to 35 percent just 6 years later. In a setting where 60 to difference impact of 5 Various factors lay behind this rapid increase. per 1,000. These figures One of these is the stipend program, supported 75 percent of births are are consistent with the by the Asian Development Bank, NORAD, the attended by TBAs, it is not meta-analysis mentioned World Bank, and the Government of Bangla- prudent to sideline them earlier, which found that desh. Introduced in 1994, this program pays a training TBAs reduced stipend to all girls in rural areas as long as they in any strategy for peri-neonatal mortality attend school on at least 75 percent of school improving health by 7 per 1,000 (Sibley and days, maintain a passing grade, and remain un- outcomes. Doing so in Sipe 2004). The cost of married (see Annex I for further discussion of Bangladesh was the result training a single TBA is in the stipend program). the range $350-$400, and The beneficial impact of female education on of faddism on the part of one TBA attends be- infant and child mortality is shown above. Figure the international tween 10 and 15 births a 4.2 confirms the bivariate relationship, with community, rather than a year.20 Hence, training under-five mortality decreasing monotonically as one TBA saves 0.05­ education increases. Multivariate analysis gener- decision based on solid 0.105 lives per year, sug- ally supports this relationship, even though the local evidence. gesting a cost of between effect may be underestimated, since the equation 2 4 I M P A C T O F S P E C I F I C I N T E R V E N T I O N S O N C H I L D H E A L T H A N D F E R T I L I T Y Secondary Enrollments Have Risen Rapidly in the F I G U R E 4 . 2 1990s: Educational Attainment of Women 17 to 24 Years Old 50 200 45 1993 180 Under-five mortality 1999 40 160 1,000) 35 140 (per women 30 120 of 25 100 20 80 mortality 15 60 Percentage 10 40 Under-five 5 20 0 0 No education Primary Secondary Higher also includes some behavioral factors that may be five mortality. The total fertility rate is around 3.3, affected by education (for example, number and so that 300 women will, over their lifetimes, give timing of birth), as well as household income birth to 1,000 children. If these 300 women receive (proxied by the wealth index). The results are secondary education (rather than primary), then somewhat mixed for primary, showing no edu- the number of child deaths averted is around 25 cation effect for neonatal and a beneficial impact (using a conservative estimate based on child of primary on postnatal, but the regression for in- deaths only). The stipend cost per girl for the five fant mortality suggests that lower secondary ed- years of secondary, including administrative costs, ucation compared with no education saves is US$90. Therefore, the For child deaths, both around 15 lives per 1,000 live births. For child cost of getting the 300 girls maternal primary and deaths, both primary and secondary education to school is US$27,000 are robust in reducing mortality. A child born to ($300 $90). The cost per secondary education are a mother with primary education is around death averted is thus robust in reducing 20 percent less likely to die than a child born to US$1,080 (=$27,000/ 25). mortality. A child born to a mother with no education; for children of This figure is an underesti- a mother with primary mothers with secondary education, this figure mate, since infant deaths is 80 percent.21 Child mortality for children of have been excluded from education is around women with no education stood at 42 per 1,000 the calculation, and various 20 percent less likely to by the end of the 1990s. Educating a mother to indirect behavioral chan- die than a child born to a primary level would reduce the probability of pre- nels are not considered. mature death from 0.042 to 0.034; and educating But it is also an over- mother with no education, her to secondary, to just 0.007, reductions of estimate for the reasons increasing to 80 percent 8 and 35 per 1,000 respectively. The gain from explained in the next for women with secondary secondary compared with primary is thus 27 child paragraph. education. lives saved per 1,000 live births. The above calculation Some approximate figures illustrate the assumes that the presence of the program is the method used to calculate the cost-effectiveness of reason the girls go to school, though in fact female secondary education in reducing under- some would have gone to school in the absence 2 5 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? of the stipend. Determining the marginal impact cause an increase in income of 15 percent,24 of the stipend on girls' enrollment is not that which translates into approximately an 11 per- straightforward. Best estimates suggest that of cent increase in the wealth index used to every 10 girls receiving the stipend, 2 of those measure economic well-being in DHS data.25 would not have gone to school in the absence of Given the regression coefficients, electrification- the stipend (see Annex I). This raises the cost induced income effects have reduced infant mor- per death averted by a factor of five, to $5,400.22 tality by around 5 per 1,000, and for children at While, perhaps unsurprisingly, more expensive around 10 per 1,000. Assigning the whole "TV ef- than the cost of direct health interventions dis- fect" adds another 8 per 1,000, and direct effects cussed above, this figure is not high by interna- another 2 per 1,000. Electrification thus reduces tional standards. the probability of under-five death by 0.025 (25 From 1993 to 1999 an additional 1.4 million per thousand). During the 1990s a household girls attended secondary school, of which one connection cost US$500,26 so that connecting million were from rural areas. Over this period, 1,000 households cost US$0.5 million. Connect- Electrification can affect the World Bank provided ing these 1,000 households will avert 25 deaths, 0.8 million stipend years, health outcomes through at a cost of US$20,000 each.27 As the program which, assuming the sti- rolls out, connections become more expensive, several channels: pend was paid to each girl currently reaching around US$1,000 per house- (1) income effects; for four years, means that hold, bringing the cost per life saved to (2) quality of health care, 0.2 million additional girls US$40,000. Rural electrification is having a mea- were paid to go to school. notably the cold chain; surable impact on child health outcomes. Since These 0.2 million girls will there are many benefits to electrification other (3) sanitary environment, go on to have 660,000 than reduced mortality, it is unsurprising that the including boiled water; children. Since 25 deaths cost per life saved is high relative to that of other and (4) access to media, are averted per 1,000 live interventions. births, then 16,500 under- and thus to health Access to media variables--particularly TV five deaths will be averted and radio--are significant in the outcome re- messages. by this increased enroll- gressions for mortality (as well as those for ment. As above, it is assumed that the whole en- fertility and nutrition). These results provide rollment increase is on account of stipends. evidence of the importance and effectiveness of Adjusting for the likely marginal impact of the pro- information, education, and communication gram on enrollments reduces the number of (IEC) campaigns. Given the diverse nature of deaths averted by the World Bank's support of the these campaigns (see Chapter 2 on the different stipend program to 3,300 (that is, just over 1 per- ministries involved), and different channels cent of the actual number of under-five deaths through which households access information, each year). it is not possible to quantify their impact and Rural Electrification and Access to Media cost-effectiveness. Electrification can affect health outcomes through several channels: (1) income effects; Fertility Reduction (2) quality of health care, notably the cold The cross-country evidence presented at the chain;23 (3) sanitary environment, including beginning of this chapter feeds directly into a boiled water; and (4) access to media, and thus debate as to the causes of fertility decline in to health messages. There is evidence of the im- Bangladesh. Some have claimed that the usual portance of each of these effects in the results, socioeconomic factors explaining lower fertility partly evident in a direct effect from the electri- have been absent and that the country's interna- fication variable in multivariate analysis. tionally renowned family planning program ex- A study of the impact of rural electrification plains the success in fertility reduction. Others conducted for USAID found electrification to argue that there has been progress--notably, re- 2 6 I M P A C T O F S P E C I F I C I N T E R V E N T I O N S O N C H I L D H E A L T H A N D F E R T I L I T Y duced poverty and a measure of women's em- Nonetheless, fertility Socioeconomic changes powerment--that can explain lower fertility.The decline has slowed, so it is do explain some of evidence presented above shows that both ex- legitimate to ask how mo- Bangladesh's fertility planations have played a part. Socioeconomic mentum may be main- changes, including the usual demographic tran- tained. The government's reduction in recent sition effect of mortality reductions driving new HNP Strategic Invest- decades--but not all of down fertility, do explain some of Bangladesh's ment Plan highlights the it. There is a substantial fertility reduction in recent decades--but not all role of increasing the age residual, which is most of it. There is a substantial residual, which is at marriage as a means of most plausibly explained by the presence of a reducing fertility, and sev- plausibly explained by the successful family planning program. eral programs, including presence of a successful Why can the unexplained fertility reduction the counseling provided family planning program. most plausibly be attributed to family planning? under the World Bank­ Evidence to support this argument comes from supported Bangladesh Integrated Nutrition Pro- a number of ways in which Bangladesh is excep- ject (BINP), promote getting married later. It is a condition of the female secondary school tional. First, figure 4.1(b) already showed the stipend program that recipients remain unmar- large reduction in fertility to be greatly in excess ried. It is true that the age at marriage in of what would have been expected from the Bangladesh is low, with half of all women marry- country's income growth over this period. ing by age 14. It is also true that there is a well- Second, figure 3.4 showed that contraceptive established international pattern whereby knowledge was almost universal in Bangladesh increasing the age at marriage drives down fer- by the early 1980s. Bangladesh is an outlier in tility. But this pattern should not be expected to this respect, meaning that contraceptive knowl- be observed in Bangladesh for two reasons: (1) edge is much higher than expected for a country raising the age at marriage of girls aged 13 or less of Bangladesh's income and female education has no effect on the age at which they have their levels. Furthermore, as discussed below, Bang- first child (as the age at marriage has risen, the ladesh is also an exception in the low level of its gap from marriage to first birth has fallen: see fertility given the low age at which women get Annex F); and (2) if a woman plans to have only married. These discrepancies are best explained three to four children, as the majority of by the presence of the family planning program. Bangladeshi women do, Finally, multivariate analysis confirms the links Fertility decline has then this can be accom- between family planning acceptance and fertility plished whether child- slowed, so it is legitimate in Bangladesh (Annex F). Less directly, the analy- bearing begins at 15 or 20. to ask how momentum sis shows the importance of access to media in The effect of expanding may be maintained. lowering family size, indicating the contribution education is muted since of IEC campaigns to spreading awareness of Bangladesh has already attained levels of fertility small family size and contraceptive knowledge. comparable to those in countries with higher The argument as to the impact of the family levels of education. Hence, raising the age at planning program in the past has significance marriage, while desirable for both maternal and to the present day, since the claimed fertility child health (children born to young mothers plateau in the 1990s has cast new doubt on the have a greater chance of premature death),28 will program's effectiveness. However, it was shown have little impact on fertility. At the same time, in the last chapter that fertility decline did in- there remains a substantial proportion of high- deed continue throughout the 1990s, with con- fertility households. Reducing fertility among traceptive usage continuing to grow. There are these would make an important contribution to valid concerns among policymakers about con- continued fertility decline. One means of doing traceptive discontinuation, but it would be a mis- this would be to attempt to restore the use of take to conclude that the program has stalled. permanent contraception to its previous level-- 2 7 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Higher-fertility women are long-term methods ac- counts for close to 10 percent of the observed less likely to be utilizing counted for 30 percent reduction in stunting. of contraceptive use in · Mother's education to secondary level im- health services, so that the 1991, but only 12 per- proves child nutrition, whereas children of case for home visits should cent by 2004.29 Higher- mothers with primary education alone have be reconsidered as a part fertility women are less no nutritional advantage over those whose of the wider strategy for likely to be utilizing mothers have no education. The expansion health services, so that fertility reduction. of secondary education accounts for about the case for home visits 6 percent of the observed fall in stunting. should be reconsidered as a part of the wider These nutritional gains can be added to the strategy for fertility reduction. In addition, beneficial impact of the Female Secondary though it is difficult to suggest appropriate pol- School Stipend Program (FSSAP). icy responses, attempts can be made to tackle · A household having electricity reduces the son preference, which creates a barrier to fertil- probability of stunting; service expansion ac- ity decline. counts for about 4 percent of the decline in stunting in the late 1990s. Nutrition · Children of lower birth order have better nu- In recent years the main focus for improving nu- tritional status, capturing the advantage of tritional outcomes has been through BINP, smaller family size in limiting resource com- which is the subject of the next chapter. How- petition among young children. The reduc- ever, BINP has had a limited geographical focus, tion in mean birth order, as a result of falling while, as shown in Chapter 3, nutritional status fertility, accounted for about 6 percent of the has been improving across the country. DHS col- improvement in nutritional status. lected anthropometric data in 1996 and 1999, · The regression results leave a fair amount of which thus allows some analysis of the factors be- the reduction in stunting unexplained. Possi- hind this improvement. OED analysis shows that: ble reasons for this decline are picked up in · Wealth is a significant determinant of a child's the next chapter, which discusses the Bang- nutritional status, and the increase in the ladesh Integrated Nutrition Project in greater wealth index between the two surveys ac- depth. 2 8 5 A Closer Look at Nutrition: The Bangladesh Integrated Nutrition Project T he Community-Based Nutrition Component of BINP was to improve nutritional status through nutritional counseling and supplementary feeding for malnourished children and pregnant women. Both coun- seling and feeding suffered from problems of inappropriate targeting strate- gies and a failure to reach intended groups. While counseling has changed women's knowledge, it has had less of an impact on behavior. There is a knowledge-practice gap, which is explained by resource constraints faced by women, including lack of time, that prevent them from putting advice into practice. The impact of the project on anthropometric outcomes has not been large, so that the approach does not appear to be a cost-effective means of tackling poor nutrition. While rapid strides were being made in re- level. This chapter takes a look at the evidence ducing mortality and fertility in the 1980s, mal- regarding the effectiveness of the BINP's main nutrition showed no improvement; it affects component, the Community-Based Nutrition close to 70 percent of all children under five. In Component (CBNC). order to address this remaining problem, the BINP, and notably the CBNC, has been the government undertook a pilot nutrition inter- subject of several studies: the independent eval- vention supported by the World Bank, the uation of BINP for the While rapid strides were Bangladesh Integrated Nutrition Project (BINP), project (Karim and others which was to initiate a national program whose 2003), a study by Save the being made in reducing ultimate goal was to reduce malnutrition to the Children (2003), an evalu- mortality and fertility in extent that it ceased to be a public health prob- ation commissioned by the 1980s, malnutrition lem. More specific targets--such as the reduc- the government's evalua- showed no improvement; tion of severe malnutrition by 40 percent, and tion department (Imple- it affects close to reduction of moderate malnutrition by 25 per- mentation, Monitoring and cent in the project areas--were also set. The Evaluation Department, 70 percent of all children project is now being scaled up to the national (IMED); Haider and others under five. 2 9 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? The key assumption 2004) and two PhD the- do" (BINP SAR, para. 1.13, p. 4; World Bank behind the Community- ses from the University 1995). Therefore, changing bad practice to good of Cambridge. For this will bring about nutritional improvements. There Based Nutrition study, OED has obtained are a number of steps in the causal chain behind Component is that "bad the BINP evaluation and this approach: practices" are responsible Save the Children data · The right people (those making decisions re- for malnutrition in sets.1 The analysis is garding undernourished children) are tar- extended from that con- Bangladesh. geted with nutritional messages. tained in the afore- · These people participate in project activities, mentioned reports through the application of a and so are exposed to these messages. theory-based evaluation framework, and it im- · Exposure leads to acquisition of the desired proves on the quality of the control for the BINP knowledge. evaluation data set by using a nationally repre- · Acquisition of the knowledge leads to its sentative sample survey to construct a new con- adoption (that is, a change in practice). trol using propensity score matching.2 · The new practices make a substantial impact Overview of the Project on nutritional outcomes. BINP had three components: (1) national nutri- A feeding program for malnourished children tion activities (US$20.6 million), including insti- and pregnant women was implemented alongside tutional development, IEC, and monitoring and growth monitoring. For this program to work: evaluation (M&E); (2) community-based nutri- tion (US$39.1 million); and (3) intersectoral nu- · The target groups have to enroll in the trition program development (US$7.6 million), program. supporting schemes such as home gardening · The criteria are correctly applied in selecting and poultry rearing. those to receive supplementary feeding. CBNC was the main component, the one ad- · Those selected for supplementary feeding dressed here. In each project thana a number of attend sessions to receive the food. community nutrition promoters (CNPs) were · There is no leakage (for example, selling of recruited--local women with children of their food supplements) or substitution (reducing own who had achieved at least an eighth-grade other food intake). education. The CNPs implemented activities at · The food is of sufficient quantity and quality to the community level: monthly growth monitor- have a noticeable impact on nutritional status. ing for children under 24 months old, supple- If project design fails to take account of one mentary feeding for malnourished children and of these steps, there is a missing link in the pregnant women, and nutritional counseling in causal chain. If activities take place correspond- a variety of settings. The CNPs were overseen by ing to a step but are ineffective, then there is a a community nutrition officer and supervisors at weak link. the thana level who were staff of the imple- menting nongovernmental organization (NGO).3 Project Coverage and Targeting The Theory Underlying the CBNC BINP nutritional counseling activities target The key assumption behind CBNC is that "bad pregnant and lactating women and adolescent practices" are responsible for malnutrition in girls.4 Clearly strategies to affect health behavior Bangladesh. This point of view was strongly ar- need to influence the attitudes of all those in- gued in the BINP appraisal document: "behaviors volved in making health decisions. In Bangla- related to feeding of young children have at least desh, decisions regarding health and nutrition as much (if not more) to do with the serious do not rest solely with the mother, but also the problem of malnutrition in Bangladesh as pov- husband, and frequently the mother-in-law. Data erty and the resultant household food insecurity from the Demographic and Health Survey show 3 0 A C L O S E R L O O K A T N U T R I T I O N : T H E B A N G L A D E S H I N T E G R A T E D N U T R I T I O N P R O J E C T that only one in five women is solely responsible traditional constraints on Strategies to affect health for decisions regarding their own health and women's mobility. The behavior need to influence that of their children, falling to only one in 10 multivariate analysis shows the attitudes of all those of women living with their mother-in-law (see that these factors do in- Annex E). While women are most likely to have deed play a role, with involved in making health responsibility over deciding what to cook (two- constraints on mobility, decisions. In Bangladesh, thirds of women are solely responsible for this particularly for women liv- decisions regarding health decision, though this is so for only 42 percent of ing with their mothers- women living with their mother-in-law), which in-law, being especially im- and nutrition do not rest clearly matters for child nutrition, the effective- portant in the two more solely with the mother, but ness of this decisionmaking power is con- conservative thanas in- also the husband, and strained by the fact that it is men who do the cluded in the survey (Raj- frequently the mother-in- shopping in Bangladesh. This is reflected by the nagar and Shahrasti). fact that only 17 percent of married women in Low growth, identified law. male-headed households are responsible for de- by growth monitoring, was cisions about daily purchases, and only 10 per- to be addressed in two ways: nutritional counsel- cent of those women living with their mother- ing and supplementary feeding. The feeding is in-law have this responsibility. Hence, there said to have been intended as an example to apears to be a need for the project to broaden mothers, the heart of the strategy being counsel- the target audience for its nutritional messages.5 ing to achieve behavior change.7 The growth mon- Similarly, just over half (52 percent) of mar- itoring sessions themselves are too chaotic a setting to provide nutritional counseling. But the ried women can visit the health center either CNPs work full time in their position and provide alone or with their children. But this percentage advice through different forums, such as the vari- falls to 39 percent for women living with their ous group counseling sessions or one-to-one mother-in-law (Annex E), showing that this do- meetings with parents. However, over one-third mestic arrangement is an important factor in of women whose children were receiving supple- restricting women's mobility. mentary feeding said that they had neither dis- Ideally all children in the project area should cussed nutrition directly with the CNP, nor sat in participate in growth monitoring, with a target any meeting where it was being discussed. set in the project appraisal document that Supplementary feeding was provided to chil- 80 percent of 0-to-24-month-olds should be reg- dren who were detected to be severely mal- istered, and 80 percent of these (64 percent of nourished (less than ­4 standard deviations all children) receive at least 18 out of 24 monthly from the reference me- weighings. The data show that over 90 percent Just over half of married dian weight for age) or of children were weighed at least sometimes, experiencing growth fal- women can visit the with 88 percent being weighed on a regular tering8 for a period of health center either alone basis.6 Hence the project's coverage targets three months, extended were met, although these allowed for some chil- or with their children. But for a further month if they dren to go unmonitored. If the children not at- did not show the required this percentage falls to tending have little risk of malnutrition, then weight gain. Analysis of 39 percent for women their omission can be seen as an efficiency gain register and field data living with their mother- for the project rather than a shortcoming. Multi- from the Cambridge stud- variate analysis shows no income bias in partici- ies showed a reasonably in-law, showing that this pation in growth monitoring, though the most low Type II error:9 only domestic arrangement is educated mothers, who presumably need it 16 percent of children re- an important factor in least, were less likely to attend (figure 5.1 and ceiving food supplemen- restricting women's Annex G). Fieldwork suggested that factors con- tation should not have straining participation include remoteness and been doing so.10 How- mobility. 3 1 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Various Factors Affect Women's Participation, F I G U R E 5 . 1 but Restrictions on Women's Mobility in More Conservative Areas Are the Most Important 1.0 0.9 Living in monitoring 0.8 Rajnagar or 0.7 Shahrasti growth in 0.6 0.5 Living with 0.4 mother-in- law in participation 0.3 of Rajnagar 0.2 or Shahrasti 0.1 Probability0.0 Base value Living with mother- Higher education No water or in-law in Rajnagar sanitation or Shahrasti (remote location) ever, of those enrolled, only one-quarter (26 per- not malnourished (that is, the children were cent) received supplementation for the recom- above ­2 SDs WAZ). mended three months, the majority dropping Turning to monitoring of pregnancy weight out sooner. And Type I error was very high: over gain, close to three-quarters of pregnant women two-thirds (69.8 percent) of eligible children attended weighing sessions, and just under half were not being fed.11 received supplementary feeding, with both these While an acceptably low percentage of ineli- percentages being a bit lower at endline than the gible children were being fed, the criteria them- baseline. There is no pattern between attendance selves are open to question. Growth faltering is at weighing sessions and the mother's nutritional quite normal, so enrolling growth faltering chil- status, which is to be expected. Supplementary dren, regardless of their nutritional level, will food was meant to be received by women not mean enrolling perfectly well-nourished chil- having normal body mass index. However, by the dren: one study found that 37 percent of a group endline, about 60 percent of eligible women were of U.S. children from well-off backgrounds not receiving the supplement. At the same time, would qualify for supplementary feeding under 40 percent of those who were receiving the sup- the criteria used in the Tamil Nadu Integrated plement were not eligible.12 For both children and Nutrition Project in India For both children and mothers, there is evi- (Martorell and Shekar dence of both leakage and substitution of the mothers, there is evidence 1992). The Save the Chil- food supplement. For example, 32 percent of of both leakage and dren register data show women said they had shared their food supple- that over 40 percent of ment with someone else. Many of those who substitution of the food those enrolled in the were not sharing said they did not eat more supplement. feeding program were during pregnancy than usual, indicating that the 3 2 A C L O S E R L O O K A T N U T R I T I O N : T H E B A N G L A D E S H I N T E G R A T E D N U T R I T I O N P R O J E C T BINP-provided food was substituting for other crease in knowledge in the There have been problems foodstuffs. This was possible, since many women project area but not in the in the targeting of feeding and children, contrary to project design, con- control. At the baseline, a sumed the food at home. programs, especially the slightly higher percentage In summary, enrollment in growth-monitor- of women in the control exclusion of eligible ing sessions has been at a reasonable level for thought it advisable to eat participants among both both children and pregnant women. However, more during pregnancy children and pregnant attendance at these sessions has not provided than was the case in the women. opportunities for nutritional counseling for a project area. This propor- sizeable minority of women. There have been tion had not changed in the control areas, whereas problems in the targeting of feeding programs, in the project area it rose by 28 percent (which is especially the exclusion of eligible participants therefore the double-difference effect, since there among both children and pregnant women. In was no change in the control). the case of pregnant women, a considerable Of course, other factors may be affecting number of feeding beneficiaries are in fact ineli- knowledge, such as superior education in the gible. Such Type II error is less of a problem for project area. Multivariate regression analysis was child feeding, though the entry criteria them- used to control for these factors (Annex G). The selves appear inappropriate, and only a minority probability of a mother knowing a specific prac- of enrolled children complete the full three tice was modeled as a function of mother's age, months of feeding. education, education of the household head, Acquiring Knowledge whether the woman was a group member, and The central thrust of the project design was to participation in project activities. In the Save the change nutritional behavior of child caretakers. Children data, variables were included in the There are a number of nutritional practices analysis of whether a woman simply lived in the considered adverse to child nutrition. Some are project area, or whether she also participated in simple differences in habit, such as cutting veg- nutrition education activities. The BINP evalua- etables before washing them rather than vice tion dataset allowed the inclusion of additional versa, which is nutritionally disadvantageous. measures of socioeconomic status; however, the Others, such as eating less during pregnancy only available measure of participation in project ("eating down"), result from different percep- activities related to supplementary feeding and tions of health risks and benefits (mothers per- attending an antenatal clinic.14 ceive the benefit of a lower-risk delivery of a In summary, knowledge of the prescribed nu- smaller child, discounting the risks to low-birth- tritional practices is higher weight children).13 And others stem from tradi- in the project areas than Data from both Save the tional beliefs that appear to have no plausible the control, the difference Children and the BINP health-related rationale, such as avoiding meat, being greatest for women evaluation show that fish, and eggs during pregnancy. who have participated in knowledge of good Data from both Save the Children and the BINP project activities. But this evaluation show that knowledge of good nutri- knowledge is not univer- nutritional practices is tional practices is indeed higher in project areas sal, which is consistent indeed higher in project than in the control. For example, 63 percent of with the evidence of weak areas than in the control. women in project areas say that the baby should links in the causal chain be fed the colostrums (the first milk), compared through incomplete enrollment, partial mistar- with only 52 percent in the control area, a differ- geting of supplementary feeding (and so the ence of 11 percent. The BINP dataset allows ex- associated counseling), and failure to transmit amination of the double difference for two counseling messages to all women whose chil- variables, for both of which there has been an in- dren have nutritional problems. 3 3 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Turning Knowledge into Practice: or that other project activities not captured The Knowledge-Practice Gap in the participation variables, for example, Although the project has had success in pro- women's group meetings, are also channels for moting nutritional information, a considerable communication of nutrition education. Accord- knowledge-practice gap remains: that is, women ing to these regression results, simply living in do not put into practice the "good behaviors" the project area raises a women's probability of (figure 5.3). This gap exists for every practice, having a piece of nutrition knowledge by 7 per- and is extremely large in the case of exclusive cent, but full participation in project activities breastfeeding. The gap exists in both project and increases this probability to between 10 and control areas, with little evidence that the gap is 23 percent (figure 5.2). That is, the propor- any less in project areas than control. tion of women aware of the importance of colostrums feeding is 23 percent greater for Although the project has The multivariate anal- ysis shows that attending women participating in project activities than for had success in promoting nutritional counseling women in the control area. Other determinants nutritional information, a indeed has a significant of a women's nutritional knowledge are found not to vary much between the project and con- considerable knowledge- impact on a woman's nutritional knowledge, trol, and so account for only a small amount of practice gap remains: that though being in receipt the difference in knowledge between the two is, women do not put into of supplementary feed- areas (Annex G). Multivariate regression analysis (Annex G) practice the "good ing does not.15 However, even when these partici- and the results of qualitative fieldwork commis- behaviors." pation variables are in- sioned by OED identify a common set of factors cluded, the BINP project dummy is still that explain the knowledge-practice gap. Re- significant. This means either that there are source and time constraints are foremost among spillover effects (women who get the knowledge these. Women who have work to do, including in nutrition sessions communicate it to others) caring for children and elderly relatives, are less Women Living in Project Areas Are More Likely to F I G U R E 5 . 2 Have Nutritional Information, Especially if They Participate in Project Activities 1.0 Control Project area Participate in project activities 0.8 knowledge 0.6 having of 0.4 0.2 Probability 0.0 Rest Colostrum Breastfeeding 3 4 A C L O S E R L O O K A T N U T R I T I O N : T H E B A N G L A D E S H I N T E G R A T E D N U T R I T I O N P R O J E C T The Knowledge-Practice Gap in Project Areas: More F I G U R E 5 . 3 Women Say They Know Good Behavior than Actually Practice It 100 Knowledge 90 Practice 80 70 mothers 60 of 50 40 30 Percentage 20 10 0 Rest during pregnancy Colostrum Exclusive Increase food intake breastfeeding during pregnancy likely to be able to rest or avoid hard work dur- of both males and females--that will also close ing pregnancy (table 5.1). Women engaged in the gap. agricultural work may also not have time to In summary, the project does increase knowl- breastfeed, or not be able to do so if they are edge about nutritional practices. However, there with the child away from the home. Although is a gap between knowledge and practice, and the the project has some effect in reducing the project does not have an impact in reducing the gap, the multivariate analysis shows the magni- size of the gap. But since knowledge is more wide- tude of this effect to be very small. There are spread, and the gap is the same in project and con- other channels--such as increasing education trol areas, then the promoted practices are more Many Factors Prevent Women from Putting Nutritional Advice into T A B L E 5 . 1 Practice, Though the Project Partially Overcomes Some of These Main determinants Project effect Moderating project impacts Rest during pregnancya Agricultural work, children, elderly male None None in household, poverty. Colostrum Children reduce gap. None None Breastfeeding Agricultural work. None Reduces effect of living with mother-in- law and being poor. More food during Poverty, children; having a vegetable Reduces gap Bigger effect in working season. pregnancy garden reduces the gap. Avoid hard work during Children, agricultural work (including Reduces gap Reduction of gap larger for poor, but pregnancy vegetable garden), but lower for female- smaller reduction during working headed households and women not in season. paid employment. a. The same results are found in both the Save the Children and the BINP data. Source: Annex G. 3 5 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? widespread in the project area. So, are these prac- feeding has a weight gain equal to 0.1 WAZ for tices beneficial to nutritional outcomes? those under 12 months old, and over 0.4 for those over one year of age. The NSP data show The Nutritional Impact of that WAZ falls by about 0.5 WAZ over a three- BINP Interventions month period for children under 12 months, and then stabilizes: hence, supplementary feed- Child Nutrition ing has a positive impact for children of this age Existing studies have found somewhat different of around 0.6 WAZ. The impact is greater still results with respect to the impact of BINP, though for children who entered the program more where an impact has been found its magnitude is malnourished. not that great. This study used the nationally rep- resentative Nutritional Surveillance Project (NSP) Low Birth Weight data to construct a control group using propen- BINP aimed to increase pregnancy weight gain sity score matching. These results find a signifi- both by encouraging women to eat more during cant project impact from BINP at midline on all pregnancy and by providing supplementary three z-scores of just over 0.1. However, at end- feeding to malnourished pregnant women. line the impacts are smaller and most are not While there was a striking improvement in preg- significant. This figure means that the project re- nancy weight gain in project areas, there was an duced malnutrition in project areas by at most even greater improvement in the control, sug- 5 percent--well short of the project target, so the gesting that non-project factors played a large project failed to achieve its central objective. role in the observed improvements. Nonethe- It is difficult to construct a suitable control for less, multivariate analysis suggests that there was supplementary feeding. However, the best esti- some project effect, of about 200 grams per mate possible suggests that a child receiving month, mainly through the likelihood that Qualitative Perspectives of the Knowledge-Practice Gap: B O X 5 . 1 The PPS-BD Study OED commissioned the Participatory Practitioner's Society­Bangladesh (PPS-BD) to conduct a qualitative study to examine the existence of traditional beliefs related to child and maternal nutrition and to detect the reasons for mothers' inability to transform health knowledge into practice. The research was carried out in two communities falling under BINP and two in non-BINP villages. In addition to focus groups and key informant interviews, semi-structured interviews were conducted with 150 individuals. These interviews began with a structured questionnaire asking similar questions to those used in the BINP evaluation to identify the presence of a knowledge-practice gap for both health and nutritional practices. Where a discrepancy between knowledge and practice was identified, unstructured questioning was used to probe the reasons for the gap. In general the study found that harmful health beliefs, such as eating less and working more during pregnancy, were on the de- cline, and were more prevalent among older women. In project areas, the reduction was mainly attributed to BINP initiatives; in the control areas, the diffusion of mass-media, national campaigns and health workers were considered the main reasons for the change. Regarding the knowledge-practice gap, women pointed to the lack of resources as the reason for their inability to eat more during pregnancy. Women also mentioned excessive workload and family opposition to the need for taking rest when preg- nant. Cost, distance, and dissatisfaction with the services offered were the reasons women stated for not seeking antenatal care. In addition, a short questionnaire was administered to a sample of 10 CNPs in each project thana in order to assess the quality of counseling and to test the CNPs' ability to interpret the growth charts used in the child-growth-monitoring sessions. These inter- views revealed a modest understanding of the charts, and great variation as to the counseling messages given in each case. 3 6 A C L O S E R L O O K A T N U T R I T I O N : T H E B A N G L A D E S H I N T E G R A T E D N U T R I T I O N P R O J E C T women in project areas would eat more during pregnancy than those in the control.16 The Cost of Nutrition project-related weight gain is thus not much T A B L E 5 . 2 Improvements and more than 1 kilogram over six months, which, Mortality Reduction (US$) evidence suggests, increases birth weight by only 20 grams. That is, the amount of pregnancy BINP weight gain appears insufficient to have a sub- Midterm (1998) Endline (2003) Rice ration stantial impact on low birth weight. Weight-for-age 187 333 110 This conclusion is supported by direct esti- Height-for-age 241 490 NA mates of project impact on low birth rate, al- Per life saved 2,328 4,095 2,223 though these suggest a slightly larger impact of about 80 grams, again mainly as a result of the mother not eating down. However, there is malnutrition. The cost (using CBNC costs associ- a much stronger impact of around 270 grams ated with children--that is, excluding those for for children born to women with poor nutri- pregnant women) per life saved from improved tional status, which is reassuring, as women nutritional status is also shown. For comparative of low body mass index (BMI) need a larger purposes, the results of a simulation assuming that pregnancy weight gain to avoid low birth the project funds were instead used to purchase weight than do women with a higher than aver- rice that was given to families with children age BMI. The project also seems to smooth out is also shown;17 this proves to be a more cost- the seasonal effects that come from mothers effective route to improving child nutrition. being less well nourished in the lean season, which is most likely a result of supplementary feeding. Other Sources of Improved Nutritional Outcomes Cost-Effectiveness Analysis Both the OED results and the BINP evaluation The impact results can be used to calculate the find a stronger impact on child nutrition at cost to move a child out of malnutrition, the re- midterm than endline. Between midterm and sults of which are shown in table 5.2. The first two endline, pregnancy weight gain was greater in rows of the table show the average amount of control thanas than those falling under BINP. The funds to be invested on a child over a year in order impact of BINP appears to have waned over time. to move the child out of malnutrition, measured One reason for this may be the difficulty of main- by WAZ and HAZ. Based on estimates of the rela- taining good imple- tionship between child malnutrition and mortal- Both the OED results and mentation over time, or ity, the last row shows the amount of money to be as the program is rolled the BINP evaluation find a spent in order to save one life. For the BINP pro- out. Unfortunately, the stronger impact on child ject, these figures were obtained using data on evaluation data set has nutrition at midterm than project costs, population of project areas, and pro- few process variables to ject effects on malnutrition rates at the midterm explore this hypothesis. endline. One reason for and the endline (see Annex G). The rice ration A second explanation is this may be the difficulty simulation uses the relationship between food in- that there were external of maintaining good take and physical growth observed in other stud- factors that allowed implementation over time, ies, and simulates the nutritional effects of non-project areas to investing BINP funds on the same areas and pop- catch up. As already or as the program is rolled ulation covered by the BINP project in 1998. The seen in Chapter 4, there out. A second explanation unit cost of saving one life and a malnourished was already an improve- is that there were external child is less under the rice ration simulation, be- ment in anthropomet- factors that allowed non- cause, given the assumptions used, this interven- ric outcomes in the late tion would produce a larger reduction in 1990s not explained by project areas to catch up. 3 7 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Rice Production and Daily Energy Supply Grew F I G U R E 5 . 4 Rapidly in the Late 1990s 45,000 2,300 40,000 2,200 35,000 Daily energy supply (right axis) 2,100 (calories) 30,000 supply 25,000 2,000 tons 000 20,000 Rice production (left axis) 1,900 energy 15,000 daily 1,800 10,000 capita 1,700 Per 5,000 0 1,600 1970 1974 1978 1982 1986 1990 1994 1998 2002 Greater food availability, the determinants of early to late 1990s, and a further 15 percent from 1999 to 2002 (Annex J). combined with rising wealth, maternal educa- tion, and other factors. Greater food availability, combined with ris- incomes and lower prices, There is a clear candi- ing incomes and lower prices, presents a very presents a very plausible date for such an external plausible explanation for improved nutritional explanation for improved factor: daily energy sup- status across the country. It is possible that at the ply grew rapidly in Bang- time of the midterm, these factors had not fully nutritional status across ladesh in the second part passed through to improved nutritional status, the country. of the 1990s as a result of allowing some effect of BINP to emerge. But by a yield-driven increase in the endline, project impacts had been damp- rice production (figure 5.4 and Annex J). Famine ened by improved energy intake in all areas, theory shows the importance of entitlements: it since the project is most effective for severely is not simply food availability that matters, but malnourished groups. whether people can afford to buy that food. In Bangladesh real incomes have been growing Testing the Theory in Practice--How throughout the 1990s, at over 5 percent a year, Well Did the Causal Chain Operate? picking up in the second half of the 1990s. Al- BINP succeeded in achieving high levels of partic- though growth in Bangladesh has not been pro- ipation in its activities in project areas, enrolled poor, real income growth has been above 2 large numbers in supplementary feeding, brought percent for the lowest income groups (World about significant increases in nutritional knowl- Bank 2004). Meanwhile the rice price has been edge, and, to a lesser extent, effected changes in stable in nominal terms, representing a real fall practice. However, nutritional outcomes in terms in the average rice price of 14 percent from the of low birth weight have been disappointing. 3 8 A C L O S E R L O O K A T N U T R I T I O N : T H E B A N G L A D E S H I N T E G R A T E D N U T R I T I O N P R O J E C T While child nutrition, especially for children par- feeding programs (see BINP succeeded in ticipating in supplementary feeding, appears to table 5.3). A substantial achieving high levels of have been better, the overall difference in perfor- knowledge-practice gap mance compared to the control is not great. persisted, so many women participation in its The longer the causal chain, the more likely it did not put the advice they activities in project areas, is that final outcomes will not be realized on ac- received into practice, count of missing or weak links in the chain, since especially if they were re- enrolled large numbers in there are more opportunities for external factors source or time con- supplementary feeding, to undermine the logical flow from inputs to out- strained. Those receiving brought about significant comes. There were two missing links in the BINP supplementary feeding chain: the first was the relative neglect of some key often shared it with others increases in nutritional decisionmakers regarding nutritional choices or substituted it for their knowledge, and, to a lesser (men and mothers-in-law), and the second was regular foodstuffs. This list the focus on pregnancy weight gain rather than of weak links in the chain extent, effected changes in prepregnancy nutritional status. Participation lev- explains why project im- practice. However, els of the target audience were high, but many pact was muted by the time nutritional outcomes women escaped exposure to nutritional mes- final outcomes were con- sages, and there was a high Type I error in the sidered. While attention have been disappointing. T A B L E 5 . 3 Links in the Causal Chain Assumption Children Mothers Attend growth- Over 90 percent of all children attend growth- Over 70 percent participate in monitoring preg- monitoring sessions. monitoring sessions. nancy weight gain. Targeting criteria is Nearly two-thirds of eligible children are not fed 60 percent of eligible women are not receiving correctly applied; partici- (reasons: don't attend growth monitoring in the supplementary feeding. pants stay in the program first place, wrong application of targeting criteria, to receive food. drop out of feeding). Acquire knowledge and One-third of mothers of children receiving supple- There is a knowledge-practice gap driven by put it into practice. mentary feeding do not receive nutritional coun- material resource or time constraints. seling. There is a knowledge-practice gap (see mothers). No leakage or One-quarter of children are fed at home, increas- One-third admit sharing food, and there is substi- substitution. ing possibility of both leakage and substitution. tution for those who do not. At most, 40 percent of eligible women receive full supplementation. Feeding and nutritional Supplementary feeding has a positive impact on Pregnancy weight gain is too little to have a no- advice have an impact on child nutritional status, especially for the most table impact on low birth weight, except for the nutritional status. malnourished children. There is only weak evi- most malnourished mothers. Moreover, mother's dence of any impact from nutritional counseling. pre-pregnancy nutritional status is more important than pregnancy weight gain. Consequently, birth weight gains are slight, though they are greater for children of severely malnourished mothers. Eating more during pregnancy is the main channel for both pregnancy weight gain and higher birth weight. 3 9 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? can be paid to each of these weak links, the BINP reducing malnutrition rates appears high, ap- experience does demonstrate the difficulty of im- parently more costly than simply buying food plementing complex designs. and giving it away, even allowing for a 25 percent Given that the nutritional gains from the proj- administrative cost to do so. The efficacy and ect are small, it does not appear to be a partic- efficiency of the program must be improved in ularly cost-effective intervention. The cost for order to justify its continued existence. 4 0 6 Lessons Learned E xternally supported interventions have outside the health sector can avert deaths at a had a notable impact on MCH-related out- relatively low cost (table 6.1). These reductions comes in Bangladesh. Immunization has in mortality are on top of the other benefits from proved particularly cost-effective, and has saved these interventions, such as higher income from the lives of up to two million children under the rural electrification and lower fertility from age of five. female education. Immunization has prevented up to two mil- lion children from dying prematurely since the Although interventions from many sectors launch of the Expanded Program of Immuniza- affect MCH outcomes, this fact need not imply tion, at a cost of less than US$200 per death that multisectoral interventions are always averted. Among the benefits that may be attrib- needed. uted to World Bank support are 200,000 lives saved through immunization and 3,300 saved This study has shown that interventions in through support for the Female Secondary one sector affect outcomes in other sectors. But School Assistance Project. they were not multisectoral interventions. The female stipend program is implemented under World Bank support to sectors outside of health the auspices of the Ministry of Education, and has contributed to better child health outcomes. electrification by rural electrification boards. There was no need for the Ministry of Health Interventions in different sectors affect ma- ternal and child health. The World Bank has sup- and Family Welfare to implement its own edu- ported both female secondary schooling and cation stipend programs or construct electricity rural electrification, both of which have reduced generators--nor even for them to be consulted under-five mortality. on the implementation of these programs by the responsible line ministries--for these inter- Small amounts of money save lives . . . though the sectoral impacts to be felt. It is important that amount varies significantly by intervention. the presence of intersectoral impacts not be confused with the need for multisectoral in- Immunization is the most cost-effective terventions or overcentralization of sectoral means of saving lives. But even interventions activities. 4 1 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? duction. However, there is a simple argument Cost­Effectiveness of why this effect will be considerably less in Interventions in Reducing T A B L E 6 . 1 Bangladesh than elsewhere--that is, fertility has Under-Five Mortality already been reduced to levels compatible with (US$ per death averted) a far greater age at marriage than currently ob- served. Based on current international norms, Intervention Lower estimate Upper estimate the average age at marriage could rise by over Immunization 100 300 two years and Bangladesh would still have lower Training TBAs 220 800 fertility than other countries with similar marital Female Secondary School 1,080 5,400 ages. While increasing the age at marriage has Stipend Program benefits for child health, its impact on fertility is BINP 2,300 4,100 Rural electrification 20,000 40,000 likely to be limited. Strategies to reduce fertility Source: OED analysis. would better target activities toward the pockets of high fertility. Gender issues are central to health strategies in World Bank support for training traditional birth Bangladesh . . . more attention is needed to attendants has reduced neonatal mortality...but redressing gender biases to maintain momentum this program has now been abandoned following in mortality decline and fertility reduction. But the international trend toward support for skilled traditional attitudes are not the absolute birth attendants. constraint on service provision as is sometimes suggested. Analysis of DHS data shows that TBA training in the 1990s reduced infant mortality. However, The study finds strong evidence of son pref- support to training TBAs stopped as the interna- erence, which both acts as a barrier to fertility re- tional Safe Motherhood Initiative chose to duction and creates excess mortality among emphasize instead the role of skilled birth atten- girls. This problem is most severe in the more dants. While training TBAs has had limited im- conservative parts of the country. Tackling gen- pact on maternal mortality in Bangladesh, as a der bias should be a central aspect of health result of weak emergency obstetric care, TBAs sector strategies. continue to attend over half of all births. Given this situation, one that is likely to take time to It is often argued that traditional attitudes change, it is unwise to make such a wholesale toward women in Bangladesh (purdah, or policy turnaround. In Bangladesh the Bank fol- seclusion of women) act as a constraint on lowed this trend and dropped the training of women's access to health services. There is TBAs from its projects, even though there was some evidence that this is the case, which un- no local evidence to support such a decision. derlines the importance of reaching all decision- makers with behavior-change communication. Programs should be based on local evidence, Men have limited exposure to existing health rather than general conventional wisdom. and nutritional counseling, so additional chan- nels should be sought, such as through religious The abandonment of training TBAs is an ex- authorities. However, the extent of these re- ample of the general finding of the importance strictions can be overstated. They are strongest of taking context into account in program in the most conservative areas, which is where design. efforts are likely to be more effective. A second example is the effect that raising the age at marriage can have on fertility. There is a The Bank's BINP has improved nutritional status, well-established relationship by which an in- but by much less than planned. Either the crease in the age at marriage drives fertility re- efficacy and efficiency of the program must be 4 2 L E S S O N S L E A R N E D improved, or alternatives to scaling up should creasing the size of the food supplement in be considered. this period, restricting it to those months, or adjusting the eligibility criteria by time of year. The limited impact of BINP and its cost raises · Discouraging women from eating down dur- serious doubts as to the justification for scaling ing pregnancy has some benefit for birth the project up to the national level in its current weight. But all forms of knowledge transmitted form. This limited impact points to either a fail- by the project suffer from a knowledge- ing in design, in implementation, or both, and practice gap, so attention needs to be paid to hence shortcomings in Bank performance. In both the resource constraints that create this order for the project to be justified, its efficacy gap and transmitting knowledge to other key and efficiency must be improved. In seeking to actors: mothers-in-law and husbands. do this, several lessons emerge from this analy- sis which should be borne in mind: However, based on current performance, scaling up will prove very costly, with limited nu- · Supplementary feeding for children does tritional gain. If efficiency cannot be improved have a positive impact, especially for the most through better implementation, then there are malnourished children. NNP has revised the other ways of spending the money that would eligibility criteria so that only growth-faltering have a greater nutritional impact, and other children with WAZ<­2SDs receive feeding, as channels could be sought for conveying nutri- well as those with WAZ<­4 SDs. The latter tional information, tailoring it to the differing group constitutes a very small percentage of needs of different socioeconomic groups. children, so there appears to be scope for raising this threshold to, say, ­3 SDs, to cap- Rigorous impact evaluation can show which ture more children for whom feeding seems government programs and external support most successful. There is also some mistar- are contributing most to meeting poverty- geting, which is likely to be best addressed by reduction goals. further training of community nutrition pro- moters in interpreting growth charts. This study has demonstrated the impact of · Supplementary feeding for pregnant women selected interventions and strategies on mater- appears to be a flawed approach on two nal and child health outcomes. These results grounds: (1) the pregnancy weight gain show that impact evaluation can be used to achieved is mostly too small to have a notable quantify the contribution of selected activities impact on birth weight; and (2) it is pre- toward meeting the MDGs' and countries' own pregnancy weight that evidence suggests to poverty-reduction goals. It can show which in- be the more important determinant of birth terventions work and which do not, or which are weight. The program would be more suc- more cost-effective than others. cessful if it restricted its attention to the most malnourished of women, improved targeting National surveys can be used for evaluation to reduce Type II error, and made additional purposes, but some adaptations would make efforts to discourage leakage and substi- them more powerful--notably, a more detailed tution. However, the fact that it is pre- community questionnaire. pregnancy weight that matters suggests that a different approach altogether ought perhaps This study relied on the Demographic and be considered, such as school feeding pro- Health Survey to evaluate the impact of selected grams or targeting adolescent females in interventions on maternal and child health out- poorer areas. comes. Was this a successful approach to impact · For both types of feeding programs, there is evaluation? The results reported above show evidence of a greater impact in the lean sea- that this approach can be used for certain inter- son. There are grounds for considering in- ventions. But it could have been used to exam- 4 3 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? ine a broader range of interventions, and to macies, traditional birth attendants, NGO pro- unpack the causal chain in greater detail, if grams, and government health workers and the more detailed information had been available. frequency of their visits. These data would aid in The contrast can be made with the analysis of the investigation of changing patterns of health BINP, for which the availability of a broader usage. The addition of these instruments is the range of data allowed the discussion of process easiest, and probably least costly, change to make. issues, and so examination of the whole causal However, some modifications of the women's chain. questionnaire would also be helpful in analyzing The use of DHS to evaluate HPNSP is being patterns of health-seeking behavior, including discussed. The experience of undertaking this use of nongovernment services and problems study has some lessons for doing this. The first experienced in accessing services. Seeking mod- is that a more elaborate community survey ifications of an existing survey instrument in this should be used, including a health facility sur- way will prove a more cost-effective approach to vey.1 The community survey should record in- evaluation than undertaking separate, stand- formation on the various sources of health care alone surveys, and build institutional ownership within the community, including private phar- of data and results within Bangladesh. 4 4 ANNEX A: TRENDS IN MATERNAL AND CHILD HEALTH OUTCOMES Since the early 1970s, fertility rates, infant and (1979: 82­83) argued that the rate reported in the child mortality, and malnutrition rates have de- WFS is likely to be low, because of the under- clined substantially, according to the World De- reporting of births during or in the aftermath of velopment Indicators (WDI) of the World Bank. the independence war in the early 1970s. Since However, because social indicators are often un- the 1980s, when the largest reduction in the TFR reliable as a result of weak data collection sys- occurred, the various sources largely agreed with tems, it is necessary to ask about the quality of one another. There is no doubt that a profound the data. In Bangladesh many surveys provide demographic change take place in Bangladesh high quality, nationally representative data, in- cluding the Demographic Health Survey (DHS), Malnutrition Contraceptive Prevalence Survey (CPS), Malnutrition, as measured by weight-for-age Bangladesh Fertility Survey (BFS), Bangladesh z-scores (WAZ) and height-for-age z-scores (HAZ), Retrospective Survey of Fertility and Mortality dropped from an estimated 70 percent in 1983 (BRSFM), World Fertility Survey Bangladesh to about 45 percent in 2000 (figure A.2). These (WFS), the Helen Keller International Health and figures, reported by both WHO and WDI, are Nutritional Survey (HNS), and Health and Demo- based on many sources, including DHS, child graphic Survey (HDS), and Maternal Mortality nutrition surveys by the Bangladesh Bureau of Survey (reported in Streatfield and others 2003). Statistics, and various surveys by Helen Keller This annex compares the WDI data with these International (HKI). Even so, comparing this with sources to check data quality and consistency. Fertility Total Fertility Rate, The Total Fertility Rate (TFR) indicates the ex- F I G U R E A . 1 1960­2000 pected number of births per woman during her lifetime, if she lived out her childbearing years. 8 Between 1970 and the early 1990s, the TFR in Other sources Bangladesh dropped by 50 percent (figure A.1). 7 WDI During the 1990s, however, the trend plateaued; 6 the estimated TFR remained virtually unchanged between 1993/94 and 1999/2000. This stagnation 5 has caused major concern among policymakers. 4 However, this report suggests that fertility con- 3 tinued to fall in the 1990s (see Annex F). The TFR reported in the WDI differs little from 2 those in national surveys (figure A.1). The higher 1 rate in the early 1970s reported by the latter data 0 sources reflects the different rates reported in the 1960 1965 1970 1975 1980 1985 1990 1995 2000 BRSFM for 1974, and the WFS, which is an aver- age of the TFR between 1971 and 1975. The CPS Sources: WDI (2004), BRSFM, CPS. 4 5 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? greatest since the 1980s. Figure A.4 shows that F I G U R E A . 2 HAZ & WAZ 1983­2000 the rate of under-five mortality fell from 250 per 1,000 live births in 1960 to less than 75 per 1,000 in 2002. That is a reduction in infant mortality by 80 more than two-thirds (from 150 per 1,000 to 70 WAZ (WDI) about 45 per 1,000 live births) and a reduction in 60 child mortality by more than three-quarters 50 (from about 115 per 1,000 to 25 per 1,000 chil- HAZ (WDI) 40 dren surviving to age 1 year). The mortality series Percent30 are plotted alongside the change in fertility to 20 show that the decline in mortality occurred 10 about a decade earlier than that of fertility. These trends are consistent with the demographic tran- 0 1984 1986 1988 1990 1992 1994 1996 1998 2000 sition model, which predicts that improvements in nutrition/health and mortality rates lead to re- Source: WDI (2004). ductions in fertility, though with a lag. Figure A.5 plots infant, child, and under-five HKI data for WAZ in the 1990s (figure A.3) reveals mortality rates presented in the DHS Reports a discrepancy; the latter reports a malnutrition (1994, 1997, 2001), along with OED's calcula- prevalence 5 to 10 percent higher between 1991 tions based on pooling all three sets of data. and 1998. However, the HKI survey was not na- Each data point is plotted midway through the tionally representative until 1998. Its orientation five-year period. Rather than estimate mortality initially was disaster-prone areas, where people for each round, OED's calculations pool the are more likely to suffer from malnutrition. Thus, data from the three survey rounds; pooling the HKI data overstate the national level for this makes estimates more efficient by increasing earlier period, and an "end-to-end" comparison sample size. (Otherwise, the number of deaths would overstate the downward trend. on which the mortality calculations are based is relatively low.) Mortality Both methods calculate mortality using the Childhood mortality has dropped appreciably synthetic cohort probability model, (Rutstein since the 1960s, though the decline has been 1984), which examines a hypothetical cohort subject to the age-specific mortality rates of one particular period; unless mortality is constant (or F I G U R E A . 3 WAZ, 1990s the period of analysis is long), the same person will not experience age-specific rates observed in one period during his or her life. The advan- 80 tage of the synthetic cohorts method is that mor- 70 HKI tality rates are more sensitive to period-specific 60 events such as natural disaster and economic 50 WDI crisis, affecting all cohorts simultaneously. 40 Although the two data series are similar, the es- Percent timates based on the pooled sample (the dashed 30 lines in the figure, denoted by ^) on average 20 overestimate mortality. This difference is proba- 10 bly because the calculation of mortality rates uses 0 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 longer recall periods. The events therefore are recorded closer to the present than they occurred Sources: WDI (2004), HKI. (that this, "forward telescoping"). 4 6 T R E N D S I N M A T E R N A L A N D C H I L D H E A L T H O U T C O M E S Fertility and Mortality: The Demographic Transition F I G U R E A . 4 in Bangladesh 300 8 7 250 TFR U5MR 6 1,000) 200 5 rate (per 150 IMR 4 rate fertility 3 100 Total CMR 2 Mortality 50 1 0 0 1960 1965 1970 1975 1980 1985 1990 1995 2000 Source: WDI (2004). F I G U R E A . 5 Mortality Trends 200 U5MR^ 175 U5MR 150 IMR^ 125 IMR 100 75 CMR CMR^ 50 25 0 1981 1983 1985 1987 1989 1991 1993 1995 1997 Note: ^Calculated from pooled DHS data. Sources: Mitra and others (1994, 1997), NIPORT and others (2001); authors' own calculations based on DHS data. 4 7 ANNEX B: CROSS-COUNTRY ANALYSIS OF CHILD HEALTH AND NUTRITION OUTCOMES Some initial insights regarding the main determi- of children who are underweight, using a cross- nants of child health and nutritional outcomes country data set culled from various sources. can be sought from cross-country regressions. The first model regresses nutritional status on This is not a novel area of analysis. Work ema- what they call the "basic determinants" of per nating from the World Bank (Reutlinger and capita GDP and democracy (measured by the Selowsky 1976; Berg 1983), and followed up more Freedom House political freedom index). The recently in a series of papers by Haddad and oth- second model, of underlying determinants, in- ers (e.g., Smith and Haddad 2000), has shown cludes access to safe water, female secondary that income growth translates into improved nu- school enrollments; the ratio of male to female tritional status, but at too low a rate to achieve in- life expectancy, as a measure of gender in- ternational targets. Concerning mortality, there equality; and per capita dietary energy supply. has also been debate about the respective role of The logic of the approach is that the latter var- economic growth versus direct support for lower iables, which directly affect nutritional status, mortality via health service provision (e.g., Filmer depend, in turn (though only in part), on the and Pritchett 1999 and the debate around the "basic determinants." The basic determinants World Health Report 2000; see Pedersen 2002). equation is thus a reduced-form equation, This annex presents cross-country regression though one that appears to suffer from omitted results drawing on two data sources. The first set variable bias, as there are other determinants of data updates the work of Haddad and Smith, of the underlying variables than GDP and de- using the same sources and series, but including mocracy alone. Using the results from a fixed-ef- more recent data. Their main results are repli- fects estimate of their underlying determinants cated and extended through the addition of model, they examine the relative importance of further variables and a closer look at the non- the different variables in the explained change in income drivers of changes in nutritional out- nutrition. Female education comes out as the comes, specifically for the case of Bangladesh. DHS data, customarily used for the sort of most important factor (43 percent), followed by household modeling presented in Annexes C food supply (21 percent), then health environ- and D, can also be used for cross-country anal- ment (safe water, 19 percent), and finally ysis using national averages of the various women's status (11 percent). variables, which are provided on the Macro In- The analysis for this report has updated their ternational Web site. This second set of esti- data set with five modifications: (1) the analysis mates is provided to corroborate the first set of is carried out for four dependent variables: HAZ, results, and as some additional variables can be WAZ, under-five mortality, and the total fertility included. rate; (2) more recent data have been obtained; (3) rather than using data for specific years, Drivers of Maternal and Child Health decade averages are used for the 1970s, 1980s, and Nutritional Outcomes in Bangladesh 1990s, and the current decade; (4) female liter- Smith and Haddad (2000) estimate two sets of acy is used rather than secondary enrollment, as models for the determinants of the proportion the stock of female education appears more 4 9 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? appropriate than the new flow into that stock, example, under-five mortality, attributable to the which is what the enrollment rate measures; income effect. This effect is highest in the case (5) income inequality has been added to the of mortality, in which case it is just under one- data set, although this variable restricts sample third, and lowest in the case of fertility, where it size so that estimates are presented with and is just 16 percent. without the variable. As in Smith and Haddad, re- The same argument may be made graphically, sults are presented from both ordinary least as is done in figure B.1. The four graphs in figure squares (OLS) and fixed effects (FE) models, B.1 show this clear link between income and using time dummies in the latter and regional four social outcomes--under-five mortality, the and time dummies in the former. total fertility rate, and the prevalence of stunting The results from the "basic determinants" and underweight among under-fives--using the model demonstrate the extent to which income cross-section data. growth has been behind improvements in ma- In the 1980s, Bangladesh lay above the fitted ternal and child health and nutrition outcomes. line (the solid line in each figure) for under-five There are both direct and indirect channels from mortality and fertility, meaning that those indi- income to better health and nutrition. The direct cators were worse than should be expected for a channel is from the higher consumption of food country at its income level. If these indicators and health services made possible by higher in- had improved following the internationally es- come. Indirect channels include increasing both tablished relationship with income, then subse- the supply and demand for education, improved quent observations for Bangladesh would lie water supply, and other facilities. To the extent along the dashed line. But these later observa- that factors positively affecting health and nutri- tions lie below the fitted line, showing that tion outcomes are positively correlated with in- Bangladesh now does better than expected for a come, but not wholly explained by income, then country at its income level. the regression of any one outcome on income In the case of the nutrition indicators, the ob- will provide an upward biased estimate of the in- servations from all three decades lie above the come effect, which may be regarded as an fitted line. Bangladesh continues to have worse "upper limit" estimate of this effect. nutritional outcomes than the average for coun- Table B.1 shows the extent to which income tries at a similar income level. But this discrep- growth in Bangladesh has determined the ancy has narrowed over time. As for mortality change in the four outcomes being considered and fertility, the rate of progress in nutritional here. The results are based on the model 1 OLS status exceeds that to be expected from income estimates for each variable, shown in tables growth alone. B.2(a)­B.5(a). Multiplying the coefficient on in- Estimation found that inequality affects nu- come by the change in Bangladesh's (log) in- tritional outcomes once an interactive term is in- come gives the change in the outcome--for cluded. Table B.6 shows some simulations to Income Growth Accounts for at Most One-Third of the Reduction in T A B L E B . 1 Mortality in Bangladesh . . . And Less Than a Fifth of Lower Fertility 2000 income- Percent reduction 1980 actual 2000 actual based estimate explained by income Under-five mortality 205.0 77.5 163.1 32.9 Total fertility rate 5.6 3.0 5.2 16.0 Stunting 67.6 44.7 62.3 23.1 Underweight 69.5 47.7 64.6 22.4 Source: Calculated from data used for figure B.1. 5 0 C R O S S - C O U N T R Y A N A L Y S I S O F C H I L D H E A L T H A N D N U T R I T I O N O U T C O M E S Bangladesh's Improvement in Social Outcomes Is Greater Than Can F I G U R E B . 1 Be Explained by Economic Growth Alone (a) Under-five mortality (b) Total fertility rate 400 9 ths) bir 350 8 live 7 300 000 rate 6 250 1980s (per 5 200 1980s fertility 4 tality 1990s Income-driven trajectory 150 3 mor 1990s Total 2000s 100 Fitted line Income-driven trajectory 2 50 2000 Fitted line 1 Under-five 0 0 0 2,000 4,000 6,000 8,000 10,000 12,000 14,000 16,000 18,000 0 2,000 4,000 6,000 8,000 10,000 12,000 14,000 16,000 18,000 GDP per capita (US$) GDP per capita (US$) (c) Stunting (HAZ) (d) Underweight (WAZ) 90 90 80 80 age) 70 1980s 70 1980s age) for 60 1990s 60 for 1990s 50 50 2000 Income-driven trajectory 2000 (weight 40 (height 40 Income-driven trajectory 30 30 20 Stunting 20 10 Fitted line Fitted line Underweight 10 0 0 ­10 0 2,000 4,000 6,000 8,000 10,000 12,000 14,000 16,000 18,000 0 2,000 4,000 6,000 8,000 10,000 12,000 14,000 16,000 18,000 GDP per capita (US$) GDP per capita (US$) help interpret this result. First note, that for any nutrition, while others are pushed further into given level of inequality, HAZ falls as income food deprivation. rises, which is to be expected. However, HAZ is The next step in the analysis is the estimation also decreasing in inequality up to a certain level of the underlying determinants model. While of income (log GDP of 7.7, or just over $2,200 Smith and Haddad look at the contribution of per person), after which level inequality is detri- the four variables to the estimated change in the mental to nutrition. The most likely explanation data set as a whole, the estimates here are for the for this result is that at lower income, the aver- contribution to the actual change just for the age energy supply is below the minimum re- case of Bangladesh. The "unexplained change" is quired to avoid malnutrition, so that if resources captured through the residual. In addition, a fur- are equally distributed, everyone is malnour- ther explanatory variable (immunization) is ished. With inequality, at least some escape mal- added to the under-five mortality regressions 5 1 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Height for Age Z-Score HAZ Fixed TABLE B.2(a) TABLE B.2(b) (HAZ), OLS Effects Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. t Coeff. t Coeff. t Coeff. t Coeff. t Log GDP per ­13.16*** ­54.19*** ­7.28*** ­33.23*** ­23.40** ­9.88*** ­15.77*** ­55.77*** ­11.54*** ­35.11*** capita (­12.72) (­5.23) (­3.53) (­2.88) (­2.16) (­4.73) ­5.13 ­3.15 ­3.06 ­2.24 Democracy ­0.84 ­0.18 ­0.59 0.00 ­0.38 ­0.67 0.10 ­1.54 0.94 0.47 (­1.54) (­0.20) (­0.91) (0.00) (­0.40) (­1.10) 0.16 ­1.19 1.47 0.41 Adjusted Gini ­5.89*** ­0.08* ­4.38** ­2.16 ­4.93 ­3.89 coefficient (­3.75) (­1.65) (­2.63) (­1.33) ­1.54 ­1.43 Gini*GDP 0.77*** ­0.03 0.56** 0.28 0.61 0.44 (3.73) (­0.67) (2.59) (1.33) 1.50 1.29 Safe water ­0.18*** ­81.55*** ­0.18** ­0.19** ­0.10** 0.01 0.00 ­0.11* ­3.76 (­3.01) (­2.48) (­2.46) (­2.13) 0.13 0.07 ­1.98 Female ­0.10*** ­0.01*** ­0.10* ­0.14** ­0.01 ­0.30*** ­0.23*** 0.05 literacy ­2.66 (­3.23) (­1.65) (­2.37) (­0.17) ­3.91 ­2.99 0.26 Male:female ­99.72*** ­66.80 5.14 ­26.33 ­63.03 ­30.87 ­41.88 life expectancy ­3.64 (­1.80)** (0.13) (­0.97) ­1.43 ­0.72 ­0.78 Daily energy ­0.01*** ­0.01** ­0.01 ­0.01*** ­0.02*** ­0.01*** ­0.02** supply ­5.46 (­2.03) (­1.56) (­3.36) ­4.15 ­3.08 ­2.36 Sub-Saharan ­10.55*** ­4.88** Africa (­3.44) (­2.42) Middle East ­6.80* & N. Africa (­1.87) South Asia 9.79*** (3.43) 1980s ­7.83** (­2.18) 1990s ­4.47 (­1.21) Intercept 134.65*** 448.42 189.19*** 206.75*** 402.11*** 255.56*** 168.17 151.19*** 478.68*** 153.25*** 190.50*** 418.78** 18.02 5.75 6.91 7.70 4.39 2.82 6.38 6.45 3.44 3.38 4.33 3.37 No. of 196 96 152 84 84 152 96 152 84 observations Note: *, **, *** significant at greater than 10 percent, 5 percent, and 1 percent respectively. and under-five mortality itself appears as an ex- water, female literacy, and greater gender equal- planatory variable in the fertility regression. The ity have all contributed to the observed im- results are summarized in table B.7. provements. The included variables do less well In the case of Bangladesh, the model does at explaining weight for age and fertility, ac- best in explaining the reduction in under-five counting for only about one-third of the ob- mortality and height for age, with the included served decline. That means that there have been variables accounting for about 80 percent of the other factors, not included in the model, behind observed improvement. Higher immunization improved outcomes. In the case of fertility, this coverage has been a major factor in lower mor- analysis feeds directly into a debate about the tality in Bangladesh, accounting for close to one- extent to which reduced fertility is a result of third of the reduction. For both variables, safe changing socioeconomic conditions. The results 5 2 C R O S S - C O U N T R Y A N A L Y S I S O F C H I L D H E A L T H A N D N U T R I T I O N O U T C O M E S Weight for Age Z-Score WAZ Fixed TABLE B.3(a) TABLE B.3(b) (WAZ), OLS Effects Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. t Coeff. t Coeff. t Coeff. t Coeff. t Log GDP per ­12.57*** ­48.47*** ­5.37** ­28.68** ­7.40*** ­23.93** ­10.68*** ­36.99** ­5.60** ­17.79 capita ­11.73 ­4.53 ­2.48 ­2.29 ­3.83 ­2.17 ­4.82 ­2.64 ­2.16 ­1.21 Democracy ­0.13 1.42*** 0.28 0.84 ­0.04 0.27 ­0.21 ­0.93 0.73 ­0.31 ­0.25 1.45 0.42 0.80 ­0.06 0.29 ­0.45 ­0.94 1.56 ­0.31 Adjusted Gini ­5.26*** ­4.19** ­2.74* ­2.55 ­1.80 coefficient ­3.26 ­2.32 ­1.68 ­1.01 ­0.71 Gini*GDP 0.65*** 0.49** 0.34 0.35 0.19 3.05 2.07 1.59 1.09 0.61 Safe water ­0.13*** ­0.07 ­0.09 ­0.09** ­0.11* ­0.02 ­0.02 ­0.02 ­2.66 ­1.34 ­1.09 ­2.05 ­1.64 ­0.70 ­0.51 ­0.45 Female ­0.10*** ­0.08* ­0.08 ­0.05 ­0.02 ­0.14** ­0.13** ­0.08 literacy ­2.63 ­1.69 ­1.17 ­1.18 ­0.37 ­2.32 ­2.02 ­0.44 Male:female ­125.87*** ­100.89*** ­101.58** ­38.40 ­37.04 ­30.10 ­24.14 ­83.91 life expectancy ­4.51 ­3.59 ­2.51 ­1.42 ­0.90 ­0.92 ­0.78 ­1.69 Daily energy ­0.01*** ­0.01*** ­0.01*** ­0.01 ­0.01** ­0.02*** ­0.01*** ­0.01* supply ­4.85 ­3.11 ­2.69 ­4.09 ­2.60 ­5.58 ­4.90 ­1.92 Sub-Saharan ­4.53*** ­3.72 Africa ­2.26 ­1.15 South Asia 17.13*** 16.09*** 6.28 4.34 Europe and 9.12* 7.21 central Asia 1.64 0.88 Intercept 119.35*** 405.22*** 202.79 202.74*** 414.92 155.39*** 290.41*** 105.06*** 307.78 104.58*** 130.96*** 307.79** 7.73 5.02 7.27 5.71 3.06 6.25 2.78 3.06 3.87 2.65 No. of obs. 208 99 165 159 86 159 86 99 165 159 86 Note: *, **, *** significant at greater than 10 percent, 5 percent, and 1 percent respectively. do not support the idea that changing condi- to national estimates of infant and child mortal- tions have played no role, but this role does ap- ity and nutrition. Although DHS data do not in- pear to be relatively minor (with the main clude income, it has become common practice change coming from under-five mortality). to create an asset index from information on There is a large unexplained residual, for which ownership of a list of consumer durables, and the most likely candidate is family planning sometimes also housing quality and education. programs. For this analysis, an asset index was made by tak- ing a simple average of the percentage of house- Estimates Using DHS Data holds owning the assets included for that Demographic Health Surveys (DHS) are con- country. This simple measure appears to be a ducted under the auspices of Macro good proxy for income, as shown by the clear International. The company's Web site allows negative relationship with nutritional depriva- data to be downloaded from 126 surveys in the tion and mortality; see, for example, the scatter form of national averages--for example, the per- plot for the percentage of children stunted centage of women in the survey age range (usu- (figure B.1). The simple correlation coefficient ally 15-50) having primary education, in addition between the asset index and the outcomes mea- 5 3 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Under­Five Mortality Under­Five Mortality TABLE B.4(a) TABLE B.4(b) (U5M), OLS (U5M), Fixed Effects (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Log GDP per ­0.71*** ­0.65*** ­32.38*** ­24.84*** ­16.57* ­1.00*** ­1.66*** 13.92 19.19 13.52 capita ­14.45 ­9.11 ­4.45 ­3.54 ­1.71 ­6.37 ­5.58 1.08 1.53 0.58 Democracy ­0.08*** ­0.11*** 4.98** 3.11 ­3.17 ­0.05 ­0.07 ­2.12 ­2.62 ­3.60 ­3.27 ­3.00 2.45 1.64 ­1.19 ­1.55 ­0.98 ­1.02 ­1.29 ­0.93 Adjusted Gini 0.02*** 1.11** 0.01 ­0.47 Coefficient 3.02 2.52 0.31 ­0.39 Safe water ­0.92*** ­0.65*** ­0.40** ­0.10 ­0.33* ­0.30* ­0.18 0.12 ­5.04 ­3.25 ­2.17 ­0.43 ­1.85 ­1.72 ­1.00 0.47 Female ­1.34*** ­1.06*** ­1.10*** ­1.05*** ­1.33*** ­1.27*** ­0.77* ­1.44 Literacy ­9.62 ­6.46 ­6.62 ­5.34 ­4.21 ­3.61 ­1.98 ­1.04 Male:female ­100.67 25.34 ­171.70** ­71.48 ­277.36** ­308.70** ­378.51*** 81.01 life expectancy ­1.13 0.29 ­2.05 ­0.70 ­2.02 ­2.31 ­2.88 0.44 Daily energy ­0.03*** ­0.01 0.00 0.00 ­0.04*** ­0.04*** ­0.03** ­0.03 supply ­4.00 ­0.90 ­0.34 ­0.40 ­2.98 ­3.07 ­2.12 ­0.79 Immunization ­0.20 ­0.40** ­0.35* ­0.83*** ­0.55*** ­0.65*** ­0.32 ­0.79 ­1.04 ­2.16 ­1.75 ­3.39 ­3.35 ­3.87 ­1.54 ­2.34 Sub­Saharan 23.37*** 18.94** Africa 3.57 2.30 South Asia ­29.05*** ­14.28 ­2.98 ­1.34 Middle East & ­15.87 ­16.40 North Africa ­1.55 ­1.51 1980s ­12.11* 4.09 ­18.62** ­0.73 ­1.78 0.51 ­2.50 ­0.06 1990s ­12.46 ­26.15*** ­1.65 ­2.72 Intercept 10.23*** 9.15*** 451.24*** 465.86*** 595.32*** 397.75*** 12.37*** 17.43*** 638.48*** 575.33*** 536.75*** 136.90 29.08 16.42 5.02 5.44 7.31 4.16 10.42 7.54 4.30 3.59 3.46 0.45 No. of obs. 212 149 145 145 76 149 145 76 Note: *, **, *** significant at greater than 10 percent, 5 percent, and 1 percent respectively. sures used here ranges from ­0.47 (weight for ables were identified using two criteria: (1) rela- height) to ­0.65 (weight for age). tively low correlation with other health variables, DHS includes a large number of variables re- and (2) being available for most or all of the lated to access to, and use of, health services. 126 observations. The four selected were: per- These measures are correlated, of course, so centage of women using modern contraceptives that including them individually in the regres- (which partly proxies for access to health sup- sions creates a multicollinearity problem. To get plies), the percentage of assisted deliveries around this problem, an index of health services (which gives some idea of the availability of was constructed using four variables. The vari- trained health workers), and, as measures of 5 4 C R O S S - C O U N T R Y A N A L Y S I S O F C H I L D H E A L T H A N D N U T R I T I O N O U T C O M E S T A B L E B . 5 ( a ) Fertility (OLS) (1) (2) (3) (4) (5) (6) (7) (8) (9) GDP per capita (logged) ­1.01*** ­1.33*** ­2.88*** 0.12 ­0.06 ­0.08 ­0.05 ­7.28 ­7.15 ­2.75 0.48 ­0.23 ­0.20 ­0.22 Democracy ­0.06 ­0.13 ­0.12 0.12 0.04 0.11 0.13* ­0.89 ­1.39 ­1.25 1.70 0.46 1.00 1.78 Inequality 0.05*** ­0.19 0.03** 0.03 3.02 ­1.21 2.30 1.28 Inequality x GDP 0.03 .50 Safe water ­0.02*** ­0.01 ­0.02*** ­0.02*** ­0.02** ­0.02** ­3.84 ­0.93 ­4.08 ­2.81 ­2.44 ­2.53 Female literacy ­0.02*** ­0.01 ­0.03*** ­0.03*** ­0.03** ­0.02*** ­5.53 ­1.35 ­5.27 ­4.92 ­2.47 ­3.53 Female:male life expectancy 0.78 1.10 1.62 ­1.62 ­6.88 1.32 0.26 0.24 0.52 ­0.49 ­1.28 0.38 Daily energy supply 0.00*** 0.00 0.00*** 0.00** 0.00 0.00 ­3.29 ­1.43 ­2.89 ­2.24 ­1.31 ­2.38 Under-five mortality (lagged) 0.01* 0.00 1.77 0.63 South Asia ­0.19 ­0.51 Sub-Saharan Africa 0.75*** 2.85 Middle East and North Africa 0.66* 1.80 Europe and Central Asia ­1.42** ­2.08 1980s ­0.43 ­1.42 1990s ­0.84*** ­2.64 2000 ­1.41*** ­4.19 Intercept 12.64*** 13.29*** 24.95*** 8.69*** 4.80 7.00** 10.17*** 14.83*** 7.69** 12.63 9.16 3.16 2.94 1.01 2.28 3.23 3.02 2.19 Note: *, **, *** significant at greater than 10 percent, 5 percent, and 1 percent respectively. 5 5 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? T A B L E B . 5 ( b ) Fertility (Fixed Effects) (10) (11) (12) (13). (14) (15) (16) (17) GDP per capita (logged) ­0.81 ­3.35*** ­5.05** 1.24* ­0.48 ­0.15 ­1.39 ­4.86 ­2.15 1.91 ­0.91 ­0.26 Democracy ­0.19 ­0.14 ­0.17 0.18 ­0.04 ­0.02 ­1.58 ­0.86 ­0.99 1.55 ­0.37 ­0.18 Inequality 0.06 ­0.26 0.02 0.02 1.42 ­0.61 0.62 0.80 Inequality x GDP 0.04 0.76 Safe water 0.00 0.00 ­0.01 0.00 0.00 0.61 0.28 1.05 ­0.79 0.42 Female literacy ­0.07*** ­0.07*** ­0.09*** ­0.10*** ­0.07* ­5.26 ­3.65 ­5.86 ­3.23 ­1.92 Female:male life expectancy 6.27 7.33 5.48 ­4.07 4.33 0.82 0.94 0.71 ­0.76 ­0.78 Daily energy supply 0.00** 0.00* 0.00*** 0.00 0.00 2.64 ­1.78 3.24 1.16 0.85 Under-five mortality (lagged) 0.00 0.01 0.00 0.84 0.92 0.78 South Asia Sub-Saharan Africa Middle East and North Africa Europe and Central Asia 1980s ­0.26 ­0.88 1990s 0.60 ­1.22 2000 Intercept 11.53** 28.31*** 41.81** 7.06 3.85 ­0.12 15.43** 12.03 2.63 5.27 2.24 0.89 0.43 ­0.01 2.25 1.61 Note: *, **, *** significant at greater than 10 percent, 5 percent, and 1 percent respectively. 5 6 C R O S S - C O U N T R Y A N A L Y S I S O F C H I L D H E A L T H A N D N U T R I T I O N O U T C O M E S antenatal services, the proportion of women having no antenatal services at all and the pro- HAZ for Different portion receiving no tetanus vaccination. The T A B L E B . 6 Combinations of Income weights were derived from principal compo- and Inequality nents analysis, the last two variables entering with negative weights as should be expected. Gini coefficient 30 40 50 60 70 Choice of Regressors 6.0 84.8 71.9 59.1 46.2 33.4 The approach adopted for the nutrition regres- 6.5 69.1 60.1 51.1 42.1 33.1 sions assesses the robustness of the coefficients 7.0 53.4 48.3 43.1 38.0 32.9 of the various variables. Table B.8 shows the sim- 7.5 37.6 36.4 35.2 33.9 32.7 Ln GDP ple regression estimate for each of the variables 8.0 21.9 24.6 27.2 29.8 32.4 8.5 6.2 12.7 19.2 25.7 32.2 (equations 1-4 in table B.8), and the results 9.0 ­9.5 0.9 11.2 21.6 32.0 when all variables are included in a single multi- 9.5 ­25.2 ­11.0 3.2 17.5 31.7 ple regression (equation 5 in the table). In addi- tion, to pursue a robust regression approach, all possible pairs and triplets of these variables were estimated. The main findings are as follows: cant in the multiple regressions (for WHZ only when the South Asia dummy is added, · The asset index is always significant in the though the dummy just removes the signifi- simple regression, but never so in the multi- cance of the newspaper variable in the WAZ ple regression. Indeed, assets are not signifi- equation). cant in two of the four regressions with three · The South Asia dummy (added in equation 6 regressors for both HAZ and WHZ, and once of table B.8 for each regressand) is always sig- for WAZ. This result means that the effect of nificantly positive, showing the worse nutri- assets found in the simple regression is pick- tional status of that region once a range of ing up the impact of the other omitted vari- determinants is allowed for. ables that are correlated with assets, which they all are, but particularly the education These results were used as point estimates to variables. Of course the correlation with as- model the drivers of change in nutritional status. sets is in this case likely be at least in part The calculations were performed as follows: since income (for which assets are proxying) is one determinant of education. · For each variable, table B.9 records the quar- · Health service inputs remain significant in two of the three nutrition equations, the ex- tile values of the regressors. ception being wasting (weight for height), · These regressors are used to calculate the ex- the latter being the only case when the health pected value of the dependent variable, for a index is not significant in any of the other country with the respective set of characteris- regessions (it is not so in two of the four tics. Although they need not be exactly equal, triplets for this regressand). the corresponding values do correspond · The education variable is significant in all re- quite closely to the quartile values for the ap- gressions. In the case of HAZ, secondary edu- propriate outcome variable in each case. cation worked better than primary and so was · The absolute change is the change in the out- used instead.1 come that would be caused by a country mov- · The various access to media variables were ing from across the inter-quartile range. This each tried, retaining in each case that which figure is calculated as the product of the dif- best fit for that dependent variable: watching ference in the value of the independent vari- TV for HAZ and WHZ and reading the news- able and the estimated coefficient. These paper for WAZ. These variables are signifi- changes sum to the predicted change in the 5 7 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? outcome variable. Hence, the changes from Contribution to Welfare T A B L E B . 7 each separate regressor can be expressed as a Outcomes in Bangladesh percentage of the total change. Share of The results show that the smallest effect usu- Coeff. 1980 2000 Change actual ally comes from the change in the asset index (a) HAZ (which recall was anyhow insignificant), with Safe water ­0.11 ­4.84 ­10.67 ­5.8 25.5 education and media accounting for the bulk of Female literacy ­0.29 ­5.9 ­9.1 ­3.2 13.8 the difference for the first two outcomes. In Gender inequality ­90.99 ­90.0 ­92.6 ­2.6 11.3 these first two cases, health care accounts for a Daily energy ­0.01 ­27.1 ­29.5 ­2.4 10.4 bit under 20 percent of the difference, and supply rather more for WAZ, for which assets also ap- Constant 176.72 176.7 176.7 0.0 0.0 Predicted 53.6 34.8 ­18.8 82.1 pear more important. Actual 67.6 44.7 ­22.9 100.0 What these results show is that improve- Residual 14.0 9.9 ­4.1 17.9 ments in education and access to media (pre- (b) WAZ sumably as a medium of health and nutritional advice) can bring about substantial reductions in Safe water ­0.02 ­0.99015 ­2.18282 ­1.2 5.5 Female literacy ­0.14 ­2.9 ­4.4 ­1.5 7.1 nutritional deprivation. It may not be accurate to Gender inequality ­30.10 ­29.8 ­30.6 ­0.9 3.9 say these reductions can be brought about "in- Daily energy ­0.01 ­33.4 ­36.3 ­2.9 13.5 dependent of income growth" since that growth supply may be needed to generate both supply and de- Constant 104.58 104.6 104.6 0.0 0.0 mand for education and media services. Having Predicted 38.5 31.0 ­7.5 34.5 Actual 69.5 47.7 ­21.8 100.0 said that, neither of these variables is driven Residual 30.9 16.7 ­14.2 65.5 solely by income, so there is scope for auton- (c) U5M omous improvements in nutrition. Somewhat similar results emerge for the mor- Safe water ­0.33379 ­14.6869 ­32.378 ­17.7 13.9 tality regressions (table B.10), although only the Female literacy ­1.32569 ­26.6 ­40.8 ­14.2 11.1 Gender inequality ­277.356 ­274.5 ­282.3 ­7.9 6.2 multiple regression results are shown here. The Daily energy ­0.04125 ­83.0 ­90.3 ­7.3 5.7 dependent variables are neonatal mortality (first supply month), postnatal mortality (months 1­11), and Immunization ­0.54844 ­5.3 ­43.9 ­38.6 30.3 child mortality. As with nutrition, assets are sig- Constant 638.4797 638.5 638.5 0.0 0.0 nificant in a simple regression but never so in Predicted 249.1 148.8 ­100.3 78.7 Actual 205.0 77.5 ­127.5 100.0 the multiple regression. In these regressions it Residual ­44.1 ­71.3 ­27.2 21.3 was possible to include both primary and sec- (d) TFR ondary education and get significant results, de- spite the very high level of correlation between Safe water 0 0 0 0.0 0.0 these two regressors. For infants, secondary is Female literacy ­0.01192 ­0.2 ­0.4 ­0.1 5.0 Gender inequality 0 0.0 0.0 0.0 0.0 significant and primary is not, having even the Daily energy ­0.00051 ­1.0 ­1.1 ­0.1 3.5 wrong sign for postnatal mortality (which is also supply true if secondary is omitted). But for child mor- Lagged under- 0.007539 1.7 1.0 ­0.7 26.6 tality, primary is significant and secondary not. A five mortality possible interpretation of this finding is that the Constant 5.093549 5.1 5.1 0.0 0.0 Predicted 5.5 4.6 ­0.9 35.1 child rearing skills needed to ensure child health Actual 5.6 3.0 ­2.6 100.0 are of a more rudimentary nature than those re- Residual 0.1 ­1.6 ­1.7 64.9 quired for infants. The health index is significant in all three regressions. In addition to health, a variable is included on the percentage of women 5 8 C R O S S - C O U N T R Y A N A L Y S I S O F C H I L D H E A L T H A N D N U T R I T I O N O U T C O M E S Contribution of Different Factors to Improved Welfare F I G U R E B . 2 Outcomes in Bangladesh 100 90 80 70 60 50 Percent 40 30 20 10 0 HAZ WAZ U5M TFR Residual Under-five mortality Immunization Daily energy supply Gender inequality Female literacy Safe water The Income-Outcome F I G U R E B . 3 Relationship in Cross- Country DHS Data 60 50 SDs) 40 (HAZ<2 30 20 Stunting 10 0 0 10 20 30 40 50 60 70 Asset index 5 9 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? T A B L E B . 8 Regression­Based Estimates of Contribution to Welfare Changes (1) (2) (3) (4) (5) (6) Height for age Asset index ­0.73*** ­0.07 0.11 ­7.19 ­0.38 ­0.75 Health service index ­0.23*** ­0.12* ­0.10** ­6.49 ­2.01 ­2.12 Secondary education ­0.26*** ­0.18* ­0.14* ­6.60 ­1.85 ­1.85 Watch TV ­0.21*** ­0.13** ­0.15*** ­5.78 ­2.17 ­2.99 South Asia dummy 14.93*** 4.99 Intercept 48.20*** 35.21*** 36.22*** 37.27*** 43.37*** 41.96*** 17.32 26.85 24.14 18.95 14.23 17.12 Number of observations 81 86 104 79 47 47 R-squared 0.40 0.33 0.30 0.30 0.57 0.73 Weight for height Asset index ­0.27*** 0.01 0.00 ­4.85 0.12 0.03 Health service index ­0.10*** ­0.04 ­0.04 ­5.65 ­1.45 ­1.63 Primary education ­0.12*** ­0.13*** ­0.12*** ­8.58 ­4.87 ­4.31 Watch TV ­0.08*** ­0.04 ­0.04* ­3.96 ­1.55 ­1.64 South Asia dummy 3.02*** 1.88 Intercept 14.88*** 9.88*** 15.05*** 10.51*** 18.58*** 17.74*** 9.93 15.43 15.02 10.12 10.85 10.29 Number of observations 81 86 104 79 47 47 R-squared 0.23 0.28 0.42 0.17 0.70 0.73 Weight for age Asset index ­0.87*** ­0.18 ­0.32*** ­7.52 ­1.18 ­2.78 Health service index ­0.262*** ­0.15** ­0.16*** ­6.91 ­2.19 ­3.24 Primary education ­0.3*** ­0.14* ­0.07 ­8.58 ­1.98 ­1.44 Read newspaper ­0.36*** ­0.18** ­0.11* ­5.92 ­2.26 ­1.83 South Asia dummy 18.93*** 6.50 Intercept 46.16*** 30.471*** 42*** 32.8*** 48.363*** 42.59*** 14.61 21.85 17.1 14.8 11.42 13.73 Number of observations 81 86 104 79 47 47 R-squared 0.42 0.36 0.42 0.29 0.66 0.84 Notes: (1) In the multiple regression for HAZ, primary education in fact has the wrong sign, though it is not significant. *, **, *** significant at greater than 10 percent, 5 percent, and 1 percent respectively. 6 0 C R O S S - C O U N T R Y A N A L Y S I S O F C H I L D H E A L T H A N D N U T R I T I O N O U T C O M E S T A B L E B . 9 Regression­Based Estimates of Contribution to Welfare Changes Percentage difference First Second Coefficient quartile Median quartile Absolute Share Height for age Asset index ­0.11 18.8 23.1 31.6 ­1.39 8.3 Health service index ­0.10 10.0 26.3 42.2 ­3.16 18.9 Secondary education ­0.14 12.2 24.3 43.6 ­4.55 27.2 Watch TV ­0.15 24.9 35.7 77.2 ­7.63 45.6 South Asia dummy 14.93 0 0 0 -- -- Intercept 41.96 1 1 1 -- -- Predicted 33.5 28.1 16.8 ­16.7 100.0 Weight for height Asset index 0.002 18.8 23.1 31.6 0.03 ­0.3 Health service index ­0.04 10.0 26.3 42.2 ­1.44 16.1 Primary education ­0.12 43.3 71.9 89.2 ­5.35 60.1 Watch TV ­0.04 24.9 35.7 77.2 ­2.15 24.1 South Asia dummy 3.02 0 0 0 -- -- Intercept 17.74 1 1 1 -- -- Predicted 11.3 6.8 2.4 ­8.9 100.0 Weight for age Asset index ­0.32 18.8 23.1 31.6 ­4.04 24.5 Health service index ­0.16 10.0 26.3 42.2 ­5.16 31.3 Primary education ­0.07 43.3 71.9 89.2 ­3.32 20.1 Read newspaper ­0.11 13.1 26.2 49.8 ­3.97 24.1 South Asia dummy 18.93 0 0 0 -- -- Intercept 42.59 1 1 1 -- -- Predicted 30.5 23.0 14.0 ­16.5 100.0 aged 15­19 having no children. This variable assets. Both education and health (including picks up two effects: (1) the direct adverse effect reproductive health captured in the no-birth of younger mothers on child survival chances, variable) account for between one-third and and (2) as a proxy for fertility. Finally, electricity one-half of the decline in mortality. is used here instead of the media variable (there is a high correlation between electricity and each Conclusions of the media variables). Although only signifi- This annex has presented cross-country evi- cant in one of the three regressions (CMR) (and dence of the determinants of child health and just not so for PNM), electricity proved powerful nutritional outcomes. It has also analyzed the in many specifications estimated during model importance of the different drivers of changes in specification testing, usually being the variable these outcomes, at both the general level and that knocked assets out of the equation. specifically in the case of Bangladesh. The moti- The simulations were calculated in the same vation behind this analysis is to help identify the way as were those for nutrition with one differ- selection of sectors for study in this evaluation. ence. In this case the coefficients were obtained To this end, the main findings are as follows: from re-estimated equations including only the significant variables. This was done to avoid per- · While health inputs do matter for child health verse effects, such as the positive coefficient on outcomes, they are by no means the sole-- 6 1 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Mortality Regression T A B L E B . 1 0 Estimates Using DHS Cross­Country Data Neonatal Postnatal Child mortality mortality mortality Asset index 0.21 0.27 0.40 1.42 1.04 0.81 Primary education ­0.08 0.17* ­0.58*** ­1.40 1.73 ­3.02 Secondary education ­0.20** ­0.48*** 0.19 ­2.39 ­3.24 0.65 No births, 15­19 ­0.22* ­0.57*** ­1.29*** ­1.85 ­2.84 ­3.32 Health index ­0.11** ­0.20** ­0.38** ­2.25 ­2.33 ­2.32 Electricity ­0.02 ­0.18 ­0.39* ­0.36 ­1.53 ­1.75 Sub-Saharan Africa 5.19 0.78 8.04 1.55 0.13 0.72 South Asia 13.08*** ­14.57** ­42.52*** 3.73 ­2.41 ­3.64 Middle East and 2.33 4.64 ­14.69 North Africa 0.68 0.78 ­1.28 Intercept 60.86*** 96.83*** 208.71 5.42 5.01 5.57 Number of observations 75 75 75 R-squared 0.77 0.71 0.78 Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively nor are they always the most important-- nutrition information. For mortality, elec- determinant of these outcomes. It makes trification provides a better fit. Although this good sense to incorporate determinants from variable is highly correlated with access to other sectors. media, electrification may provide additional · Education emerges as a very robust determ- channels for better health than simply health inant of child health and nutritional out- information. comes. The evidence is mixed on whether · Activities in support of economic growth are primary education is sufficient for there to be one way in which external assistance can a beneficial effect, or whether secondary ed- improve health and nutrition outcomes. ucation is required. The quality of education However, the evidence suggests that these is a factor underlying this distinction. If many channels have not been the dominant ones, children are still illiterate after primary school especially in Bangladesh. graduation, then its beneficial effects will be · In summary, these findings support the choice limited. of sectors for this study--that is, health and · Access to media is important for nutrition, nutrition, female secondary education, and presumably reflecting access to health and rural electrification. 6 2 C R O S S - C O U N T R Y A N A L Y S I S O F C H I L D H E A L T H A N D N U T R I T I O N O U T C O M E S T A B L E B . 1 1 Mortality Simulations Change Bottom Upper Coefficient quartile Median quartile Absolute Share Neonatal mortality Primary education ­0.11 43.3 71.9 89.2 ­5.3 27.0 Secondary education ­0.16 12.2 24.3 43.6 ­5.0 25.4 No births, 15­19 ­0.42 76.5 84.9 91.0 ­6.2 31.5 Health index ­0.10 10.0 26.3 42.2 ­3.2 16.1 Sub-Saharan Africa 0.99 0 0 0 0.0 South Asia 8.57 0 0 0 0.0 Middle East and North Africa 2.22 0 0 0 0.0 Europe and Central Asia 3.33 0 0 0 0.0 Intercept 85.69 1 1 1 0.0 Predicted 45.3 35.0 25.8 ­19.5 Postnatal mortality Secondary education ­0.32 12.2 24.3 43.6 ­10.2 30.2 No births, 15­19 ­0.55 76.5 84.9 91.0 ­8.0 23.7 Health index ­0.11 10.0 26.3 42.2 ­3.4 10.1 Electricity ­0.22 14.5 33.0 70.6 ­12.1 36.0 Sub-Saharan Africa ­4.29 0.0 0.0 0.0 0.0 South Asia ­19.87 0.0 0.0 0.0 0.0 Middle East and North Africa 2.62 0.0 0.0 0.0 0.0 Intercept 111.05 1.0 1.0 1.0 0.0 Predicted 60.9 46.6 27.2 ­33.6 100.0 Infant mortality Primary education ­0.51 43.3 71.9 89.2 ­23.4 34.7 No births, 15­19 ­1.30 76.5 84.9 91.0 ­18.9 28.1 Health index ­0.37 10.0 26.3 42.2 ­11.8 17.6 Electricity ­0.24 14.5 33.0 70.6 ­13.3 19.7 Sub-Saharan Africa 7.88 0.0 0.0 0.0 0.0 South Asia ­41.81 0.0 0.0 0.0 0.0 Middle East and North Africa ­14.13 0.0 0.0 0.0 0.0 Intercept 214.42 1.0 1.0 1.0 0.0 Predicted 85.6 49.8 18.3 ­67.3 100.0 6 3 ANNEX C. NEONATAL, POSTNATAL, AND CHILD MORTALITY IN THE 1990S Reducing mortality in childhood has been a grassroots level and increased awareness of major component of health and population sec- health care issues among the population through tor strategy in Bangladesh, and is now included the intervention of mass communications chan- in the Poverty Reduction Strategy (Bangladesh nels." This paper analyzes the determinants of Government/IMF 2003). Health and nutrition mortality during the 1990s using Demographic programs have received substantial donor and Health Survey (DHS) data, and attempts to support (Annex H). Infant and child survival ascertain the weight of various factors in achiev- prospects have improved dramatically in recent ing the improvement in infant and child survival decades. Under-five mortality rates fell by more using decomposition methods. than one-third during the mid-1980s to late- 1990s (figure C.1).1 Analysis suggests that mor- Modeling Childhood Mortality Risk tality differentials have narrowed between The usual point of reference for analyzing mor- "poor" and "non-poor" socioeconomic groups tality determinants, at least in the demographic (Appendix C.3). literature, is Mosley and Chen's (1984) frame- Gender discrimination has important impli- work, which models mortality outcomes as cations for child survival prospects in South being determined proximately by factors such as Asia (Sen 1998; Croll 2001), and discrimination nutrient availability, use of health services, in- against girl children in Bangladesh is among juries and maternal fertility, which in turn de- the most severe in the world (see below). While pend on the underlying biological conditions mortality has declined among both sexes, post- (child-specific), socioeconomic and behavioral neonatal mortality among girls (that is, girls aged conditions (mother or household-specific), and between 1 and 60 months of age) remains environmental conditions including service pro- higher than among boys, and the difference be- vision (community-specific). tween the sexes was relatively stagnant during The economic literature derives from the the 1980s and 1990s (figure C.2). Socioeco- household's utility maximization problem. House- nomic inequalities in mortality have narrowed holds have a utility function--determining their over time, falling more clearly for male children preference orderings over consumption of differ- and neonates than all other groups of under- ent goods, including child health and nutritional fives (Appendix C.3); in contrast, socioeconomic status and leisure--which is maximized subject inequality in post-neonatal mortality is higher to the household's labor constraint and the among girls and has remained constant over health and nutrition production functions. The time at best. health and nutrition production functions are The general improvement in survival chances specified as: in Bangladesh has been well documented. Ten H and N = F( NUTR-1, H-1, INC, MOTHED, years ago, Kabir and Amin (1993, p. 10) stated CHILD, HHCARE, ENV ) that "the fall in infant mortality [since the early 1980s] may be attributed to the large-scale im- H and N are, respectively, health and nutritional munization of children from the early 1980s, de- status produced by each child, which depend on velopment of the health infrastructure at the the child's nutrition and health histories, NUTR-1 6 5 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? However, it also emphasizes that decisions on F I G U R E C . 1 Mortality Rates (per 1,000) health are taken jointly with other household decisions such as labor supply. Thus, current in- come is considered endogenous to health sta- 180 tus, so that its inclusion in multiple regression 1985­89 160 1990­94 analysis results in inconsistent parameter esti- 140 1995­99 mates. A solution is to use wealth as a proxy for 120 income, as it is unlikely to be endogenous to cur- 100 rent nutrition/health choices--thus producing consistent estimates.2 Furthermore, determining 80 the impact of household decision variables-- that 60 is, variables determined in part behaviorally, 40 such as utilization of health care--is problematic 20 because of unobserved heterogeneity. In other 0 words, where heterogeneous household behav- Neonatal Postnatal Child Under 5 ioral characteristics cannot be fully accounted Source: Calculated from DHS data.. for in the set of explanatory variables, they are subsumed into the error term. Hence, if these characteristics are correlated with other ex- and H-1, household income, INC, mother's edu- planatory variables, inconsistent estimates result cation and knowledge, MOTHED, child-specific (Rosenzweig and Schultz 1983). Two ap- factors, CHILD (such as gender, age, birth proaches can be taken to ameliorate this prob- order), utilization of health services, HCARE, and lem. The "health production function" approach factors determining exposure to environmental estimates proximate determinants of health risk, ENV (for example, water, sanitation and pol- status using multi-stage estimation where the lution) (see Chamarbagwala and others 2004). first-stage equation predicts the behavioral The economic model is generally consistent factors for each observation. However, this ap- with demographic interpretations in terms of proach requires identifying instruments exoge- identifying determinants of health status. nous to child health status.3 The alternative is to estimate reduced-form "health demand func- tions," linking health outcome to underlying variables affecting behavioral choices--for ex- Female-to-Male Mortality ample, presence of health and family planning F I G U R E C . 2 Ratio (percent) facilities in the community, or distance from that facility. The demand function approach is a popular 180 1985­89 solution, particularly in analysis of mortality using 160 1990­94 survival models. However, by not including prox- 140 1995­99 imate determinants as explanatory variables, the 120 approach is likely to understate the importance 100 of factors such as health care. Moreover, it is 80 these proximate determinants which are the most interesting for policymakers since they are 60 most amenable to policy intervention. The re- 40 mainder of this section reviews empirical evi- 20 dence regarding which variables affect mortality 0 outcomes, drawing on a meta-analysis carried Neonatal Postnatal Child out for this study (Chamarbagwala and others Source: Calculated from DHS data. 2004). 6 6 N E O N A T A L , P O S T N A T A L , A N D C H I L D M O R T A L I T Y I N T H E 1 9 9 0 s Child-Specific Factors hand, mortality is usually high among firstborns, Biological factors are important causes of death particularly in infancy, partly because these in infancy, particularly in the first month of life. births are more likely to occur before the Mother's age is usually found to have a convex mother has reached physical maturity. On the (U-shaped) effect on infant mortality. Infants other hand, high birth order (i.e., being born born to mothers under 20 and over 35 are at later than other siblings) is expected to increase higher risk of death, because women under 20 mortality risk for two reasons: physiologically, are less likely to have fully developed reproduc- because women who have had many pregnan- tive systems, while the reproductive systems of cies are more likely to be physically depleted, women over 35 may be deteriorated. The rela- and behaviorally, particularly where birth spac- tionship between mother's age and post-infant ing is lower, due to constraints on household re- survival is due to other factors, for example be- sources. However, women who have had more cause children of older women are more likely to children are likely to be those experiencing be competing with other (younger) siblings; the higher mortality rates among their offspring. relationship between mother's age on child mor- This will lead to spurious correlation between tality is therefore more likely to be linear, al- birth order and mortality risk due to unobserved though children of very young mothers may correlations between siblings (frailty effects). It suffer through the lack of maturity of the mother. is important to control for this source of hetero- This is more likely to be a factor when first births geneity in mortality risk across families (see occur at a young age, as is the case in Bangladesh. below). Child's gender has a strong relationship with Children are at higher risk of mortality when survival chances. For biological reasons, males born after short intervals.4, This reflects a num- 5 have a higher natural risk of death at all ages. ber of factors (Koenig and others 1990): However, it is well known that preference for boys over girls is strong throughout Asia, includ- · Biological factors among prenates (fetal ing Bangladesh (Croll 2001), with discrimination growth harmed by weakened nutritional re- aggravated during crisis periods (see Bairagi serves of the mother, who has been recently 1986). Human interventions override biological pregnant and lactating) and among infants factors at later stages in infancy, to the extent where potential milk production is impaired. that, while neonatal mortality is typically higher · Socioeconomic and behavioral factors, i.e., among boys, under-five mortality rates are competition among post-neonatal siblings for higher among girls. In the multivariate regres- nutrition and mother's care, where parents' sion studies included in the meta-analysis, the choices determine which child is favored. coefficient on female child was insignificant or Either the younger child is neglected in favor significantly negative for infant mortality, of the older child, or vice versa, so that short whereas in studies of child mortality the coeffi- subsequent birth interval becomes an impor- cient was insignificant or significantly positive tant mortality risk factor--e.g., due to a new (Chamarbagwala and others 2004). The esti- pregnancy, children are weaned earlier with- mated effect of sex on mortality is likely to be out receiving adequate nutritional substitutes conservative due to sex-selective omissions in for breast milk. Alternatively, both may be reported birth histories, which Bairagi, Islam, disadvantaged due to limited total family and Barua (1999) note will artificially lower mor- resources. tality estimates among girls, although this effect · Disease, since a higher number of young chil- is less in countries where the technology is not dren in the household means that the likeli- widely available. hood of contracting contagions such as The relationship between birth order and measles is higher. mortality risk is likely to be convex (U-shaped), reflecting mother's age, social preferences, and Son preference may also have an indirect ef- food availability for older children. On the one fect on girls' mortality rates when parents 6 7 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? choose to reduce birth interval because they are and child. However, where home delivery is the unhappy with the current gender composition norm, it can be the case that only complicated of their offspring (Koenig and others 1990). deliveries take place in health facilities, thus giv- Miller and others (1992) find that mortality ing a perverse relationship. Immunization risk is higher among children born less than should prevent exposure to major killers such as 15 months after the preceding child than those measles, lowering risk of death among children born more than 15 months after or those who immunized as well as having positive external ef- are firstborns, reflecting the likelihood that, for fects for child survival among the non-immu- birth intervals less than 15 months, conception nized in a given community. The meta-analysis occurred before the preceding child was of Chamarbagwala and others (2004) indicates 6 months old, therefore increasing chances that that these factors are indeed significantly corre- the child would have been weaned early.6 The lated with mortality in multivariate studies. meta-analysis found that short preceding birth Sanitary toilet facilities and safe drinking interval increases infant mortality risk, while water improve survival chances by reducing short subsequent birth interval increases child exposure to diseases such as diarrhea. While it mortality risk (Chamarbagwala and others 2004). can be reasoned that water and sanitation are The effects of birth interval, gender, and birth more likely to have greater effects as children get order are interlinked. In particular, in South Asia, older and start using the services directly, mortality risks are likely to be high among girls Chamarbagwala and others (2004) show that, in who have one or more older sisters (Muhuri and fact, differentials in water and sanitation vari- Preston 1991; Muhuri and Menken 1997; Masset ables are more likely to be significant determi- and White 2003) or among girls with no male sib- nants of infant mortality in multivariate studies. lings (Croll 2001).7 Some important interactions Electrification is usually found to have a posi- to consider are between birth order and sex, and tive impact on infant survival, but an insignificant between sex of index child and sex composition impact on child survival (Chamarbagwala and of older siblings (Masset and White 2003).8 others 2004). The channels through which ac- Birth order and birth interval are proxies for cess to electricity affects mortality are, however, fertility, so these relationships between birth contentious, though possible candidates are that order, birth interval, and mortality risk indicate electrification facilitates use of modern medicine that lower fertility means lower mortality. Al- in health facilities (e.g., a refrigerator allows though birth order and interval have the direct hygienic storage of vaccines), improves trans- effects described above, they are also picking up fer of health information (through television the simultaneous relationship between declin- and radio), and improves domestic air quality. ing fertility and mortality that occurs as the Electrification is, of course, highly correlated demographic transition proceeds and parents with locality (particularly rural versus urban) and decide to invest in child quality rather than quan- it is also likely that electrification expansion fa- tity. There will therefore be relationships be- vors less remote and better off communities tween these variables and those such as use of (which are able to pay for the services), raising health services for the child. issues of reverse causality at community level. Controlling for socioeconomic status, broadly, Community-Level Factors should account for this. Access to decent health care is an important de- Water, sanitation, electricity, and health ser- terminant of infant and child survival. Babies de- vices are labeled as community-level variables, livered in modern health facilities by trained because communities gain access to them at the health professionals will have lower risks of same time (e.g., due to electrification programs, death, since presence of a health professional re- construction of a health center, etc.) and be- duces the mortality risk associated with compli- cause it is recognized that there are strong posi- cations during and after labor, while hygienic tive externalities from improved services, e.g., in conditions reduce risk of infection of mother terms of disease control (the degree of coverage 6 8 N E O N A T A L , P O S T N A T A L , A N D C H I L D M O R T A L I T Y I N T H E 1 9 9 0 s in the community has an important effect on (ORS) to treat diarrhea. Educated mothers are reducing morbidity and mortality among the more likely to seek treatment from health pro- community). However, where intra-community fessionals, and see this as their right. Moreover, variation occurs due to variation in household educated mothers may also have greater deci- income and preferences and where externalities sion-making ability within their families and, do not exist, "community level" may be an erro- therefore, greater say on child nutrition and neous classification. This is likely to be particu- health decisions (Caldwell 1979). Chamarbagwala larly true for aspects of health care such as and others (2004) find that parental, particularly antenatal care, where access is determined at the maternal, education has a positive effect on infant community level by presence of a nearby health and child survival in studies of mortality. facility, but whether parents seek professional In a highly patriarchal society such as Bang- health care is determined at family level by such ladesh, marriage disruption caused by divorce, factors as income and maternal education and separation, or abandonment is potentially disas- preferences. trous for women and therefore children. Where Most multivariate analyses control for physi- women have no support network, or are dis- cal locality (region and rural/urban sector), to owned by their families, they will be highly vul- represent unobserved determinants of mortality nerable and may have to remarry, or accept a risk, mainly in terms of access to health care and new husband they know is already married. In clean water and sanitation, and possibly geo- such cases, the children may not fare better if graphic socio-cultural factors. Chamarbagwala they are not welcome to the new husband's fam- and others (2004) showed that urban location is ily. The effect of divorce is likely to be greater for frequently related to significantly lower mortal- infants because young children suffer more from ity risk in multivariate models. reduced quantity and quality of mother's care, e.g., because she is working to support the fam- Household-Level Socioeconomic Factors9 ily (Bhuiya and Chowdhury 1997). Socioeconomic status determines nutrition, ill- ness, and ability to access health care. House- Unobserved Factors hold socioeconomic status represents direct Unobserved factors determining mortality determinants of mortality such as nutrient intake among families include genetic or other family- and likelihood of affording health treatment. specific practices altering childhood survival Multivariate studies generally find income or prospects and are referred to in the literature as wealth to have a significantly positive impact on "frailty" effects. As noted above, it is important to child survival, though no impact on infant sur- account for sources of unobserved heterogene- vival prospects (Chamarbagwala and others ity in empirical estimation, since correlation be- 2004). tween observations can lead to inconsistent and Education of household members is often inefficient statistical estimates, as demonstrated used to proxy socioeconomic status as an indica- for survival models by Heckman and Singer tor of potential earnings capacity of the labor (1984). Statistical tools can be used to account force. In analyses of child health, however, there for frailty, e.g., by assuming that observations at is a clear distinction between mother's and fa- child level are correlated among mothers; how- ther's education. Whereas father's education may ever, when modeling frailty among siblings, it is affect child health indirectly through socioeco- also important to control for the death of the nomic status, mother's education has clear direct preceding child separately, since the association links to children for a number of reasons. between immediate pairs of siblings may be Education can proxy for basic knowledge about stronger (Zenger 1993). nutrition and child care, including knowledge of good caring practices, health problems and treat- Data and Methodology ment, and the importance of immunization and Data used for modeling mortality in Bangladesh techniques such as use of oral rehydration salts come from the nationally representative Demo- 6 9 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? graphic and Health Survey. DHS collects data on age-specific baseline hazard, which is the complete fertility histories of all ever-married mortality risk at each age in the case where all women aged 10-49, as well as data on health sta- explanatory variables are equal to zero (i.e., tus and health care, contraception, women's equivalent to the constant term in a standard lin- knowledge, beliefs and caring practices, house- ear regression); xi is the matrix of covariates and hold composition, and other socioeconomic the estimated parameters. Child mortality is factors such as water and sanitation facilities, modeled using the proportional hazards specifi- household asset ownership, and educational cation. For neonates and postnates, however, it attainment. Data were collected over three is reasonable to drop observations on, respec- rounds during the 1990s: 1993/94, 1996/97, and tively, children born in the month of the inter- 1999/2000. view and children born in the year before the Most empirical studies analyze mortality de- interview and estimate using a probit model. terminants for infants (up to one year old) and The probit model is summarized as: children (between one and approximately five years old) separately. Biological and environ- P yi = 1 = xi ( ) ( ) mental factors should be more important for infants, while socioeconomic and behavioral fac- where P(yi=1) is the probability of the event oc- tors become more important as children get curring for individual i, in the case of neonatal older, though some behavioral factors, such as death in the first month after birth or for post- delivery assistance and antenatal care, are also nates between the second and twelfth months important for neonatal survival prospects.10 It is following birth, and is the standard normal also useful to divide infant mortality further into cumulative density. neonatal (first month after birth) and postnatal A determinant can alter risk of death through (second to twelfth months). Appendix C.1 sum- two routes: where the effect of the factor marizes results of other multivariate studies changes over time (i.e., changes in the coeffi- examining childhood mortality in Bangladesh. cient estimate); and where the distribution of The major econometric issue with mortality the factor changes among the population (i.e., analysis is that survey data are right-censored: changes in the mean of the variable). A decom- survival status of children born within the period position analysis is undertaken following model of analysis is unknown past the survey date. This estimation in order to determine the latter. introduces a bias in conventional regression es- As noted above, there are likely to be issues of timates, such as those obtained from probit/logit unobserved heterogeneity of important proxi- regression, unless observations are omitted for mate causes of health status relating to behav- children born within x time periods before the ioral choices which will cause biased estimation survey, where x is the mortality rate being of the "true" (technological) impact of the ex- analyzed. Survival analysis, by accounting for planatory variables on health outcomes. Be- right-censoring, enables all observations to havioral factors, such whether the mother seeks be included and is therefore statistically more antenatal care during pregnancy and delivery efficient. assistance, are likely to be endogenous to health A frequently used model of mortality is the status--i.e., she seeks health care during preg- Cox proportional hazards model.11 The model nancy because she is experiencing complica- assumes that the risk of death for any age can be tions or previously experienced complications calculated by adjusting the baseline risk for that during birth, which are likely to be correlated age by an exponential set of factors. The model among her children due to frailty. On the other can be summarized as follows: hand, immunization is likely to be less endoge- hi t = h0 t exp xi ( ) ( ) ( ) nous than antenatal care, since it is preventative rather than curative.12 But even where reverse where hi(t) represents the hazard (mortality) causality is not likely to be a serious issue, unob- rate at time (t) for individual i and h0(t) is the served heterogeneity in preferences will lead to 7 0 N E O N A T A L , P O S T N A T A L , A N D C H I L D M O R T A L I T Y I N T H E 1 9 9 0 s erroneous estimates of the effect of decision Estimation Results variables, particularly those relating to caring The estimation strategy is to pool data for the practices. Some sources of heterogeneity can be three survey rounds and distinguish between controlled for with other explanatory variables-- two periods, the late-1980s to early-1990s and for example, maternal education and indicators the mid-late 1990s. Due to the likelihood that of knowledge of modern health care practices. the relationships between some covariates and In addition, modeling frailty should account for mortality change during early childhood, sepa- some of the unobserved heterogeneity (though rate regressions are estimated for neonatal mor- it is unable to account for the likelihood of be- tality (less than 1 completed month), postnatal havior changing from birth to birth as it is as- mortality (from 1 to 11 completed months), and sumed that frailty is constant across births for child mortality (from 12 to 59 months). the same mother). Table C.1 presents bivariate estimates of mor- It is particularly likely that antenatal visits and tality rates by various factors. Mortality rates are delivery assistance are endogenous to child lower among wealthier households, though health, both because of reverse causality and there is little difference between mortality rates unobserved heterogeneity with respect to moth- among the bottom four quintiles. Mortality is ers' behavioral choices. Under these conditions, higher in households with no electricity, poor consistency requires use of instrumental va- sanitation, and non-piped (direct into the home) riables. Greene (1998; see also Greene 2000, drinking water. The Divisions of Chittagong13 p. 849) indicates that an endogenous system of and Barisal have the highest mortality rates while limited dependent variable equations can be ap- Khulna has the lowest; girls fare worst in Chit- propriately modeled as a multivariate probit tagong. Mortality rates are lower among children model. Suppressing individual subscripts, the whose mothers have formal education, espe- trivariate model can be expressed as: cially when they have at least upper secondary education (grade 10 onwards), are not divorced P y1 = 1, y2 = 1, y3 = 1 = 3(1x1 + y2 + y3, ( ) or remarried, and are more mobile (can go out 2 x2, 3 x3, 21. , 31, 32 ). and/or can go to the health center). Child- specific characteristics associated with higher 3 indicates the trivariate standard normal den- mortality risk include multiple birth, short pre- sity; y1, y2, and y3 are, respectively, the probabili- ceding birth interval (less than 15 months), and ties of dying, receiving antenatal care, and delivery death of previous child. Higher mortality rates assistance; the equation estimating probability of are experienced by boys in the neonatal period, dying includes antenatal care in the explanatory but by girls in the post-neonatal period. Children variables set, where and are the estimated pa- whose mothers received antenatal care, two rameters on antenatal visits and assisted delivery; or more tetanus toxoid (TT) vaccinations or 2x2 and 3x3 are the vectors of estimated coeffi- who have been breastfed, have better survival cients and explanatory variables of the equations chances.14 Mortality rates are lower among chil- for antenatal visits and assisted delivery; repre- dren living in communities where over half of sents the estimated correlation between error the children have received at least one vacci- terms of each equation--their statistical signifi- nation or received Vitamin A in the preceding cance, therefore, provides evidence for endogene- six months. Babies delivered by a doctor have ity. Identifying instruments used in the equations higher neonatal mortality rates than those deliv- for antenatal visits and delivery assistance are ered by nurses or traditional birth attendants presence of health and information facilities in the (TBAs), suggesting that doctor-assisted births community, distances from Upazilla and District are more likely to occur for high-risk cases. headquarters; in addition, the equation for ante- natal visits includes the non-self geographical area Neonatal Mortality mean (by maternal education attainment and sex Table C.2 presents estimates of the probit analy- of child) of the dependent variable. sis of neonatal mortality. The table reports coef- 7 1 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? T A B L E C . 1 Mortality Rates by Selected Characteristics Neonatal Infant Child mortality rate mortality rate mortality rate (standard error) (standard error) (standard error) All children 0.047 (0.002) 0.077 (0.002) 0.033 (0.001) Characteristics of household Wealth quintile 1 0.054 (0.004) 0.095 (0.005) 0.051 (0.004) Wealth quintile 2 0.050 (0.004) 0.085 (0.005) 0.044 (0.004) Wealth quintile 3 0.050 (0.004) 0.078 (0.005) 0.031 (0.003) Wealth quintile 4 0.049 (0.004) 0.075 (0.004) 0.027 (0.003) Wealth quintile 5 0.034 (0.003) 0.054 (0.004) 0.017 (0.002) No electricity 0.052 (0.002) 0.084 (0.002) 0.037 (0.002) Electricity 0.034 (0.003) 0.057 (0.004) 0.021 (0.002) Unsanitary toilet 0.052 (0.002) 0.083 (0.003) 0.038 (0.002) Sanitary toilet 0.038 (0.003) 0.063 (0.003) 0.022 (0.002) Non-piped water 0.048 (0.002) 0.078 (0.002) 0.033 (0.001) Piped water 0.030 (0.007) 0.049 (0.009) 0.016 (0.005) Characteristics of mother Mother not attended school 0.052 (0.002) 0.088 (0.003) 0.042 (0.002) Mother attended some primary 0.048 (0.004) 0.079 (0.005) 0.029 (0.003) Mother finished primary 0.039 (0.005) 0.065 (0.006) 0.022 (0.004) Mother attended lower secondary 0.042 (0.004) 0.054 (0.004) 0.010 (0.003) Mother attended upper secondary or higher 0.014 (0.005) 0.019 (0.006) 0.003 (0.001) Mother married 0.047 (0.002) 0.077 (0.002) 0.032 (0.001) Mother divorced 0.061 (0.014) 0.113 (0.019) 0.047 (0.011) Mother married once 0.047 (0.002) 0.076 (0.002) 0.032 (0.001) Mother remarried 0.047 (0.006) 0.095 (0.009) 0.048 (0.007) Mobility index=0 0.059 (0.004) 0.095 (0.005) 0.036 (0.005) Mobility index=1 0.048 (0.002) 0.077 (0.003) 0.034 (0.002) Mobility index=2 0.037 (0.003) 0.064(0.004) 0.028(0.003) Characteristics of child Child male 0.053 (0.002) 0.081 (0.003) 0.029 (0.002) Child female 0.042 (0.002) 0.073 (0.003) 0.037 (0.002) Single birth 0.044 (0.002) 0.072 (0.002) 0.032 (0.001) Multiple birth 0.261 (0.025) 0.404 (0.028) 0.105 (0.025) Previous interval >15 0.045 (0.002) 0.074 (0.002) 0.031 (0.001) Previous interval <15 0.099 (0.011) 0.171 (0.014) 0.064 (0.009) Previous child survived 0.046 (0.002) 0.074 (0.002) 0.031 (0.001) Previous child died 0.063 (0.006) 0.102 (0.007) 0.048 (0.005) First­born child 0.069 (0.004) 0.099 (0.004) 0.023 (0.002) Second or third birth order 0.038 (0.002) 0.063 (0.003) 0.033 (0.002) Fourth or fifth birth order 0.037 (0.003) 0.067 (0.004) 0.038 (0.003) Sixth or higher birth order 0.044 (0.004) 0.089 (0.006) 0.041 (0.004) Rural*Born Dec­Jan 0.052 (0.004) 0.078 (0.005) 0.031 (0.003) Rural*Born Feb­Mar 0.048 (0.004) 0.074 (0.005) 0.039 (0.004) Rural*Born Apr­May 0.039 (0.004) 0.067 (0.006) 0.044 (0.005) Rural*Born Jun­Jul 0.050 (0.005) 0.083 (0.006) 0.039 (0.005) Rural*Born Aug­Sep 0.049 (0.004) 0.079 (0.006) 0.034 (0.004) Rural*Born Oct­Nov 0.053 (0.004) 0.091 (0.005) 0.030 (0.003) Delivered by nurse 0.034 (0.006) 0.068 (0.009) 0.017 (0.008) Delivered by trained TBA 0.043 (0.006) 0.067 (0.007) 0.034 (0.011) 7 2 N E O N A T A L , P O S T N A T A L , A N D C H I L D M O R T A L I T Y I N T H E 1 9 9 0 s T A B L E C . 1 Mortality Rates by Selected Characteristics (continued ) Neonatal Infant Child mortality rate mortality rate mortality rate (standard error) (standard error) (standard error) Characteristics of child (continued) Delivered by untrained TBA 0.045 (0.002) 0.072 (0.003) 0.030 (0.003) Delivered by other 0.053 (0.003) 0.089 (0.004) 0.034 (0.002) Not breastfed 0.686 (0.014) 0.766 (0.014) 0.074 (0.022) Breastfed 0.003 (0.000) 0.029 (0.001) 0.026 (0.002) No antenatal visits 0.051 (0.002) 0.084 (0.003) 0.035 (0.002) Some antenatal visits 0.036 (0.003) 0.058 (0.004) 0.019 (0.003) Less than 2 TT vaccinations 0.056 (0.002) 0.092 (0.003) 0.035 (0.002) 2 or more TT vaccinations 0.039 (0.002) 0.064 (0.003) 0.023 (0.003) Characteristics of community Located in Barisal 0.047 (0.005) 0.083 (0.007) 0.038 (0.005) Located in Chittagong 0.049 (0.003) 0.082 (0.004) 0.042 (0.003) Located in Dhaka 0.046 (0.003) 0.081 (0.004) 0.035 (0.003) Located in Khulna 0.043 (0.004) 0.060 (0.005) 0.014 (0.003) Located in Rajshahi 0.049 (0.003) 0.074 (0.004) 0.026 (0.003) Female child*located in Barisal 0.031 (0.006) 0.069 (0.009) 0.034 (0.007) Female child*located in Chittagong 0.045 (0.004) 0.079 (0.005) 0.050 (0.004) Female child*located in Dhaka 0.041 (0.004) 0.076 (0.005) 0.038 (0.004) Female child*located in Khulna 0.041 (0.006) 0.060 (0.007) 0.012 (0.003) Female child*located in Rajshahi 0.044 (0.005) 0.070 (0.006) 0.032 (0.004) Less than half of children in community received vitamin A supplement 0.049 (0.002) 0.081 (0.002) 0.035 (0.002) More than half of children received vitamin A supplement 0.039 (0.033) 0.059 (0.004) 0.021 (0.003) Less than half of children in community vaccinated 0.052 (0.004) 0.088 (0.005) 0.053 (0.006) More than half of children in community vaccinated 0.046 (0.002) 0.075 (0.002) 0.030 (0.001) ficient estimates, as well as marginal effects between death of previous child and birth inter- (dF/dx), which indicate the effect of a unit val, suggests that newborns are less likely to die change in the explanatory variable on the prob- when there are fewer children competing for ability of death. The regressions bear out the resources. prior expectation that biological characteristics Maternal characteristics are important deter- and inadequate access to public services are key minants of neonatal mortality. Newborns have factors impeding neonatal survival prospects. better survival chances if their mothers are more Biological factors associated with higher mobile (able to go out to the health center). neonatal mortality are gender (male), multiple Mother's schooling has a negative effect on birth, short preceding birth interval, death of neonatal mortality, particularly upper secondary previous sibling, and very low or very high birth (grades 10 through 12) and higher education, as order (the estimated turning point is between compared to the reference category of no birth order 6 and 7).15 Mother's age has a convex schooling; mothers who have primary education effect on mortality risk, the coefficients suggest- also have fewer neonatal fatalities, though the ing that the mother's optimal age for childbirth estimate is imprecise.16 It is possible to examine is during her late 20s. While these factors are some of the channels through which education likely to reflect biological characteristics, the sig- lowers child mortality.17 Specification (2) in- nificantly negative coefficient on the interaction cludes a variable indicating number of modern 7 3 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? T A B L E C . 2 Neonatal Mortality Determinants: Probit Regression Estimates (1) (2) (3) (4) dF/dx Coeff. z dF/dx Coeff. z dF/dx Coeff. z dF/dx Coeff. z Wealth ­0.004 ­0.044 ­0.42 ­0.002 ­0.021 ­0.20 ­0.003 ­0.041 ­0.38 ­0.002 ­0.018 ­0.17 Electricity ­0.009 ­0.114** ­2.03 ­0.009 ­0.111* ­1.95 ­0.009 ­0.113** ­1.98 ­0.009 ­0.110* ­1.94 Piped water & sanitary toilet ­0.027 ­0.485*** ­2.62 ­0.028 ­0.502*** ­2.72 ­0.028 ­0.505*** ­2.72 ­0.028 ­0.505*** ­2.73 Mother no formal ed + Mother primary ­0.005 ­0.066 ­1.47 ­0.004 ­0.047 ­1.05 ­0.005 ­0.055 ­1.17 ­0.004 ­0.047 ­1.05 Mother lower secondary ­0.004 ­0.043 ­0.66 ­0.001 ­0.013 ­0.20 ­0.003 ­0.031 ­0.43 ­0.001 ­0.014 ­0.22 Mother upper secondary or more ­0.023 ­0.373** ­1.97 ­0.020 ­0.321* ­1.68 ­0.022 ­0.359* ­1.80 ­0.020 ­0.320* ­1.68 Contraceptive knowledge ­0.004 ­0.053*** ­4.29 ­0.004 ­0.054*** ­4.32 ­0.004 ­0.053*** ­4.30 Father no formal ed + father primary 0.003 0.037 0.82 Father lower secondary 0.002 0.027 0.47 Father upper secondary or more 0.006 0.066 0.67 Multiple birth 0.244 1.204*** 10.66 0.243 1.203*** 10.63 0.242 1.203*** 10.62 0.242 1.203*** 10.62 Age at birth ­0.004 ­0.047** ­2.15 ­0.004 ­0.043** ­2.01 ­0.004 ­0.044** ­2.04 ­0.004 ­0.043** ­2.01 Age at birth sq 0.000 0.001** 1.99 0.000 0.001* 1.81 0.000 0.001* 1.84 0.000 0.001* 1.81 Prev child died 0.019 0.200*** 3.30 0.018 0.191*** 3.16 0.018 0.191*** 3.15 0.018 0.191*** 3.15 Prec interval <15 0.085 0.619*** 6.46 0.083 0.615*** 6.42 0.083 0.615*** 6.43 0.083 0.613*** 6.40 Prev child died x prec Interval ­0.027 ­0.473*** ­3.07 ­0.027 ­0.484*** ­3.14 ­0.027 ­0.485*** ­3.16 ­0.027 ­0.481*** ­3.12 Birth order ­0.017 ­0.197*** ­5.68 ­0.016 ­0.186*** ­5.39 ­0.016 ­0.186*** ­5.37 ­0.016 ­0.186*** ­5.38 Birth order sq 0.001 0.016*** 5.39 0.001 0.015*** 5.12 0.001 0.015*** 5.11 0.001 0.015*** 5.11 Female child ­0.010 ­0.123*** ­3.59 ­0.010 ­0.123*** ­3.58 ­0.010 ­0.123*** ­3.59 ­0.026 ­0.315*** ­2.78 Female head of hh ­0.004 ­0.046 ­0.57 ­0.004 ­0.051 ­0.62 ­0.004 ­0.053 ­0.64 ­0.004 ­0.051 ­0.62 Mother's mobility ­0.006 ­0.069** ­2.46 ­0.005 ­0.057** ­2.02 ­0.005 ­0.056** ­1.99 ­0.005 ­0.057** ­2.02 Mother divorced ­0.001 ­0.013 ­0.10 ­0.001 ­0.010 ­0.08 ­0.001 ­0.009 ­0.07 ­0.001 ­0.012 ­0.10 Mother remarried ­0.002 ­0.026 ­0.35 ­0.002 ­0.030 ­0.40 ­0.003 ­0.030 ­0.41 ­0.002 ­0.028 ­0.37 Antenatal visits ­0.009 ­0.110** ­2.42 ­0.008 ­0.095** ­2.10 ­0.008 ­0.097** ­2.13 ­0.008 ­0.095** ­2.10 TBA delivery ­0.007 ­0.081** ­2.22 ­0.006 ­0.075** ­2.04 ­0.006 ­0.074** ­2.02 ­0.006 ­0.075** ­2.05 TTBA delivery 0.009 0.096 0.82 0.010 0.111 0.94 0.010 0.112 0.94 0.010 0.109 0.92 TTBA delivery x period 1995­99 ­0.017 ­0.239 ­1.55 ­0.017 ­0.245 ­1.59 ­0.017 ­0.249 ­1.62 ­0.017 ­0.243 ­1.58 Barisal + Chittagong 0.002 0.020 0.30 ­0.002 ­0.030 ­0.45 ­0.002 ­0.027 ­0.40 ­0.009 ­0.118 ­1.38 Dhaka ­0.001 ­0.012 ­0.18 ­0.003 ­0.032 ­0.48 ­0.002 ­0.027 ­0.41 ­0.008 ­0.105 ­1.23 Khulna ­0.005 ­0.067 ­0.90 ­0.006 ­0.075 ­0.99 ­0.006 ­0.070 ­0.93 ­0.013 ­0.183* ­1.83 Rajshahi ­0.002 ­0.026 ­0.38 ­0.003 ­0.034 ­0.50 ­0.003 ­0.030 ­0.45 ­0.009 ­0.114 ­1.29 Female x Chittagong 0.020 0.214* 1.66 Female x Dhaka 0.017 0.182 1.39 Female x Khulna 0.026 0.256* 1.69 Female x Rajshahi 0.019 0.198 1.48 Rural ­0.001 ­0.016 ­0.24 ­0.001 ­0.012 ­0.17 ­0.001 ­0.012 ­0.18 ­0.001 ­0.010 ­0.15 Born In Jan-Feb x rural + born In Feb-Mar x rural ­0.006 ­0.068 ­1.07 ­0.006 ­0.070 ­1.09 ­0.006 ­0.070 ­1.09 ­0.006 ­0.071 ­1.11 Born in Apr-May x rural ­0.012 ­0.158** ­2.27 ­0.012 ­0.162** ­2.32 ­0.012 ­0.161** ­2.32 ­0.012 ­0.163** ­2.35 Born in Jun-Jul x rural ­0.003 ­0.038 ­0.57 ­0.003 ­0.041 ­0.62 ­0.003 ­0.041 ­0.62 ­0.003 ­0.041 ­0.63 Born in Aug-Sep x rural ­0.007 ­0.093 ­1.52 ­0.008 ­0.098 ­1.60 ­0.008 ­0.098 ­1.61 ­0.008 ­0.098 ­1.61 Born in Oct-Nov x rural ­0.003 ­0.035 ­0.62 ­0.003 ­0.038 ­0.68 ­0.003 ­0.038 ­0.68 ­0.003 ­0.039 ­0.69 Period 1995­99 ­0.005 ­0.063* ­1.72 ­0.006 ­0.075** ­2.04 ­0.006 ­0.073** ­1.99 ­0.006 ­0.075** ­2.05 Constant ­0.346 ­1.21 ­0.204 ­0.71 ­0.201 ­0.70 ­0.129 ­0.45 # Obs 18,646 18,646 18,646 18,646 # Deaths 876 876 876 876 Wald chi-squared 318.5*** 329.5*** 330.7*** 334.2*** Log-likelihood ­3,272.9 ­3,262.2 ­3,261.7 ­3,261.0 Pseudo R-squared 0.060 0.063 0.063 0.063 Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively; + reference category; x interactive term; dF/dx marginal effect. 7 4 N E O N A T A L , P O S T N A T A L , A N D C H I L D M O R T A L I T Y I N T H E 1 9 9 0 s contraceptive methods known by the mother-- piped water into the dwelling and sanitary toilet a proxy for general knowledge of modern health facilities (septic tank). In addition, newborns liv- and child care practices--which is highly signifi- ing in electrified dwellings have significantly bet- cantly negative; schooling remains significant ter survival chances than others; this result is (though its estimated impact is marginally re- significant in specifications including a wide va- duced). This suggests that the impact of mater- riety of socioeconomic controls such as wealth nal formal schooling on neonate health operates and parental education. However, when rural outside knowledge, though it is possible that and urban regressions are estimated separately, maternal education is approximating household electricity is significant in urban only (results not socioeconomic status, if the wealth index mea- reported), which suggests electricity may be sures this inaccurately. Indicators of paternal ed- proxying for factors such as location in a slum. ucational attainment were added, in order to Newborns whose mothers had one or more isolate the impact of socioeconomic status from antenatal visit are estimated to have significantly maternal education with more certainty (spec- lower mortality rates. Children delivered by ification 3). Inclusion of paternal schooling TBAs had higher survival chances, while a posi- enhances significance of maternal schooling, tive impact on survival of trained TBAs was possibly by indicating educated women's en- found for the later period (late 1990s) only, sug- hanced decision making in the household. gesting that quality of care provided by trained- Socioeconomic status is a less important de- TBAs improved over time. However, as noted terminant of neonatal mortality risk: there is no above, these coefficients may overstate the im- estimated impact of wealth (nor of paternal ed- pact of health care on survival. In the case of de- ucation).18 Similarly, location (division of resi- livery assistance, home delivery assisted by a dence, rural/urban) does not exert an important TBA is the norm--over 60 percent of children effect on neonatal survival chances, once con- are delivered in this way--but the decision to trolling for other factors.19 Interactions between use a TBA may be affected by whether she is per- division of residence and gender of child were ceived to deliver better quality services due to included in order to test for regional differences modern training. The results of re-estimation in treatment of girls and boys (table C.2, specifi- using a trivariate probit model to account for the cation 4). The results provide some evidence of endogeneity are presented in table C.3. gender discrimination, since newborns are sig- The probit for one or more antenatal visits is nificantly less likely to die in Khulna, the wealth- relatively well fit (pseudo-R-squared=0.17) and iest division by the late 1990s, though only if identifying instruments are highly significant: they are boys, while newborn girls in Chittagong presence of a satellite clinic offering antenatal are significantly more likely to die than boys. care in the community significantly increases ac- Rural seasonality is also a significant factor be- cess to antenatal care, as does access to a com- hind neonatal survival, which is most likely for munity television in rural areas, which is a proxy a child born in April-May (which follows the for information transfer; distances from Upazilla boro rice harvest and is a period of high food and district headquarters reduce the likelihood availability) or in August-September (when ma- that women receive antenatal care; in addition, ternal calorie deficiency is least severe); in con- the non-self mean of the dependent variable is trast neonatal mortality risk is highest during highly significant. Turning to the remaining ex- December-January (when maternal calorie defi- planatory variables, pregnant women are more ciency is also highest) (see Bloem and others likely to demand antenatal care if they are edu- 2003). cated, if they are older, if they have good knowl- Services such as clean water and sanitation, edge of modern health practices (proxied by electrification, and antenatal and delivery care knowledge of contraception), when they are are key determinants of neonatal survival. identified as the head of the household, and Newborns have significantly better survival when they are more freely mobile. Short pre- chances when they live in households with both ceding birth interval has a negative impacts on 7 5 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Determinants of Neonatal Death, Antenatal Visits, and Delivery T A B L E C . 3 by Trained TBA: Trivariate Probit Estimates (1) (2) Neonatal Antenatal TTBA Neonatal Antenatal TTBA Death Visit Delivery Death Doctor/Nurse Delivery Coeff. z Coeff. z Coeff. z Coeff. z Coeff. z Coeff. z Wealth ­0.028 ­0.27 0.574*** 8.17 0.223** 2.32 0.016 0.15 0.642*** 9.52 0.223** 2.32 Electricity ­0.115** ­2.02 0.252*** 6.82 0.086* 1.67 ­0.098* ­1.71 0.266*** 7.58 0.086* 1.67 Piped water & sanitary toilet ­0.504*** ­2.73 ­0.483*** ­2.61 Mother primary ­0.049 ­1.09 0.137*** 4.23 0.070 1.60 ­0.039 ­0.87 0.168*** 5.46 0.070 1.59 Mother lower secondary ­0.023 ­0.34 0.382*** 8.44 0.163*** 2.84 0.014 0.21 0.432*** 10.14 0.162*** 2.82 Mother upper secondary or more ­0.346* ­1.80 0.682*** 6.58 ­0.312** ­2.37 ­0.274 ­1.42 0.728*** 7.85 ­0.311** ­2.37 Contraceptive knowledge ­0.054*** ­4.34 0.093*** 10.02 0.053*** 4.37 ­0.049*** ­3.94 0.100*** 11.41 0.053*** 4.33 Multiple birth 1.208*** 10.65 0.011 0.09 0.298** 2.12 1.200*** 10.58 ­0.231* ­1.78 0.303** 2.16 Age at birth ­0.044** ­2.03 0.021*** 5.87 0.011** 2.30 ­0.042* ­1.92 0.022*** 6.49 0.011** 2.30 Age at birth sq 0.001* 1.83 0.001* 1.78 Prev child died 0.190*** 3.15 0.054 1.30 ­0.006 ­0.11 0.191*** 3.15 0.017 0.39 ­0.005 ­0.09 Prec interval <15 0.617*** 6.46 ­0.165* ­1.79 0.617*** 6.40 ­0.104 ­1.10 Prev child died x prec interval ­0.490*** ­3.18 0.284** 2.16 ­0.480*** ­3.11 0.276** 2.04 Birth order ­0.184*** ­5.33 ­0.146*** ­6.44 ­0.048*** ­3.08 ­0.193*** ­5.56 ­0.130*** ­5.76 ­0.048*** ­3.07 Birth order sq 0.015*** 5.10 0.003 1.52 0.016*** 5.18 0.002 1.05 Female child ­0.123*** ­3.58 0.021 0.93 ­0.001 ­0.05 ­0.123*** ­3.59 0.015 0.64 ­0.001 ­0.04 Female head of hh ­0.049 ­0.60 0.113** 2.02 0.133* 1.85 ­0.041 ­0.49 0.133** 2.47 0.134* 1.86 Mother's mobility ­0.059** ­2.08 0.102*** 4.82 ­0.056** ­1.98 0.070*** 3.36 Mother divorced ­0.007 ­0.05 ­0.121 ­1.06 ­0.012 ­0.09 ­0.131 ­1.15 Mother remarried ­0.028 ­0.37 ­0.086 ­1.54 ­0.030 ­0.40 ­0.064 ­1.12 Antenatal visits ­0.031 ­0.47 Antenatal visit by doctor/nurse ­0.278*** ­4.01 TBA delivery ­0.074** ­2.02 ­0.074** ­2.01 TTBA delivery ­0.015 ­0.11 0.007 0.05 TTBA delivery x period 1995­99 ­0.234 ­1.53 ­0.245 ­1.59 Chittagong ­0.031 ­0.46 0.020 0.42 ­0.025 ­0.39 ­0.029 ­0.43 0.030 0.66 ­0.027 ­0.41 Dhaka ­0.032 ­0.48 ­0.009 ­0.19 0.009 0.14 ­0.032 ­0.48 0.026 0.60 0.008 0.12 Khulna ­0.080 ­1.07 0.086 1.64 ­0.268*** ­3.54 ­0.079 ­1.05 0.076 1.54 ­0.271*** ­3.59 Rajshahi ­0.037 ­0.54 0.084* 1.75 ­0.093 ­1.41 ­0.037 ­0.54 0.045 0.98 ­0.094 ­1.43 Rural ­0.003 ­0.04 ­0.303*** ­7.41 0.157*** 2.71 ­0.029 ­0.43 ­0.271*** ­6.99 0.158*** 2.72 Born in Feb-Mar x Rural ­0.070 ­1.10 ­0.067 ­1.05 Born in Apr-May x Rural ­0.161** ­2.32 ­0.159** ­2.28 Born in Jun-Jul x Rural ­0.042 ­0.63 ­0.041 ­0.62 Born in Aug-Sep x Rural ­0.098 ­1.61 ­0.093 ­1.53 Born in Oct-Nov x Rural ­0.039 ­0.69 ­0.034 ­0.60 Period 1995­99 ­0.075** ­2.03 0.240*** 9.15 0.239*** 6.72 ­0.072* ­1.94 0.082*** 3.25 0.238*** 6.71 Distance to Thana hq ­0.013*** ­3.69 ­0.021*** ­4.43 ­0.011*** ­3.21 ­0.022*** ­4.46 Distance to District hq ­0.005*** ­3.80 ­0.003* ­1.81 ­0.003** ­1.99 ­0.003* ­1.83 Antenatal clinic 0.107*** 3.98 0.103*** 2.92 0.088*** 3.41 0.102*** 2.88 Community tv x rural 0.073** 2.34 0.056* 1.83 Community tv ­0.273*** ­6.80 ­0.273*** ­6.80 Mean-antenatal visit 1.379*** 7.58 0.776*** 5.19 Constant ­0.210 ­0.74 ­1.494*** ­12.20 ­2.134*** ­13.98 ­0.218 ­0.76 ­1.731*** ­14.59 ­2.132*** ­13.91 Rho (2,1) ­0.035 ­1.29 0.001 0.02 Rho (3,1) 0.056 1.49 0.058 1.52 Rho (3,2) 0.172*** 8.17 0.151*** 7.37 # Obs 18,646 18,646 # Deaths 876 876 Wald chi-squared 3,172.2*** 3,188.0*** Log likelihood ­5361.0 ­5169.0 Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively; X interactive term. 7 6 N E O N A T A L , P O S T N A T A L , A N D C H I L D M O R T A L I T Y I N T H E 1 9 9 0 s demand for antenatal care, though the positive correlated between antenatal care and neonatal coefficient on the interaction between previous mortality equations (RHO 2,1) and positively interval and death of previous sibling indicates correlated between trained-TBA and neonatal that women with fewer young children are mortality equations (RHO 3,1), though the coef- better able to attend the clinics. Finally, women ficients are marginally insignificant.21 The in- demand less antenatal care the higher the child's significance of antenatal care may be because birth order, which is to be expected as they quality of service is not being accounted for--it have experience from previous births. House- is highly unlikely that antenatal care has no im- hold variables--electrification and, particularly, pact on neonatal survival. Table C.3 specification wealth--exert a strong positive influence on de- (2) accounts for quality of antenatal care by mand for antenatal care. Finally, there is evi- re-estimating the trivariate probit where the sec- dence that pregnant women are more likely to ond equation models antenatal visits to a quali- access services in the Divisions of Khulna and fied doctor or nurse. The fit of the probit for Rajshahi and during the mid-late 1990s and gen- antenatal visit attended by doctor or nurse is erally less likely to access them in rural areas.20 again high (pseudo-R-squared=0.16) and iden- Identifying variables are also highly significant tifying instruments are highly significant, but, in in the probit for trained-TBA-assisted delivery contrast to above, accounting for quality of an- (pseudo-R-squared=0.05): presence of a com- tenatal care and potential endogeneity with re- munity antenatal clinic increases the likelihood spect to child health, antenatal visits (with of receiving the delivery assistance of a trained qualified doctor/nurse) significantly reduce risk TBA, while distances from Upazilla and District of neonatal mortality.22 headquarters and access to community tele- vision (by promoting use of professional atten- Postnatal Mortality dants, i.e., doctors, nurses, midwifes, family For the postnatal infant period, services are less welfare officers) reduce it. The remaining ex- prominent determinants of mortality risk (table planatory variables are of the expected sign: C.4). Access to electricity and clean water and household-level variables such as wealth, elec- sanitation are not significantly correlated with tricity, rural location, female headship, mother's postnatal mortality. Whether the child received age and contraceptive knowledge are positively vitamin A supplementation in the past six correlated with trained-TBA delivery incidence. months, proxied by the (non-self) share of chil- Multiple births are also more likely to be deliv- dren in the community receiving vitamin A, has ered by trained TBAs. Children in Khulna Divi- the expected negative effect on postnatal mor- sion are less likely to be delivered by trained tality, though is marginally insignificant. Child- TBAs, but the incidence of delivery by health level factors such as multiple birth, short professionals is highest there. For maternal edu- preceding birth interval, and mother's age of cation, while mother's education to lower delivery have a negative impact on postnatal secondary level is associated with greater prob- mortality. However, it is also clear that socioe- ability of receiving delivery assistance from a conomic level is an important determinant of trained TBA, women who are educated beyond postnatal mortality, demonstrated most obvi- this level are less likely to. ously by the significance of wealth, as well as the The trivariate probit specification does not estimated linear negative effect of maternal de- significantly alter the estimated effect of trained livery age and the observation that the effect of TBA on neonatal mortality, which remains mar- fate of the preceding child is entirely through ginally insignificant in the latter period. How- the interaction with the preceding interval, both ever, the coefficient on antenatal care in the of which suggest competition for resources neonatal mortality regression becomes insignifi- among siblings increases postnatal mortality. cant when using the multivariate probit specifi- Of mother-level variables, coefficients on ma- cation (table C.3 specification 1). The estimates ternal mobility index and marital status are in- of Ú (RHO) indicate that errors are negatively significant. Maternal educational attainment to 7 7 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? T A B L E C . 4 Postnatal Mortality Determinants: Probit Estimates (1) (2) (3) (4) dF/dx Coeff. z dF/dx Coeff. z dF/dx Coeff. z dF/dx Coeff. z Wealth ­0.019 ­0.320** ­2.04 ­0.018 ­0.301* ­1.92 ­0.015 ­0.248 ­1.55 ­0.010 ­0.167 ­1.01 Electricity 0.002 0.027 0.38 0.002 0.033 0.45 0.002 0.035 0.49 0.003 0.042 0.58 Piped water & sanitary toilet ­0.003 ­0.053 ­0.25 ­0.004 ­0.072 ­0.35 ­0.005 ­0.083 ­0.40 ­0.003 ­0.059 ­0.28 Mother no formal ed + Mother primary 0.003 ­0.048 ­0.84 ­0.002 ­0.032 ­0.55 ­0.002 ­0.032 ­0.55 0.000 0.002 0.04 Mother lower secondary ­0.015 ­0.314*** ­3.15 ­0.014 ­0.286*** ­2.85 ­0.014 ­0.282*** ­2.81 ­0.009 ­0.162 ­1.63 Mother upper secondary or more ­0.019 ­0.476 ­1.59 ­0.017 ­0.428 ­1.43 ­0.017 ­0.428 ­1.44 ­0.007 ­0.129 ­0.43 Contraceptive knowledge ­0.003 ­0.045*** ­2.72 ­0.003 ­0.045*** ­2.72 ­0.003 ­0.044*** ­2.64 Father no formal ed + father primary ­0.005 ­0.086 ­1.52 Father lower secondary ­0.007 ­0.131* ­1.83 Father upper secondary or more ­0.018 ­0.427*** ­2.79 Multiple birth 0.149 1.020*** 8.30 0.150 1.028*** 8.36 0.149 1.025*** 8.33 0.149 1.031*** 8.36 Age at birth ­0.001 ­0.013* ­1.91 ­0.001 ­0.013** ­1.98 ­0.001 ­0.014** ­2.00 ­0.001 ­0.012* ­1.82 Prev child died 0.002 0.031 0.40 0.002 0.026 0.34 0.001 0.021 0.27 0.001 0.017 0.22 Prec interval <15 0.067 0.636*** 5.59 0.067 0.639*** 5.60 0.065 0.631*** 5.56 0.064 0.628*** 5.53 Prev child died x prec interval ­0.015 ­0.335* ­1.77 ­0.015 ­0.340* ­1.80 ­0.015 ­0.339* ­1.80 ­0.014 ­0.329* ­1.75 Birth order ­0.005 ­0.075** ­2.00 ­0.004 ­0.065* ­1.75 ­0.005 ­0.083** ­2.15 ­0.005 ­0.084** ­2.16 Birth order sq 0.001 0.013*** 4.26 0.001 0.012*** 3.97 0.001 0.012*** 3.98 0.001 0.012*** 3.95 Female child 0.004 0.062 1.38 0.004 0.060 1.32 ­0.004 ­0.060 ­0.76 ­0.004 ­0.060 ­0.75 Female x birth order 0.002 0.035* 1.88 0.002 0.034* 1.85 Female x older sister 0.011 0.159 1.43 0.011 0.160 1.44 Female x older sister x wealth ­0.039 ­0.656* ­1.70 ­0.038 ­0.656* ­1.68 Female head of hh 0.007 0.117 1.26 0.007 0.116 1.24 0.007 0.110 1.18 0.007 0.120 1.28 Mother's mobility 0.000 ­0.002 ­0.05 0.001 0.010 0.26 0.001 0.009 0.25 0.001 0.009 0.24 Mother divorced 0.015 0.202 1.37 0.015 0.205 1.38 0.015 0.214 1.44 0.015 0.210 1.41 Mother remarried 0.006 0.098 1.13 0.006 0.098 1.13 0.006 0.100 1.14 0.006 0.096 1.10 Barisal + Chittagong ­0.006 ­0.104 ­1.30 ­0.008 ­0.143* ­1.74 ­0.008 ­0.145* ­1.76 ­0.008 ­0.153* ­1.86 Dhaka ­0.003 ­0.051 ­0.63 ­0.004 ­0.065 ­0.80 ­0.004 ­0.068 ­0.84 ­0.005 ­0.083 ­1.02 Khulna ­0.013 ­0.276*** ­2.69 ­0.013 ­0.278*** ­2.71 ­0.013 ­0.280*** ­2.72 ­0.014 ­0.296*** ­2.89 Rajshahi ­0.009 ­0.158* ­1.88 ­0.009 ­0.164* ­1.95 ­0.009 ­0.166** ­1.98 ­0.009 ­0.176** ­2.09 Rural ­0.013 ­0.189** ­2.16 ­0.013 ­0.189** ­2.16 ­0.013 ­0.191** ­2.17 ­0.013 ­0.190** ­2.15 Born in Jan-Feb x rural + Born in Feb-Mar x rural ­0.002 ­0.029 ­0.34 ­0.002 ­0.029 ­0.35 ­0.002 ­0.035 ­0.42 ­0.002 ­0.038 ­0.45 Born in Apr-May x rural 0.009 0.129 1.46 0.008 0.128 1.45 0.008 0.127 1.43 0.008 0.122 1.38 Born in Jun-Jul x rural 0.005 0.082 0.92 0.005 0.085 0.95 0.005 0.085 0.96 0.005 0.081 0.91 Born in Aug-Sep x rural 0.001 0.024 0.29 0.001 0.024 0.29 0.001 0.024 0.29 0.001 0.023 0.28 Born in Oct-Nov x rural 0.010 0.150** 2.03 0.010 0.151** 2.04 0.010 0.150** 2.02 0.010 0.148** 2.00 Mean-vitamin A ­0.008 ­0.139 ­1.45 ­0.008 ­0.132 ­1.36 ­0.008 ­0.134 ­1.38 ­0.008 ­0.135 ­1.40 Period 1995­99 ­0.003 ­0.054 ­1.17 ­0.004 ­0.066 ­1.43 ­0.004 ­0.065 ­1.42 ­0.004 ­0.075 ­1.61 Constant ­1.434*** ­7.04 ­1.277*** ­6.07 ­1.218*** ­5.67 ­1.222*** ­5.67 # Obs 14,285 14,285 14,285 14,285 # Deaths 451 451 451 451 Wald chi-squared 196.1*** 202.6*** 209.4*** 214.7*** Log-likelihood ­1,865.6 ­1,861.1 ­1,857.6 ­1,853.0 Pseudo R-squared 0.058 0.060 0.062 0.064 Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively; + reference category; x interactive term; dF/dx marginal effect. 7 8 N E O N A T A L , P O S T N A T A L , A N D C H I L D M O R T A L I T Y I N T H E 1 9 9 0 s lower secondary level leads to lower mortality (2) reports results for rural areas only and in- risk in postnatal infancy, as does mother's health cludes a variable indicating whether the house- knowledge, as proxied by knowledge of contra- hold was situated in a Bangladesh Integrated ceptives (specification 2). However, the magni- Nutrition Project (BINP) area. While being in a tude and significance of maternal education is BINP area does not significantly reduce infant reduced when paternal education is included mortality, the interaction with mother's schooling (specification 4). suggests that better educated mothers are more Girls of high birth order are more likely to be able to benefit from the BINP project, e.g., be- discriminated against and the interaction be- cause they are more likely to understand child tween gender of child and sex composition of growth charts and to follow educational messages elder children provides some evidence that girls such as beneficial food practices (see Annex G).25 have higher mortality risk if they have living In neither specification is mother's agency signif- elder sisters at birth, though only among poorer icantly correlated with infant mortality. households (specification 3). There is some locational variation in postnatal Child Mortality mortality in multivariate analysis, as children in Table C.6 presents multivariate analysis of child rural areas and in Khulna and Rajshahi have lower mortality using Cox proportional hazards estimated mortality risk when controlling for model.26 The model is stratified by poverty sta- other factors.23 The negative estimated impact of tus (where poverty is defined using a threshold urban areas on postnatal survival may reflect that, of the wealth index), allowing the baseline haz- accounting for socioeconomic factors and ser- ard to differ between poor and non-poor.27 The vices, urban areas are more risky environments most significant factors explaining differentials for infants, e.g., due to increased exposure to dis- in child mortality are household wealth, vacci- ease in areas of high population density, greater nations, and maternal education. likelihood of poor communities being directly Coefficients on (non-self) community shares adversely affected by annual floods (e.g., because of children receiving vitamin A supplementation they are forced to live on river-banks), and due to and having no vaccinations have the expected greater environmental pollution. Seasonality is signs, indicating that access to health services again a significant factor in rural areas: postnatal has a direct positive effect on child survival infants are more likely to die if born in the prospects. Formal education of mothers reduces months of October and November, which is the child mortality risk, but in contrast with results counterpart to the higher neonatal mortality ob- for infants, the effect of mother's knowledge of served in December-January. modern contraception is negative but insignifi- It is possible to account for (self-reported) an- cant in the child mortality regressions (spec- nual food availability at household level and ification 2); this may indicate that contraceptive mother's recollection of birth weight in the knowledge is a better representation of knowl- 1999/2000 survey round, as well as mother's edge of good care practices for infants regarding agency, using an index constructed from health care and feeding. Access to piped drink- mother's reported decisionmaking in the house- ing water and sanitary toilet facilities negatively hold.24 Table C.5 presents results for 1999/2000 correlates with child mortality, though, some- survey data for infant mortality, which pools what surprisingly, this effect is insignificant. observations for neonates and postnates in order However, the interaction between schooling to maximize number of observed deaths. Spec- and water and sanitation is significantly negative, ification (1) indicates that low birth weight (mod- indicating complementarities in the improve- erately small and very small) children are ment of child health status: intuitively, access to significantly more likely to die in the infant pe- piped water and clean sanitation is unlikely to riod, but whether the household experiences improve survival chances alone, mothers must food deficit during some or all of the year has no know how to use facilities properly and children significant effect on infant mortality. Specification must also learn to use them, which is more likely 7 9 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? T A B L E C . 5 Infant Mortality Determinants in 1999: Probit Estimates (1) (2 Rural) dF/dx Coeff. z dF/dx Coeff. z Wealth ­0.017 ­0.156 ­0.88 ­0.009 ­0.083 ­0.40 Electricity ­0.007 ­0.066 ­0.78 ­0.009 ­0.083 ­0.82 Piped water & sanitary toilet ­0.017 ­0.178 ­0.66 Mother no formal ed + Mother primary ­0.016 ­0.156** ­2.14 ­0.018 ­0.171** ­2.02 Mother lower secondary ­0.012 ­0.111 ­1.08 ­0.001 ­0.013 ­0.12 Mother upper secondary or more ­0.043 ­0.621** ­2.36 ­0.044 ­0.691 ­1.59 Contraceptive knowledge ­0.009 ­0.081*** ­3.97 ­0.011 ­0.097*** ­4.14 Multiple birth 0.341 1.352*** 7.77 0.362 1.418*** 7.34 Age at birth ­0.004 ­0.038 ­1.09 ­0.002 ­0.018 ­0.45 Age at birth sq 0.000 0.001 1.04 0.000 0.000 0.44 Prev child died 0.000 0.002 0.02 ­0.006 ­0.054 ­0.42 Prec interval <15 0.111 0.642*** 4.05 0.102 0.614*** 3.25 Prev child died x prec interval ­0.027 ­0.302 ­1.10 ­0.014 ­0.143 ­0.47 Birth order ­0.022 ­0.200*** ­3.45 ­0.024 ­0.218*** ­3.27 Birth order sq 0.002 0.017*** 3.26 0.002 0.018 3.10 Female child ­0.009 ­0.082 ­1.46 ­0.005 ­0.049 ­0.77 Female head of hh 0.006 0.055 0.44 0.001 0.008 0.06 Mother's mobility ­0.001 ­0.005 ­0.09 0.004 0.034 0.49 Mother's agency 0.000 0.004 0.37 0.001 0.006 0.52 Barisal + Chittagong ­0.010 ­0.091 ­0.84 ­0.011 ­0.100 ­0.83 Dhaka ­0.005 ­0.046 ­0.41 ­0.009 ­0.085 ­0.68 Khulna ­0.024 ­0.258** ­2.05 ­0.025 ­0.269* ­1.88 Rajshahi ­0.009 ­0.080 ­0.70 ­0.008 ­0.077 ­0.61 Rural ­0.013 ­0.113 ­1.16 Born in Jan-Feb x rural + Born in Feb-Mar x rural ­0.027 ­0.287** ­2.45 ­0.027 ­0.291** ­2.47 Born in Apr-May x rural 0.001 0.006 0.06 0.000 0.003 0.03 Born in Jun-Jul x rural 0.000 ­0.002 ­0.01 0.001 0.006 0.05 Born in Aug-Sep x rural ­0.003 ­0.025 ­0.25 ­0.001 ­0.009 ­0.08 Born in Oct-Nov x rural ­0.003 ­0.027 ­0.28 ­0.002 ­0.021 ­0.23 Mean-vitamin A ­0.018 ­0.164 ­1.49 ­0.012 ­0.106 ­0.83 Food deficit 0.005 0.041 0.58 0.005 0.047 0.60 Low birth weight 0.023 0.192*** 2.87 0.025 0.209*** 2.77 BINP 0.022 0.180 1.01 BINP x mother's schooling ­0.055 ­0.509*** ­2.87 Constant ­0.141 ­0.300 ­0.491 ­0.94 # Obs 5,626 4,172 # Deaths 382 289 Wald chi-squared 175.2*** 154.3*** Log-likelihood ­1,272.5 ­939.7 Pseudo R-squared 0.080 0.087 Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively; + reference category; x interactive term. 8 0 N E O N A T A L , P O S T N A T A L , A N D C H I L D M O R T A L I T Y I N T H E 1 9 9 0 s T A B L E C . 6 Child Mortality Determinants: Cox Regression Estimates (1) (2) (3) (4) (5) Haz. Haz. Haz. Haz. Haz. ratio Coeff. z ratio Coeff. z ratio Coeff. z ratio Coeff. z ratio Coeff. z Wealth 0.269 ­1.313*** ­3.03 0.278 ­1.280*** ­2.98 0.279 ­1.278*** ­2.98 0.313 ­1.161*** ­2.71 0.272 ­1.301*** ­3.02 Electricity 1.044 0.043 0.31 1.041 0.040 0.29 1.041 0.040 0.28 1.053 0.052 0.37 1.021 0.021 0.15 Piped water & sanitary toilet 0.683 ­0.381 ­0.64 1.797 0.586 0.98 1.814 0.596 0.99 1.840 0.610 1.01 1.769 0.570 0.93 Mother no schooling + Mother primary 0.836 ­0.179 ­1.58 0.853 ­0.159 ­1.39 0.860 ­0.151 ­1.32 0.914 ­0.090 ­0.78 0.834 ­0.181 ­1.57 Mother lower secondary 0.747 ­0.291 ­1.52 0.789 ­0.237 ­1.25 0.811 ­0.210 ­1.10 0.957 ­0.044 ­0.21 0.731 ­0.313* ­1.66 Mother higher secondary or more 0.111 ­2.198** ­2.14 0.147 ­1.914* ­1.89 0.157 ­1.849* ­1.83 0.216 ­1.533 ­1.49 0.121 ­2.112** ­2.09 Water & sanitary x mother ed 0.281 ­1.270** ­1.98 0.277 ­1.283** ­2.00 0.284 ­1.260* ­1.95 0.276 ­1.287** ­1.98 Contraceptive knowledge 0.966 ­0.035 ­1.08 0.961 ­0.039 ­1.22 0.964 ­0.037 ­1.15 0.964 ­0.036 ­1.13 Age at birth 0.971 ­0.029** ­2.29 0.971 ­0.029** ­2.32 0.967 ­0.034*** ­2.58 Birth order 1.106 0.100*** 2.94 1.105 0.100*** 2.94 1.090 0.087** 2.10 0.837 ­0.177 ­1.59 Female 1.284 0.250*** 2.78 1.287 0.252*** 2.81 0.509 ­0.675** ­1.99 0.788 ­0.238 ­1.56 Female x birth order 1.279 0.246** 2.56 0.603 ­0.507* ­1.76 Female x birth order sq 0.978 ­0.022** ­2.32 0.969 ­0.032** ­2.49 Previous child died 1.317 0.275** 2.35 1.312 0.272** 2.31 1.283 0.249** 2.11 1.086 0.083** 2.02 Maternal mobility 0.984 ­0.016 ­0.22 0.994 ­0.006 ­0.08 0.988 ­0.012 ­0.16 0.504 ­0.685** ­2.02 0.836 ­0.179 ­0.64 Mother divorced 1.450 0.372 1.55 1.436 0.362 1.51 1.460 0.378 1.58 1.284 0.250*** 2.59 Mother Remarried 1.369 0.314** 1.99 1.372 0.317** 2.00 1.393 0.331** 2.09 0.978 ­0.022** ­2.33 Barisal + Chittagong 0.980 ­0.021 ­0.13 0.954 ­0.047 ­0.30 0.718 ­0.332 ­1.59 0.985 ­0.015 ­0.20 0.992 ­0.008 ­0.10 Dhaka 0.796 ­0.228 ­1.39 0.789 ­0.237 ­1.44 0.651 ­0.429** ­1.96 1.456 0.376 1.56 1.404 0.339 1.42 Khulna 0.438 ­0.826*** ­3.51 0.438 ­0.827*** ­3.52 0.444 ­0.812*** ­2.58 1.385 0.326** 2.06 1.300 0.262* 1.66 Rajshahi 0.660 ­0.416*** ­2.44 0.659 ­0.417** ­2.45 0.503 ­0.688*** ­2.89 Female x Chittagong 1.735 0.551* 1.75 0.715 ­0.335 ­1.60 0.717 ­0.332 ­1.59 Female x Dhaka 1.496 0.403 1.22 0.638 ­0.449** ­2.04 0.642 ­0.444** ­2.02 Female x Khulna 1.008 0.008 0.02 0.436 ­0.829*** ­2.63 0.443 ­0.814*** ­2.59 Female x Rajshahi 1.702 0.532 1.52 0.498 ­0.697*** ­2.92 0.500 ­0.692*** ­2.91 Rural 0.929 ­0.073 ­0.48 0.929 ­0.074 ­0.49 0.926 ­0.077 ­0.51 1.737 0.552* 1.76 1.748 0.559* 1.78 Mean-no vaccination 1.579 0.457*** 2.57 1.551 0.439** 2.45 1.550 0.439** 2.46 1.509 0.412 1.25 1.476 0.389 1.18 Mean-vitamin A 0.638 ­0.449** ­2.19 0.639 ­0.448** ­2.17 0.653 ­0.427** ­2.07 1.009 0.009 0.02 0.961 ­0.040 ­0.09 Period 1995­99 0.649 ­0.432*** ­3.27 0.642 ­0.443*** ­3.35 0.645 ­0.439*** ­3.32 1.714 0.539 1.54 1.645 0.498 1.42 # Obs 22,527 22,527 22,527 0.932 ­0.071 ­0.47 0.932 ­0.070 ­0.46 # Deaths 558 558 558 1.543 0.434** 2.43 1.568 0.450** 2.52 Wald chi-squared 129.1*** 130.4*** 144.4*** 0.657 ­0.420** ­2.04 0.625 ­0.470** ­2.26 Log-likelihood ­5206.0 ­5203.6 ­5197.5 0.638 ­0.450*** ­3.40 0.627 ­0.467*** ­3.54 Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively; + reference category; x interactive term. where their mothers are educated. Maternal ed- icantly higher chances of dying than boys be- ucation becomes insignificant when controlling tween age 1 and 5 years. Breaking down this ef- for paternal education (specification 4). Of the fect, it is girls of higher birth order who fare worse remaining mother-level variables, mobility is in- (though, as estimated by the quadratic term, the significant, but children of mothers who are re- birth order effect levels off after the sixth child) married have significantly higher mortality risk, (specification 3). Children living in Khulna and which may indicate children are at risk of being Rajshahi Divisions have significantly better sur- rejected by the new family. vival chances than elsewhere, while interactions The regressions indicate evidence for dis- with gender find that girls living in Chittagong crimination against girl children, who have signif- Division face the highest levels of discrimination. 8 1 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Maternal age at birth, birth order, and fate of tive impact on child mortality, though is margin- previous child are all significantly correlated ally insignificant, while low birth weight children with child mortality risk; maternal age has a lin- are also more likely to die in the post-infant ear positive relationship and birth order a linear period. There is some evidence that children of negative relationship with child mortality. These food deficit households are more likely to die, factors may reflect fertility and therefore be en- especially if they are female--this result is statis- dogenous to mortality. Thus, specification (5) tically significant in rural areas (specification 2). omits these variables, the main effect being to However, the equation including both wealth improve significance of education variables. and incidence of food deficit is over-specified: Finally, we re-estimate the child mortality the estimated coefficient on food deficit, with- model for 1999/2000 survey round, in order to out controlling for wealth, is positive and signif- include measures of food deficiency, low birth icant at less than 5 percent for both girls and weight, mother's agency, and effects of the BINP. boys (results not reported). Finally, the regres- Table C.7 indicates mother's agency has a nega- sion model for rural areas suggests that the ef- T A B L E C . 7 Child Mortality Determinants in 1999: Cox Regression Estimates (1) (2 Rural) Haz. ratio Coeff. z Haz. ratio Coeff. z Wealth 0.152 ­1.886*** ­3.06 0.105 ­2.259*** ­3.14 Electricity 0.983 ­0.017 ­0.09 1.112 0.106 0.51 Piped water & sanitary toilet 0.372 ­0.988 ­0.96 Mother no schooling + Mother primary 0.932 ­0.071 ­0.45 0.978 ­0.023 ­0.13 Mother secondary or higher 0.835 ­0.181 ­0.71 0.963 ­0.038 ­0.13 Contraceptive knowledge 0.947 ­0.054 ­1.13 0.952 ­0.049 ­0.94 Age at birth 0.961 ­0.039** ­2.26 0.962 ­0.038** ­2.00 Birth order 1.145 0.136** 2.41 1.163 0.151** 2.47 Female 0.763 ­0.271 ­0.76 0.681 ­0.383 ­0.94 Female x birth order 1.289 0.254* 1.83 1.320 0.277* 1.79 Female x birth order sq 0.973 ­0.027* ­1.93 0.970 ­0.031** ­1.96 Previous child died 0.872 ­0.137 ­0.74 0.843 ­0.171 ­0.84 Maternal mobility 0.979 ­0.022 ­0.18 0.975 ­0.025 ­0.19 Mother agency 0.964 ­0.037 ­1.60 0.962 ­0.039 ­1.51 Mother divorced 1.664 0.509 1.47 1.673 0.514 1.38 Mother remarried 1.440 0.365* 1.66 1.495 0.402* 1.72 Barisal + Chittagong 0.975 ­0.026 ­0.11 0.871 ­0.138 ­0.55 Dhaka 0.837 ­0.177 ­0.72 0.802 ­0.221 ­0.84 Khulna 0.517 ­0.660** ­2.06 0.531 ­0.634* ­1.86 Rajshahi 0.727 ­0.318 ­1.25 0.703 ­0.352 ­1.32 Rural 0.837 ­0.178 ­0.97 Mean-no vaccination 2.050* 0.718* 1.79 2.081 0.733* 1.64 Mean-vitamin A 0.950 ­0.052 ­0.16 0.890 ­0.117 ­0.30 Food deficit 1.300 0.263 1.32 1.501 0.406* 1.74 Food deficit x male 0.778 ­0.250 ­0.88 0.651 ­0.429 ­1.30 Low birth weight 1.550 0.438*** 2.60 1.597 0.468** 2.47 BINP 1.282 0.249 0.74 BINP x mother education 0.529 ­0.637** ­2.05 # Obs 11,493 8,522 # Deaths 249 196 Wald chi-squared 108.3*** 105.3*** Log-likelihood ­2,417.2 ­1,808.6 Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively; x interactive term. 8 2 N E O N A T A L , P O S T N A T A L , A N D C H I L D M O R T A L I T Y I N T H E 1 9 9 0 s fect of BINP on child mortality was dependent time. The decompositions for neonatal and post- on the level of maternal education: BINP had no natal mortality (table C.8) indicate that, in addi- effect on mortality for children, unless, as in- tion to the large (adverse) impacts of mother's dicated by the significantly negative interaction age at birth and birth order on mortality, expan- between BINP and mothers' schooling, their sion of electrification, antenatal care by trained mothers were educated. doctor or nurse, and share of mothers complet- ing secondary school were the biggest contribu- Decompositions tors to reducing neonatal mortality, while the Following Fairlie (2003), for the probit model improvement in wealth had the biggest impact on the total decomposition between two periods is reducing postnatal mortality. The decomposition estimated as: for child mortality (table C.9) indicates that im- provements in wealth, maternal secondary edu- Y2 - Y1 = (N N 2 X i2 )- (N N1 X i1 ). cation, and immunization have contributed the i=1 2 i=1 1 largest impact on reducing child mortality during the decade.29 That is, the change in mean probability of dying over time is calculated as the difference in Cost Effectiveness Analysis means of the predicted probabilities over the Analysis of cost effectiveness requires an esti- two sub-samples, corresponding to periods 1 mate of the actual number of lives saved by an in- and 2. However, in order to calculate the specific tervention. Table C.10 shows the actual numbers contributions to the change for an individual ex- that are used as the base case for the simula- planatory variable requires: tions. Data on population, crude birth rate, and 1 (Wi1 + N1 Zi2 - Wi1 + Zi1 ) ( ) infant and child mortality are taken from Khan N1 i=1 and Yoder (1998) and the World Development Indicators database. The crude birth rate is used where vector W and variable Z comprise the total to calculate the number of live births each year. explanatory variables set, X. The contribution of The infant mortality rate can be used to calculate variable Z to the total change over time is calcu- how many of those born in each year will die as lated as the change in the average predicted infants, giving total infant deaths. Where neces- probability from replacing the period 1 distribu- sary for the calculations it is assumed that half of tion with the period 2 distribution of variable Z, infant deaths are neonatal (figures not shown). holding constant the distributions of the other It is thus possible to calculate how many of the variables (Fairlie 2003: 4). Calculation of the de- children born that year survive to age one. Using composition requires matching observations the CMR it can be calculated how many of those from the two sub-samples; observations from children will die before their fifth birthday. both periods are matched according to the rank These deaths are spread over the four years in of predicted probabilities calculated from the re- the proportions, 0.5, 0.3, 0.15, and 0.05. Hence, gression estimates.28 the number of child deaths each year can be cal- For the proportional hazards model, Masset culated. These can be added to infant deaths to and White (2003) have shown that the percent- get under-five deaths (not shown). age contribution of each variable to the change The first simulation uses the result of the in predicted hazard can be calculated at the cross-country analysis that immunization ac- means of the data: counts for 30 percent of the decline in mortality Y2- 1 = exp ( Z2 - Z1 ) [ ] - 1 since the 1980s. Immunization coverage in- . Y1 creased markedly from 1987. The simulation as- sumes no reduction in mortality rates from 1988 Tables C.8 and C.9 present results of the de- (table C.11); the number of deaths are calculated compositions of coefficient estimates into their as for the base case. The difference in the num- shares of the change in predicted mortality over ber of under-five deaths between the two cases 8 3 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Decomposition (%) which is the figure used here. To use this infor- mation it is necessary to know the mortality rate T A B L E C . 8 for Neonatal and Postnatal Probit Models for unimmunized children. This can be done using the fact that the observed mortality rate is a weighted average of that for immunized and Neonatal Postnatal unimmunized children: Wealth 0.30 ­2.28 Electricity ­2.44 0.74 MR = i MRIMM + 1 - i MRUNIMM ( ) Piped water & sanitary toilet ­0.47 ­0.01 Mother primary ­0.02 0.02 where I is the rate of immunization coverage. Mother lower secondary 0.78 ­0.93 The odds ratio tells us that Mother upper secondary or higher ­1.51 ­0.31 Contraceptive knowledge 0.12 3.01 MRUNIMM 3 Multiple birth 0.85 4.13 = MRIMM 2 Age at birth 14.62 3.79 Age at birth sq 14.89 so given i (for which DPT coverage is used) and Prev child died ­0.88 ­0.14 MR, it is possible to calculate MRUNIMM. This mor- Prec interval <15 0.76 4.50 tality rate is used to calculate the number of Prev child died x prec interval 2.00 2.01 deaths in the absence of immunization for in- Birth order 39.12 21.88 fants and children separately (children not Birth order sq 2.17 13.52 shown). It is assumed that immunization only Female child 0.49 0.63 Female x birth order 0.18 affects postnatal deaths among infants. Female x older sister 0.94 The number of deaths averted provides the Female x older sister x wealth 0.28 denominator for the cost-effectiveness calcula- Female head of hh 0.00 0.08 tion. Budget data for the numerator are taken Mother's mobility ­0.78 0.20 from Khan and Yoder (1998). EPI expenditure Mother divorced 0.03 ­0.38 data are provided from the start of the program Mother remarried 0.07 ­0.05 until 1998. A more detailed calculation for 1997/ Antenatal visits by doctor/nurse ­4.27 98 captures all costs associated with immuniza- Chittagong ­0.26 ­0.40 tion services, including those outside of EPI, Dhaka 0.27 1.34 which are about 30 percent greater than the Khulna ­0.35 ­0.76 direct EPI costs. Two calculations are made: Rajshahi 0.28 2.19 Rural 0.79 5.72 (1) total immunization costs in 1998/deaths Born in Feb-Mar x rural 0.56 0.20 averted in 1998; and (2) program costs (from Born in Apr-May x rural 0.83 0.03 Khan and Yoder inflated for additional years and Born in Jun-Jul x rural 0.17 0.01 30 percent to capture immunization costs not Born in Aug-sep x rural 0.45 ­0.04 paid through EPI)/deaths averted over the whole Born in oct-Nov x rural 0.18 0.61 period. Mean-vitamin A ­0.65 The results of these calculations are summa- Note: x interactive term. rized in table C.12. The range is quite large, from just under US$100 to nearly US$300 per life is the number of deaths averted by the reduction saved. These results compare with the US$137 in mortality; 30 percent of these are attributed to reported by Khan and Yoder (1998). These cal- immunization. culations suggest that that result may be some- The second simulation is based on the DHS thing of an underestimate, but that a value of analysis, which finds that immunization reduces US$200 per life saved is not unreasonable. the probability of death by 50 percent. The haz- ard ratio for the Cox regression using 1999 data Conclusions was 2.05 (table C.7), but using pooled data, and The exercise to examine the determinants of a without-self mean of immunization to account mortality in the neonatal, postnatal, and child for potential endogeneity, the ratio falls to 1.55, periods has yielded some interesting results: 8 4 N E O N A T A L , P O S T N A T A L , A N D C H I L D M O R T A L I T Y I N T H E 1 9 9 0 s · Household incomes, as represented in this Decomposition for Child paper by wealth, exert a significant effect on T A B L E C . 9 Mortality: Proportional post-neonatal survival, particularly in post-in- Hazards Model fancy when income contributed the most of any one variable to reducing mortality over Late 80s Mid- Decomposition time. Coeff. ­early 90s late 90s (%) · Mothers' education is an important factor be- Predicted hazard at hind variation in mortality risk across children means 0.053 0.041 and over time. Mothers' education, partic- Wealth ­1.308*** 0.240 0.332 ­11.4 ularly at secondary or higher level, has a sig- Electricity 0.023 0.178 0.288 0.3 Piped water & 0.495 0.025 0.023 ­0.1 nificantly positive impact on child survival; sanitary toilet 0.495 0.025 0.023 ­0.1 this is observed over the impact of household Water & sanitary wealth, as well as mothers' knowledge of x mother ed ­0.787*** 0.059 0.058 0.1 modern contraception, an indicator of Mother some primary ­0.180 0.258 0.274 ­0.3 general health knowledge, which itself is a Mother compl primary ­0.290 0.131 0.174 ­1.3 Mother secondary significant determinant of infant mortality dif- or more ­2.20** 0.019 0.029 ­2.2 ferentials. The results suggest that mother's Contraceptive health knowledge has an important effect on knowledge ­0.035 3.663 3.767 ­0.4 survival chances, but also that children of bet- Female ­0.181 0.495 0.492 0.1 ter educated mothers are more likely to sur- Maternal mobility ­0.01 1.117 1.202 ­0.1 Mother divorced 0.335 0.019 0.024 0.1 vive. The latter may be due to more highly Mother remarried 0.253 1.056 1.061 0.1 educated mothers having greater say in Chittagong ­0.349* 0.313 0.287 0.1 household decisionmaking. Evidence also in- Dhaka ­0.46** 0.295 0.305 ­0.4 dicates that neonates have better survival Khulna ­0.834*** 0.109 0.103 0.5 chances when their mother is free to leave Rajshahi ­0.715*** 0.219 0.242 ­1.6 Female x Chittagong 0.561* 0.150 0.138 ­0.7 the dwelling and visit the health center, while Female x Dhaka 0.389 0.146 0.151 0.2 divorce and remarriage can be detrimental to Female x Khulna ­0.036 0.058 0.051 0.0 survival in childhood. Female x Rajshahi 0.502 0.109 0.118 0.5 · Health services contribute significantly to im- Rural ­0.054 0.899 0.832 0.4 proving childhood survival chances. Mortality Mean-no vaccination 0.447** 0.340 0.180 ­6.9 Mean-vitamin A ­0.466** 0.255 0.271 ­0.7 is lower among newborns whose mothers at- Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively; x interactive term. tended antenatal visits with a trained doctor or nurse and immunization had a strong im- pact on reducing mortality in childhood in Bangladesh in the 1990s. These effects are ro- girls are more likely to die in both the post- bust to use of methods that attempt to ac- natal and post-infant periods. Analysis of DHS count for endogeneity and unobserved 1999/2000 data indicated that girl children in heterogeneity in mothers' preferences. food deficit households are more likely to be · Access to piped drinking water and sanitary discriminated against. toilet facilities were found to have a signifi- · The analysis of DHS 1999/2000 survey round cant effect on reducing mortality among data showed that infants and children living in neonates, and among children in households BINP areas tended to have better survival where the mother has formal schooling. chances where their mothers were formally · The multivariate analysis indicates that educated, suggesting educated mothers were gender discrimination against girls remains more able to understand and act upon the strong. While the effect is quantitatively great- messages of the project and/or were more est among children, as would be expected, likely to take complementary actions such as the evidence suggests that higher birth order use modern medical care. 8 5 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? T A B L E C . 1 0 Under­Five Death Calculation Based on Actual Data (base case) Child Child Crude No. of Children deaths deaths Popu- birth children Infant alive at of this Year 1 Year 2 Year 3 Year 4 per Year lation IMR CMR rate born deaths age one cohort (.5) (.3) (.15) (.05) year 1981 89912 126 83.9 43.0 3866 485 3381 284 142 85 43 14 1982 92061 122 80.6 42.0 3867 472 3395 274 137 82 41 14 1983 94169 119 77.2 41.0 3861 458 3403 263 131 79 39 13 1984 96335 115 73.8 40.0 3853 444 3409 252 126 75 38 13 260 1985 98531 112 70.4 39.0 3843 430 3412 240 120 72 36 12 249 1986 100758 113 63.0 38.0 3829 431 3398 214 107 64 32 11 230 1987 103025 113 55.6 37.0 3812 431 3381 188 94 56 28 9 207 1988 105282 107 54.8 35.6 3744 402 3342 183 91 55 27 9 192 1989 107587 102 53.9 34.1 3671 373 3298 178 89 53 27 9 183 1990 109922 96 53.1 32.7 3592 345 3247 172 86 52 26 9 176 1991 111455 92 47.0 31.2 3482 320 3162 149 74 45 22 7 162 1992 113729 88 40.9 29.8 3389 298 3091 127 63 38 19 6 143 1993 115981 84 41.7 29.5 3417 287 3130 131 65 39 20 7 134 1994 118161 80 43.0 29.1 3441 274 3167 136 68 41 20 7 134 1995 120382 75 44.3 28.8 3465 260 3205 142 71 43 21 7 138 1996 122561 71 41.3 28.4 3486 247 3239 134 67 40 20 7 137 1997 124706 67 38.4 28.1 3504 233 3271 125 63 38 19 6 131 1998 126579 62 35.4 28.1 3559 222 3337 118 59 35 18 6 124 1999 128797 58 32.5 28.1 3624 211 3413 111 55 33 17 6 116 2000 131050 54 29.6 28.2 3690 199 3491 103 52 31 15 5 109 2001 133345 51 27.9 28.2 3758 192 3566 100 50 30 15 5 103 2002 135684 48 26.3 28.2 3826 184 3643 96 48 29 14 5 99 8 6 N E O N A T A L , P O S T N A T A L , A N D C H I L D M O R T A L I T Y I N T H E 1 9 9 0 s T A B L E C . 1 1 Mortality Simulations Simulation 1: no mortality reduction since 1985 Simulation 2: immunization reduces mortality by 50 percent Difference Difference Infant Child vs base IMR for IMR for Infant Child vs base Year IMR CMR deaths deaths case IMR immunized unimmunized CMR deaths deaths case 1981 126 84 485 126 84 126 84 487 305 0 1982 122 81 472 122 82 122 81 473 294 0 1983 119 77 458 119 79 119 77 460 281 0 1984 115 74 444 260 0 115 77 116 74 446 269 204 1985 112 70 430 249 0 112 75 113 70 433 257 198 1986 112 70 429 243 11 113 76 114 63 437 230 199 1987 112 70 427 240 29 113 78 116 56 444 206 198 1988 112 70 419 236 62 107 76 113 55 424 204 182 1989 112 70 411 233 88 102 81 122 54 446 222 155 1990 112 70 402 228 109 96 83 125 53 448 232 125 1991 112 70 390 222 130 92 81 122 47 425 204 107 1992 112 70 380 216 155 88 75 113 41 382 168 101 1993 112 70 383 214 176 84 74 112 42 381 179 96 1994 112 70 385 214 192 80 74 110 43 380 194 84 1995 112 70 388 215 206 75 65 97 44 337 190 79 1996 112 70 390 217 224 71 63 95 41 332 185 73 1997 112 70 392 218 246 67 60 90 38 315 173 67 1998 112 70 399 220 273 62 57 86 35 306 166 63 1999 112 70 406 224 302 58 53 80 33 289 155 59 2000 112 70 413 228 333 54 50 75 30 276 145 56 2001 112 70 421 232 358 51 47 71 28 265 140 54 2002 112 70 429 236 382 48 45 67 26 256 136 51 Decomposition for Child T A B L E C . 1 2 Mortality: Proportional Hazards Model Simulation 1 Simulation 2 1998 only 224 292 Whole program 192 90 8 7 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? APPENDIX C.1: Mortality Regression Results for Bangladesh Data Estimation Study Dataset Dependent variable Explanatory variables method Bairagi, Sutradhar, DSS 1966-94 Infant mortality Mother literate(--), boy(+++), muslim(---), mother's age (---), birth order(--) + Hazard and Alaml (1999) square(+++). Child mortality Mother literate(---), boy(---), muslim(+++), mother's age (--), birth order(+++) + Hazard square(---). Bhuiya and DSS 1975-89 Infant mortality Mother's education(---), mother's age(---), divorced(+++), girl(--), MCH project Logit (infant) Chowdhury (1997) area. Child mortality Mother's education(---), mother's age(+++), divorced(++), girl(+++), MCH project Hazard area(---). (child) Bhuiya and DSS 1982-84 Under-three mortality (chil- Mother's education (---), wealth(---), mother's age, girl child(+++), age of Hazard Streatfield (1991) dren surviving to 6 mos.) child(+++), boys with more highly educated mothers (-), girls with more highly ed- ucated mothers(+), MCH project area (Matlab) (---). Howlader and DHS 1996 Neonatal mortality Male(+++), mother aged over 20 (---), birth order(+++), birth interval over 18 mos Logit Bhuiyan (1999) (--), previous child died(+++), mother's education(--), urban, flush toilet, piped drinking water, electricity(---), landholding, antenatal care(--), assisted delivery, hospital delivery, immunization(---), tetanus vaccinatoin(+++), breastfed, visited by family planning worker. DHS 1996 Infant mortality Male(+++), mother aged over 20 (---), birth order(+++), birth interval over 18 mos Logit (--), previous child died(+++), mother's education(--), urban, flush toilet(---), piped drinking water, electricity(---), landholding, antenatal care(--), assisted de- livery, hospital delivery, immunization(---), tetanus vaccination(+++), breastfed(+), visited by family planning worker(--). DHS 1996 Child mortality Male(---), mother aged over 20 (---), birth order(---), birth interval over 18 mos, Logit previous child died(+++), mother's education(--), urban(---), flush toilet, piped drinking water, electricity, landholding(---), antenatal care(--), assisted delivery, hospital delivery, immunization(---), tetanus vaccination(-), breastfed(+++), visited by family planning worker(--). Kabir and Amin Bangladesh Prop. of dead children born Mother's education (---), father's education (---), father's oocupation (sig), reli- OLS (2003) Fertility Survey to each mother relative to gion, rural(+++), region (sig), water supply, toilet (---). (BFS) 1989 expected mortality rate (standardised by age of deaths) Miller and others DSS and Mortality among children Short conception interval interacted with higher birth order(+++), proportion of Hazard (1992) Determinants of <24 mos. previous children dead(+++), mother's age(---), mother's education(-), breastfeed- Natural Fertility ing(---). Survey (DNFS), Matlab 1983/84 Muhuri and DSS 1981/82 Child mortality Mother's education, wealth(---), boy with two or more older brothers and any sis- Logit Menken (1997) ters (+++), girls with 1 or more older sister and any brothers(+++), second child with older brother(+++), birth-to-conception interval(---), birth-to-conception in- terval interacted with girl sex by age of child(+++ for older girls), living in MCH project area (Matlab)(---), family with more schooling lives in control area(---). Muhuri and DSS 1981/82 Under-five mortality (chil- Mother's education(---), wealth(---), girl with one older brother, girls with two or Logit Preston (1991) dren surviving past 6 mos.) more older brothers(---), girl with one or more older sister(+++), birth order dum- mies, girl in poor household(-), born in MCH project area (Matlab)(---), girl born in Matlab. Razzaque and DSS Neonatal mortality Mother aged under 20-34(---), boy(+++), wealth, famine-born, famine-conceived, Logit others (1990) famine-born interacted with sex, wealth and mother's age. DSS Post-neonatal mortality Mother aged under 20-34, boy, wealth(---), famine-born(+++), famine- Logit conceived, famine-born interacted with sex, wealth and mother's age. DSS Child mortality Mother aged under 20-34, boy(---), wealth(---), famine-born, famine-conceived, Logit boys born in famine(+++). Zenger (1993) DSS 1968-82 Neonatal mortality Mother's education, mother under 20 years old(+++), mother over 29 years old(+), Logit girl child(---), Hindu(+++), birth order(+++), previous child died(+++), post-famine birth(---). child died(+++), post-famine birth(---). Notes: +++, ++, + indicates positive correlation at significance level <1%, <5%, <10%; ---, --, ­ indicates negative correlation at significance level <1%, <5%, <10%. 8 8 N E O N A T A L , P O S T N A T A L , A N D C H I L D M O R T A L I T Y I N T H E 1 9 9 0 s APPENDIX C.2: Measuring Socioeconomic Status: Wealth, Assets, and Welfare Outcomes in Bangladesh This paper reviews measures of socioeconomic is because economic indicators are relatively easy status (SES), with a particular focus on wealth. to measure--unlike concepts of social standing, The objective is to provide guidance in con- which are measured with much greater difficulty structing a suitable index, which can be used to if at all--and because social and economic indi- measure SES using Demographic and Health cators are assumed to be strongly correlated. Surveys (DHS). DHS collect data on asset own- However, most studies also account for factors ership at the household level, which can be con- that arguably reflect social standing to a greater verted into indicators of household wealth extent than income/wealth alone, including status. There are various ways of doing this, and human capital and/or occupation. indices have been constructed from different Household income and consumption expendi- types of assets using different estimation meth- ture are the most commonly used measures of odologies. However, attention to methodology living standards. Consumption and income are es- is important because alternative indices have timated from surveys, such as the World Bank- been shown not only to identify different sponsored Living Standards Measurement Surveys households as "poor," but also to generate sig- (LSMS), which collect detailed household-level nificantly different measures of health inequality information on income and consumption items, between "non-poor" and "poor." This paper re- including imputed estimates of self-employed in- views the types of indicators that are usually come and consumption of own produce. White used for socioeconomic status, focusing specifi- and Masset (2003) provide a description of the cally on assets to measure SES and summarizes steps that must be taken to ensure comparability the issues involved in aggregating assets into a of income and expenditure across households composite index of wealth. Finally, the paper and time. presents results of construction of an asset index Indicators of physical asset ownership such as for Bangladesh DHS data. durable goods, housing quality, and access to basic services such as water and sanitation, are Indicators of Socioeconomic Status often used to measure socioeconomic status Socioeconomic status is a well-established term, where income or expenditure data are unavail- originating in sociological literature (see Bollen able, as in the case of research based on Demo- and others 1999).1 Socioeconomic status (SES) graphic and Health Survey (DHS) data. DHS can be broadly defined as indicating an individ- collect nationally representative information on ual or household's relative position according to health, nutrition, and fertility. The DHS asset some social and/or economic stratification scale, module can be used to construct an index of where the latter may be determined with refer- wealth, though the main problem is in aggrega- ence to a broad or narrow set of characteristics. tion, since data on prices, which would facilitate The term "status" implies a relative approach to comparisons of different assets, are not available. measuring well-being, though some measures, There are also issues regarding the choice of as- such as income/consumption, can be inter- sets to include in the index and adjustment for preted absolutely, given adjustments for dif- household size, composition, and economies of ferences across space and time. Given this scale. These topics are taken up below. conceptual ambiguity, socioeconomic status is However, it should be clear that income/ measured in a variety of ways, the most common consumption and assets are not equivalent indi- being with economic indicators (such as in- cators. There is a theoretical justification for come, expenditure, and asset ownership). This using assets to measure SES. Economic theory 8 9 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? distinguishes current income from permanent sets can enter separately in regressions. The main income (wealth), the latter providing a better arguments against this are: (1) reduction in de- representation of living standards over time. If grees of freedom, but this is not usually an issue credit and insurance markets operated perfectly with analysis of household survey data sets, given (and foresight were perfect), households would that the number of observations is often in the be able to plan expenditures and borrow/save thousands; and (2) a high correlation between accordingly to minimize consumption fluctua- the different assets will result in multicollinearity, tions across seasons and over economic and life undermining their statistical significance. cycles, in order to stabilize life-time consump- Where households need to be ranked by tion. Under these conditions, current expendi- wealth, in order to determine the correlation be- ture provides the best indication of permanent tween wealth and another variable, an index can income (wealth). In practice, though, particu- be generated by aggregating assets: larly in poor countries, household expenditure tends to fluctuate with current income because Ai = 1ai1 +L+ nain (1) problems in saving and borrowing (and access- ing insurance) prevent consumption smoothing where Ai is the total asset score for each house- over time (Baulch and Hoddinott 2000).2 Under hold i, anare individual asset types and are the n such conditions, assets are preferable to con- weights. Variables indicating ownership/access sumption for measuring wealth, and there is to assets are measured categorically (1 = yes; some empirical evidence supporting this; for ex- 0 = no). In all other cases, scale equivalence ample, using survey data from Indonesia, Nepal must be ensured, where necessary by adjusting and Pakistan, Filmer and Pritchett (1998) argue variables to range between 0 and 1, otherwise assets to be preferable to consumption for mea- the index will contain "hidden weights" since suring wealth because they rank households variables ranging over bigger scales will implic- more consistently over time. itly be weighted more heavily.3 Young Lives On the other hand, income represents (2002) cites as an example of the latter the edu- human capital attainment and remittance cation component of UNDP's Human Develop- flows--important determinants of current living ment Index (which is measured as one-third standards that physical assets are less likely to ac- mean years of schooling to two-thirds literacy), count for. Note also that, for these reasons, the which in the first year of implementation failed stock of physical wealth would tend to be more to adjust for scale equivalence of literacy (mea- unequally distributed than income flows. sured in percentage terms) and mean years of Human capital assets have been included in schooling (ranging between about 2 and 10); in asset indices by some (Sahn and Stifel 2000) to the resulting education index, the majority of reduce the bias in measuring wealth when rely- variation was due to literacy because of its much ing on physical assets alone; the problem with greater variance. One problem with rescaling is doing so is that it precludes estimation of a sep- that it is sensitive to outliers, which may there- arate education effect. fore have to be removed from the data before- hand. Generating an Index The following hypothetical example (table This section describes the main issues in gener- C2.1) demonstrates the calculation of a wealth ating an index from asset data collected by score for two households, assuming a simple household surveys. index of seven household durable goods (own- Before moving on, it is important to note that ership of a wardrobe, chair, and table, which all there is no reason to generate an index unless have unitary weights; ownership of telephone, it is needed for some analytical purpose, e.g., TV, and bicycle, which each carry weights of 2; in order to analyze the relationship between and ownership of a motor vehicle, with weight "wealth" and another variable. If not, and house- 3). Household 1 owning all of the items would hold wealth merely needs to be controlled for, as- therefore have asset score 12; household 2 own- 9 0 N E O N A T A L , P O S T N A T A L , A N D C H I L D M O R T A L I T Y I N T H E 1 9 9 0 s ing chair, table, and bicycle would be given an Simple Example of asset score of 4. T A B L E C 2 . 1 Asset Index In the preceding example, weightings were chosen arbitrarily, though were intended to be Weighting Household 1 Household 2 sensible. An alternative is to allow weightings to be determined by the variation in the data using Wardrobe 1 Y N Chair 1 Y Y statistical methods, for example principal Table 1 Y Y components analysis (PCA), as used by Filmer Telephone 2 Y N and Pritchett (1998). Table C2.10 presents a cal- TV 2 Y N culation of asset scores for two hypothetical Bicycle 2 Y Y households using weightings estimated for Motor vehicle 3 Y N Bangladesh by Gwatkin and others (2000).4 Total value 12 4 Household A--which has electricity, radio, TV, Note: Y = household owns item; N = household does not own item. bicycle, piped drinking water outside residence, septic tank, cement flooring, brick walls, and fin- ished (cement/concrete) roof--has a high asset ceiling and number of rooms in the score of 3.4. Household B--which is "poorer"-- dwelling; works its own land, owns a radio and bicycle, has ­ productive assets, such as landholdings, a surface well for drinking water, uses a pit la- livestock, machinery, human capital; fi- trine, and has wooden flooring and rudimentary nancial assets, such as savings. (wooden) walls and (tin) roof, and is assigned an (2) Basic amenities, such as drinking water asset score of 0.7. If the same dwelling had only source, toilet facility, electricity, and fuel natural (earth) flooring, (bamboo) walls and used for cooking, which are likely to be (thatched) roof, it would have an asset score strongly determined by location. of ­0.2. There are three conceptual issues associated The distinction between "household vari- with generating an index: choosing the individ- ables" and basic amenities is important, due to ual components that make up the index, deter- the fact that many amenities are either publicly mining the weighting scheme and controlling provided or dependent on the availability of in- for the size and structure of the household frastructure, which is in turn strongly correlated (Gwatkin and others 2000). These issues are ex- with locational factors such as rural or urban res- plored in the following sub-sections. idence, region, etc. In other words, lack of a par- ticular facility does not necessarily indicate Choosing the Components of the Index "poverty" at the level of the household. Of Asset information commonly collected in house- course, the classification presented above is pri- hold surveys can be categorized into two main marily illustrative and simplified. Basic amenities groups, or five sub-groups. Table C2.2 presents coverage will vary by country/region, and will a list of assets for which data are collected in therefore be less indicative of location in coun- three such surveys, the Bangladesh Household tries/regions where coverage is greater. Other Income and Expenditure Survey (HIES) 2000/01, factors classified as household assets may also be the Bangladesh DHS 1999/2000, and the Young locational in nature, e.g., productive assets and Lives project.5 The variables include: (landline) telephone, and households are un- likely to own durables such as a television or re- (1) Household assets, such as: frigerator where electricity is unavailable. On the ­ consumer durables, such as a bicycle, other hand, toilet facility may be household spe- motor vehicle, refrigerator, radio, table, cific.6 TV, and telephone; One of the key problems in using assets, ­ housing quality, such as the types of ma- broadly defined, is that some assets, while use- terial used to construct floor, walls and ful to proxy SES, are also important indicators of 9 1 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? T A B L E C 2 . 2 Assets Included in Three Household Questionnaires Asset group Bangladesh HIES (2000/01) Bangladesh DHS (1999/2000) Young Lives (2002) Productive Land ownership Land ownership Land ownership assets Agricultural assets including tractor, farm Agricultural assets including tractor, equipment, livestock, trees farm equipment, livestock Financial assets Savings, investments Savings Household TV, radio, cassette player, VCR, dish TV, radio, almirah (wardrobe), (Specified to be in working condition): durables antenna/decoder, camera/camcorder, bicycle, table/chair, bench, watch/clock, TV, radio, refrigerator, bicycle, motorcycle/scooter, motor vehicle, refrigerator/ cot/bed, bicycle, motorcycle, motorbike/scooter, motor vehicle, freezer, washing machine, fans, heaters, sewing machine, telephone landline telephone, mobile phone, sewing machine, pressure lamps/petromax, car- sewing machine (plus two other coun- pet, furniture, kitchen items, watch, clock, tele- try-specific assets) phone/mobile phone, other Housing quality Floor, roof material Floor, roof, wall material Floor, roof material Number of rooms & area of dwelling Number of rooms Home ownership Home ownership Home ownership Amenities Toilet facility Toilet facility Toilet facility Drinking water source Drinking water source Drinking water source Electricity Electricity Electricity Fuel used for cooking other aspects of quality of life of interest to the re- taker, embodied in existing knowledge and abil- searcher. That is to say, including different types ity to process new information (as strongly af- of assets with separate relationships with a vari- fected by education outcomes), is likely to have able of interest, such as health status, may ob- important direct effects on child health out- scure effects operating directly from indirect comes. Other productive assets are more likely effects operating via economic status (Mont- to have indirect implications for health via wealth gomery and others 2000). For example, aspects (e.g., where they act as methods of saving). of housing quality such as flooring and amenities Therefore, an alternative classification may be such as water, sanitation, and energy source7 are more relevant to a study examining causal rela- likely to have separate direct effects on health tionships, by classifying variables according to outcomes, e.g., by affecting exposure to infec- their "direct" and "indirect" impacts on a variable tions, in addition to any indirect effect which of interest (Houweling and others 2003). Note, operates through their use as an indicator of however, that the definition of "direct" and "in- income or wealth. In comparison, consumer direct" depends on the specific health outcome durables are likely to influence health status only in question. Table C2.3 presents a matrix show- indirectly through their correlation with income ing the relationship between these alternative and wealth, though possibly ownership of a classification schemes for health/survival status. radio, TV, and telephone gives the household Given the variety of assets that are often avail- better access to information regarding good able, and the fact that they may indicate or de- health practices and emergencies, especially in termine (directly or indirectly) separate aspects remote areas. Human capital of the main care- of well-being, there is usually a large menu of po- 9 2 N E O N A T A L , P O S T N A T A L , A N D C H I L D M O R T A L I T Y I N T H E 1 9 9 0 s Asset Classification for Direct and Indirect Impacts T A B L E C 2 . 3 on Health Status Impact on health/survival Asset type Direct Indirect Consumer durables Table, bed, radio, TV, bicycle, motor vehicle, tele- phone, refrigerator Housing quality Type of material used on construct floor, number Type of material used to construct walls and ceiling of people per room Productive assets Human capital (education, knowledge) Landholdings, livestock, farm implements, machinery Amenities Drinking water source, toilet facility, fuel source tential indices to choose from. However, recent highly skewed), it may be necessary to include research has demonstrated empirically that dif- some productive assets in order to distinguish ferent asset indices are not equivalent, as they between households. Inclusion of agricultural are often imperfectly correlated across house- land may be desirable given a large rural popu- holds due to the different types of assets in- lation, despite the fact that it would also tend to cluded. For example, Houweling and others mis-categorize households in urban areas as (2003) demonstrate that only about half of poorer than they may actually be. households in Indonesia and Uganda are identi- · Access to certain assets will be determined by fied in the same wealth quintile of the popula- location as opposed to purchasing power of tion by asset indices created from household the household. Most obviously, households durables ownership alone and an index that also in rural areas are less likely to have access to includes housing quality and access to amenities amenities such as electricity, water, and sani- (water, sanitation, and electricity); most (about tation, though some would not be considered three-quarters) of the remaining households poor as more conventionally defined. There- move one wealth group up or down. fore, amenities which are highly locational In sum, there are good reasons to exclude should not be used to measure household certain types of assets from an index of house- wealth. The effect of including assets deter- hold socioeconomic status: mined by location is to overstate measured inequalities in household wealth. However, · Different types of assets represent different as- given that inclusion of multiple asset types as pects of "wealth." Productive assets, while rep- explanatory variables in regression analysis resenting household wealth status within will undermine precision of estimates where livelihood groups, are unsuitable for measuring the assets are strongly correlated, it may be wealth between livelihood groups--for exam- sensible to construct separate indices for ple, households that do not own cattle may not "household assets" and "amenities," where be poor, but simply do not rely on herding as a the latter includes locational items (Hou- livelihood. Productive assets should therefore weling and others 2003).8 be excluded from the asset index. However, · The distinction between "direct" and "indi- where productive assets represent an impor- rect" determinants of a variable of interest is tant aspect of wealth, such as land or livestock important. Only indicators having indirect ef- ownership in rural areas (or, for example, fects (via "wealth") on a dependent variable, where there are limited numbers of non-pro- say, health status, should be included as com- ductive assets, or where their distribution is ponents of an index used to measure in- 9 3 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? come/wealth in causal analyses. This is crucial Including the same list of assets for different for policy analysis because, given public re- countries, and weighting these assets by the source constraints, policy may be more effec- same amounts, enables comparison across tive where it focuses on improving those countries--this is relatively easy with DHS data assets that are the main drivers of change, since the questionnaire has a standard asset list rather than on measures to raise general included in all surveys. However, some assets wealth. The effect of including assets that may represent household wealth to different ex- have direct impacts on a variable of interest is tents across countries (or even regions/sectors to overstate estimated socioeconomic in- within countries) and thus are not strictly com- equalities in that variable.9 For example, Hou- parable; e.g., bicycles do not represent the pres- weling and others (2003) estimate that tige in Asia, where in many countries they are inclusion of water and sanitation variables in relatively common and cheap, that they do in the index augments socioeconomic inequali- most African nations, where they are rare and ex- ties in under-five mortality, in 50 percent of pensive. Sahn and Stifel (2000) generate sepa- cases, by attributing some of the impact of rate weighting schemes for 11 African countries these direct factors to economic status. Ex- to reflect local preferences for each asset. But cluding water and sanitation from a full index by doing so they lose the ability to make inter- that also includes consumer durables, hous- country comparisons. Young Lives (2002), on ing quality, and access to electricity supply, the other hand, uses a common weighting sys- reduces inequality in under-five mortality in tem across countries, though it allows for two Chad and Malawi by more than 30 percent of locally specific assets for each country, and re- the inequality index; in Brazil, Indonesia, and tains comparability across countries by rescaling Pakistan, estimated inequality is reduced by (see endnote 4). between 10 and 30 percent; in Bolivia, Kenya, Tanzania, and Uganda, inequality is insignifi- Determining Weighting Coefficients cantly affected; and in Cameroon estimated The weightings assigned to the individual asset inequality rises. Excluding housing quality, components of the index can be determined water, and sanitation from the full index re- using a variety of methods (Filmer and Pritchett duces estimated inequality by over 10 percent 1998). First, weights may be assumed to be equal in 3 out of 10 (Brazil, Indonesia, and Kenya) across assets, thus the in equation (1) are but increases inequality in a further 3 (Chad, equal to one and the index is simply the sum of Malawi, and Tanzania). Additional exclusion individual asset indicators. However, as Filmer of electricity produces estimates of inequality and Pritchett (1998: 4) state, this method "has as that are lower than the estimate produced by its only appeal not seeming as arbitrary as it re- the full index in Bolivia, Brazil, Indonesia, ally is." In addition, equal weighting will give the Kenya, and Pakistan but higher in Chad, least possible categories of the asset ownership Malawi, and Tanzania. index, making it more difficult to distinguish households, which is likely to be problematic The cost of excluding certain asset types is given the limited number of assets on which data that it may prevent sufficiently continuous rank- are usually collected; more complex weighting ing of households, particularly in poor countries systems give more thorough rankings. where ownership of durables may be highly lim- Second, assets may be assigned unequal ited among poorer segments of the population, weights using more arbitrary, but reasonable, or where the survey only collects limited infor- methods, e.g., by making reasoned choices based mation on durables.10 Houweling and others on other information sources. In Young Lives (2003) demonstrate for DHS data that the lack of (2002), the wealth index is calculated as the aver- the variation in the data may necessitate inclu- age of three components: (1) housing quality, sion of more types of assets in order to stratify which is the average of number of people per poorer households evenly into wealth quintiles. room (rescaled to vary from zero to one), and 9 4 N E O N A T A L , P O S T N A T A L , A N D C H I L D M O R T A L I T Y I N T H E 1 9 9 0 s dummies indicating finished floor, walls, and roof; first is used to generate the asset index. Sahn and (2) the sum of household ownership of a range of Stifel (2000) argue that, since FA accounts for the durable assets as a continuous variable scaled be- variation in assets using a smaller number of tween zero and one; (3) the average value of covariates, it is therefore preferable to PCA for dummy variables indicating access to drinking generating the wealth index.12 However, the water, toilet, electricity, and cooking fuel. A par- methods tend to produce highly similar asset ticular advantage of the index is that by scaling rankings. total durables ownership between 0 and 1, it does There are two main concerns with PCA/FA. not give undue weight to durables (of which there First, using an index may entail losing a lot of in- are more variables) over house quality and ser- formation regarding variance of asset ownership vices, which an equally weighted index would. across households, but this is the case with any Third, where data are available, weights could index. Second, PCA/FA replaces arbitrariness be determined by asset prices. Many surveys, with statistical credibility, but does not neces- such as the DHS, do not collect data on asset sarily estimate weightings that are meaningful in prices, though average national or regional rela- a socioeconomic sense--a more meaningful tive prices could potentially be used instead index may be created using an "arbitrary" but (Young Lives 2002). However, it is not clear reasonable weighting system, as in Young Lives whether a weighting scheme based on prices is (2002). In Gwatkin and others (2000) index de- necessarily ideal--though this would create an rived from PCA, a household that works its own index analogous to household income and con- or its family's land is assigned negative weight- sumption measures. Only under a highly restric- ing, which obviously reflects location (urban res- tive set of assumptions do prices reflect what is idents are more likely to have access to water, measured at a theoretical level, namely marginal sanitation, and electricity utilities--as well as, utilities in consumption. For example, where therefore, other electrical assets--but less likely there exist imperfect competition, production, to work in agriculture). However, in rural areas or consumption externalities, missing markets it is the wealthier who own land (and the land- or imperfections in information--i.e., nearly all less that are usually poorest), therefore the co- of the time, and especially in developing coun- efficient on land ownership, if indicating wealth, tries--prices are likely to deviate substantially should enter positively. The negative coefficient from marginal utility. on land is therefore a good example of why Fourth, assets can be linearly aggregated into using productive assets can be misleading. a single index using econometric methods. Another way to determine weightings econo- Filmer and Pritchett (1998, 1999) and Gwatkin metrically is to use the coefficients estimated and others (2000) used principal components from a fully specified regression of the dependent analysis (PCA) to generate asset weightings, variable of interest, say, height-for-age z-score, on while Sahn and Stifel (2000) used factor analysis the asset variables and all other explanatory vari- (FA). PCA/FA are data reduction methods that ex- ables (Lensink and White 2000). Given estimated amine the statistical relationships between vari- coefficients, the asset index can be calculated and ables.11 The first principal component/factor is the regression re-estimated, replacing the indi- the linear combination of variables explaining the vidual assets with the asset index, in order for the maximum possible variation in the data, the sec- impact of "wealth" to be determined. ond principal component/factor is that account- Table C2.11 indicates some of the different ing for the largest of the remaining variation, etc. ways in which assets have been aggregated in In analyses of asset ownership the characteristic previous research. The studies were selected that explains the largest variation (that estimated mainly, though not exclusively, as relating to by the first principal component/factor) is as- DHS data and estimation of child mortality and sumed to be "wealth." However, there are prob- nutrition. Madise and others (1999) use indicator lems in attributing socioeconomic meaning to variables for whether the household owns any higher order components/factors, thus, only the one or more than one of a set of items. Many re- 9 5 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? cent studies have used an asset index similar to of members, and the scale coefficient, which the one of Filmer and Pritchett (1998), which ag- ranges between 0 and 1. Zero represents no gregates household and community-level assets economies of scale (thus the adjusted index is into a single index. Some separate the two, e.g., simply equal to the unadjusted index expressed Kishor and Parasuraman (1998) and Madise and in per capita, or per equivalent adult, terms) and others (1999), showing that amenities (water and one indicates perfect economies of scale (house- sanitation) have separate impact on mortality, on hold size does not alter the asset index). top of household wealth. Often, the asset index Empirical estimates of the scale coefficient usu- is used to categorize households into asset quin- ally are between 0.15 and 0.3 in developing tiles,13 and/or is used to calculate socioeconomic countries (White and Masset 2003). For richer measures of health inequality. countries where economies of scale are likely to be greater (given the lower proportion of food Controlling for Household Size and in the household budget, and greater coverage Economies of Scale of utilities), the coefficient will be higher; e.g., Since data are often collected at household level, Wagstaff (2000) cites a study of OECD countries an issue arises as to whether adjustment should that implied the coefficient of scale was equal to be made for the size and possibly composition of around 0.6 (Buhmann and others 1988). the household. In the case of wealth, where Deciding on an overall scale coefficient may be durable goods are used as a store of value, e.g., problematic, given that different assets are likely as insurance to be sold in the event of an adverse to have different scale economies, which may also household shock, the size of the household de- depend on the particular use the index is trying to termines the amount of insurance per capita a capture. However, it is possible to adjust some given asset can provide.14 For other household components of the asset index selectively. goods, such as bicycle ownership, their utility Most studies using an asset index (e.g., under normal conditions also implies adjust- Gwatkin and others 2000) do not indicate that ment for household size. they adjust the asset index for household size, Other assets are consumed publicly by the which means that, implicitly, the scale coeffi- household--i.e., economies of scale in house- cient is assumed to equal 0 (i.e., it is assumed hold consumption entail that assets can be used that adding another person to the household by one member without reducing consumption does not alter the value of the index, nor the possibilities for another--and therefore adjust- weights attached to specific variables in the ment for household size is not necessary. index). Wagstaff (2000) takes an intermediate Indeed, many items in the asset index can be stance by setting the scale coefficient as equal to considered public goods, although to varying 0.5 for expenditure data--in his estimation for degrees, e.g., consumption of some utilities nine developing countries, assuming zero economies of scale tended to reduce socioeco- such water (unless metered), electricity (for nomic inequalities in mortality. Filmer and heating), aspects of housing quality such as ceil- Pritchett (1999) and Sahn and Stifel (2000) ing, wall, and floor materials, some durable report that results are insensitive to different goods (e.g., refrigerators, radio/TV), and to a assumptions about scale coefficients. lesser extent dwelling size (until congestion is reached) (White and Masset 2003). Constructing a Wealth Index The following formula is used to adjust household-level variables for household size and for Bangladesh economies of scale: The preceding sections examined major con- ceptual and practical issues regarding the gener- Ai* = Ai / n1- (2) ation of an asset index. The aim is to provide where Ai* represents the adjusted value of the recommendations for generating an asset index asset index, Ai the unadjusted index at house- using DHS survey data for Bangladesh, to be hold level, n the composition-adjusted number used subsequently in bivariate and multivariate 9 6 N E O N A T A L , P O S T N A T A L , A N D C H I L D M O R T A L I T Y I N T H E 1 9 9 0 s analyses of child mortality and nutritional out- (2) Housing quality, which is itself the mean of comes. First, given the importance of separating two components: assets that are likely to be direct determinants of ­ Wall material dummy (weight): natural/ health outcomes from those that have indirect other (0), wood (0.25), tin (0.5), brick (1). effects, by representing "wealth," the paper rec- ­ Roof material dummy (weight): natural/ ommends that direct factors should be ex- other (0), tin (0.5), cement/tile (1). cluded, and estimated separately from wealth. Direct determinants of child nutrition and sur- Since the weightings are time insensitive, this vival are drinking water source, toilet facility, and enables comparisons between households over floor material of the household. This constrains time, to determine whether wealth poverty is which assets can be included in the index, and increasing or decreasing over time. Due to the therefore the possibilities for differentiating be- arguably strong public good aspects of the non- tween households, especially given that other durables items, and given that the DHS do not items that, while having arguably less direct ef- collect data on number of items owned of each fects on health and nutrition, are representative household durable, no adjustment is made for household size or composition--the economies of location, such as electricity, and therefore of scale in the household are assumed perfect. should be excluded. We are left with household Table C2.4 shows the actual weighting as- durables--ownership of almirah (wardrobe), cribed to each variable in the asset index, as well clock, cot, bicycle, radio, table, TV--and housing as the raw Eigenvectors calculated using princi- quality as measured by wall and roof material.15 pal components analysis. In order to make com- A wealth index is constructed based on parisons between each, the scores are rescaled weightings that are both intuitive and in part de- to lie in the interval (0, 1). PCA ranks housing termined by the variation in the data. The struc- ture of the index follows from Young Lives (2002), in that it is the average of household Weights for Bangladesh durables ownership and housing quality, where T A B L E C 2 . 4 DHS 1993/94, 1996/97, total durables owned and housing quality are and 1999/2000 scaled comparably. However, unlike Young Lives (2002), which calculates all weights arbitrarily, Rescaled scores the weightings in the index for Bangladesh for Scores for Score for durable goods are partly chosen by the variation Asset variable intuitive index PCA index Intuitive PCA in the data--specifically, individual weights for Household durable goods each durable owned is set equal to (1­P) where Radio 0.081 0.273 0.326 0.887 P is the proportion of households owning the TV 0.101 0.314 0.403 0.948 item; total ownership of durables is rescaled ac- Bicycle 0.095 0.176 0.379 0.743 cordingly. This scheme allows much greater dif- Wardrobe (Almirah) 0.086 0.334 0.344 0.978 ferentiation between households, as compared, Table 0.050 0.326 0.201 0.966 Clock 0.062 0.349 0.247 1.000 for example, to a scheme based on equal weight- Cot 0.025 0.259 0.100 0.866 ings of durables.16 Weightings for housing quality are determined intuitively, though the Housing quality weightings chosen are not incontrovertible, Jute/bamboo/mud wall 0.000 ­0.325 0.000 0.000 since choice of building material is partly deter- Wooden wall 0.063 0.024 0.250 0.517 mined by location.17 Each household's asset Tin wall 0.125 0.137 0.500 0.685 Brick wall 0.250 0.316 1.000 0.951 score comprises the mean of: Other wall 0.000 ­0.061 0.000 0.392 Bamboo/thatch roof 0.000 ­0.284 0.000 0.061 (1) Household durables: weighted sum of Tin roof 0.125 0.149 0.500 0.703 dummy variables for seven goods, scaled to Cement/tile roof 0.250 0.235 1.000 0.830 range between 0 and 1.18 Other roof 0.000 ­0.040 0.000 0.423 9 7 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? quality similarly to the intuitive index, with the inequality in the asset index makes intuitive exception of "other" categories. The main differ- sense since households attain assets as they get ence between the indices is that, as expected, richer (and once attained, assets are kept), PCA generates scores that give greater weight whereas the index is made up of a limited range to durable goods. However, the indices are of assets, and is unable to distinguish multiple strongly correlated: the Spearman's rank corre- ownership of the same asset. Table C2.7 breaks lation between the indices is equal to 0.99, while down the change in the asset index over time the correlation coefficient is 0.98. into shares made up by changes in ownership of The index enables a fairly consistent ranking individual assets, where each asset is weighted of households, meaning that they can be identi- with its asset score, indicating that more than fied into nearly equal wealth quintiles. Table three-quarters of the change in the average value C2.5 shows the distribution of households by of the index comprises increases in ownership of quintile for each year. As with remaining calcula- TVs, clocks, tin walls and roofs, brick walls and tions, survey weights have been rescaled so that cement roofs. households have an equal chance of being in each year. The results indicate that household Linking Wealth and Expenditure wealth is improving over time on average and for Wealth indices are used as a measure of eco- all quintiles. The Bangladesh PRSP (Bangladesh/ nomic well-being, partly as DHS collects data on IMF 2003) indicates that consumption poverty assets but not expenditure. But economic analy- declined from 59 to 50 percent between 1991/92 sis may tell us the income/growth effect of an in- and 2000. A poverty line is, therefore, chosen to tervention. Is a 5 percent increase in income the define 50 percent of households as poor based same as a 5 percent increase in the wealth index? on their wealth in 1999/2000, which is equal to a Thinking about how the wealth index is con- value of the wealth index of 0.30. Table C2.6 structed suggests it is probably not so. Wealth shows poverty estimates and standard errors for indices are constructed on variables such as "poverty" as defined here, using the poverty whether the household owns a radio and a bicy- index due to Foster and others (1984).19 Wealth cle (or sometimes the number of such items). poverty is estimated to be declining much faster Once households receive a certain level of in- than income poverty. Asset ownership among come they will own these things, but not buy the poorest is also growing fastest, so that over- more as income rises. The wealth index is thus all inequality is decreasing; in contrast, income unlikely to pick up increases in income above a inequality, as measured by the Gini index, rose certain level. At the very least this argument sug- considerably during the 1990s from about 25 to gests that the elasticity of the wealth index with 30 percent (Bangladesh/IMF 2003). This fall in respect to household expenditure should be Shares of Population in Wealth Quintile and Mean Value T A B L E C 2 . 5 of Index by Year 1993/94 1996/97 1999/2000 Percent of Mean value of Percent of Mean value of Percent of Mean value of Quintile population wealth index population wealth index population wealth index Q1 24 0.010 22 0.022 20 0.058 Q2 17 0.100 19 0.131 20 0.185 Q3 19 0.179 20 0.227 20 0.304 Q4 21 0.323 20 0.380 21 0.462 Q5 19 0.641 20 0.693 19 0.752 Total 100 0.244 100 0.287 100 0.350 Note: Households are ordered into quintiles separately by year. 9 8 N E O N A T A L , P O S T N A T A L , A N D C H I L D M O R T A L I T Y I N T H E 1 9 9 0 s less than one, and probably that it will be lower Estimates of Poverty at higher levels. T A B L E C 2 . 6 and Inequality This issue was examined using data from the Bangladesh Household Income and Expendi- Headcount Poverty Squared Gini ture Survey (HIES) of 2000-01. HIES and DHS index gap index poverty gap index questionnaires do not cover the same house- 1993­94 0.690 0.430 0.335 0.515 hold assets, but are very similar. The variables (0.011) (0.011) (0.011) (0.007) used by DHS, and the corresponding variables 1996­97 0.616 0.353 0.261 0.467 used from HIES are in table C2.8. (0.012) (0.010) (0.010) (0.006) Table C2.9 shows the expenditure elasticity of 1999­2000 0.501 0.247 0.169 0.400 the wealth index, obtained from a regression of (0.013) (0.010) (0.008) (0.007) the log of the wealth index on the log of per Note: Standard errors in parantheses. capita household expenditure. The estimated elasticity is 0.8. Two other models are presented Contribution to Change in T A B L E C 2 . 7 in the second and third column. The first in- Asset Index Over Time cludes a quadratic term of the log of per capita expenditure. The second includes an intercept Percent Percentage contribution to and slope dummy for the top 25 percent of the owning asset change in wealth index per capita expenditure distribution. As ex- Asset variable 1993 1996 1999 1993­99 pected, the expenditure elasticity is decreasing at higher levels of household expenditure. Radio 25 32 32 5.3 TV 7 11 17 9.5 Bicycle 16 19 20 3.6 Almirah (wardrobe) 22 27 26 3.2 Table 49 55 63 6.6 Clock 35 46 55 11.7 Cot 72 78 84 2.8 Jute/bamboo/mud wall 73 71 61 0.0 Wooden wall 3 2 3 0.0 Tin wall 9 12 19 11.8 Brick wall 11 12 17 14.2 Other wall 4 3 0 0.0 Bamboo/thatch roof 37 29 20 0.0 Tin roof 53 59 70 20.0 Cement/tile roof 5 6 9 9.4 Other roof 6 6 0 0.0 9 9 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Asset Variables Expenditure Collected by DHS Elasticities of the T A B L E C 2 . 8 and Similar T A B L E C 2 . 9 Wealth Index Variable of HIES (standard errors in brackets) DHS HIES Top Radio Radio Double Log- intercept TV TV Variable arithmic quadratic and slope Bicycle Bicycle Almirah (wardrobe) Drawing room furniture Log of expenditure 0.81*** 2.21*** 0.87*** Table Dining room furniture (0.02) (0.27) (0.04) Clock Clock Bed Bedroom furniture Log of expenditure ­0.10*** Wall material Wall material squared (0.02) Finished Brick/cement Intercept of top 25% 2.53*** Rudimentary CI sheet/tile/wood (0.38) Natural Hemp/hay/bamboo Slope of top 25% ­0.35*** Roof material Roof material (0.05) Finished Cement Rudimentary CI sheet/tile/wood Constant ­6.94*** ­11.7*** ­7.34*** Natural Hemp/hay/bamboo (0.14) (0.95) (0.24) Observations 6887.81*** 6887.77*** 6887.53*** R-squared 0.30 0.31 0.31 Plot of the Predicted Values Against per F I G U R E C 2 . 1 Capita Expenditure from the 3 Models Above 1 Log-linear Top 25% 0 Quadratic ­1 ­2 ­3 ­4 0 5,000 10,000 15,000 Monthly per capita household csm 1 0 0 N E O N A T A L , P O S T N A T A L , A N D C H I L D M O R T A L I T Y I N T H E 1 9 9 0 s Asset Score for Bangladesh DHS 1996-97, Generated by Principal T A B L E C 2 . 1 0 Components Analysis as Reported in Gwatkin and others (2000) Household Ownership score if: (1=yes, 0=no) Asset score Asset factor Has Does not Household Household Household Household Asset variable scores asset have asset A B A B Has electricity 0.118 0.214 ­0.065 1 0 0.214 ­0.065 Has radio 0.071 0.104 ­0.049 1 1 0.104 0.104 Has television 0.125 0.347 ­0.045 1 0 0.347 ­0.045 Has bicycle 0.032 0.065 ­0.016 1 1 0.065 0.065 If household works own or family's agric. land ­0.022 ­0.042 0.012 0 1 0.012 ­0.042 If piped drinking water in residence 0.104 0.526 ­0.021 0 0 ­0.021 ­0.021 If piped drinking water outside residence 0.020 0.179 ­0.002 1 0 0.179 ­0.002 If has tubewell for drinking water ­0.063 ­0.022 0.186 0 0 0.186 0.186 If has a surface well ­0.009 ­0.066 0.001 0 1 0.001 ­0.066 If rain for drinking water ­0.001 ­0.023 0.000 0 0 0.000 0.000 If uses river, canal or surface water for drinking ­0.009 ­0.049 0.002 0 0 0.002 0.002 Other source of drinking water 0.000 0.005 0.000 0 0 0.000 0.000 If uses septic tank or toilet 0.127 0.370 ­0.043 1 0 0.370 ­0.043 If has pit latrine ­0.016 ­0.041 0.006 0 1 0.006 ­0.041 If uses a water-sealed or slab latrine 0.017 0.033 ­0.009 0 0 ­0.009 ­0.009 If has open latrine ­0.033 ­0.059 0.019 0 0 0.019 0.019 If uses a hanging latrine ­0.012 ­0.054 0.003 0 0 0.003 0.003 If uses bush, field as latrine ­0.054 ­0.092 0.031 0 0 0.031 0.031 If has other type of latrine ­0.001 ­0.031 0.000 0 0 0.000 0.000 If has earth or bamboo as principal floor in dwelling ­0.152 ­0.057 0.401 0 0 0.401 0.401 If has wood, plank principal floor in dwelling 0.010 0.146 ­0.001 0 1 ­0.001 0.146 If has cement principal floor 0.153 0.415 ­0.056 1 0 0.415 ­0.056 If has other type of flooring 0.003 0.135 0.000 0 0 0.000 0.000 If has cane, palm, trunks for walls ­0.108 ­0.074 0.158 0 0 0.158 0.158 If has rudimentary walls ­0.004 ­0.021 0.001 0 1 0.001 ­0.021 If has tin for walls 0.012 0.034 ­0.004 0 0 ­0.004 ­0.004 If has bricks, cement blocks, concrete walls 0.143 0.361 ­0.057 1 0 0.361 ­0.057 If has other material for walls ­0.010 ­0.054 0.002 0 0 0.002 0.002 If has natural material roofing ­0.061 ­0.095 0.039 0 0 0.039 0.039 If has rudimentary roofing 0.003 0.003 ­0.004 0 1 ­0.004 0.003 If has finished roof 0.125 0.498 ­0.031 1 0 0.498 ­0.031 If has other roofing ­0.014 0.053 0.004 0 0 0.004 0.004 Total asset score 3.379 0.659 1 0 1 T A B L E C 2 . 1 1 Studies Using Assets to Measure SES 102 Asset type Description Asset/income Household Household dwell- Non-housing Author of study Country measure consumer durables ing characteristics Amenities (production) items Hill and Child mortality, Kenya Asset index (PCA), generating Radio, TV, refrigerator, Quality of floor, roof Electricity others (1996) DHS wealth quintiles bicycle, motorbike, car Filmer and Educational en- Indian states Asset index (PCA), generating Clock/watch, bicycle, radio, TV, Number of rooms, Toilet facilities, drinking water, elec- HH owns more than 6 acres of land Pritchett rollment, NFHS wealth quintiles sewing machine, refrigerator, building materials, tric lighting (1998) car, motorcycle cooking source Filmer and Educational at- 33 countries Asset index (PCA), generating Clock/watch, bicycle, radio, TV, Number of rooms, Toilet facilities, drinking water, elec- HH owns more than 6 acres of land Pritchett tainment, DHS across the wealth quintiles sewing machine, refrigerator, building materials, tric lighting (1999) world, including car, motorcycle cooking source Bangladesh Gage, Som- Immunization 11 Sub-Saharan Asset index (asset dummies Radio, motorcycle, car Non-mud floor Some toilet facility, piped drinking merfelt, and and diarrhea, African summed--equal weightings) water, electricity Piani (1996) DHS countries Gwatkin and Health, nutrition Bangladesh Asset index (PCA), generating Radio, TV, bicycle Floor material, wall Toilet facilities, drinking water, elec- HH works own or family's agric. land others and population, wealth quintiles material, roof mate- tricity (2000) DHS rial Kishor and Infant and child India HH asset index (ownership dummies Asset (weight in parentheses): No toilet or water, either toilet or Parasuraman mortality, NFHS summed-- weights chosen appar- clock (1), sewing machine (2), water, both toilet and water. (1998) ently subjectively), toilet and water sofa (2), fan (2), radio (2), bicycle facilities index (ranges from 0-2) (2), refrigerator (3), TV (3), scooter (3), car (4) Madise, Mat- Nutrition, DHS 6 in Sub- Asset dummies: variables for hh items `Modern hh items': working Floor material, wall Electricity, pit/bucket toilet, flush toilet thews, and Saharan Africa (at least one item, two or more items) radio, TV, bicycle, motorcycle, material Margetts and community items entered sepa- car (1999) rately Masset and Infant and child India HH asset index--non-weighted, Radio, TV set, refrigerator, bicy- White (2002) mortality, NFHS scaled to one cle, motorcycle, car, land, live- stock, clock and sewing machine Sahn and `Poverty' across 11 in Sub- Asset index (FA) Radio, TV, refrigerator, bicycle, Floor quality Toilet facilities, drinking water Years of education of household head Stifel (2000) countries and Saharan Africa car, motorcycle (or indicator variable where years ed- time ucation is not available) Sastry (1996) Child mortality, Brazil Asset dummies measured at Water supply, sanitation, electricity, DHS variant community level (proportions per garbage collection/public cleaning population) (hh level SES measured service (also health facilities and edu- using hh income) cation per population and number of TV stations) Wagstaff and Nutrition, LSMS 19 Income, asset index (PCA) Radio, refrigerator, TV, Rooms per person, Water supply, sanitation Watanabe countries motorcycle floor quality (2002) N E O N A T A L , P O S T N A T A L , A N D C H I L D M O R T A L I T Y I N T H E 1 9 9 0 s APPENDIX C.3: Socioeconomic Inequality in Mortality in Bangladesh During the 1990s This annex presents estimates of intertemporal The poverty line is defined as the value of wealth movements in socioeconomic inequality in mor- that divides the population of households in half tality among under-fives in Bangladesh. Socio- in the 1999/2000 survey round, which corre- economic status is measured using the wealth sponds to the national estimate of expenditure index, which is calculated as the rescaled poverty in 2000 of 50 percent (Government of weighted sum of household ownership durables Bangladesh 2002). As shown in the table, under- and indicators of housing quality (wall and roof five mortality fell slightly faster for the poor than materials). Mortality estimates are calculated the non-poor, with the largest difference being from three rounds of the Bangladesh DHS for reductions of neonatal mortality, and in fact (1993/94, 1996/97, and 1999/2000); data are reversed for postnatal. pooled across surveys, so estimates become In order to examine changes over time be- more precise as we move back in time, as they tween periods 1 and 2, equation (1) can be de- are based on a larger sample. Period mortality composed into the change in mortality rates rates are estimated using the synthetic cohorts among the poor, the change in mortality among probabilities method (Rutstein 1984). the rich, and the change in poverty:1 Examining health outcomes by income poverty, to which ill-health is often closely MR = MR p. p2 + MRn. 1 - p2 ) p n linked, it is possible to decompose changes in + p. MR1 - MR1 ( )( (2) mortality by poor and non-poor. Total mortality That is, the change in mortality is the in each period is equal to the weighted sum of weighted sum of the change in mortality for the mortality among the poor and mortality among poor and non-poor, plus the product of propor- the non-poor: tion of people moving out of poverty times the MRt = pt.MRt + 1 - pt .MRt p ( ) n (1) mortality differential for poor and non-poor. Table C3.2 presents the results of the decompo- where MRt is the mortality rate period t, p, and n sition,2 indicating that, for each age group, the superscripts denote poor and non-poor, and reduction in mortality over time has been largely p equals the share of the population living in due to the reduction in mortality among the poverty. Table C3.1 provides some indication poor. In addition, virtually all the change is ex- that the improvement in survival has been plained by reduction of mortality within groups, achieved among both "poor" and "non-poor." and very little as a result of households moving T A B L E C 3 . 1 Estimates of Mortality (per 1,000) by Poverty Status Neonatal Postnatal Child Under-five Poor Non-poor Poor Non-poor Poor Non-poor Poor Non-poor 1980­84 101 70 55 41 91 42 233 148 1985­89 71 60 52 29 76 37 190 123 1990­94 67 50 38 33 60 28 158 109 1995­99 44 38 33 19 38 19 112 75 Percentage change ­56.4 ­45.7 ­40.0 ­53.7 ­58.2 ­54.8 ­51.9 ­49.3 1 0 3 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Mortality Rate C = 2 cov yi, Ri ( ) µ (3) T A B L E C 3 . 2 Decomposition by Poverty Status where is the mean for variable y whose in- equality is being measured and cov is the covari- Decomposition (% of total oMR) ance between each individual i's value of y and oMRpoor oMRnon-poor oPoverty oMR the individual's fractional rank in the wealth dis- tribution, Ri. Unweighted standard errors are cal- Neonatal 64.9 32.2 2.9 100.0 culated from: Postnatal 53.1 44.8 2.1 100.0 Child 69.4 27.7 2.9 100.0 2 Under five 63.4 34.2 2.4 100.0 var(C) = 1/ n i ai - (1 + C)2 (4) where ai = yi /µ 2Ri - 1 - C + 2 - qi ( ) -1- qi (5) out of poverty. These findings support the argu- ment made in Annex B that factors other than in- and qi = 1/(µn) i yi (6) come growth are the main drivers of mortality reduction in Bangladesh.3 (Kakwani and others 1997). Examination of con- The preceding analysis has examined some centration curves for the 1990s for under-five, in- distributional characteristics of health status fant, and child mortality indicates that mortality among children, but only by looking at two is more concentrated among the poor during groups, "poor" and "non-poor," and therefore childhood than infancy (figure C3.1). This is ex- abstracting from a large amount of distributional pected since socioeconomic characteristics de- information. It is possible to utilize information termine survival to a greater extent as the child on the entire wealth distribution by using con- grows older and biological factors become less centration curves and concentration indices. important. Concentration curves have also, gen- There are various ways of measuring socioeco- erally, moved toward the line of equality, with nomic inequalities in health (and other welfare much of the reduction in inequality arising from outcomes) (see Mackenbach and Kunst 1997), inward movements of the lower half of the dis- though a common and easily interpreted mea- tribution of wealth, particularly for neonatal sure is the concentration index, which is calcu- infants, showing that improvements in survival lated from the concentration curve in the same prospects have been faster for newborns of way as the Gini coefficient is obtained from the poorer people. Table C3.3 presents concentra- Lorenz curve. The concentration curve is similar tion indices and t-statistics, confirming these to the Lorenz curve, except that where the trends. Inequality in under-five mortality has de- Lorenz curve ranks observations by the variable creased over the period, mainly caused by a re- in which inequality is being measured, the con- duction in mortality inequality among children centration curve ranks them according to an- other variable (e.g., to generate socioeconomic and among neonates (table C3.4). However, dif- inequalities in health, the concentration index ferences between concentration indices over ranks households by SES and calculates the re- time are not significant at reasonable levels, with sultant inequality in health status. Unlike Gini the exception of neonates, for whom the con- coefficient, the concentration index, C, can take centration index fell significantly between 1980- negative values: C<0 indicates poorer socioeco- 84 and 1995-99.4 nomic groups have worse health status; C=0 in- Examining socioeconomic inequalities in dicates perfect equality in health status by mortality risk by child's gender, the conclusion socioeconomic status; C>0 indicates poorer so- emerges that inequalities between rich and cioeconomic groups have better health status. poor are smaller for boys than for girls, and the The concentration index can be calculated inequalities among boys have also been declin- from the following formula (Kakwani 1980): ing faster (table C3.5). By differentiating be- 1 0 4 N E O N A T A L , P O S T N A T A L , A N D C H I L D M O R T A L I T Y I N T H E 1 9 9 0 s Concentration Curves for Infants, Children, and Under-Fives, F I G U R E C 3 . 1 all Children and by Gender Concentration curves for under-five mortality Concentration curves for neonatal mortality 1 1 .8 .8 Early 1990s deaths .6 Early 1990s deaths .6 Late 1990s Late 1990s 45 degree line 45 degree line prop. .4 prop. .4 Cum. .2 Cum. .2 0 0 0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1 Cum. prop. live births Cum. prop. live births Concentration curves for child mortality (Children aged 12-59 mos.) Concentration curves for postnatal mortality 1 1 .8 Early 1990s .8 Early 1990s Late 1990s deaths .6 deaths .6 45 degree line 45 degree line prop. .4 prop. .4 Late 1990s Cum. .2 Cum. .2 0 0 0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1 Cum. prop. live births Cum. prop. live births Concentration curves for infant mortality 1 .8 deaths .6 Early 1990s Late 1990s 45 degree line prop. .4 Cum. .2 0 0 .2 .4 .6 .8 1 Cum. prop. live births Sources: Adapted data from WDI; graph, bars from DHS. 1 0 5 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? much of the decline in socioeconomic inequal- Concentration Indices for ity in mortality occurred for children and, to a T A B L E C 3 . 3 Infants, Children, and lesser extent, for neonates, while inequalities in Under-Fives postnatal mortality widened during the mid- 1990s, though none of these differences are Infant t stat. Child t stat. Under five t stat. statistically significant. 1980­84 ­0.10 ­9.72 ­0.22 ­14.56 ­0.15 ­17.38 It is possible to decompose socioeconomic 1985­89 ­0.08 ­7.90 ­0.21 ­13.83 ­0.14 ­16.25 inequalities in mortality risk into inequalities in 1990­94 ­0.09 ­7.29 ­0.21 ­11.89 ­0.13 ­13.25 the determinants of mortality; similarly changes 1995­99 ­0.08 ­4.52 ­0.20 ­6.93 ­0.12 ­7.45 in inequalities can be decomposed into changes in inequality of the determinants. Following Wagstaff (2000), the concentration index can be expressed as the weighted sum of the concentration indices ci of the n variables de- Concentration Indices for T A B L E C 3 . 4 termining mortality: Neo- and Postnates C* = y1c1 + y2c2 +L+ yncn (7) Neonatal t stat. Postnatal t stat. 1980­84 ­0.08 ­6.24 ­0.12 ­7.12 where the weights are calculated by multiplying 1985­89 ­0.06 ­4.89 ­0.11 ­5.95 the coefficient estimate by the mean of that vari- 1990­94 ­0.08 ­5.10 ­0.11 ­5.41 able and dividing by the linear prediction of the 1995­99 ­0.04 ­2.82 ­0.14 ­3.88 regression model at the means of the data. Note that what is really being examined in this analy- sis is socioeconomic inequality in the underlying tween infants and children (table C3.6) and latent variable in the case of probit regressions then, among infants, between neonates and (for neonatal and postnatal mortality) and the postnates (table C3.7), it becomes clear that all log of the hazard ratio in the case of the survival of the apparent decline in mortality inequality model (for child mortality). The change in the between "poor" and "non-poor" girls has oc- concentration index over time t can also be de- curred for the neonatal period--postnatal and composed into the proportional contribution of child mortality inequality has remained constant the change in concentration indices for each or deteriorated among girls. However, tests variable i: show that differences between girl concentra- tion indices over time are not statistically signif- = y(ci )/C* (8) icant, with the exception of girl neonates for whom the reduction in inequality in mortality Table C3.8 presents a summary of the de- risk between 1980-84 and 1995-99 is marginally composition of the changes in concentration in- dices over time for determinants of mortality significant at the 10 percent level. For boys, used in subsequent regressions, with the full re- sults, including concentration indices for each Concentration Indices for variable, given in table C3.9. Socioeconomic in- T A B L E C 3 . 5 Under-Five Mortality by equalities in determinants of mortality have gen- Child's Gender erally declined over time; this is true for factors such as wealth, education attainment, and ma- Male t stat. Female t stat. ternal mobility. It is not true for the measures of 1980­84 ­0.14 ­11.67 ­0.16 ­12.97 health: there has been no change for antenatal 1985­89 ­0.13 ­11.03 ­0.14 ­11.81 care and a worsening for vitamin A and having 1990­94 ­0.13 ­8.99 ­0.14 ­9.89 no vaccination. For neonatal mortality, fertility- 1995­99 ­0.09 ­4.16 ­0.14 ­6.47 related changes, such as birth order have played 1 0 6 N E O N A T A L , P O S T N A T A L , A N D C H I L D M O R T A L I T Y I N T H E 1 9 9 0 s Concentration Indices by T A B L E C 3 . 6 Child's Gender Infant Child Male t stat. Female t stat. Male t stat. Female t stat. 1980­84 ­0.09 ­7.00 ­0.11 ­6.63 ­0.20 ­8.69 ­0.23 ­11.51 1985­89 ­0.10 ­6.77 ­0.05 ­4.16 ­0.18 ­7.95 ­0.23 ­11.44 1990­94 ­0.10 ­5.63 ­0.07 ­4.55 ­0.18 ­6.90 ­0.22 ­9.89 1995­99 ­0.08 ­3.12 ­0.08 ­3.29 ­0.15 ­3.52 ­0.24 ­6.16 Concentration Indices by T A B L E C 3 . 7 Child's Gender Neonatal Postnatal Male t stat. Female t stat. Male t stat. Female t stat. 1980­84 ­0.09 ­4.83 ­0.08 ­3.96 ­0.10 ­4.83 ­0.15 ­5.23 1985­89 ­0.07 ­4.14 ­0.04 ­2.60 ­0.14 ­5.27 ­0.07 ­3.07 1990­94 ­0.07 ­3.74 ­0.08 ­3.43 ­0.15 ­4.48 ­0.08 ­3.13 1995­99 ­0.06 ­2.58 ­0.03 ­1.52 ­0.12 ­2.03 ­0.17 ­3.71 Summary of Contribution T A B L E C 3 . 8 to Changes in Concentration Index Neonatal Postnatal Child Wealth 0.012 0.001 ­0.164 Education ­0.019 0.002 ­0.085 Fertility related ­0.149 ­0.017 0.023 Electricity ­0.069 0.000 0.005 Water and sanitation ­0.015 0.000 0.003 Gender related ­0.014 ­0.001 0.061 Locational ­0.003 ­0.004 0.010 Other 0.121 0.001 0.060 Total ­0.136 ­0.018 ­0.087 the largest role, followed by a reduction in the the most important single variable contributing inequality in access to electricity. The decompo- to reduction in socioeconomic inequality in sitions for postnatal mortality are unrevealing, mortality is wealth, followed by maternal edu- due to the small estimated change in postnatal cation, which accounts for just over half the mortality during the period. Finally, for children change accounted for by wealth. 1 0 7 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Decomposition of Changes T A B L E C 3 . 9 in Socioeconomic Inequality in Mortality by Age Group Early Late 1990s 1990s Neonatal Postnatal Child Wealth 0.514 0.412 0.012 0.001 ­0.164 Electricity 0.593 0.478 ­0.069 0.000 0.005 Water & sanitation 0.899 0.846 ­0.015 0.000 0.003 Mother primary ed 0.030 ­0.049 ­0.011 0.000 ­0.048 Mother finished primary 0.313 0.182 ­0.027 0.000 ­0.011 Mother secondary ed 0.552 0.425 0.025 0.001 ­0.025 Mother finished second or higher 0.763 0.741 ­0.006 0.001 ­0.001 Contraception 0.041 0.037 ­0.020 ­0.001 ­0.003 Multiple birth 0.112 ­0.087 0.102 0.000 Age at birth ­0.002 ­0.003 ­0.027 ­0.001 ­0.004 Age at birth sq ­0.004 ­0.008 0.042 Prev child died ­0.126 ­0.117 ­0.005 0.000 ­0.001 Prec interval <15 0.004 ­0.064 0.043 0.000 Prev child died x prec higher interval ­0.011 ­0.128 ­0.028 0.000 Birth order ­0.034 ­0.058 ­0.398 ­0.009 0.030 Birth order sq ­0.048 ­0.097 0.297 ­0.006 Female child ­0.008 ­0.005 0.004 0.000 0.004 Female x birth order ­0.036 ­0.066 0.000 0.052 Female head of hh 0.003 0.004 0.002 0.000 0.000 Mother's mobility 0.026 0.017 ­0.015 0.000 ­0.001 Mother divorced ­0.258 ­0.083 0.001 0.000 ­0.004 Mother remarried ­0.009 ­0.016 ­0.006 ­0.001 0.010 Antenatal visits 0.304 0.304 ­0.001 Chittagong 0.010 ­0.026 ­0.011 0.000 ­0.017 Dhaka 0.058 0.069 0.003 0.000 0.007 Khulna ­0.046 0.104 0.035 0.001 0.059 Rajshahi ­0.062 ­0.107 ­0.009 ­0.001 ­0.033 Rural ­0.048 ­0.073 ­0.021 ­0.004 ­0.006 Born In Feb-Mar x rural ­0.082 ­0.065 0.004 0.000 Born In Apr-May x rural ­0.029 ­0.083 ­0.029 0.000 Born In Jun-Jul x rural ­0.064 ­0.037 0.003 0.000 Born In Aug-Sep x rural ­0.069 ­0.076 ­0.003 0.000 Born In Oct-Nov x rural ­0.053 ­0.099 ­0.008 0.000 Mean_vitamin A 0.005 0.076 0.001 0.037 Mean_no_vaccine ­0.035 ­0.086 0.024 1 0 8 ANNEX D. CHILD MALNUTRITION DURING THE 1990S Bangladesh has made good progress in reducing measures nutrition cumulatively over time. rates of malnutrition among children, particu- Weight for height is sensitive to short-term fluc- larly in the 1990s, when malnutrition is esti- tuations in nutritional state. Weight for age is a mated to have fallen by almost one-third composite measure of weight for height and (Deolalikar 2002; Annex A). However, malnutri- height for age; it is criticized due to its inability tion remains a big problem. Stunting was esti- to distinguish children who are truly thin from mated at nearly 50 percent in 2000 (see figure those who weigh less because they are shorter. D.1), among the highest rates in the world. Of However, it has been shown to be more closely countries with available data, only a dozen oth- related to mortality risk in Bangladesh than ers, mainly in South Asia and Sub-Saharan Africa, other indicators (Chen and others 1980).2 have worse stunting rates (UNICEF 2002, table Child nutritional status is determined by 2). Similarly, underweight prevalence in Bang- comparing the anthropometric indicator for ladesh, though falling, remains one of the each child in the survey population to the ex- world's highest. Bangladesh seems to perform pected measurement for a child of the same much better compared with other countries in age/height and sex from a healthy reference incidence of wasting, though it should be borne population.3 This comparison can be done in a in mind that, due to seasonal variation in nutri- variety of ways, the most effective of which is to tion, it is difficult to make sound comparisons of normalize the anthropometric measure by cal- short-term indicators across surveys that may culating z scores: well have been conducted at different times of year. zi( y,g) = xi - µ( y,g) This paper examines the determinants of nu- ( y,g ) tritional status among children, in order to as- certain the contribution of various factors to where xi indicates the anthropometric measure nutrition differentials across children and im- for individual i and and are the respective provements in mean outcomes over time. Bang- median and standard deviation of the anthropo- ladesh experiences higher malnutrition than its metric measure for the appropriate age or income level alone would predict (see Annex B), height y and gender g in the reference popula- and discrimination against girls in nutrition and tion.4 health care has been well-documented (for a rel- The basic model of child nutritional status is atively early example, see Chen and others due to UNICEF (1990), in which nutrition de- 1981). pends proximately on dietary intake and health status. Thus, high rates of child malnutrition are Modeling Child Nutritional Status likely to be closely linked to the underlying Nutritional status is measured anthropometri- household food insecurity and the exposure to cally,1 using three indicators: height for age infectious disease such as diarrhea, due to poor (stunting), weight for height (wasting), and sanitary environment and inadequate medical weight for age (underweight). Height for age is treatment. A third underlying factor is the quan- an indicator of long-term malnutrition because it tity and quality of child care, which determines 1 0 9 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? F I G U R E D . 1 Malnutrition Prevalence in Bangladesh 0.6 1997 2000 0.5 0.4 0.3 Percent 0.2 0.1 0 Stunting Severe Underweight Severe Wasting Severe stunting underweight wasting Source: Calculated from DHS data. how effectively income is converted into nutri- short-term measures, age should exert a non- tion and the share allocated to children (and, im- linear concave impact on nutritional status: nu- portantly, which children receive the most), as trition deteriorates in the period in which well as the healthiness of the home and com- children are most susceptible to disease--from munity environments. Note that these three the onset of weaning (which should be at age underlying factors may be complements or sub- 6 months) until age 24 months--and improves stitutes in the production of child health. thereafter. For long-term measures the relation- The meta-analysis conducted as a back- ship between age and malnutrition is also likely ground paper for this study (Charmarbagwala to be concave, though because stunting is often and others 2004) provides a summary of the lit- permanent, the function is unlikely to fall after erature modeling nutritional status, as indicated 24 months of age. Gender-based discrimination by height for age z score. In multivariate analysis, in South Asia means that the coefficient on a fe- nutritional status can be modeled as a function male dummy may be negative; girls who have of child-specific factors, d, household-specific older sisters or who are competing with the first- factors, h, and community factors, e: born male are particularly likely to be discrimi- nated against (Croll 2001).5 Surprisingly, the z = n(d, h, e). meta-analysis found that boy children were sig- Child-specific factors include demographic nificantly more likely to be malnourished than variables such as child's age, gender, birth order girls (Charmarbagwala and others 2004). How- and birth interval, and behavioral factors such as ever, this finding may be due to mortality selec- breastfeeding and immunization. The relation- tivity, which was not allowed for in previous ship between age and nutritional status depends studies (see below). on whether the anthropometric indicator mea- More generally, children of high birth order sures short-term or long-term malnutrition. For (born latest) and children born closely together 1 1 0 C H I L D M A L N U T R I T I O N D U R I N G T H E 1 9 9 0 s are more likely to have lower nutritional states, weight; DHS round 2000 also provides data on due to socioeconomic factors (resource compe- mother's recollection of low birth weight. tition) and biological factors (e.g., physically de- Household-specific factors include parental pleted mothers may give birth to low birth- or family resources such as income, household weight babies and may be unable to breastfeed). size and composition, parental education, Mozumder et al. (2000) find that short sub- mother's age and mobility, and sex of the house- sequent birth interval impacts negatively on hold head; these are proxies for food security weight for age, but no statistically significant im- and the quality and quantity of care provided to pact of preceding interval, which they attribute children. to the likelihood that "the new infant holds an Income--broadly defined to include imputed advantageous position with the mother, com- own production--is a key determinant of house- pared with any older siblings, because of breast- hold nutritional intake. In the health production feeding" (p. 295). One issue concerning use of function literature, income is considered jointly household composition and demographic var- determined with nutrition and health, leading to iables is that they may be endogenous to de- biased estimates of the coefficient on the in- cisions on child health, leading to biased come variable in standard regression analysis regression estimates. As Horton (1986) argues, (see Annex C). The extent to which current in- parents are likely to make joint decisions on come and child health are endogenously deter- child quantity and child quality (demonstrated, mined can be questioned for extreme cases such for example, by nutritional status). Regression as long-term malnutrition (stunting) and mor- estimation should account for this source of tality (see Charmarbagwala and others 2004). bias, e.g., through use of instruments or by However, estimation techniques usually instru- dropping endogenous demographic variables ment for income, a suitable instrumental vari- to assess robustness of the other coefficient able being wealth.7 The meta-analysis found estimates. income/wealth to be strongly correlated with Child immunization is an important variable child nutrition, with the clear majority of studies to control for--vaccination reduces chances of finding a significantly positive effect (see Char- contracting debilitating disease. A mother's de- marbagwala and others 2004). cision to immunize her child depends partly on Food availability is highly seasonally depen- availability of health services and income, but dent in rural areas of Bangladesh (see Annex J). also her preference regarding modern medical Models of short-term nutritional status should care, which is unobservable.6 Due to unobserv- therefore include the month that measurement able heterogeneity, behavioral variables such as was taken (interacted with a rural sector immunization may be correlated with the error dummy). The wet season in July-October, which term in the regression equation, leading to occurs before the main rice harvest, aman, is a biased regression estimates (Rosenzweig and critical time of year for child health, due to Schultz 1983; Thomas and others 1991; Alder- greater prevalence of water-borne disease (diar- man and Garcia 1994). Consistent estimation rhea, malaria) and food scarcity (Muhuri 1996; requires use of instrumental variables (the "pro- Annex I). DHS data were collected between duction function" approach). November and March, limiting the possibilities In addition to the above, determinants of to explore seasonal variation in short-term mal- long-term malnutrition (height for age) should nutrition to this 5-month period. DHS 2000 control for genetic health endowment and birth round does, however, provide data on the weight, which can be an important determinant household's (self-reported) food availability dur- of ill-health and height throughout childhood ing the year. and onwards. Commonly used indicators of Household size is likely to be positively cor- health endowment are (log of) mother's height related with child nutrition, reflecting availability and, due to seasonal variation in food availabil- of replacement caretakers, including older chil- ity, month of birth may be a good proxy for birth dren and grandparents (people who may other- 1 1 1 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? wise be considered economic "dependents"). rural child nutritional status over urban, urban The meta-analysis found evidence for a sig- location was found to exert a positive impact on nificantly positive effect of household size on child nutritional status in the meta-analysis, pos- nutrition, and a significantly negative effect of sibly due to better access to health facilities in presence of young children in the household urban areas and other factors relating to better (Charmarbagwala et al. 2004). Estimation should communications and physical infrastructure account for potential endogeneity of household (Charmarbagwala and others 2004). composition and demographic variables, as ex- Variables such as water and sanitation are key plicated above. complements to food availability in determining Maternal education usually has a positive ef- child nutrition, because diseases such as diar- fect on child nutrition, as indicated by the meta- rhea diminish the body's nutrient intake; the analysis, which found education to be positively meta-analysis found that water and sanitation correlated with child nutritional status, par- were positively correlated with nutrition (ibid.). ticularly of mothers. The relationship between These are often termed "community variables" maternal education and child well-being reflects because access is often determined by location a number of factors: greater educational attain- and because there are likely to be positive spill- ment means greater income earning oppor- overs from one person's consumption of clean tunities; schooling may impart knowledge of water and sanitation to another person (e.g., by modern caring techniques directly; literacy reducing exposure to contagious diseases).9 means better ability to assimilate new informa- However, it may be so that those households tion from newspapers; exposure to new envi- with direct access to a facility, such as electricity, ronments due to schooling makes women more derive greater benefit than do other members of receptive to modern medical treatment; educa- the community. tion improves self-confidence and therefore de- Mortality selectivity, a final estimation issue, cision-making ability in the family; and schooling concerns the lower censoring of malnutrition provides the opportunity to form social net- data due to child death. Malnourishment is as- works (Alderman et al. 2003). Some of the ef- sociated with increased mortality risk (Gomez et fects of education can therefore be broken down al. 1956; Briend and others 1986), the risk of by controlling for income, father's education, death from malnutrition being greatest between mother's literacy, and knowledge of health and ages 6 and 36 months in one Matlab study family planning.8 (Fauveau et al. 1990). Since anthropometric data Other indicators of women's power in the cannot be collected on children who are dead household likely to determine child nutritional and these children are more likely to have been status include women's agency, mobility, and malnourished in life, the sample of live children age and whether the household head is female. is unlikely to be random (Lee and others 1997; The share of economic resources devoted to Charmarbagwala and others 2004). In conse- children is often greater in households where quence, the nutrition model should be esti- women have greater say in decision making, mated conditional on survival probability; this is though in the case of female-headed households possible because DHS collect complete fertility (FHHs) the positive impact on child health and and mortality histories of eligible mothers, en- nutrition may be counter-balanced by the abling survival probability to be estimated. Lee greater likelihood of both monetary and time and others (1997) do not find evidence for non- poverty. random selectivity in survival with respect to an- Community factors include location (urban/ thropometric measures.10 rural, division of residence) and environmental resources including access to clean drinking Data and Model water, adequate sanitation, electricity and health The data used in this study are from the nation- services. Despite the apparent logic that greater ally representative Demographic and Health food availability in the countryside would favor Surveys (DHS) collected in 1996/97 and 1999/ 1 1 2 C H I L D M A L N U T R I T I O N D U R I N G T H E 1 9 9 0 s 2000. DHS compile complete fertility histories of gression equations is bivariate normal (see ever-married women aged 10-49, as well as data Greene 2000). Unlike linear simultaneous equa- on health status and health care, knowledge of tions systems, which require some difference in health and family planning, socioeconomic fac- the variable set in each equation for identifica- tors such as education, household asset owner- tion, the Heckman two-step model is identified ship, water and sanitation, and information on simply by the non-linearity of the selection facilities operating in the community. In the sur- probit. However, there should also be a theoret- vey rounds analyzed here, DHS also collected ical justification of the model, which comprises data on height and weight of children aged less variables determining selection probability but than 5 years old, enabling calculation of anthro- exogenous to child nutritional status. Variables pometric indicators. satisfying this criterion are the community (non- Multivariate analysis of z scores is usually car- self) means of attended births and mortality ried out on individual child-level data using rates; means are estimated by mother's educa- ordinary least squares (OLS): tion level and sex of child, in order to improve zi = Xi + ei accuracy of the instrument. However, it would be incorrect to use a stan- where X is the vector of explanatory variables, dard Heckman procedure incorporating a first- the vector of coefficients to be estimated and stage probit on the full sample of dead and live e a normally distributed random error. However, children, since mortality/survival data are right- as noted above, the error term in OLS nutrition censored (see Annex C). The solution adopted equations is unlikely to be random due to selec- in this paper is to model the selection mecha- tivity in survival; if the unobserved factors deter- nism using survival analysis, the particular mining survival subsumed into the OLS error are method being Cox's proportional hazards model correlated with the Xs, OLS estimates of will be (Annex C). The convenience of survival analysis biased. They will also be biased for any X that is is that the predictions of the regression model correlated with e for reasons of endogeneity. are the individual hazard rates, which are the An appropriate model that accounts for same as inverse Mill's ratios. Note, however, that sample selection is Heckman's two-step proce- the estimated hazard rate is equal to the inverse dure. In the first stage the non-random selection Mill's ratio in the case of upper censoring (see variable--the probability of survival S--is esti- Greene 2000): mated using a probit model: Prob(Si = 1) = (Wi) i = Wi ( ) ( ) where is the set of coefficients estimated on W 1 - Wi explanatory variables and indicates the cumu- whereas the correct Mill's ratio for the nutrition lative standard normal density. From the fitted model is the lower-censoring case (above). The values of the probit model, the inverse Mill's solution is to model non-selection (mortality) ratio is calculated and included as an explanatory using a hazards model, generating the hazard variable in the second-stage regression model of rate which can be shown to be equal to the neg- nutrition determinants: ative of the inverse of Mill's ratio in the case of lower-censoring for the selection model.12 zi = Xi + Wi ( ) Wi ( ) + ui Results Table D.1 presents descriptive statistics of child being the probability density function of the nutritional status, as measured by anthropo- normal distribution and the estimated coeffi- metric variables, by various correlates.13 On the cient on the inverse of Mill's ratio.11 The proce- whole, children have better mean anthopometric dure produces consistent estimates assuming scores where they live in households that are that the error distribution of selection and re- non-poor (measured using a threshold value of 1 1 3 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? ously? Multivariate regression Weighted Means of HAZ, WAZ, and analysis enables this question T A B L E D . 1 WHZ for Regression Sample by to be answered, while also ac- Various Characteristics counting for censoring and di- rection of causality. Table D.2 HAZ score WAZ score WHZ score shows results of the Cox pro- mean mean mean portional hazards models for 1997 2000 1997 2000 1997 2000 under-fives. Deaths of other children in the community All children ­2.31 ­1.99 ­2.24 ­2.01 ­1.11 ­1.04 are a significantly positive de- Non-poor ­1.94 ­1.72 ­1.95 ­1.78 ­0.99 ­0.94 Poor ­2.53 ­2.22 ­2.41 ­2.21 ­1.18 ­1.13 terminant of risk of death of Household head male ­2.30 ­1.99 ­2.24 ­2.01 ­1.11 ­1.04 the index child; the coeffi- Household head female ­2.37 ­1.82 ­2.22 ­1.91 ­1.05 ­1.02 cient on (non-self) share of No electricity ­2.45 ­2.15 ­2.34 ­2.15 ­1.15 ­1.10 children delivered by a doctor Electricity ­1.83 ­1.59 ­1.88 ­1.68 ­0.99 ­0.90 Unsanitary toilet ­2.48 ­2.17 ­2.35 ­2.16 ­1.13 ­1.11 or nurse has a negative sign Sanitary toilet ­2.05 ­1.81 ­2.08 ­1.86 ­1.09 ­0.98 but is estimated imprecisely; Non-piped water ­2.34 ­2.01 ­2.26 ­2.03 ­1.12 ­1.06 other determinants were Piped water ­1.29 ­1.32 ­1.46 ­1.33 ­0.84 ­0.67 identified and discussed else- Mother no schooling ­2.52 ­2.20 ­2.40 ­2.19 ­1.17 ­1.12 where (Annex C). The results Mother some primary education ­2.33 ­2.12 ­2.22 ­2.12 ­1.07 ­1.10 Mother completed primary ­2.19 ­1.87 ­2.11 ­1.89 ­1.01 ­0.97 are discussed for determi- Mother some secondary educ ­1.77 ­1.66 ­1.89 ­1.74 ­1.03 ­0.92 nants of height for age z score Mother completed secondary ­1.42 ­1.26 ­1.64 ­1.44 ­1.00 ­0.80 (HAZ), weight for age z score Mother higher educated ­0.98 ­0.98 ­1.25 ­1.18 ­0.79 ­0.72 (WAZ), and weight for height Mother mobility low ­2.30 ­2.06 ­2.24 ­2.07 ­1.12 ­1.06 z score (WHZ) separately. In Mother mobility medium ­2.38 ­1.97 ­2.27 ­1.99 ­1.11 ­1.03 Mother mobility high ­1.76 ­1.79 ­1.86 ­1.87 ­1.01 ­1.02 each case, standard OLS Mother agency low ­2.10 ­2.10 ­1.07 regressions and two-step re- Mother agency medium ­1.92 ­1.96 ­1.02 gressions controlling for sur- Mother agency high ­1.86 ­1.91 ­1.03 vival selectivity are presented. Female child ­2.34 ­2.03 ­2.27 ­2.05 ­1.09 ­1.03 Modeling is undertaken for Male child ­2.27 ­1.94 ­2.20 ­1.97 ­1.13 ­1.05 Barisal Division ­2.42 ­2.06 ­2.15 ­2.10 ­0.90 ­1.12 children aged between 6 and Chittagong Division ­2.44 ­2.06 ­2.35 ­2.04 ­1.18 ­1.04 60 months,14 and separately Dhaka Division ­2.31 ­1.97 ­2.23 ­1.97 ­1.09 ­1.00 for 1997 and 2000 data. A Khulna Division ­2.00 ­1.77 ­2.04 ­1.82 ­1.08 ­0.96 Chow test indicated that the Rajshahi Division ­2.18 ­1.93 ­2.18 ­2.03 ­1.15 ­1.11 Sylhet Division ­2.61 ­2.25 ­2.40 ­2.20 ­1.08 ­1.08 coefficients from separate re- Rural ­1.79 ­1.62 ­1.85 ­1.73 ­0.98 ­0.96 gressions for each year were Urban ­2.36 ­2.06 ­2.27 ­2.06 ­1.12 ­1.06 jointly significantly different from one another. the wealth index), in which the head is female, Height for Age which have electricity, sanitary toilet facilities and The HAZ regressions provide a reasonably good piped water into the dwelling, and which are lo- fit of the data, as indicated by the values of R- cated in rural areas and outside of Chittagong and squared (tables D.3-D.4). The OLS regressions Sylhet Divisions. Children fare better, on average, indicate that: the higher their mother's level of education, the higher their mother's mobility and agency, and if · Household wealth exerts a significantly posi- they are of a non-Muslim religious group. tive effect on nutritional status. To what extent are these trends reproduced · Services are important determinants of HAZ, when controlling for many factors simultane- as indicated by the positive effect of access to 1 1 4 C H I L D M A L N U T R I T I O N D U R I N G T H E 1 9 9 0 s electricity and, in 1997, use of a sanitary toilet facility. In contrast, although the estimated Under-Five Mortality: T A B L E D . 2 coefficient is positive, access to piped water Cox Regression Results into the dwelling does not appear to signifi- cantly improve HAZ. 1997 2000 · Maternal education at secondary or higher HAZ Ratio z HAZ ratio z levels results in significantly better child height. Children also have greater HAZ where Wealth 0.834 ­0.63 0.711 ­1.15 their mothers have more say in household Electricity 0.836 ­1.22 0.845 ­1.15 Water & sanitation 0.383 ­1.56 0.569 ­1.18 decisionmaking regarding cooking, pur- Mother primary ed 0.951 ­0.45 0.860 ­1.23 chases, and health care, measured by the Mother lower secondary 0.822 ­1.02 0.974 ­0.15 agency index. Children of shorter and Mother higher secondary 0.741 ­0.56 0.269** ­2.18 younger mothers have lower HAZ. Contraceptive knowledge 0.924*** ­2.65 0.880*** ­3.57 · Girls under five are significantly shorter than Multiple birth 7.254*** 12.45 6.534*** 9.53 boys, controlling for other factors, particularly Mother age at birth 0.977* ­1.80 0.988 ­0.82 Previous child dead 1.095 0.70 1.022 0.13 in poorer households, as indicated by the in- Preceding interval <15 2.163*** 4.74 2.789*** 4.56 teraction between gender and household Prev dead * prec interval 0.511 ­1.58 wealth. Female child 0.973 ­0.32 0.879 ­1.39 · Of the fertility-related factors, children of Birth order 0.868** ­2.00 0.726*** ­3.79 higher birth order and those born less than Birth order sq 1.209*** 3.59 1.343*** 5.08 Female head 1.032 0.19 1.010 0.05 24 months after the preceding child tend to Maternal mobility 0.837** ­2.51 0.856* ­1.78 be shorter. Children living in households with Maternal agency 1.009 0.53 more adults relative to children fare better Chittagong Division 0.824 ­1.21 0.702* ­1.79 nutritionally. These variables were omitted in Dhaka Division 0.823 ­1.28 1.013 0.07 order to test for presence of bias on other co- Khulna Division 0.716* ­1.76 0.594** ­2.31 efficients, the effect being to increase the Rajshahi Division 0.823 ­1.27 0.867 ­0.74 Sylhet Division 0.988 ­0.07 1.180 0.82 magnitude and significance of the coefficient Rural 0.747* ­1.91 0.745** ­2.18 on the inverse Mill's ratio (results not re- Mean­deaths in community 3.988*** 2.62 3.406*** 2.61 ported). Mean­doctor/nurse delivery 0.768 ­0.72 0.928 ­0.24 · Children born in rural areas in the lean season Food deficit 0.988 ­0.10 are shorter on average, as are children born in Low birth weight 1.414*** 3.29 BINP 1.306 0.92 Barisal, Chittagong, and, especially, Sylhet, BINP * mother educ 0.413*** ­3.01 the poorest division. # Obs 6332 6974 · Results from 2000 indicate that children living # Deaths 610 522 in households with food deficit during part or Wald chi­squared 319.333*** 350.888*** all of the year and low birth weight children Log­likelihood ­4984.2 ­4400.6 are significantly more likely to be short for Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. their age. · It was possible to estimate the effect of whether the household resides in a cluster Table D.3 also shows results of HAZ regres- covered by the Bangladesh Integrated Nu- sions controlling for selectivity in survival using trition Project (BINP) in 2000.15 The BINP vari- the two-step method. Estimated coefficients on able is insignificant alone, but significantly selection terms (LAMBDA) in each equation are positively correlated with HAZ when included significantly positive, providing statistical evi- alongside the interaction between BINP and dence for selectivity in survival, and indicating maternal education, which itself is signifi- that the unobserved characteristics determining cantly negative, suggesting that the effect of mortality risk and malnutrition are positively cor- BINP is stronger for mothers with low levels related--in other words, children who are more of education. likely to die are estimated to be those more 1 1 5 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? T A B L E D . 3 Results of OLS and Two-Step Regressions of HAZ Score 1997 (OLS) 1997 (2­step) 2000 (OLS) 2000 (2­step) Coeff. t Coeff. t Coeff. t Coeff. t Wealth 0.525*** 3.00 0.502* 2.86 0.481*** 3.28 0.401*** 2.73 Finished floor 0.057 0.59 0.058 0.60 ­0.023 ­0.29 ­0.013 ­0.17 Electricity 0.228*** 3.25 0.214*** 3.05 0.164*** 2.98 0.145*** 2.62 Sanitary toilet 0.102** 1.96 0.100* 1.92 0.033 0.73 0.031 0.68 Piped water 0.165 1.16 0.141 0.98 0.031 0.32 ­0.007 ­0.07 Mother primary educ 0.072 1.24 0.062 1.08 0.036 0.73 ­0.003 ­0.05 Mother lower secondary 0.316*** 3.92 0.294*** 3.62 0.243*** 3.58 0.222*** 3.25 Mother higher secondary 0.687*** 4.23 0.659*** 4.04 0.505*** 4.15 0.413*** 3.32 Adult/child ratio 0.071*** 2.93 0.070*** 2.89 0.014 0.79 0.012 0.70 Mother age 0.023*** 3.59 0.021*** 3.20 0.025*** 4.42 0.023*** 4.10 Female head ­0.043 ­0.44 ­0.038 ­0.38 0.063 0.71 0.067 0.74 Mother mobility 0.019 0.53 0.008 0.21 0.035 1.03 0.008 0.23 Mother agency 0.014* 1.89 0.013* 1.83 Log mother height 0.968*** 3.86 0.971*** 3.85 0.743*** 3.49 0.743*** 3.50 Female child ­0.154** ­2.32 ­0.160** ­2.42 ­0.115* ­1.86 ­0.146** ­2.39 Female*wealth 0.237 1.37 0.248 1.44 0.195 1.41 0.221 1.62 Birth order ­0.113** ­2.56 ­0.120*** ­2.75 ­0.164*** ­3.83 ­0.198*** ­4.54 Birth order square 0.037 0.89 0.050 1.23 0.075* 1.87 0.110*** 2.66 Prec interval<24 ­0.126** ­2.37 ­0.115** ­2.16 ­0.220*** ­4.36 ­0.192*** ­3.76 Child's age ­0.071*** ­10.89 ­0.069*** ­10.46 ­0.045*** ­7.95 ­0.042*** ­7.42 Child's age sq/10 0.083*** 8.39 0.081*** 8.15 0.051*** 6.04 0.048*** 5.66 Rural 0.155* 1.74 0.139 1.55 ­0.050 ­0.72 ­0.084 ­1.18 Born Feb­Mar*rural ­0.139* ­1.86 ­0.145* ­1.95 ­0.251*** ­3.41 ­0.253*** ­3.47 Born Apr­May*rural ­0.137* ­1.71 ­0.142* ­1.77 ­0.078 ­1.07 ­0.082 ­1.14 Born Jun­Jul*rural ­0.149* ­1.83 ­0.152* ­1.87 ­0.015 ­0.20 ­0.017 ­0.23 Born Aug­Sep*rural ­0.012 ­0.16 ­0.016 ­0.21 0.040 0.54 0.035 0.49 Born Oct­Nov*rural 0.006 0.08 0.003 0.04 0.042 0.64 0.042 0.65 Chittagong Division ­0.026 ­0.31 ­0.041 ­0.49 ­0.052 ­0.67 ­0.091 ­1.15 Dhaka Division 0.144* 1.86 0.126 1.62 0.044 0.57 0.056 0.72 Khulna Division 0.437*** 4.93 0.408*** 4.54 0.190** 2.41 0.134* 1.67 Rajshahi Division 0.370*** 4.69 0.350*** 4.41 0.196** 2.48 0.177** 2.23 Sylhet Division ­0.130 ­1.40 ­0.117 ­1.26 ­0.087 ­1.00 ­0.024 ­0.27 Mean­measles vacc. 0.046 0.15 0.050 0.16 0.139 0.62 0.144 0.63 Food deficit ­0.073 ­1.57 ­0.077* ­1.66 Low birth weight ­0.424*** ­8.23 ­0.358*** ­6.78 BINP 0.241* 1.92 0.266** 2.09 BINP*educ ­0.183*** ­2.91 ­0.231*** ­3.56 Constant ­6.753*** ­5.27 ­6.613*** ­5.13 ­5.448*** ­4.97 ­5.121*** ­4.67 Lambda 0.815** 2.00 1.993*** 3.70 # Obs 4,258 4,258 4,719 4,719 F-statistic 24.7*** 24.1*** 21.0*** 21.2*** R-squared 0.159 0.160 0.153 0.158 Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. 1 1 6 C H I L D M A L N U T R I T I O N D U R I N G T H E 1 9 9 0 s likely to be malnourished. The correct specifica- tricity, and maternal education (particularly sec- tion is therefore the two-step method, which is ondary) have significantly positive effects on reported for all subsequent HAZ regressions. child weight.17 Age of the child has a significantly The routes through which education affects convex relationship with WAZ. The survival se- child anthropometry can be examined using in- lection terms (LAMBDA) are again significant, dicators of women's literacy and knowledge indicating that the two-step regressions ac- (table D.4). Maternal literacy and, to a lesser ex- counting for non-random selection bias are the tent, contraception knowledge (a proxy for better specification.18 knowledge of health and sanitary practices) have However, there are some differences with positive impacts on HAZ of children when used height for age results. For example, piped water independently of education. However, inclusion in 2000 has a significantly positive effect on of schooling variables (specification 2) knocks weight for age. On top of the effect of household out the significance of literacy, while leaving the wealth, children living in households with fin- impact of contraception knowledge virtually un- ished, as opposed to mud and wooden, floors affected. This suggests the beneficial impact of are estimated to be heavier for their age, consis- maternal schooling on child nutritional status tent with a hypothesis that cleanable floors can may operate partially through improving liter- help reduce the spread of pathogens. Children acy, though there are likely to be other impor- measured during the interviews carried out in tant factors at work. November and December were also found to The preceding analysis attempted to control have the lowest WAZ scores on average. Measles for child immunization using the (non-self) vaccination, measured using the (non-self) share share of children receiving measles vaccination, of children vaccinated in the community has a which, though a positive sign, as expected, was positive effect on WAZ, which is significant in insignificant in all regressions. We now restrict 2000. analysis to children aged over 12 months in order that the individual child's immunization Weight for Height status can be used. Table D.5 shows results of re- In contrast with HAZ and WAZ, the WHZ score gression for children aged over 12 months, in- regressions fit much more poorly (table D.7). cluding the indicator of whether the child Significantly positive impacts on WHZ are found received measles vaccination. In two-step re- for wealth, finished flooring, piped water gressions (specification 1), the effect of measles (though sanitary toilet is estimated to have a per- is significantly positive in 2000. However, due to the possibility of upwards bias on the measles verse negative impact in 1997), and measles vac- vaccination coefficient due to unobserved het- cinations in 2000. The selection terms in the erogeneity, the second columns (specification two-step regressions do not provide evidence 2) for each year report regression results ac- for non-random sample selection bias; we may counting for the endogeneity of vaccinations (as expect mortality to be less closely related to a well as the survival selectivity).16 The estimated short-term malnutrition indicator such as WHZ coefficients on RHO indicate that the errors are than longer-term one (HAZ or WAZ). correlated between the equations (though the correlation is marginally insignificant), while the Decompositions estimated coefficient on measles becomes in- Decompositions examining the contribution of significant. Thus, though evidence is not con- different variables to the improvements in an- clusive, the results appear to provide indication thropometric outcomes over time are presented of upwards bias in the estimated effect of vacci- in table D.8, which shows the means of the ex- nation due to unobserved heterogeneity. planatory variables used in the calculations (which are restricted to those variables which Weight for Age are statistically significant in regression analysis), Results for weight for age (table D.6) are broadly and the percentage change in anthropometric similar to those for height for age. Wealth, elec- measure due to each variable. The most impor- 1 1 7 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Impact of Maternal Education on HAZ score (OLS and T A B L E D . 4 Two-Step Regressions) 1997 (1) 1997 (2) 2000 (1) 2000 (2) Coeff. t Coeff. t Coeff. t Coeff. t Wealth 0.536*** 3.05 0.491*** 2.76 0.423*** 2.89 0.392*** 2.65 Finished floor 0.087 0.89 0.061 0.63 0.022 0.29 ­0.012 ­0.16 Electricity 0.214*** 3.05 0.215*** 3.06 0.138** 2.50 0.144*** 2.60 Sanitary toilet 0.099* 1.91 0.095* 1.83 0.039 0.88 0.031 0.69 Piped water 0.208 1.45 0.145 1.01 0.015 0.15 ­0.008 ­0.08 Mother primary educ 0.052 0.81 ­0.019 ­0.34 Mother lower secondary 0.271** 2.37 0.174* 1.89 Mother higher secondary 0.628*** 3.49 0.368*** 2.72 Literacy 0.171*** 2.66 0.019 0.22 0.151*** 2.86 0.054 0.76 Contraception 0.021 1.29 0.019 1.17 0.002 0.10 0.000 ­0.01 Adult/child ratio 0.074*** 3.04 0.071*** 2.92 0.012 0.72 0.012 0.68 Mother age 0.023*** 3.59 0.021*** 3.24 0.025*** 4.56 0.023*** 4.09 Female head ­0.038 ­0.38 ­0.036 ­0.36 0.051 0.57 0.063 0.71 Mother mobility 0.013 0.37 0.003 0.07 0.017 0.51 0.008 0.24 Mother agency 0.014* 1.85 0.013* 1.83 Log mother height 0.987*** 3.85 0.966*** 3.82 0.748*** 3.49 0.741*** 3.49 Female child ­0.164** ­2.47 ­0.159** ­2.40 ­0.142** ­2.32 ­0.145** ­2.37 Female*wealth 0.269 1.55 0.250 1.45 0.221 1.61 0.220 1.61 Birth order ­0.136*** ­3.12 ­0.121*** ­2.77 ­0.210*** ­4.86 ­0.198*** ­4.54 Birth order square 0.060 1.46 0.051 1.25 0.117*** 2.85 0.110*** 2.67 Prec interval<24 ­0.106** ­1.99 ­0.113** ­2.13 ­0.183*** ­3.61 ­0.191*** ­3.76 Child's age ­0.069*** ­10.42 ­0.070*** ­10.51 ­0.042*** ­7.40 ­0.042*** ­7.39 Child's age sq/10 0.081*** 8.10 0.082*** 8.21 0.048*** 5.60 0.048*** 5.64 Rural 0.133 1.47 0.142 1.58 ­0.087 ­1.23 ­0.084 ­1.19 Born Feb­Mar*rural ­0.144* ­1.94 ­0.146** ­1.96 ­0.254*** ­3.48 ­0.253*** ­3.47 Born Apr­May*rural ­0.134* ­1.67 ­0.143* ­1.78 ­0.080 ­1.10 ­0.083 ­1.15 Born Jun­Jul*rural ­0.155* ­1.91 ­0.153* ­1.89 ­0.019 ­0.26 ­0.018 ­0.25 Born Aug­Sep*rural ­0.012 ­0.15 ­0.017 ­0.22 0.037 0.50 0.035 0.48 Born Oct­Nov*rural 0.006 0.08 0.002 0.02 0.045 0.69 0.043 0.66 Chittagong Division ­0.022 ­0.26 ­0.026 ­0.31 ­0.068 ­0.86 ­0.087 ­1.09 Dhaka Division 0.139* 1.80 0.134* 1.72 0.074 0.96 0.061 0.78 Khulna Division 0.419*** 4.67 0.413*** 4.60 0.145* 1.81 0.138* 1.71 Rajshahi Division 0.354*** 4.49 0.357*** 4.49 0.204*** 2.56 0.182** 2.29 Sylhet Division ­0.094 ­0.98 ­0.090 ­0.94 ­0.002 ­0.02 ­0.021 ­0.23 Mean­measles vacc. 0.309 1.01 0.045 0.14 0.269 1.24 0.131 0.58 Food deficit ­0.086* ­1.86 ­0.077* ­1.67 Low birth weight ­0.359*** ­6.78 ­0.357*** ­6.73 BINP 0.251** 1.96 0.269** 2.11 BINP*educ ­0.223*** ­3.36 ­0.232*** ­3.57 Constant ­6.840*** ­5.23 ­6.673*** ­5.16 ­5.217*** ­4.72 ­5.106*** ­4.66 Lambda 0.795* 1.91 0.739* 1.77 1.978*** 3.57 1.994*** 3.52 # Obs 4,258 4,258 4,719 4,719 F-statistic 23.6*** 22.8*** 20.9*** 20.2*** R-squared 0.157 0.160 0.155 0.158 Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. 1 1 8 C H I L D M A L N U T R I T I O N D U R I N G T H E 1 9 9 0 s Impact of Measles Vaccination on HAZ Score, Children T A B L E D . 5 Aged 12-59 Months 1997 2000 (1) (2) (1) (2) Coeff. t Coeff. t Coeff. t Coeff. t Dep var. = HAZ score Wealth 0.606*** 3.82 0.600*** 3.78 0.507*** 3.62 0.519*** 3.70 Finished floor 0.107 1.02 0.109 1.04 ­0.021 ­0.27 ­0.022 ­0.28 Electricity 0.221*** 2.93 0.209*** 2.73 0.112** 1.99 0.121** 2.14 Sanitary toilet 0.124** 2.25 0.124** 2.24 0.049 1.03 0.051 1.07 Piped water 0.074 0.49 0.075 0.49 ­0.076 ­0.77 ­0.074 ­0.75 Mother primary educ 0.035 0.59 0.025 0.41 ­0.017 ­0.32 ­0.006 ­0.12 Mother lower secondary 0.282*** 3.44 0.271*** 3.25 0.245*** 3.63 0.268*** 3.86 Mother higher secondary 0.602*** 3.71 0.588*** 3.61 0.468*** 3.82 0.490*** 3.99 Adult/child ratio 0.091*** 3.51 0.091*** 3.52 0.018 0.93 0.017 0.91 Mother age 0.022*** 3.32 0.021*** 3.11 0.021*** 3.57 0.022*** 3.73 Female head ­0.060 ­0.59 ­0.065 ­0.62 0.087 0.93 0.086 0.92 Mother mobility 0.005 0.13 0.002 0.05 0.024 0.70 0.028 0.81 Mother agency 0.012 1.53 0.013* 1.68 Log mother height 0.986*** 4.33 0.986*** 4.33 0.670*** 3.28 0.673*** 3.29 Female child ­0.110** ­2.44 ­0.104** ­2.29 ­0.104*** ­2.64 ­0.111*** ­2.78 Birth order ­0.118*** ­2.69 ­0.115*** ­2.61 ­0.195*** ­4.28 ­0.200*** ­4.36 Birth order square 0.039 0.98 0.040 1.00 0.119*** 2.75 0.119*** 2.76 Prec interval<24 ­0.109** ­1.96 ­0.110** ­1.97 ­0.181*** ­3.40 ­0.182*** ­3.43 Child's age ­0.017* ­1.85 ­0.017* ­1.86 0.000 0.05 0.000 0.06 Child's age sq/10 0.015 1.19 0.015 1.20 ­0.006 ­0.53 ­0.006 ­0.54 Rural 0.180* 1.89 0.182* 1.90 ­0.055 ­0.75 ­0.061 ­0.83 Born Feb­Mar*rural ­0.254*** ­3.16 ­0.254*** ­3.16 ­0.372*** ­4.73 ­0.372*** ­4.74 Born Apr­May*rural ­0.280*** ­3.18 ­0.280*** ­3.18 ­0.222*** ­2.91 ­0.222*** ­2.90 Born Jun­Jul*rural ­0.209** ­2.40 ­0.209** ­2.40 ­0.108 ­1.36 ­0.108 ­1.35 Born Aug­Sep*rural ­0.041 ­0.50 ­0.041 ­0.51 ­0.001 ­0.01 0.000 ­0.01 Born Oct­Nov*rural 0.024 0.32 0.024 0.32 0.045 0.68 0.046 0.69 Chittagong Division ­0.048 ­0.54 ­0.038 ­0.42 ­0.059 ­0.73 ­0.061 ­0.74 Dhaka Division 0.172** 2.09 0.180** 2.19 0.126 1.57 0.109 1.35 Khulna Division 0.481*** 5.00 0.470*** 4.82 0.194** 2.34 0.197** 2.38 Rajshahi Division 0.406*** 4.82 0.401*** 4.73 0.267*** 3.21 0.262*** 3.15 Sylhet Division ­0.050 ­0.51 ­0.036 ­0.36 0.042 0.44 0.022 0.23 Measles vacc. 0.018 0.30 0.135 0.86 0.105** 2.03 ­0.068 ­0.50 Food deficit ­0.097** ­1.99 ­0.097** ­1.98 Low birth weight ­0.349*** ­6.28 ­0.347*** ­6.24 BINP 0.257* 1.89 0.274** 2.01 BINP*mother educ ­0.230*** ­3.58 ­0.233*** ­3.63 Constant ­7.733*** ­6.59 ­7.798*** ­6.62 ­5.647*** ­5.30 ­5.556*** ­5.21 Lambda 0.606*** 3.82 0.689* 1.78 1.812*** 3.34 1.862*** 3.40 Dep var. = Measles vaccination Wealth 0.173 1.04 0.214 1.42 Electricity 0.398*** 4.61 0.158*** 2.28 Previous child died ­0.058 ­0.71 0.085 1.04 Prec interval < 15 0.153 1.10 ­0.192 ­1.57 Birth order ­0.109*** ­4.94 ­0.091*** ­4.40 (Continued ) 1 1 9 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Impact of Measles Vaccination on HAZ Score, Children T A B L E D . 5 Aged 12-59 Months (continued) 1997 2000 (1) (2) (1) (2) Coeff. t Coeff. t Coeff. t Coeff. t Dep var. = HAZ score Measles vacc. 0.018 0.30 0.135 0.86 0.105** 2.03 ­0.068 ­0.50 Mother's age at birth 0.036*** 4.78 0.027*** 4.01 Female child ­0.170*** ­3.43 ­0.145*** ­3.04 Female head 0.144 1.24 0.020 0.17 Mother primary educ 0.287*** 4.18 0.165*** 2.69 Mother lower secondary 0.371*** 3.58 0.473*** 5.26 Mother higher secondary 1.206*** 3.84 0.974*** 3.56 Contraception 0.038** 2.02 0.055*** 2.99 Maternal mobility 0.085* 1.89 0.063 1.40 Maternal agency 0.021** 2.28 Mother remarried ­0.113 ­1.06 ­0.467*** ­4.22 Community TV 0.063 0.88 0.187*** 2.65 Distance to Thana hq ­0.004 ­0.65 ­0.026*** ­4.22 Chittagong Division ­0.270*** ­2.59 0.026 0.26 Dhaka Division ­0.229** ­2.39 ­0.282*** ­2.97 Khulna Division 0.481*** 3.68 0.138 1.27 Rajshahi Division 0.192* 1.92 ­0.060 ­0.60 Sylhet Division ­0.332*** ­2.94 ­0.262** ­2.44 BINP 0.298** 2.45 Constant ­0.179 ­0.71 0.335 1.45 Rho ­0.053 ­0.83 0.088 1.36 Chi­squared (rho = 0) 0.69 1.95 Prob > chi-squared 0.41 0.16 # Obs 3,779 3,779 4,243 4,243 F-statistic 18.3*** 19.3*** R-squared 0.135 0.153 Wald chi-squared 590.4*** 721.2*** Log likelihood ­8204.6 ­8857.8 Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. tant factors explaining mean improvements in Bangladesh. Presence of BINP in the community anthropometry over time are general increases has been beneficial to the nutritional status of in wealth, secondary education of mothers (par- children of lesser educated mothers. This is in ticularly up to Grade 10) and electricity. The contrast to the effect of BINP on survival (Annex large estimated effect of changes in birth order C), which was estimated to be beneficial for bet- over time on HAZ and WAZ indicate that reduc- ter educated mothers only. One explanation for tions in fertility have also played an important these seemingly contradictory results is that cer- role in improving nutrition. tain aspects of BINP contribute to malnutrition reduction, e.g., supplementary feeding and nu- Conclusion trition education, because they are most benefi- The conclusion that emerges from this analysis cial to less educated mothers; in contrast, BINP is that improvements in wealth and maternal reduces mortality most effectively where moth- secondary education had strong effects on re- ers are more willing to seek modern medical ducing child malnutrition in the late 1990s in care, which is more likely among the educated. 1 2 0 C H I L D M A L N U T R I T I O N D U R I N G T H E 1 9 9 0 s T A B L E D . 6 Results of OLS and Two-Step Regressions of WAZ Score 1997 (OLS) 1997 (2-step) 2000 (OLS) 2000 (2-step) Coeff. t Coeff. t Coeff. t Coeff. t Wealth 0.620*** 5.51 0.61*** 5.40 0.489*** 4.78 0.444*** 4.30 Finished floor 0.151* 1.89 0.15* 1.90 0.090 1.40 0.095 1.49 Electricity 0.171*** 2.91 0.16*** 2.72 0.115*** 2.73 0.102** 2.42 Sanitary toilet ­0.014 ­0.35 ­0.02 ­0.40 0.014 0.41 0.013 0.38 Piped water 0.049 0.40 0.03 0.25 0.172* 1.92 0.145 1.61 Mother primary educ 0.073* 1.68 0.07 1.51 0.012 0.31 ­0.015 ­0.40 Mother lower secondary 0.130** 2.13 0.11* 1.84 0.155*** 2.96 0.141*** 2.70 Mother higher secondary 0.474*** 3.13 0.45*** 2.98 0.335*** 3.13 0.270** 2.50 Adult/child ratio 0.051*** 2.94 0.05*** 2.89 0.018 1.23 0.017 1.16 Mother age 0.014*** 2.94 0.01** 2.55 0.013*** 3.15 0.012*** 2.84 Female head 0.017 0.23 0.02 0.29 0.025 0.38 0.027 0.42 Mother mobility 0.000 ­0.01 ­0.01 ­0.32 0.013 0.52 ­0.006 ­0.23 Mother agency 0.007 1.32 0.007 1.28 Log mother height 0.858*** 4.73 0.86*** 4.72 0.466*** 3.70 0.467*** 3.72 Female child ­0.068** ­2.13 ­0.07** ­2.19 ­0.053* ­1.89 ­0.069** ­2.44 Birth order ­0.063* ­1.78 ­0.07** ­1.96 ­0.064** ­2.01 ­0.088*** ­2.70 Birth order square 0.009 0.28 0.02 0.58 0.018 0.59 0.042 1.36 Prec interval<24 ­0.066 ­1.60 ­0.06 ­1.40 ­0.099*** ­2.63 ­0.080** ­2.10 Child's age ­0.031*** ­6.06 ­0.03*** ­5.67 ­0.022*** ­4.76 ­0.020*** ­4.34 Child's age sq/10 0.045*** 5.84 0.04*** 5.59 0.030*** 4.47 0.028*** 4.15 Rural 0.175** 2.43 0.16** 2.19 0.072 1.37 0.045 0.85 Interview Feb*rural ­0.079 ­1.31 ­0.08 ­1.29 ­0.102* ­1.95 ­0.103** ­1.99 Interview Mar*rural ­0.004 ­0.04 ­0.01 ­0.11 ­0.114 ­1.45 ­0.110 ­1.40 Interview Nov*rural ­0.127** ­2.48 ­0.13** ­2.49 ­0.094 ­1.64 ­0.084 ­1.46 Interview Dec*rural ­0.189*** ­3.50 ­0.18*** ­3.42 ­0.106* ­1.78 ­0.101* ­1.70 Chittagong Division ­0.186*** ­2.69 ­0.20*** ­2.85 0.012 0.16 ­0.008 ­0.11 Dhaka Division ­0.056 ­0.90 ­0.07 ­1.12 0.026 0.36 0.041 0.57 Khulna Division 0.165** 2.28 0.14* 1.94 0.197*** 3.19 0.159** 2.54 Rajshahi Division 0.033 0.48 0.02 0.27 0.088 1.39 0.078 1.23 Sylhet Division ­0.108 ­1.40 ­0.10 ­1.30 ­0.051 ­0.74 ­0.005 ­0.07 Mean­measles vacc. 0.238 1.17 0.24 1.18 0.376 1.87 0.378 1.88 Food deficit ­0.053 ­1.53 ­0.056 ­1.62 Low birth weight ­0.434*** ­11.49 ­0.388*** ­10.01 BINP 0.225*** 2.64 0.243*** 2.82 BINP*educ ­0.076 ­1.45 ­0.109** ­2.04 Constant ­6.611*** ­7.15 ­6.50*** ­6.99 ­4.495*** ­6.87 ­4.282*** ­6.53 Lambda 0.61* 1.78 1.393*** 3.81 # Obs 4,258 4,258 4,719 4,719 F-statistic 16.5*** 16.0*** 19.3*** 19.2*** R-squared 0.121 0.122 0.152 0.156 Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. 1 2 1 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? T A B L E D . 7 Results of OLS and Two­Step Regressions of WHZ Score 1997 (OLS) 1997 (2-step) 2000 (OLS) 2000 (2-step) Coeff. t Coeff. t Coeff. t Coeff. t Wealth 0.376*** 3.32 0.37*** 3.29 0.205** 2.29 0.193** 2.13 Finished floor 0.171** 2.08 0.17** 2.08 0.142** 2.53 0.143** 2.56 Electricity 0.027 0.47 0.02 0.40 0.037 0.95 0.033 0.85 Sanitary toilet ­0.097** ­2.42 ­0.10** ­2.44 ­0.008 ­0.25 ­0.008 ­0.26 Piped water ­0.059 ­0.50 ­0.07 ­0.55 0.204** 2.55 0.196** 2.43 Mother primary educ 0.048 1.08 0.05 1.02 ­0.007 ­0.21 ­0.015 ­0.43 Mother lower secondary ­0.055 ­0.82 ­0.06 ­0.90 0.032 0.64 0.028 0.56 Mother higher secondary 0.134 0.83 0.13 0.78 0.047 0.47 0.029 0.29 Adult/child ratio 0.011 0.61 0.01 0.60 0.015 1.12 0.015 1.10 Mother age 0.002 0.32 0.00 0.20 ­0.002 ­0.42 ­0.002 ­0.53 Female head 0.061 0.84 0.06 0.86 ­0.006 ­0.11 ­0.005 ­0.10 Mother mobility ­0.008 ­0.27 ­0.01 ­0.37 ­0.003 ­0.13 ­0.009 ­0.36 Mother agency ­0.001 ­0.31 ­0.002 ­0.32 Female child 0.054* 1.66 0.05 1.63 0.034 1.31 0.029 1.13 Birth order ­0.014 ­0.42 ­0.02 ­0.48 0.034 1.19 0.028 0.93 Birth order square 0.001 0.03 0.00 0.14 ­0.028 ­1.10 ­0.022 ­0.80 Prec interval<24 ­0.005 ­0.13 0.00 ­0.06 0.014 0.42 0.020 0.57 Child's age ­0.008 ­1.44 ­0.01 ­1.33 ­0.010** ­2.38 ­0.010** ­2.24 Child's age sq/10 0.016** 2.09 0.02** 2.02 0.017*** 2.73 0.017*** 2.62 Rural 0.106 1.45 0.10 1.36 0.159*** 3.39 0.152*** 3.17 Interview Feb*rural 0.032 0.52 0.03 0.53 ­0.072 ­1.53 ­0.072 ­1.54 Interview Mar*rural ­0.040 ­0.45 ­0.04 ­0.47 ­0.087 ­1.31 ­0.086 ­1.30 Interview Nov*rural ­0.125** ­2.34 ­0.12** ­2.35 ­0.199*** ­3.77 ­0.196*** ­3.70 Interview Dec*rural ­0.170*** ­3.03 ­0.17*** ­3.01 ­0.207*** ­3.88 ­0.205*** ­3.85 Chittagong Division ­0.280*** ­3.72 ­0.28*** ­3.77 ­0.036 ­0.54 ­0.042 ­0.62 Dhaka Division ­0.194*** ­2.87 ­0.20*** ­2.94 ­0.044 ­0.69 ­0.040 ­0.62 Khulna Division ­0.134* ­1.70 ­0.14* ­1.78 0.127** 2.25 0.116** 2.00 Rajshahi Division ­0.281*** ­3.77 ­0.29*** ­3.82 ­0.010 ­0.19 ­0.013 ­0.24 Sylhet Division ­0.086 ­1.02 ­0.08 ­0.99 0.007 0.13 0.021 0.34 Mean­measles vacc. 0.285 1.22 0.29 1.22 0.439** 2.50 0.440** 2.51 Food deficit ­0.011 ­0.35 ­0.012 ­0.38 Low birth weight ­0.257*** ­7.76 ­0.244*** ­6.91 BINP 0.141 1.61 0.146** 1.66 BINP*educ 0.019 0.39 0.009 0.19 Constant ­1.119*** ­5.97 ­1.08*** ­5.53 ­1.140*** ­7.82 ­1.078*** ­6.88 Lambda 0.21 0.75 0.394 1.11 # Obs 4,258 4,258 4,719 4,719 F-statistic 3.88*** 3.75*** 6.98*** 6.82*** R-squared 0.029 0.029 0.053 0.054 Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively.%. 1 2 2 C H I L D M A L N U T R I T I O N D U R I N G T H E 1 9 9 0 s Decompositions for HAZ, T A B L E D . 8 WAZ, and WHZ Regressions Mean (xj) Percentage change xj 1997 2000 HAZ WAZ WHZ Wealth 0.28 0.33 8.30 9.19 3.99 Finished floor 1.70 2.56 Electricity 0.23 0.29 3.56 2.51 Sanitary toilet 0.41 0.51 1.18 ­0.31 Piped water 0.41 0.56 Mother lower secondary 0.13 0.19 4.44 2.82 Mother higher secondary 0.02 0.03 1.53 1.00 Adult/child ratio 1.22 1.32 0.48 0.66 Mother's age 26.07 26.19 1.02 0.52 Log mother's height 5.00 5.00 0.51 0.32 Female child 0.50 0.49 0.42 0.20 Female child*wealth 0.14 0.16 2.06 Birth order 3.14 2.94 14.38 6.38 Birth order sq 1.44 1.28 ­6.59 ­2.52 Prec interval <24 0.41 0.42 ­0.65 ­0.27 Child's age 32.16 32.06 1.67 0.79 0.38 Child's age sq/10 12.77 12.71 ­1.21 ­0.70 ­0.41 Rural 0.91 0.83 2.32 ­1.25 ­4.21 Born Feb/Mar 0.16 0.13 2.80 Born Apr/May 0.13 0.13 0.17 Born Jun/Jul 0.12 0.11 0.10 Born Aug/Sep 0.13 0.13 0.00 Born Oct/Nov 0.17 0.17 ­0.04 Interview Feb ­1.45 ­1.01 Interview Mar ­1.24 ­0.98 Interview Nov 1.76 4.10 Interview Dec 3.75 7.64 Chittagong 0.24 0.22 0.87 0.08 0.40 Dhaka 0.31 0.31 0.05 0.04 ­0.04 Khulna 0.10 0.11 0.04 0.05 0.03 Rajshahi 0.21 0.23 1.06 0.47 ­0.08 Sylhet 0.06 0.07 ­0.08 ­0.02 0.07 Mean­measles vacc. 0.79 0.92 Note: % change gives the share of the total change explained due to each variable xj, that is: j2(xj2­xj1)/[ j ( j2(xj2­xj1)) + j (xj1( j2­ j1))]. 1 2 3 ANNEX E: WOMEN'S AGENCY, HOUSEHOLD STRUCTURE, AND HEALTH OUTCOMES Bangladesh is often characterized as a country in in which the adult male is absent for prolonged which women face many restrictions. This is cer- reasons for work is not recorded in the roster, so tainly so: recent data from DHS show that just that the wife may appear as the household head. over one in five women are solely responsible The first question asked about each person in for decisions relating to their own health, and the roster is their relationship to the household the large majority is not permitted to travel out- head. The responses to this question are tabu- side their household unaccompanied. But it is lated in table E.1, for all 54,627 respondents of also easy to overstate the case, and to ignore that the DHS 1999/2000 survey and, in the second things are changing, and changing quite rapidly. column, excluding the 3,190 respondents who This paper begins by presenting data from were not usual residents. The final two columns the three DHS surveys from Bangladesh on tabulate the share for usual residents by sex. women's role in decisionmaking (women's The "core" of each household is the head, agency) and women's mobility. These issues are their husband or wife and children. Just under linked to household structure, which is shown half of household members are children, around here to be a critical determinant of female au- one-third of men are household heads, and one- tonomy. Other determinants that may lay be- third of women are the wife of the head. As ex- hind the changes that took place in the 1990s, plored below, extended households are not such as increased women's work outside the uncommon, as shown by 4.0 percent who are home and the spread of secondary education, parents or parents-in-law and the 2.5 percent are also analyzed. who are a son or daughter-in-law. The gender A related issue is age at marriage. The data disaggregation shows these to be daughters-in- show that the age at marriage is increasing. In law: it is uncommon for a man to live with his most countries, raising the marriage age has parents-in-law (only 0.4 percent of men are the been a critical factor in reducing fertility. How- son-in-law of the head compared with the 5 per- ever, it is argued that this is not likely to be the cent of women who are daughters-in-law). case in Bangladesh. Comparing the shares for the whole roster for those of "usual residents" shows that grandchil- Household Structure dren, siblings, and other relatives are the most Household surveys contain a household roster likely visitors. The small percentage of "not re- in which basic information on all household lated" are generally not visitors, so these people members is collected. In DHS, as in most other are most likely to be live-in servants. The cate- surveys, all people who spend the night before gory co-spouse received zero responses. This the survey are included in the roster, with a fur- does not mean there are no polygamous house- ther question asking if that person is a "usual res- holds, since the relationship shown is that to the ident" of the household. The roster also includes household head; rather, it means there are no people deemed to be household members but polygamous households in which one of the absent the previous night. Not collecting this in- wives was identified as the head. For just eight formation means that female-headed house- people in total, the relationship was not known holds are "over-identified," that is, a household or not stated. 1 2 5 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? in some household surveys). It is not possible Relationship to Household T A B L E E . 1 with DHS data to distinguish these two Head (percent) household types,2 and White notes that the distinction is often more nominal than real. Usual residents only · Nuclear: a household including the spouse Whole roster Total Male Female of the head, but no parent, parent-in-law, or Head 18.0 19.1 34.7 3.3 son or daughter-in-law Wife or husband 15.8 16.7 0.2 33.5 · Female-headed household: we do not simply Son/daughter 43.9 45.3 50.6 40.0 take those households for which a woman is Son/daughter-in-law 2.8 2.7 0.4 5.0 listed as the head, as some of these women Grandchild 5.0 4.1 4.3 3.9 Parent 3.5 3.6 1.3 6.0 have husbands or live with another adult Parent-in-law 0.5 0.3 0.1 0.6 male (table E.1). Rather, we classify as FHH Brother/sister 3.4 3.1 3.9 2.4 those cases where a woman is identified as Co-spouse 0.0 0.0 0.0 0.0 the head and there is no spouse, male parent, Other relative 5.3 3.2 2.9 3.6 or son-in-law listed in the roster. Adopted/foster child 0.3 0.3 0.3 0.2 · Single-male-headed household: where there Not related 1.5 1.5 1.4 1.6 Don't know/missing 0.0 0.0 0.0 0.0 is no spouse listed. · Polygamous: where the head has more than Absolute number 54,627 51,437 25,877 25,560 one wife. Source: DHS 1999/2000. The last column of table E.2 shows the break- The data in the household roster can be used down of the 9,853 households covered by DHS to identify five household types, where this clas- 1999/2000 into these five categories. The major- sification is fully exhaustive and mutually exclu- ity of households are nuclear, though just over sive (that is, it covers all households with no one-third are extended. The share of female- overlaps):1 headed households is likely overstated for the reasons given above: that is, many of these · Extended: a household including either a par- households will have an absent male (husband) ent, parent-in-law, or son or daughter-in-law who supports the family and visits occasionally. of the household head. White (1992) dis- A woman left alone after death or desertion of tinguishes between extended and joint the spouse is more likely to return to her par- households, where the latter live in the same ents' household.3 The share of female-headed compound but separate production and con- households can also be overstated where co- sumption (crucially, they do not eat together, spouses live in separate households, which which is the definition of household adopted would also explain why the share of polygamous households is low, at less than one percent. Household Types by Unsurprising is that the share of single-male- T A B L E E . 2 Income Group (percent) headed households is low, since men typically remarry on death of, or separation from, the Income group spouse (White 1992; Kabeer 2004). Also unsurprising is the correlation between Low Middle High Total household size and structure: the majority of Extended 24.6 37.1 41.3 33.5 households between two to six people are nu- Nuclear 64.8 54.6 50.9 57.5 clear, but larger households are more likely to be Female head 8.8 6.6 6.4 7.4 Single male headed 1.0 0.8 0.5 0.8 extended (figure E.1). The largest household in Polygamous 0.8 0.9 0.9 0.8 the dataset is an extended family of 25 people Absolute number of consisting of an 80-year-old man and his 60-year- households 3,639 3,865 2,350 9,854 old wife, their four sons with their wives, one un- Source: DHS 1999/2000. married son aged 30, and 14 grandchildren.4 1 2 6 W O M E N ' S A G E N C Y , H O U S E H O L D S T R U C T U R E , A N D H E A L T H O U T C O M E S F I G U R E E . 1 Household Size and Household Structure 2,500 Extended Nuclear 2,000 Female headed Single male headed Polygamous 1,500 households of 1,000 Number 500 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 >15 Household size There is also a relationship between house- agency. The most recent survey contained a hold structure and economic well-being as mea- number of questions on women's say in deci- sured by an asset index based on housing quality sionmaking. Specifically, if the woman had a say and ownership of consumer durables (see in decisions relating to: Appendix C.2 on the construction of the asset index); see table E.2. Extended families are more · Her own health care common among higher-income groups, with · Child health care the converse being true for nuclear families. · Large household purchases Given that there is also the link between house- · Household purchases for daily needs hold size and structure, this finding means that · Visits to family, friends, or relatives large (extended) households are better off: a fact · What food to cook each day. that runs counter to "standard economics poverty profiles," in which household size is cor- For each question respondents could answer related with poverty, but is of no surprise to an- whether they make the decision alone, whether thropologists, who report that households with the husband makes it, "someone else" makes it, resources need and attract labor, mostly through or whether they decide "jointly with husband" or the family.5 For example, White (1992) attributes "jointly with someone else." the likelihood of a household disintegrating to Both DHS 1999/2000 and the earlier surveys its lack of resources. asked questions on women's mobility--that is, their ability to go to the market or health center Patterns of Women's Agency and Mobility either alone or with other family members. Measurement However, only one question (frequency of going The Bangladesh Demographic Health Surveys shopping) has been asked the same way in all contain a number of questions on women's three surveys (table E.3). 1 2 7 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? ranked/ordered with "don't go" as lowest rank- Questions on Women's ing, followed by going "with others" (including Mobility Contained in T A B L E E . 3 husband), with children, and finally alone. Different DHS Questionnaires Women's Agency: Descriptive Analysis Table E.4 summarizes the six aspects of women's DHS93 DHS96 DHS99 agency collected by DHS 1999/2000. These data If do go outside village/town/city alone X X are tabulated for three groups: all respondents If can go outside village/town/city alone X X X (which are ever-married women aged 11-49), If go to health center/hospital alone X X If can go health center/hospital alone X X X currently married women in male-headed Frequency of going shopping X X X households (since we argue below that women With whom do they go shopping X X in female-headed households largely make deci- Frequency of going outside village/town/city Xa X X sions by default), and women who live with their a. In 1993 the question was whether they go to another part of their village/town/city rather than if they mother-in-law, since the literature suggests that leave it. these women will have less say in decisionmak- ing. The data indeed show that women living For questions on frequency of going shop- with their mother-in-law have a diminished role ping or going outside village/ town/city, respon- in decisionmaking compared with other dents could answer "never," "several times a women, whereas "someone else" (presumably year," "once a month or more." Answers were the mother-in-law) is much more likely to have ranked/ordered accordingly. For questions on a say--and implicitly that women in female- whether they go or can go outside village, town, headed households decide more than those in or city or to health center, respondents could male-headed ones. We return to this issue choose to answer one of the following: "don't below. go," "alone," "with husband," "with children," or Other than the decision about what to cook, "other." For our purposes, going "with husband" fewer than one in five women are solely respon- and "other" were pooled and the responses sible for the decisions shown, but over half are T A B L E E . 4 Women's Say in Decisionmaking by Position in Household (percent) Own health Child's health Large purchases Daily purchases Visits Cooking All respondents (n = 9,716) Respondent 20.6 19.2 10.3 19.5 14.0 66.8 Joint 36.1 47.1 51.0 43.8 48.3 20.4 Husband 36.9 26.9 29.4 27.2 29.5 4.6 Someone else 6.4 6.9 9.3 9.6 8.2 8.2 Currently married women in male-headed households (n = 8,706) Respondent 15.7 14.3 6.1 15.5 9.3 66.8 Joint 36.8 50.0 53.6 46.1 51.1 20.4 Husband 41.0 29.9 32.5 30.2 32.7 4.6 Someone else 5.1 5.8 7.7 8.2 6.9 8.2 Living with mother-in-law (n = 2,017) Respondent 12.2 10.3 3.2 9.8 5.2 42.8 Joint 36.2 48.5 50.5 45.2 48.1 31.3 Husband 39.5 27.1 26.1 22.8 26.7 3.3 Someone else 12.2 14.5 20.12 22.2 20.0 22.6 1 2 8 W O M E N ' S A G E N C Y , H O U S E H O L D S T R U C T U R E , A N D H E A L T H O U T C O M E S involved in the decision either solely or jointly. percent) are married women and usual residents Women are least commonly solely responsible of a male-headed household. Over one-third of for the decision regarding large purchases; but these have a low level of agency, though nearly husbands are also less commonly solely respon- one-half have a medium level. We can distin- sible for this decision, which is the most com- guish the just over one-fifth of these women mon joint decision. who live with their mother-in-law from those Women's involvement in decisionmaking on who do not. There is a significant difference in matters related to their own health and that of agency between these two groups: women living their children is low. Forty-three percent of with their mother-in-law have a significantly women have no say in the decisions on their lower say in decisionmaking, with nearly half own health; this number rises to 52 percent for having low agency, and less than 10 percent hav- women living with their mother-in-law. With re- ing high agency. spect to decisions related to child health, these Agency is highest among the various cate- figures are 34 and 42 percent, respectively. gories of female-headed households. Nearly Using the DHS 1999/2000 data, an index of 95 percent of unmarried women heading their women's agency (AGENCY) was constructed in households with no man present have high two stages. First, the response to each of the six agency. Perhaps more surprising is the high questions was recoded so that the respondent agency of female-headed households where the deciding by herself = 2, with husband or some- woman is married but the man absent. Closer in- one else = 1, and someone else decides = 0. spection of the data shows that in these cases, Second, the responses to the six questions were the woman makes day-to-day decisions by her- summed together to give a 13-point scale from self, but the decision on large purchases is in 0 to12. A score of zero indicates that the respon- many cases made either jointly or just by the dent has no say in any of the six decisions, and husband. The third category comprises female- 12 means she is solely responsible for all of headed households living with at least one adult them. For purposes of presentation, three cate- male. These are women who are divorced or gories of AGENCY were created: low for AGENCY widowed who live with either or both a parent or between 0 and 4, medium for 5 to 7, and high for an adult son (defined as age greater than 15). values of AGENCY between 8 and 12 inclusive. These women also have a high level of agency. Table E.5 tabulates the categorical version of An implication of this analysis is that when the agency variable against the status of the looking at the correlates of agency, it makes woman in the household. The majority of the sense to focus on married women in male- women in the sample (8,706 out of 10,543; 83 headed households, who are the bulk of the T A B L E E . 5 Women's Agency by Status in the Household Women's agency (percent) Number of Woman's marital status Household type Low Medium High observations Male-headed households 37.9 48.1 14.0 8,706 Married o/w Living with mother-in-law 49.2 43.4 7.5 2,015 Other 34.5 49.5 16.0 6,691 Female-headed (husband not present) 3.1 12.9 84.0 136 FHH with no adult male 2.7 3.3 94.0 162 Unmarried FHH with adult male 3.3 16.9 79.9 112 Other 35.7 31.3 33.0 574 Total 36.9 45.0 18.2 10,543 1 2 9 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? sample. Including women in female-headed agency. A women's agency increases with households would undermine the power of the household wealth, her age, whether she has had results, since these women are the decision- a child or not (with a son counting for more than makers by virtue of their status, regardless of a daughter), and if she is working (with working their other characteristics (wealth, religion, and for cash having more of an impact). A slight kink the like). in the pattern is the lopsided U-shaped relation- Bivariate tabulations are shown for the former ship between high agency and asset quintile, group of women against various characteristics with women in the lowest quintile having slightly in table E.6. The relationship between the vari- greater agency than those in the second and ables shown is statistically significant in all cases. third. This result may reflect that poorer house- The main area where there is no difference is the holds are more likely to be nuclear, and women similar distribution of agency between Muslim have higher agency in these households. Multi- and Hindu women, although the small group of variate analysis is used below to separate out women from other religions do have higher these effects. Agency by Selected Women's Mobility: Descriptive Analysis Characteristics of Married An analysis of women's mobility using the same T A B L E E . 6 Women in Male-Headed methodology as above reveals that mobility and Households agency are related to the same factors, even though, as shown below, the two are only im- perfectly correlated. Agency (percent) Total Women's mobility is most restricted when it Number of comes to going shopping, indicating that pur- Low Medium High Percent observations dah and social norms continue to render the Asset quintile (poorest = 1) marketplace off-limits to women. Thus, a whole 1 41.9 44.7 13.4 100.0 1,980 68 percent of all women and three-fourths of 2 40.2 49.0 10.8 100.0 1,374 women who live with their mother-in-law never 3 40.3 47.2 12.5 100.0 1,552 go shopping. In comparison, only 4 to 5 percent 4 37.5 49.2 13.3 100.0 1,697 of women cannot leave their village, town, or city 5 29.0 51.3 19.8 100.0 2,103 of residence or go to the health center. Religion As is the case for intra-household agency, Muslim 38.3 47.3 14.4 100.0 7,553 women who live with their mother-in-law have Hindu 36.5 52.7 10.9 100.0 1,047 less mobility than other women. Fewer of these Other 21.4 61.1 17.5 100.0 106 women can leave their village, town, or city Number and sex of children alone, and more can only go if accompanied by their husband or someone else. The pattern is No children 53.2 42.1 4.7 100.0 840 Daughter(s) only 40.6 46.7 12.8 100.0 1,295 the same for going to the health center and At least one son 35.5 49.1 15.4 100.0 6,571 going shopping. An index of women's mobility (MOBILITY) Age group was constructed in the same way as the agency 11­19 53.8 40.3 5.9 100.0 1,318 index, using DHS 1999 data. As before, the re- 20­29 39.3 48.8 11.9 100.0 3,280 sponse to each of the three questions was re- 30­39 31.5 50.2 18.3 100.0 2,547 40­49 31.3 50.1 18.6 100.0 1,561 coded so that the respondent going out alone = 2, with husband or someone else = 1, and not at Work status all = 0. Second, the responses to the questions Not working 40.5 46.6 13.0 100.0 6,993 were summed together to give a 10-point scale Working (not for cash) 35.0 49.4 15.6 100.0 175 from 0 to 9. A score of zero indicates that the re- Working (for cash) 27.3 54.5 18.2 100.0 1,538 spondent cannot go out for any of the three pur- Total 37.9 48.1 14.0 8,706 poses, and 12 means she can go out alone for all 1 3 0 W O M E N ' S A G E N C Y , H O U S E H O L D S T R U C T U R E , A N D H E A L T H O U T C O M E S of them. Again, three categories of MOBILITY Women's Mobility were created: low for agency between 0 and 3, T A B L E E . 7 (percent) medium for 4 to 6, and high for values 7 to 10. Table E.8 illustrates women's levels of mobil- Can leave Can go to Go ity by their status in the household. Only one in village/town health center shopping eight married women who live in male-headed All respondents (n = 9,716) households have high mobility, almost half have medium mobility, and 40 percent have low mo- Alone 17.0 30.0 13.8 bility. As is the case for women's agency, women With children 49.5 24.2 6.4 With husband or someone else 29.2 40.6 11.7 who live with their mother-in-law have signifi- Cannot go 4.3 5.2 68.1 cantly lower mobility than others, with over half having low mobility and only 7 percent having Currently married women in male-headed households (n = 8,704) high mobility. Similarly, a much larger share of Alone 14.8 28.4 11.5 women whose husbands are away have high With children 50.1 23.4 6.1 mobility (over one-third). Unsurprisingly, un- With husband or someone else 30.7 43.0 12.2 Cannot go 4.3 5.2 70.2 married female heads of household with no adult male present are those with the highest Living with mother-in-law (n = 2,017) level of mobility, with half having high mobility. Alone 10.2 22.6 6.6 Analyzing women's mobility by selected char- With children 38.3 16.4 3.9 acteristics for married women in male-headed With husband or someone else 46.2 54.3 15.2 households (table E.9) reveals a pattern similar Cannot go 5.3 6.7 74.4 to that of women's agency. There is a U-shaped relationship between assets and mobility, with women in wealth quintiles 1 and 2 having asset quintile,6 while, as illustrated in table E.8, a greater mobility than the women in quintile 3, larger proportion of women who work have and women in the poorest quintile also having high mobility. This is consistent with the notion greater mobility than women in quintile 4. Such that women from better-off households can a relationship is consistent with the notion that more easily afford to adhere to purdah and the poorest women in Bangladesh often are hence to have more restricted mobility. those who adhere least to purdah because their The difference between Hindu and Muslim financial situation forces them to venture out- women is small. Just over 2 percent more from side to work: the largest percentage of women the former group have high mobility, whereas who undertake paid work belong to the poorest the percentage is 40 percent higher for the small T A B L E E . 8 Women's Mobility by Status within the Household Women's mobility (percent) Number of Woman's marital status Household type Low Medium High observations Male-headed households 41.5 46.2 12.3 8,706 Married o/w Living with mother-in-law 55.1 37.4 7.6 2,015 Other 37.4 48.8 13.7 6,691 Female-headed (husband not present) 14.0 54.7 31.3 136 FHH with no adult male 8.7 40.7 50.5 162 Unmarried FHH with adult male 5.7 55.3 39.1 112 Other 26.0 45.0 29.0 574 Total 40.0 45.7 14.3 10,543 1 3 1 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? perfectly, correlated. Table E.10 shows the cross- Mobility by Selected tabulation of the two. If there were no relation, Characteristics T A B L E E . 9 then the entry in each cell would be 1/3 (33 per- for Married Women in cent), which is far from the case. Applying a chi- Male-Headed Households squared test shows the two to be highly related. This finding is somewhat contrary to that of Mobility (percent) Total some anthropologists, who argue that women's Number of physical mobility does not necessarily reflect Low Medium High observations their intra-household agency, and that activities Asset quintile (poorest = 1) outside the homestead such as paid employ- 1 42.6 45.7 11.6 1,980 ment usually serve to further the household 2 42.1 47.7 10.1 1,374 interest (White 1992, p. 79). The imperfect cor- 3 46.1 45.2 8.7 1,552 relation, however, could indicate that this is true 4 43.0 45.6 11.4 1,697 for some women but not all. 5 33.8 46.8 19.4 2,103 Religion Changes over Time Muslim 42.1 46.5 11.5 7,553 Women's overall mobility has increased signifi- Hindu 40.1 46.0 13.9 1,047 cantly over just a few years (see tables E.11a and Other 16.9 28.3 54.8 106 E.11b). Almost a quarter of women could not Number and sex of children leave their village, town, or city at all in 1993, but only 4 percent could not do so in 1999. Similarly, No children 73.5 19.7 6.8 840 women who could not go to the health center Daughter(s) only 48.6 39.0 12.4 1,295 At least one son 36.1 50.9 13.0 2,135 decreased from over a fifth to less than 6 per- cent. At the same time, however, we see a very Age group large increase in the percentage of women who 11­19 69.4 26.5 4.1 1,318 can go with "others" and a decrease in those 20­29 42.5 45.7 11.8 3,280 who can go alone--a change that may be par- 30­39 32.8 51.5 15.7 2,547 tially explained by a slight difference in the way 40­49 29.2 55.7 15.1 1,561 these questions were asked in 1993 and in the Work status two later surveys. Finally, the proportion of Not working 44.1 46.2 9.7 6,993 women who never go shopping has decreased Working (not for cash) 27.6 57.3 15.2 175 from over 80 percent to under two-thirds. This is Working (for cash) 32.5 44.6 22.9 1,538 concurrent with an increase in the proportion of Total 40.0 45.7 14.3 10,543 women who go shopping once a month or more, which increased from less than one in 20 in 1993 to almost one in 5 in 1999. Thus, while sample of women of other religions. Whether going shopping is the area where women's mo- a woman has children--and whether at least bility continues to be most severely restricted, one child is a son--has a significant impact on there has been quite a significant change over a women's mobility, to an even greater extent period of only six years, indicating a change of than it impacts women's agency. Hence, low mo- social norms and the practice of purdah. This bility is characteristic of nearly three-quarters of does not mean that purdah is disappearing, but, women with no children, just under half of as suggested by Kabeer (2002), it is adapting to women with only daughters, and just over a changing socioeconomic conditions. third of women who have at least one son. Multivariate Analysis Agency and Mobility Regression analysis was performed for the Although agency and mobility are associated agency and mobility indices and the health deci- with the same factors, they are highly, but not sion variables. The indices are discrete variables 1 3 2 W O M E N ' S A G E N C Y , H O U S E H O L D S T R U C T U R E , A N D H E A L T H O U T C O M E S in the ranges 0-12 and 0-9, so might appear to be Women's Agency and able to be treated as continuous variables using T A B L E E . 1 0 Mobility (as percent of ordinary least squares. However, a tabulation agency category) shows both series to be right censored (obser- vations are bunched at the upper limit), so that Agency a probit model is appropriate. The recoded Low Medium High Total agency and mobility variables related to health can be used as dependent variables in an or- Low 48.6 51.3 22.7 45.5 dered probit model. The sample in all cases is Mobility Medium 46.0 30.6 42.4 36.2 High 5.4 18.1 34.9 18.3 married women in male-headed households. By and large, the results support the patterns shown in bivariate tabulations. The following re- Changes in Mobility, sults are particularly striking: T A B L E E . 1 1 a 1993­99 (percent) · Both age and age squared are significant, in- dicating that both agency and mobility in- Can go outside Can go to village/town/city health center crease with age, but then decrease after a certain threshold.7 The coefficients show this 1993 1996 1999 1993 1996 1999 threshold to be in the early forties (see table No 24.9 15.6 4.3 22.0 13.1 5.6 A.3). This variable may also pick up a "time ef- With other 4.0 7.8 30.9 5.6 40.4 40.0 fect," where older households afford women With children 34.0 49.0 47.2 38.1 25.8 23.3 less agency and mobility. Having a child also Alone 37.2 27.6 17.6 34.3 20.7 31.1 increases agency, with the effect being Number of stronger if the child is male. observations 7,308 7,308 10,543 9,639 9,127 10,543 · Both agency and mobility increase with edu- cational status (the reference category is no education), though the effect of primary is Changes in Mobility, T A B L E E . 1 1 b not always significant. Given that there is a 1993­99 (percent) strong correlation between education and the level of the asset index (table A.2), it is Frequency of shopping striking that both sets of variables are fre- 1993 1996 1999 quently significant, showing that the impact Never 81.9 78.8 64.4 of education can operate independently of Once/yr or less 6.7 6.2 5.2 that through wealth. Several times/yr 7.2 8.8 11.6 · The three media variables, especially TV, have Once/month or more 4.1 6.2 18.9 a strong effect, though reading a paper is not Number of always significant.8 observations 9,639 9,127 10,543 · Daughters-in-law have a strongly significant lower level of agency and mobility. Mobility is higher for mothers-in-law, but not agency. have more agency and mobility. The Hindu · The inverted U with respect to wealth is evi- dummy is negative but not significant. dent for agency but not mobility. · Working has a strong positive effect, as does In summary, the policy implications are that: being a member of a credit organization. · Regional effects are high, with women in · It is true that women often do not have a Sylhet and Barisal faring worst.9 Rural women say--so health and nutrition messages need also have lower mobility and agency. to reach a wider audience. · The small number of women from other re- · But agency and mobility can change over ligions (the reference category is Muslim) time, and are doing so quite rapidly in Bang- 1 3 3 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? ladesh. Policy relevant factors that can affect from the three DHS surveys show that 99.4 per- this rate of change are communications cent of women over 30 have been married at (IEC), and education. some point (that is, are currently or formerly married). Figure E.2(a) shows the marital status Age at Marriage in Bangladesh for women in different age groups.10 This graph The age at marriage is a well-known determinant shows that, over 30, only a tiny proportion of of maternal and child health outcomes. Children women have never been married. It also shows born to young mothers are more likely to die, so that the vast majority of women are married by that raising the age of marriage, especially in a their early 20s, and that virtually all are married country such as Bangladesh, in which births out- by the time they reach 25. As might be expected, side of marriage are extremely rare, will reduce older women are more likely to be formerly mar- infant mortality. Traditionally, rising age at mar- ried, since their husbands predecease them. riage has been a driver of fertility reduction, Husbands dying before their wives is not only a though there are reasons to think that this effect result of the shorter life expectancy of males, but may be muted in Bangladesh, where age at mar- also of the large age differential that is common riage is indeed low, but fertility has already been in Bangladesh. Figure E.2(b) shows that men substantially reduced (Annex F). Data from the marry later than women, with the majority mar- Demographic and Health Survey show surpris- rying by the time they are 30, rather than by their ingly large changes in the age at marriage over a early 20s, as is the case for women. Most men very short period. This paper analyzes the de- under 20 are not married, whereas most women terminants of age at marriage, and thus identifies are. factors behind these changes. However, the age of marriage is changing. Despite the short period covered by the DHS Marriage in Bangladesh surveys, which might be expected to be too Virtually all women in Bangladesh get married. short to pick up significant social trends, women Data on over 110,000 women over the age of 10 in Bangladesh are getting married later. As F I G U R E E . 2 Marital Status by Age (a) Women (b) Men 9,000 8,000 Formerly married Formerly married 8,000 Currently married 7,000 Currently married Never married Never married 7,000 6,000 6,000 5,000 men women 5,000 of of 4,000 4,000 Number 3,000 Number 3,000 2,000 2,000 1,000 1,000 0 0 10­15 16­20 21­25 26­30 31­35 36­40 41­50 51­60 61­70 >70 10­15 16­20 21­25 26­30 31­35 36­40 41­50 51­60 61­70 >70 Age range Age range 1 3 4 W O M E N ' S A G E N C Y , H O U S E H O L D S T R U C T U R E , A N D H E A L T H O U T C O M E S shown in table E.12, the proportion of women married before they are 22 fell from 86 percent T A B L E E . 1 2 Median Age at Marriage in the 1993 survey to 80 percent in 1999. The change is not among the youngest group--just 20­24 25­29 30­34 35­39 40­44 45­49 20­49 under 5 percent of 10-15 year olds are married in all three surveys-- but among those aged 16-21. 1993/94 15.3 14.8 14.2 13.9 13.6 13.6 14.1 1999/2000 16.1 15.4 14.9 14.5 14.0 13.8 14.7 The assertion that the age at marriage is rising is contrary to the DHS reports, which show only a slight rise in the median age at marriage for Percentage of Women women aged 20-49, from 14.1 to 14.7 years. T A B L E E . 1 3 Ever Married However, this calculation is based on the age of marriage for women who have married. If a 1993 1996 1999 larger proportion of women are marrying later, 10­15 4.9 4.7 4.9 then they simply drop out of this calculation, 16­18 30.7 30.4 28.1 creating a downward bias in the sample estimate 19­21 50.6 48.9 47.1 of the age of marriage. The presentation in table Total married before 21 86.3 84.0 80.1 E.12 both avoids this problem, and shows that Source: Calculated from DHS 1993 and 1999. there is indeed a problem in basing the calcula- tion simply on the age of marriage of those who Age at Marriage by have married. Alternatively, the median age at T A B L E E . 1 4 Selected Characteristics marriage can be calculated using the whole (percent) sample of women, rather than the ever-married sample. Age range A first look at why the age at marriage may be Number of 16­18 19­21 22­25 >26 observations falling comes from bivariate tabulations of age at marriage and various individual and household Education status characteristics. There is a problem in using this None 76.7 16.6 4.9 1.8 1,033 Primary 76.9 17.4 4.1 1.6 827 approach, as DHS collects detailed data only Secondary 69.4 21.2 7.9 1.5 1,228 from ever-married women. Hence, the younger Higher 31.9 39.3 21.8 7.0 514 women in the sample are not representative, as Asset quintile they are married, and so, for example, are less 1 78.3 14.9 5.5 1.3 457 likely to be at school than unmarried girls in the 2 73.3 19.1 5.7 1.9 419 same age range. For this reason, girls under 15 3 74.3 18.0 5.7 2.1 529 are excluded from the tables. This problem can 4 70.3 22.1 6.1 1.5 787 be tackled more satisfactorily in a multivariate 5 59.1 25.5 11.8 3.5 1,410 setting, as seen below. Region Looking at age at marriage by educational at- Barisal 69.3 21.0 6.8 2.9 309 tainment (DHS 1999), we see that women's level Chittagong 68.9 21.5 7.9 1.7 836 Dhaka 67.1 23.9 7.2 1.8 930 of education affects their age at marriage posi- Khulna 71.6 21.3 5.2 1.9 536 tively. As illustrated in table E.14, women of higher Rajashahi 69.8 18.1 9.7 2.4 507 educational status marry later, especially those Sylhet 60.3 21.5 13.0 5.2 484 with higher education. But notably more women Total 2,444 777 294 87 with secondary education get married in their 20s, compared with women with less education. The regional pattern is not very strong, except that censoring problem. But the DHS does not have women in Sylhet appear to get married older. much data on unmarried women. An alternative Multivariate analysis may be done in two ways. is to use Heckman, first estimating a selection First is a Cox hazards model (where the hazard is equation for if a woman is married or not, and getting married). This model takes care of the then at what age those who are married got mar- 1 3 5 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? ried, using a sample selection bias term from The results (table E.15) show the expected re- the first-stage equation. The selection equation lationship with education. There is a U-shape should contain at least one variable not in the age with assets: poorer and richer women tend to at marriage equation. In this case identification is get married later, presumably as the poor need easy: age is the overwhelming determinant of time to save for a dowry (and are considered less whether a woman is married or not, and that is all desirable brides and so a "last choice") and the we need to enter in the selection equation. rich get married later for social reasons. Regression Results for Age T A B L E E . 1 5 at Marriage Heckman model Cox Hazards model Coefficient z statistic Hazard ratio z statistic No education ­5.46 ­41.53*** 5.06 33.40*** Primary ­4.93 ­37.68*** 3.86 27.71*** Secondary ­3.51 ­27.56*** 2.01 14.63*** Quintile 2 ­0.15 ­1.71* 1.08 2.15** Quintile 3 ­0.17 ­1.86* 1.07 1.90* Quintile 4 ­0.07 ­0.75 1.05 1.43 Quintile 5 0.02 0.17 1.01 0.36 Intercept 19.96 138.64*** Selection equation t statistic Age 0.21 68.22*** Intercept ­3.84 ­65.76*** /athrho ­0.52 ­16.26*** /lnsigma 1.02 141.33*** Rho ­0.48 Sigma 2.76 Lambda ­1.33 No. of observations 15,773 15,573 Of which censored (Heckman) 5,230 Of which failures (Cox) 10,543 Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. 1 3 6 W O M E N ' S A G E N C Y , H O U S E H O L D S T R U C T U R E , A N D H E A L T H O U T C O M E S T A B L E E . 1 6 Women's Agency and Mobility: Survey-Ordered Probit Regression Own Coefficient t Child Coefficient t Cangohealth Coefficient t age 0.048 4.38*** age 0.035 2.98*** age 0.072 6.7*** age_sq ­0.001 ­3.43*** age_sq 0.000 ­2.35** age_sq ­0.001 ­5.26*** primary ­0.013 ­0.36 primary 0.045 1.26 primary 0.033 0.99 secondary 0.010 0.22 secondary 0.068 1.61 secondary 0.151 3.51*** tertiary 0.226 3.18*** tertiary 0.224 3.39*** tertiary 0.359 4.64*** child 0.058 0.95 child 0.065 1.22 child 0.299 5.54*** child_son ­0.015 ­0.38 child_son 0.091 2.6*** child_son 0.077 2.08** listen_radio ­0.026 ­1.7* listen_radio ­0.007 ­0.44 listen_radio 0.021 1.39 watch_tv 0.023 1.44 watch_tv 0.066 4.28*** watch_tv 0.066 4.17*** read_paper 0.015 0.87 read_paper 0.038 2.24** read_paper 0.058 3.1*** wealth2 ­0.049 ­1.14 wealth2 ­0.015 ­0.37 wealth2 ­0.025 ­0.55 wealth3 ­0.046 ­1.01 wealth3 ­0.029 ­0.64 wealth3 ­0.103 ­2.21** wealth4 0.014 0.31 wealth4 0.080 1.8* wealth4 ­0.055 ­1.16 wealth5 0.118 2.1** wealth5 0.176 3.37*** wealth5 ­0.080 ­1.5 DIL ­0.100 ­2.74*** DIL ­0.145 ­4.22*** DIL ­0.135 ­4.39*** MIL 0.021 0.33 MIL 0.016 0.23 MIL 0.003 0.05 belong credit ­0.059 ­1.47 belong credit 0.009 0.23 belong credit 0.229 5.56*** other org ­0.014 ­0.33 other org ­0.022 ­0.6 other org 0.105 2.5*** work 0.175 4.98*** work 0.182 5.04*** work 0.191 4.98*** rural ­0.105 ­2.21** rural ­0.098 ­2.64*** rural ­0.157 ­3.77*** Hindu 0.011 0.22 Hindu ­0.021 ­0.52 Hindu ­0.076 ­1.22 other_rel 0.229 2.49** other_rel 0.269 2.4** other_rel 0.422 2.35** Chittagong 0.117 2.04** Barisal 0.082 1.44 Sylhet ­0.134 ­1.18 Sylhet ­0.141 ­1.59 Chittagong 0.075 1.43 Khulna 0.075 1.27 Rajashahi 0.082 1.72* Sylhet ­0.154 ­1.99* /cut1 ­0.106 ­0.7 /cut1 0.767 4.68*** /cut1 0.445 2.6*** /cut2 1.572 10.08*** /cut2 1.902 11.68*** /cut2 1.923 11.32*** /cut3 2.227 14.09*** Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. 1 3 7 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? T A B L E E . 1 7 Women's Agency and Mobility: Tobit Estimates Agency Coefficient t Mobility Coefficient t age 0.206 8.48*** age 0.172 10.5*** age_sq ­0.003 ­6.74*** age_sq ­0.002 ­8.11*** primary 0.129 1.72* primary ­0.016 ­0.31 secondary 0.184 1.88* secondary 0.259 3.92*** tertiary 0.680 4.21*** tertiary 0.968 8.84*** child 0.433 3.4*** child 0.437 5.09*** child_son 0.146 1.65* child_son 0.131 2.2** listen_radio ­0.019 ­0.57 listen_radio 0.047 2.06** watch_tv 0.174 5.22*** watch_tv 0.183 8.09*** read_paper 0.071 1.78* read_paper 0.138 5.16*** wealth2 ­0.080 ­0.82 wealth2 ­0.042 ­0.63 wealth3 ­0.089 ­0.92 wealth3 ­0.221 ­3.39*** wealth4 ­0.037 ­0.37 wealth4 ­0.168 ­2.53** wealth5 0.239 2.13** wealth5 ­0.158 ­2.08** DIL ­0.755 ­10.08*** DIL ­0.397 ­7.86*** MIL ­0.049 ­0.35 MIL ­0.011 ­0.12 belongcredit 0.087 1 belongcredit 0.261 4.45*** otherorg 0.112 1.38 otherorg 0.348 6.3*** work 0.619 8.08*** work 0.700 13.46*** rural ­0.392 ­5.35*** rural ­0.643 ­12.94*** Hindu ­0.058 ­0.63 Hindu ­0.061 ­0.98 other_rel 0.646 2.38** other_rel 1.773 9.45*** Sylhet ­0.529 ­5.35*** Chittagong ­0.038 ­0.65 _cons 1.149 3.29*** Sylhet ­0.124 ­1.71* Khulna 0.465*** _se 2.728 (Ancillary parameter) Rajashahi ­0.082 _cons 0.687** se _1.838533 (Ancillary parameter) Note: P<0.1 ** P<0.05 *** p<0.01, DIL= daughter-in-law, MIL= mother-in-law Agency: Mobility: Number of observations = 8,706 Number of observations = 8,704 LR chi2(23) = 992.60 LR chi2(26) = 2,040.09 Prob > chi2 = 0.0000 Prob > chi2 = 0.0000 Log likelihood = ­20,858.077 Pseudo R2 = 0.0232 Log likelihood = ­17,428.529 Pseudo R2 = 0.0553 1 3 8 ANNEX F: FERTILITY This annex presents analysis on fertility trends Cohort fertility--the average number of children and determinants in the 1990s and strategies for born to women of a particular age--would re- achieving continued fertility decline to acceler- main unchanged. But the total fertility rate is cal- ate achievement of a stable population. The first culated using the synthetic cohort method, issue addressed is that of fertility trends in the which is an average of the age-specific fertility 1990s. This issue is of policy importance, since rates of women currently alive. Hence, TFR will there is a widespread perception that fertility de- temporarily fall as a result of postponed births cline leveled off in the 1990s, suggesting that the (because of the reduction in the age-specific fer- hitherto successful family planning program had tility rate among the women postponing births), run out of steam. But it is shown here that fertil- only to return to its original level when these ity decline continued throughout the 1990s, al- women give birth in the next period (see beit at a slower rate than previously. box F.1). It is plausible that such a tempo effect The regression model results feed directly into in Bangladesh accelerated fertility decline in the strategy formulation for maintaining fertility de- 1980s, but served to counteract a continuing un- cline. First, the results clearly indicate that son derlying decline in the early 1990s. If this argu- preference is a barrier to reducing fertility. Ex- ment is correct, then the fact that TFR stagnated perience in Bangladesh and elsewhere shows that rather than rose in the 1990s indeed implies that son preference exerts itself more strongly as fer- there were other changes taking place to exert a tility falls, so that policies to address the problem, downward pressure on the TFR. while not obvious, are required. The regression However, while there is evidence of such a results show that, contrary to normal international tempo effect, there is no need to rely on that to patterns, age at marriage has limited potential to explain the claimed fertility plateau for a number reduce fertility, and why this is so is discussed. of reasons. First, it has not been usual to rely on While desirable for other reasons, strategies to direct estimates of fertility to estimate fertility in raise the age at marriage will have, at best, limited the past, so it is not clear why they should be the impact on fertility. It is proposed instead that at- reference point for the 1990s. Second, even if tention should focus on high-fertility households. the direct estimates are used, there are reasons to believe that fertility was underestimated in What Has Been Happening to Fertility? 1993/94 on account of displacement in birth re- Following the rapid decline during the 1980s, porting. But, third, if indirect estimates are used, DHS data suggest that fertility leveled off in the they show both that the direct estimates under- 1990s, declining slightly from 3.4 to 3.3 between estimate fertility (particularly in 1993/94, rein- 1993 and 1996, and then remaining at that level forcing the previous point) and that fertility did until 1999. continue to decline in the 1990s. These argu- It has been suggested that this slackening ments are considered in turn. pace of fertility reduction comes from a tempo effect. Suppose that during the 1980s, women Direct Versus Indirect Estimates began postponing births, while maintaining the Table F.1 summarizes direct fertility estimates number of children they had at the same level. from various surveys, together with indirect es- 1 3 9 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? B O X F . 1 The Tempo Effect from Birth Spacing Number of Births per 100,000 Women The table shows the number of births per 100,000 women by five-year age cohorts. These numbers are Period 1 Period 2 Period 3 Period 4 used to calculate the age-specific fertility rates 15­19 40,000 40,000 20,000 20,000 shown in the bottom part of the table. 20­24 50,000 50,000 50,000 70,000 The fertility rate is the sum of these age-specific rates 25­29 20,000 20,000 20,000 20,000 (multiplied by five, as each rate covers a five-year 30­34 8,000 8,000 8,000 8,000 period). This is different from the cohort fertility rate, 35­39 5,000 5,000 5,000 5,000 which is the sum of the age-specific rates for an age 40­44 1,200 1,200 1,200 1,200 cohort across its reproductive years. For example, 45­49 400 400 400 400 Cohort A is aged 15­19 in period 1, 20­24 in period 2, and so on. Age-Specific Fertility Rate In the example shown, the cohort aged 15­19 in period 1 2 3 4 3 decides to postpone 2,000 births from age 15­19 until they are aged 20­24. There is no change in the number 15­19 0.40 0.40 0.20 0.20 of children they will have over their life span, so the 20­24 0.50 0.50 0.50 0.70 cohort fertility rate is unchanged. 25­29 0.20 0.20 0.20 0.20 But the fertility rate is not measured in this way (as it 30­34 0.08 0.08 0.08 0.08 could only be observed for women who have completed 35­39 0.05 0.05 0.05 0.05 their child bearing years). The total fertility rate tem- 40­44 0.01 0.01 0.01 0.01 porarily drops in period 3, as the age-specific fertility 45­49 0.00 0.00 0.00 0.00 rate for 15­19s has declined. But in the next period the TFR 6.23 6.23 5.23 6.23 rate for 20­24s rises, as the postponed births take place. timates made using data from these surveys. Direct and Indirect Two points can be noted. First, the direct esti- T A B L E F . 1 Fertility Estimates from mates are systematically lower than the indirect Various Sources ones. Second, the trend in fertility decline is far less from the direct estimates than from the in- TFR estimates direct ones. For example, both CPS 1983 and Year Survey Direct Indirect* 1989 yielded direct TFR estimates of 4.9. But 1974 BRSFM 4.8 7.3 these results have not been used to argue that 1975 BFS 5.4 7.4 there was a fertility plateau in the 1980s. To the 1983 CPS 4.9 7.0 contrary, this is seen as the period of most rapid 1985 CPS 4.6 6.5 decline. 1989 CPS 4.9 5.9 1989 BFS 4.6 5.4 1991 CPS 4.2 n.a. Underestimation of TFR in DHS 1993/94 1993/94 BDHS 3.4 4.3 The argument that the direct estimate of TFR in 1996/97 BDHS 3.3 4.0 the 1993/94 DHS was an underestimate is based 1999/2000 BDHS 3.3 3.7 on an apparent displacement of births, with Note: 1974­89 from Cleland and others (1994) and 1993­99 from OED analysis. births that were reported as taking place more than three years ago, actually being more recent. The evidence for this phenomenon comes from comparing the number of births reported in the 1 4 0 F E R T I L I T Y three years preceding the survey with the num- Number of Births in Selected ber reported for the three years before that (that T A B L E F . 2 Periods Before DHS is, 4-6 years before the survey). Since they cover similar periods, the ratio of these two amounts Three-year intervals Five-year intervals will be one, except that declining fertility or postponed births raise it above one. Births in X years Births in X years preceding the survey preceding the survey Table F.2 shows this analysis. The ratio for the 1993/94 survey is greatly in excess of that for the X = 3 X = 4­6 X = 5 X = 6­10 other surveys. In that survey, 26 percent more years years Ratio years years Ratio births were reported for 4-6 years before the sur- 1993/94 3,800 4,795 1.26 6,947 8,226 1.18 vey, compared with those reported for within 1996/97 3,655 3,799 1.04 6,187 7,102 1.15 3 years before the survey. This ratio compares 1999/2000 4,167 3,940 0.95 6,852 6,864 1.00 with just 4 percent for 1996/97 and a 5 percent re- duction in 1999/2000. Yet comparing the ratio for five-year periods there is little difference between Indirect Fertility Measures the 1993/94 and 1996/96 surveys. These data thus Appendix F.1 presents indirect estimates of fer- show that in 1993/94 children under three were tility using three different approaches, using BFS being reported as being older. The effect of this 1989 and the three DHS. One set of estimates birth displacement is exactly the same postpone- are also made using 1991 and 2001 census data. ment of births shown in box F.1. That is, it will re- Figure F.1 summarizes the results. duce reported fertility. But since, in this case, it is The data shown here confirm that direct believed that births are being incorrectly attrib- methods overestimate fertility. This gap is uted to a later period, this finding represents a largest for 1993/94, which supports the argu- downward bias of the direct fertility estimate. ment that birth displacement exacerbated un- F I G U R E F . 1 Indirect Estimates of Total Fertility Rate 8 7 Rele: census data rate 6 Rele: survey data Proximate determinant fertility method 5 Total Regression method 4 3 1980 1985 1990 1995 2000 1 4 1 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? derestimation in that year. The results also show inroads into fertility. That fertility was being re- that fertility continued to decline during the duced points to the continuing success of these 1990s, albeit at a slower rate than had been the programs. Nonetheless, it is the case that the rate case from 1989-93. of fertility reduction was slowing down, suggest- ing that thought does need be given to strategy Summary for attaining the goal of a stable population. Three arguments have been presented to counter the commonly held view that there was a fertility Modeling Fertility Determinants plateau in Bangladesh in the 1990s. This finding Table F.3 presents the results of regression mod- matters, since the perceived plateau gave weight els of two dependent variables that proxy for fer- to the view that family planning services were be- tility: (1) whether a woman desires to have more coming moribund and could not make additional children and (2) whether a woman has had a T A B L E F . 3 Logistic Regression Results of Fertility Proxies Dependent variable Birth in the Desire more children Desire more children last three years Current age > = 25 years ­0.84*** ­0.84*** ­1.81*** Age at first marriage > = 15 years 0.25*** 0.25*** 0.53*** Division (Ref Sylhet) Barisal ­0.72*** ­0.72*** ­0.34*** Chittagong ­0.09 ­0.09 ­0.1* Dhaka ­0.64*** ­0.65*** ­0.33*** Khulna ­1.00*** ­1.00*** ­0.54*** Rajshahi ­0.912*** ­0.92*** ­0.47*** Rural (ref urban) 0.28*** 0.31*** 0.01 Education (ref No education) 1­5 0.08 0.06 ­0.05 6­9 ­0.00 ­0.05 0.09** 10+ ­0.13 ­0.19** 0.22*** Non-Muslim ­0.35*** ­0.356*** ­0.23*** Number of Living Children (Ref None) 1 ­1.62*** ­1.62*** 2 ­3.71*** ­3.7*** 3+ ­4.82*** ­4.81*** All living children are females 0.46*** A living son ­0.26*** Number of children* all living children female ­0.19*** Male child died* all living children female 0.04 Respondent currently working ­0.18*** ­0.19*** ­0.21*** Watch television every week ­0.14** ­0.12*** Listen radio every week ­0.06 ­0.08*** Access to electricity 0.02 ­0.043 ­0.23*** Survey (Ref BDHS 93) BDHS 96 ­0.02 ­0.03 ­0.05 BDHS 99 ­0.05 ­0.06 0.03 Constant 4.07 4.03 ­0.1 ­ 2 Log likelihood 16,763 16,785 Model chi­square 11,158 11,171 Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. 1 4 2 F E R T I L I T Y birth in the last three years. Some of the ex- Fourth, access to electricity has an unex- pected patterns emerge here: all regions have pected, though insignificant, positive impact lower fertility than Sylhet, non-Muslims have on the desire to have more children. However, lower fertility, women with more children are the coefficient becomes negative (though still less likely to desire another child, and older insignificant) once the media variables are re- women are less likely to have had a child in the moved. This result shows that the effect elec- last three years or to desire another. Working tricity has on lowering fertility is by facilitating women have lower fertility, which may reflect access to media, notably TV, which has a signifi- the effect of women's agency. Some of the other cant fertility-reducing effect when included. This results require more comment. variable is thus a measure of the impact of IEC First, the age at marriage is significantly po- through one specific channel. sitively related to the two fertility measures, a Finally, it might be expected that the survey result that is robust to different model specifica- dummy coefficients would be negative, showing tions. This finding is contrary to the conven- an "autonomous" downward trend in fertility by tional wisdom that increasing age at marriage is factors not included in the survey. This pattern a driver of lower fertility, as indeed has been the is present, though not significant, for the "desire case in many countries (including, quite prob- more children regression." But the result does ably, Bangladesh in the past). However, as dis- not appear for the child in last three years result. cussed in more detail below, since fertility is Rather than evidence of such a downward trend, already low, there is limited possibility for in- this result reinforces the argument of birth dis- creasing age at marriage to reduce fertility--the placement in the 1993/94 survey. effect is rather to postpone births (which will, as explained above, create a tempo effect that tem- Characteristics of High-Fertility Women porarily accelerates the rate of fertility decline). Although fertility has fallen, there remain a sig- Hence, women who have married later are more nificant number of women bearing high num- likely to want more children and to have had a bers of children. The following groups of women child in the last three years. were defined as having "high fertility": (1) those Second, the level of education has the ex- aged 35 or older with five or more children; pected effect on the desire for more children-- (2) those aged 35 or older with four or more chil- more educated women are less likely to want dren; (3) those aged 25 or older with four or another child. But there is the opposite rela- more children; and (4) those aged 25 or older tionship for women who have had a child in the with four or more children, and who want more last three years--those women having reached children. at least grade 6 are more likely to have had a Age 35 was chosen as the cutoff age after child in the last three years, and those at grade which very few (around 3 percent) of the births 10 and above even more likely to have done so. take place. The problem with this cutoff is that it This result once again reflects the marriage post- cannot capture fertility behavior of women cur- ponement effect--women attending school rently in their reproductive years. Hence, the after the onset of their reproductive years are lower cutoff of 25 was also used. postponing births rather than having fewer than The percentage of women in the high-fertility the (already low) average. category using the various definitions is shown Third, there is considerable evidence of son in table F.4. Except under the fourth definition, preference. This preference is picked up by the a significant minority of the population is identi- four interactive variables used in the final re- fied as having "high fertility." The fact that many gression. A woman is more likely to have had a women with four children do not want more child in the last four years if all her children are children, but many will go on to have them, rep- girls (although this effect is mitigated somewhat resents either unmet demand for contraception, if she has a lot of them), and less likely to if she incorrect usage, contraceptive failure, or that it already has a son. is not solely the woman's decision. By all defini- 1 4 3 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Women in High Fertility highest share in the middle quintiles). The pro- T A B L E F . 4 Categories by Year and portion of high-fertility women is highest in Definition (percent) Sylhet and lowest in Khulna. There is a more marked pattern for education. Close to half of Year 1 2 3 4 women with no education have a high level of 1999 17.2 22.4 34.1 1.8 fertility, compared with just 6 percent of those 1996 18.2 22.3 37.3 2.0 with higher education. 1993 23.3 29.0 39.2 2.4 Note: See text for definitions. Strategies for Fertility Reduction The evidence is clear that Bangladesh has tions there is a clear drop in women in the high- achieved substantial fertility reduction, in excess fertility category between 1993 and 1999. of what is expected from a country at its level of Figure F.2 shows the relationship between socioeconomic development by virtue of its ef- high fertility and selected characteristics from fort in family planning (see Annex B). Although the 1999 DHS data. Although there are variations fertility reduction has slowed, it is of course nec- by region, religion, and wealth, these are not essary to continue to disseminate family plan- that marked, in that there is a substantial pro- ning messages and support contraceptive supply portion of high-fertility women in each group. in order to preserve gains to date and seek addi- There is a slight deterioration in the share by tional progress. Given the slowdown in fertility wealth quintile (though by other definitions and reduction, which is not unexpected, what poli- in other years there is an inverted U, with the cies are most likely to prove successful for mov- F I G U R E F . 2 Correlates of High Fertility 45 50 40 45 40 women 35 women 35 30 30 25 25 20 20 high-fertility 15 high-ferility 15 10 10 5 Percent Percent 5 0 0 1 2 3 4 5 None Primary Secondary Higher Asset quintiles 45 40 40 35 women 35 women 30 30 25 25 20 20 15 high-fertility 15 high-fertility 10 10 5 5 Percent Percent 0 0 Barisal Chittagon Dhaka Khulna Rajashahi Sylhet Muslim Hindu Other 1 4 4 F E R T I L I T Y ing toward the goal of bringing fertility down to Bangladesh Is an Exception the replacement level? It is first argued that rais- to the Normal Relationship ing the age at marriage, while it has other bene- F I G U R E F . 3 Between Age at Marriage fits, is unlikely to make much of an impact on and Fertility fertility. Rather, attention should be paid to the three interrelated areas of high-fertility families, 8 son preference, and continuing reductions in under-five mortality. 7 6 Why Raising the Age at Marriage Will Have a Senegal Limited Impact on Fertility in Bangladesh rate 5 There is a well-established inverse relationship 4 Egypt between the age of marriage and fertility, with an fertility 1992/93 Peru 1999/00 increasing age at marriage often being a driver of 3 1996/97 Bangladesh fertility reduction. DHS data from around the Total 2 world, shown in figure F.3, bear this out. There is a clear negative relationship: the fitted regres- 1 sion line suggests that increasing the age of mar- 0 riage by one year reduces fertility by one-third of 12 14 16 18 20 22 24 26 a child.1 However, Bangladesh is a clear outlier Median age at first marriage in this figure, with the three DHS surveys for that country laying in the lower left quadrant. Bang- Source: ORC Macro 2004. MEASURE DHS STATcompiler. http://www.measuredhs.com, November 2 2004. ladesh has the lowest age at marriage of all coun- tries shown, but well below-average fertility, with an age at first marriage typical for a country in shows no such shift. Women are getting married which the median age at marriage is over 20, later, but not having their first child at an older rather than 14-15. This evidence both supports age than before. Thus, it should be the case that the view that Bangladesh has already, through the lag between getting married and having the other means (that is, family planning programs), first child has shortened, which figure F.4(c) reaped the gains to be had from reducing the shows to be the case: in 1999, 17 percent of age at marriage, and suggests that there may be women had their first child within a year of mar- rather limited fertility gains from increasing the riage, compared to just 9 percent doing so in age at marriage. This latter point is further sup- 1993. ported by the pattern displayed by data, with the Figure F.4(d), which shows the scatter plot of data points being more or less horizontal. This age at first birth against age at marriage, digs a bit pattern is in contrast to other countries, for ex- deeper into what is going on here. The plot is ample Egypt, Peru, and Senegal, for which the fitted with a locally weighted regression line, line joining the data points has the expected which shows there is no relationship between downward slope. age of marriage and age at first birth (that is, the Figures F.4(a)-(c) show the age at marriage regression line is flat) up to about 13 years of and at first birth, and the lag between them. age.3 Close to 20 percent of ever-married women During a short period in the 1990s there was a in the 1990s had married by 12, and 40 percent surprisingly large change in the age at marriage by 13. But, as the graph shows, increasing the (see also Annex E). For example, in 1993, 26 per- age at marriage within this age range will have no cent of the women interviewed had been mar- effect at all on the age at first birth. Girls getting ried by age 12, compared with just 14 percent in married below age 13 are generally considered 1999.2 These changes are shown by a shift of the to be below reproductive age and only excep- cumulative distribution in figure F.4(a) to the tionally bear children (only 4 percent of births right. But the age at first birth, in figure F.4(b), are to girls aged 13 and younger). Second, the 1 4 5 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? F I G U R E F . 4 Changes to Age at First Marriage and First Birth (a) Age at first marriage, 1993 and 1999 (b) Age at first birth, 1993 and 1999 100 100 1993 90 90 1999 80 80 women women of 70 of 70 60 60 1993 50 50 percent 1999 percent 40 40 30 30 20 20 Cumulative 10 Cumulative 10 0 0 5 10 15 20 25 30 10 12 14 16 18 20 22 24 26 28 30 Age at first marriage Age at first birth (c) Lag from marriage to first birth, 1993 and 1999 (d) Relationship, age at marriage and first birth 100 Lowess smoother 90 40 80 birth women 1st of 70 at 60 30 1999 50 1993 percent 40 20 30 respondent 20 of Cumulative 10 Age bandwidth = .8 10 0 <1 1 2 3 4 5 6 7 8 9 10 0 10 20 30 40 Period from marriage to first birth Age at first marriage Source: Calculated from DHS 1993 and 1999. slope of the line, which is constant after the onset of a woman's reproductive life.4 It has, on structural break at around 13 years, is less than 1 average, no effect at all for women marrying at 13 (it is 0.8, see footnote 3). Hence, an increase in or younger. And it has a muted effect on older the age of marriage of one year leads to a one women. less a one year increase. This means that as the It is usually argued that delaying marriage re- age at marriage increases, the absolute gap be- duces fertility, since a woman starts reproduc- tween marriage and first birth declines. The re- tion later, and so has less time to achieve her gression results suggest this gap is on average reproductive potential. But in Bangladesh the 2.7 years for a woman marrying at 15, but only majority of women are having three or four chil- 1.8 years for one marrying at 20. This is precisely dren during their reproductive lives. This is quite the phenomenon shown in panels (a) and (c) of possible to achieve if childbearing starts at 20 figure F.4--as the age at marriage increases the rather than 15. Hence, even if delaying marriage interval to first birth declines. delayed first births, which it does to only a lim- The point being made above is that delaying ited extent, it still would not affect the total num- marriage only has a limited effect of delaying the ber of children they have during their lifetime. 1 4 6 F E R T I L I T Y Strategies to Reduce Fertility: acerbated by the spread and escalation of dowry.5 Addressing High-Fertility Households, Strategies that may be adopted include Behavior- Son Preference, and High Mortality Change Communication, particularly including As shown above, one-third of all women aged work with religious authorities, expanding fe- 25 have four or more children. There seems to be male education, and employment and income- greater potential in persuading these families to generating opportunities for women. These last have 3 children than in persuading those having two are areas of already sizeable programs in 2-3 children to have fewer. That son preference Bangladesh. is a barrier to fertility decline in Bangladesh has Bangladesh has followed international norms been argued elsewhere (Chowdury and Bairagi in that fertility decline has been linked to declin- 1990; Bairagim 1996; and Khan and Khanum ing under-five mortality. There remain pockets 2000), and is supported in the regression results of high mortality, especially in Sylhet where use presented above. Reducing son preference of health services is low. Concerted efforts to in- would hence have a beneficial impact on fertility. crease immunization coverage and other inter- This is easier said than done, however: son pref- ventions to reduce mortality can be expected to erence is a deeply ingrained cultural tradition, ex- have an impact on fertility in these areas. 1 4 7 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? APPENDIX F.1: Indirect Estimates of Fertility in the 1990s This appendix provides indirect estimates of fer- obtained from the two CWRs and arrive at a tility in Bangladesh for the 1990s using three decadal estimate (Rele 1976). Also, one may use methods: Rele method, Proximate Determinant smoothed age data. But this procedure has method (Bongaarts decomposition), and the re- some disadvantages as estimates with the gression method. The data for the analysis are smoothed age distribution are influenced by the taken from BFS 1989 and the three DHS surveys, particular method used in smoothing the age and, for one set of calculations, census data. data. Rele (1987) refines the method for situations Rele Method more typical in some developing countries, in- This method infers fertility levels from child- cluding India. The refined method requires two woman ratios (CWR), utilizing the close linear or more successive censuses preferably with a relationship between the CWR and the gross re- 10-year gap between them and use of weights3 production rate (GRR)1 shown to exist at any derived on the basis of extent of under-enumer- given level of mortality (measured by life ex- ation/over­enumeration of two types of CWRs to pectancy, e00). Rele (1976) provides the param- adjust preliminary estimates made under this eters required for implementation of the method. If weights cannot be computed very ac- approach.2 Using these coefficients, GRR is esti- curately for some populations, it may not affect mated from CWR using suitable values of e00 in a much the final or adjusted estimate for GRR/ given population and then applying the sex-ratio TFR. In other words, the estimates are not that at birth estimate to obtain TFR from GRR. sensitive to some variation in the weights. The method uses two types of CWR in esti- Estimates for TFR are obtained using both mating GRR--children aged 0-4 years divided by survey and census data. The survey data are women aged 15-49 (CWR1), and children aged those from three BDHS conducted in the 1990s 5-9 divided by women aged 20-54 (CWR2). The and the BFS-1989. In the case of population cen- e00 value used corresponds to the period when sus, data from only 1991 and 20014 census were the children in the CWR were born. The fertility used. The value of life expectancy at birth (e00) is estimates also refer to the same period viz., esti- assumed on the basis BBS estimates for the mates obtained through using CWR1 produces same. They are given in tables F1.1 and F1.2, dis- an estimate for five years preceding the cen- playing the estimated TFR as obtained under this sus/survey and that with CWR2 produces for 5 to method. Un-smoothed age distribution data 10 years preceding that. have been used for estimating TFR under this This method has certain strengths and weak- method. nesses. Among the former, the estimated fertil- TFR estimates applying the Rele method ity is insensitive to the errors in e00, so that an using survey data with overlapping periods are approximate value of e00 gives a fairly accurate es- shown in table F1.2. Given the overlapping peri- timate of fertility. Among the weaknesses, the ods, a final estimate has been derived using the derived fertility estimates depend on the accu- weights suggested in the refined Rele method racy of CWR. In developing countries, due to er- (1987). rors in age reporting, notably exaggeration of age at young ages and underreporting of very Proximate Determinant Method young children, the reported age distributions Davis and Blake (1956) were the first investiga- are often distorted. To overcome this problem it tors to develop a framework outlining the inter- was initially suggested to average the two rates mediate fertility variables. That seminal work 1 4 8 F E R T I L I T Y identified 11 intermediate variables through Rele Estimates of GRR which fertility is affected in a population. Later T A B L E F 1 . 1 and TFR Obtained from on, Bongaaarts (1978) revised this framework in- Survey Data corporating only 8 of them, which he termed proximate determinants of fertility. As he sug- Preliminary Final gested, each of these eight variables directly estimates estimates influences the fertility and they together deter- Period e00 GRR TFR GRR TFR mine the level of fertility in a population. How- ever, using the data from both developed and BFS­1989 1979­84 50 3.26 6.69 3.09 6.30 developing countries, Bongaarts and Potter 1984­89 53 2.56 5.25 (1983) observed that 96 percent of the variance BDHS-1993/94 in total fertility rates could be explained by four 1984­89 53 2.99 6.12 2.86* 5.86* principal proximate determinants: marriage, 1989­94 56 2.13 4.37 contraception, induced abortion, and lactational BDHS­1996/97 infecundability. The other four, which can be 1987­92 55 2.73 5.61 2.55** 5.22** 1992­97 57 1.92 3.94 disregarded, are: frequency of intercourse, steril- BDHS-1999/2000 ity, spontaneous intrauterine mortality, and du- 1990­95 57 2.32 4.76 2.20*** 4.51*** ration of the fertile period. These last four 1995­2000 60 1.83 3.76 1.85 3.78 factors all relate to the natural fertility. Notes: * for 1984­89; ** for 1987/89­92/94; and *** for 1990/92­95/97. The key contribution of the Bongaarts frame- work is to set up the procedures for parceling out the effects on fertility of different proximate vari- infecundability. Each index measures the extent ables. The forces of each of the intermediate vari- to which fertility is reduced from the maximal ables in the Bongaarts framework is measured by levels by specified proximate determinants. The an index `C' that varies from zero to one. When it indexes are measured as follows: is zero, none of the potential for fertility is trans- lated into births. When it is 1.0, the controls exert Cm = m(a)× g(a). g(a) no restraining effect whatsoever. In Bongaarts' framework, the total fertility rate Where m (a) = age specific proportions cur- (TFR) in a population is expressed as the product rently married among female and g(a) = age of four indexes measuring the fertility inhibiting specific marital fertility rate, effect of four principal intermediate variables and the total fecundity rate (TF). The total fecundity Cc = 1 - 1.08 × u × e. rate is the average number of live births expected Where u = proportion currently using contracep- among women who during their entire repro- tion among married women of reproductive age ductive period remain married, do not use con- and e = average use effectiveness of contraception, traception, do not have any induced abortion, and do not breastfeed their children. The TF does not vary much between populations, lying Rele Estimates of GRR between 13 and 17 births per woman with a stan- T A B L E F 1 . 2 and TFR Obtained from dard value of 15.3 (Bongaarts and Potter 1983). Census Data The following equation summarizes the basic structure of the Bongaarts model by relating the Preliminary Final estimates estimates fertility measures to the proximate determinants: Period e00 GRR TFR GRR TFR TFR = Cm × Cc × Ci × Ca × TF 1981­86 51.0 3.44 7.04 3.33 6.82 Where Cm is the index of marriage, Cc is the 1986­91 54.0 2.93 6.01 2.91 5.97 index of contraception, Ca is the index of in- 1991­96 57.0 2.37 4.86 2.38 4.89 duced abortion, and Ci is the index of lactational 1996­01 60.0 1.91 3.91 1.90 3.89 1 4 9 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? However, the BDHS have shown an increase Ci = 20 . 18.5 + i in the discontinuation rate of contraceptive methods over the 1990s. Hence, the Cc estimates Where i = average duration of post-partum in- for 1996/97 and 1990-2000 using contraceptive fecundity caused by breastfeeding or post-par- use-effectiveness from 1993/94 or the period tum abstinence, around that may overestimate the contraceptive effect on TFR in the former two periods. Thus, Ca = TFR . this may at least partly offset the opposing effect TFR+.4 × (1+u)× TA of induced abortion noted above. Where TA is total abortion rate. The Ci has been estimated using direct esti- Cm, Cc, and Ci are calculated for different mates of the post-partum amenorrhea period by years using the information from three BDHS of different surveys (table F1.4). This has been the 1990s and BFS of 1989. Data for Ca are not falling over time, exerting an upward pressure available, hence, the value of it has been as- on fertility. sumed 1.0 all through. This assumption most The values of Cm, Cc, and Ci as estimated for likely produces an upward estimation of TFR for different periods are presented in table F1.5, to- population that may practice some degree of in- gether with the corresponding TFR. TFR is duced abortion in Bangladesh (Begum 2003), al- shown to have continued to fall during the though the extent is unknown. Such practice 1990s, with the bulk of the decline coming from may also be on the rise over the years. increased contraceptive usage. As noted, Ci ex- In estimating Cc the knowledge about the erted a countervailing tendency. contraceptive use-effectiveness is required. Regression Method There is no uniform view about this. Using the information of 1993/94 BDHS, Islam and others From international comparisons of the family (1996) estimated this to be 0.93, but using planning program performance and fertility village-level statistics (Matlab) Bairagi and other change, Blanc (1990) has derived a set of regres- (1996) estimated a value of 0.81. However, the sion equations that can be used to predict TFR Matlab estimate is more akin to the experiences from the information on CPR. Blanc proposed of the developing countries (Ross and Frank- five regression equations. However, to arrive at enberg 1993). Here Cc is estimated using both a single value of TFR the average of five TFR val- sources (table F1.3) giving two Cc estimates and, ues is suggested. Past users of this method have consequently, two TFR estimates, which are argued that this method is likely to produce an averaged to arrive at a single figure for TFR. underestimation of TFR in a country such as Bangladesh (Kantner and Frankenberg 1986; Kantner and Noor 1992), and suggested that es- timates obtained through this procedure may be Methods-Specific Use-- treated as the lower-bound estimates. The TFR T A B L E F 1 . 3 Effectiveness of estimation formulae under this method are as Contraception follows; to arrive at a single value for the TFR, five different estimates obtained under five dif- Method BDHS 1993/94 Matlab ferent formulae are average out: Pill 0.96 0.74 TFR = 7.30 - (6.42 * CPR) IUD 0.99 0.92 Injectables 0.98 0.99 TFR = 6.83 - (6.20 * CPR) Condom 0.87 0.57 TFR = 7.38 - (7.20 * CPR) Sterilization 1.00 1.00 TFR = 7.28 - (6.55 * CPR) Periodic abstinence 0.81 0.70 TFR = 7.15 - (6.56 * CPR). Withdrawal 0.81 0.70 Others 0.63 0.70 The results are presented in the next section, Source: Islam, Mamun, and Bairagi 1996. alongside those of the other methods. 1 5 0 F E R T I L I T Y Estimated TFR Estimated Duration of The estimates for TFR as obtained under the dif- T A B L E F 1 . 4 Post­Partum Amenorrhoea ferent methods are presented in tables F1.6 and (months) F1.7. The Rele and Proximate Determinant methods are mainly relied upon to arrive at an Period Duration (months) estimate of TFR, as the Regression method is be- 1989 11.9 lieved to produce an underestimate. 1993/94 11.5 The estimates obtained using census and sur- 1996/97 10.9 vey data under the Rele method and those ob- 1999­2000 9.5 tained through Proximate Determinant methods Source: Survey reports. are in close agreement, while the Regression method, as expected, produced consistently lower estimates for all the periods. Among the while the Proximate Determinant method esti- former two, the major discrepancy is for TFR es- mated the same level for the year 1989. Thus, on timates for the early 1980s. The Rele method, by the basis of Rele and Proximate Determinants using census data, produced an estimate of 6.82 method, we can conclude that the fertility level for the TFR for the 1981-86 period and, using in Bangladesh before 1990 was around 6 per survey data, produced an estimate of 6.30 for the woman. A similar result was found by Cleland 1979-84 period. Both estimates are obtained and others (1994), who report a TFR of 5.86 for using primarily the CWR2, which used the 1988-89 using CPS data, and Islam and others 5-9 age group as the numerator, hence it may be (1996), who report a TFR of 5.83 using BFS. As a that there was some higher enumeration of this crude measure, the average number of children age group in the census, which is less of a prob- born to women aged 35 or over in DHS 1992/93 lem in the survey data, which are collected with was 5.9, which fits with these estimates. greater care and attention. Both in 1961 and The Rele method, using census data for the 1974 the population censuses have excessively first half of 1990s viz., for 1991-96, produced an enumerated the 0-9 age and particularly the estimate of 4.9, while using survey data pro- 5-9 age (Bangladesh 1961; Begum 1976, 1990). duced an estimate of 4.51 for 1990/92-1995/97 For the 1986-91 period, the Rele method, period. The Proximate Determinant method using census data, produced an estimate of 6.0 produced an estimate of 4.5 for 1993/94, which for the TFR and, using survey data, produced an corresponds roughly to the mid-period of the estimate of 5.9 for 1984-89 period. Hence, ac- above two periods noted for the Rele method. cording to this method, the TFR in the country From these rates, it appears the TFR in 1993/94 in the 1987-89 period was around 6 per woman, in Bangladesh was around 4.5. T A B L E F 1 . 5 Estimated Index Values of the Proximate Determinants Cc Using BDHS use- Using Matlab use- Cm x Cc Year Cm effectiveness effectiveness Ci x Ci * TFR* BFS 1989 .826 .695 .730 .658 .387 5.99 BDHS 1993/94 .761 .550 .610 .667 .294 4.50 BDHS 1996/97 .756 .503 .573 .680 .276 4.23 BDHS 1999­2000 .745 .466 .545 .714 .269 4.11 Percent change, 1989­99 ­9.8 ­32.900 ­25.3 8.5 ­30.500 ­31.40 Share 32.200 95.6* ­27.900 100.000 Note: * Calculated using average of two Ccs.. Growth rate of TFR differs from that of product term due to rounding. 1 5 1 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Second, as these estimates show, the fertility Estimates of TFR Using T A B L E F 1 . 6 did decline in the country over the 1990s albeit Rele Method at a slower rate than that noticed between the late 1980s and early 1990s. For example, TFR de- Period Census estimates Period Survey estimates clined by about 25 percent between 1989 and 1981­86 6.82 1979­84 6.3 1993/94, while during 1993/94 and 1999/2000 1986­91 5.97 1984­89 5.86 1991­96 4.89 1987­89­1992­94 5.22 the magnitude of decline has been only about 1996­2000 3.93 1990­92­1995­97 4.51 11 percent. 1995­2000 3.78 The dynamic behind the slowing fertility de- cline in the second half of the 1990s can be dis- cerned from the trend in index values of various proximate determinants over this time (table For 1996/97 the Proximate Determinant F1.8). As they suggest, for a big drop in TFR method has produced an estimate of 4.23 and, between 1989 and 1993/94, both marriage and for 1999-2000, 4.11. On the other hand, the Rele contraception played a big role. Contraception method using census data produced an estimate use during that period increased 14 percentage of 3.93 for the second half of 1990s viz., for 1996- points from 31 percent to 45 percent, while the 2000, which corresponds to the mid-period of never-married group in the 10-49 age group in- 1998, while using survey data produced an esti- creased from 20 to 32 percent. Since 1993/94, mate of 3.78 for 1995-2000, corresponding to the marriage played only a marginal role in inhibit- year 1997. Hence, from these estimates we can ing fertility. Such role of marriage improved by conclude that the TFR in the country during less than 1 percent during 1993/94-1996/97 and 1996/97 was perhaps around 4.2 and that during by another 1.5 percent during 1996/97­1999- 1999-2000 it was around 4. 2000, while during the 1989­1993/94 period Summary alone it improved by 8 percent. Two major conclusions can be drawn from this Similarly, during the 1989­1993/94 period, analysis. First, national sample surveys con- while the fertility inhibiting effect of contracep- ducted in the country consistently underesti- tion increased by 19 percent, it improved by a mated the true fertility level in the country. much smaller degree after that. During 1993/94­ According to these estimates, of three BDHS 1996/97 the fertility-inhibiting effect of contra- conducted in the country in the 1990s, under- ception increased by 7 percent and by 6 percent estimation of TFR by highest margin has taken during 1999-2000. While the lesser degree of in- place in 1993/94. crease in the contraception use after 1993/94 compared to 1989-1993/94 may in part be re- sponsible for this, there is also evidence that in Estimates of TFR Using the latter part of the 1990s, contraception use re- Proximate Determinant lied more on relatively less-effective methods T A B L E F 1 . 7 Method and and the use-effectiveness of the method experi- Regression Method enced some decline during this period. Throughout this period the role of post- Proximate determinant method partum infecundability has been in the opposite Using Cc Using Cc Average direction viz., it favored fertility increase. Not estimated from estimated from of two Regression only this, this role increased noticeably over Year 1993/94 BDHS Matlab data estimates method time, offsetting part of the fertility-inhibiting ef- 1989 5.85 6.14 5.99 5.16 fect exerted by other proximate determinants. 1993/94 4.27 4.74 4.50 4.25 During the 1989­1993/94 period the fertility in- 1996/97 3.96 4.51 4.23 3.96 hibiting effect of Ci slackened by 1.4 percent, it 1999­2000 3.79 4.43 4.11 3.65 slackened by another 1.9 percent during the 1 5 2 F E R T I L I T Y 193/94­1996/97 period and another 5 percent Percentage Change in TFR during 1996/97­1999/2000. and Different Indices of The outcome of all these effects has been T A B L E F 1 . 8 Proximate Determinants continuous weakening of collective strength of Since 1989 the fertility-inhibiting effect of various proxi- mate determinants. During the 1989­1993/94 Combined effect of period mentioned above, three proximate de- Period Cm Cc* Ci Cm, Cc, Ci TFR terminants viz., Cm, Cc, and Ci together could 1989­1993/94 ­8.0 ­18.6 +1.4 ­24.0 ­17­22 improve their fertility-depressing effect by an- 1993/94-1996/97 ­0.7 ­7.2 +1.9 ­6.1 ­6­15 other 24 percent. While they could continue im- 1996/97-1999/2000 ­1.5 ­6.0 +5.0 ­2.5 ­5­10 proving such an effect throughout the rest of * Change for Cc has been calculated from average value. 1990s, they could do so only to lesser degree. During 1993/94­1996/97 such improvement has increased another 6 percent, and during 1996/97 and 1999/2000 it improved by another 2.5 percent. 1 5 3 ANNEX G: ANALYSIS OF BINP'S COMMUNITY-BASED NUTRITION COMPONENT The Community-Based Nutrition Component, Bangladesh as poverty and the resultant house- the central component of BINP, focused on hold food insecurity do" (BINP Staff Appraisal growth monitoring, combined with nutritional Report, World Bank 1995, para 1.13, p.4). counseling and food supplementation. The im- Therefore, changing bad practice to good will pact of the component has been studied in an in- bring about nutritional improvements. There are dependent evaluation conducted for the project a number of steps in the causal chain behind this (Karim and others 2003), by Save the Children in approach:1 their report Thin on the Ground (Save the Children 2003), in an evaluation commissioned · The right people (those making decisions by GoB's Implementation, Monitoring and Eval- with respect to undernourished children) are uation Department (Haider and others 2004), targeted with nutritional messages. and in research carried out under the supervi- · These people participate in project activities, sion of Professor Mascie-Taylor at the University and so are exposed to these messages. of Cambridge, U.K. (Mascie-Taylor 2004). For · Exposure leads to acquisition of the desired this report, further analysis has been conducted knowledge. using existing data, but partly different ap- · Acquisition of the knowledge leads to its proaches to those used in these other studies. adoption (that is, a change in practice) For this study we have had access to the project · The new practices make a substantial impact evaluation dataset, the data collected by Save on nutritional outcomes. the Children, and data from Helen Keller Inter- national's Nutritional Surveillance Survey. Thanks A feeding program for malnourished children are due to the respective agencies for making and pregnant women was implemented along- these data available, and to World Bank staff in the side growth monitoring. For this program to Nutrition Hub for facilitating access to the evalua- work: tion dataset. This annex presents analysis relating to both · The target groups have to enroll in the child nutrition and low birth weight, whose program. prevalence is intended to be reduced through · The criteria have to be correctly applied in se- improved weight gain among pregnant mothers. lecting those to receive supplementary feed- The annex is structured to reflect the logic of ing. project design. The key assumption behind the · Those selected for supplementary feeding Community-Based Nutrition Component is that have to attend sessions to receive the food. "bad practices" are responsible for malnutrition · There can be no leakage (such as selling of in Bangladesh. This point of view was strongly food supplements) or substitution (reducing argued in the appraisal document for the pro- other food intake). ject: "behaviors related to feeding of young chil- · The food has to be of sufficient quantity and dren have at least as much (if not more) to do quality to have a noticeable impact on nutri- with the serious problem of malnutrition in tional status. 1 5 5 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Project Coverage and Targeting · There is no sex bias in participation. Partic- ipation has a non-linear relationship with the Participation in Growth Monitoring for Children child's age, with those aged 9-10 months Under BINP, growth monitoring sessions are most likely to attend. held monthly at which the Community Nutrition · Both younger mothers and daughters-in-law Promoter (CNP) weighs all children at the Com- are less likely to attend (daughter-in-law munity Nutrition Center (CNC).2 Their weight is meaning here a woman living with her marked against age on a growth monitoring mother-in-law). The daughter-in-law dummy chart by the CNP. Mothers of children were alone is significantly negative (results not asked whether or not they attended the growth shown), but is no longer so when an interac- monitoring sessions with their child, with possi- tive term is included for the daughter-in-law ble responses of never, sometimes, and almost dummy and the dummy for the two more always at midterm, and never, sometimes, al- conservative thanas, Rajnagar and Sharasti.3 most always, and always at endline. The results The significant negative coefficient on this in- show that around 85 percent of mothers reply teractive term shows the effect of restrictions that they always participate (table G.1). There is on female mobility in attending growth mon- little change in the overall average between end- itoring sessions, although they only have an line and baseline, but considerable variation effect in two of the six thanas. If the mother is among thanas, with participation consistently employed outside the household, which is lowest in Rajnagar. positively associated with women's agency A simple bivariate analysis shows that more (Annex E), she is more likely to participate. It educated mothers are more likely to attend than is also the case that Hindu women are more less educated, though this difference is no likely to participate than Muslim women. longer significant by the endline (table G.2). · Qualitative analysis shows that remoteness However, at midterm it appears that the poorer (travel time) is also a constraint on participa- mothers (lower housing quality) are more likely tion. There is no direct measure of this vari- to participate, but this pattern is reversed by able in the dataset, but it is likely that type of endline. Multivariate analysis of participation water supply is a reasonable proxy for this was used to disentangle these effects, using both variable (it is difficult to imagine a direct effect a logit specification of participate/don't par- of this variable), the results confirming that ticipate and a multinomial logit of the level of inferior forms of water supply (indicating re- participation. moteness) significantly reduce participation.4 The multivariate results, which are very simi- · The results confirm that the most educated lar for midterm and endline, show the following: women are significantly less likely to partici- T A B L E G . 1 Participation in Growth Monitoring by Thana (percent) Midterm Endline Thana Never Occasionally Regularly Never Occasionally Regularlya Banaripara 2 3 94 6 4 91 Faridpur 8 13 80 4 6 90 Gabtoli 2 5 93 2 3 95 Mohammadpur 4 8 88 4 3 94 Rajnagar 15 15 70 16 9 75 Shahrasti 2 12 87 9 7 83 Total 5 9 85 7 5 88 a. Combines "almost always" and "always." Source: BINP evaluation dataset. 1 5 6 A N A L Y S I S O F B I N P ' S C O M M U N I T Y - B A S E D N U T R I T I O N C O M P O N E N T pate. There is no income-bias in participation Rele Estimates of GRR in the multivariate analysis, this result being T A B L E G . 2 and TFR Obtained from robust to several different specifications (that Census Data is, models that do not have other variables correlated to income, such as water and sani- Non- Occasional Regular tation and parent's education). However, at participant participants participants midterm the less poor are less likely to at- Comparison Compared to Compared to Compared to tend, based on both the housing quality index for calculation occasional regular non- and the categorical income variable (based on of t-test participant participant participant the enumerator's assessment). By contrast, at Midterm endline those with better housing are more likely to participate, but women in house- Mother's education 2.1 2.2*** 2.8** House index 4.5*** 4.1** 4.2*** holds with sizeable landholdings less likely to do so (reflecting a time constraint for such Endline women, which is shown to be important Mother's education 4.0 4.1 4.1 when it comes to putting knowledge into House index 4.5 4.7 4.7* practice). Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. Of these factors, it is the more conservative Participation in Growth T A B L E G . 3 ( a ) nature of Rajnagar and Shahrasti that appears to Monitoring: Midterm account for lower participation in these thanas. The coefficients on the other variables are not Logit Multinomial logit such as to have such a large difference, and the Occasional Regular differences in these explanatory characteristics Participants participants participants between the thanas are not so great (see fig- Variable Coeff. t stat. Coeff. t stat. Coeff. t stat. ure G.1). Child sex ­0.02 ­0.12 0.12 0.59 ­0.04 ­0.26 The logic of the project is that attendance at Child age 0.19 1.77* 0.20 1.62* 0.20 1.79* growth monitoring provides a context for nutri- Child age squared ­0.01 ­1.85* ­0.01 ­1.39 ­0.01 ­1.92* tional counseling, especially for women with Mother's age ­0.04 ­3.08*** ­0.05 ­2.60*** ­0.04 ­3.04** Female-headed malnourished children. The monitoring sessions household 0.00 ­0.01 ­0.21 ­0.39 0.04 0.08 are not a good setting for such counseling, but Mother works 1.34 1.23 0.15 0.11 1.45 1.33 BINP provides other opportunities for this, in- Daughter-in-law ­0.19 ­0.54 ­0.15 ­0.36 ­0.20 ­0.56 cluding group sessions. As shown in table G.4, Daughter-in-law in Rajnagar and fewer women have participated in nutrition dis- Shahrasti ­1.13 ­2.15** ­0.81 ­1.37 ­1.19 ­2.22** cussions than participate in growth monitoring. Widow ­0.45 ­0.49 0.04 0.04 ­0.56 ­0.61 Save the Children data also show participation in Number of pregnancies 0.18 1.63* 0.01 0.09 0.21 1.82* nutritional counseling at just under 50 percent. Primary education 0.13 0.52 0.05 0.19 0.14 0.57 Secondary education 0.42 1.35 0.27 0.79 0.45 1.42 Targeting of Children's Supplementary Feeding Higher education ­0.41 ­0.96 ­0.49 ­0.91 ­0.40 ­0.94 Per capita income 0.00 ­1.68* 0.00 ­1.76* 0.00 ­1.67* Growth monitoring is used to provide supple- House index ­0.22 ­2.46** ­0.35 ­3.26*** ­0.20 ­2.20** mentary feeding to children who are either fal- No drinking water ­0.98 ­2.12** ­0.63 ­1.30 ­1.08 ­2.23** tering in their growth or severely malnourished No latrine ­0.01 ­0.04 ­0.10 ­0.34 0.00 0.01 (third-degree level of malnutrition by the Faridpur ­1.64 ­3.91*** ­0.32 ­0.74 ­1.76 ­4.03*** Gabtoli ­0.21 ­0.35 0.23 0.48 ­0.23 ­0.38 Gomez classification, which corresponds to ap- Mohammadpur ­0.76 ­1.84* 0.19 0.43 ­0.82 ­1.96*** proximately less than ­4 SDs from the WAZ ref- Rajnagar and erence median). Table G.5 presents the official Shahrasti ­1.21 ­2.96*** 0.08 0.19 ­1.32 ­3.20*** Constant 4.55 4.58*** 2.07 1.84* 4.38 4.35*** screening criteria, though field experience sug- Pseudo R­squared 0.12 0.07 gests that the actual criteria applied vary from Observations 3,541 3,541 thana to thana. The majority of the children ad- Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. 1 5 7 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? given a food supplement. The first screening Participation in Growth T A B L E G . 3 ( b ) should happen as soon as the pregnancy is de- Monitoring: Endline tected in order to measure the pre-pregnancy weight. Table G.8 shows participation rates in Logit Multinomial logit weighing and feeding sessions. As expected, Occasional Regular some women participate only in the weighing Participants participants participants sessions. Surprisingly, there are also women Variable Coeff. t stat. Coeff. t stat. Coeff. t stat. who participate only in the feeding sessions. Child sex ­0.02 ­0.12 0.12 0.59 ­0.04 ­0.26 Participation in weighing sessions is around Child's age 0.01 0.29 ­0.01 ­0.53 0.01 0.39 75 percent, and just under half of pregnant Child's sex ­0.21 ­1.14 ­0.51 ­2.01** ­0.18 ­0.98 women received supplementary feeding. There Number of ­0.03 ­0.30 ­0.07 ­0.54 ­0.03 ­0.28 is little change in participation rates between the pregnancies two surveys. If the cutoff point used to detect Mother's age 0.04 1.46 0.02 0.56 0.04 1.52 Mother's work 1.05 1.73* 1.14 1.65* 1.08 1.77* pre-pregnancy malnutrition is a body mass index Primary education 0.25 1.01 0.21 0.63 0.25 1.03 (BMI) below 18.5, as stated in the project doc- Secondary education 0.36 1.19 0.46 1.18 0.37 1.22 ument, then fewer women are receiving sup- Higher education ­0.68 ­1.80* ­0.87 ­1.51 ­0.67 ­1.76* plementary feeding than should be doing so, Hindu household 0.83 2.22** 0.46 0.97 0.85 2.27** particularly at the midterm when the proportion House index 0.21 1.88* 0.19 1.34 0.21 1.87* of women below this cutoff point is 68 percent. Land owned 0.00 ­2.10** 0.00 ­1.63* 0.00 ­2.02** No drinking water ­0.50 ­2.16** ­0.64 ­2.08** ­0.50 ­2.15** By endline, 47 percent are below this threshold, No latrine ­0.25 ­1.09 ­0.34 ­1.10 ­0.25 ­1.07 suggesting few women are excluded if only eli- Faridpur 0.44 1.06 0.91 1.67* 0.42 1.02 gible women participate, but this is doubtful, as Gabtoli 0.79 1.58 0.56 0.86 0.82 1.64* is shown below. Mohammadpur 0.27 0.67 ­0.10 ­0.17 0.28 0.70 Table G.9 shows the percentages of women Rajnagar ­0.94 ­2.81** 0.35 0.74 ­1.04 ­3.09*** Shahrasti ­0.45 ­1.51 0.45 1.02 ­0.50 ­1.68* of different nutritional status that participate in Constant 1.58 1.78* ­0.28 ­0.24 1.42 1.60 project activities. BMI measurement is from Pseudo R­square 0.08 0.07 women who were not more than six months Observations 2538 2538 pregnant in order to avoid the bias induced by Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. the pregnancy weight gain. This is not a correct indicator of the mother's pre-pregnancy status, mitted into supplementary feeding qualify by but is the best that can be done in the absence virtue of growth faltering; analysis of register of data on pre-pregnancy weight. data collected by Save the Children show only Participation in weighing sessions is indepen- 5 percent of supplementary feeding participants dent of mothers' nutritional status. Since all to be severely malnourished (WAZ < ­4 SDs), women are meant to participate, this is the ex- though another 60 percent are malnourished by pected finding if there is no bias in favor of, or the conventional criterion of WAZ < ­2 SDs against, nutritionally deprived women. How- (table G.6). ever, coverage of supplementary feeding is cor- However, not all children enrolled in supple- rectly correlated to nutritional status in both mentary feeding appear to be eligible (table surveys, apparently more so at the endline, al- G.7).5 A likely explanation for this is the inability though a third of even the most severely mal- of all CNPs to accurately interpret the growth nourished women do not participate. However, charts, a finding from a limited sample encoun- over 30 percent of women with normal BMI sta- tered during OED's fieldwork.6 tus also received supplementary feeding. At the endline, 40 percent of the recipients of supple- Targeting of Activities for Pregnant Women mentary feeding were women of normal BMI sta- Under the project, pregnant women are meant tus. In summary, targeting errors at the endline to be weighed monthly and screened for malnu- were around 60 percent for Type I error (miss- trition. Women below a certain cutoff point are ing women who should have benefited) and 1 5 8 A N A L Y S I S O F B I N P ' S C O M M U N I T Y - B A S E D N U T R I T I O N C O M P O N E N T Various Factors Affect Women's Participation, but F I G U R E G . 1 Restrictions on Women's Mobility in More Conservative Areas Are the Most Important 1.0 0.9 0.8 Living in monitorinng Rajnagar or 0.7 Shahrasti growth 0.6 in 0.5 Living with 0.4 mother-in- law in participation 0.3 Rajnagar of 0.2 or Shahrasti 0.1 Probability0.0 Base value Living with mother- Higher education No water or in-law in Rajnagar sanitation or Shahrasti (remote location) 40 percent for Type II error (giving it to women birth weight. In addition, mothers are advised to who were not eligible). Improved coverage of give colostrums to the newborn and to practice poorly nourished women, especially the most exclusive breastfeeding of the infant up to the severely malnourished, would increase program age of 5-6 months. For older children, advice is impact on pregnancy weight gain and, hence, given on good nutrition. As outlined above there low birth weight. are several steps in the underlying theory behind this approach. This section investigates the third Knowledge Acquisition and the and fourth of these steps, that is, whether knowl- Knowledge-Practice Gap edge is acquired and if it is put into practice, fail- The main objective of the project is to improve ure to do so is labeled the knowledge-practice children's nutritional status by changing the gap. pregnancy, lactation, and feeding practices of mothers. Pregnant women are advised, individ- Evidence from the Save the Children Data ually and in group sessions, to take more food The Save the Children study used three types of and rest during pregnancy in order to avoid low nutrition knowledge to calculate the knowledge- Mother's Participation in Group Discussions on Nutritional Topics T A B L E G . 4 (percent) Malnutrition Birth weighing Weight gain Iodine Vitamin A Prevention Participated 66.4 66.2 62.2 47.5 48.6 52.2 Did not participate 30.8 30.9 34.7 49.1 47.9 44.3 Not invited 2.8 2.9 3.1 3.4 3.5 3.5 Source: BINP midterm data. 1 5 9 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? practice gap. These are the knowledge on rest Eligibility Criteria during pregnancy, colostrums, and exclusive T A B L E G . 5 for Children's breastfeeding reported in table G.10. Supplementary Feeding Regarding knowledge, all the differences in knowledge between project and control areas Eligibility conditions are statistically significant, though these differ- Normal children ences are not that great. There are differences in From 6 to 12 months Less than 600 gram gain in 2 months practice between project and control areas, but From 12 to 24 months Less than 300 gram gain in 2 months only one of them, rest during pregnancy, is sta- I and II degree malnutrition tistically significant. Knowledge is generally high, being between 50 and 80 percent in both project From 6 to 12 months Less than 600 gram gain in 3 months From 12 to 24 months Less than 300 gram gain in 3 months and control areas. The knowledge-practice gap is small in the case of colostrums, moderate in the case of rest, and extremely high in the case of exclusive breastfeeding. The project improves Malnutrition Rates at both knowledge and practice, but does not re- T A B L E G . 6 Feeding Start by duce the knowledge-practice gap. Feeding Episode Regression analysis can be used to answer two questions: (1) What are the obstacles that Observations Percent prevent the knowledge being put into practice? < ­4 81 5.4 and (2) Does the project help to reduce the > = 4 & < ­2 903 59.5 knowledge-practice gap in general, or for > = ­2 & < 0 522 39.4 mothers with particular characteristics? The > 0 11 0.7 knowledge-practice gap is modeled using a pro- All children 1,517 100.0 bit regression, where the dependent variable is Source: Calculated from Save the Children register data. one if there is a knowledge-practice gap. The sample is composed of mothers who have the Eligible and Non­Eligible for knowledge about the three nutritional practices T A B L E G . 7 Supplementary Feeding described above. The estimation of this model Among the Fed Children can produce biased coefficients, because unob- served determinants of the knowledge can have Observations Percent effects on the practice as well. In order to avoid this bias, a two-step Heckman approach is Eligible 1,283 84.6 Non-eligible 233 15.4 adopted, first estimating a selection equation All children 1,516 100.0 explaining the knowledge among mothers, and then a gap equation for mothers who have the knowledge. The results are in table G.11. The explanatory variables for the selection Participation of Pregnant equation (the "knowledge" equation) are T A B L E G . 8 Women in Weighing and mother's age, the education of the mother and Feeding Sessions (percent) of the head of household, the membership of a group, and residence in a project area. The var- 7 Midterm Endline iables included in the knowledge-practice gap Weighing sessions 73.2 71.6 equation are possible constraints that can pre- Feeding sessions 49.2 44.8 vent mothers from putting in practice their Of which nutritional knowledge. These include time con- Only weighing 30.9 36.7 straints due to house and farm work, resource Only feeding 6.9 9.3 availability, and women's agency within their Weighing and feeding 42.3 35.8 households. The number and type of obstacles 1 6 0 A N A L Y S I S O F B I N P ' S C O M M U N I T Y - B A S E D N U T R I T I O N C O M P O N E N T used in the regression is limited by the informa- Participation in BINP tion contained in the dataset. In order to assess T A B L E G . 9 Activities by BMI Status the project effect on the knowledge-practice gap, a project dummy is included together with Normal Mild Moderate Severe some interactive terms to pick up possible dif- ferential project impact on different groups. Midterm The multivariate analysis shows that attending Weighing sessions 72.8 74.2 72.6 72.5 nutritional counseling indeed has a significant im- Supplementary feeding 39.4 49.5 56.0 58.6 pact on a woman's nutritional knowledge, though Endline being in receipt of supplementary feeding does Weighing sessions 71.8 70.0 72.4 74.2 not.8 However, even when these participation Supplementary feeding 34.3 51.4 60.2 68.9 variables are included, the BINP project dummy Recipients of supplementary feeding by BMI status is still significant. This means either that there are spillover effects (women who get the knowledge Midterm 25.2 31.6 22.9 20.3 in nutrition sessions communicate it to others) or Endline 39.8 33.2 16.3 10.7 that other project activities not captured in the Note: BMI categories used here are those recommended by WHO (1995): greater than 18.5 is normal, be- tween 17 and 18.5 is mild thinness, between 16 and 17 is moderate, and below 16 is severe. participation variables--for example, women's group meetings--are also channels for commu- nication of nutrition education. According to feeding for daughters-in-law and better-off fami- these regression results, simply living in the pro- lies (who are otherwise less likely to breastfeed). ject area raises a women's probability of having a piece of nutrition knowledge by 7 percent, but The BINP Evaluation Sample full participation in project activities increases this The BINP evaluation data included information al- probability to 10 to 23 percent (figure G.2). That lowing calculation of the gap for a larger range of is, the proportion of women aware of the impor- behavior. However, this information was collected tance of colostrums feeding is 23 percent greater for women participating in project activities than for women in the control area. Other determi- Nutrition Knowledge nants of women's nutritional knowledge are of Pregnant and T A B L E G . 1 0 found to not vary much between the project and Lactating Mothers control and so account for only a small amount of (Save the Children data) the difference in knowledge between the two Project (%) Control (%) t stat. areas. The gap equations show that some of the ob- Rest during pregnancy stacles included help determine the gap. Time at Knowledge 77.1 69.4 4.00*** work on the farm and in the house, attending Practice 58.4 52.3 3.56*** children, and other household members' needs, Knowledge-practice 30.7 34.7 ­2.15** all reduce mothers' ability to rest during preg- gap nancy. There are fewer obstacles to the mother's Colostrum use of colostrums for lactation, which is a rea- Knowledge 75.4 71.9 4.33*** sonable result. The number of pregnancies the Practice 62.7 51.5 1.53 mother has reduces the gap, perhaps as these Knowledge-practice 18.3 21.1 ­1.30 women have learned from previous experience. gap Work on the farm and general lack of resources Breastfeeding seem to increase the knowledge-practice gap in Knowledge 77.6 68.7 5.06*** the case of breastfeeding, which is also larger for Practice 3.6 4.9 ­1.16 daughters-in-law. The project area dummy is Knowledge-practice 95.7 94.2 1.08 never significant, but the project does seem to gap close the knowledge-practice gap for breast- Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. 1 6 1 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? mainly in the endline survey of 2003, and only for Knowledge and Knowledge- T A B L E G . 1 1 two behaviors (food intake and rest during preg- Practice Gap Equations nancy) is it possible to observe that the change occurred over time (tables G.12 and G.13). Rest during pregnancy Colostrum Colostrum The knowledge and the practices related to pregnancy are higher in project areas, and the Coeff. t stat. Coeff. t stat. Coeff. t stat. differences are highly significant. Differences in Knowledge-practice gap equation both knowledge and practice are smaller for breastfeeding and colostrum. The knowledge- Daughter-in-law ­0.04 ­0.40 ­0.07 ­0.68 0.30 2.30** H/h owns land 0.00 1.63* 0.00 ­0.07 0.00 0.95 practice gap is always very large, except for Farming household 0.14 3.44*** ­0.07 ­1.67* 0.12 1.99** colostrums. The gap is significantly smaller, both Durables index 0.00 ­0.16 ­0.01 ­0.77 ­0.10 ­3.46*** in size and statistically, in project areas com- No. of children 0.09 4.58*** ­0.06 ­2.69*** 0.02 0.53 pared to control areas. However, in three cases Mother's education ­0.01 ­0.30 0.03 1.01 0.04 1.38 (breastfeeding, colostrums, and hard work dur- Father's education ­0.04 ­1.35 0.03 1.00 ­0.04 ­1.36 ing pregnancy), the gap is as large in project Elderly male in areas as in control areas. household 0.11 1.90* 0.03 0.60 0.03 0.35 Changes in knowledge and practice over time Project area 0.13 1.25 0.03 0.26 ­0.13 ­0.58 can be observed only for the food intake and rest Project * durables ­0.01 ­0.72 ­0.01 ­0.69 0.07 2.49** during pregnancy. Note that the three surveys Project * daughter- did not phrase the questions in the same way, in-law ­0.06 ­0.57 ­0.01 ­0.09 ­0.28 ­1.81* Project * children ­0.03 ­1.28 0.03 1.00 ­0.01 ­0.22 and that the baseline knowledge and practice of Intercept ­0.95 ­11.07*** ­1.08 ­10.17*** 1.57 5.66*** rest during pregnancy in control areas is unrea- sonably high. Hence, the analysis over time is Knowledge (selection) equation limited to midterm and endline data only. Table Mother's age 0.04 2.27** ­0.01 ­0.50 0.03 1.49 G.13 shows that knowledge and practice of rest Mother's age squared 0.00 ­2.18** 0.00 0.01 0.00 ­2.13** and food intake during pregnancy are increasing Mother primary educ. 0.26 3.59*** 0.18 2.51** 0.20 2.29** over time in both project and control areas. The Mother secondary knowledge-practice gap is increasing in project educ. 0.44 4.94*** 0.43 5.20*** 0.54 5.43*** areas over time, as the practice does not increase Mother tertiary educ. 0.36 1.97** 0.98 5.80*** 0.91 4.61*** as rapidly as the related knowledge. Education h/h head 0.02 3.00*** 0.02 3.04*** 0.01 2.07** The knowledge-practice gap is much smaller Female headed for health matters than for nutrition, though household 0.33 3.04*** 0.22 2.28** 0.41 3.90*** Group member 0.07 1.92* 0.07 2.04** 0.08 1.86* there are significant origin differences in knowl- Project area 0.16 2.16** 0.21 2.44** 0.22 2.94*** edge of the treatable diseases such as goiter and Supplementary night blindness. These differences are much feeding ­0.07 ­0.95 0.01 0.13 ­0.02 ­0.26 smaller in the case of diarrhea and ARI, probably Attending nutrion because the latter diseases are much more fre- meeting 0.27 3.35*** 0.23 3.24*** 0.19 2.08** quent, as are prevention campaigns all over the Participated in nutri- country. The same differences are observed in tion discussions 0.13 1.68* 0.12 1.45 0.01 0.16 the practices of prevention/treatment when any Project * primary of these diseases occur. education 0.03 0.28 0.00 ­0.02 0.10 0.93 Table G.14 reports the regression results of the Project * secondary knowledge-practice gap using the BINP data. The education 0.05 0.47 0.11 1.05 0.07 0.58 knowledge equation shows that parental educa- Project * tertiary education 0.40 1.75* 0.11 0.49 0.04 0.18 tion matters, but that there is also a large project Intercept ­0.38 ­1.33 0.02 0.07 ­0.04 ­0.14 effect. Some clear patterns emerge with respect Selection term 1.44 5.82*** 1.89 3.33*** 0.61 2.74*** to the knowledge-practice gap. Women in poor N 7,216 7,217 7,216 households are more likely to have any of the Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. three gaps: resource constraints prevent women 1 6 2 A N A L Y S I S O F B I N P ' S C O M M U N I T Y - B A S E D N U T R I T I O N C O M P O N E N T Women Living in Project Areas Are More Likely to Have F I G U R E G . 2 Nutritional Information, Especially if They Participate in Project Activities 1.0 Control Project area Participate in project activities 0.8 knowledge 0.6 having of 0.4 0.2 Probability 0.0 Rest Colostrum Breastfeeding from putting knowledge into practice. However, Mothers' Knowledge and being a member of an income-generation group T A B L E G . 1 2 Practices Related to Major and having a vegetable garden make it more likely Child Diseases (BINP data) that a woman will not reduce food intake during pregnancy, though the vegetable garden in- Project (%) Control (%) t stat. creases the probability of engaging in hard work. Women who are pregnant in the working season Diarrhea are also more likely to not adopt changed prac- Knowledge 91.2 84.5 2.88*** tices for all three cases.9 Having more children Practice 92.6 84.0 5.1*** also reduces the likelihood of resting more, not Knowledge-practice 1.3 4.8 ­5.16*** gap eating less, and not engaging in hard work. There is a project effect in reducing the likelihood of eat- ARI ing down, which is possibly the result of food sup- Knowledge 82.0 73.7 3.96*** plementation rather than the advice per se, but Practice 87.2 82.4 3.67*** there is also a project impact in reducing the Knowledge-practice 3.7 5.9 ­1.74* knowledge-practice gap for hard work. gap Goiter Leakage of Supplementary Feeding Knowledge 52.3 32.4 9.25*** Supplementary feeding may not have the de- Practice 52.8 32.5 9.37*** sired nutritional impact if there is either leakage Knowledge-practice 2.4 6.2 ­1.89* (it is not consumed by the targeted individual) gap or substitution (the target individual consumes Night blindness less of other foodstuffs, so that the supplemen- Knowledge 72.1 57.0 3.94*** tary feeding merely takes their place). In princi- Practice 73.4 57.7 4.09*** ple, BINP avoids leakage by having feeding Knowledge-practice 2.0 3.3 ­0.96 sessions at which the food is consumed. How- gap ever, as shown in table G.15, in practice, many Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. 1 6 3 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Nutrition Knowledge of Pregnant and Lactating Mothers T A B L E G . 1 3 (BINP data) Baseline Midterm Endline Project (%) Control (%) t stat. Project (%) Control (%) t stat. Project (%) Control (%) t stat. Food intake Knowledge 62.8 65.4 ­0.48 83.6 51.5 15.4*** 91.9 67.5 17.43*** Practice 47.5 74.6 ­7.52*** 55.7 21.9 17.17*** 58.6 29.4 8.75*** Knowledge-practice gap 49.9 20.5 6.92*** 34.3 61.3 ­11.31*** 37.7 61.4 ­5.63*** Rest Knowledge 81.2 61.4 8.18*** 89.0 58.0 23.36*** Practice 64.0 35.2 11.61*** 67.0 46.2 6.98*** Knowledge-practice gap 22.8 46.9 ­9.54*** 27.8 40.0 ­3.49*** Hard work Knowledge 93.9 82.7 5.35*** Practice 52.6 44.1 2.98*** Knowledge-practice gap 44.8 52.5 ­2.62** Smoking Knowledge 88.0 65.4 4.76*** Practice 80.5 85.5 ­3.13*** Knowledge-practice gap 17.9 10.4 4.35*** Iodized salt Knowledge 94.5 77.4 13.09*** Practice 79.2 42.7 7.53*** Knowledge-practice gap 17.6 39.3 ­5.21*** Breastfeeding Knowledge 96.6 89.6 5.86*** Practice 62.5 61.1 0.34 Knowledge-practice gap 37.0 39.6 ­0.65 Colostrum Knowledge 97.4 92.4 4.87*** Practice 95.6 92.3 3.73*** Knowledge-practice gap 2.7 4.1 ­1.85* Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. children and pregnant women get the food out- pregnancy, mainly as it is believed to make for an side of the feeding sessions, making leakage a easier childbirth. Through its nutritional coun- stronger possibility. In addition, Save the Chil- seling BINP encourages women to eat more not dren found that less than half of mothers less, and facilitates doing so through the supple- (42 percent) consumed the whole sachet at the mentary feeding program. Table G.16 shows that CNC, and one-third (32 percent) admitted hav- women receiving supplementary feeding were ing shared it with someone else. more likely to eat more than those not receiving Evidence on substitution comes from the it. However, 32 percent of women receiving sup- question regarding whether mothers ate more plementary feeding ate the same or less during during pregnancy or not. It is relatively common pregnancy, suggesting that they were substitut- practice to "eat down" (that is, eat less) during ing the food they received for their normal diet. 1 6 4 A N A L Y S I S O F B I N P ' S C O M M U N I T Y - B A S E D N U T R I T I O N C O M P O N E N T Determinants of Knowledge-Practice Gap During Pregnancy T A B L E G . 1 4 (BINP data) Food intake Rest Hard work Coeff. t stat. Coeff. t stat. Coeff. t stat. Gap equation Mother's age at marriage ­0.01 ­0.92 0.03 3.08*** 0.03 2.99*** Mother is head of household 0.04 0.21 ­0.17 ­1.09 ­0.39 ­2.37** Mother is widow or divorced 0.55 2.03** 0.27 0.87 0.10 0.38 Children alive 0.08 4.93*** 0.14 8.81*** 0.14 8.15*** Housewife ­0.02 ­0.19 ­0.15 ­1.15 ­0.42 ­4.49*** Land owned 0.01 0.30 0.01 0.87 0.01 0.39 Farming 0.08 1.94* 0.06 1.24 0.24 5.15*** Poor household 0.21 2.64** 0.25 2.84*** 0.12 2.07** Member of community-based income- ­0.10 ­2.10** ­0.05 ­1.17 ­0.01 ­0.13 generating group Vegetable garden ­0.13 ­2.68*** ­0.02 ­0.42 0.14 2.96*** Pregnant in working season 0.14 2.17** 0.20 3.35*** 0.26 2.91*** Project area ­0.55 ­2.92*** ­0.04 ­0.53 ­0.14 ­1.81* Poor in project area 0.03 0.37 ­0.13 ­1.62 ­0.21 ­2.12** Pregnant in working season in project area ­0.22 ­2.72*** ­0.01 ­0.15 0.25 3.46*** Constant 0.17 0.68 ­1.49 ­6.71*** ­0.81 ­4.56*** Knowledge equation Mother's age 0.01 0.82 0.01 2.24** 0.01 0.58 Mother's education 0.05 5.89*** 0.05 6.36*** 0.05 5.04*** Father's education 0.02 2.54** 0.01 1.25 0.02 2.54** Project area 0.96 17.90*** 1.03 18.04*** 0.59 6.43*** Constant 0.08 0.57 ­0.24 ­2.32** 0.63 4.43*** Observations 4,967 4,967 4,967 F statistic 27.6 30.63 17.9 P­value 0.000 0.000 0.000 Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. It is of course possible that the women re- datasets. These are the endline evaluation of ceiving the supplement did not eat more as they BINP by Karim and others (2003), the evaluation gave the good away (leakage) rather than sub- conducted by Save the Children (2003), and stituting the food for their ordinary diet. But the IMED (Haider and others 2004). The results of Save the Children data suggest this cannot be these studies are summarized in table G.17.11 entirely so: of the 68 percent of women who did The BINP evaluation conducted by Karim and not share their BINP food, 63 percent said they others (2003) compares project outcomes of six ate the same or less, thus providing clear evi- upazillas where the BINP project was imple- dence of substitution. mented, and two upazillas where the project was not implemented (see the technical note, avail- The Nutritional Impact of able on request, for a review of the quality of these data). The data were collected in three BINP Interventions rounds: at the onset of project implementation, Child Nutrition10 at a midterm point and at the end. There are sev- Three evaluations of BINP have been conducted eral problems with these surveys that limit their so far that compare project impact on child nu- use for empirical analysis. First, the sample size tritional outcomes using large household survey of the baseline survey is very small, and the qual- 1 6 5 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? thanas do not represent a sample of household Sources of Supplementary T A B L E G . 1 5 "without the project."13 Finally, the control Food (percent) samples are particularly small, especially at the baseline. Source Children Pregnant women Save the Children collected data between CNC (feeding sessions) 75.3 49.6 March and May 2002 in three of the first-phase Pushti Apa (CNP) brings it to the home 16.8 28.2 BINP thanas, and in three control thanas. Somebody from the household brings it from the nutrition center 7.7 22.1 According to Save the Children, these control Take/took it preparing by own n.a. 0.1 thanas do not have serious problems with "con- Total 100 100 tamination" and lack of regional representative- Number of observations 1,710 2,643 ness, as did the BINP control group. However, the control sample is very small, and there is the possibility of seasonal bias, which cannot be as- ity of the data is not very good.12 Second, the sessed since dates of anthropometric measure- questionnaires used in the three rounds are ments are not reported in the dataset. Finally, different, and similar questions are often non- the IMED study collected data from 12 project comparable. Third, no longitudinal data across upazillas from all 3 phases of BINP, and 12 con- villages were collected. This set of problems re- trols selected from neighboring upazillas. Both duces the scope for analyzing changes in out- studies have to rely on single difference esti- comes over time. There are also problems that mates of project impact, and so depend on limit the use of the cross-sectional aspect of strong assumptions regarding the quality of the these data. First, anthropometric measurements control. of children were taken at different times of the According to the BINP evaluation data, there year. Given the high seasonality of child malnu- was no difference between project and control trition in Bangladesh, this is a potential source of areas for the mean of any of the three z-scores at bias. Second, while the sample of project thanas the baseline. There is a significant project effect is representative of all Bangladeshi divisions, the on weight for age at the midterm. Height for age, sample of control thanas is not. If there are sub- however, remained unchanged, resulting in a stantial differences in nutrition determinants be- very large effect on weight for height. At the end- tween regions, comparisons of outcomes can be line, the data show a significant project effect on biased. Third, some of the selected control height for age. However, it is not clear how this thanas may have benefited from nutrition pro- could happen, since the project did not succeed jects of other NGOs. In one of the thanas, BRAC in improving weight for age for the same chil- (the implementing NGO of the first phase BINP dren. The endline difference in weight for age is thanas) operated nutrition activities in the same a combination of the difference in weight for age period as the BINP was implemented (Karim and and height for age. Since heights are increasing others 2003). There is a risk that the control and weights are stable, weight for height de- creases. Eating Habits and The Save the Children data show no signifi- T A B L E G . 1 6 Supplementary Feeding cant differences between the project and con- trol. The IMED data show not much difference Eating habits during either, though there is a slight difference in se- pregnancy (%) vere stunting, partly offset by higher moderate Number stunting in project areas.14 This finding is consis- Ate more Ate same or less of observations tent with the BINP evaluation, which also sug- Received supplementary feeding gested that some children were moved from Yes 67.6 32.4 318 severely to moderately stunted. No: project area 61.3 38.7 762 In summary, these studies find either no pro- No: control 24.6 75.4 353 ject effect, or where there is one, it appears rela- 1 6 6 A N A L Y S I S O F B I N P ' S C O M M U N I T Y - B A S E D N U T R I T I O N C O M P O N E N T tively small.15 There are two problems con- BINP Project Impact from nected to the estimation of project effects pre- T A B L E G . 1 7 Other Studies sented above: (1) the possible poor quality of the control and (2) the estimated effects are av- Number of erage population effects that conflate two issues, observations project participation and effect on those who Project Control Difference Project Control participate. This study improves on these esti- mates of project effects by using national data to BINP evaluation create a comparison group, and by separating Baseline participation (analyzed above) from impact on HAZ ­2.36 ­2.49 0.13 403 153 participants. WAZ ­2.30 ­2.24 ­0.06 425 161 The comparison group for this study is con- WHZ ­0.77 ­0.73 ­0.04 393 150 Midterm structed based on the characteristics of the child HAZ ­1.98 ­1.89 ­0.08 3,488 1,147 and of its household of origin using data from WAZ ­2.04 ­2.14 0.09*** 3,502 1,149 Helen Keller's Nutritional Surveillance Survey WHZ ­0.98 ­1.19 0.21*** 3,520 1,149 (Box G.1), using the method of propensity score Endline matching (see Appendix 3). Project effects are HAZ ­1.90 ­2.07 0.17*** 2,554 837 WAZ ­1.87 ­1.94 0.07 2,567 842 estimated only for the BINP evaluation samples WHZ ­0.84 ­0.76 ­0.08** 2,548 837 of midterm and endline children. For several reasons a similar analysis cannot be applied to Save the Children the children sampled by the Save the Children HAZ ­1.73 ­1.78 0.05 1,640 817 survey.16 WAZ ­1.87 ­1.92 0.05 1,640 817 NSP data were used only from the rounds that WHZ ­1.04 ­1.06 0.03 1,640 817 correspond to the time of collection of the IMED (0­59 months) midterm and endline BINP surveys. This restric- WFA: severe 14.0 14.0 0.0 3,024 tion is necessary given the high sensitivity of nu- WFA: moderate 13.0 12.9 0.1 3,024 tritional indicators to seasonal variations in HFA: severe 7.0 8.0 ­1.0 3,024 Bangladesh. The rounds used are therefore HFA: moderate 7.7 7.4 0.3 3,024 the October/November and December/January Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. rounds for the midterm evaluation, and the December/January and the February/March rounds for the endline evaluation. These data specification of the selection model, including are complemented with the data from the com- their definition in the two datasets used (the parison sites that were collected for the BINP BINP evaluation and the NSP data). The second evaluation. There are 7,091 control observations column describes how the variables from the two against 3,000 project observations at the datasets compare. The third column explains the midterm, and 6,247 control observations against inclusion of the variable in the model when the 2,269 project observations at the endline.17 reason is not obvious. The results from these To create a match, it is necessary to estimate equations were used to estimate the propensity the propensity to participate using a model that: scores for the purposes of matching.18 (1) avoids explanatory variables influenced by Table G.19 shows the estimated average pro- the project, (2) includes all possible observable ject effects and the effects of the project on the determinants of participation, and in particular participants. The effect considered is the differ- those that are influencing both outcomes and ence in the mean z-score of the three nutritional participation. It was shown above that various indicators for children aged 6 to 23 months, be- factors influence participation. Model specifica- tween project area and non-project areas. tion is more constrained here, since only vari- Average treatment effects are calculated using all ables also available in NSP can be included in the project observations, independently of project model. Table G.18 lists the variables used in the participation. It is possible to imagine that even 1 6 7 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? for age give similar results. On the contrary, the The Helen Keller BINP evaluation shows a positive difference in B O X G . 1 International Nutritional weight for age but not in height for age at the Surveillance Project (NSP) midterm, and a positive difference in height for age but not weight for age at the endline, al- In 1990, the NSP started collecting data on health and nutrition of though it is not clear why, given that height for mothers and children of Bangladesh. Data are collected bi-monthly in age scores are the cumulated results of weight "rounds" corresponding to six Bangladeshi seasons. At each round, for age scores. The results also differ from the the data collection takes place in six or seven weeks, from roughly Save the Children evaluation in that there is a 10,000 rural households. The household is interviewed only if it con- project effect on weight for age at the endline, tains at least one physically able child under five years of age (aged though this effect is very weak. between six and 59 months prior to 2000), and if the mother is avail- These results can be summarized as follows: able. Starting from February 1998, the data are nationally representa- tive. Four thanas were randomly selected from each administrative · There is evidence of a project effect for all nu- division, which makes a total of 24 thanas. At each survey round, tritional indicators at the midterm, and for 15 mauzas (10 prior to 2000) are randomly selected with probability weight-for-age at the endline. proportional to size from each thana. One village is randomly selected · The size of the effect is small (negligible in from each mauza (administrative unit within a thana), and 25 house- the case of the endline data), a reduction of holds are systematically sampled from each village. Thus, while the 0.1 z-scores is equivalent to a 3­4 percent re- 24 thanas that were selected in 1998 are always the same, the villages duction of malnutrition rates. and the households interviewed differ at every round. The data con- · There is evidence that the project perfor- stitute a series of repeated cross-sections. The dataset contains mance was worse at the endline with respect 49 variables, including location variables, household and child char- to the midterm. acteristics and anthropometric measurements. Table G.20 presents the estimated projects ef- fects corresponding to different levels of project participation.20 There is evidence of a project ef- children not directly participating in the project fect only for children who participated regularly activities are nevertheless project beneficiaries, in the project. as was indeed shown in the "knowledge equa- Tables G.21 and G.22 show project impact ac- tions" reported above. But it was also shown that cording to nutritional status.21 Children measur- there was a bigger impact on knowledge from ing between ­3 and ­2 z-scores are normally fuller participation. Hence, the average treat- considered malnourished. Children scoring less ment effects are included only for completeness, than ­3 are normally considered severely mal- and to offer a comparison with the project ef- nourished. The category of the severely mal- fects on the project participants.19 The rows "av- nourished children is split into those who score erage treatment effect for the treated" show the between ­3 and ­4, and those who score less project effects only for the children who regu- than ­4. A z-score of ­4 is approximately the cut- larly attended the growth monitoring sessions. off point used to define the third-degree level of As expected, table G.19 shows, in general, malnutrition in the Gomez classification. This greater project effects compared with the aver- category is interesting because children scoring age treatment effect. less than ­4 have much higher mortality risk These results differ from the BINP evaluation than other children, and because the project is results (reported in the first column of "average targeting children of third degree of Gomez mal- treatment effect") in two ways. First, there is a nutrition for food supplementation. Figure G.3 much larger project effect at the midterm, but provides a visual representation of the same dis- no project effect at the endline. Second, these tribution of z-scores. Z-scores in healthy popula- results are more coherent, since at each point in tions have a distribution that is very close to the time they show that weight for age and height normal distribution. The percentage of children 1 6 8 A N A L Y S I S O F B I N P ' S C O M M U N I T Y - B A S E D N U T R I T I O N C O M P O N E N T T A B L E G . 1 8 Definition of the Variables Used in the Model Variable Specification Notes Child age Age is in completed months. Child sex Gender of the child is 1 for female. Child is breastfed Whether the child is breastfed at the time of the interview. Children over 6 months only, and there- fore is independent of the project, which recommends exclusive breastfeeding for children under 6 months. Season of birth of The season in which the child was born obtained using date of birth and This is a weak correlate of low birth the child date of interview. weight, which in turn is a determinant of future nutritional status. Mother's age Age in completed years. Female-headed The mother is head of the household. household Widow or divorced Whether the marital status of the mother is widow or divorced. Mother's height Mother's height measured in centimeters. This predicts child's height. Largely miss- ing before 1998. Mother is pregnant Whether the mother is pregnant at the time of the interview. Pregnant mothers are specifically targeted by the project. Household size This is the number of people who live together and take food from the same pot (NSP). The definition used by BINP is not known. Number of children Number of children under five. Prior 2000 only the number of children under five aged between 6 and 59 months was recorded. Occupation of head NSP: Occupation of the person that provides the majority of household We assume that the main earner and of household income. BINP: occupation of the head of household. BINP midterm and head of household are the same endline surveys used different categories Mother's education Four levels of education are defined: illiterate, primary (including uncom- This is a correlate of wealth. pleted), secondary (including uncompleted) and higher. The variable is orig- inally the number of completed years of schooling in both NSP and BINP. Father education Same as for the mother. House This is the area size of the house where household lives. It is measured in square feet based on measurements of width and length. Land This is the land owned by the household in decimals. Land that is rented, sharecropped, and homestead garden are not included. Landlessness Household with no land. Without well Household without a tubewell from which obtaining drinking water (so use pond or river water or rain water). Without latrine Households without an open or closed latrine. These households dispose of excreta in rivers or in the bush. 1 6 9 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? T A B L E G . 1 9 Project Impact Using Propensity Score Matching One-to-one Nearest neighbor BINP means Regression matching matching Kernel matching Weighting Midterm: average treatment effect HAZ ­0.09 0.09*** 0.08*** 0.10*** 0.10*** 0.08*** WAZ 0.10** 0.13*** 0.14*** 0.14*** 0.14*** 0.13*** WHZ 0.21** 0.14*** 0.16*** 0.13*** 0.13*** 0.12*** Midterm: average treatment effect on the treated HAZ ­0.09 0.11*** 0.10*** 0.10*** 0.11*** 0.10*** WAZ 0.10** 0.14*** 0.12*** 0.13*** 0.14*** 0.14*** WHZ 0.21** 0.13*** 0.10*** 0.12*** 0.12*** 0.13*** Endline: average treatment effect HAZ 0.17*** 0.01*** 0.01 0.01 0.01 0 WAZ 0.07 0.01*** 0.01 0.02 0.01 0.01 WHZ ­0.08** ­0.01* 0.01 0.01 0.01 ­0.02 Endline: average treatment effect on the treated HAZ 0.17*** 0.02 0.08** 0.03 0.03 0.01 WAZ 0.07 0.04* 0.09** 0.06* 0.03 0.03 WHZ ­0.08** 0.02 0.03 0.04 0.03 0.01 Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. below ­2 z-scores is normally 2.3. The vertical reduces the number of severely malnourished lines in the charts help to assess the difference children (less than ­4 z-scores), which in turn can in malnutrition rates between project and non- be found in excess in the category of malnour- project areas. ished between ­4 and ­3 z-scores, as many of The shapes of the midterm and endline distri- them have been upgraded to this category. butions are very similar, although project effects at the endline are much smaller. For all nutri- Supplementary Feeding Program tional indicators the project has the effect of shift- As noted above, it proved difficult to construct a ing the distribution to the right. In particular, it control for the supplementary feeding program reduces the number of malnourished children since we do not have data on growth faltering between ­3 and ­2 z-scores, and increases the among the control, and the determinants for the number of normal children. In addition, the selection equation seemed not to work well, so weight for age charts clearly show that the project the problem of endogeneity of program place- ment remained, resulting in an apparent negative program impact. The best source for analysis is Project Effect by Level the Save the Children register data, which in- T A B L E G . 2 0 of Participation cludes the weight for age z-scores of all children who had received supplementary feeding, both Midterm Endline during the feeding and at other times. The re- Irregular Regular Irregular Regular gression results for the change in the WAZ between starting and ending feeding are shown in HAZ ­0.06 0.08*** 0.03 0.01 WAZ 0.05 0.13*** 0.02 0.03 table G.23. WHZ 0.07 0.13*** ­0.02 0.02 The main results are that supplementary Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. feeding is much more effective for children who 1 7 0 A N A L Y S I S O F B I N P ' S C O M M U N I T Y - B A S E D N U T R I T I O N C O M P O N E N T are malnourished, and particularly those who Midterm Malnutrition Rates are severely malnourished. The results also pro- T A B L E G . 2 1 by Nutrition Status vide evidence of the adverse effects of substitu- tion. Finally, there are very striking seasonal HAZ WAZ WHZ effects from the weight gain experienced during supplementary feeding. Z­scores Project Control Project Control Project Control The lean season in Bangladesh, as deter- Z > ­3 & < ­2 27.6*** 31.9 36.8** 39.3 14.8* 16.6 mined by WAZ for children under five using the Z > ­4 & < ­3 13.7 12.6 16.4 15.3 2.3** 1.4 NSP data, runs from April to November, with the Z < ­4 5.2 4.9 1.6*** 3.2 0.3*** 0.1 start and end months varying from year to year Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. (see Annex J). Children receive supplementary feeding for 3-4 months, so four categories can be to 0.3 WAZ for older children. However, as identified for the start-end points shown in figure G.4, the NSP data show that the WAZ falls for children across Bangladesh at a rate · Good-Lean, starting feeding in January to of about 0.5 WAZ over a three-month period. March, and in some years, starting in April or This occurs since the curves in the growth mon- May: the coefficients show these children ex- itoring chart are very steep for children aged perience no, or negative, weight gain from below 12 months (figure G.5), so that substan- the seasonal effect, picking up the effect of re- tial weight gain is needed to not fall further duced food intake in these months below the curve. Hence, compared to this refer- · Lean-Lean, starting feeding in June to August, ence group, a zero gain of WAZ is actually a gain and in some years early September: the coef- of 0.5 WAZ, which is not insubstantial. ficients show a greater weight gain for chil- Finally, table G.25 reports regressions using dren starting and finishing feeding in the lean the BINP endline data for WAZ and midterm and season. HAZ at endline, these measures being selected · Lean-Good, starting feeding September to since they are the ones for which a significant November, and in some years August: as ex- impact appears in the double difference and pected there is a positive weight gain for PSM analyses. A project dummy is used, inter- these children, partially reflecting greater acted with various characteristics. Program ac- food availability at the end of the period, this tivities cannot be used as they would suffer from effect is greatest for children starting Novem- selection bias (they are indeed found to have a ber as it takes time for the effect of increased negative effect when introduced into the regres- food availability to feed through to improved sion). The project dummies are significant, WAZ. though not large. The project mutes the advan- · Good-Good, starting feeding in December tage of more educated women, that is, nutri- and January, and in some years November: tional counseling is substituting for education, the coefficients for November and December and to a lesser extent offsetting the advantages are high reflecting the cumulative effect of three months of better food supply. The regression results can be used to calcu- Endline Malnutrition Rates T A B L E G . 2 2 late the weight gain during supplementary feed- by Nutrition Status ing for a child with average characteristics. Table G.24, shows the results, varying only the charac- HAZ WAZ WHZ teristic shown. Children who enter the program Z­scores Project Control Project Control Project Control more malnourished gain more from the pro- Z > ­3 & < ­2 29.6 30.2 33.9*** 35.1 14.7 13.1 gram than better nourished children.22 Children Z > ­4 & < ­3 11.0 11.5 12.4 12.3 1.9 1.6 under 12 months appear to show negligible gain Z < ­4 3.1 3.0 1.4 1.7 0.0 0.0 in WAZ during the program, compared to close Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. 1 7 1 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? F I G U R E G . 3 Distribution of Z-Scores in Project and Non-Project Areas Midterm HAZ Midterm WAZ Midterm WHZ .4 .5 Without the project .4 Without the project Without the project .4 .3 .3 .3 .2 .2 Density Density Density.2 .1 .1 With project With project With project .1 0 0 0 ­6 ­4 ­3 ­2 0 2 4 6 ­6 ­4 ­3 ­2 0 2 4 6 ­6 ­4 ­3 ­2 0 2 4 6 HAZ WAZ WHZ Endline HAZ Endline WAZ Endline WHZ .4 .4 .4 Without With project Without the project Without the project the project .3 .3 .3 .2 .2 .2 Density Density Density .1 .1 .1 With project With project 0 0 0 ­6 ­4 ­3 ­2 0 2 4 6 ­6 ­4 ­3 ­2 0 2 4 6 ­6 ­4 ­3 ­2 0 2 4 6 Height-for-age z-score Weight-for-age z-score Weight-for-height z-score of higher economic status. It also reduced the Concepts and Definitions "daughter-in-law" disadvantage. Low birth weight--that is, below 2,500 grams-- is determined by two factors: gestational age at Low Birth Weight birth and intrauterine growth (WHO 1995). In At 40 percent, low birth weight incidence in the first case, the infant has a low birth weight Bangladesh was one of the highest in the world in because he or she was born prematurely, which the 1990s (Mason and others 2001). The social does not imply a failure to grow during preg- and economic consequences of low birth weight nancy. In the second case, the infant has a low are well documented (see, for example, Alderman birth weight because its intrauterine growth was and Behrman 2003). Hence, among BINP's ob- impaired by the health and nutrition history of jectives is "reduction of low birth weight occur- the mother (Intrauterine Growth Retardation, rence by half of the baseline level," which is to be IUGR). The two types of low birth weight should achieved through the "improvement in maternal be considered separately because they have dif- weight gain by at least 50 percent in at least ferent determinants and different consequences 50 percent of the pregnant women"23 (World Bank for child health. Premature birth is not affected 1995). These goals are to be achieved through nu- by malnutrition, and infants born prematurely trition education and food supplementation. All have better chances of surviving and catching up pregnant women of project villages are regularly in infancy and childhood than those children weighed and counseled on the proper diet to fol- with low birth weight resulting from IUGR. In low in order to gain the necessary weight. Women order to isolate birth weight from premature identified as malnourished are given a daily food birth effects, infant weight can be measured supplement from the first screening to the sixth against the weight for gestational age of a refer- month of lactation after birth. ence population. 1 7 2 A N A L Y S I S O F B I N P ' S C O M M U N I T Y - B A S E D N U T R I T I O N C O M P O N E N T Gestational weight gain has four compo- Determinants of nents: fat stores, breast and uterine growth, vol- T A B L E G . 2 3 Weight Gain During ume of plasma, and fetus (Kramer 1987). Supplementary Feeding Mother's fat stores are the only component of weight gain that has an influence on the growth Variable Coeff. t stat. of the fetus. But mothers have different fat stores Female child 0.07 1.73* levels before pregnancy, which has two impor- Month at which supplementary feeding started tant implications: (a) heavier mothers need February 0.11 1.12 lower weight gains during pregnancy, and March ­0.10 ­1.11 (b) the effect of additional food intake has a April ­0.06 ­0.60 larger effect on infants of mothers that are thin- May ­0.01 ­0.18 ner before pregnancy. The Institute of Medicine June 0.20 2.12** July 0.17 2.16** (1990) has produced charts of optimal gesta- August 0.09 1.12 tional weight gain for women of different BMI, September 0.10 1.04 based on a sample of U.S. women who delivered October 0.05 0.69 infants of normal weight. These charts recom- November 0.17 1.73* mend a weekly gain of 0.4 kilograms during the December 0.19 2.31** Age at which supplementary feeding started second and third trimester for woman of normal From 12 to 17 months 0.25 7.06*** pre-pregnant BMI, and of 0.5 kilograms for From 18 to 23 months 0.28 4.76*** women who are underweight. Nutritional status at start of supplementary feeding It is common in Bangladesh for women to re- Normal child (above ­2 z-score) 0.24 1.05 duce food intake during pregnancy, a practice Between ­2 and ­4 z-scores 0.57 2.51** often referred to as "eating down." This practice Below ­4 z-score 0.91 3.70*** Child shared the food with other household members 0.00 ­0.11 is well documented for India and is explained by Child was not given other food than supplementary ­0.07 ­1.91* the mothers' intention to reduce infant body food at meal time size, thus facilitating delivery.24 CARE (Vemury Child was ill during supplementary feeding ­0.02 ­0.46 1981) conducted a survey study on food habits Child missed some of the sessions 0.00 ­0.74 in six developing countries and found that some Constant ­0.52 ­2.27** R-squared 0.21 60 percent of the 700 Bangladeshi women inter- Observations 716 viewed believed that pregnant women should Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. eat less during pregnancy. The reasons given were the prevention of difficult delivery and cus- size by restricting maternal food intake was an tomary beliefs (see also Hossain and Choudury understandable goal." Maternal mortality at de- 1987). The project document and Bangladeshi livery is very high in Bangladesh (estimated in women thus share the belief that infant body the range of 320 to 400 per 100,000 live births for size can be changed during pregnancy by eating the period 1998 to 200125). And nearly 80 per- more or less than usual. They do not have the cent of these deaths are due to obstetric causes, same view, however, on the delivery outcome. Larger infants, according to the project, do not create an additional risk at delivery. However, Weight Gain from the review of knowledge on nutrition during Supplementary Feeding T A B L E G . 2 4 pregnancy done by the Institute of Medicine for Different Categories (1990) documents that in Western countries be- of Children fore World War II, women were commonly ad- vised during pregnancy to avoid excessive Severely Very severely weight gain or to restrict their food intake, con- Malnourished malnourished malnourished cluding that "in a period when maternal mortal- ­0.11 0.22 0.56 ity was extremely high and cesarean deliveries < 12 months 12­17 months 17­23 months were a desperate alternative, limitation of fetal 0.02 0.27 0.29 1 7 3 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? F I G U R E G . 4 BINP Growth Monitoring Chart 1 7 4 A N A L Y S I S O F B I N P ' S C O M M U N I T Y - B A S E D N U T R I T I O N C O M P O N E N T F I G U R E G . 5 Mothers' Weight Gain by Month of Gestation Midterm Endline 15 15 Reference Reference (kg.) 10 (kg.) 10 Project gain gain Control Control 5 5 Weight Project Weight 0 0 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 Month of gestation Month of gestation like obstructed labor, abortion, and eclampsia Table G.26 shows the difference, and the dif- (Faveau and others 1988). Sixty percent of ference in difference, in the percentage of low Bangladshi women reported a complication dur- birth weight and in the average newborn weight ing pregnancy, delivery, or after delivery (NIPRT between project and control areas. Data on birth 2002). Yet, very few women in Bangladesh de- weight were not available at the baseline and liver in hospitals or are assisted by professionals. therefore the table only contains the differences The belief of "eating down" is thus understand- between the midterm and the endline. Low birth able, and its effects on delivery are in fact not weight is defined as a birth weight below 2.5 fully understood. However, some studies kilograms. The ratio of birth weight over the me- (quoted in Pojda and Kelley 2000, p. 24) report dian weight of a reference population of the that increases in head circumference due to same sex and gestational month is used as mean food supplements does not increase mortality weight score.27 risk from obstructed labor.26 There was a reduction in low birth weight in- cidence in project areas from 26 percent Project Impact (midterm) to 16 percent (endline). If a similar This section uses the data collected for the BINP reduction had been observed between the base- evaluation at midterm and endline. The baseline line and the midterm, the stated project objec- survey did not collect information on birth tive of reducing the proportion of low birth weight and pregnancy weight gains. The mid- weight by 50 percent would have been more term and endline surveys are cross-sections than achieved. This result seems impressive, using different samples for children, newborns, even just considering the change between the and mothers anthropometric measurements. midterm and the endline. However, only a small Two problems arise when comparing the nutri- part of this decline can be attributed to project tional outcomes of project and control areas. activities: there was an even larger decline in the First, not all women interviewed in BINP areas incidence of low birth weight in the control participated in project activities. Second, project areas. Table G.26 shows a difference of 4 and and control areas may differ in some important 2 percentage points between project and con- determinants of nutritional outcomes indepen- trol areas at the midterm and the endline re- dently of project implementation. The second spectively, indicating a positive effect of the problem is handled through the difference and project assuming the control is good. But the dif- double difference approaches (see appendix 1). ference in difference over time and areas is pos- 1 7 5 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? itive (though not statistically significant) since the model include mother's height and pre- the rate of reduction in low birth weight has pregnancy weight. been higher in control areas than in project The month of measurement is included in the areas. model to detect seasonal effects on birth weight. The bivariate analysis shows no difference in The effect is strong and the pattern clear. Birth weight gain. But this analysis does not allow for weight is higher in the included months, possible differences in determining characteris- December-March. However, there is a large dif- tics between the project and control, for which ference in the effect of the seasonal variable be- multivariate analysis is needed. The explanatory tween the two surveys, which indicates that the variables used are those identified in the litera- model should be better specified. The impli- ture as main determinants of birth weight (e.g., cation is that mother's nutrition is critical in Kramer 1987). In addition, seasonality is added the last months of pregnancy. However, the sea- as a potential determinant of birth weight as are sonal effect is much less relevant in project areas project inputs. Not all relevant explanatory vari- (shown by interacting the seasonal variables ables are available in the BINP data, and the with the project dummy, results not shown midterm and endline survey used different ques- here). The project tends to smooth out the sea- tionnaires. Thus, the list of regressors is incom- sonal pattern of mother's nutrition. The impli- plete and different across surveys. The results cation is that if the data had been collected are shown in Table G.27. during the "hungry" season they would have Gestational age at birth and the sex of the probably shown a positive impact of the project child have the expected values and are highly on birth weights. significant. Using the mean birth score over the The model includes a variable taking the reference population values gives very similar value of one if the mother says she ate more than results. Newborns from primiparous women usual during pregnancy, which is one channel have lower birth weight, which is a common re- for project impact if women change this practice sult in the literature, though a clear explanation in response to being in the project area. Women for this finding does not exist. Mother's age is may have eaten more during pregnancy in the normally a determinant of birth weight. project areas both as a result of the availability of Pregnancy outcomes are usually less favorable food supplementation and in response to the for very young and very old women (Kramer nutritional education against eating down. 1987). The coefficients confirm this finding, Whether the woman received food supplemen- though are only significant in the pooled regres- tation was also included in the model, but was sion; in which case the turning point is 32 not significant. This result presumably reflects years.28 Illness of the mother during pregnancy attrition from leakage and substitution (see ta- (available for endline only) has the expected bles G.15 and G.16)--it is only if the woman negative sign, but the coefficient is not statisti- actually eats more that matters, which some cally significant. A woman who experienced mis- women receiving food supplements do, but a carriage, abortion, or stillbirth is more likely to sizeable number do not. "Eating more" increases have a child of lower weight. Among the socioe- birth weight by 44 grams in the midterm survey conomic variables used, only father's education and by 88 grams in the endline. The effect is and income have an effect on birth weight. Birth highly significant in both cases. The effect of weight increases with household income and "eating down" (mothers stating they ate less education of the father. Land ownership also has than usual during pregnancy) is also significant a positive and significant effect, but the value of and negative, reducing birth weight by 45 grams the coefficient is negligible. The R-squareds are (only available for endline). The effect of a low in particular for the midterm model, which mother changing from eating down to eating contains only a small number of regressors. more is thus 133 grams. An increase of just over Important explanatory variables missing from 80 grams from eating more is not very large, but is similar to the effect on birth weight observed 1 7 6 A N A L Y S I S O F B I N P ' S C O M M U N I T Y - B A S E D N U T R I T I O N C O M P O N E N T in other projects where pregnant women were Determinants of the given food supplements.29 In addition, the effect T A B L E G . 2 5 WAZ and HAZ with is larger for women of poorer nutritional status. Interactive Terms The interaction of the "eating more" dummy with the income level of the household of resi- Midterm Endline dence (results not shown) produces a much Coeff. t stat. Coeff. t stat. larger coefficient (+ 270 grams) for the "desti- tute" women. Father's education 0.01** 1.98 0.02*** 4.05 BINP may have affected birth weight through Land 0.00 1.40 0.00 0.01 Female head of household ­0.12 ­1.18 0.29* 1.92 other channels, such as the quality of diet. These Water 0.13*** 4.82 0.05 1.46 channels are captured by the project dummy in Sanitation 0.02 0.68 0.08** 1.94 the midterm and endline equations.30 However, Female child 0.02 0.62 0.08** 2.06 this dummy is negative in both cases, though in- Child age ­0.06*** ­21.00 ­0.06*** ­15.14 significant at endline. But interpreting this neg- Household size ­0.02*** ­3.14 0.00 0.41 Birth order ­0.05*** ­3.63 ative coefficient as an adverse project impact Birth interval 0.00*** 3.85 means assuming that all unobserved determi- Mother's height 0.04*** 14.75 0.05*** 14.39 nants of birth weight were the same in the pro- Project area 0.15** 2.50 0.18 1.62 ject and control areas prior to the project. Primary education 0.18** 1.96 ­0.07 ­0.64 Pooling the data, and so estimating the double Secondary education 0.20*** 2.75 0.10 1.12 Higher education 0.01 0.03 0.30* 1.85 difference effect, allows us to control for these Primary education*project area ­0.15 ­1.48 0.10 0.80 unobservables. In the pooled equation, the co- Secondary education*project area ­0.06 ­0.82 0.02 0.15 efficients on the BINP area dummy are now un- Higher education*project area 0.39** 2.15 0.04 0.24 explained area differences, with the project 2nd wealth quantile 0.17** 2.36 0.18 1.61 3rd wealth quantile 0.26*** 3.56 0.36*** 3.21 effect shown by the difference in difference co- 4th wealth quantile 0.25** 2.32 0.42*** 3.53 efficient (see Appendix 1). This coefficient is 2nd wealth quantile*project area ­0.10 ­1.15 ­0.05 ­0.41 negative but relatively small and very insignifi- 3rd wealth quantile*project area ­0.19** ­2.34 ­0.18 ­1.43 cant. There is a large and significant time effect 4th wealth quantile*project area 0.00* 0.00 ­0.17 ­1.25 on birth weight in both areas, where birth Early marriage ­0.15** ­2.03 Early marriage*project area 0.11 1.28 weight has increased on average by 100 grams, Daughter in law ­0.01 ­0.13 indicating a large autonomous change, most Daughter in law*project area 0.12 1.00 likely as a result of improving mother's nutri- Age difference ­0.13 ­1.52 tional status. Age difference*project area 0.10 1.02 The total project effect is calculated by com- Seasonal dummies (not shown) Constant ­7.79*** ­18.45 ­9.28*** ­17.39 bining both the effect through mother's eating R­square 0.19 0.19 patterns and the project effect coefficient (the Observations 4,753 3,369 BINP dummy in the midterm and endline equa- Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. tions and the difference in difference coefficient for the pooled data). The results from the pooled regression suggest an increase in birth weight is not large at all. What explains this rela- weight of just over 80 grams through eating tively weak performance? more (on the assumption that a woman does eat more, which not all do), with no other project ef- Project Impact on Pregnancy Weight Gain fects. The figure of 80 grams is thus an upper es- The project aimed to increase birth weight by in- timate of the average effect. This is an average creasing mothers' weight gains during preg- effect, since it was also shown that the impact is nancy. The first hypothesis to explain the greater during the lean season and higher for absence of a reduction in the incidence of low poorer (and therefore probably less well nour- birth weight in project areas compared to con- ished) women. The average impact on low birth 1 7 7 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? trol areas, is that mothers' pregnancy weight did advice of eating more during pregnancy has the not increase as expected. expected positive sign of increasing weight gain, Figure G.5 plots weight gains of project and and this appears to be the main channel through control areas for month of gestation compared which the project has an effect. While the single to a reference population. The reference popu- difference estimate of the pure project effect is lation is the one used by the Institute of positive, it becomes insignificant in the double Medicine (1990) for underweight women, de- difference estimate (in fact negative and of suffi- fined as women of BMI below 19.8, which in- cient magnitude to offset the eating effect). cludes most the women in the sample. The lines Even at the most optimistic estimate, the for the project and control areas are obtained overall project effect is to increase weight gains cumulating median weight gains at each month by not much more than 200 grams per month in of gestation. Few observations are available for comparison to control areas (adding the coeffi- weight gains before the third month of gesta- cient on the project dummy from the single dif- tion, therefore the curves before that date are ference equation to the eating more effect), not reliable. Both graphs show a growth pattern equaling a total weight gain of just over 1 kilo- inferior to the reference population, and no dif- gram over a period of six months.31 Kramer's ference between control and project at the (1987) meta-analysis of the determinants of low midterm, though there is some difference at the birth weight suggests the effect of gestational endline. Similarly, table G.28 shows the differ- weight gain on birth weight is 20 grams per one ences between areas and over time in mean kilogram of total weight gain. Hence, these re- weight gains. Weights were measured at one- sults suggest that birth weight will increase by month intervals for samples of pregnant women just 20 grams as a result of greater pregnancy at different stages of gestation. The table shows weight gain. Weight gains of pregnant women a significant project effect and a positive project in BINP areas would have to much higher in effect over time compared to control areas. order to exert a significant effect on birth However, as in the case of birth weight, these weights. comparisons are of little use unless differences There is, however, an increase over time in in the determinants of weight gain are allowed gestational weight gain (over 300 per month) in for with multivariate analysis. both project and control areas, which equates to Table G.29 presents the results of regressions an increase in birth weight of around 60 grams. of weight gains of pregnant women measured in As mentioned above, this increase is largely a re- grams. The same model is used for the midterm, sult of improved maternal nutrition. For exam- the endline, and the pooled data. Weight gain is ple, mother's weight, which in Kramer's review inversely correlated to women's age, though the increases birth weight by 9.5 grams per kilo- link is not clear (Institute of Medicine 1990). gram, has increased by an average of 3.4 kilo- Weight gains are higher in the second trimester grams between the two surveys. Mother's of gestation and than decrease in the last height, which in Kramer's review increases birth trimester. The variables indicating the month of weight by 8 grams/centimeter, has increased by measurement depict the shape of this function. two centimeters between the surveys. Hence, The data show again the presence of seasonality the difference over time in birth weights de- in the consumption of food. February is the pends in large part on non-project effects. month in which weight gains are larger. Weight It remains to be explained why gestational gains increase with income, and landless house- weight gain has not increased much more in holds are disadvantaged. Weight gains are larger project areas with respect to control areas. Two in project areas, but the rate of change over time reasons are possible: (i) low participation rates, is the same as in control areas. There is a time in- particularly by women most in need and, crease of some 300 grams in monthly gain, (ii) though women participate and receive the which is common to project and control areas. A services, there are constraints that prevent the variable representing the mother following the realization of the expected outcomes. 1 7 8 A N A L Y S I S O F B I N P ' S C O M M U N I T Y - B A S E D N U T R I T I O N C O M P O N E N T Cost Effectiveness Low Birth Weight, Birth Table G.30 shows BINP project costs from 1996 Weights, and Mean through 2003 by main component from the im- T A B L E G . 2 6 Newborn Weight: BINP plementation completion report (ICR). The and Control Areas Community-based Nutrition Component ac- counts for more than 50 percent of total project Low birth weight Birth weight Mean weight costs. (%) (grams) (%) Table G.31 shows the change in child malnu- Differences trition produced by the project as measured by Midterm difference ­4.3* ­34.3 0.2 PSM at the midterm and the endline, and the Endline difference ­1.7 ­30.9 0.4 consequent percentage reductions of malnour- Time effect ­12.4*** 89.1*** 2.3*** ished children.32 Difference in difference 2.6 3.5 0.1 Cost effectiveness estimates are calculated in Mean levels two ways: using total project costs and excluding all costs that are not directly related to children Midterm project 25.6 2677 75.8 Midterm control 29.9 2712 75.6 (table G.32 and G.33; Appendix 4 explains the Endline project 15.9 2770 78.2 basis of these calculations). The figures show the Endline control 17.6 2800 77.8 cost per child to be spent in order to obtain a re- * One percentage point corresponds to approximately 30 grams. duction of malnutrition by 1 percent in the pro- ject areas of the first phase. The children Determinants of Birth considered are between 6 and 23 months of age. T A B L E G . 2 7 Weight (gm) (OLS) Note that costs have almost doubled between the midterm and the endline, because the pro- Midterm Endline Pooled data ject has become much less effective in reducing malnutrition rates. There is a reduction of costs Coeff. t stat. Coeff. t stat. Coeff. t stat. by around 50 percent when funds spent for im- 9th gestational proving mothers' nutritional status are ex- month 104.4 3.01*** 167.9 3.66*** 133.1 4.77*** 10th gestational cluded. month 93.5 2.29** 239.9 4.47*** 154.8 4.72*** In order to estimate the cost of saving the life Female child ­58.0 ­3.18*** ­69.2 ­3.60*** ­61.2 ­4.52*** of a child through the BINP nutritional interven- Parity ­34.2 ­1.32 ­108.5 ­3.63*** ­52.1 ­2.66*** tion, we use a method developed by Pelletier Age 13.4 1.19 14.3 0.91 20.5 2.11** Age square ­0.2 ­0.89 ­0.2 ­0.83 ­0.3 ­1.88* and Frongillo (2003, p. 118). This method esti- Mother ill 19.8 0.63 mates the reduction in mortality rates following Pregnancy history ­188.6 ­2.26** a reduction in the prevalence underweight in Father's education 14.0 4.33*** Mother's education ­1.2 ­0.33 the population. A simple formula relates the Land owned 0.2 2.16** 0.1 3.14*** number of deaths averted to pre- and post-in- Middle income 56.8 2.52** tervention mortality rates. The parameters used Rich 115.7 2.79*** by the formula are obtained from regressions of December 360.6 4.38*** 112.0 2.75*** January 355.7 4.42*** 81.2 0.63 111.0 2.84*** mortality rates on underweight prevalence rates February 422.5 5.25*** 78.5 2.08** 161.3 4.71*** using a panel dataset of 59 developing countries. March 283.5 2.94*** 67.8 1.91* 128.2 3.86*** The results are presented in table G.34 for 1998 BINP area ­70.7 ­2.72*** ­36.6 ­1.32 ­52.2 ­2.14** and 2003 using overall project costs of each spe- More food during 43.2 2.21** 88.1 3.75*** 86.4 6.11*** Less food during cific year. As above, there are two estimates, one pregnancy ­44.5 ­1.74* for all project costs, and another for only costs Time 105.8 3.46*** spent on children. Difference in difference ­12.8 ­0.40 Constant 2,118.6 10.86*** 2,387.9 10.51*** 2,220 15.02*** Cost Comparison with Alternative Interventions R-square 0.05 0.12 0.07 To benchmark the figures reported above for Observations 1,560 1,362 2,922 BINP, two alternative scenarios are presented: Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. 1 7 9 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? simply giving the project budget away as a cash Difference in Monthly transfer, and giving away an equivalent value of Pregnancy Weight Gain T A B L E G . 2 8 rice. The rice distribution is assumed to be to all Between Project and households of project areas with a child aged 6 Control Areas to 23 months, funded with the monthly cost of Grams BINP project in 1998. The cost per household of US$1.80 per month corresponds to 5.6 kilo- Midterm difference 0.5 grams of rice (or 19,720 calories), after allowing Endline difference 211.0*** for 25 percent administrative costs for distribu- Time effect ­137.9 tion.33 The estimate of the percentage of the Difference in difference 210.4** administrative cost over total project cost is de- Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. rived from a review of food supplementation projects (Beaton 1982). Food supplementation programs often do Determinants of not achieve the desired results for two reasons. T A B L E G . 2 9 Pregnancy Weight First, the food distributed can be shared among Gain (OLS) household members, or can substitute rather than complement current feeding. Second, not Midterm Endline Pooled data all calories ingested are transformed into physi- Coeff. t stat. Coeff. t stat. Coeff. t stat. cal growth. Part of the energy obtained through the supplement is dispersed in additional physi- Mother's height 4.5 1.99** cal activity, and in the maintenance of the in- Mother's age ­4.5 ­1.34 ­6.6 ­2.12** ­4.8 ­1.70* creased body size (basal metabolic rate). To 4th gestational month 110.7 1.38 433.9 7.90*** 129.5 2.01** 5th month 548.8 7.15*** 243.7 4.50*** 455.4 7.38*** account for the first problem we assume that 6th month 709.1 9.35*** 332.6 6.08*** 607.6 9.94*** half of the food ingested is substituting rather 7th month 992.9 13.09*** 293.6 5.33*** 849.0 13.88*** then complementing the child's normal dietary 8th month 1027.0 12.79*** 477.4 8.07*** 899.6 13.83*** intake. Only 50 percent of the rice delivered is 9th month 1179.6 10.80*** 713.2 4.47*** 1081.5 11.37*** therefore actually eaten by the child in excess of 3rd and 4th week of his or her previous food intake. To account for January 226.0 2.59*** 219.6 2.67** the second problem, we make the additional as- 1st and 2nd of* sumption that 45 Kcal/day are required for an February 454.8 5.22*** 463.8 5.51** 3rd and 4th of extra weight gain of 1 gram (see Beaton and February 305.5 6.01*** 315.0 6.37*** Ghassemi 1982), rather than the normally used 1st and 2nd of March 103.1 2.18** 107.2 2.32** value of 5 Kcal/day (see Waterlow 1992). 3rd and 4th of March ­85.7 ­1.27 ­26.4 ­0.30 The results are shown in table G.35. Two 1st and 2nd of April ­55.3 ­0.87 30.8 0.28 BINP costs are shown, one is obtained consider- Middle income 75.6 1.89* 85.3 2.14** 68.2 2.00** ing all project costs, while the other uses only High income 127.8 1.95* 73.0 1.01 115.9 2.04** costs directly attributable to children.34 The costs Landless ­97.7 ­2.63*** ­1.6 ­0.04 ­81.0 ­2.53** BINP area 162.0 2.84*** 136.8 3.00*** 156.8 2.92** from the income transfer are also shown for Mother ate more comparative purposes. These costs are far during pregnancy higher, since only a percentage of the income in BINP area 73.8 1.77* 67.4 1.70* 76.1 2.12** will be used for food (the elasticity of food ex- Time 332.0 2.84** penditure is about 0.3), and not all of the incre- Difference in mental food expenditure will benefit children. difference ­74.5 ­0.72 Of course there are other benefits to the income Constant 298.1 2.37** 273.2 0.78 392.4 3.59*** transfer (and households save resources if they R-square 0.06 0.10 0.06 substitute for the rice ration). Observations 7,562 1,433 8,995 The cost of reducing malnutrition with the Notes: *, **, *** significant at greater than 10%, 5% and 1% respectively. rice ration is US$80, compared with US$200­400 1 8 0 A N A L Y S I S O F B I N P ' S C O M M U N I T Y - B A S E D N U T R I T I O N C O M P O N E N T for BINP, and the cost per life saved around Project Costs in US$1,800 compared with US$2,300­5,000 for T A B L E G . 3 0 US$ Million (actuals) BINP. Hence, BINP does not compare favorably with simply spending project resources to buy US$ million % shares rice for children in the project area. The argu- ment against such an approach is that it implies National Nutrition Component (NNC) 18.4 31.7 Community-based nutrition component 32.3 55.6 a permanent subsidy, whereas behavior change (CBNC) should become self-sustaining once affected. Inter-sectoral nutrition program 6.5 11.2 However, the evidence shows limited success in development (INC) behavior change, partly as resource constraints Total baseline cost 57.3 prevent recommended changes from being put Physical contingencies 3.4 5.9 into practice. Moreover, there may be lower cost Price contingencies 5.0 8.7 models of BCC given the pervasive presence of NGOs, women's groups, and health and family Total project cost 65.7 100.0 planning officials,35 plus media saturation, which has proved successful in conveying other mes- sages. Reduction in Malnutrition T A B L E G . 3 1 at the Midterm and Endline Calculated Using PSM Conclusions Child Nutritional Outcomes Midterm Endline Overall project impact on child nutritional out- Z­score % change Z­score % change comes appears disappointing despite high levels Weight-for-age +0.13 ­5.3 +0.05 ­2.0 of participation in project activities. Although par- Height-for-age +0.10 ­4.1 +0.03 ­1.3 ticipation was high it was lower in more traditional thanas, which was partly a result of the project not having a broader target audience encompassing though that of reducing low birth weight was all key decisionmakers. Lack of decision-making not. However, the BINP evaluation dataset power of participants was also a factor behind the shows that there was an even larger increase in knowledge-practice gap, though lack of resources pregnancy weight gain in the control areas than and time played a larger part. There were other in the project areas, suggesting that outside fac- weak links in the chain, notably poor targeting, so tors accounted for these improvements. Greater that some eligible children did not receive feed- food availability has improved mothers' pre- ing, and mothers and some pregnant women did pregnancy nutritional status, as shown by rising not receive nutritional counseling. Cost per Child of Reducing Low Birth Weight Malnutrition Rates by 1% Low birth weight is a correlate of a child's sub- T A B L E G . 3 2 in First Phase BINP Areas sequent nutritional and health status, and so of (US$/% reduction in those mortality. BINP addresses low birth weight below ­2z) through increasing pregnancy weight gain by discouraging eating down and provided food Midterm (1998) Endline (2003) supplements to malnourished women. BINP All project costs aimed to reduce "low birth weight occurrence Weight-for-age 4 7 by half of the baseline level" through the "im- Height-for-age 5 10 provement in maternal weight gain by at least 50 Only child costs percent in at least 50 percent of the pregnant women." The BINP evaluation found that the Weight­for-age 2 3 pregnancy weight gain target was achieved, Height-for-age 2 5 1 8 1 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? ment that lack of income is sometimes the bind- Cost of Upgrading One ing constraint on nutritional practices rather Child from Malnutrition T A B L E G . 3 3 than lack of knowledge. in First Phase BINP Areas (US$) How Well Did the Causal Chain Operate? Midterm (1998) Endline (2003) BINP succeeded in achieving high levels of par- ticipation in its activities in project areas, en- All project costs rolled large numbers in supplementary feeding, Weight-for-age 396 706 and brought about significant increases in nu- Height-for-age 509 1037 tritional knowledge and associated changes in Only child costs practice. However, nutritional outcomes in terms of low birth weight have been disappoint- Weight-for-age 183 327 Height-for-age 236 480 ing. While child nutrition, especially for children participating in supplementary feeding, appears to have been better, the overall difference in per- formance compared to the control is not great. BMI in both project and control areas, and this The longer the causal chain, the more likely it has been the main driving force behind greater is that final outcomes will not be realized on ac- pregnancy weight gain. count of missing or weak links in the chain. There has been a project effect, but it has There were two missing links in the BINP chain: been relatively small. Multivariate analysis shows the first was the relative neglect of some key de- a small project effect on both pregnancy weight cisionmakers regarding nutritional choices (men gain and low birth weight, mainly through dis- and mothers-in-law), and the second the focus couraging the practice of eating less during preg- on pregnancy weight gain rather than pre-preg- nancy (eating down), assisted by providing nancy nutritional status. Participation levels of feeding supplementation to pregnant women the target audience were high, but many women with less than normal body mass index. These ef- escaped exposure to nutritional messages, and fects are greater for more disadvantaged (and so there was a high Type I error in the feeding pro- presumably less well-nourished) women and grams (see table G.36). A knowledge practice during the lean season. gap persisted, so many women did not put the Potential beneficial impacts of the project are advice they received into practice, especially if muted for reasons of poor targeting, failure to they were resource or time constrained. Those amend eating practices even when the message receiving supplementary feeding often shared it is gotten across (the knowledge-practice gap) with others or substituted it for their regular and the leakage and substitution of supple- foodstuffs. This list of weak links in the chain ex- mentary feeding. It is shown that resource plain why project impact was muted by the time constrained women are more likely to have a final outcomes are considered. While attention knowledge-practice gap, supporting the argu- can be paid to each of these weak links, the BINP experience does demonstrate the difficulty of implementing complex designs. Number of Deaths Averted T A B L E G . 3 4 by BINP Project and Cost Should BINP Be Expanded Nationally? per Life Saved (US$) Despite recent progress, Bangladesh has contin- uing high levels of malnutrition. Several factors 1998 2003 lie behind improving nutrition, notably im- As a percentage of all (estimated) deaths 26.1 10.5 proved agricultural performance. Nonetheless, Number of lives saved 296 129 there appears to be a case for direct nutritional Cost of one life saved I 4,925 8,661 interventions. But in its current form, BINP has Cost of one life saved II 2,328 4,095 had a rather meager impact on nutritional out- 1 8 2 A N A L Y S I S O F B I N P ' S C O M M U N I T Y - B A S E D N U T R I T I O N C O M P O N E N T Cost of Reducing Malnutrition and Saving Lives from Hypothetical T A B L E G . 3 5 Income Transfer and Rice Ration BINP Simulations All costs Only children costs Income transfer Rice ration Extra weight gain (grams/month) 18 18 13 110 Reduction in prevalence of underweight 7.5 7.5 5.4 37.1 Percentage of deaths averted (number of deaths 26.1 26.1 22.7 70.0 in parenthesis) (296) (296) (258) (794) Cost of removing a child from underweight (US$) 396 183 555 80 Cost per life saved (US$) 4,925.0 2,328.0 5,654.0 1,835.0 comes, especially in comparison to its cost. to target resources more finely, toward less well There are two ways to respond to these findings. nourished mothers and children, with a larger The first would be to attempt to improve project supplement in the lean season. The alternative is effectiveness by strengthening the causal chain. a less sophisticated form of nutritional program, The most plausible way of doing this would be such as a rice ration. Although such schemes are T A B L E G . 3 6 Links in the Causal Chain Assumption Children Mothers Attend growth monitoring Over 90 percent of children attend growth Over 70 percent participate in monitoring sessions monitoring sessions pregnancy weight gain Targeting criteria correctly ap- Nearly two­thirds of eligible children not fed 60 percent of eligible women not receiving plied; participants stay in pro- (reasons: don't attend growth monitoring in first supplementary feeding gram to receive food place, wrong application of targeting criteria, drop out of feeding) Acquire knowledge and put it One­third of mothers of children receiving sup- There is a knowledge­practice gap, driven by into practice plementary feeding do not receive nutritional material resource or time constraints. counseling. There is a knowledge practice gap (see mothers). No leakage or substitution One­quarter of children fed at home, increasing One­third admit sharing food, and there is possibility of both leakage and substitution. substitution for those who do not. At most, 40 percent of eligible women receive full supplementation. Feeding and nutritional advice Supplementary feeding has a positive impact Pregnancy weight gain is too little to have a no- have an impact on nutritional on child nutritional status, especially for the table impact on low birth weight, except for status most malnourished children. There is only most malnourished mothers. Moreover, weak evidence of any impact from nutritional mother's pre­pregnancy nutritional status is counseling. more important than pregnancy weight gain. Consequently, birth weight gains are slight, though greater for children of severely mal- nourished mothers. Eating more during preg- nancy is the main channel for both pregnancy weight gain and higher birth weight. 1 8 3 184 MAINTAINING MOMENTUM F I G U R E G . 6 Calculation of BINP Project Costs PSM results in 6 thanas: Change in WAZ z-scores TO over 8.5 months: 0.13 in 1998 and 0.05 in 2003. 2015? Change in Change in malnutrition mortality Probit function:- Pelletier method: 5.33% in 1998and 6x1000 lives saved 1.99% in 2003. in 98, 2 in 2003 ICR costs Project costs in Present value of Total project cost was 65,610 1st phase thanas project costs Cost of upgrading 1 from Nov 1996 to Mar 2003. The project operated for 3,141 Project costs are actualized Cost of saving 1 life malnourished child Excluding contingencies and months in 59 thanas. Each using a 12% year interest rate, Project cost/ lives Project cost/ children other expenses not related to year the project operated in and discounting a 5% saved. In 1998: upgraded. In 1998: nutrition project, project costs six 1st phase thanas for 72 devaluation rate. Current 1459/(.006*49023)=4925. (1459*(8.5/12))/(0.0533* over the period were 46,589. months. The project costs in figures are multiplied by In 2003: 49023)=396. this area per year are the (1+0.07)^ number of years. The 1113/(0.002*56054)=8661. In 2003: 706. fraction present values are 1,459 in 1998 (72/3141)*46,589=1,041. and 1,113 in 2003. Targeted population Interpolation of 1991 and 2001 census data in 6 thanas: 1,446,964 pop. in 1998, and 1,564,367 in 2001. Children aged 6 to 23 months (4% of pop. In DHS): 57,879 in 1998 and 62,575 in 2003. Participating children (85% in 1998 and 90% in 2003): 49,023 in 1998, and 56,054 in 2003. A N A L Y S I S O F B I N P ' S C O M M U N I T Y - B A S E D N U T R I T I O N C O M P O N E N T Simulation of an Income Transfer Equivalent to the F I G U R E G . 7 Project Cost Annual household income transfer (US$) Project cost in 1998/participating children = (1,459,000 / 49,023) = 29.8. Discount a 15% administrative cost for the transfer: 29.8 * 0.85 = 25.3 Additional per capita expenditure per month (Taka) Discount a saving rate of 10%. Multiply by the exchange rate (48.5 Taka per US$). Divide by 12 months and average household size (5.4). (25.3 * 0.9 * 48.5) / (12 * 5.4) = 17 Extra calories consumed per child per day (kcals) Use calories elasticity of expenditure (0.363), mean per child expenditure (652) and mean calories consumption of children 6 to 23 (917). 0.363 * ((17 + 652) / 652) ­ 1) * 917 = 8.7 Extra weight gain per child per month (grams) 20 calories are required for 1 gram weight change: 8.7 / 20 * 30 = 13 Reduction in prevalence of underweight (%) Number of lives saved (n) 1 kg weight gain is required for an increase of Using Pelletier and Frongillo method, this 0.86 in average z-score. The year change in weight gain and changes in z-scores and z-score is: 13 / 1000 * 12 / 0.86 = 0.13. malnutrition rates results in 258 deaths With the use of a probit function this is averted. equivalent to a reduction in malnutrition of 5.4%. Cost of upgrading a Cost of saving 1 life (US$) malnourished child (US$) Project costs/number of Project cost/children upgraded. lives saved. 1459 / 0.054 / 49023 = 555 1458 / 258 = 5654 1 8 5 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Simulation of a Rice Ration Equivalent to the F I G U R E G . 8 Project Cost Resources available for rice purchase (Taka) Project cost (1,458) minus 25% administrative costs multiplied by exchange rate (Taka 48.5 / US$). 1,458 * 0.75 * 48.5 = 53,035. Rice delivered to each child per month (Kg.) Divide available Taka for purchase by the number of children (49,023), the number of months (12) and the rice price in Taka per kilo (16). 53,035 / 49,023 / 12 / 16 = 5.6 Extra calories consumed per child per day (kcals) Multiply food delivered by the calories per kilo of rice (3,500). Discount a 50% substitution rate. Divide by the number of days in a month (30). 5.6 * 3,500 * 0.5 / 30 = 329 Extra weight gain per month per child (grams) Multiply the extra calories times the number of days in a month (30). Detract 50% of Kcals intake because spent into increased physical activity. Divide by the number of calories required for a 1 gram increase of weight for a child who is gaining weight rapidly (45). 329 * 30 * 0.5 / 45 = 110 Reduction in prevalence of underweight (%) Number of lives saved (n) 1 kg weight gain is required for an increase of Using Pelletier and Frongillo method, this 0.86 in average z-score. The year change in weight gain and changes in z-scores and z-score is: 110 / 1,000 * 12 * 0.86 = 1.1 With the use malnutrition rates results in 794 deaths of a probit function this is equivalent to a reduction averted. in malnutrition of 37.1%. Cost of upgrading a Cost of saving 1 life (US$) malnourished child (US$) Project costs/number of Project cost/children upgraded. lives saved. 1,458,000 / (0.371 * 49,023) = 80 1,458,000 / 794 = 1,835 1 8 6 ANNEX H: DFID AND WORLD BANK PROGRAMS IN BANGLADESH This annex provides data on selected interven- for, electrification has a significant impact in re- tions mentioned in the report. ducing mortality. These econometric estimates are likely to underestimate the impact of elec- Rural Electrification trification since electrification will also affect When the Rural Electrification Board (REB) was mortality indirectly, notably by the increase in created in 1976, only 3 percent of the rural pop- economic opportunities such as irrigation and ulation had access to electricity. REB developed small businesses. a plan for rural electrification based on coopera- The ICR for REL III provides cost benefit cal- tives, called Palli Bidyut Samities. The plan had culations for four new cooperatives. The total five phases that would complete rural electrifi- cost for 93,214 connections, of which 80,210 are cation by 2005. Different phases were supported residential, is given as Tk1, 686 million. This a by different donors, including USAID, Finland, discounted valued, presumably discounted back Norway, Japan, and CIDA, as well as the World to the year before the initial investment, 1991. Bank, which has had three projects, shown in Thus, to arrive at the capital cost in U.S. dollars table H.1. for the year 2000, this figure is multiplied by a There have been some implementation dif- CPI-based inflator, and then divided by the ex- ficulties. Under REII there were financial prob- change rate for that year: lems, as REB was not paying its debt to government, as it in turn was not receiving pay- Capital cost in 2000 US$ = 1,686 × (cpi in 2000 / CPI in 1991) / ments from the cooperatives. In addition, RE & (Tk per US$ in 2000) REII fell short of connection targets, but REIII exceeded them. = 1,686 × (1129 / 82)52 DHS data show the percentage of rural = 550 households with electricity to have risen from The cost of a household connection is estimated 5 percent in 1993 to 10 percent in 1996 and on a pro rata basis: 15 percent in 1999 Figure H.1 shows the strong correlation be- Cost of h/h connection = $550 × (80,210 / 93,214) tween under-five mortality at all ages, with infants = $470 of households with no electricity 50 percent This calculation thus confirms that during the more likely to die than those with electricity, ris- 1990s the cost per connection was approxi- ing to an increased change of 75 percent for chil- mately US$500 per household. However, since dren. Of course, electrification is correlated with less-accessible (and hence more costly) house- many household characteristics that are also cor- holds are now being connected, the cost per related with mortality, notably household eco- connection has risen. nomic well-being, but also location, which may proxy for access to other facilities and general The Female Secondary School environmental conditions. Assistance Project Multivariate analysis takes into account these The stipend program grew out of a USAID- additional factors. Even once they are controlled supported, NGO-implemented program in one 1 8 7 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? T A B L E H . 1 World Bank Rural Electrification Projects Period IDA amount Connections Rural electrification 1982­86 US$49 milliona 70,953 residential connections, 7,436 commercial, and 1,145 irrigation. Rural electrification II 1985­94 US$80 million 85,766 residential connections, 16,004 commercial, 2,406 irrigation. Rural electrification III 1990­99 US$110 million 216,791 new residential connections and 140,066 renovated; 45,875 commercial new and renovated, 9,367 irrigation new and renovated. a. US$40 million initial credit and US$90 million supplemental credit. upazilla in the early 1980s. The program was sub- The scheme is administered through four sequently expanded, reaching seven upazillas by separate project offices. In principle, the scheme the time NORAD took it over in 1992. The World should be the same in each upazilla, regardless Bank program, under the Female Secondary of which funding source it falls under. However, School Assistance Project (FSSAP), began in some projects have additional components, 110 upazillas the following year. The Asian De- such as the component to improve school facil- velopment Bank also began its support in 1993, ities under the Bank project. In addition, the providing finance for 53 upazillas. In 1994, the strength of monitoring and supervision appears government decided to expand the program, to vary (e.g., it has not been possible to obtain which had proved very popular, to all 410 rural data on the number of beneficiaries under the upazillas, meeting program costs from its own government scheme). funds in areas not covered by the three external The key characteristics of the program are as agencies. Figure H.2 shows the number of schol- follows: arships supported by each agency; comparable data for GoB are not available. Under FSSAP and · The stipend is paid to all girls attending rural the follow-on FSSAP II, the World Bank financed secondary schools which are enrolled in the 6.5 million girl-years of stipends over the period program. The stipend is paid into a bank ac- 1994-2002, equivalent to paying stipends for close count in the girl's name (where there are no to 2 million girls attending secondary school. local branches of the bank administering the scheme, it opens booths for this purpose). Tuition is paid directly to the school. Bivariate Relationship · The school has to certify that the girl attends F I G U R E H . 1 Mortality and Electrification school on at least 75 percent of school days, and that she is attaining passing grades 90 (45 percent). No electricity 80 · The girl must remain unmarried. Electricity 70 The stipend, which increases by grade, is in- 1,000 60 tended to cover a proportion of the direct costs per 50 of the girl's education. It does not fully cover 40 these costs, especially as it has remained the 30 Mortality same in nominal terms since the scheme began. 20 The stipend has not changed in nominal 10 terms over the lifetime of the project. Including 0 the two exam payments of Tk 250, the total NNM IMR CMR amount paid per girl during the course of her ed- Source: Calculated from DHS data. ucation is Tk 3,914 (using a simple average of 1 8 8 D F I D A N D W O R L D B A N K P R O G R A M S I N B A N G L A D E S H F I G U R E H . 2 Number of Stipends Financed by Each Agency 1600 ADB NORAD 1400 IDA (000s) 1200 1000 awardees 800 stipend 600 of 400 Number 200 0 1994 1995 1996 1997 1998 1999 2000 2001 2002 government and nongovernmental tuition fees, stipends. The OED PPAR of FSSAP suggested table H.2). At the 2000 exchange rate this that other factors may have been more impor- amount equals US$75.30. Allowing an adminis- tant than the stipend, notably (1) the increasing trative markup of 20 percent1 brings the cost per number of girls completing primary school, girl to US$90.40. which meant that girls enrollment at secondary An alternative way to arrive at the estimated level grew much more rapidly than it did for cost per girl is to divide the cost of the World boys over the decade before the program, and Bank FSSAP, Tk 4,518.2 million, by the number (2) free tuition for girls in grades 6-8, which was of girl years supported by the project (4.904 mil- established in 1992-93. Nonetheless, the OED lion), which gives a figure of Tk 920 per girl year, study concludes that "project probably had an which is US$17.8 per girl year. Since the stipend effect on enrollments" (World Bank 2003 p.10). runs for five years, this multiplies to US$88.60 Female transition rate jumped from under 60 per girl. Hence, both approaches result in an percent in 1990­91 to over 70 percent in 1992, estimate of US$90 in stipend payments per girl but then again to 79 percent in 1994 and climb- during the course of her secondary education. Female secondary enrollment has grown Stipend Payments rapidly since the stipend has been introduced, at T A B L E H . 2 under FSSAP (Tk) a rate of over 10 percent a year since 1993. However, the important question is the extent Stipend Tuition Tuition to which this expansion in female enrollments Grade (monthly) (govt.) (nongovt.) SSC exam fee has been the result of the stipend program. An upper limit of the program impact is given by the 6 25 10 15 .. 7 30 12 15 .. whole rise in female secondary enrollments in 8 35 12 15 .. rural areas since 1993. But this figure would 9 60 15 20 250 clearly be an overestimate. Enrollments had any- 10 60 15 20 250 how been rising before the introduction of Source: FSSAP ICR, p. 31. 1 8 9 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? ing above 80 percent in the following years There are alternative means of constructing the (World Bank 2000 p. 78). "without stipend enrollment figure." Options A more rigorous attempt to determine are to base the counterfactual on the growth of project impact used econometric estimating ex- enrollments among rural girls prior to the in- ploiting the fact that the program was rolled out troduction of the stipend, enrollment growth over time. This study found a significant impact for rural boys during the program period, or en- of the program on enrollments using two dif- rollment growth for urban girls during the pro- ferent methods: the first, using household data, gram period. Finally, estimates can be made showed a 12 percent increase in enrollments for using the impact study mentioned in the previ- each year the program was in operation. The ous paragraph. second approach, using school-level data, Unfortunately data are available for girls and showed an initial impact of 8 percent, falling to boys enrollments, and rural and urban, but not 4 percent in subsequent years. An earlier paper the two disaggregations together. Hence single by a member of the same research team ad- difference estimates can be made comparing dresses this point more specifically, suggesting boys versus girls (7.4 percent), before versus that the program caused an increase in annual after for girls (-0.7 percent as the rate of increase enrollments of 2 percent above the trend pre- was higher prior to the program), and double vailing before the program's introduction. difference (0.3 percent); see table H.3. Finally, These results need to be assessed against the using the econometric estimate of a 2 percent in- actual increase that has taken place to know cremental growth rate suggests that just under what share of the enrollment increase is attrib- one-fifth of enrollment growth is the result of utable to stipends. the program. This latter estimate, which falls in A variety of methods may be used to gauge the rather wide range given by the simple esti- this last figure, that is the increment in enroll- mates (from none to two-thirds), is used for the ments on account of stipends. The upper limit cost-effectiveness calculations. is to assume that the whole increase is the result of the stipend, which is clearly an overestimate. A Note on DFID Aid to Bangladesh Figures H.3 and H.4 show two figures related to U.K. aid to Bangladesh. Figure H.3 shows the Growth in Secondary substantial sectoral shift that has taken place T A B L E H . 3 Enrollments (percent) during the 1990s, away from infrastructure, partly to health. Figure H.4 shows the number of Boys Girls Difference separately funded health projects, showing that 1986­92 4.2 11.3 7.1 despite the move to a sector wide approach, 1993­99 3.2 10.6 7.4 there in fact remain a substantial number of dis- Double difference 0.3 crete activities. 1 9 0 D F I D A N D W O R L D B A N K P R O G R A M S I N B A N G L A D E S H F I G U R E H . 3 Sectoral Composition of DFID Aid to Bangladesh 40 Agriculture and rural development 35 Transport and energy 30 Health and population Roads and bridges (%) 25 Emergency relief and flood rehabilitation aid Commodity aid 20 DFID of 15 10 Share 5 0 1991­94 1995­2000 2001­04 Number of DFID Health F I G U R E H . 4 Projects 45 40 35 projects 30 25 ongoing 20 of 15 10 Number 5 1990­91 1992­93 1994­95 1996­97 1998­99 2000­01 2002­03 1 9 1 ANNEX I. AGRICULTURAL PRODUCTION, NATURAL DISASTERS, SEASONALITY, AND NUTRITIONAL OUTCOMES Trends in Rice Production only one or two crops are grown in the lowlands Production of rice, the main food crop in Bang- (Ateng 1998, p. 144). Of the net cropped area2 ladesh,1 has increased by over 60 percent over in Bangladesh, 36 percent is single cropped, the past four decades as a result of a 50 percent 51 percent double cropped, and 13 percent triple increase in yields and a close to 25 percent in- cropped (annually). That gives a cropping inten- crease in the area grown (figure I.1). Most of the sity of 177 percent (tables I.2 and I.3; Mian and increase in production is due to the introduction others 2001, p. 37). Despite the lower prevalence of high-yielding varieties, which, among other of triple cropping, this area has expanded from things, have enabled production of rice during around 480,000 hectares in 1965­69 to an average different seasons. of 968,000 hectares during the triennium ending 1996­97. Thus, it accounts for an important part Cropping Seasons of production increase in this period. Bangladesh has two cropping seasons, Kharif and The increase in multiple cropping has oc- Rabi. Kharif lasts from April through November curred as a result of the introduction of new tech- and is divided into early Kharif and late Kharif, nology spearheaded by the green revolution the former being the pre-monsoon months of (Alauddin and Hossain 2001: 34). Whereas rice April through July and the latter June through cultivation has traditionally been determined November. Rabi, the winter season, lasts from by the monsoon, the spread of small-scale, me- November through May and is colder and drier. chanically powered irrigation equipments, HYVs Figure I.2 provides an overview of crops, seasons, (high-yielding varieties), Boro technology and an and weather. irrigation policy, which enables farmers to invest Each season is suitable for producing different in shallow tube wells sited at their discretion, varieties of rice. The main rice crop is Aman, led to an annual surge in Rabi season Boro rice which is grown during late Kharif, followed by production (Ateng 1998, p. 144; Rogal, Harriss, and Aus, which is grown during early Kharif, and Boro Bose 1999, p. 99). Therefore, while MV (modern rice grown during the Rabi season. The former variety) rice was introduced in the late 1960s, dif- variety is the largest in terms of area grown fusion in the dry season really took off in the mid- whereas Boro rice has the highest yield and the 1980s along with changes in government policies highest output. Aus rice is the smallest in term of in favor of privatization in the procurement and area grown, production, and yield (table I.1). distribution of small-scale irrigation equipment The main determinant of the cropping pattern and chemical fertilizers, liberalization of trade, in Bangladesh is land elevation. It affects the an- and reduction in tariffs for imported agricultural nual extent and duration of flooding, which in equipment. Consequently, while 63 percent of turn influences cropping calendars. Land eleva- the rice harvest today is monsoon-related, the re- tion similarly influences the number of crops maining 37 percent is produced in the dry Rabi grown annually in specific areas. For instance, season when farmers grow HYV Boro rice using about 71 percent of the total cultivable area is in supplementary groundwater irrigation. This type the highlands and medium highlands where two of production has experienced a 6.9 percent to three crops are grown each year. By contrast, annual growth during 1972­73 to 1993­94 com- 1 9 3 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? F I G U R E I . 1 Rice Production, Area, Yield, 1961­2002 45,000 4.00 40,000 3.50 35,000 3.00 30,000 2.50 ha 25,000 000/t Yield (right axis) 2.00 t/ha 20,000 Production (left axis) 000 1.50 15,000 1.00 10,000 Area (left axis) 0.50 5,000 0 0.00 1961 1966 1971 1976 1981 1986 1991 1996 2001 Source: IRRI 2004 F I G U R E I . 2 Crop Seasons and Weather Jan Feb March April May June July Aug Sept Oct Nov Dec Rabi season-- Rabi Season--Boro rice Boro rice Crop seasons Early Karif season ­ Aus rice Late Kharif season--Transplanted Aman rice Harvest T. aman HYV Boro Aus B. aman T. aman Season Dry season (and pre-monsoon hot season) Monsoon season Dry season Average rainfall, mm 19 33 56 104 190 317 439 309 240 175 33 5 (Dhaka) Average rainfall, mm 10 31 137 396 553 834 827 633 524 250 27 10 (Sylhet) Source: Adapted from Ateng 1998, supplemented from Bangladesh Bureau of Statistics 2000; National Environment Agency, Singapore. 1 9 4 A G R I C U L T U R A L P R O D U C T I O N , N A T U R A L D I S A S T E R S , S E A S O N A L I T Y A N D N U T R I T I O N A L O U T C O M E S pared with 2.8 percent annual rice production Rice: Production, Area, over the same time (Ateng 1998, p. 144; Rogal, T A B L E I . 1 Yield by Season, Harriss, and Bose 1999, p. 99). 1992­94 Average Despite the increase in multiple cropping fol- lowing the spread of irrigation and production of Share of Share of dry season Boro rice, the possibilities for triple Area total Output total Yield cropping--and in particular for growing three (000 ha.) (percent) (000 t) (percent) (t/ha.) rice crops--are limited. This is so first because Aus 1,767 17.4 733 6.1 1.15 the growing seasons of both Aus and Aman rice Aman 5,763 56.9 4,898 40.9 1.64 overlap with that of Boro rice (see figure I.2), Boro 2,605 25.7 6,342 53.0 2.58 and second because dry season Boro rice often Total / average 10,135 100 11,973 100 1.8 replaces other dry season non-cereal crops. Source: Baffes and Gautam 1996. Some argue that at least in lower lands, ex- pansion of irrigated Boro rice cultivation has taken place on land that would otherwise have Land Utilization by been involved in a complex rain-fed double and T A B L E I . 2 District (as percent of triple cropping pattern involving broadcast Aus,3 total net area cropped) broadcast Aman, and a range of other winter crops. Since most non-cereal crops are grown on Single cropped Double cropped Triple cropped non-irrigated land and compete for land in the 1989/90 dry season these crops are thus liable to be re- Chittagong 35.0 52.4 12.6 placed with the expansion of irrigation ( Jaim Dhaka 41.5 40.9 17.6 1984, p. 36; Ateng 1998, p. 146­8). This argu- Khulna 72.2 23.8 3.9 ment is supported by looking at the proportion Rajshashi 58.6 34.5 6.9 of Aus and Boro rice cropped areas under triple All Bangladesh 43.5 44.5 11.9 cropping. Ten percent of MV Aus cropped area 1997/98 is under rice-only triple cropping and 17 percent Chittagong 35.7 58.4 13.2 is under rice and other triple cropping, while Dhaka 39.6 56.9 13.6 7 percent of Boro rice is under rice-only triple Kulna 66.9 25.5 4.1 cropping, and only 1 percent under rice and Rjshashi 52.4 44.3 7.7 other triple cropping (Mahmud, Rahman, and All Bangladesh 39.0 56.1 13.8 Zohir 1994, p. 30). Source: Chowdhury and Zulfikar 2001. Because Aman rice is the largest crop in terms of area grown and makes up close to 60 per- not to suffer during disasters (table I.4), some- cent of rice grown area (table I.1), November- times even increasing, presumably reflecting re- December remains the main harvest season with lief food aid. the widest food availability. May-June, when Boro rice is harvested, is the second most important Seasonality and Nutrition harvest. Cropping patterns determine harvest seasons, which in turn influence work, employment, food Aggregate Output, Disasters, and Daily Energy Supply As rice is the major agricultural crop, the trend in T A B L E I . 3 Land Use by Rice Sort, 1993 daily energy supply per capita closely follows rice production, notably the upsurge in the late 1990s (figure I.3). The figure also shows the major dis- Single cropped Double cropped Triple cropped asters during this period, which for the most part Aus 2 71 27 seem to have not had a large an impact an aggre- Boro 33 58 8 gate production. Daily energy supply also appears Source: Mahmud et al. 1994. 1 9 5 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? F I G U R E I . 3 Rice Output and Daily Energy Supply 45,000 2300 40,000 2200 35,000 Daily energy supply (right axis) 2100 30,000 25,000 2000 person tons per 00020,000 Rice Production (left axis) 1900 15,000 Calories 1800 10,000 1700 5,000 0 1600 1970 1974 1978 1982 1986 1990 1994 1998 2002 Source: www.em-dat.net intake, and nutrition in rural areas. As the largest agricultural technology has made significant con- crop, Aman is the most important in this respect. tribution to employment creation, rather than the Demand for agricultural labor is highest during reverse, especially HYV Boro rice (Alauddin and the Aman harvest. Also, opportunities for em- Tisdell 1998: 86­9). ployment are enhanced during production and Seasonal patterns affect nutrition in three harvest of Boro and Aus--for instance during the ways: (1) food availability, (2) labor requirements, months of January and March when the trans- and (3) rainfall affecting health, with children par- plantation of HYV Boro takes place (Pitt and ticularly prone to disease in the rainy season. For Khandker 2002: 12). The introduction of new households working their own land the harvest season is one of extra effort, as food is not widely available. But for wage laborers this is a time of in- Top 10 Disasters in creased income, though rice prices will be higher T A B L E I . 4 Bangladesh by Number until harvest is completed. of People Affected Children Disaster Date Affected (millions) Two studies (Brown 1982; Bloem, Moench- Flood July 1987 73.0 Pfanner, and Panagides 2003) conducted more Flood August 1988 73.0 than two decades apart focused on seasonal Flood July 1974 38.0 variations in children's nutritional status. Both Flood June 2004 33.6 found significant differences across seasons. A Flood May 1984 30.0 third study (Becker and Weng 1998) found signif- Drought July 1983 20.0 Flood July 1968 15.9 icant seasonal variation in neonatal, post-neonatal, Wind storm May 1965 15.6 and child mortality. Wind storm April 1991 15.4 Using various anthropometric indicators,4 Flood July 1998 15.0 Brown (1982) found the period of poorest nu- 1 9 6 A G R I C U L T U R A L P R O D U C T I O N , N A T U R A L D I S A S T E R S , S E A S O N A L I T Y A N D N U T R I T I O N A L O U T C O M E S tritional status to begin with the height of the that the lean season falls between April and No- monsoon rains in the Kharif season continuing vember, with the start and end points varying a until roughly the time of the Aman rice harvest. little from year to year. Measuring weight for age, arm circumference, These findings indicate a temporal association and triceps skinfold thickness, they found a between cropping patterns and children's nutri- large decline between June and September. The tional status. Seasonality in nutrition, however, latter was the month with the highest prev- is furthered by the greater prevalence of floods alence of child underweight. August was found and diseases during the monsoon in the late to be the month with by far the lowest mean Kharif season. Floods are extremely common in percentage of expected monthly increments Bangladesh and affect hundreds of thousands of of weight for age (<10 percent) (Brown 1982: people each year, particularly during the months 308­9). of June, July, and August (table I.5). Of the Using NSP data, Bloem, Moench-Pfanner, and 47 floods that occurred between 1970 and 2004, Panagides (2003) also found that wasting (WHZ 39 took place between June and September. Nine < -2 SD) in children under five peaked between of 19 diarrhoeal epidemics similarly took place June to September with 28 percent of children during those months. under five being wasted. The prevalence of wast- The link between floods and health is indi- ing in children aged 0­23 months rose from cated by the fact that the monsoon season has "serious severity" to "critical severity" in August/ the highest occurrence of diarrhoeal deaths, September (Bloem, Moench-Pfanner, and Pana- and highest levels of child mortality (Becker and gides 2003, p. 90­1). Such findings were com- Weng 1998). Brown, Black, and Becker also find plemented by other evidence that food scarcity the prevalence of certain infectious diseases to is greatest in the months from August to Oc- be generally higher during that season (1982: tober. For instance, in 1992, 1993, and 1994, 312). Finally, examining the 1998 floods, Del September was the month with the highest per- Ninno and others (2001) found that children in centage of households that took a loan for food severely exposed villages had a slightly in- or made a distress sale of assets (2003: 55). creased illness risk and for those in severely ex- OED's own analysis of WAZ for under-fives is posed villages the risk more than trebled; the shown in figure I.4. The pattern shown confirms incidence of fever rose by 55 percent and that of Seasonal Patterns in Weight for Age Z-score F I G U R E I . 4 for Under-Fives ­1.7 ­1.8 score­ ­1.9 z ­2.0 age for ­2.1 ­2.2 Weight ­2.3 ­2.4 Feb. June Oct. Feb. June Oct. Feb. June Oct. Feb. June Oct. Feb. June Oct. Feb. June Oct. 1998 1999 2000 2001 2002 2003 1 9 7 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? August-September) (Bloem, Moench-Pfanner, T A B L E I . 5 Disasters by Season and Panagides 2003). Such seasonal variations in female malnour- ishment are most likely explained by the signifi- March- June- October- May September November cant increase in women's workload. A qualitative study of women's schedules and time constraints Floods 7 39 1 found them to vary according to the agricultural Diarrhoeal epidemics 6 9 4 season as harvest seasons and rice processing sig- Source: EM Dat 2004. nificantly affects time use (Levinson and others 2002). The busiest seasons for women are the diarrhea by 25 percent, which will affect nutri- crop harvest seasons from mid-April to mid-June tional outcomes. and from mid-November to mid-January. The rice harvest months of mid-April to mid-May and Women mid-November to mid-December (figure I.5) are The OED analysis of the NSP data shows that sea- the times when women generally prepare the sonal variations in underweight are not the same least number of meals because they lack time to for children and women. Whereas children's do so. Most women in the study said that they weight for age is lowest during the monsoon in managed to complete their tasks only by sleep- June-July and the "hungry months" preceding the ing, resting, and bathing less. Finally, the study Aman harvest, and highest during November and found that women considered time a more seri- December, women's Body Mass Index (BMI) ous problem with respect to self-care than to peaks in August-September and is lowest in childcare. For example, they regarded eating December. Maternal calorie deficiency (CED) is more food during pregnancy as difficult given most severe in December and January and least their time constraints, whereas providing com- so in August-September (although seasonal plementary food to their children or giving them variation is not found to be very large: 43 per- more than three meals a day was given higher cent in December-January versus 36 percent in priority. F I G U R E I . 5 Seasonal Patterns of Women's Body Mass Index 19.8 19.6 19.4 index 19.2 mass 19 Body 18.8 18.6 18.4 Feb. June Oct. Feb. June Oct. Feb. June Oct. Feb. June Oct. 2000 2001 2002 2003 1 9 8 ANNEX J: APPROACH PAPER Background and Rationale and sustaining a reasonable level of income and Two of the eight Millennium Development Goals consumption. Female education, clean water (MDGs) refer to maternal and child health, and and, most important, sanitation have an impor- child malnutrition is a Goal 1 indicator. The re- tant and well documented impact on (in sults-based rationale of the MDGs requires un- particular) child survival. Female empowerment derstanding the main drivers behind changes in is vital to enable women and families access to MDG indicators: what can government policy health care" (DFID, 2000, p. 9). and interventions, assisted by external partners, Bangladesh has been chosen for a number of do to accelerate the pace of improvement and so reasons. It is a country that has made impressive secure the achievement of the goals? progress in reducing both under five mortality The impact studies being undertaken by and fertility. External partners, including both OED, under the OED-DFID partnership agree- the World Bank and DFID, have been active in ment, seek to address this issue, focusing in this supporting large-scale interventions with spe- case on maternal and child health (MCH) out- cific objectives to improve maternal and child comes in Bangladesh. The study will build upon health. Improved access to health services espe- the methods developed during the first impact cially for poor women and children was included evaluation carried out under the partnership under the first objective in DFID's country strat- agreement, which analyzed the impact of the ex- egy for Bangladesh formulated in 1998. During ternal support to basic education in Ghana from recent years external interventions in the popu- the Bank, DFID, and USAID.1 This approach is lation and health sectors have been coordinated based on the combination of rigorous statistical under a sector program in which both DFID analysis of outcomes with contextual analysis and the Bank have been active partners. Finally, based on qualitative material. A joint evaluation there are a substantial number of large data sets under the partnership agreement interventions appropriate for the analysis. supported by both the Bank and DFID will be in- cluded in the study. The Bangladesh Context The focus of the study will not, however, be Bangladesh presents a mixed picture with re- restricted to interventions in health, nutrition spect to MCH indicators. Over the 1990s infant and population (HNP), since MCH outcomes are and child mortality have been halved: unlike also affected by activities in other sectors. In the most countries, Bangladesh has a sufficient rate words of the 2004 World Development Report of progress to meet the MDG target of a two- "the determinants of supply and demand oper- thirds reduction by 2015. Fertility also fell rapidly ate through many channels" (WDR, 2004, p.27). from the 1980s, although there are fears in some OED's own review of the MDGs similarly quarters that the indicator has now reached a stressed the need for multi-sectoral strategies to plateau. However, despite low levels of fertility, achieve single-sector targets.2 DFID's target maternal mortality is high: at between 320 and strategy paper, Better Health for Poor People, 400 per 100,000 live births, it is greater than that states that, "good health is determined by many in both India and Pakistan and six times that in factors, the most important of which is reaching Sri Lanka. Intermediate indicators show a similar 1 9 9 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? T A B L E J . 1 MCH­Related Millennium Development Goals (a) Goal Target Indicators Goal 1: Eradicate extreme Target 2: Halve, between 1990 and 2015, the Prevalence of underweight children (under five years of age) poverty and hunger proportion of people who suffer from hunger Proportion of population below minimum level of dietary energy consumption Goal 4: Reduce child Target 5: Reduce by two­thirds, between 1990 Under­five mortality rate mortality and 2015, the under­five mortality rate Infant mortality rate Proportion of one­year­old children immunized against measles Goal 5: Improve maternal Target 6: Reduce by three­quarters, between Maternal mortality ratio health 1990 and 2015, the maternal mortality ratio Proportion of births attended by skilled health personnel discrepancy. Immunization rates have grown DFID's support for HPSP has been principally rapidly in Bangladesh and are today notably to Strengthen Health and Population for the higher than those in neighboring countries. But Less Advantaged (SHAPLA), through two mech- only 13 percent of births are delivered by skilled anisms: (i) time slice financing of the HPSP bud- birth attendants, compared to close to half in get, and (ii) project support focused in key areas India. of HPSP to strengthen government's implemen- The study will focus on interventions from tation capacity and to minimize risk. Both Bank several sectors. The chosen interventions will be and DFID financing to the population and health those supported by DFID and the World Bank, sector has had several components, which will and other donors subject to the required infor- be assessed separately for their inclusion in the mation being available. Amongst these sectors study. Other donor interventions to be included will be health and population, which has ac- in the study will be provisionally identified at the counted for 17 percent of DFID aid to Bang- design stage. ladesh in recent years (Annex J.1b) and, together In line with the multi-sectoral philosophy of with nutrition, 7 percent of Bank lending (Annex the MDGs (i.e., the notion that the relevant in- J.1a). As described below, the other sectors are terventions for each outcome are in a range of to be identified based on an analysis of the de- sectors) the study will consider interventions terminants of MCH outcomes. from outside the HNP sector. For example, in- On the Bank's part there have been five come-generation activities will also positively af- Health and Population projects in Bangladesh: I fect MCH outcomes, as will improved water (1974-80), II( 1979-86), III (1986-92), IV (1992- supply and better education. The analysis of 98) and the Health and Population Program other sectors will be carried out through a two Project (ongoing since 1998), the last three of stage approach described below. which have been co-financed by DFID. In addi- tion there is the Bangladesh Integrated Nutrition Evaluation Questions Project (BINP, 1995-02), which is being followed The evaluation questions to be addressed are: up by the National Nutrition Project (NNP, on- going since 2000). All these projects have spe- (a) What is happening to maternal and child cific objectives related to maternal and child health indicators (including nutrition and health. The study will include, amongst other in- fertility) in Bangladesh? Are the poor shar- terventions, PHP III, IV and HPSP, and BINP. ing in the progress that is being made? 2 0 0 A P P R O A C H P A P E R (b) What have been the main determinants of tions contained in project design. For example, MCH outcomes in Bangladesh over this since health and nutrition interventions attempt period, and which of these have accounted to bring about behavior change, they are partly for substantial changes? built on the assumption of irrational beliefs re- (c) What have been the primary drivers of garding health and nutrition practices. But the changes in the determinants of MCH out- data show some discrepancies between knowl- comes since the mid-80s?3 edge and practices, suggesting that projects may (d) What have been the range of government change knowledge but not practices. The most and externally-supported interventions that recent progress report for Bangladesh's Health may have affected the changes in the deter- and Population Sector Program notes scant im- minants of MCH outcomes? provement in service use. The study will use (e) To the extent that interventions have both quantitative and qualitative methods to brought about positive impacts, have they analyze issues such as patterns of service use and done so in a cost effective manner? the rationale for various health and nutritional practices. Evaluation Approach and The determinants given in a regression model Data Requirements of MCH outcomes are not usually direct mea- The analytical framework for the evaluation is sures of policies or programs. The task for an im- provided by the combination of two widely- pact evaluation is to link outputs from policies adopted frameworks--UNICEF's conceptual and programs with the determinants and hence framework of the causes of malnutrition4 and to the final outcomes. This study will examine the Mosley-Chen analytical framework of child the link between MCH outcomes and DFID and survival5--together with the insights from Bank-supported interventions, together with household models of the demand for health those of other donors where possible, through a used in economics.6 The first two frameworks two stage approach. A regression-based decom- have the same outcome variables (malnutrition position analysis of the determinants of MCH and mortality) and essentially the same struc- outcomes will be used to identify which factors ture, in which the basic causes come from a have been most important in changing out- country's socio-economic and institutional comes in Bangladesh. This analysis will be used conditions, but these are mediated through a to identify which DFID, Bank and other inter- number of proximate causes observed at the ventions have directly supported improvements community, household and individual levels. in MCH. The most important interventions will Maternal health is both a determinant of child be selected for a more detailed analysis, to quan- welfare outcomes and an outcome in its own tify the size and cost effectiveness of their im- right. Economic models also highlight the pact. In addition to HNP, it is envisaged that two importance of behavioral factors ("tastes and or three other sectors will be selected. The ac- preferences") in affecting demand for health tivities of the DFID and the Bank within these services. sectors will be examined, along with those of The framework identifies the determinants of other donors where this proves feasible. The MCH outcomes in broad terms thus providing a final selection of sectors will be an outcome of theoretical basis for the selection of variables for the first stage of the analysis. modeling these outcomes. These determinants An exception to this approach is BINP, which have been examined more closely in a meta- has an unusually rich evaluation dataset com- analysis of infant and child mortality and nutri- prising a baseline, mid-term, and endline surveys tion, prepared as a background paper for this for both project sites and control communities. study (see Appendix J.3 for a summary of deter- The data collected can be mapped onto the logic minants identified by this analysis). of the project, since intermediate process in- The adoption of a theory-based evaluation ap- dicators relating to behavioral factors were in- proach will direct attention to further assump- cluded in the surveys in addition to outcome 2 0 1 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? data and socio-economic status (SES) variables. organizations may include a component of ca- These data are available to the study team to- pacity development through supported learn- gether with the relevant documentation. Save ing-by-doing. DFID staff in the Dhaka office shall the Children, which has been critical of the ap- facilitate the inclusion of DFID-supported in- proach adopted by BINP, conducted their own terventions in the study and assist, together survey in 2002 and these data will also be avail- with Bank staff, liaison with other agencies as able to the study team. necessary. Several other nationally-representative data sets exist for the relevant period. These are listed Schedule and Task Management in Appendix 2. Surveys likely to be used are: The study will be undertaken in the following stages: · Demographic Health Surveys: 1993/94, 1996/ 97 and 1999/2000 · Inception--comprising an inception field · Household income and expenditure surveys: visit (February 2004), preparation of the ap- 1991/92, 1995/96 and 2000 proach paper followed by preliminary field- · Helen Keller International Nutritional Surveil- work (April 2004) and development of the lance Project study design (April-May). · Maternal mortality survey, 2003. · Analysis first phase--analysis of household data sets of the determinants of health, nu- The first two of these surveys have been ac- trition, and fertility outcomes. Background quired for the study. The other two are not in studies will be carried out in parallel. The re- the public domain, but can be analyzed through sults of these studies will be presented in collaborative arrangements with their owners. Dhaka in late July or early August 2004 to feed A review of existing material shall be under- into the discussions concerning future sup- taken during design to determine to what extent port. Identification of projects to be analyzed the analysis can rely on secondary data and by the study will be carried out in this phase. to what extent commissioned studies are re- · Analysis second phase--collection and analy- quired. Relevant areas of analysis are: (a) trends sis of project-level data from DFID, the Bank in government expenditure,7 (b) facility and and other donors as possible. staffing survey, (c) demand for health services, · Report draft for OED management review to and (d) behavioral changes related to health be ready by late October 2004, with submis- and nutritional practices. At least two of these is- sion to CODE of the final report targeted for sues will be examined through qualitative field- January 2005. work, using participatory analysis of such issues as perception of service quality and the rationale The evaluation will be carried out by a team of underling health and feeding practices. OED staff and consultants with the assistance of Bangladeshi government officials, DFID staff Collaboration with Other Agencies under the Task Management of Howard White Collaboration shall be sought with relevant gov- (OEDST). External peer reviewers will be ap- ernment officials or research institutions in pointed to review the proposed evaluation de- Bangladesh. Commissioned studies from local sign and draft final report. 2 0 2 A P P R O A C H P A P E R Overview of Bank APPENDIX J.1a Operations in Bangladesh Since 1990 Commitments Sector US$ million Percent Economic policy 224.8 4.4 Education 419.9 8.1 Energy and mining 1,252.7 24.3 Environment 151.3 2.9 Financial sector 315.2 6.1 Global information/communications technology 39.8 0.8 HNP 349.3 6.8 Private sector development 362.8 7.0 Public sector governance 185.9 3.6 Rural sector 908.5 17.6 Transport 866.0 16.8 Urban development 48.4 0.9 Water supply and sanitation 35.8 0.7 Total 5,160.4 100.0 Source: calculated from OED database. Overview of DFID APPENDIX J.1b Operations in Bangladesh, 1998­2001 (£ millions) 1998/99 1999/00 2000/01 2001/02 Total Share Education 7 10 10 9 36 10.6 Population and 13 18 14 13 58 17.0 health Natural resources 10 22 29 31 92 27.0 Roads/bridges 11 8 11 11 41 12.0 Water and 0 4 7 7 18 5.3 sanitation Energy 6 14 7 7 34 10.0 Small business/ 2 6 5 5 18 5.3 micro-credit Good government 4 5 5 2 16 4.7 Other 5 6 9 8 28 8.2 Total 58 93 97 93 341 100.0 Source: DFID Bangladesh Country Strategy Paper, 1999: Annex 2. 2 0 3 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? A P P E N D I X J . 2 List of Data Sets Survey (b) Content Availability Demographic Health Child health and nutrition, child care practices, SES, Freely downloadable from Macro International website Survey comprehensive birth history. Includes some "women's agency" questions. BBS Health and demo- As for DHS To determine if in Bank or not, otherwise purchase from graphic survey BBS BINP evaluation data set Nutritional status, beliefs and practices, SES Available in Bank and clearance has been obtained for use, awaiting data SCF Bangladesh nutrition Nutritional status and household characteristics Was promised sometime ago, need to follow up data set Household Income and Standard LSMS type survey, but no anthropometrics. Only Most recent acquired, others available in Bank and being Expenditure Survey most recent has health module, but all have health expen- followed up ditures Health facility survey Not known yet Acquired Helen Keller Interna- Nutritional status + SES. Every two months since 1990, Proposed to contract HK to conduct analysis. tional National Nutrition with nationally representative sample since 1996. Surveillance Survey Content changes but a common core. Vital registration sample Births, deaths and basic household and individual charac- Can be purchased from BBS. Possibly analyze in collabo- survey (BBS) teristics. Cause of death (quite aggregated). ration with BIDS National Nutrition Anthropometric measurement, SES. Can be purchased from BBS. Possibly analyze in collabo- Survey (BBS) ration with BIDS UNICEF Multiple Range of social indicators, representative at district level. Should be available from MICS website, for which clear- Indicators Cluster Survey ance to download already obtained UNICEF maternal health Use, deliveries and materials availability Statistics officer can perform analysis on request facility survey Maternal Mortality Maternal death (sisterhood and h/h based), SES Not available but can influence form of analysis and re- Survey (USAID financed) ceive results Note: Though both BBS surveys began in the 1980s, data may only be available for the last one or two rounds. All the above are household surveys. Some are supplemented by community­level data collection of variables such as facilities and prices. These are all national surveys; there are also more local data, notably those collected in Matlab thana that may also be used. 2 0 4 A P P R O A C H P A P E R A P P E N D I X J . 3 Determinants of Outcome Variables Child Household Community Nutrition Child characteristics: gen- Economic characteristics: per capita income, asset ownership, Basic characteristics: der, age, birth order, sickness, public works participation, paternal labor supply, maternal labor rural/urban, region ethnicity, lagged height supply, poverty status, land ownership, occupation, remittances Facilities: electricity, water, sanitation, travel time to health Care practices: number of Parental characteristics: mother's and father's education, center, school, road conditions, antenatal visits, attended parental height, mother's age, maternal nutrition knowledge, ma- hospital, TV, radio birth, breastfed ternal literacy, maternal employment, marital status, ethnicity, re- ligion Other: famine, season, price of drugs, food prices, literacy, Nutrition­related factors: per capita calories, per capita pro- hospital beds teins, parental education, household size, mother's height, food prices Note: some studies use a com- munity fixed effects Care­related factors: sibling help, adult help, mother works full model. time Demographic characteristics: household size and composition, no. of dependents, sex of household head Mortality Child characteristics: gen- Economic characteristics: income (total or per capita), asset Basic characteristics: der, age, ethnicity, birth order ownership, dwelling space, land ownership, parental employment rural/urban, region (first birth), birth interval, birth group, paternal and maternal labor supply cohort, birth weight, length of Facilities: electricity, water, pregnancy (premature), twin Parental characteristics: mother's and father's education, sanitation, schools, health fa- mother's age, mother's height, no. of children born to mother, ma- cility, health workers, doctors, Care practices: (duration of) ternal health knowledge, marital status, religion, ethnicity, caste paved road breastfeeding, antenatal care, attended birth, immunization, Mortality-related factors: fate of previous children, fate of Other: level of child mortality exposure to water-borne dis- other relatives/household members in community, disease prev- ease alence, population (growth Care-related: mother works full time, sibling/adult help rate), temperature and Note: some studies use inter- precipitation active term between breast- Demographic characteristics: household size and composition, feeding and water and sex of household head sanitation variable Other: persons per room, migration status Fertility Not applicable Economic characteristics: income per capita, asset ownership Basic characteristics: including, land ownership, occupation, employment status rural/urban, region Preferences: female education, male education, contraception Facilities: electricity, access and abortion practices, religion, ethnicity, access to social security to mass media (radio, TV), ac- in old age (including land ownership), farm size cess to family planning facility, schools Other factors: woman's status, woman's age, marriage age, mor- tality among children and other household members, breastfeed- Other: mortality rates, ing duration disease prevalence, literacy Source: derived from a review of statistical studies. 2 0 5 ENDNOTES Chapter 1 10. Moore (2003) argues that the remote nature of 1. Data from the 2003 Maternal Mortality Survey many communities, resulting in control of local poli- were not available for analysis. tics by a small number of dominant families, in part re- 2. The under-five mortality rate in 1970 was 239 per sults from difficulties in moving around the country. 1,000 live births, compared with 202 for India, 181 for Thus, it may be argued that improved infrastructure Pakistan (excluding Bangladesh), and just 100 for Sri has a beneficial effect on political life, which in turn Lanka. will improve service delivery. 3. Social indicators are often built on fairly shaky 11. The data from a fourth DHS, conducted in statistical foundations. However, as shown in Annex 2004, were not available for this study. A, many surveys have been conducted in Bangladesh, and so the data are unusually reliable. The exception Chapter 2 has been maternal mortality, but a new Maternal Mor- 1. Bangladesh has six administrative divisions, fur- tality Survey will improve the situation. ther subdivided into 64 districts. Each district has sev- 4. Whether or not fertility decline continued into eral thanas (formerly upazillas), of which there are 490 the 1990s is a matter of debate. This issue is explored in total. Each (covering some 15­20 villages) is further in Chapter 3 and Annex F. subdivided into several unions. 5. For example, the United Kingdom financed 15 separate activities under the Fourth Population 2. One might wonder if the assessment a few years and Health Project, totaling UK£15.5 million (close to earlier can thus be faulted. While the earlier OED re- US$30 million at current exchange rates). Close to half port missed the emerging success of the program, its of this total was for support to NGO programs, with findings reflected accepted opinion of the program at substantial amounts also for strengthening nurse the time. Perceptions regarding the success of the education (UK£2.8 million) and medical colleges Bangladesh program changed with the data from the (UK£3.1 million) (see Annex I). BFS and CPS carried out in 1989. 6. Rural Electrification I: 1982­86; II: 1985­94; and 3. For example, Caldwell and others (1999) and III: 1990­99. Kabeer (2001). Chapter 4 will examine the evidence 7. The World Bank has lent US$234 million for flood regarding this debate in more detail. relief and drainage programs in the last two decades, 4. See, for example, the staff appraisal report for and DFID £59 million since 1991 (see Annex I). the Bank's Fourth Population and Health Project. 8. The 1998 floods were far greater in terms of the 5. Data from the Health Facility Surveys show a de- affected area and infrastructure destroyed than those cline the proportion of households using government in 1988. But fewer than 1,000 people lost their lives in health and family planning services from 13 to 10 per- the 1998 floods compared with more than 2,400 in cent from 1999 to 2003 (Cockcroft Milne, and Anders- 1988. And only 600 died in the 2004 floods, which ap- son 2003). peared to be at least comparable to those in 1998. 6. A recent study found absenteeism among rural 9. Among older children, drowning is now the public health providers to average 26 percent, but major cause of death. But, as these figures show, these drownings are mainly accidents rather than the was higher for doctors (40 percent), particularly in direct consequence of flooding. Figures on indirect smaller subcenters (76 percent); see Chaudhury and deaths from flooding as a result of disease and mal- Hammer 2003. nutrition are less easy to obtain, though there is 7. DFID is implementing a public-private partner- such an effect (for some estimates, see Del Ninno and ship program under which the clinics are utilized by others 1991). an NGO using a matching grant provided by DFID. 2 0 7 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? 8. This figure includes support to NGO programs 6. The plateau has been attributed to a tempo ef- under the joint finance scheme (JFS). fect. If women postpone births, this will temporarily 9. Other than this contribution, the United King- decrease the total fertility rate (which is based on a dom has not been a main supporter of immunization, synthetic cohort), but it will then rise again. Against though it did assist in the program to eliminate polio the background of an underlying downward trend, (see box 4.2). the tempo effect would first accelerate the rate of de- 10. The Maternal Mortality Survey reported an cline, and then temporarily halt it (see Annex F). even higher share of 74.4 percent of births being at- tended by TBAs (NIPORT, Mitra and Associates, and Chapter 4 ORC Macro 2003, p. 52). 1. The dashed line has the same slope as the fitted 11. The Safe Motherhood Initiative is an interna- line, but with the intercept adjusted to pass through tional effort launched in Nairobi in 1987 that is the 1980s observation for Bangladesh. There is an as- supported by several international agencies, including sumption here that the outcome-income relationship the World Bank. is the same for different countries over time. This is clearly not so, most notably in the case of Bangladesh. Chapter 3 The point of the analysis is to show how much 1. The usual pattern is that mortality falls much Bangladesh departs from international norms, so as to more rapidly for older children, so that the remaining pose the question as to why this is. deaths become concentrated first among infants and, 2. It is an upper estimate, since the calculations are within that category, among neonates (and for neo- based on a simple regression. This equation is un- nates, the first days of life are the most risky). This pat- doubtedly mis-specified, resulting in omitted variable tern is less marked in Bangladesh than elsewhere, bias. Income is positively correlated with several with neonatal mortality falling as fast as that for post- determinants of these outcomes--such as education nates (up to one month), and not too far behind that and immunization--so that there will be an upward of children one to four years. bias on the estimated regression coefficient for 2. The poor are identified as the bottom 50 per- income. cent, ranked by a wealth index, in the 1992­93 DHS 3. The cross-country analysis conducted for this re- data. The wealth index poverty line from that survey port builds on earlier work by Smith and Haddad was applied to the subsequent surveys so that data are (2000), who estimated two models for the determi- comparable between years. This conclusion is robust nants of the proportion of children who are under- to the choice of the poverty line; the concentration weight. Their first model regresses nutritional status curve for the early 1990s has first-order dominance on what they call the "basic determinants" of per over that for the late 1990s up to around the seventh decile. Analysis shows that group reductions of mor- capita GDP and democracy (measured by the Free- tality accounted for far more of the fall in mortality dom House political freedom index). The second than did movements out of poverty (that is, into a model, of underlying determinants, includes access to lower mortality group); see Annex C. safe water, female secondary school enrollments, the 3. Data were examined from 125 DHSs. Ranking ratio of male to female life expectancy as a measure of these surveys by female:male mortality in descending gender inequality, and per capita dietary energy sup- order, the ratio from the three Bangladesh surveys ply. The logic of the approach is that the latter vari- ranks eighth, ninth, and twelfth--that is, in the worst ables, which directly affect nutritional status, depend 10 percent. There is no gender bias in neonatal mor- in turn (though only in part) on the basic determi- tality, with girls less likely to die than boys in the first nants. They then examine the relative importance of month of life. A bias begins to emerge for postnatal the different underlying variables in the explained mortality, although not as marked as that for child change in nutrition. Female education is shown as the mortality. most important factor (43 percent), followed by food 4. The downward trend in disaster-prone areas, supply (21 percent), and then health environment compared with the lack of a trend nationally, may be (safe water, 19 percent), and finally women's status, taken as crude evidence of the effectiveness of flood- with 11 percent. Whereas Smith and Haddad's esti- protection measures mentioned in Chapter 1. mates are calculated across the whole sample, those 5. Direct estimates are based on a weighted aver- for this study are based on the values of the regressors age of the age-specific fertility rates of all women in- for Bangladesh. Specifically, we calculated b(x1­x0)/y, terviewed, whereas indirect measures utilize various where b is the estimated regression coefficient, x the proxy measures to estimate fertility. value of the explanatory variable for Bangladesh in the 2 0 8 E N D N O T E S 1980s (0) and the current decade (1), and y the rele- 15. In China between 1950 and 1980, the maternal vant outcome. mortality ratio declined from 1,500 to 115 per 100,000 4. Following Smith and Haddad (2000), these esti- live births. The vast majority of births during this mates are made excluding income from the model. period were delivered by minimally trained village Hence, some of the explained variation results from the birth attendants (backed up by a strong obstetric care income-driven component of these determinants, and emergency referral network) (Levitt 1999). but there is also an independent element. To test for 16. However, the study of Goodland and others this independent element the equations were re- (2002) finds that training TBAs did not reduce post- estimated including income per capita as an explana- partum infection, which they attribute to the impor- tory variable (see Annex B). An independent effect was tance of pre-existing reproductive tract infection. This found for most determinants, with the exception of is not something to which EOC would have made a gender equality and safe water in some specifications. difference. 5. Bangladesh has six divisions: Barisal, Chittagong, 17. Multivariate analysis suffers from the same Dhaka, Khulna, Rajshahi, and Sylhet. Sylhet was for- problems illustrated in the bivariate analysis (table merly part of Chittagong, so that Chittagong (includ- 4.3). That is, in the neonatal mortality regression, the ing Sylhet) is used for the regional analysis. coefficient for births attended by trained TBAs is sig- 6. Many care-related variables are difficult to in- nificantly positive (that for untrained TBAs is signifi- clude for reasons of endogeneity. Controlling for en- cantly negative), but the interaction of the year with trained TBA is negative, and of sufficient magnitude to dogeneity of antenatal care shows it to be significant give trained TBAs an advantage over untrained by only if provided by a trained doctor or nurse. It is not 1999. This result remains when a trivariate probit is possible to include breastfeeding in the equations, as estimated (to handle the endogeneity of antenatal the only available question ("never breastfeed") is a care and choice of birth attendant). The trivariate re- function of child health. sults show which women use TTBAs: the coefficient 7. Analysis using DHS data measures economic for the later period is highly significant, picking up the well-being using a wealth index rather than income. increased availability of TTBAs, so is being a female Annex C analyzes the relationship between such household head, a multiple birth, having attended wealth indexes and income/expenditure. ANC, and the mother's age. 8. First-born children have a higher risk of infant 18. This issue of difficult birth selectivity is yet death than do subsequent births. Lowering fertility more severe for births at health facilities, making it means increasing the proportion of first-born children, very difficult to measure the beneficial impact of such exerting an upward influence on infant mortality. facilities by this sort of statistical analysis. 9. When interacting the female dummy with the 19. This is the single difference in 1999 (7 = 38-31); Chittagong dummy, the female dummy is no longer that is, just after training was stopped. The double- significant, but that for the interactive term is signifi- difference estimate from the bivariate analysis is 30 cantly positive. deaths averted, as the mortality rate was much higher 10. The hazard ratio for the Cox regression using for births attended by trained TBAs than untrained in 1999 data was 2.05 (table C.7). Using pooled data, and 1993. a without-self mean of immunization to account for 20. Costs are estimated from USAID implemented potential endogeneity, the ratio falls to 1.55. TBA training programs in Afghanistan, Indonesia 11. Khan and Yoder (1998). The Bangladesh EPI and Madagascar (which give estimates in the range office currently reports a figure of $10 per child. US$350-500). Data are scarce on the number of deliv- 12. Khan and Yoder (1998) arrive at an estimate of eries per year, though it appears to vary widely. $136 per life saved. 21. These coefficients are averaged across esti- 13. Moreover, the cost of treatment of children mates from different specifications. who become sick can be shown to exceed the cost of 22. This calculation omits the supply-side costs avoiding sickness, so that immunization represents a associated with higher enrollments. saving of health expenditures. 23. The cold chain is a system of conservation and 14. Once immunization reaches a certain thresh- distribution of vaccines to maintain them in cold tem- old, then there is a reduction in the cross-infection, so peratures, between 32º and 46.4ºF, which is necessary that unimmunized children are also less at risk of the to guarantee their effectiveness. disease. The threshold varies by disease but is mostly 24. The study found that income in households in the 80-90 percent range (Fine 1993). with electricity was 65 percent higher than that in 2 0 9 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? households of non-electrified villages, and 126 per- Chapter 5 cent higher than households without electricity in 1. Thanks are due to the staff the World Bank Nu- electrified villages. Rather than tackle the endogene- trition Hub for facilitating access to the BINP evalua- ity problem econometrically, the study breaks down tion data set, to Save the Children for providing its income by source to identify those that may be attrib- data, and the staff of Helen Keller International Dhaka uted to electrification. On this basis it is concluded for providing the Nutritional Surveillance Project data. that 16.5 percent of the income of electrified house- 2. Theory-based evaluation is propounded in holds is attributable to electrification (Barkat and Weiss (1998); see Carvalho and White (2004) for an others 2002). application to development projects. 25. Assuming a savings propensity of 0.1 and using 3. In some thanas BINP was implemented by the the wealth-expenditure elasticity of 0.8 estimated Government of Bangladesh, so that this last level of using HIES data (Annex C). supervision was missing. 26. The figures of US$500 and US$1,000 per con- 4. In addition to the adolescent girls' forum, there nection were provided by the task manager for the is a married couples' forum. However, during field- Bank's rural electrification projects. Cost-benefit cal- work, only wives were found to be present at such culations in the ICR for RE III present cost figures for forums. four new rural electricity cooperatives, giving a dis- 5. The BINP appraisal document makes some men- counted cost of Tk 1,686 million cost for 80,210 new tion of other decisionmakers, but the main focus of residential connections. These figures give a cost of both design and implementation has been mothers. US$470 per residential connection during the 1990s 6. Calculated from endline data using responses (see Annex I). "always" or "almost always" attend. 27. The number of deaths averted in a household 7. However, the same supplement was provided depends on household composition. If all children throughout supplementary feeding, which does not are already over five, no deaths are averted. However, appear to be a good example regarding nutritional if the mother is just entering her childbearing years, on average just over three children will be born who practices. are less exposed to risk of premature mortality on 8. During fieldwork, differing criteria appear to account of electrification. An analysis of household have been applied to detecting growth faltering, which structure suggests that on average 1.2 children per may in part reflect poor understanding by CNPs household will benefit from the reduced risk due to (found in interviews with 20 CNPs in two project electrification. Since all figures are indicative of broad thanas for the OED qualitative study, see box 5.1) or orders of magnitude, a figure of one child per house- confusion arising from the transition to new criteria hold is used in the cost-effectiveness calculations. under NNP. For normal children, the official criteria However, unlike immunization, which benefits just appear to have been less than 600 gram gains in two that child (possibly others through the herd immu- months for those aged 6­12 months and 300 grams in nization effect discussed above), and schooling, two months for those aged 12­24 months. For chil- which benefits the children of a single woman, elec- dren with WAZ < ­2 SDs, these figures apply to a trification will benefit subsequent generations, so that three-month period. the cost calculations have an upward bias from this 9. Type I error is the proportion of the target group factor. But in comparison to the female schooling cal- not receiving the benefit, and Type II error the pro- culations, they have an overestimate, since the elec- portion of those receiving the benefit who are not el- trification figures include the income effect, which igible to do so. was excluded for schooling. 10. This analysis requires information on pre- 28. The effect of delayed marriage in reducing feeding nutritional status, which is not available in the mortality introduces a "second-order effect" by which BINP evaluation and Save the Children data sets. Save delaying marriage will indeed reduce fertility. the Children also collected register data, but only for 29. Of course this decline partly represents the fact children in the feeding program, so Type I error can- that women subject to permanent methods by ques- not be calculated from these data. The Type II error tionable, semi-coercive means in the early years of the calculated from the Save the Children register data is program are leaving the group of women covered by 15.4 percent, almost identical to that found in the the DHS. Cambridge studies. 2 1 0 E N D N O T E S 11. A number of factors probably lay behind poor Annex B targeting, one of which is certainly the difficulty CNPs 1. In the multiple regression for HAZ primary have in interpreting the quite complex eligibility crite- education in fact has the wrong sign, though it is not ria. Impressions gathered in fieldwork suggest this to significant. be a factor, though there are no numbers to present. The Save the Children data show that the majority of Annex C mothers do not understand the growth charts. 1. The data used in calculations are three DHS 12. Data are not available on pre-pregnancy rounds, collected in 1993/94, 1996/97, and 1999/2000. Mortality rates are calculated using the synthetic co- weight. The BMI used here is calculated from data for hort probabilities method (Rutstein 1984). Data from women not more than six months pregnant, so there the survey rounds are pooled; thus, the rate for 1985- will have been some pregnancy weight gain, thus pro- 89 includes information from each survey round, viding an upward bias to the estimate of Type II while that for 1995-99 includes data on under-fives error--but a downward bias on the estimate of from the 1999/2000 round and under-twos from the Type I error. 1996/97 round. 13. Eating down was recommended medical prac- 2. DHS collects information on household durable tice in European countries for this reason until the goods ownership and housing quality, which can be 1940s. There is not unanimity regarding the accepted used to construct a wealth index. DHS does not col- wisdom, with a recent Australian study finding that lect data on income or expenditure. women delivering children with a higher birth weight 3. Information on differences in health outcomes are more likely to require a Caesarian section. and parental behavior within the same family can be 14. One striking impact of BINP appears to have used to get around the endogeneity problem, though been a much greater utilization of antenatal care this method can only be applied to families for which (ANC) services, which is attributed to the one-stop there are two or more live births (Rosenzweig and service provided under BINP. In practice, this means Schultz 1983). that the CNP coordinated with health workers to pro- 4. The positive correlation between preceding vide services at the same time and place. Exposure to birth order and mortality could be spurious due to ANC under these circumstances implies exposure to prematurity. Premature births have higher mortality risk and, by definition, a lower birth interval, so the nutritional messages. effect of a regression coefficient on preceding birth 15. That is, women could be in receipt of supple- interval could simply be picking up the relationship mentary feeding without acquiring nutritional infor- between prematurity and perinatal mortality (Miller mation. This finding is not a good one for BINP, since 1989). Miller and others (1992) find that controlling supplementary feeding was intended mainly as a tool for gestation period reduces the estimated effects of for demonstrating feeding practices, not as a feeding low birth interval by as much as one-third. program. However, it was already shown that one- 5. Subsequent birth interval is also likely to affect third of mothers whose children were enrolled in mortality risk of children through a number of chan- supplementary feeding did not discuss nutrition with nels, including premature weaning and reduced quan- the CNP. tity and quality of parental resources devoted to the 16. Participation in supplementary feeding was not older child in infancy (e.g., because the mother is found to have a significant impact. This impact may physically tired and therefore less able to provide ad- have been muted by leakage and substitution. equate care for sick children), and in childhood due 17. The assumptions underlying this calculation to competition among siblings for limited familial re- are laid out in Annex G. A 25 percent administrative sources, if the family believes that he or she has overhead was allowed. already passed the critical stage for survival (Muhuri and Menken 1997). However, the relation between subsequent birth interval and mortality is likely to be Chapter 6 highly endogenous because birth interval is likely to 1. While there has been a health facility survey in be shorter if the older child has died; this effect oper- Bangladesh during HPSP, the analytical possibilities ates both behaviorally, where parents choose to re- from a survey are far greater when the facility survey place the lost child, and biologically, as death of the can be merged with a household survey. previous infant or young child interrupts breastfeed- 2 1 1 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? ing causing ovulation to return early (Palloni and are alive at the time of interview, so immunization sta- Tienda 1986; Koenig and others 1990). tus cannot be used. The same procedure is used for 6. It is possible to test between the competing Vitamin A supplementation, which is also only avail- explanations of why preceding birth interval affects able for live children. mortality by interacting preceding birth interval with 13. Chittagong includes Sylhet, which became a mortality status of the previous child, with a signifi- separate Division in 1996/97. cant negative effect on the interaction term providing 14. Due to extreme endogeneity of breastfeeding support for the resource competition theory (Bairagi, with respect to child health status--e.g., low birth Islam, and Barua 1999). weight or sick children may be unable to breastfeed-- 7. Muhuri and Preston (1991) argue that mortality variables indicating breastfeeding are omitted from risk is likely to be greater for both (younger) boys and the explanatory variable set in multivariate analysis. girls the greater the number of (older) siblings of the 15. The non-linear variable x enters the probit same sex there are, since children of the same sex are equation in the form (b1x + b2x2 + ...) where ... in- better substitutes for one another in the family econ- dicates the remaining arguments of the equation. Tak- omy than children of the opposite sex (though the ing the partial differential with respect to x, setting effect for girls will be higher). equal to zero and solving for x gives the maximum/ 8. Regarding the latter, it is important to use sibling minimum at x = ­b1/(2b2). sex composition at birth of the index child, as op- 16. The effect of education is stronger when moth- posed to sex composition of siblings at interview date, ers have completed primary education than when since the latter will be endogenous to mortality. they have not (results not reported). 9. Household size and age composition are poten- 17. It was also possible to test the impact of edu- tially important determinants of child health status, cation via (self-reported) literacy, though coefficients both indirectly as a measure of the ratio of hands to on the literacy variable were insignificant in all regres- mouths available, but also directly by indicating the sions for all three age periods. quantity and quality of care a given child may be able 18. These results are from the multivariate analysis. to receive (Chamarbagwala and others 2004). How- As shown in table C.1, there is higher mortality in ever, most studies, including this one, are unable to lower quintiles than higher, though the differential is include directly indicators of household size to ex- less marked for neo-nates than for children at older plain fertility and mortality differentials because ages. household composition is endogenous to fertility de- 19. Rural is insignificant when interactions be- cisions and mortality. tween month of birth and rural are dropped (results 10. Note that some variables indicate different un- not shown). derlying cause for different age groups. Males have 20. A variable indicating that the household re- higher biological mortality risk than females at all sided in a Bangladesh Integrated Nutrition Project ages, which is reflected in higher observed mortality (BINP) area was included in a antenatal care probit rates among male neonates; higher observed mortal- over the rural 1999/2000 survey round sub-sample (re- ity rates among girl children in South Asia, on the sults not reported)--coefficients were positive but in- other hand, is entirely behavioral, due to son prefer- significant (p-value=0.20). ence. Similarly, short birth interval and higher birth 21. Wald tests can not reject the null hypothesis of order are known to increase mortality risk in (neo- insignificance at p-values 0.20 and 0.14, respectively. natal) infancy, possibly for biological reasons such as 22. There is no statistical evidence for endogeneity physical depletion of the mother, but for older chil- between quality antenatal care and neonatal mortality, dren effects will be entirely behavioral, and probably indicated by the insignificance of RHO 2,1 (a Wald test modified by sex of the child. for insignificance does not reject the null hypothesis 11. The advantage of the Cox semi-parametric at p-value 0.98). estimator is that, in comparison to fully parametric 23. Interactive terms between division of residence proportional hazards models such as the Weibull, as- and gender are not significant determinants of post- sumptions do not have to be made about the baseline natal mortality (results not reported). hazard. 24. The agency index is the simple sum of indica- 12. For immunization of children, the variable used tors of whether the woman has final say on own and in this paper is the geographical mean (by cluster and children's health care, household purchases for large gender) of children immunized. This is because im- items and daily needs, what should be cooked each munization data are only collected on children who day and visits to family, friends or relatives. 2 1 2 E N D N O T E S 25. Another possibility is that wealthier house- out, but not good at identifying poor households holds were better able to benefit from the BINP. How- (Grosh and Glewwe 2000). Under these circum- ever, an interactive term between BINP and wealth is stances, wealth may be a better indicator of differ- insignificant, while the interaction between BINP and ences in current income across households, and mother's education remains significant. The same re- assets are a better predictor of this. The problem with sult is found for child mortality. wealth is that the data collected on assets are usually 26. As well as the coefficients, hazard rates are re- limited (e.g., do not include financial savings), and in ported, indicating the effect of a unit change in the ex- poorer countries, all the better-off have most or all of planatory variable on the risk of death in childhood. the items and many of the poorest have none, making 27. A test for proportional hazards based on Scho- it difficult to distinguish between them. enfeld residuals indicates the null hypothesis (that the 3. Young Lives (2002) gives a formula for scaling a proportional hazards assumption is not violated) is variable X between 0 and 1, where welfare increases not rejected for each explanatory variable separately, with X: while the joint test for all variables does not reject the Xi* = ( Xi - Xmin) / ( Xmax - Xmin) null at p-value=0.62. 28. For neonatal mortality regressions, period 2 where Xi* denotes the scaled observation of individ- (mid-late-1990s) contains 88 more observations than ual i in variable X, and min and max refer to the re- period 1 (early-mid-90s), therefore, in order to match spective minimum and maximum of variable X. To observations from each period, 88 observations are ensure spatial and temporal comparability, the min- omitted, at random, from the period 2 sub-sample. ima and maxima may have to be pre-set rather than For postnatal mortality regressions, period 2 contains determined by the data. 1,009 fewer observations than period 1; these obser- 4. Asset scores are calculated according to the for- vations for period 1 are deleted at random. mula gik = fk (ai ­ a)/s, where gik represents the asset 29. Appendix C.3 presents decompositions of score of asset k for household i, fk is the "raw" asset changes in socioeconomic inequality in mortality by score generated by the principal components analy- changes in inequality in the determinants of mortality, sis, and (ai ­ a)/s is the standardized value of asset vari- based on the regression results. The most important able ai using mean Ç and standard deviation s variables explaining reduction in socioeconomic in- (Gwatkin and others 2000, Annex B). equality in mortality among neonates are birth order 5. Young lives is collecting panel data with specific and electricity. For children, the reduction in wealth interest in children's well-being in Ethiopia, India, inequality contributed the most of any single variable Peru, and Vietnam; see www.younglives.org.uk to reducing inequality in mortality, followed by the 6. The sub-categories may also misclassify. For ex- improved distribution of education among women. ample, bicycles and motor vehicles may serve produc- tive purposes; livestock and land represent livelihood Appendix C.2 and production activities, but may also be used as a form 1. Bollen and others (1999, p. 4) state that the term of savings to insure against adverse circumstances, or as socioeconomic status "appears to have become com- collateral against credit, and therefore fulfill a similar monplace only after the U.S. Census Bureau pub- role to financial assets. Indeed, it may be instructive to lished a report in 1964 using a composite SES index classify assets according to function, since items are held for comparison of different groups." for different purposes--consumption, production, and 2. There is also the problem of lumpy consump- savings, with some items simultaneously fulfilling dif- tion items--expenditure data collected in household ferent functions. In this context, the term "assets" seems surveys will be sensitive to large non-recurring con- more appropriate for production and savings, and it is sumption items such as healthcare bills, which would for this reason that Young Lives (2002) distinguishes a tend to artificially inflate the household's estimated "wealth index" (which includes only household items) welfare, particularly if during the preceding or subse- from an "asset index" (which includes items used in pro- quent period the household saved for the item by re- duction and/or as a store of value), where both indices ducing expenditure. While measures are usually taken can include consumer durables. when constructing household expenditure aggre- 7. Physical contact with certain types of fuel can be gates to exclude some non-recurring items, it has hazardous to health, as can environmental pollution often been argued that because of these problems generated by burning fuel in unventilated areas. household expenditure data are satisfactory for calcu- 8. In principle, where the locational nature of as- lating poverty levels, as these differences get ironed sets varies by country or within country by region, 2 1 3 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? cluster-level data could be used to determine if a 17. In a recent field visit, it was observed that urban household potentially has access to a service (water, houses (and rural houses near main roads) were electricity, etc.), and the asset included in the index mainly constructed of corrugated metal sheets, while for those clusters with access and excluded for those the main building material in rural areas were mud without, with suitable rescaling across assets to en- and thatch (Howard White, personal communica- sure comparability across all clusters. tion). In addition, some mud buildings were observed 9. In order for different SES indicators to produce to be actually brick with mud daubing (which is used different estimates of socioeconomic inequality in an- for traditional decorative patterns)--whereas usually other variable, the indicators must produce different enumerators fill in construction material from own rankings of households and these rank differences observation so these houses may be misclassified. must be correlated with the welfare outcome (Wag- 18. Unweighted means of, or proportions of staff and Watanabe 2002). households owning, variables are as follows: radio 10. Inherent subjectivity associated with choosing (0.303); TV (0.137); bicycle (0.189); almirah (ward- assets to be included would also tend to favor inclu- robe) 0.265; table (0.570); clock (0.471); cot (0.787). sion of all available assets (Gwatkin and others 2000), 19. The poverty index calculates poverty P using though differentiation according to asset category-- the formula: e.g., household assets, community assets--can limit P() = 1/ n.i (z - Ai ) / z ( ) the degree of subjectivity. 11. In PCA the structure of the model is deter- where z represents the poverty line. The most com- mined by the variance of the data; in FA the structure mon used values of are 0 (the poverty headcount is determined by the covariances (or correlations) index), 1 (the poverty gap index) and 2 (the squared between variables. poverty gap index). 12. PCA has the advantage that, exclusion of higher order principal components does not theoretically re- Appendix C.3 sult in mis-specification (omitted variables bias) in 1. The alternative decomposition is MR = multivariate analysis, since the principal components MRp.p1 + MRn.(1­p1) + 2Dp.(MR2 ­MR2 ). When p n are constructed as orthogonal to (i.e., not correlated applied to the data, this method accords more weight with) one another; bias may still arise through corre- to reductions in mortality among the poor. lation of the omitted principal component with other 2. The decomposition is of pt.MRt + (1­pt).MRt as p n explanatory variables (Filmer and Pritchett 1999, opposed to the total mortality rate since the synthetic p. 118, fn. 8). cohort method used to calculate mortality probabili- 13. Wagstaff (2000) notes that deciles could be used ties takes a different sample for each group. in principle, though sample size is often too small to 3. It is supportive, though not conclusive, as there make this classification statistically meaningful. may be economic improvement for those who none- 14. Given whether the shock is covariate or idio- theless remain poor. But expenditure data from syncratic (a covariate shock inducing many house- household surveys show rising inequality, with much holds simultaneously to try to sell their assets would higher average growth for the top 50 percent than the depress prices). bottom 50 percent (World Bank 2002). 15. It could be argued that wall and roof type have 4. The t-statistic is calculated by: direct implications for health and nutrition--though t =(C2 - C1 ) (C2 ) + Var(C1 ) less substantially than flooring--by increasing expo- Var sure to infection. Nevertheless, given the limited al- where C1 and C2 represent the estimate of the con- terative assets available for the index, these variables centration index in 1980-84 and 1995-99 and Var is the were included in the index. variance operator. 16. The weighting scheme that would maximize vari- ance determines each score according to the variance Annex D of the asset, equal to P.(1­P), where P is the proportion 1. Anthropometry measures the body's quantita- owning the asset; this accords greatest weight to the as- tive change in response to changes in the balance be- sets with largest variation in ownership. This scheme tween net energy requirements and net energy intake was deemed unsuitable for the asset index, since it pro- from food. duced scores that appeared unreasonable; for example, 2. According to Chen and others (1980), in a ownership of a table and clock each received scores longitudinal study, weight for age was found to be the almost twice as large as ownership of a television. strongest predictor of mortality, followed by height 2 1 4 E N D N O T E S for age and finally weight for height. Of course, short 12. To see the intuition behind this, note that a children may well be more at risk where height indi- model estimating P(Yi=1)=(Zi) generates inverse cates low birth weight or a history of sickness. Mill's ratio (Zi)/ (Zi). It follows that the model esti- 3. The reference population is the NCHS/WHO mating P(Yi=0)=1­(Zi) leads to Mill's ratio -(Zi)/ (1978) reference. (1­(Zi) due to symmetry of the standard normal dis- 4. Z scores are preferable to other measures, such tribution. as the percentage of the reference median, because 13. Comparable trends are found when tabulating they account for distributional variation specific to dif- incidence of stunting, wasting, and underweight by ferent ages or heights in the reference population. these variables. Children with z scores between -2 and -3 are chron- 14. Modeling anthropometry at ages below ically malnourished and those below -3 z scores 6 months is complicated by problems in physically severely malnourished. measuring young babies with accuracy; also the 5. Using multivariate analysis of panel data, Bairagi WHO/NCHS reference population was bottle-fed, (1986) finds evidence that nutrition discrimination while WHO now recommends breastfeeding during against girls increases following famine. this period. 6. Controlling for education and health knowledge 15. This is a crude estimate of the BINP effect; a in the regression equation should proxy some com- more comprehensive analysis is provided in Annex G. ponent of this heterogeneity. 16. The treatment effects regression is discussed in 7. The wealth index is calculated as the weighted Greene (2000). sum of household durable goods ownership and qual- 17. No impact of contraceptive knowledge and lit- ity of housing (wall and roof) building materials. This eracy on weight for age could be found when these excludes factors that are expected to have "direct" ef- variables were included in the regressions. fects on child health and nutrition, such as drinking water source, sanitation, and dwelling flooring, in Annex E order that these effects can be estimated separately. 1. The following calculations were made including 8. Income, maternal education and access to clean only "usual residents" from the household roster. water and sanitation may potentially be substitutes in Those people identified as not related to the house- the production of child health. This was tested for in hold head were not included since, as mentioned regression analysis using interactive regressors, and above, these people are most likely live-in servants. no significant impact was found. 2. As discussed below, DHS 1999/2000 contains 9. For Peru, Alderman and Behrman (2003) esti- data on if a woman makes the decision on what to mate positive community externalities in female edu- cook each day by herself. Where this is the case for cation, water, and sanitation on top of the observed more than one woman, then this fact may indicate a effects for each child by including geographical area joint household. non-self means of these variables in the set of ex- 3. For example, there are 70 women in the dataset planatory variables. The meta-analysis found commu- aged over 35 and living with a parent, half of these nity-level access to clean water to be marginally were formerly married--and the other half are cur- significantly positive, though sanitation was insignif- rently married, there are no never-married women in icant (Charmarbagwala and others 2004). this group (there are only 19 never-married women 10. Their data are the 1981-92 Nutrition Survey of from the 6,209 aged over 35 in the whole dataset). Rural Bangladesh, which is a nationally representative 4. The age of the oldest son is given as 50. The ages sample of 385 households. of the adults--reported as 30, 35, 40, 45, 50, 60, and 11. The sign of the estimated coefficient on the in- 80--clearly suffer from what is called "heaping," i.e., verse Mill's ratio reflects the correlation between rounding by respondents to the nearest 5 or 10. error terms in selection and regression equations, Nonetheless, it seems likely that the oldest son at least providing statistical evidence for non-random selec- is the product of a previous marriage. tion (Greene 2000). In the case of nutrition, the coef- 5. See White (2001) for more discussion of this ficient is expected to be positive, indicating that the contradiction in findings. unobserved characteristics determining survival, such 6. There is a strict monotonic inverse relationship as preferences regarding child care, broadly defined, between asset quintile and the proportion of women are also positively correlated with those influencing in either paid or unpaid work. Twenty-five percent of nutrition. women in the poorest quintile engage in paid work, 2 1 5 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? and 4.6 in unpaid. For the top quintile these figures 2. The CNC is a facility provided by the community, are 14.8 and 0.9 percent, respectively. it may sometimes be the health clinic, but is more usu- 7. The age variables are more significant than one ally at the home of one of the Community Nutrition measuring the age gap between husband and wife, Committee members. which is only significant when age is not included. 3. It is also more common for a woman to be living 8. Analysis of other data shows that respondents with their mother-in-law in these thanas, being so for saying they can read the paper often prove to have 15 and 12 percent of respondents in Rajnagar and Sha- poor literacy, which will undermine the power of this rasti respectively, compared to 9 percent on average. variable. 4. The BINP dataset does not contain data on dis- 9. Dummies were initially included for all regions tances to health facilities. Using data of the Bang- with Dhaka as the reference region. Those with ladesh Household Income and Expenditure Survey insignificant coefficients were dropped in the results (HIES) of 2000-01, it was found that the time distance reported here. to the health provider sought for treatment was sub- Annex F stantially longer for household not having access to 1. The fitted line is FERT=10.8-0.33 AGEM, drinking water (i.e., accessing water from rivers, R2 = 0.24, n =126. ponds, rain, etc.). The difference was statistically sig- 2. The median age at marriage has shown only a nificant using a t-test. small change. But this statistic is a problematic mea- 5. A similar figure for Type II error was reported in sure as it is necessarily based on ever-married women, the Cambridge study. excluding from the calculation women not yet mar- 6. Save the Children data show that most mothers ried. It is also insensitive to movements within the cannot interpret growth charts. Proponents of growth bottom half of the distribution. monitoring question whether this is necessary for the 3. Alternatively, OLS can be used with a spline func- approach to work, though Christiaensen and Alder- tion, which gives age_fb= 5.24 + 0.83 age_marr man (2004) have found a mother's understanding of + 10.2 dum--0.82 sldum, where dum=1 for women her child's growth performance to be a significant de- married at 13 or less, and sldum is the slope dummy terminant of child nutrition in Ethiopia. The World for that variable for these women. Bank Nutrition Toolkit No. 4, "Promoting the Growth 4. Of course, any delay will have a temporary of Children: What Works" (Griffith, Dickin and Favin tempo effect on the TFR. 5. Until some generations ago, bridewealth was 1996), states that mothers helping weighing the child practiced in Bangladesh, but has now been commonly and interpreting the growth pattern are part of a good replaced by dowry which was previously only found growth monitoring program (p. 87), as this increases among Hindus. The former practice is consistent with the mother's understanding of child growth and of its Islam whereas the latter is not. importance (p.35). 7. Equal to one if any household member is in- Appendix F.1 volved in an association, savings group, or income- 1. The GRR is the average number of daughters generating activity. that would be born alive to a woman (or group of 8. That is, women could be in receipt of supple- women) during her lifetime based on currently pre- mentary feeding without acquiring nutritional infor- vailing age if she passed through her childbearing mation. This finding is not a good one for BINP, since years conforming to the age-specific fertility rates. supplementary feeding was intended mainly as a tool 2. GRR= a + b CRW, where a and b are given in for demonstrating feeding practices, not as a feeding Rele (1976). The TFR= GRR x 2.05 (assumed sex ratio program. However, it was already shown that one- at birth). third of mothers whose children were enrolled in sup- 3. Suggested weights to arrive at a final estimates plementary feeding did not discuss nutrition with the are 0.7 for GRR obtained from CWR2 and 0.3 for GRR obtained from CWR1. CNP. 4. Results of the 2001 population census are provi- 9. This variable is obtained using the information sional ones calculated based on 5 percent sample on the date of interview and the age of the mother's schedule. Final results are yet to be published. child. A woman is considered as pregnant during the working season if the last six months of pregnancy Annex G were mostly in the months between November and 1. This study focuses on protein-energy malnutri- April. tion, as do other studies of BINP. Data have not been 10. A longer version of this section is available on collected to analyze micronutrient components. the study website. 2 1 6 E N D N O T E S 11. All malnutrition prevalence rates have been cal- irregular participation, and 2 for continuous participa- culated using the old NCHS reference population tion. The predicted probability are used to weight the growth charts. The National Nutrition Project (NNP) observations. Outcomes of children whose mother has adopted the new (2000) NCHS growth charts in made the choice 2 are weighted by the inverse of the 2004. The introduction of the new charts will have predicted probability of that choice. The project effect predictable effects on the calculation of malnutrition is calculated taking the difference between this value rates. and the mean outcomes of the children who did not 12. There are little more than 150 control observa- participate. tions in the baseline survey (see last column of table 21. Tables G.21 and G.22, and the charts in figure G.17 for actual figures). G.3 are based on matched observations of BINP eval- 13. The fact that knowledge has not risen in con- uation sample and NSP data, performed with the one- trol thanas suggest that contamination is not a serious to-one matching method. problem. 22. The category of normally nourished is not 14. The text of the IMED report claims a larger im- shown in table G.24 as they constitute less than 1 per- pact if considering only the weight for height propor- cent of program participants. tion of children aged 0-23 months. 23. These are the quantitative targets defined in 15. One possibility, which cannot be explored with the project document (World Bank 1995, p. 15). It is the data to hand, is that improved nutrition was ex- not clear how the latter can be measured, since the pended in extra physical activity rather than weight same women will not be pregnant at midterm and gain. endline as were at baseline. Karim and others (2003) 16. It is not possible to discern from the Save the rephrased the target as "improvement of weight gain Children questionnaire the children who participated to 7.0 kg. in at least 50% of pregnant women." in GMP from those who did not. In addition, there is 24. See in particular the reviews by Chatterjee little information that can be used to identify partici- (1991) and Nag (1994). pants in a model of participation. Finally, the absence 25. Estimates by NIPORT (2002). of data on the date of the interview is also a concern, 26. It is not clear, however whether the absence of because of the sensitivity of nutritional indicators to an effect of food supplementation program on mater- seasonal effects. nal mortality risk is the result of the absence of an ef- 17. After selecting NSP data of the same months in fect of larger body size on delivery, or of the absence which the BINP data were collected, some discrepan- of an effect of food supplementation on body size. cies in the temporal distribution of the observations 27. In order to measure intrauterine growth it is BINP and NSP data still remains. In order to correct necessary to adjust birth weight by gestational age at this source of seasonal variation in nutritional indica- birth in order to isolate the component influenced by tors, the data were seasonally adjusted. nutrition. Several reference populations are available. 18. It is questionable whether the equation can be The one used here are a Canadian reference popula- estimated using the HKI data since households out- tion (Kramer 2001) for birth weight for gestational side the project area could not participate even if they age. BINP data report only the month of birth as the wanted to. An alternative procedure is to estimate the 8th, 9th, or 10th, while reference weights by gestational participation equation using only the project sample age are normally reported by week. The recom- and then apply those coefficients to the NSP data to mended weight at the 40th week of gestation is used calculate the scores. This approach was also used, as a reference for the 9th month reported weight; the with the final results not being much different from 39th week for the 8th month and the 41st week for the those shown here. 10th month, assuming that the distribution of births in 19. Participation here refers to participation in the the 8th and 10th months tend to concentrate around growth monitoring sessions. An attempt was made to the 40th week. This ratio is not very precise, but it is construct a control for children participating in sup- better to adjust for sex, and partially for gestational plementary feeding. Data on the eligibility criteria are age, than not adjusting at all. not available for the control, so that proxies had to be 28. The turning point is obtained as ­b1/(2*b2), used, but created a poor control. Negligible project ef- where b1 and b2 are the regression coefficients of fect was found. mother's age and mother's age squared respectively. 20. Project effects are estimated by running ordered 29. See the review of studies on the effect of sup- logit models of participation, where the dependent plementation project on low birth weight in Waterlow variable takes three values: 0 for non-participation, 1 for (1992) and Institute of Medicine (1990). These review 2 1 7 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? report the following changes in birth weight: Guat- 2. Net cropped area refers to the area actually emala (+100), Colombia (+ 180, but only for thin cropped, whereas effective gross cropped area refers women), Gambia (+ 120, but only in hungry season), to the total yearly area cropped counting areas under Chile (+ 60), Indonesia (no effect). No effect on birth double and triple cropping two and three times weight was observed in similar projects carried out in respectively. Canada, Scotland, and the United States. 3. Transplanted, broadcast, and deepwater rice are 30. Given the time required to identify pregnant different varieties of rice, which exist within each rice women and enroll them in BINP activities, pregnant sort. Each of these contains various subgroups. women will be covered by project activities for about 4. The study looked at weight for age (WAZ); six months rather than the full nine months of preg- height for age (HAZ); arm circumference (AC) for age; nancy. In any case, weight gain is relatively small in the and triceps skinfold thickness (TSFT) for age. The lat- first trimester (see figure G.5). ter two had better indicators of seasonal variations in 31. Percent changes in malnutrition are calculated nutrition given their independence from height. using a probit function in the following way. If Z is the average z-score before the project and dZ is the Annex J change in the average z-score of table 2, the percent 1. Books, Buildings and Learning Outcomes: an reduction in malnutrition is obtained as (­2 ­ Z) ­ impact evaluation of World Bank support to basic (­2 ­ Z + dZ). Malnutrition rates are calculated education in Ghana, OED, January 2004. using ­2 z-score as cutoff point (this includes mild and 2. 2002 Annual Review of Development Effective- severe malnutrition). ness - Achieving Development Outcomes: The Millen- 32. Data on calories and rice price are obtained nium Challenge, OED, report no. 25159. from the IFPRI data of 1998. 3. Determinants are the variables entering the mul- 33. A large fraction (58 percent) of the nutrition tiple regression analysis, such as mother's education, component of the project is invested on food pur- and drivers the policies and programs which drive chase and preparation. It is estimated that 85 percent changes in these determinants (e.g. policy on school of this food is consumed by mothers enrolled in the fees, and female scholarships). supplementary feeding program. The costs of reduc- 4. UNICEF (1990) Strategy for Improved Nutrition ing malnutrition and of saving one life reported in the of Children and Women in Developing Countries. second column are calculated after subtracting costs New York: UNICEF. that are directed to mothers. 5. W. Henry Mosley and Lincoln C. Chen (1984) 34. Use of current government workers or local- "An Analytical Framework for the Study of Child Sur- level voluntary workers was ruled out of the BINP de- vival in Developing Countries." Population Develop- sign because of the workload. But a large part of this ment Review Vol. 10 Supplement: Child Survival: workload relates to growth monitoring rather than Strategies for Research, 25-45. BCC. 6. For example, J.M. Currie "Child Health in Devel- Annex H oping Countries" in Culyer and Newhouse (eds), 1. The ratio of the total budget for FSSAP to the Handbook of Health Economics Amsterdam, Elevier stipend component is 1.20, giving an overhead of Science BV, 2000. 20 percent. 7. Public expenditure reviews are available from 1990, 1996, 1997 and 2003. Annex I 1. Rice accounted for almost three-quarters of total crop production in the late 1990s (Mian and others 2001, p. 39). 2 1 8 REFERENCES Agarwal, B. 1990. "Gender Relations and Food Secu- Bairagi, R., M. Mazharul Islam, and M. K Barua. 1996. rity: Coping with Seasonality, Drought, and Fam- "Contraceptive Failure: Levels, Trends and Deter- ine in South Asia." PEW/Cornell Lecture Series on minants in Matlab, Bangladesh." Dhaka: Popula- Food and Nutrition Policy, November 13, Wash- tion Studies Center, ICDDR'B. [Photocopy.] ington, DC: CFNPP. Bairagi, R., S. Chandra Sutradhar, and N. Alam. 1999. Alauddin, M., and M. Hossain. 2001 Environment and "Levels, Trends, and Determinants of Child Mor- Agriculture in a Developing Economy: Problems tality in Matlab, Bangladesh, 1966­94." Asia-Pacific and Prospects for Bangladesh. Northampton, MA: Population Journal 14 (2): 51­68. Edward Elgar. Bangladesh Bureau of Statistics. 2000. Statistical Poc- Alauddin, M., and C. A. Tisdell. 1998. The Environ- ketbook of Bangladesh. BBS, Ministry of Planning, ment and Economic Development in South Asia: Bangladesh. An Overview Concentrating on Bangladesh. Lon- Bangladesh Government. 2003. "A National Strategy don: Macmillan. for Economic Growth, Poverty Reduction, and So- Alderman, H., and J. R. Behrman. 2003. "Estimated Eco- cial Development." Economic Relations Division, nomic Benefits of Reducing LBW in Low-Income Countries." University of Pennsylvania, Philadelphia, Ministry of Finance. Dhaka. for Human Development Network/ Nutrition, ------. 2002. "Interim Poverty Reduction Strategy World Bank, Washington, DC. [Photocopy.] Paper." Dhaka. Alderman, H., and M. Garcia. 1994. "Food Security and Bangladesh Government/International Monetary Health Security: Explaining the Levels of Nutri- Fund. 2003. "Bangladesh: Interim Poverty Reduc- tional Status in Pakistan." Economic Development tion Strategy Paper." Washington, DC. and Cultural Change 42 (3): 485­507. Barkat, Abul, S.H. Khan, M. Rahman, S. Zaman, A. Pod- Alderman, H., J. Hentschel, and R. Sabates. 2003. der, S. Halim, N.N. Ratna, M. Majid, A.K.M. Maksud, "With the Help of One's Neighbors: Externalities A. Karim, and S. Islam. 2002. Economic and Social in the Production of Nutrition in Peru." Social Impact Evaluation Study of the Rural Electrifica- Science and Medicine 56 (10): 2019­31. tion Program in Bangladesh. Human Develop- Ateng, B. 1998. "Comparative Advantage and Crop ment Research Centre and NRECA International Diversification." In Bangladesh Agriculture in the Ltd. 21st Century, ed. R. Faruqee. Dhaka: The Univer- Baulch, Bob, and John Hoddinott, eds. 2000. "Eco- sity Press. nomic Mobility and Poverty Dynamics in Develop- Baffes, J., and M. Gautam. 1996. "Is Growth in ing Countries." Journal of Development Studies Bangladesh's Rice Production Sustainable?" World 36. Bank Policy Research Working Paper 1666, Wash- Beaton, G. H, and H. Ghassemi. 1982. " Supplemen- ington, DC. tary Feeding Programs for Young Children in Bairagi, R. 1996. "Is Son Preference an Obstacle to Developing Countries." American Journal of Clin- Fertility Transition in Bangladesh?" In IUSSP Com- ical Nutrition: 863­916. mittee on Fertility and Family Planning, Population Becker, S., and S. Weng. 1998. "Seasonal Patterns of Council Islamabad, Rockefeller Foundation, and UNFPA Pakistan Office, Comparative Perspec- Deaths in Matlab, Bangladesh." International tives on Fertility Transition in South Asia. Liege: Journal of Epidemiology 1999 (27): 814­23. IUSSP. Begum, Sharifa. 2003. "Abortion in Rural Bangladesh: ------. 1986. "Food Crisis, Nutrition, and Female Nature, Causes, Problems, and Practices." BIDS Re- Children in Rural Bangladesh." Population and search Report 174. Dhaka. Development Review 12 (2): 307­15. ------. 1990. "Population Birth, Death, and Growth ------. 1980. "Is Income the Only Constraint on Rates in Bangladesh: Census Estimates." The Child Nutrition in Rural Bangladesh?" Bulletin of Bangladesh Development Studies, Vol.18, No.2. the World Health Organization 58 (5): 767­72. Dhaka. 2 1 9 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? ------. 1976. "An Estimation of Birth and Death Rate in Chatterjee, M. 1991. Towards Better Health for In- Bangladesh, 1951­60. " Bombay, IIPS. [Photocopy.] dian Women: The Dimensions, Determinants, Berg, A. 1983. Nutrition Review. Washington, DC.: and Consequences of Female "Illnesses" and Population and Human Resources Department, "Death." Washington, DC.: World Bank. World Bank. Chaudhury, Nazmul, and Jeffrey Hammer. 2003. Bhuiya, A., and M. Chowdhury. 1997. "The Effect of "Ghost Doctors: Absenteeism in Bangladeshi Divorce on Child Survival in a Rural Area of Health Facilities." World Bank Working Paper Bangladesh." Population Studies 51: 57­61. 3065, Washington, DC. Bhuiya, A., and K. Streatfield. 1991. "Mothers' Edu- Chen, L., A. Chowdhury, and S. Huffman. 1980. "An- cation and Survival of Female Children in a thropometric Assessment of Energy-Protein Mal- Rural Area of Bangladesh." Population Studies 45: nutrition and Subsequent Risk of Mortality among 253­64. Pre-School Aged Children." American Journal of Blanc, Ann. 1990. The Relationship between Contra- Clinical Nutrition 33: 1836­45. ceptive Prevalence and Total Fertility in DHS Chen, L., E. Huq, and S. D'Souza. 1981. "Sex Bias in Countries. Columbia, MD: Demographic and the Family Allocation of Food and Health Care in Health Surveys Project. Rural Bangladesh." Population and Development Bloem, M., R. Moench-Pfanner, and D. Panagides, eds. Review 7: 55­70. 2003. Health and Nutritional Surveillance for De- Chowdhury, M, and R.. Bairagi. 1990. "Son Preference velopment. Singapore: Helen Keller International. and Fertility in Bangladesh." Population and De- velopment Review 16: 4. Bollen, Kenneth A., Jennifer L. Glanville, and Guy Chowdhury, M. A. T., and H. Zulfikar. 2001. An Agri- Stecklov. 2002. "Economic Status Proxies in Stud- cultural Statistical Profile of Bangladesh, 1947­ ies of Fertility in Developing Countries: Does the 199. Bogor, Indonesia: CGPRT Centre. Measure Matter?" Population Studies 56: 81­96. Chowdhury, Zafrullah. 1996. The Politics of Essential ------. 1999. "Socioeconomic Status and Class in Drugs. The Makings of a Successful Health Strat- Studies of Fertility and Health in Developing Coun- egy: Lessons from Bangladesh. Dhaka: University tries." University of North Carolina at Chapel Hill. Press Limited. Bongaart, John. 1978. "A Framework for Analyzing the Christiaensen, L., and H. Alderman. 2004. "Child Mal- Proximate Determinants of Fertility." Population nutrition in Ethiopia: Can Maternal Knowledge and Development Review 4(1). Augment the Role of Income?" Economic Devel- Bongaart, John, and R. G. Potter. 1983. Fertility opment and Cultural Change 52 (2): 287­312. Behavior: Biological Analysis of the Proximate Cleland, John. 1994. The Determinants of Reproduc- Determinants of Fertility. New York: Academic. tive Change in Bangladesh: Success in a Chal- Briend, A., C. Dykewicz, K. Graven., R. Mazumder, B. lenging Environment. Washington, DC.: World Wojtyniak., and M. Bennish. 1986. "Usefulness of Bank. Nutritional Indices and Classifications in Predicting Cleland, John, and Kim Streatfield. 1992. "The Demo- Death of Malnourished Children." British Medical graphic Transition: Bangladesh, UNICEF." Dhaka: Journal 293: 373­75. Program Planning Unit. [Photocopy.] Brown, K. H., R. E. Black, and S. Becker. 1982. "Sea- Cockcroft, Anne, Debbie Milne, and Neil Andersson. sonal Changes in Nutritional Status and the Preva- 2003. Bangladesh: HPSP, 1998­2003. Third Ser- lence of Malnutrition in a Longitudinal Study of vice Delivery Survey 2003. Dhaka: CIET Canada. Young Children in Rural Bangladesh." The Ameri- Croll, E. 2001. Endangered Daughters: Discrim- can Journal of Clinical Nutrition 36. ination and Development in Asia. London: Buhmann et al. 1988. "Equivalence Scales, Well- Routledge. Being, Inequality, and Poverty." Review of Income Davis K., and Judith Blake. 1956. "Social Structure and and Wealth 34: 115­42. (Cited in Wagstaff 2000, Fertility: An Analytical Framework." Economic De- p. 23.) velopment and Cultural Change 4 (2). Caldwell, J., B. Barkat-Khuda, I. Cladwell, I. Pieris, and Del Ninno, C., P. A. Dorosh, L. C. Smith, and D. K. Roy. P. Cladwell. 1999. "The Bangladesh Fertility De- 2001. The 1998 Floods in Bangladesh--Disaster cline: An Interpretation." Population and Devel- Impacts, Household Coping Strategies, and Re- opment Review 25(1): 67­84. sponse. Washington, DC.: International Food Pol- Carvalho, S., and H. White. 2004. "Theory-Based Eval- icy Research Institute. uation: The Case of Social Funds." American Jour- Deolalikar, A. 2002. "Poverty and Child Malnutrition in nal of Evaluation 25(2): 141­60. Bangladesh." Bangladesh Poverty Assessment Chamabagwala, Rubiana, Martin Ranger, Hugh Background Paper 7. Waddington, and Howard White. 2004. "The De- EM-DAT. 2004. OFDA/CRED International Disaster terminants of Child Health and Nutrition: A Meta- Database, www.em-dat.net. Brussels: Université Analysis." OED Working Paper, Washington, DC. Catholique de Louvain 2 2 0 R E F E R E N C E S Fairlie, R. 2003. "An Extension of the Blinder-Oaxaca Standards Measurement Study 3 Vols. Washing- Decomposition Technique to Logit and Probit ton, DC: World Bank. Models." Economic Growth Center Discussion Gwatkin, Davidson R., Shea Rustein, Kierston John- Paper 873. New Haven, CT: Yale University. son, Rohini P. Pande, and Adam Wagstaff 2000. Faruqee, R., ed. 1998. Bangladesh Agriculture in the "Socio-Economic Differences in Health, Nutrition 21st Century. Dhaka: The University Press Ltd. and Population in Bangladesh." HNP/Poverty The- Faveau, V., A. Briend., J. Chakraborty, and A. Sarder. matic Group of the World Bank, Washington, DC. 1990. "The Contribution of Severe Malnutrition to Haddad, L. H. S. Alderman, L. Appleton, L. Song, and Child Mortality in Rural Bangladesh: Implications Y. Yohannnes. 2003. "Reducing Malnutrition: How for Targeting Nutritional Iinterventions." Food Far Can Income Growth Take Us?" World Bank and Nutrition Bulletin 12: 215­22. Economic Review 17 (1). Faveau V., M. A. Koenig, J. Chakraborty, and A. I. Haider, S. J, D. Hussain, I. Nayer, A. Maleque, Ashaduz- Chowdhury. 1988. "Causes of Maternal Mortality in zaman, and A.R. Khan. 2004. "Impact Evaluation of Rural Bangladesh, 1976­85." Bulletin of the World Bangladesh Integrated Nutrition Project." Dhaka: Health Organization 66 (5): 643­51. IMED and Read. Filmer, Deon, and Lant Pritchett. 1998. "Estimating Heckman, J., and B. Singer. 1984."Econometric Dura- Wealth Effects without Income or Expenditure tion Analysis." Journal of Econometrics 24 (1­2): Data--Or Tears: Educational Enrolment in India." 63­132. World Bank Policy Research Working Paper 1994, Horton, S. 1986. "Child Nutrition and Family Size in Washington, DC the Philippines." Journal of Development Eco- Filmer, Deon, and Lant Pritchett. 1999a. "The Effect of nomics 23: 161­16. Household Wealth on Educational Attainment: Hossain, Shahadat, et al. 1991. "Fertility in Bangla- Evidence from 35 Countries." Population and desh: Levels and Trends." In Proceedings of the Development Review 25 (1): 85­120. Workshop on Dissemination of Current Statistics, ------. 1999b. "The Impact of Public Spending on November 3­4, 1991, Dhaka. Health: Does Money Matter?" Social Science and Hossain, M., D. Lewis, M. L. Bose, and A. Chowdhury. Medicine 49 (10): 1309­23. 2003. "Rice Research, Technological Progress, and Fine, P. 1993. "Herd Immunity: History, Theory, Prac- Impacts on the Poor: The Bangladesh Case (Sum- tice." Epidemiology Review 15: 265­302. mary Report)." EPTD Discussion Paper 110. Wash- Foster, A., J. Greer, and E. Thorbecke. 1984. "A Class ington, DC: IFPRI. of Decomposable Poverty Measures." Econo- Hossain, Z., and A.F. H. Choudhury. 1987. Folk Di- metrica 52: 761­66. etary Practices and Ethnophysiology of Pregnant Gage, A., E. Sommerfelt, and A. Piani. 1996. "House- Women in Rural Banglades. Michigan State Uni- hold Structure, Socioeconomic Level, and Child versity Working Paper 137, Ann Arbor. Health in Sub-Saharan Africa." Demographic and Houweling, Tanya, Anton Kunst, and Johan Macken- Health Surveys Analytical Reports 1. bach. 2003. "Measuring Health Inequality Among Gomez, F., R. Galvin, S. Frenk, J. Cravioto, R. Chavez, Children in Developing Countries: Does the and J. Vasquez. 1956.. "Mortality in Second and Choice of the Indicator of Economic Status Mat- Third Degree Malnutrition." Journal of Tropical Pediatrics 2: 77­83. ter?" International Journal for Equity in Health Goodburn, Elizabeth A., Mushtaque Chowdhury, 2(1): 8­21. Rukhsana Gazi, Tom Marshall, and Wendy Gra- Howlader, A. A., and M. U. Bhuiyan. 1999. "Mothers' ham. 2000. "Training Traditional Birth Attendants Health-Seeking Behaviour and Infant and Child in Clean Delivery Does not Prevent Postpartum In- Mortality in Bangladesh." Asia-Pacific Population fection." Health Policy and Planning 15 (4): Journal 14 (1): 59­75. 394­99. Huq, Md. Nazmul, and John Cleland. 1990. Green, W. 1998. "Gender Economics Courses in Lib- Bangladesh Fertility Survey­1989. Dhaka: NIPORT. eral Arts Colleges: Further Results." Journal of Institute of Medicine. 1990. Nutrition During Preg- Economic Education. nancy. Washington DC: . National Academies Press. Greene, W. H. 2000. Econometric Analysis, 4th ed. International Rice Research Institute. 2004. http:// New York: Prentice-Hall. www.irri.org/science/ricestat/index.asp Griffiths M., K. Dickin, and M. Favin. 1996. Promoting Islam, M. A, M. M. Islam, and N. Chakraborty. 2001. the Growth of Children: What Works. Rationale "Plateauing of Fertility Level in Bangladesh: Ex- and Guidance for Programs. Washington, DC: ploring the Reality." Paper 18. Dhaka: CPD-UNFPA World Bank. Program on Population and Sustainable Develop- Grosh, M., and P. Glewwe. 2000. Designing House- ment, CPD. hold Survey Questionnaires for Developing Jaim, W. M. H. 1984. "Changes in Output and Em- Countries: Lessons from 15 Years of the Living ployment due to New Technology: A Study in an 2 2 1 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? Irrigated Area of Bangladesh." Economic Affairs Mortality in Rural Bangladesh." Demograph 27 (2): 29: 35­42. 251­65. Kabeer, Naila. 2002. The Power to Choose: Bangla- Kramer M. S. 2001. "A New and Improved Population- deshi Garment Workers in London and Dhaka. Based Canadian Reference for Birth Weight for London: Verso. Gestational Age." Pediatrics (August). ------. 2001. "Ideas, Economics, and the Sociology ------. 1987. "Determinants of Low Birth Weight: of Supply: Explanations for Fertility Decline in Methodological Assessment and Meta-Analysis." Bangladesh." Journal of Development Studies. Bulletin of the World Health Organization 65 (5): Kabir M., and Abdullah A. Mamun Chowdhury. 2004. 663­737. "Plateauing Fertility in Bangladesh: Correlates and Lavy, V. J. Strauss, D. Thomas, and P. de Vreyer. 1996. Proximate Determinants." In Fertility Transition "Quality of Health Care, Survival, and Health Out- in Bangladesh: Evidence and Implication, Rob comes in Ghana. " Journal of Health Economics Ubaidur, N. Ameen, and N. Piet Pelon, eds. Dhaka: 15: 333­57. UNFPA. Lee, L.-f., M. Rosenzweig, and M. Pitt. M. 1997. "The Ef- Kabir, M., and R. Amin. 1993. "Factors Influencing fects of Improved Nutrition, Sanitation, and Water Child Mortality in Bangaldesh and their Implica- Quality on Child Health in High-Mortality Popula- tions for the National Health Programme." Asia- tions." Journal of Econometrics 77, 209­35. Pacific Population Journal 8 (3): 31­46. Levinson, M., O. Halpern, Z. Mahmud, S. A. Chowd- Kakwani, N. 1980. Income Inequality and Poverty: hury, and J. F. Levinson. 2002. Nutrition-Related Methods of Estimation and Policy Applications. Caring Practices and Women's Time Constraints: New York: Oxford University Press. A Study in Rural Bangladesh. TFTS Nutrition, Kakwani, N., A. Wagstaff, and E. van Doorslaer. 1997. Food Policy and Applied Nutrition Program, Dis- "Socioeconomic Inequalities in Health: Measure- cussion Paper 18. ment, Computation and Statistical Inference." Levitt, M. J. 1999. "When the Training of TBAs is Cost Journal of Econometrics 77 (1): 87­104. Effective: Trained TBAs and Neonatal Essential Kantner, A., and Elizabeth Frankenberg. 1986. "Levels Care in South Asia." In Improving Newborn Infant and Trends in Fertility and Mortality in Bangla- Health in Developing Countries, Anthony Costello desh." Dhaka: Bangladesh Contraceptive Preva- and Dharma Manandhar, eds. London: Imperial lence Survey­1985: Secondary Analysis, Mitra and College Press. Associates. Mabud, M. A. 1987. "Levels and Trends in Fertility and Kantner Andrew, and Ali Noor. 1992. "Population Dy- Demographic Prospects in Bangladesh." In Recent namics in Bangladesh: Current Demographic Esti- Trends in Fertility and Mortality in Bangladesh, mates and Future Prospects." Dhaka: East­West M. A. Mabud, ed. Proceedings of a National Semi- Center and Office of the Population and Health, nar, June 11­13, 1987: Dhaka: PDEU, Planning USAID. [Photocopy.] Commission. Karim, R., S. A. Lamstein, M. Akhtaruzzaman, K. M. Mackenbach, J. P., and A. E. Kunst. 1997. "Measuring Rahman, N. Alam. 2003. The Bangladesh Inte- the Magnitude of Socio-Economic Inequalities in grated Nutrition Project: Endline Evaluation of Health: An Overview of Available Measures Illus- the Community-Based Nutrition Component. The trated with Two Examples from Europe." Social Institute of Nutrition and Food Sciences University Science and Medicine 44, 757­71. of Dhaka, Bangladesh, and The Friedman School Madise, N. .J., Z. Matthews, and B. Margetts. 1999. of Nutrition Science and Policy, Tufts University. "Heterogeneity of Child Nutritional Status Be- Khan, M. A., and P. A. Khanum. 2000. "Influence of tween Households: A Comparison of Six Sub- Son Preference on Contraceptive Use in Bangla- Saharan African Countries." Population Studies 53: desh." Asia-Pacific Population Journal, Septem- 331­43. ber: 43­54. Mahmud, W. 1998. "Comparative Advantage and Crop Khan, M. M., and R. A. Yoder. 1998. "Expanded Pro- Diversification." In Bangladesh Agriculture in the gram on Immunization in Bangladesh: Cost, Cost- 21st Century, R. Faruqee, ed. Dhaka: The Univer- Effectiveness, and Financing Estimates." Special sity Press Ltd. Initiatives Report 6. Bethesda: Partnerships for Mahmud, W., S. H. Rahman, and S. Zohir. 1994. Agri- Health Reform. cultural Growth through Crop Diversification in Kishor, Sunita, and Sulabha Parasuraman. 1998. Bangladesh. Working Papers on Food Policy in "Mothers' Employment and Infant and Child Mor- Bangladesh 7. Washington, DC: IFPRI. tality in India." National Family Health Survey Martorell, R., and M. Shekar. 1992. Growth-Faltering Subject Reports 8. Rates in California, Guatemala, and Tamil Nadu: Koenig, Michael, James Phillips, Oona Campbell, and Implications for Growth-Monitoring Programs. Stan D'Souza. 1990. "Birth Intervals and Childhood Food and Nutrition Bulletin 5 (3). 2 2 2 R E F E R E N C E S Mason, J. B, J. Hunt, D. Parker, and U. Johnson. 2001. National Environment Agency, Singapore. http://app. "Improving Child Nutrition in Asia." Food and Nu- nea.gov.sg/cms/htdocs/article.asp?pid=1099 trition Bulletin 22 (3) (Suppl.). Tokyo: United Na- NIPORT (National Institute of Population Research tions University Press.. and Training). 2002. "Bangladesh Maternal Health Masset, E., and H. White. 2003. "Infant Mortality in Services and Maternal Mortality Survey 2001, Andhra Pradesh, Analyzing Changes over Time and Measure/DHS+." Between States." Young Lives Working Paper 6. NIPORT (National Institute of Population Research Masset, Edoardo, and Howard White. 2002. "Infant and Training), Mitra and Associates, and ORC and Child Mortality in Andhra Pradesh: Analyzing Macro. 2001. Bangladesh Demographic and Health Changes over Ttime and Between States." Young Survey: 1999 ­ 2000. Lives Working Paper 8. OED (Operations Evaluation Department). 2004a. Mian, W. M. A. (ed). 2001. Agricultural Research in "Impact Evaluation of External Support on Mater- Bangladesh in the 20th Century. Dhaka: Bangla- nal and Child Health in Bangladesh." Design Paper, desh Agricultural Research Council and Bangla- May 2004. Washington, DC: The World Bank. desh Academy of Agriculture. [Photocopy.] Miller, Jane E. 1989. "Is the Relationship between ------. 2004b. Books, Buildings, and Learning Out- Birth Intervals and Perinatal Mortality Spurious? comes: An Impact Evaluation of World Bank Sup- Evidence from Hungary and Sweden." Population port to Basic Education in Ghana. Howard White, Studies 43 (3): 479­95. task manager. Washington, DC: World Bank. Ministry of Health and Population Control, Bangla- Palloni, A., and M. Tienda. 1986. "The Effects of Breast- desh. 1978. World Fertility Survey: Bangladesh feeding and Pace of Childbearing on Mortality at Fertility Survey, 1975, First Report. Dhaka. Early Ages." Demography 23 (1): 31­52. Mitra, S., A. Al-Sabir, A. Cross, and K. Jamil, K. 1997. Pedersen, K. M. 2002. "The World Health Report 2000: Bangladesh Demographic and Health Survey, Dialogue of the Deaf?" Health Economics 11: 1996­97, Calverton, MD: National Institute of Pop- 93­1003. ulation Research and Training (NIPORT), Mitra Pelletier, D. L., and E. A. Frongillo. 2003. "Changes in and Associates and Macro International Inc. Child Survival Are Strongly Associated with Mitra, S., M. Ali, S. Islam, A. Cross, A., and T. Saha. Changes in Malnutrition in Developing Countries." 1994. Bangladesh Demographic and Health Sur- Journal of Nutrition 133: 107­ 19. vey, 1993­94, Calverton, MD: National Institute of Pitt, M., and S. R. Khandker. 2002. "Credit Pro- Population Research and Training (NIPORT), Mitra grammes for the Poor and Seasonality in Rural and Associates and Macro International Inc. Bangladesh." Journal of Development Studies Moore, Mick. 2003. "Politics and the Drivers of Pro- 39 (2). Poor Change in Bangladesh." Report to DFID. Podja, J., and L. Kelley. 2000. Low Birth-Weight. ACC/ Dhaka. SCN (Administrative Committee on Coordination/ Mosley, W., and L. Chen. 1984. "An Analytical Frame- Sub-Committee on Nutrition), Nutrition Policy work for the Study of Child Survival in Developing Paper 18. Countries." Population and Development Review Rabbani, AKMA Ghulam, and Shahadat Hossain. 1984. [Suppl. to Vol. 10]. "Levels of Fertility and Mortality from the National Mozumder, A., Barkat-E-Khuda, T. Kane, A. Levin, and Sample Vital Registration System." Paper pre- S. Ahmed. 2000. "The Effect of Birth Interval on sented at the National Seminar on Recent Trends Malnutrition in Bangladeshi infants and Young in Fertility and Mortality in Bangladesh January Children. Journal of Biosocial Science 32, 289­ 29­31, 1984. Dhaka: PDPU, Planning Commission. 300. Razzaque, A., N. Alam, L. Wai, and A. Foster. 1990. Muhiri, P., and J. Menken. 1997. "Adverse Effects of "Sustained Effects of the 1974­75 Famine on Infant Next Birth, Gender, and Family Composition on and Child Mortality in a Rural Area of Bangladesh." Child Survival in Bangladesh." Population Studies Population Studies 44: 145­54. 51 (3): 279­94. Rele, J. R. 1987. "Fertility Levels and Trends in India, Muhuri, P., and S. Preston. 1991. "Effects of Family 1951­81." Population and Development Review Composition on Mortality Differentials by Sex 13 (3). among Children in Matlab, Bangladesh." Popula- ------. 1976. Fertility Analysis Through Extension of tion and Development Review 17 (3): 415­33. Stable Population Concepts. Westport, CT: Green- Muhuri, R. 1996. "Estimating Seasonality Effects on wood Press Child Mortality in Matlab, Bangladesh." Demogra- Reutlinger S., and M. Selowsky. 1976. Malnutrition phy 33 (1): 98­110. and Poverty. World Bank Staff Occasional Paper Nag, M. 1994. "Beliefs and Practices about Food Dur- 23. Washington, DC. ing Pregnancy." Economic and Political Weekly, Rogal, B., B. W. Harriss-White, and S. Bose. eds. 1999. September 10: 2427­38. Sonar Bangla? Agricultural Growth and Agrar- 2 2 3 M A I N T A I N I N G M O M E N T U M T O 2 0 1 5 ? ian Change in West Bengal and Bangladesh. UNICEF. 2002. State of the World's Children 2003. Thousand Oaks, CA: Sage Publications. New York. Rosenzweig, M., and T. P. Schultz. 1983. "Estimating a ------. 1990. Strategy for Improved Nutrition of Household Production Function: Heterogeneity, Children and Women in Developing Countries. the Demand for Health Inputs, and their Effects on New York. Birth Weight." Journal of Political Economy 91 Vemury, M. 1981. Rural Food Habits in Six Devel- (5): 723­46. oping Countries: CARE Study of Environmental, Ross, J., and E. Frankenberg. 1993. Findings from Two Social, and Cultural Influence on Food Con- Decades of Family Planning Research. The Popu- sumption Patterns. New York: CARE. lation Council. Wagstaff, A. 2000a. "Unpacking the Causes of Inequalities Rutstein, S. 1984. "Infant and Child Mortality: Levels, in Child Survival: The Case of Cebu, The Philippines." Trends, and Demographic Differentials." WFS Washington, DC, The World Bank. [Photocopy.] Comparative Studies 43. Voorburg, The Nether- ------. 2000b. "Socioeconomic Inequalities in Child lands: International Statistical Institute. Mortality: Comparisons Across Nine Developing Sahn, David, and David Stifel. 2000. "Poverty Compar- Countries." Bulletin of the World Health Organi- isons Over Time and Across Countries in Africa." zation 78 (1). World Development 28 (12): 2123­55. Wagstaff, Adam, and Naoko Watanabe. 2002. "What Sastry, Narayan. 1996. "Community Characteristics, Difference Does the Choice of SES Make in Health Individual and Household Attributes, and Child Inequality Measurement?" World Bank Develop- Survival in Brazil." Demography 33 (2): 211­29. ment Research Group, Washington, DC. Save the Children. 2003. Thin on the Ground. Ques- Waterlow, J. C. 1992. Protein-Energy Malnutrition. tioning the Evidence Behind World Bank-funded London: Edward Arnold. Community Nutrition Projects in Bangladesh, Weiss, C. 1998. Evaluation. Upper Saddle River, NJ: Ethiopia, and Uganda, London: Save the Children Prentice-Hall. UK. White, Howard, and Edoardo Masset. 2003. "The Sen, A. 1998. "Mortality as an Indicator of Economic Importance of Household Size and Composition Success and Failure." Economic Journal 108 in Constructing Poverty Profiles: An Illustration (446): 1­25. from Vietnam." Development and Change 34 (1): Sibley, L. M., and T. A. Sipe. 2004. "What Can a Meta- 105­26. Analysis Tell Us About Traditional Birth Attendant WHO Expert Committee. 1995. Physical Status: The Training and Pregnancy Outcomes?" Midwifery 20: Use and Interpretation of Anthropometry. WHO 51­60. Technical Report Series 854. Geneva: WHO. Smith, L., and L. Haddad. 2000. Explaining Child World Bank. 2002a. "Poverty in Bangladesh: Building Malnutrition in Developing Countries: A Cross- on Progress." Washington, DC. Country Analysis. Washington, DC: IFPRI. ------. 2002b. "Country Assistance Strategy for Streatfield, P., A. Mercer, A. Siddique, A. Ziauddin, A., Bangladesh." Report 21326 ­ BD. Washington, DC. Ashraf. 2003. "Status of Performance Indicators, ------. 1995. Staff Appraisal Report. Bangladesh In- 2002, Health Population Sector Programme, tegrated Nutrition Project. Washington, DC: Pop- 1998­2003." A Report for the Health Programme ulation and Human Resources Division. Support Office for the Annual Programme Review Young Lives: An International Study of Childhood 2002. Dhaka: International Centre for Diarrhoeal Poverty. 2002. Justification of the Content of the Disease Research, Bangladesh. Household Questionnaires, Section 9: Socioeco- Thomas, D., J. Strauss, and M.-H. Henriques. 1991. nomic Status. http://younglives.org.uk "How Does Mothers' Education Affect Child Zenger, E. 1993. "Siblings'Neonatal Mortality Risks Height?" Journal of Human Resources 26 (2): and Birth Spacing in Bangladesh." Demography 30 183­211. (3): 477­88. 2 2 4 OED PUBLICATIONS Study Series 2003 Annual Review of Development Effectiveness: The Effectiveness of Bank Support for Policy Reform Agricultural Extension: The Kenya Experience Assisting Russia's Transition: An Unprecedented Challenge Bangladesh: Progress Through Partnership Bridging Troubled Waters: Assessing the World Bank Water Resources Strategy The CGIAR: An Independent Meta-Evaluation of the Consultative Group on International Agricultural Research Debt Relief for the Poorest: An OED Review of the HIPC Initiative Developing Towns and Cities: Lessons from Brazil and the Philippines The Drive to Partnership: Aid Coordination and the World Bank Financial Sector Reform: A Review of World Bank Assistance Financing the Global Benefits of Forests: The Bank's GEF Portfolio and the 1991 Forest Strategy and Its Implementation Fiscal Management in Adjustment Lending IDA's Partnership for Poverty Reduction Improving the Lives of the Poor Through Investment in Cities India: The Dairy Revolution Information Infrastructure: The World Bank Group's Experience Investing in Health: Development Effectiveness in the Health, Nutrition, and Population Sector Jordan: Supporting Stable Development in a Challenging Region Lesotho: Development in a Challenging Environment Mainstreaming Gender in World Bank Lending: An Update The Next Ascent: An Evaluation of the Aga Khan Rural Support Program, Pakistan Nongovernmental Organizations in World Bank­Supported Projects: A Review Poland Country Assistance Review: Partnership in a Transition Economy Poverty Reduction in the 1990s: An Evaluation of Strategy and Performance Power for Development: A Review of the World Bank Group's Experience with Private Participation in the Electricity Sector Promoting Environmental Sustainability in Development Reforming Agriculture: The World Bank Goes to Market Sharing Knowledge: Innovations and Remaining Challenges Social Funds: Assessing Effectiveness Uganda: Policy, Participation, People The World Bank's Experience with Post-Conflict Reconstruction The World Bank's Forest Strategy: Striking the Right Balance Zambia Country Assistance Review: Turning an Economy Around Evaluation Country Case Series Bosnia and Herzegovina: Post-Conflict Reconstruction Brazil: Forests in the Balance: Challenges of Conservation with Development Cameroon: Forest Sector Development in a Difficult Political Economy China: From Afforestation to Poverty Alleviation and Natural Forest Management Costa Rica: Forest Strategy and the Evolution of Land Use El Salvador: Post-Conflict Reconstruction India: Alleviating Poverty through Forest Development Indonesia: The Challenges of World Bank Involvement in Forests Uganda: Post-Conflict Reconstruction Proceedings Global Public Policies and Programs: Implications for Financing and Evaluation Lessons of Fiscal Adjustment Lesson from Urban Transport Evaluating the Gender Impact of World Bank Assistance Evaluation and Development: The Institutional Dimension (Transaction Publishers) Evaluation and Poverty Reduction Monitoring & Evaluation Capacity Development in Africa Public Sector Performance--The Critical Role of Evaluation Multilingual Editions Allègement de la dette pour les plus pauvres : Examen OED de l'initiative PPTE Appréciation de l'efficacité du développement : L'évaluation à la Banque mondiale et à la Société financière internationale Determinar la eficacia de las actividades de desarrollo : La evaluación en el Banco Mundial y la Corporación Financiera Internacional Côte d'Ivoire : Revue de l'aide de la Banque mondiale au pays Filipinas: Crisis y oportunidades Reconstruir a Economia de Moçambique http://www.worldbank.org/oed THE WORLD BANK ISBN 0-8213-6376-X