Report No. 30266-IN India Attaining the Millennium Development Goals in India How Likely and What Will It Take to Reduce Infant Mortality, Child Malnutrition, Gender Disparities and Hunger-Poverty and to Increase School Enrollment and Completion? December 2, 2004 Human Development Unit South Asia Region Document of the World Bank ACKNOWLEDGEMENTS This report was prepared by Ani1B. Deolalikar, who was Lead Economist (SASHD) until September 2003 and currently Consultant (SASHD), under the overall guidance o f Charles Griffin, who was Sector Director (SASHD) at the time o f the report preparation, and Michael Carter, Country Director for India (SACIN). Inputs from Peeyush Bajpai, Laveesh Bhandari, Amaresh Dubey, and Mehtab ul-Azam (consultants) and from V. Selvaraju and V. Sundararaman (SASHD) are gratefully acknowledged. The peer reviewers were Jere Behrman, Raghav Gaiha, Robert M. Hecht, Raghbendra Jha, and Maureen Lewis. Sudesh Ponnappa and Karthika Nair (SASHD) provided very helpful team assistance. Comments on an earlier draft from a technical review committee headed by Shanta Devarajan, Chief Economist (SAR), were helpful in preparing the current version o f the report. This report has been shared with the Government o f India, but does not necessarily bear their approval for all its contents, especially where the Bank has stated its judgments, opinions, and conclusions. CONTENTS PREFACE ......................................................................................................................................................................... i EXECUTIVE SUMMARY .......................................................................................................................................... i 1. INTRODUCTION .............................................................................................................................................. 1 2. INFANT AND CHILD MORTALITY ......................................................................................................... 4 A. Overall Trends B. Inter-State Variations.......................................................................................................................................... S C. Intra-State Variations.......................................................................................................................................... 7 E. Economic Growth, Public Spending on Health, and Infant Mortality Reduction ..................................... 9 F. Government Health Expenditure I1 G. Proximate Causes of Infant Mortality. ........................................................................... I S H. Socioeconomic and Policy Correlates ofhfant Mortality......................................................................... I 9 I.MultivariateAnalysis ofhfant Mortality ...................................... 24 J. Simulations to 2015........... 27 Annex II.I:Infant Mortality, Government Health Expenditure, and Per Capita Income across States, 1980-99...... ................................................... 32 3. CHILD MALNUTRITION ............................................................................................................................ 37 A. Patterns and Trends ...................................................................................................................................... 37 B. Public Spending on Nutrition........................................................................................................................... 40 C. Intrastate Variations............................................................................................ D. Concentration of ChildMalnutrition E. Proximate Causes of Child Malnutrition....................................................................................................... 43 F. Socioeconomic Correlates of ChildMalnutrition......................................................................................... 45 G. Multivariate Analysis of Child Malnutrition................................................................................................. 47 H. Simulations to 2015 48 I.ICDSandChildMalnutrition S I 4. PRIMARY SCHOOLING .............................................................................................................................. 54 A. Overall Trends.................................................................................................................................................... 54 B. Economic Growth, Public Spending, and School Enrollments.................................................................. 55 C. Household Survey-Based Enrollment Estimates................................. D. Concentration of Out-of-School ChildrenAged 6-1I E. Primary Completion Rate F. Socioeconomic Differences in Primary SchoolAttendance and Completion G. Infrastructure and Schooling........................................................................................................................... 64 H. Teachersand Primary Completion Rates...................................................................................................... 65 I.MultivariateAnalysis of Primary Attendanceand Completion 68 J. Simulations to 2015................................................. 70 Annex I K I:Gross Primary Enrollment Rates, Government Elementary Education Expenditure, and Per CapitaIncome across States, 1980-99............................................................................................... 80 5. GENDER DISPARITY INSCHOOLING ................................................................................................. 83 A. Trends and Patterns .................................................................................... .83 B. Economic Growth, Public Spending, and Gender Disparity. ..................................... 85 C. Household Survey-BasedEstimates of Gender Disparity D. Socioeconomic Differences in Gender Disparity................................................................. E. Infrastructure and Gender Disparity .......................................................................................... F. MultivariateAnalysis of Gender Disparity .................................................................................................... 89 G. Simulations .................... 91 6. HUNGER POVERTY ...................................................................................................................................... 95 B. Socioeconomic Variations in CalorieDeficiency ...................................................................... 97 C. Role of Infrastructure ................................................................................................. 98 D. MultivariateAnalysis of CalorieDeficiency .................................. ..99 E. Simulations ........................... 101 7. CONCLUSION ............................................................................................................................................... 103 ANNEX TABLES.............................................................................................................................. 110 REFERENCES..................................................................................................................................... 126 List of Tables Table Number Table Title 11.1 Infant mortality rates by child birthorder, sex and by state groups, 1994-98 11.2 Infantmortalityrate, byvarious individual household, child and community characteristics and by poor and nonpoor states, 1994-98 11.3 Infant mortality rate, by village availability o f government anti-poverty programs and health infrastructure and by poor and non-poor states, 1994-98 11.4 Projected decline ininfant mortality rate with various interventions inthe poor and nonpoor states 11.5 Assumptions about various interventions to reduce the infant mortality rate in the poor states, 1998-99 to 2015 Annex 11.1 Fixed-effects, l o glinear regressiono f infant mortality rate, pooleddata for 1980-1999 across Indianstates 111.1 Projected decline inchild underweightrates (percentage points) with various interventions inthe poor and norrpoor states 111.2 Assumptions about various interventions to reduce the child underweightrate in the poor states, 1998-99 to 2015 IV.1 Projected increase inschool attendance and primary completion rates (percentage points) with various interventions inthe poor states IV.2 Assumptions about various interventions to increase the school attendance and the net primary school attendance rate inthe poor states, 1999-2000 to 2015 Annex IV.1 Fixed-effects, l o glinear regression ofthe gross primary enrollment rate, pooled data for 1980-1999 across Indian states v.1 Assumptions about various interventions to reduce the gender disparity in primary and secondary school enrollment rate inthe poor states, 1999-2000 to 2015 v.2 Assumptions about various interventions to reduce the headcount ratio o f calorie deficiency inthe poor states, 1999-2000 to 2015 Annex 1 Maximum likelihood probit estimates of the probability o f an infant death inthe four years preceding the survey, 1994-98 Annex 2 Maximum likelihood probit estimates o f the probability o f a child aged 0-35 months beingunderweight, 1998-99 Annex 3 Maximum likelihood probit estimates o fthe probability o f a child aged 0-35 months beingunderweight, 1992-93 Annex 4 Maximum likelihood probit estimates o f school and primary school attendance among 6- 11year old children, 1999-2000 Annex 5 Maximum likelihood probit estimates o f primary school completion among 12 year old children, 1999-2000 Annex 6 Maximum likelihood probit estimates o f school attendance among 6-18year olds, by sex, 1999-2000 Annex 7 Maximum likelihood probit estimates o f the probability o f a household being calorie deficient, 1999-2000 Annex 8 Data on various MD indicators inthe early and late 1990s, by region Table Number Table Title Annex 9 Data on various MD indicators inthe early and late 1990s, by region Annex 10 List o f regions ranking among the lowest quartile o f all regions inIndia on three or more o f the five millenniumdevelopment indicators considered inthis report, 1998-2000 List of Figures Figure Number Figure Title 11.1 Infant mortality rate, by residence, 1971-2000 11.2 Infantmortality rate, 1970-2000, selected countries inAsia 11.3 Relationship across states between infant mortality and under-5 mortality rates, 1998-99 11.4 Infant mortality rate across Indian states, 1981and 2000 11.5 State-specific MDGs for Infant Mortality, 2015 11.6 Contribution o f the 21 larger states to national infant deaths, 2000 11.7 Cumulative distribution o f infant deaths in India across districts and villages, 1994-98 11.8 Infant mortality rates and real public spending on health and family planning per capita, across 13 major states, 1980-99 11.9 States ranked by "innate infant mortality" (i.e., infant mortality after control for public spending on health and per capita GSDP, averaged over 1980-99) and by observed IMR averaged over 1980-1999-2000 11.10 Per capita government revenue expenditure (state plus central) on health and family welfare (nominal prices), by state, 1999-2000 11.11 Average annual increase (%) in real government health expenditure per capita, by state, 1981-99 11.12 Public spending on health as % o f gross state domestic product per capita, 1981- 82 and 1999-2000 11.13 Public expenditure on health as % o f GDP, selected countries inAsia, 2000 11.14 Functional composition o f state government health expenditures, 1981 11.15 Functional composition o f state government health expenditures, 1999 11.16 Share of central grants in total government (revenue) expenditure on health, by function, selected states, 1991-92 and 1999-2000 11.17 Ranking of states inthe performance o f government health service, 2002 11.18 Cumulative number o f children survivingout o f 1,000 born, by age (inmonths), inpoorand other states, 1994-98 11.19 Proximate causes o f infant mortality inpoor and nonpoor states, 1996-98 11.20 L o w birthweight children, infant mortality, and maternal weight, 1998-99 11.21 Infant mortality rates, by predicted probability o f a child being severely underweight Figure Number Figure Title 11.22 Relationship across regions between under-five child mortality rate and child underweight rate, 1998-99 11.23 Relationship across regions between one-year olds immunized against measles and difference between infant and under-5 mortality rate, 1998 11.24 Measles vaccination rate (%) among 12-23 month olds, by state, 1992-93 and 1998-99 11.25 Mortality rate (per 1,000 live births), by age (months) and sex, 1994-98 11.26 Infant mortality rates o f various social groups inpoor and other states, 1994-98 11.27 Infant mortality by sex and by mother's schooling, 1994-98 11.28 Under-five child mortality rate, by presence o f ICDS anganwadi center in village, 1988-92 11.29 Projected decline in infant mortality rate in the poor states, 1998-2015, under different intervention scenarios Annex Estimated elasticity o f infant mortality with respect to real government health Fig. 11.1 expenditure across states, 1980-99, by level o f state GDP per capita (nonparametric kernel estimates with control for state fixed effects and annual year dummies 111.1 Percent o f children aged 0-35 months who are underweight, 1998-99, and annual % decline inthis rate between 1992-93 and 1998-99, by state 111.2 Child underweight rate MDGs by state, 2015 (% o f children 035 months underweight) 111.3 Change in absolute number o f underweight children 0-35 months o f age between 1992-93 and 1998-99, by state ('000 children 111.4 Child underweight rate (%) in 1992-93 and percent annual decline in underweight rate, 1992-98, by state 111.5 Child underweight rate (%) and gross state domestic product per capita across states, 1998-99 111.6 Government expenditure on ICDS (child nutrition) program (excluding training) per child aged 0-6 years, 1999-2000 (nominal prices) 111.7 Underweight rate among 0-3 year olds (%) and real expenditure on ICDS program per child 0-6 years, across states, 1992-93 and 1998-99 111.8 Changes in child (0-3) underweight rates (%) and in real government expenditure on the ICDS program per child 0-6 across states, 1992-93 to 1998- 99 111.9 Changes in child underweight rates and in real per capita gross state domestic product across states, 1992-93 to 1998-99 111.10 Contribution o f 20 states to the national number o f underweight 035 month olds, 1998-99 111.11 Cumulative distribution of all underweight 035 month old children in India across villages and districts, 1998-99 111.12 Initiationo f breast- feeding after birth, by groups o f states, 1998-99 111.13 Cumulative number of diarrheal infections experienced by infants, by age (months), 1998-99 Figure Number Figure Title 111.14 Percent o f children under 3 who are underweight, by mother's weight and birth weight of child, 1998-99 111.15 Childunderweightrates (%) byper capita expenditure quintile, 1998-99 111.16 Child underweight rates (YO)of various social groups in poor and other states, 1998-99 111.17 Child(0-3 years) underweightrates (YO)by sex and bymother's schooling, 1998- 99 111.18 Childunderweightrates (%), by birthorder and sex, 1998-99 111.19 Child underweight rates (%), by infrastructure access, 1998-99 111.20 Projected decline in percent o f children (13 who are underweight in the poor states, 1998-2015, under different intervention scenarios 111.21 Percent o f children under 4 years who are underweight, by sex and presence of ICDS anganwadi center invillage, 1992-93 IV.1 Gross primary enrollment rates, India, 1950-51 to 1999-2000 IV.2 Gross primary enrollment rate, by state, 1999-2000 IV.3 Public spending on elementary education per child aged 614 years and per student enrolled inelementary school, 1998-99 IV.4 Annual % growth in elementary enrollments, population aged 614, and real government expenditure on elementary education per child aged 6- 14, by state, 1980-99 IV.5 Distribution o f enrolled students and o f public spending on education in two states, by level, 1998-1999-2000 IV.6 Gross lower primary enrollment rates and real public spending on elementary schooling per child 6-14 across states, 1980-99 IV.7 Inter-state increase in elementary school (grades 1-8) enrollment and in real government expenditure on elementary education, 1980-99 IV.8 Additional number o f primary school (grades 1-5) students enrolled per (1993- 94) Rupee increase in gross state domestic product per capita, 1980-99 IV.9 School and primary school attendance rates, by age, 1999-2000 IV.10 Percent change in age-specific school attendance and net primary attendance rates, 1993-94 to 1999-2000 IV.11 Age-specific and net primary school attendance rates for 611 year olds, by state, 1999-2000 IV.12 Percentage o f children 6-11 years attending school, by state, 1993-94 and 1999- 2000 IV.13 Contribution o f 17 states to the national number o f 6- 11year olds out o f school, 1999-2000 IV.14 Cumulative distribution o f all out-of-school 6-11 year olds in India across villages and districts, 1999-2000 IV.15 Primary completion rate (%), by state, 1999-2000 IV.16 School attendance rates (ages 611) and primary completion rate for 12-year olds, by per capita consumption expenditure quintile, 1999-2000 Figure Number Figure Title IV.17 School attendance rates (ages 611) and primary completion rate for 12-year olds, by social group, 1999-2000 IV.18 School attendance rates (ages 611) and primary completion rate for 12-year olds, by female education, 1999-2000 IV.19 School attendance rates (ages 611) and primary completion rate for 12-year olds, by access to electricity, 1999-2000 IV.20 School attendance rates (ages 611) and primary completion rate for 12-year olds, by access to pucca roads, 1999-2000 IV.21 Ranking o f states in the performance o f government school education services, 2002 IV.22 Increase inprojected % o f children aged 6- 11 attending school inthe poor states, 1999-2015, under different intervention scenarios IV.23 Increase in projected % o f children aged 611 attending primary school in the poor states, 1999-2015, under differentintervention scenarios IV.24 Increase inprojected primary completion rate (%) inthe poor states, 1999-2015, under different intervention scenarios v.1 Ratio of female to male gross primary enrollment rate, 1950-51 to 1999-2000 v.2 Ratio o f female to male gross primary enrollment rate, by state, 1999-2000 v.3 Ratio o f females to males enrolled inprimary (grades 1-5) schools, 1980-81 and 1999-2000, by state v.4 Percent change in the ratio o f females to males enrolled in primary schools, 1980-81to 1999-2000 v.5 Relationship between the ratio o f females to males in primary school and real public spending on elementary education per child 614, 14 states in 1980-81 to 1999-2000 V.6 Relationship between the ratio o f females to males in primary school and real gross state domestic product per capita, 14 states in 1980-81to 1999-2000 v.7 Age-specific andprimary school attendance, by gender, 1999-2000 V.8 Ratio o f female to male students attending primary and secondary school, by age, 1999-2000 v.9 Ratio o f girls to boys inprimary and secondary schools (%), by state, 1999-2000 v.10 % change in age- & gender-specific school attendance rates, 1993-94 to 1999- 2000 v.ll Ratio o f girls to boys inprimary and secondary schools (%), by social group and consumption quintile, 1999-2000 v.12 Ratio o f girls to boys in primary and secondary schools (%), by adult male and female schooling in household, 1999-2000 V.13 Ratio o f girls to boys in primary and secondary schools (%), by infrastructure availability, 1999-2000 V.14 Ratio o f girls to boys in primary and secondary schools (YO),by school availability and by government expenditure on education, 1999-2000 Figure Number FigureTitle V.15 Projected changes in male-female difference (in percentage points) in school attendance rate o f children aged 6-18 in the poor states, 1999-2015, under different intervention scenarios VI.1 Percent o f population that is calorie-deficient, by state, 1999-2000 VI.2 Percent change in the proportion o f population that was calorie-deficient, 1993- 94 to 1999-2000,by state VI.3 Socio-economic differences incalorie deficiency, 1999-2000 VI.4 Predicted calorie intake and requirementsper capita per day, 1999-2000 VI.5 Percentage o f population that i s calorie deficient, by adult male and female schooling inhousehold, 1999-2000 VI.6 Percentage o f population that i s calorie deficient, by infrastructure access, 1999- 2000 VI.7 Percentage o f population that is calorie deficient, by various agricultural indicators, 1999-2000 VI.8 Projected changes in the incidence o f calorie deficiency (%) in the poor states, 1999-2015, under different intervention scenarios List of Text Boxes Box Number Box Title 11.1 Home-Based Neonatal Care: Results from a Field Trial inRural Maharashtra 11.2 I s Decentralization o f Health Services Associated with Child Mortality? 111.1 The Tamil Nadu IntegratedNutrition Project IV.1 The Education Guarantee Scheme of Madhya Pradesh IV.2 The Learning Guarantee Program inKarnataka IV.3 The Sawa ShikshaAbhiyan: A Program for Universal Elementary Education in India v.1 The Bangladesh Female Secondary School Stipend Program PREFACE The Millennium Development Goals commit the international community to a comprehensive vision o f development - one that places human development as the centerpiece o f social and economic progress and puts great value on global partnerships for development. Since 2000, when the MDGs were ratified at the UnitedNations Millennium Summit, the goals have been widely accepted as a yardstick for measuring development progress across countries. However, discussion o f many o f the issues relating to the MDGs - namely, the likelihood and costs o f attaining them - has remained at a global level. There have been relatively few country studies that either have examined sub-national variations in the likelihood o f attaining the MDGs or have analyzed, inany detail, the different factors that are likely to speed the attainment o f the MDGs. This i s one o f the first detailed country reports on the MDGs done by the World Bank. A major contribution o f the report is its focus on inter-state and inter-regional variations in India's development progress. The report argues forcefully that India's performance on the MDGs will hinge critically on the MDGperformance o f its poor states, which not only have the worst human development outcomes in the country but which also account for a large and increasing proportion o f India's population. The report goes beyond mere description; it examines the association between the MD indicators and various policy and behavioral variables. Using these associations, it undertakes simulations of the likely trajectory o f the MD indicators in the poor states through 2015 under possible scenarios. Naturally, quantitative analysis o f the type presented inthis report i s subject to many limitations. There are issues o f data quality, statistical inference, and omission o f important `qualitative' variables, such as governance and accountability. It i s therefore important to view this report not as the final and complete word o f India's prospects at attaining the MDGs, but instead as one o f many contributions to a broader debate on what it will take for India, and its many states and regions, to attain the MDGs.Ingoing forward, it will be important to complement the findings o f this report with other methodological approaches. Indeed, the World Bank commissioned a number o f case studies o f innovative, on-the-ground interventions throughout India that have attempted to influence MD outcomes. These case studies and this report were discussed extensively at an MDGconference held inNew Delhi inJune 2004. We would like to express ow gratitude to the many experts from government and non- government agencies, other international organizations, and civil society who provided their time andinsights generously duringthe preparationo f this report. Michael F. Carter Country Director World Bank India Office EXECUTIVESUMMARY Introduction 1. Since the launch o f the Millennium Development Goals (MDGs) at the Millennium Summit in New York in September 2000, the MDGs have become the most widely accepted yardstick o f development efforts by governments, donors and NGOs. The MDGs are a set o f numerical and time-bound targets related to key achievements in human development. They include halving income-poverty and hunger, achieving universal primary education and gender equality, reducing infant and child mortality by two-thirds and maternal mortality by three- quarters, reversing the spread o f HIV/AIDS and other communicable diseases, and halving the proportion o f people without access to safe water. These targets are to be achieved by 2015, from their levels in 1990 (United Nations 2000). 2. Almost all the countries in the world, including India, have committed themselves to attaining the targets embodied in the Millennium Declaration by 2015. Unfortunately, there i s little understanding o f whether India will be able to attain all o f the MDGs, and whether there are some MDGs that India will be able to attain. here i s even less understanding o f what it will take -byway ofeconomicgrowth, infiastructuralinvestments, andsectoral interventions -toattain the different MDGs. Further, this report argues the importance o f disaggregating the MDGs for India, given the very large geographical and socioeconomic variations in millennium development (MD) indicators across the country. 3. This report focuses on the attainment o f five major human development-relatedMDGsby sub-national units in India - child and infant mortality, child malnutrition, schooling enrollment and completion, gender disparities in schooling, and hunger-poverty (as reflected by inadequate calorie intake). The selection o f these MDGs for detailed analysis was based in large part on the availability o f reliable sub-national data. 4. The basic premise o f this report i s that there are large disparities inthe past performance as well as future prospects o f different sub-national units in terms o f the MD indicators. India cannot hope to attain the MDGs without significant progress in the MD indicators in its poorest states - Bihar, Orissa, Uttar Pradesh, Madhya Pradesh and Rajasthan. These states not only currently account for a large proportion o f the country's population, but, because o f more rapid population growth, will account for an even larger share o f the country's population in2015. The report attempts to identify the specific interventions that will improve substantially the likelihood o f the poor states attaining the MDGs. 5. While one o f the main objectives o f this report is to present a disaggregated analysis o f MDGs, availability o f data limits the extent to which the analysis can be fully disaggregated. In addition, the simulations undertaken inthis report are based on empirical analysis o f survey data, which typically relies on many assumptions about data quality and measurement, inferences o f causality between variables, and potential biases o f statistical and econometric estimates. It i s therefore important to note at the outset that, while the results and simulations presented in this report may give an impression o f precision, they are not that. They should be treated as being indicative of possible broad trends, and could usefully be complemented with other analyses 11 using different methodological approaches. As long as the results are used with this understanding,they can be helpfulin `rough-order' planning for MDGattainment. 6. The conclusion o f this report is that attainment o f the MDGs (at least the five specific goals considered in this report) will remain challenging in the poor states o f India. Despite the grim situation, however, the poor states should be able to meet the MDGs relating to infant mortality, child malnutrition, and hunger-poverty with a combination o f interventions, including sector-specific interventions (such as nutrition supplementation and immunization), economic growth, improved coverage o f infrastructure, and wide-ranging reforms in the institutions o f service delivery. However, attaining the education goals will require considerably more effort. This report concludes that while substantial progress could be made by the poor states on increasing the rates o f net primary enrollment and primary completion, it will be challenging for them to attain the education-related MDGs o f 100% net primary enrollment and 100% primary Completion, inlarge part because o f the enormous gap between their current rates o f net primary enrollment and completion (only 50% and 54%, respectively) and the MDtargets. The simulation analysis also suggests that complete elimination o f the gender gap in primary and secondary enrollment, called for by the MDGs, will be a very challenging goal to attain. Infant and Child Mortality 7. The millenniumdevelopment goal i s to reduce infant and child mortality by twethirds between 1990 and 2015. For India, this would imply a reduction o f the infant mortality rate (IMR)to 27 andofthe under-five mortality rate (U5MR) to 32 by2015. 8. Despite infant mortality having kclined impressively in India - from 130-140 infant deaths per 1,000 live births in the early 1970s to 68 in 2000 - the absolute levels o f infant and child mortality are still too high(about 68 infant and 95 child deaths per 1,000 live birthsin 1998- 99). Nearly 1% million children die each year in the country before reaching the age o f one. In addition, India compares poorly on the pace o f IMRreduction to several other countries in South and Southeast Asia, including Bangladesh. 9. There are large inter-state and intra-state variations inIMR in the country, with the IMR ranging from a low o f 14 for Kerala to a high o f 96 for Orissa. The rate o f progress on IMR reduction has also varied significantly across sub-national units. States such as Bihar and U.P., which had among the highest IMRs in the country in 1981, were among the top performers in IMR reduction over the priod 1980-99. On the other hand, Andhra Pradesh and Karnataka had the slowest rate o f IMR decline over the two decades. Ingeneral, there was some convergence in IMRs, so that inter-state disparity in infant mortality decreasedbetween 1981 and2000. 10. An interestingfeature o f infant deaths inIndia is that they are concentrated injust a few states. The NFHS-2 data are also suggestive o f infant deaths being heavily concentrated in a rektively small number o f districts and villages in the country. For instance, during 1994-99, a fifth of the country's districts and villages accounted for one-half of all infant deaths in the country. While these numbers are not precise (owing to the dlfficulty o f measuring infant mortality for small samples), they suggest that infant mortality could be brought down considerably by first identifying and then targeting mortality-reducing interventions to those districts and villages with the largest number o f infant deaths inthe country. 11. Econometric analysis o f state-level data on IMRs over the last two decades shows some evidence, although not persistent, o f a significant inverse association between infant mortality and government health expenditure across states. In addition, some o f the more general econometric ... 111 specifications suggest a stronger inverse association for the very poor states than for the non-poor states. This may simply reflect the fact that states with high infant mortality rates tend to have a larger proportion of infant deaths occurring during the post-neonatal (as opposed to the neo-natal) period, and these deaths are more easily averted by the typical (andrelatively inexpensive) child survival interventions, such as child immunizations and Gal rehydration therapy. On the other hand, reduction o f neo-natal deaths, which are relatively more common in the better-off, low- mortality states, typically requires more expensive interventions, such as professionally-attended deliveries or deliveries n institutions as well as post-delivery and emergency hospitakbased care.' 12. The results also indicate significant associations between infant mortality on the one hand and female literacy and per capita GSDP on the other hand. Per capita income and female literacy have a significant interactive association with infant mortality such that the inverse association between infant mortality and female literacy i s stronger at higher income levels than at lower income levels. This suggests that female literacy and the level o f development in a state compkment (rather than substitute for) each other in terms o f their association with infant mortality. 13. There i s significant gender disparity in the likelihood o f children dying in India. While the probability o f death i s greater for males than for females untilage one, the reverse i s true from ages one to five. Parental neglect toward girls - symptomatic o f the generally low social status o f women -appears to be an important cause o f the gender disparity inchild mortality. Girls are less likely to receive adequate food allocations andmedicaltreatment for their illnesses than boys. 14. In addition, this report finds strong inverse associations between infant mortality and matemal education, access to roads, regular supply o f electricity and sanitation, and coverage o f tetanus immunization and antenatal care among women. The empirical results suggest that ifthe poor states were simply brought up to the national averages in terms o f coverage o f sanitation, road access, electricity, antenatal care, tetanus immunization, female schooling, and public spending on health and family welfare per capita, the cumulative reduction inthe infant mortality rate inthe poor states would be o f the order o f 12 infant deaths per 1,000 live births (or 16%). If the magnitude o f the proposed interventions were scaled up so as to bring the poor states to the mean level o f the non-poor states, the reduction in infant mortality rate in the poor states would be greater - about 36.5 infant deaths per 1,000 live births (or 48%). It thus appears that infant mortality inthe poor states could decline substantially with a combination o f interventions. 15. Finally, we undertake simulations o f the reduction in the infant mortality rate in the poor states to 2015 under the assumption that the fore-mentioned interventions are pursued gradually between now and 2015. The simulation results suggest that the two interventions most strongly associated with infant mortality reduction in the poor states are increased schooling o f mothers and additional public spending on health and family welfare. With all o f the interventions, the infantmortalityrate inthe poor states is projected to decline from 75 infant deaths per 1,000 live births to a level o f 29 by 2015 - just slightly below the MD goal. This suggests that the infant mortality MDG will be challenging but attainable in the poor states o f the country with a combination o f mortality-reducing interventions pursued simultaneously. 16. The finding about the association between increased government health expenditure and infant mortality needs two qualifications. First, merely increasing health spending will not be ' Although neo-natal mortality reduction typically requires hospital-based care, it is possible to provide a relatively inexpensive package o f home-based neonatal services, as shown by a highly-successful field trial inMaharashtra in 1995-98 (see Box 11.1for a detailed description o fthe intervention). iv enough; the composition, quality and effectiveness o f public spending i s as important as raising its quantity. This i s especially true o f the poorer states in India, which are plagued with the most serious problems o f govemance and service delivery in the health (and other) sectors. Serious attention thus needs to be paid to making government health services work, especially for the poor. As the World Development Report 2004 points out, this i s a complex and difficult task that entails creation o f the right institutions and incentives in the system to improve service delivery, such as devolving responsibility for service delivery to local governments and communities, contracting out certain types o f service delivery to the non-government sector, and empowering consumers to demandbetter services from government health facilities. 17. Second, the finding that public spending on health i s associated with infant mortality reduction in the poor states i s o f little use to policy makers. Identifying the type o f health interventions on which to spend resources in the poor states is o f much greater policy interest. The evidence in this report, as well as the results o f previous studies, suggest that a package consisting o f expanded child and matemal immunization, antenatal care coverage, nutritional supplementation (including promotion o f exclusive breast-feeding), and home-based neonatal services (including treatment o fpneumonia) is likely to be a high-impact intervention strategy. ChildMalnutrition 18. The millenniumdevelopment goal is to reduce the percentage o f underweight children by one-half between 1990 and 2015, which would imply, for India, a reduction in the child underweight rate from 54.8% in 1990 to 27.4% in2015. 19. Child malnutrition rates in India are extraordinarily high. The NFHS2 indicates that nearly one-half o f children aged 035 months were underweight or stunted in 1998-99.2 About 18-23 percent o f children are severely underweight or stunted in the sense o f being more than three standard deviations below the relevant NCHS standards. A comparison o f data from two rounds o f the NFHS shows a modest decline o f about 11% (from a rate o f 52.7% to 47%) between 1992-93 and 1998-99. This i s a much smaller decline than in neighboring Banghdesh, where underweight rates fell from 68% in 1992 to 51% in2000. 20. An average child underweight rate o f 47% masks wide variations in child malnutrition across states. Child mderweight rates vary from a low o f 24-28% inthe Northeastem states and Kerala to 51-55% in the states o f Bihar, Rajasthan, Uttar Pradesh, Madhya Pradesh and Orissa Likewise, the decline inchild underweight rates over time has also varied greatly across states. In Punjab, for instance, the child underweight rate fell at an annual rate o f 7.6% tetween 1992-93 and 1998-99, while Rajasthan saw an ncrease of 2% per annum m the child underweight rate duringthe same period. 21. Sub-state variation inthe child underweight rate i s also very large, with the rate ranging from a low o f 11% in the Hills region o f Assam to a high o f 60% in Chhattisgarh. More than a third of the regions in the country have underweight rates of 50% or greater, showing the pervasiveness o f child malnutrition in India. Evenmore worrying i s the finding that more than a quarter o f all the regions inthe country experienced an increase inchild underweight rates. These regions are scattered around the country - inthe poor states (Jharkhand, Bihar, Orissa, and Uttar Pradesh) but also inthe more prosperous states, such as Gujarat, Maharashtra and Haryana. As in the literature, a child is considered underweight when his or her weight-for-age is more than two standard deviations below the N C H S reference weight. A child is stunted when his o r her height-for-age is more than two standard deviations below the N C H S reference. Severe underweight and stunting occur when the relevant nutrition indicator is more than three standard deviations below the N C H S reference. V 22. Like infant deaths, underweight children are also concentrated geographically in a relatively small number o f districts and villages. A mere 10% o f districts and villages account for 27-28% - and a quarter o f districts and villages account for more than half - o f all the underweight children inthe country. 23. Several factors contribute to the high rates o f child malnutrition in India: poor maternal nutrition leading to low birthweights, which intum increases the risk o f child malnutrition; poor infant feeding practices (such as waiting for a day or longer after birth before breast-feeding an infant,discarding the first breast milk or colostrum, early termination of exclusive breast-feeding andintroductionof supplementary feeding); and highrates of diarrheal and other infections. 24. Much o f the economic variation in child underweight rates across economic groups occurs between the bottom four consumption quintiles and the top qumtile, with little variation among the poorest 80% o f the population. However, even the richest quintile shows relatively high child underweight rates (about 30%). The fact that nearly a third of the top consumption quintile in the country - a group that is likely to have good economic access to food - is malnourished suggests that cultural and social factors have an important role to play in determining child malnutrition in India. This i s also consistent with the finding that the child underweight rate is relatively highinprosperous states such as Gujarat and Maharashtra. 25. The other correlates o f child malnutrition are affiliation to socially-disadvantaged groups (such as scheduled castes andtribes and other backward castes), maternal age and education, and birth order (with higher birthorder girls having a greater risk of malnutrition than higher birth order boys). In addition, infrastructural variables - access to toilets, regular electricity, and good roads -are significantly associatedwith reduced child malnutrition. 26. Much o f the public spending on child nutrition in India takes place on the Integrated Child Development Services program. This program consists o f annanwadi centers (AWCs) in each village, typically staffed by a village woman with 5 8 years o f schooling and an assistant, whose functions are to provide growth monitoring, pre-school education and nutritional supplementation to targeted children aged 0 6 years inthe village. Although the program covers all the villages inthe country, recent surveys from a few states suggest that relatively few (about 10-30%) children aged 0 6 years in states such as Uttar Pradesh, Madhya Pradesh andRajasthan regularly attend the AWCs in their community. This may be because the amount o f food supplementation providedto children i s meager or irregular or both. 27. There are large disparities in public expenditure on nutrition across states. Poor, high- malnutrition states like Bihar, Uttar Pradesh, Madhya Pradesh and Rajasthan spend only Rs. 30- 50 on nutrition programs per child 06, while Gujarat, hnjab, and Haryana spend Rs. 90-100. Tamil Nadu's expenditure is close to Rs. 170, while spending inthe Northeastern states is above Rs. 500. 28. Simulations based on multivariate analysis o f child underweight rates using unit record data show that ifthe poor states were brought up to the national average interms o f coverage o f sanitation, road access, electricity, medical attention at the time o f delivery, adult female schooling, household income (consumption), and public spending on nutrition programs per child, the cumulative reduction inthe child underweight rate inthese states would be o f the order o f about 8 percentage points (or 15%). Ifthe magnitude o f the proposed interventions were scaled up, so as to bring the poor states to the average level prevailing in the non-poor states, the cumulative eduction in the child underweight rate in the poor states would be 21 percentage vi points or 38%. These are large reductions in child malnutrition. Indeed, the simulations indicate that, together, the proposed interventions are associated with a reduction o f 25 percentage points inthe childunderweight rate inthe poor states-enough for themto attain the MDgoal. 29. As noted earlier, a great deal of the spending on nutrition in India takes place via the ICDS program. The NFHS-1 data suggests that, after controlling for other factors (including household living standards and matemal education), the presence o f an ICDS anganwadi center ina village is associated with a reduction of about 5% inthe childunderweight rate, but only for boys. A 5% reduction inunderweight rates is consistent with anecdotal evidence that food rations distributed by most anganwadi centers inthe country are limited, infrequent, and often irregular. However, the finding that all o f the positive nutritional benefits o f anganwadi centers accrue to boys, not girls, i s surprising. It could reflect that parents tend to alectively bring their boys, but not their girls, for supplementary feeding at the center. Or it could indicate that anganwadi workers or helpers provide a larger allocation o f food to boys than girls. This is an issue that merits further exploration. Primary Schooling 30. The millennium development goal is to ensure that, by 2015, all children are in school, the net primaryenrollment ratio is loo%, and that all the pupils entering grade 1 are retained until grade 5 (typically the last year o f primary school). 31. India has made rapid strides in education during the last 4 5 decades. The gross primary enrollment rate, which was only 43% in 1950-51, reached 100% by 1990-91, and has fallen slightly since then. An average gross primary enrollment rate o f 95% for the country in 1999-200 for the country masks wide variations across states. Gross primary enrollment rates vary from a low o f 65% inUttar Pradeshto a higho f 139% in Sikkim. 32. However, as in other countries, gross enrollment rates obtained from school administrative records differ significantly from household survey-based estimates o f enrollment. Data from the 55'h round o f the NSS indicate a gross primary enrollment rate o f 61% and a net rate o f only 52.5% in 1999-2000. The data also show that, even at the peak attendance ages o f 9- 11years, nearly 15% o f the population does not attend school. This is indicative o f a large class o f children who never attend school. 33. Inaddition, there are large differences across states in the primary attendance rate. Attendance rates for the age group 6-11 exceed 90% in 9 states - Kerala, Tamil Nadu, Maharashtra, Goa, Himachal Pradesh, and the states o f the Northeast. At the other end, the primary attendance rates are only 75% or lower in Bihar, Orissa, Rajasthan, Uttar Pradesh, and Madhya Pradesh. With only 53% o f children aged 611 attending school, Bihar ranks as the poorest-performing state on school attendance inthe country. 34. There is a large discrepancy between the age-specific andthe net primary attendance rate in all the states. Even in the states having highattendance rates among 611year olds, the net primary attendance rate is significantly lower. InBihar, a mere 28% o f children aged 611 attend primary school! 35. Sub-state differences inthe net primary school attendance rate are very large as well. Six regions in the country - largely in Bihar, Jharkhand, Orissa and Manipur - had a net primary attendancerate o f less than40% in 1999-2000. Sixteen regions, out o f a total o f 78 regions, had a net primary attendance rate o f less than 50%. Even more discouraging i s the fact that 29 regions vii inthe country, out ofatotal of77regions, either didnotsee animprovementintheir netprimary attendance rate between 1993-94 and 1999-2000, or saw it decline. 36. Analysis of state-level data over the period 1980-99 generally seems to show a significant positive association between the gross primary enrollment rate and per-child government expenditure on elementary education across states, even after controlling for per capita income and adult female literacy. In addition, depending upon different specifications, the positive association between the gross primary enrollment rate and government elementary school expenditure i; observed to be weaker in the non-poor states than in the poor states. The latter result could reflect the fact that the only children who remain out o f school inthe non-poor states are children klonging to socially-excluded and fringe groups located in hard-to-reach areas and mainstreaming these children into the regular school system i s often difficult and expensive. 37. According to the NSS 55'h round data, there were nearly 30 million out-of-school children aged 611 in India in 1999-2000. Nearly half o f all these children come from the two states of Uttar Pradesh and Bihar. School nonattendance i s also very highly concentrated in relatively few villages in the country. A mere 10% o f villages in the country account for nearly one-half o f all out-of-school children aged 6 11, while 20% o f villages account for three-quarters o f all out-of-school children. Most likely, these villages are scheduled tribe habitations that do not have a primary school o f their own and are not within easy walking reach o f a primary school. They are also likely to be poor villages where the opportunity cost o f child labor (interms o f agricultural work) i s high. These results indicate the importance and potential effectiveness o f targeting school hterventions to those villages and districts having the largest number o f out-of- school children. 38. Getting out-of-school children into school i s only one o f the education-related millen- nium development goals. Another goal is retention of students - viz., to ensure that the entire cohort o f children who begins grade 1 remains in school until grade 5. This report uses the primary completionrate, as measured by the proportion o f 12-year old childrenwho do not report themselves as never having attended school and also report currently being inmiddle school, as a proxy for retention. The primary completion rate for the country was 61.4% in 1999-2000 - only slightly up from 58.7% in 1993-99- with large interstate variations. For instance, Keralahad the highest rate (92.1%), followed by Goa, Maharashtra, Kamataka and Tamil Nadu. The Northeastem states and Bihar, Madhya Pradesh, Uttar Pradesh and Rajasthan rank at the bottom, with primary completion rates o f 50% or lower. Intra-state variations in the primary completion rate are also very large. 39. Analysis o funit-record data from the NSS 55'h round show significant associations o f the age-specific primary enrollment rate, net primary enrollment rate, and primary completion rate with: household living standards (as proxied by per capita consumption expenditure), affiliation to historically-disadvantaged social groups (such as scheduled castes and tribes), adult male and female schooling in the household, electricity coverage, and the level o f economic development ina state (as proxiedby gross domestic product per capita). Better road accesshas a significant association with primary completion but not with enrollment. Govemment expenditure on elementary education per child 614 years in a state also has strong positive associations with primary school attendance and primary completion (but not with regular school attendance). 40. The results also suggest that crime against women, as proxied by the number o f cognizable kidnappings o f women and girls per capita in a district, i s inversely associated with both school attendance and completion for girls. viii 41. The results with respect to school infrastructure are interesting. The availability o f primary schools per 1,000 children aged 6-11ina district - an indicator of schooling quantity - is strongly and positively associated with school attendance, but not with primary completion, among 611 year olds. On the other hand, an increase in the pupil-teacher ratio at the primary level in a district - an indicator o f poor quality o f schools - i s associated with lower rates o f school attendance and primary completion. This suggests that school attendance i s currently constrained in India by the availability o f primary schools; hence, expanding the number o f primary schools (in relation to the population o f 611 year olds) would be associated with an increase in the school attendance rate. On the other hand, quality improvements, such as a reduction in the pupil teacher ratio, would be associated with an increase in primary completion (as well as inschool attendance). 42. A simulation analysis using the multivariate models estimated above suggests that simultaneous pursuit o f several interventions - economic growth, growth o f government expenditure on elementary education, electricity coverage, expansion o f male and female adult schooling, an increase in the number o f primary schools per 1,000 children aged 611, and a reduction o f the pupil-teacher ratio - would be associated with a 31.5 percentage point increase in the school sttendance rate in the poor states by 2015 - enough for them to attain universal primary enrollment. However, the same package o f interventions would be associated with an increase o f only about 27 percentage points in the net primary enrollment rate by 2015 - well short o f the 50 percentage points needed to attain the MD goal. These results suggest that while it may be possible to get all children aged 6- 11in the poor states in school by 2015 with concerted action on many fronts, raising the net primary enrollment or attendance rate to 100% will be significantly more challenging. 43. A comparable simulation analysis o fprimary completion inthe poor states suggests that, with a combinationo f the fore-mentioned interventions: the primary completionrate is likely to increase by about 29 percentage points by 2015 - signifcantly short o f the 46 percentage point increase that would be needed to attain the MDgoal o f 100% primary completion. The simulation results thus highlight the challenges that the Government o f India i s likely to face as it attempts to accomplish its ambitious goals o f universal elementary schooling andcompletion by 2010. 44. It is important to note that attainment o f the education-related MDGs in India will require, inaddition to the interventions discussed earlier, broad-ranging institutional reform inthe education sector. Various surveys indicate that the rates o f teacher absenteeism inpublic schools are very high in India, with the situation often being worse in the poor states. Teacher absenteeism is a reflection o fthe larger problem o f lack o f accountability o f school administrators and teachers to students and their parents. While there are no simple solutions to this problem, evidence from other countries suggests the need for institutional reform that empowers citizens and communities to hold the state accountable for school performance, devolves administrative and financial powers to communities, provides greater autonomy to schools, involves parents in school management, and motivates front-line workers to provide better-quality schooling. The package o f interventions considered for primary completion is slightly different from that for enrollments. First, increasing the number o f primary schools per 1,000 children aged 6-1 1 i s dropped from the projection (as it is not significantly associated with primary completion). Second, increased availability o f village roads is included in the projection, as it is significantly associated with primary completion (but not with primary enrollment. i x Gender Disparityin SchoolEnrollments 45. Another o f the MD goals is to eliminate gender disparities in schooling, such that the ratio o f girls to boys enrolled at all schooling levels, but particuhrly at the primary and secondary levels, i s 100%. 46. Schookbased administrative data suggest that India has made impressive gains in reducing the male-female gap inthe gross primary enrollment rate inthe last fifty years, with the ratio o f the female to male gross primary enrollment rate nearly doubling from 41% in 195CL51to 81% in 1993-94, where it has stayed since. Yet there are large interstate variations inthe extent o f gender disparity in schooling, with the gender gap being the largest inBihar, Uttar Pradesh and Rajasthan, where the gross primary enrollment rate for females i s about two-thirds or less than that for males, and the smallest in Punjab, Haryana, Sikkim and Kerala, where i s parity or near- parity in h e gross primary enrollment rates for boys and girls. The data also indicate wide variation in the performance o f states over time. Between 1980-81 and 1999-2000, for instance, the largest relative gains for girls occurred in Haryana, where the ratio o f females to males enrolled in primary school nearly doubled. At the other extreme, Orissa, Uttar Pradesh, and Kerala experienced small relative &clines in the female-male ratio (although it is important to note that the female-male ratio of primary school students was already over 95% in Kerala in 1980-81). 47. Household survey data from the 55'h NSS round o f the NSS also show large sub-state differences in the ratio o f female to male students in primary and secondary school. InEarly a third of the regions in the country, female students constitute fewer than three-quarters of male students at the primary and secondary levels. Even more disconcerting i s the finding that, between 1993-94 and 1999-2000, as many as 30 regions - out o f a total o f 78 - experienced a decline in the ratio of female to male students. Surprisingly, these regions are located in such states as Tamil Nadu, Andhra Pradesh, Karnataka, Maharashtra and even Kerala. 48. Multivariate analysis o f school attendance for boys and girls aged 6-18 years, usingthe NSS 55th round unit record data, indicates that household consumption expenditure per capita and access to roads are associated with increased gender disparity, as both o f these variables have stronger associations with male than with female enrollment. On the other hand, adult female schooling in the household has a much stronger association with female than with male enrollment. Interestingly, the overall volume o f public spending cn elementary and secondary education per child 6-18 is not significantly associated with gender disparity in enrollment, but the composition o f public spending is. An increase in the share o f secondary education in government educational spending i s associated with reduced gender gap in enrollment, which i s consistent with additional public spending on secondary education drawing more girls than boys into school, at least incomparisonto additional public spending on primary education. 49. Other variables that are associated with reduced gender disparity in enrollment in the multivariate analysis are the number o f elementary and secondary schools per child aged 6-18, better access to electricity, and crime against women (as measured by the number o f cognizable kidnappings o f women and girls per capita). 50. Simulations based on the multivariate analysis suggest that a package o f interventions implemented gradually over the next decade would be associated with a reduction o f 3.6 percentage points (from about 13 percentage points in 1999 to 9.4 percentage points in 2015) in the gender gap inprimary and secondary enrollments - less than a third o fthe distance to the MD goal o f a zero gender gap. X 51. One issue that the simulation analysis in unable to highlight is the role o f incentives in narrowing gender disparities. Since gender disparities in schooling outcomes are largely caused by parental discrimination against the girl child, public policies that increase the parental incentive to invest ingirls, such as tuition waivers for girls and female stipends and scholarships, are likely to work well innarrowingthe gender gap. An example o f a policy intervention that has worked remarkably well in narrowing gender disparities in secondary school enrollments is the Female Secondary School Stipend program in Bangladesh (see Box V.1). This program seeks to not only increase the enrollment o f girls at the secondary level but to also ensure that most o f them are retained untilgraduation from secondary school. Many states in India also subsidize the schooling o f girls invarious ways. 29. Finally, it is important to reiterate the message that a simple expansion o f school capacity or introduction o f a female scholarship program will not work unless it is accompanied by broad- ranging institutional reform to reduce teacher absenteeism and to make schools and school managers more accountable to students andthe community. Hunger-poverty 52. The last millennium development goal we consider in this report is the reduction of hunger-poverty. The goal here is to halve the proportion o f people who suffer from hunger between 1990 and2015. For India, this would mean bringing down the headcount ratio o f calorie deficiency from 62.2% in 1990 to 31.1% in2015. 53. Calorie deficiency i s pervasive in India. In 1999-2000, more than one-half (53%) o f India's populationconsumed fewer calories thanit required - down from a level o f 60% in 1993- 94. This i s nearly double the estimated national incidence o f consumption poverty o f 22-26%. There are large interstate variations in the extent o f hunger-poverty, with Assam topping the list with nearly 78% o f the population being calorie deficient. The other Northeastern states, Goa and Andhra Pradesh also have very high proportions o f calorie -deficient population. At the other extreme, Jammu & Kashmir, Rajasthan and Uttar Pradesh have the lowest rate o f calorie deficiency (30-38%). 54. The rate at which hunger-poverty declined during the 1990s also varies significantly across states. Uttar Pradesh as well as many o f the Northeastern states, such as Mizoram, Manipur, Arunachal Pradesh and Nagaland, experienced large declines (greater than 10%) in hunger-poverty between 1993-94 and 1999-2000, while four states - Haryana, Assam, West Bengaland Jammu & Kashmir - experienced an increase in the incidence o f hunger. Within-state differences are also large. For instance, the incidence o f hunger-poverty varies from 48% to 75% withinthe state o f Gujarat. 55. Multivariate analysis o f calorie deficiency indicates that adult schooling in a household - both that o f males and females - is strongly associated with a reduction o f hunger-poverty. Rural Note that there i s a separate MDG of halving, between 1990 and 2015, the proportion of people whose income is less than one dollar a day, which we do not consider. The notion o f hunger-poverty differs from that of consumption- or income-poverty. Using household survey data on consumption of various foods and on the age, sex andoccupations of different household members, we have calculated calorie availability per capita to the household as well as calorie requirements (using normative age-specific calorie requirements adjusted for heavy, moderate and sedentary work). Individuals residing in households whose calorie requirements exceed availability are considered calorie -deficient or hunger-poor. In 1999-2000, the rate of hunger-poverty so defined was 53% in India, as compared to a consumption-poverty rate of 22-26%. xi households and farm households are less likely to be hunger-poor than urban and non-farm households, respectively, reflecting the importance o f land in reducing vulnerability. Availability o f electricity i s not a significant correlate o f calorie deficiency, but access to safe drinkingwater is, particularly in the poor states. Access to roads also has a strong negative effect on hunger- poverty inbothpoor and non-poor states, but the effect i s nearly two times as strong in the poor states. 56. Production o f food in a district and household access to land, particularly irrigated land, are associated inversely with calorie deficiency, with the associations being much stronger in the poor states than inthe non-poor states. This suggeststhat, even after controlling for other factors, access to the means o f food production reduces the risk o f hunger and calorie deficiency for individuals andhouseholds. 57. Finally, the empirical results indicate a very strong association between economic growth and calorie deficiency, with a one percent increase inmeanconsumption expenditure per capita in a district being associated with a 0.26% reduction inthe incidence o f hunger-poverty. 58. Simulations based on the multivariate analysis discussed above suggest that the proportion o f the calorie-deficient population in the poor states is likely to &cline by about 14 percentage points (from 47% to 33%) from 1999 to 2015 with a package o f six interventions (excluding economic growth) - just slightly short o f the MD goal. However, the seventh intervention - viz., general economic growth resulting in a 3% annual rise in mean district consumption expenditure per capita - i s associated with a further reduction of 12 percentage points in the calorie deficiency rate. This implies that, together, the seven interventions considered will easily allow the hunger-poverty MDG to be attained by 2015, even in the poor states o f the country. Conclusions m,there 59. There are several major findings and implications for policy that come out o f this report. are very large disparities across different states and regions o f India interms o f their performance on the millennium development indicators. Some regions rank low on some MD indicators but not on others. However, there are 18 in the country that rank consistently low on three or more o f the five MD indicators considered in this report. Of these, 8 regions rank consistently low on 4 or more indicators, while two regions rank low on all five indicators. These two regions are the South-Westem region o f Madhya Pradesh and the Southem region o f Rajasthan. In a sense, these are the most deprived regions inthe country. 60. The inter- and intra-state disparities have important implications for the targeting o f interventions and resources to the lagging states and regions. But they also have another important implication- the need for state-specific approaches that allow for greater flexibility in addressing problems. What works in a state such as Tamil Nadu may not work well in another state, such as Orissa. Even though health and education are state-level subjects in India, the central government plays a strategic role in these sectors via its centrally-sponsored schemes. In the case o f health, for instance, centrally-sponsored schemes account for a large portion o f the public health and reproductive health activities undertaken in a state. Likewise in the education sector, centrally-sponsored schemes, such as Operation Blackboard, finance a significant portion o f non-salary costs o f education in the states. It will thus be important to make centrally- sponsored schemes more flexible and more responsive to local conditions. xii 61. A second and related finding o f this report is that many o f the millenniumdevelopment indicators have high levels o f geographical concentration in India. For instance, one-fifth o f the villages and districts account for about one-half o f all the infant deaths andunderweight children inthe country. Amazingly, three-quarters of all the out-of-school children aged 611years are concentrated inonly 20% o f the country's villages. Unfortunately, currently-available data do not allow us to identify these villages, because the sample surveys on the basis o f which these observations are made are not large or Epresentative enough at the village level. However, promising new methodologies, based on merging o f household survey and population census data, are available for identifying specific villages having the worst human development outcomes. It would be worthwhile to explore the use o f such methodologies to identify the villages with the worst MD indicators inthe country, so that policy interventions could be better targeted to these villages. 62. The results on geographical concentration o f the MD indicators suggest that targeting resources and interventions to villages and districts with the worst human development indicators i s not only desirable for improving equity, but will also be the most effective strategy for attainment of the MDGs. How best to achieve this targeting is an issue that would appear to merit priority attention. 63. Third there are other types of disparities inmost of the MD indicators, the reduction of which would help inthe attainment o f the MDGs. Gender disparity is one example. It i s estimated that overall child mortality inthe country would drop by 20% if girls had the same mortality rate as boys between the ages o f 1month and 5 years. Likewise, the overall school enrollment rate for children aged 611 years would increase by about 7% if girls o f these ages had the same enrollment rate as boys. Among older children, the equivalent increase in enrollment rates would be even greater (about 10% for 12-14year olds and 16% among children 15-18 years o fage). Just as geographical targeting i s likely to narrow regional disparities, targeting by gender can help reduce gender inequalities. Since gender disparities inhealth and schooling outcomes are largely caused by parental discrimination against the girl child, public policies that increase the parental incentive to invest in girls, such as female scholarships or health-care subsidies for girls, are likely to work well in narrowing the gender gap. (Many states in India already subsidize the schooling o f girls in various ways, although gender-targeted subsidies for health-care and nutrition are much less common.) Many empirical studies from around the world, including some evidence presented in this report, suggest that gender disparities in the health, schooling, and nutritional outcomes o f children tend to narrow with mother's schooling, as mothers (relative to fathers) tend to invest more intheir female children.' 64. Another manifestation o f disparty in human development outcomes i s along caste and tribal lines. Scheduled tribes and, inmany cases, scheduled castes have significantly higher levels o f infant mortality and child malnutrition and significantly lower levels o f schooling indicators than mainstream groups. For instance, infant mortality among scheduled tribes inthe poor states is 60% greater than among other population groups after controlling for living standards, adult schooling, and other characteristics. Given that these groups constitute between one-quarter and one-third o f the total population o f the poor states inthe country, improving their MD indicators would help significantly in the attainment o f the MDGs in these states. It i s likely that many o f the villages identified in this report as having the heaviest concentration o f infant deaths and underweight and out-of-school children are, in fact, scheduled tribekaste villages. Ifso, targeted interventions to these villages would achieve the dual objectives of reducing geographic a1as well This i s not always the case, however. For instance, it has been observed that excess female child mortality is even more pronounced among better-schooled women relative to women with n o schooling, at least in rural Punjab (Das Gupta 1987). xiii as caste disparities. For several decades, India has had an extensive system o f quotas and reservations for scheduled castes and tribes in employment and higher education opportunities, but the results in this report suggest that targetingmatemal and child health, child nutrition, and primary schooling interventions to scheduled castes andtribes i s equally important.6 65. Fourth, there i s evidence o f significant synergies among the different MDGs. For instance, a Eduction in the proportion o f underweight children is strongly associated with a reduction o f child mortality. Although matemal mortality is an MD indicator that has not been analyzed inthis report, it is clear that interventions that reduce matemal mortality, such as tetanus immunization, expansion o f antenatal care coverage, and an increase in the ratio o f professionally-attended deliveries, will also bring about large reductions in infant (especially neonatal) mortality. Likewise, reducing child malnutrition is likely to result in improvement in both schooling quantity and quality, as better nourished children are more likely to attend school andperform better in school. Thus there are synergies amongst the MDGsthat will help in their attainment, which implies that proceeding with simultaneous action on all these measures will have the greatest impact on attainment o f the MDGs. 66. Fifth, despite appreciable improvement in many of the millennium development indicators during the 1990s, the analysis inthis report suggests that attainment o f the MDGs (at least the five specific goals considered in this report) will remain extremely challenging in the poor states of India. Yet attainment o f the MDGsby the poor states - most notably, Bihar, Uttar Pradesh, Madhya Pradesh, Orissa and Rajasthan - is critical to MDGattainment by the country as a whole, because these states currently account for a disproportionately large proportion o f the country's population (and will account for an even larger population share by 2015 owing to more rapidpopulation growth). 67. Despite the grim situation, the poor states should be able to meet the MDGs relating to infant mortality, child malnutrition, and hunger-poverty with a combination o f interventions, including sector-specific interventions (such as nutrition supplementation and immunization), economic growth, improved coverage o f infrastructure, and wide-ranging reforms in the institutions o f service delivery. However, attaining the education goals will require considerably more effort. This report concludes that while substantial progress could be made by the poor states on increasing the rates o f net primary enrollment and primary completion, it will be challenging for them to attain the education-related MDGs o f 100% net primary enrollment and 100% primary completion, in large part because o f the enormous gap between their current rates o f net primary enrollment and completion (only 50% and 54%, respectively) and the MDtargets. The simulation analysis also suggests that complete elimination o f the gender gap inprimary and secondary enrollment, called for by the MDGs, will be a very challenging goal to attain. 68. m,thesimulations carried out inthis report indicate that, inthe poor states, economic growth that brings about an improvement in household living standards i s strongly associated with virtually every MD indicator. For example, real economic growth o f 3% per annum in the poor states could alone bringdown the child underweight rate from 51% in 1999 to 37% in2015. This is nearly twethirds o f the total decline inthe child underweight rate needed to be attained between 1999 and 2015 to meet the nutrition MDG. Lkewise, 3% annual growth could bring about an increase inthe net primary enrollment rate from 50% in 1999 to 74% in2015 inthe poor states - about one-half o f the total increase in the net primary enrollment rate needed to be attained between 1999 and 2015 to meet the schooling-related MDG. Another example o f the Most state governments in India do offer incentives, such as free textbooks and uniforms, to primary school-aged children belonging to scheduled castes and tribes. xiv important role o f economic growth on MDGattainment is provided by the estimates that suggest that a real growth rate o f 3% would be associated with a 12 percentage point decline in the incidence o f hunger-poverty - nearly three-quarters o f the total decline needed to be attained between 1999 and 2015 to attain the MD goal o f halving poverty. Inother words, rapid economic growth could make a very significant contribution to an improvement in most MD indicators by 2015. 69. Inspite ofthe strong associations betweenimprovementsinMDindicators and economic growth in the poor states, the latter have actually experienced anemic growth over the last two decades. Bihar's real GSDP per capita has grown at less than 1% per annum between 1980 and 1999, while Orissa and Uttar Pradesh have seen annual real growth rates o f about 2% or less. In contrast, states such as Maharashtra, Gujarat and Karnataka have seen their economies expand at real per capita rates o f 4% or more over the same time period. What these results underscore i s the importance o f accelerating economic growth inthe poor states to speed progress toward the MDGs. It is beyond the scope of this report to discuss the pecise mechanisms by which growth in the poor states could be stimulated, but investment in physical and social infrastructure, combined with improved governance, would probably rank near the top o f the list. 70. Seventh, this report also indicates the importance o f infrastructure in attaining the MDGs. For example, simply bringing the poor states to the (average) levels o f road, electricity and sanitation coverage inthe non-poor states would be associated with reductions o f 6.2 percentage points (or 12%) inthe proportion o f underweight children and o f 13.2 deaths per 1,000 live births (or 18%) inthe infant mortality rate. Universal (100%) road, electricity and sanitation coverage in the poor states would be associated with reductions o f 40% in the child underweight rate and o f 60% in the infant mortality rate. The estimated associations between schooling outcomes and physical infrastructure are generally smaller, but the results still suggest that universal electrification in the poor states would be associated with increases o f 7 percentage point in the school attendance rate (for 611year olds), 5 percentage points inthe net primary enrollment rate, and 5 percentage points inthe primary completion rate. All o f these associations are most likely underestimated, since they do not consider the indirect association between infrastructure and the MD indicators via the positive association between infrastructure and gross state domestic product. 71. Eighth, consistent with the findings o f numerous empirical studies from around the world, this report finds evidence o f strong associations between female adult schooling and virtually every MD indicator considered. An additional year o f mother's schooling i s associated with a reduction o f 4% in infant mortality and 3% in child underweightrates and with increases o f 1-2%inthe net primary enrollment and primary completionrates. Inaddition, each additional year o f adult female schooling is associated with a reduction in the incidence o f hunger-poverty by about 1.5% and of gender disparityinschool enrollment by as much as 8%. 72. These results suggest that the large increases in girls' school enrollment that are likely to occur in the coming years (in part due to the Government o f India's Sawa Shikha Abhiyan initiative) will fuel major improvements in all o f the MD indicators. Of course, given the lag between a girl's attendance in elementary school and the birth o f her children (especially since schooling typically delays both age at marriage and age o f first birth), increases in female primary enrollments may not show their full effects on fertility, child and infant mortality, child nutrition, and child schooling for several years (perhaps, a decade or longer). However, increases infemale secondary enrollments are likely to show their results much sooner - well before 2015, the MDG reference year. xv 73. Ninth,notwithstanding the importance ofthe general interventions discussed above (e.g., economic growth, infrastructure expansion, and female schooling), the analysis in this report suggests that carefully-targeted, sector-specific interventions will also be important in attaining the MDGs. For instance, a package consisting o f expanded child and maternal immunization, antenatal care coverage, nutritional supplementation (including promotion o f exclusive breast- feeding), and home-based neonatal services (including treatment o f pneumonia) i s llkely to bring about significant reductions in both infant mortality and child malnutrition. Likewise, our results suggest that an expansion in the number o f primary schools would improve access and likely raise school and primary school attendance rates among 611 year olds, while a lowering o f the pupil-teacher ratio at the primary level would raise primary completion rates. These results are consistent with evidence from around the world that expanded school access i s typically well- targeted to the poor, since poor students are almost always the last to be enrolled ina community. 74. Tenth, while this report has identified the key interventions that are likely to speed attainment o f the MDGs, especially in the poor states, it has mt prioritized these interventions. The reason for this is that prioritization o f interventions should in part be based on the cost- effectiveness o f these interventions, and the calculation o f cost-effectiveness entails detailed knowledge o f the costs o f each intervention. The latter is not trivial, especially when the interventions span several sectors (education, health, nutrition, rural infrastructure, etc.). However, this is clearly the next step in this research. In order to come up with a detailed road map o f attaining the MDGs, it will be essential to undertake the costing o f each o f the interventions considered andproposedinthis report. 75. Eleventh, none o f the fore-mentioned interventions are likely to work unless they are simultaneously accompanied by systematic reform o f the institutions o f service delivery inIndia. Additional schools are unlikely to increase enrollment or school completion ifteachers frequently are absent from school. A package o f home-basedneonatal services is unlikely to reduce neonatal mortality rates if health workers are not sufficiently motivated to reach out into the community anddeliver the package o f services to the poor and most-at-risk. The problemo f makingpublicly- provided services work, especially for the poor, has received a great deal o f attention inthe World DevelopmentReport2004. The poor states o f India typically have the most serious problems with governance and service delivery in the social (and other) sectors. Better &livery o f public services - whether in health, schooling, nutrition, or infrastructure - i s a complex and difficult task that entails creation o f the right institutions and incentives, including devolving responsibility for service delivery to local governments and communities, contracting out certain types o f service delivery to the non-government sector, empowering consumers to demand better services from government health facilities, introducing competition among public providers, and ensuring the motivation o f front-line workers. 76. Finally, the importance o f systematically monitoring MD outcomes at disaggregated levels and evaluating the impact o f public programs cannot be overemphasized. This report has highlightedthe pucity o freliable, district-level data on most MD indicators. The lack o f such data makes it virtually impossible to monitor progress toward attainment o f the MDGs at lower levels o f administration. Inaddition, despite the fact that much has been spent by the government on public programs such as the Integrated Child Development Services and the District Primary Education Program, these programs have not been adequately subjected to rigorous, independent evaluation. In order to choose the right set o f interventions with which to attain the MDGs, it i s critical to know which programs have been successful in improving MD indicators and which have not. 1. INTRODUCTION 1.1 Since the launch o f the Millennium Development Goals (MDGs) at the Millennium Summit held in New York in September 2000, the MDGs have become the most widely-accepted yardstick o f development efforts by governments, donors and NGOs. The MDGs are a set o f numerical and time-bound targets related to key achievements in human development. They include halving income-poverty and hunger; achieving universal primary education and gender equality; reducing infant and child mortality by two-thirds and maternal mortality by three-quarters; reversing the spread o f HIV/AIDS; and halving the proportion o f people without mess to safe water. These targets are to be achieved by 2015, from their level in 1990 (UnitedNations 2000). 1.2 Almost all the countries in the world, including India, have committed themselves to attaining the targets embodied in the Millennium Declarationby 2015. Unfortunately, there i s little understanding o f whether India will be able to attain all o f the MDGs, and whether there are some MDGs that India will be able to attain. There i s even less understanding o f what it will take - by way o f economic growth, public spending on social services, and social-sector reform - to attain the different MDGs. Further, this report argues the importance o f disaggregating the MDGs for India, given the very large geographical and socioeconomic variations in millennium development (MD) indicators across the country. 1.3 This report focuses on the attainment o f five major human development-related MDGs by sub-national units in India - child and infant mortality, child malnutrition, schooling enrollment and completion, gender disparities in schooling, and hunger- poverty (as reflected by inadequate calorie intake). The selection o f these MDGs for detailed analysis was based in large part on the availability o f reliable sub-national data. For example, reliable data on disease prevalence at the state or sub-state level are simply not available, and this hampers useful sub-national analysis o f the communicable disease- related MDG. The same i s true o f another important MD indicator - maternal mortality. 1.4 The basic premise of this report is that there are large disparities in the past performance as well as future prospects o f different sub-national units interms o f the MD indicators. India cannot hope to attain the MDGs without significant progress in the MD indicators in its poorest states - Bihar, Uttar Pradesh and Madhya Pradesh. These states not only currently account for a large proportion o f the country's population, but, because o f more rapid population growth, will account for an even larger share o f the country's population n 2015. The report attempts to identify the specific interventions that will improve substantially the likelihood o f the poor states attaining the MDGs. 2 1.5 Virtually all o f the analysis in this report is based on two sets o f national household surveys. First, unit record data from two rounds o f the nationally- representative National Family Health Survey (NFHS), which were collected in 1992-93 and 1998-99, are used to analyze the correlates o f infant mortality and child malnutrition (underweight) (IIPS 1995, 2001a). Second, unit record data from the two most recent quinquennial, `thick' rounds o f the National Sample Survey (NSS) - the 50th round conducted in 1993-94 and the 55th round conducted in 1999-2000 - are used to analyze the correlates o f schooling enrollment, gender disparity in schooling, and hunger-poverty, at the individual or household level (NSSO 1996a, 2001a). While schooling enrollments could have been analyzed with either the NSS or the NFHS data, we have chosen to use the N S S data, as they are more recent and are drawn from a much larger sample. Since the NSS data do not contain information on infant mortality or child weights, the NFHS data remained as the only alternative to analyze these indicators. 1.6 In addition, annual cross-state data over the time period 1980-99 on infant mortality rates, sex-specific gross primary enrollment rates, government expenditure on health, family welfare and elementary education, and gross state domestic product are also used inthis report to analyze the correlates o f changes ininfant mortality and school enrollment over time and across states. 1.7 While one o f the main objectives o f this report i s to present a disaggregated analysis o f MDGs, availability o f data limits the extent to which the analysis can be fully disaggregated. Disaggregation o f analysis to the level o f the state i s possible and i s undertaken here. However, going below the state level poses a problem, as data on very few MD indicators are available at tk district level. In addition, given the sample design and size o f the NFHS and NSS data, it i s not possible to calculate mean values o f the MD indicators reliably at the district level from the household surveys. However, many o f the surveys are representative at the regional - rather than the district - level. (Regions are a collection o f several districts grouped together on the basis o f broadly- similar agro- climatic conditions. The National Sample Surveys (NSS) have defined a total o f 78 regions for India.) Since regions, unlike districts, are not administrative units, the region disaggregated analysis that i s presentedinthis report i s illustrative o f sub-state variations in socioeconomic indicators, but is obviously not as useful as district- level analysis for policy-making purposes. 1.8 The methodological approach adopted throughout this report i s as follows. We apply econometric estimation techniques to unit record data in order to analyze the socioeconomic and policy correlates o f the selected MD indicators. These estimates are then used to simulate the likely trajectory o f the MD indicators under altemative scenarios o f change between now and 2015. 'In addition, there is another nationally -representative household survey - the Human Development Survey - conducted by the National Council o f Applied Economic Research in 1993-94 which contains information on all the indicators analyzed in this report. However, since this survey was conducted only once, it does not lend itself to the analysis o f changes in various human development indicators over the decade o f the 1990s. 1.9 By its very nature, any empirical analysis is predicated on assumptions about data quality and measuremed, inferences o f causality between variables, and potential biases o f statistical and econometric estimates. The analysis presentedinthis report i s not immune to these same concerns. It i s therefore important to note at the outset that while the results and simulations presented inthis report may give an impression o f precision, they are not that. They should be treated as indicative o fpossible broad trends, and could usefully be complemented with other analyses using different methodological approaches. As long as the results are usedwith this understanding, they can be helpful in `rough-order' planning for MDG attainment. 1.10 Finally, it is important to note an important limitation o f the simulations performed in this report. The simulations are based on statistical analysis o f household survey data. By its very nature, such analysis tends to over-emphasize readily-measurable variables, such as household income or consumption, adult schooling levels, and access to infrastructure, and under-emphasize qualitative variables, such as the quality o f institutions, governance, and empowerment. Obviously, this does not imply that the latter variables are irrelevant to the MD indicators; indeed, institutional reform and good governance are critical to the attainment o f the MDGs. While an effort i s made in this report to include a discussion o f institutions and governance, this is not a central focus o f the paper. It is therefore important to view the messages o f this report as complementing those from the numerous qualitative (and detailed) studies o f health, nutrition, schooling and poverty that have been conducted inthe past. In addition to lack o f precision, the estimates presented in this report, like other econometric estimates, may be subject to systematic biases arising from measurement errors in the independent variables and from the omission o f important variables and unobserved heterogeneity from the analysis. 4 2. INFANTAND CHILD MORTALITY 2.1 Reduction in infant and child mortality i s Figure11.1:Infant mortalityrate,by residence,1971-2000 likely the most important o f +Rural the millennium l o t a l development goals, as Predicted +Urban children are the most important assets o f a .-.-. '. nation. In India, approxi- - = . mately 1.72 million child- 9 . -. ren die each year before =2E3.8- reaching their first birthday, which represents one o f the greatest wastes 1 IMR of27 IMR of46 3.3 3 , O J . . , . . ocountry. f humanThe potential in the ;2 ~ . . . . . . . . . . . . . .~. . .. . , ~ . . .~. . . . . . . . ~. ~ ~ 2 2 5 2 2 2 2 2 2 2 2 2 2 2 , , " N millennium develop-ment goal i s to reduce infant and child mortality by two-thirds between 1990 and 2015. For India, this would imply a reduction o f the infant mortality rate (IMR) to 27 and o f the under-five mortality rate (U5MR) to 32 by 2015. A. Overall Trends 2.2 Infant mortality has declined impressively Figure 11.2: Infant mortality rate, 1970-2000, selected countries in Asia 18 in India - from 130-140 140 infant deaths per 1,000 live 0 1970-75 a1980-85 1994-2000 I20 -3 3% births in the early 1970s to 09 68 in 2000, representing an 100 annual rate o f decline o f 80 Annual rote of declrne m IMR 1970-2000 about 2.6 percent (Figure -3 9% 11.1). Further, there i s no 60 60 evidence o f a slowdown in 40 the rate of infant mortality decline over this peri~d.~ If 20 the rate of IMR decline ---- experienced during 1971- 0 S Korea Sn Lanka Thailand Indonesia 3angladesh 2000 is maintained in the A regression o f the log of the infant mortality rate from 1971 to 2000 on the time trend and the square o f the time trend results in a coefficient on the squared term that i s not significantly different from zero. However, the rate o f infant mortality decline during the 1990s was slower than during the 198Os, as Cleason et al. (1999,2000) have argued (although it was not slower than during the 1970s). 5 future, the IMR for the country as a whole could be expected to fall to a level o f 46 in 2015 - still well above the MD goal o f 27. However, as argued later in the chapter, this i s a simplistic projection that does not recognize the underlying factors that determine infant mortality. 2.3 The data inFigure 11.1also suggest that the rural areas o f the country have seen a slightly greater decline in IMR than the urban areas, although the rural-urban gap in IMRremains very wide (with the rural IMRbeing 72% greater thanthe urbanIMR). 2.4 While India has managed to reduce its IMR significantly over the last three decades, its performance on IMR Eduction pales in comparison to that o f many other countries in South, Southeast and East Asia. Infant mortality has fallen by anywhere from 3-5% annually in the countries shown in Figure 11.2, with South Korea being the stellar performer. Even within South Asia, Sri Lanka and Bangladesh have managed to reduce their IMRs at a faster rate than India (4.3% and 3.3%, respectively). Indeed, what is surprisingis that the level o f infant mortality is now higher inIndia than inBangladesh - a country whose per capita GDP i s only about one-half of India's. Health Surveys indicate Madhya Radesh * 1 3 5 ' that under-five child mortality (USMR) rates were 109.3 per 1,000 live :9 9 5 ' Arunachal Pradeshr births in 1992-93, and declined to 94.9 by 1998- f3 -7 5 . 99, an almost identical rate 5 5 . o f decline to that in the $ 3 5 . infant mortality rate. The 15 1 B. Inter-State Variations 2.6 It is almost meaningless to talk about an average infant mortality rate for India, as there are extremely wide inter-state variations within the country. The IMR in India ranges from a low o f 14 for Kerala to a higho f 96 for Orissa (Figure 11.4). Thus, Kerala i s comparable to Bulgaria, Russia and Ukraine in tenns o f its IMR, while Orissa is 6 comparable to Lesotho, Cameroon and Tanzania. This is a wide range o f IMRs for a single country. 2.7 How have different Indian states Figure11.4: Infant mortalityrate acrossIndian states, 1981 and 2000 performed in terms o f 0Levelin2000 reducing their IMRs? The -% annual change, 1981-2000 90 data shown in Figure 11.4 are intriguing, since they go against many widely- held perceptions. For instance, the state with the lowest level o f infant mortality in 1981 - Kerala - reduced its infant mortality at the fastest rate (5% annually) between 1981 and 2000. But Bihar and U.P., which had among the highest IMRs inthe country in 1981, were also among the top performers inIMR reduction over the same period. Andhra Pradesh and Karnataka - states that are normally perceived to be good human development (HD) performers - had the slowest rate o f IMR decline over the two decades. Ingeneral, there was some - although limited -convergenceinIMRs,sothatinter-statedisparityininfantmortalitydecreasedbetween 1981 and 2000.'0 2.8 While the dis- cussion on MDGs to date Figure 11.5: State-specificMDGsfor Infant Mortality, 2015 has focused on individual OIlSSa 41 countries reaching the MD Madhya Radesh UnarPradesh goals by 2015, there i s no RaJasthan reason why the MDGs All-India A" should not be applied to Bihar individual states or Gujarat Kamaraka provinces within countries. Andhra Pradesh Indeed, as long as the Haryana WBengal distribution of the countw 's hnjab population across states TamilNadu Maharashtra does not change Kerala appreciably, a national MD 0 5 I O 15 20 25 30 35 40 goal o f reducing infant mortality by two-thirds (between 1990 and 2015) could be achieved by each state reducing its IMRby two-thirds. Figure 11.5, which shows the implied2015 MD goals for each o f the states, indicates that the range o f IMRs that will need to be attained by each loFor instance, the log variance o f the infant mortality rate across states fell from 4.7 in 1971 to 4.1 in 2000. 7 state by 2015 for the comtry as a whole to meet its MD goals is wide - from a low o f 6 for Kerala and 41 for Orissa. 2.9 In fact, however, the high-mortality states in the country - Rajasthan, U.P., M.P., Bihar and Orissa - will account for a larger share o f India's population by 2015 owing to their higher fertility and population growth rates. In turn, this will mean that these states will have to reduce their IMRs to even lower levels than shown in Figure 5 in order for the country to meet its MD goal o f 27. For states like Orissa and Madhya Pradeshto reduce their IMRs by more than 55 infant deaths per 1,000 live births over the next 12 years will be a major challenge. '' C. Intra-StateVariations 2.10 The state averages o f infant mortality mask substantial intra-state variation. Unfortunately, recent estimates o f infant mortality are not available at the district level. Butestimates are available at the regional level for 1997-99 from the Sample Registration Surveys (SRS). l2Appendix Table 8, which presents regionlevel estimates o f IMR for the country, indicates that the regional IMR ranges from a low o f 7.8 per 1,000 live births to a high o f 125.3 in 1997-99. The IMR in Northern Kerala i s more than two times as high as that in Southern Kerala (19 versus 7.8 per 1,000 live births). In Kamataka, the IMR ranges from 38.8 in the Coastal & Ghats region to 76.5 in the Inland Southern region. The data inAppendix Table 8 suggest that the interior o f the country generally has higher IMRsthan the coastal regions, with a few exceptions. 2.11 Appendix Table 8 also presents regional data on infant mortality rates in an earlier period- 1988-92 (based on the 1991 Census). The table suggests that, while infant mortality fell inmost parts o f the country, a number o f regions experienced no change or even an increase in infant mortality. These regions were distributed throughout the country -- inthe West, Center and the South. D. Geographic Concentrationof InfantDeaths 2.12 It is obvious that targeting IMR-reducing interventions to populous states with high mortality will be critical to achieving the MDGs. In addition, from a welfare (and policy) perspective, it may be inportant to attain the largest reduction possible in the absolute number o f infant deaths inthe country. For this reason, it i s useful to look at the contribution o f individual states to the number o f infant deaths nationally, in addition to the rate o f infant mortality. The two measures will yield different answers if the states that have the highest IMRs are generally not as heavily populated as the states having lower IMRs. In the case o f India, however, many o f the states having the highest IMRs ''Both o f these states reduced their IMR by 33 infant deaths per 1,000 live births duringthe 12-year period from 1988 to 2000. As noted earlier, regions are a collection o f several districts grouped together on the basis o f broadly- similar agro -climatic conditions. They are not, however, administrative units.The National Sample Surveys (NSS) have defined a total o f 78 regions for India. Data from household surveys, such as the NSS, are often representative at the regional - rather than the district - level. As a result, sub-national disaggregation o f MDindicators inthis paper hasbeen limited to the levelofthe region. 8 are also among the most Figure 11.6: Contribution of the 21 larger states to national infant deaths,2000 populous states in the country (e.g., Uttar Pradesh and Madhya Pradesh). As a result, the two measures - the infant mortality rate and the absolute number o f CJ Cumulativeshare in total number o f infant deaths nationally infant deaths - paint a Share intotal numberofinfantdeathsnationally similar picture o f where the problem o f infant mortality i n I mainly lies. 2.13 Figure 11.6, which plots the individual as well as the cumulative contribution o f the 21 larger states to the total Figure 11.7: Cumulative distribution of infant deaths in India across number o f infant deaths in districts and villages, 1994-98 the country in 2000, shows that Uttar Pradesh alone contributes one-quarter of all infant deaths in the country.13 Four states - Uttar Pradesh, Madhya Pradesh, Bihar and Rajasthan - together account for slightly more than one-half o f all infant deaths in India. The 0 I O 20 30 40 50 60 70 80 90 100 geographical concentration Cumulative %ofdistricts or villages (rankedby infantdeaths) o f infant deaths points to the importance as well as the potential efficacy o f state-level targeting o f IMR-reducing interventions. 2.14 In principle, targeting interventions to a smaller geographical unit, such as a district or village, could be even more effective. The NFHS-2 data are suggestive o f infant deaths in India being heavily concentrated in a relatively small number o f districts and villages. For instance, in the period 1994-99, a mere 20% o f the villages and 22.5% of the districts in the NFHS-2 sample with the largest number of infant deaths accounted for half o f all infant deaths inthe country duringthat year (Figure 11.7). Since the NFHS- 2 covered only a fraction o f all the villages in the country and the number o f sampled households in each village i s too small to be representative, these numbers are merely suggestive o f possible patterns. There are promising new methodologies available for more accurate identification o f village HD and poverty outcomes on the basis o f merged household survey and population census data. It would be worthwhile to explore the use o f such methodologies to identify the districts and villages with the largest number o f l3The state accounts for slightly less than 17percent o f India's population. 9 infant deaths in the country, so that government initiatives could be better targeted to these districts and villages. 2.15 Concentration o f infant deaths occurs even below the level o f a district or village. Using data from the NFHS-2 for three states (Uttar Pradesh, West Bengal and Kerala), Arulampalam and Bhalotra (2002) find that the probability o f a child dying during its frst year o f life is 0.03 if the sibling born immediately prior to it survivedpast age one but rises to 0.14 if the previous sibling died in infancy. This results in a clustering or concentration o f infant deaths within particular households. In practice, however, targeting policy interventions to households i s considerably more difficult than targeting o f interventions to entire villages or districts. E. Economic Growth,PublicSpendingonHealth,andInfantMortalityReduction 2.16 It is clear that Figure 11.8: Infant mortality rates and real public spending on health and family planningper capita, across 13 major states, 1980-99 there has been a wide divergence in the 160- t performance o f different 140- states in reducing infant 120. .. mortality. An important 2e P 100- \ ' issue i s the extent to which 2 8 0 - this divergence i s 6 0 - associated with economic 3.E 8 growth and public spending 40. on health. We have merged 20. 4 state-level data on IMRs . . . . . . . . . . . . . . , 20 30 40 50 60 70 80 90 100 110 120 130 140 150 over the period 1980-2000 Real gov't health & family welfare exp. per capita (1994 Rs.) with state-level data on real gross state domestic product per capita (GSDP) and real public spending on health over the same period to explore the association between infant mortality on the one hand and public spending and economic growth on the other. l4>l5 2.17 A plot of the data suggests an inverse association between infant mortality and real government expenditure on health and family welfare (Figure 11.8). Inaddition, there i s a hint o f the association being stronger at lower levels o f public spending on health and weaker at higher levels, although with no controls for a time trend or for other variables that may also be associated with infant mortality (such as per capita income), it i s difficult to make a definitive statement about the association. l4 Data on actual (not simply budgeted) govemment health expenditure were obtained from detailed budget demand documents o f individual states. Expenditure on health includes spending on public health; urban and rural health services; medical education, training and research; general administration; and family welfare (Le., population programs, including maternal and child health). It includes expenditure incurred by a state out o f its own revenues as well as central government health allocations to that state. The association between infant mortality andprivate spending on health is o f less interest because private spending i s not a direct instrument of policy. In addition, as private expenditure is subject to household choice, it i s highly endogenous to health outcomes. 10 2.18 In Annex 11.1 an attempt is made to estimate the relationship between infant mortality, government health expenditure, and GSDP more formally and with controls for other variables. While the results o f the analysis are not unambiguously clear, there i s some evidence, although not persistent, o f a significant inverse association between infant mortality and government health expenditure across states. In addition, some, but not all, specifications seem to indicate a stronger inverse association for the very poor states than for the nonpoor states. 2.19 The econometric results presented in Annex 11.1 indicate significant associations between infant mortality on the one hand and female literacy and per capita GSDP on the other. Interestingly, per capita income and female literacy have a significant interactive association with infant mortality such that the inverse association between infant mortality and female literacy is stronger at higher income levels than at lower income levels. 2.20 The stronger inverse association between infant mortality and public spending on health for the very poor states, observed in some o f the econometric specifications, may be related to the composition o f infant deaths across poor and norrpoor states. Typically, states with highinfant mortality rates have a larger proportion o f infant deaths occurring during the post-neonatal period. These deaths are more easily averted by the typical (and relatively inexpensive) child survival interventions. As such, it i s much easier to bring down infant mortality rates in the poor, high-mortality states with proven, low-cost, post-neonatal interventions, such as child immunizations and oral rehydration therapy. On the other hand, averting neo-natal deaths, which are relatively more common in the better-off, low-mortality states, typically requires more expensive interventions, such as professionally- attended deliveries or deliveries in institutions as well as post- delivery and hospital-based emergency care. l7 highlightsthe need for state-stratified This mortality redwtionpolicies. 2.21 Innate Mortality Differences across States. An appealing feature o f the fixed- effects regression for infant mortality that has been estimated and discussed in Annex 11.1 i s that it controls for observed and unobserved (but time-invariant) heterogeneity in health conditions across states.l8The heterogeneity could arise from inter-state differences in culture, history, governance (in-cluding poor public expenditure management), and initial health conditions (typically caused by time- invariant factors, such as geographical location or isolation). An examination o f these IMR `fixed effects' indicates, predictably, that Kerala has the lowest levels o f infant mortality even after controlling for public spending on health, per capita income, and adult female literacy (Figure 11.9). Surprisingly, however, the poor, high-mortality states, such as Bihar, Uttar l6 Likewise, the inverse association between infant mortality and per capita income is stronger at higher levels o f adult female literacy. The results indicate that female literacy and the level o f development in a state complement (rather than substitute for) each other interms o f their association with infant mortality. "Althoughneo-natalmortalityreductiontypically requireshospital-basedcare, itispossibletoprovidea relatively inexpensive package of home-based neonatal services, as shown by a highly-successhl field trial inMaharashtra in 1995-98 (see Box 11.1for a detailed description o fthe intervention). '* The fixed effects are recovered by averaging the residuals o f the estimated infant mortality regression over the sample period for each state. 11 Pradesh, Madhya Pradesh and Orissa, do not rank Figure 11.9 States ranked by "innate" infant mortality (i.e., infant mortality after poorly on this indicator, control for per capita GSDP, public spendingon health, and adult female literacy, averaged over 1983,1987,1993 and 1999) and by observedIMR averaged over indicating that these states 1980-99 enjoy rehtively low infant mortality relative to their per capita income, public spending on health, a d female literacy levels. At the other extreme, Gujarat, Tamil Nadu and Karmtaka Actual IMR averaged over 1980-99 have the largest positive residuals, indicating that these states have relatively high infant mortality rates relative to their per capita income, public spending on health, and female literacy. 2.22 An encouraging interpretation o f these results is that the poor states are not intrinsically predisposed to high levels o f infant mortality. Infant mortality can be brought down significantly in these states with a combination o f interventions that include economic growth, improvements in female literacy, and increased public spending on health. F. GovernmentHealthExpenditure 2.23 Figure 11.10, which reports the amount Figure 11.10: Per capita government revenue expenditure (state plus central) on health and family welfare (nominal prices), by state, 1999-2000 o f government health Punjab expenditure per capita in Kerala the different states, TamilNadu indicates wide variations in Gujarat Kamataka public spend- ing across Rajasthan states, with the levels o f W Bengal spending in Bihar and Uttar Andhra Pradesh Maharashtra Pradesh being only half as Havana much as in Maharashtra Orissa and Andhra Pradesh and a Madhya Pradesh Wnar Pradesh mere third o f the level Bihar found in states such as 0 25 50 75 100 125 150 175 200 225 250 Kerala and Punjab. 2.24 However, the performance o f the poor states, such as Bihar, Uttar Pradesh, Madhya Pradesh, Rajasthan, and Orissa, has beenmixed when one considers the increase over time inreal public spending per capita on health (Figure 11.11). While Uttar Pradesh and Orissa had the slowest real increase in public spending on health over the period 1981-99, Rajasthan experienced the largest increase o f any state, and Bihar ranked in the middle o f the states considered inFigure 11.11. 12 2.25 Rajasthan and Bihar also saw a sharp Figure 11.11: Average annualincrease (%)in real governmenthealth expenditureper capita,by state, 1981-99 increase between 1981-82 Rajasthan and 1999-2000 in the ratio TamilNadu o f public spending on Gujarat punlab health to GDP, although, in W Bengal the case of Bihar, this AndhraPradesh largely reflected the very Bihar Kerala slow increase in real GDP Kamataka over these years (Figure MadhyaPradesh 11.12). On the other hand, Maharashtra the ratio of government Onssa UttarPradesh health expenditure to GDP Havana declined in Uttar Pradesh 1 5 2 0 2 s 3 0 3 5 4 0 4 5 and stayed stagnant in Orissa. Figure 11.12: Public spending on health as % of gross state domestic 2.26 How does India's product, 1981-82 and 1999-2000 spending on health compare 1.3 1 N to that of other developing I 2 01981 E41999 countries, especially in the I 1 region? Figure 11.13 1 0 suggests that India 0 9 generally under-spends on 0 8 health relative to other 0 7 countries in Asia (with the 0 6 exception o f Pakistan, Nepal and Indonesia). 0 5 Even within South Asia, Bangladesh and Sri Lanka spend significantly more resources on health (as a percentage o f their GDP) Figure 11.13: Public expenditureon health as % of GDP, selected than does India. countries inAsia, 2000 (Source:UNDP 2003) Thailand 2.1 2.27 In addition to the total amount o f public Sri Lanka spending on health, it i s Philippines important to also consider Bangladesh the functional composition Viemam of health spending. It is Pakistan widely recognized that Nepal health outcomes, especially infant mortality outcomes, India are more responsive to Indonesia preventive health 0.5 I.0 1.5 2.0 interventions, including communicable disease control and child immunization, than to 13 the provision of curative services, typically in hospitals. In addition, evidence from several countries, including India, indicates that public spending on immunizations and preventive care is much more pro-poor than that on hospitaLbased curative services. For instance, Mahal et al. (2002a,b) find that the bottom two consumption quintiles receive 47% o f the public subsidy for immunizations in India but less than 20% o f the public subsidy for hospitals. 2.28 Unfortunately, the lion's share o f government Figure11.14: Functionalcomposition of state governmenthealth expenditures, 1981 health expenditure in most Indian states goes toward - the provision of health loo% services, typically in 90% 80% hospitals. Relatively little i s spent on preventive health 60% activities. The functional 50% distribution of state 40% government health expefi 30% diture in 1981 and 1999 is *O% shown in Figures 11.14 and 10% 11.15. What i s disconcerting o% WPublic health Q F m l y welfare =Medical trainlng =Medical services to note i s that in virtually all the states, with the exception of Maharashtra, Tamil Nadu and Kerala, the share o f governmerl: health expenditure devoted to public health declined between 1981 and 1999. Correspondingly, the- share spent on medical services increased. Figure11.15: Functionalcomposition of state government health expenditures, 1999 2.29 Getting the states to increase their relative 100% allo-cation to preventive 90% health activities i s not 80% straight-forward, however. 70% One o f the characteristics 60% o f public spending on 50% health in India i s that the 40% central govern-ment 30% accounts for a large share 20% o f the total govern-ment 10% 0% expenditure on public BPublic health =Family welfare DMedical training OMedical services health and family welfare activities in a state - typically via centrally sponsored (vertical) schemes. The states tend to finance medical services, typically curative, out o f their own revenue.l 9 In 1999-2000, l 9Of course, strictly speaking, not all o f the state government spending on medical services is curative, since some o f it goes into the financing o fprimary and community health centers, which are responsible for many public health functions. Sub-health centers are, however, largely financed by the central government. 14 for example, central government grants accounted for one-half to nearly three-quarters o f government expenditure on public health and family welfare activities in six states for which reliable data are available (Figure 11.16). Further, there i s evidence that central government financing o f public health and family welfare activities in the states has increased somewhat over the years. 2.30 There are two ways to increase the share Figure 11.16:Share of central grants in total government(revenue) expenditure on health, by function,selectedstates, 1991-92 and 1999- o f public health and family 80 AndhraPradesh lAssam DOnssa ORaJasthan mTamilNadu @ W Bengal welfare in total government 70 health spending in the 60 states. First, the central so government could simply 40 increase the volume o f 30 resources it allocates to the 20 states for public health and family welfare activities. This is unlikely to be a we are I spending sustainable alternative for a 1999-2000 1991-92 number o f reasons, Share(%) of centralgovemment grants intotal publtcspendingbythe state including the possibility that the states could then respond by allocating even fewer o f their own resources to preventive care and M C H (and more to medical and curative services). Second, the central government could make its grants to the states as matching grants, so that the states that spend more o f their own resources on public health and family welfare activities could qualify for larger central grants. This type o f matching formula might create the necessary incentives for states to direct increasingly larger shares o f their overall health spending to communicable disease control and preventive health activities. 2.31 Quality of Public Spending. The finding that the decline in infant mortality relative to the increase in real per-capita public spending on health over 1981-99 was largest in the poor, high-mortality states, such as Uttar Pradesh, Bihar, Madhya Pradesh and Orissa (Figure II.9), should be tempered by two observations. First, these states do considerably worse than the norrpoor states, especially inthe South, interms o f targeting government health subsidies .to the poor. Based on NSS data from 1995-96, Mahal et al. (2002a) find that all the southern states, but particularly Kerala and Tamil Nadu, show a `progressive' distribution o f subsidies (in that the ratio o f subsidies to per capita household consumption expenditure falls with per capita household expenditure), whereas the poor states (such as Uttar Pradesh, Madhya Pradesh, Bihar, Rajasthan and Orissa) have a `regressive' distribution o f public subsidies on health. 2.32 Second, governance and service delivery problems in the poor states are often worse than in other states. In the poor states, there i s widespread absenteeism o f doctors and paramedics at government health centers and sub-centers; most government health facilities are in disrepair; and the availability o f drugs and medical supplies at public health facilities i s typically nonexistent. For example, a recent survey across India indicates that 58% o f health workers in primary health facilities in Bihar were absent 15 from their positions on any given day (Chaudhury et al. 2003). Inthe country as a whole, 43% o fprimary health care workers are absent from their place o f work. Another national survey conducted by the Public Affairs Centre, which ranked the performance o f government health services on a number o f indicators, such as access, usage, reliability and user satisfaction, found that the poor states Figure11.17:Rankingof statesinthe performance of government health typically rank inthe bottom services,2002 Sikkim half o f all the states in the Gujarat country in terms o f the Maharashtra Taml,Nadub h j a quality o f their public Hunachal health services (Figure Mizoram Karnataka 11.17) (PAC 2002). The Haryana Kerala poor quality o f health UnarPradesh Assam services reflects many MadhyaRaJaSthan Pradesh factors at work, including lack o f accountability for WestBengal Onssa Meghalaya public health providers and Arunachal Bihar disenchantment with Tnpura Nagaland 21 working conditions among AndhraPradesh 22 health workers. G. ProximateCausesof InfantMortality needed to help India attain "0 its infant mortality MDG, 5 Other States one needs an understanding 3 980:; - - 'Poor States ointermediate - causes oro f f the proximate - .$ '-.'.--.'... 960' infant mortality in India. 'f '. 940' An answer to this question 1b can be found in the data on 920. - .......*..-..._.__. timing o f deaths, obtained -".....-.---* 900- -- - --..__.- from analysis o f unit record 8 8 0 1 . . . 8 . r 8. I I . r I I I . . I . r . . I I . . r I I I . . . I I . v I . r . I ... . . . .... ....,.../ 2o Infant mortality rates can be calculated from the NFHS-2 for the four-year period preceding the survey (i.e., 1994-98). See Annex B for a description o f the data. 21 These figures are broadly similar to those published by the civil registration system (see Medical Causes ofDeath, 1999). 16 life (the neonatal period). Unfortunately, data on the specific medical causes o f neonatal deaths are not readily available (inlarge part because the rrajority o f infant deaths are not medically attended), but it is likely that factors such as maternal malnutrition and post- delivery complications, such as umbilical cord sepsis, are generally responsible for many neonatal deaths. (A three-year field trial to provide home-based neonatal care by village health workers in the state o f mharashtra in 1995-98 resulted in a 46% reduction in infant mortality and a 62% reduction inneonatal mortality inthe treatment area. See Box 11.1at the end o f this chapter.) 2.34 The NFHS-2 data provide some information on the proximate correlates o f infant mortality. These are shown in Figure 11.19 for two groups o f states: poor Figure 11.19: Proximate causes of infant mortality in poor and nonpoor states, 1996-98 states (i.e., those that have a 100 N OPoor states .Other states DAll states monthly per capita 90 consumption expenditure o f 7o Rs. 676 or less according to 60 the National Sample i: 50 Survey 1999-2000 round) and nonpoor states (those 2o having a per capita IO consumrdion exDenditure o f 0 more `than Rs. 676)?2 No I Yes I All <=2 5 kgsl >25 kgs I All Figure 11-19 indicates that Whether motherreceivedany Birthweight of child tetanusshots dunngpregnancy? when a woman receives I I antenatal care for thisdelivety? I I tetanus shots during her pregnancy, the subsequent survival prospects for the child from that pregnancy are significantly raised. Inthe poor states, the infant mortality rate i s nearly two times as high when the mother has not received tetanus immunization than when she has (92 versus 50).23 2.35 Likewise, children with low birth weights are more than two times as likely to die during infancy as children who weighed more than 2.5 kilograms at birth.24Medical attendance (i.e., doctor, nurse or skilled birthattendant present) at birth i s associatedwith '*Definedin this manner, the poor states are Andhra Pradesh, Assam, Bihar, Madhya Pradesh, Manipur, Orissa, Rajasthan, Sikkim, West Bengal, Uttar Pradesh and Tripura. The other (non-poor) states are: Goa, Gujarat, Haryana, Himachal Pradesh, Jammu, Kamataka, Kerala, Maharashtra, Meghalaya, Mizoram, Nagaland, Punjab, Tamil Nadu, N e w Delhi and Arunachal Pradesh. 23 The data on antenatal care and institutional deliveries were collected in the NFHS-2 only for the last two pregnancies o f every eligible woman during the 35 months preceding the survey. A s a result, the sample o f children for whom data on antenatal care, place o f delivery, and anthropometry are available is not a random sample o f the entire child population or o f all the live births in the country. Hence, the infant mortality rate obtained from this (truncated) sample is not comparable to the IMR reported in the succeeding section o f this chapter (which is based on all live births during the four years preceding the survey). 24 Note that data on birth weights were not available for all children in the NFHS-2 survey. Indeed, the data suggest a strong selection bias in the women who reported their child's birth weight; the infant mortality rate for this group o f children is significantly lower than the IMR for children with missing birthweights. 17 a large reduction in infant mortality, especially in the poor states. Interestingly, however, receipt o f any antenatal care by a woman i s even more strongly associated with infant survival than professional attendance at birth. Inthe poor states, the infant mortality rate for children whose mothers did not obtain any antenatal care i s 83% greater than that for children whose mothers obtained some care. Thus, antenatal care and tetanus immunization of a pregnant woman appear to be more strongly associated with infant survival prospects than professional attendance at birth. 2.36 A child's birth weight is, in large part, Figure11.20:Low birth weight children, infant mortality,and maternal weight, 1998-99 influenced by the mother's 901 nutritional status at 79 Mother'sweight (kgs ) delivery. This i s observed in Figure 11.20, which shows that while 32% o f children aged 035 months whose mother's weight was less than 35 kgs were likely to be o f bw birth weight, only 15% had low birth weight when the mother's Poor states INon-poorstatesI All India Poor states I Non-poorstates I All India weight was 50 kgs or more. % of children0-35 months old who had low birth Infant mortality rate(per 1,000 live births) weight (<2,500 gms) Since infant mortality is related to low birthweights, this implies that infant mortality i s also determined inpart by maternal weight. Again, Figure 11.20 indicates this to be the case, but only in the poor states o f the country. In these states, infant mortality is nearly 30% higher when the mother weighs less than 35 kgs than when she weighs 50 kgs or more. Figure 11.21: Infant mortality rates, by predictedprobability of a child beingseverely underweight 2.37 After the first Predictedprobabdity of beingseverely underweight month o f life, child 9o malnutrition becomes an important contributing 70 factor to infant and child 60 mortality in India. 50 Malnutrition sets in early, 40 often owing to improper 30 feeding practices, such as 20 10 early termination o f exclusive breast-feeding Poor states Other states None Pnmary Secondary Higher and introduction o f Incomelevel o f Mothei`s state education level (inadequate) spplementary feeding. In addition, even during the exclusive breastfeeding period, infants may be malnourished owing to insufficient quantities o f breast milk - in turn the result o f poor nutrition and heavy workload o f poor women. Malnourished infants are more prone to 18 diarrheal, respiratory and other infections, which, when untreated, can lead to infant death. 2.38 It i s often difficult to establish the correlation between child malnutrition and mortality in the absence of longitudinal data, since anthropometric data from most cross- sectional surveys are available only for children living at the time o f the survey. We have used the NFHS-2 data to predict the probability o f a child being severely ~ n d e r w e i g h t , ~ ~ based on a maximumlikelihood probit regression o f underweight status on various child and household factors (e.g., age, sex, birth order, maternal schooling, schooling o f the household head, etc.). Although the probit regression is estimated only on a sample o f living children (obviously), the risk of malnutrition can be predicted for all children - living and dead - based on their characteristics at birth (such as sex and birth order) as well as on the characteristics o f their households. Infant mortality rates for children at different predicted risks o f severe weight malnutrition are shown in Figure 11.21. The results are striking; they suggest that, in both poor and nonpoor states, the risk o f infant death increases dramatically with the probability o f being severely underweight. An infant with a 20% risk o f being severely underweight Figure 11.22: Relationship across regionsbetweenunder-five child mortality rate and child underweight rate, 1998-99 has an infant mortality rate that i s more than two times I 6 O1 % that o f an infant with a less- than-10% risk o f being severely underweight. This difference in infant mortality i s observed even for women with primary, secondary and higher schooling, suggesting that maternal schooling cannot easily reverse the low 10 15 20 25 30 35 40 45 50 55 60 % o f children0-35 monthsunderweight survival prospects o f a malnourished infant. Figure 11.23: Relationship across regions betweenone-year olds immunized against measles and the mortality rate of 1-5 year olds, 1998-99 2.39 Data from the NFHS-2 on under-five mortality rates and child underweight rates for 69 regions also show a positive, albeit not perfect, association between the two rates (Figure 11.22). Of course, some regions, such as the Central Region of 0 4 1 Uttar Pradesh, Meghalaya 15 25 35 45 55 65 75 85 95 105 % ofchildren aged I@e., 12-23 months) immunizedagainst measles 25 As in the literature, a child i s considered severely underweight when his or her weight-for-age is more thanthree standarddeviationsbelow the NCHSreferenceweight. 19 and Sikkim, have unusually high under-5 mortality rates relative to their child underweight rates, while other regions, such as NorthernKerala and Coastal Maharashtra, have unusually low under-5 mortality. 2.40 In addition to nutrition, immunization plays an important role in enhancing a child's survival prospects, especially beyond infancy. Indeed, child immunization i s an important enough inputinto child mortality to be considered as a separate MDG (with the goal being universal immunization o f one-year olds against measles). The evidence from NFHS-2 relating to the role o f child immunization on mortality i s shown in Figure 11.23 and Appendix Table 8. Two observations can be made from these figures; first, there are very large regional variations in the child immunization rate. (See Appendix Table 8 for regional data on measles vaccination coverage.) The percentage o f one-year olds immunized against tetanus ranges from 16% (in Northern Bihar) to 100% (in Eastern Maharashtra). Second, higher measles vaccination rates are associated with lower child mortality between ages 1 and 5 years (Figure 11.23). ""1 2.41 Figure 11.24 shows the measles Figure11.24: Measlesvaccinationrate (%)among 12-23 montholds,by state, vaccination rate among 12- 1992-93 and 1998-99 YU -" _I 23 month olds by states in 1992-93 and 1998-99. Most 01992-93 1998-99 7o states experienced 4 relatively slow progress in so expanding measles 40 immunization over that 6- 30 year period, with the 2o overall rate o f measles lo vaccination in the country 0 increasing from 42% to bC"@ $ J ~~~~~~~~~~~~~~~~~~ 51%. In three states - Madhya Pradesh, +Y Q@ 8K~ Rajasthan, and Assam - the measles vaccination rate actually fell over the six years. In Bihar, which had only 15% o f children immunized against measles in 1992-93, the immunization rate increased by merely 2 percentage points between 1992-93 and 1998- 99. Thus, there appears to be significant potential for reducing infant - especially, post- neonatal - mortality in the poor states o f the country via an expansion o f child immunization. -- H. Socioeconomic andYolicy Correlates 01lnlant-Mortality . .- -. ..- -. I I " * 2.42 Inthis section, we present descriptive statistics on infant mortality based on the NFHS-2, tabulated by various socioeconomic characteristics o f households, access to infrastructure, and availability o f government programs. It should be noted that the associations shown and discussed below are simple correlations, not necessarily causal relations. 20 2.43 Gender disparity and birth order. Much has been written about sex differentials ininfant mortality inIndia. Indiais said to be one ofthe few countries inthe world where females have a higher infant mortality rate than males. The NFHS-2 data do not show a significant disparity in average male and female infant mortality rates, but this in itself is evidence of parental discrimination against female infants, as one would expect the infant mortality rate for males to be well above that for females in a nondiscriminatory environment. Further, the data show large gender differences in infant mortality for higher birth-order children (Table 11.1). Girls o f birth order 4 or more experience significantly higher rates o f infant mortality than boys o f similar birth order (84 versus 75), with this difference being larger inthe nonpoor states than inthe poor states. Table 11.1:Infant mortality rates by child birth order, sex, and by group of states, 1994-98 Poor States Other States All States BirthOrder ofchild Male Female All Male Female All Male Female All 1 91 76 84 50 41 46 76 63 70 2-3 65 65 65 45 44 45 58 58 58 4 & above 79 88 83 59 71 65 75 84 80 Total 76 76 76 49 48 49 68 67 67 2.44 Child mortality betweenthe ages o f one and five is also significantly greater for females than for males. A girl in India i s 40% more likely to die between her first and fifth birthdays than is a boy. Thus, child mortality would drop by 20% if girls had the same mortality rate as boys between the ages o f 1 month and 5 years (Victora et al. 2003). Figure 11.25, which plots the monthly mortality rate for children by sex (from the NFHS-2 data), indicates that while the probability o f death is greater for males than for females until age one, the reverse i s true from ages one to five. Parental neglect toward girls - symptomatic o f the generally low social status Figure 11.25: Mortality rate (per 1,000 live births), by age (months)and sex, 1994-98 treatment for their illnesses 4 0 . -Males than boys (Das Gupta 1987, 3 0 . Filmer et al. 1998). 20. I O - 21 groups as well (Figure 11.26). Of the three groups, STs have the highest infant mortality, followed by SCs. Although SC/STs in the poor states have the highest absolute IMRs o f any group in the country, the relative position o f SC/STs vis-a-vis the nonSC/ST/OBC groups i s worse in the nonpoor states relative to the poor states. For instance, inthe poor states, STs have an IMR that i s 54% greater than that o f forward castes, but this differential i s only about 37% inthe nonpoor states. 2.46 Female schooling. As i s widely Figure 11.26: Infant mortality ratesofvarious socialgroups in poor and other states, 1994-98 observed in many countries 88 UPoor States OtherStatesm All States (including India), mother's 85 83 schooling i s strongly 80 associated with infant 75 mortality (Figure 11.27). 70 What is interesting i s that 65 while both male and female 60 infants benefit (in terms o f 55 significantly reduced risk o f 50 mortality) from having their 45 mothers even slightly 40 schooled (Le., 1-5 years), Scheduledcastes Scheduledtribes Otherbackward SCs, STs or Forwardcastes All groups (SCS) (STs) castes(OBCr) OBCs female infants enjoy a significant (relative) Figure11.27: Infant mortality by sex and by mother's schooling, 1994-98 survival advantage over 9 0 1 83 their male counterparts only 80 when the mother has 11 or 70 more years o f schooling. 60 Male 0Female 2.47 Rural 50 infrastructure. Rural 40 infrastructure can have 30 powerful influences on health outcomes. For 20 instance, rural roads enable I O easier access to health 0 centers and referral district None 1-5 years 6-8years 9-10 years 11-12years >I2 years Mother'sschooling hospitals, thereby reducing the risk o f an infant dying because o f neonatal and post-neonatal infections. Access to safe drinking water and sanitation are important environmental hygiene interventions that significantly reduce the exposure o f an infant to water- and vector-borne diseases and increase the probability o f his or her survival. Likewise, access to electricity can also improve infant survival probabilities by improving hygiene, cooking and health practices inthe household as well as the health practices o flocalhealth providers. 22 2.48 The NFHS-2 data show si nificant association between infant mortality and access to infrastructure (Table II.2)!6 The infant mortality rate is nearly 50% greater among households without access to electricity as compared to households with electricity, although this difference is somewhat smaller in the poor states than in the non-poor states. It i s not just the availability o f electricity but also the regularity in its supply is associatedwith health outcomes. The regularity o f electricity supply ina village i s associated with significantly lower levels o f infant mortality, although this relationship too i s more predominant inthe better-offstates than inthe poor ones. 2.49 Access to sanitation i s another infrastructural intervention that i s observed to have strong association with health outcomes. Households with no toilet access have an infant mortality rate that is nearly double that o f households with toilet access. This pattern does not differ much across poor and nonpoor states. Table 11.2: Infant mortality rate, by various individual household, child and community characteristics and by poor and non-poor states, 1994-98 Poor Other All Characteristic states states states Household has: No electricity 85 66 82 Electricity 64 45 55 Electricity supply in village is: Irregular 81 57 75 Regular 78 49 65 Household has: No pipedwater 80 54 75 Pipedwater 59 45 51 Household has: Some toilet access 49 34 42 No toilet access 84 60 78 % of villages in district havingpucca road: 50% or fewer 81 53 77 50-90% 78 52 71 >90% 59 45 51 Social group of household: Scheduled Caste 84 58 77 Scheduled Tribe 88 65 81 Other Backward Caste 80 47 70 SC, ST or OBC 83 53 74 Total 76 49 67 2.50 Government programs and health infrastructure. In the poor states, there i s no discernable association between infant mortality and the presence o f two large national government anti-poverty programs in a village - the Intensive Rural 26 Itis worth reiterating that the descriptive statistics shown in Table 11.2 and similar tables throughout this report represent simple associations, not causal relations. 23 Development Program (IRDP) and the National Rural Employment Program (NREP) (Table 11.3). However, in the nonpoor states, there i s some evidence that the NREP, which creates employment through rural public works, i s associated with significantly lower infant mortality (17%). 2.5 1 Interms ofhealth infrastructure, one ofthe few mriables that is associatedwith infant mortality is the presence o f a sub-primary health center in a village. Having a sub- health center in a village i s associated with a reduction o f &9% in the risk o f infant mortality inboth poor and nompoor states. Table 11.3: Infant mortality rate, by village availability of government anti-poverty programs and health infrastructure and bv Door and non-Door states. 1994-98 Poor Other All Characteristic states states states Government anti-poverty programs in village: Intensive Rural Development Program (IRDP) 76 53 68 No IRDP 76 45 67 National Rural Employment Program (NREP) 76 41 63 N o NREP 76 50 68 Availability of sub health center in village N o 82 56 76 Yes 75 52 67 All 49 76 67 2.52 Given that child malnutrition i s an important Figure 11.28: Under-five child mortality rate, by presence of ICDS anganwadi center invillage, 1988-92 proximate cause o f infant ON0 ICDS anganwadi center in village and child mortality, one 140 OICDS anganwadi center mvillage might expect child 129 mortality to be associated with the presence o f a national program such as the Integrated Child Development Services (ICDS), which seeks to provide growth monitoring, nutritional I I Poorstates IOther statesI I None I Pnmary I Secondary I Hagher I supplementation, and early Income levelofstate Mother's education level childhood education to preschool children. Since the large portion o f ICDS beneficiaries are aged 3-6 years, the program is unlikely to influence infant mortality rates, except perhaps very indirectly. While the NFHS-2 did not obtain information on whether an ICDS anganwadi center operated in a respondent's village o f residence, the NFHS- 1 (conducted in 1992-93) did. Analysis o f unit record data from the NFHS-1 suggests that the presence o f an ICDS anganwadi center in a village i s associated with a substantial - 17% - decline in under- 24 five mortality rates (Figure 11.28). This association holds across all levels o f maternal schooling, although it i s observed only inthe poor (and not inthe nonpoor) states2' I.MultivariateAnalysisofInfantMortality 2.53 In order to undertake further simulations about the likelihood of India and its individual states meeting the IMR MDG, we have estimated a multivariate model o f infant mortality usingthe NFHS-2 unit record data (at the child The multivariate model has the advantage o f controlling for several variables that may be simultaneously associated with infant mortality. The estimation results are reported in Annex Table 1, while only the broad findings o f the empirical analysis are discussed here.29 2.54 Even after controlling for other factors associated with infant mortality, scheduled castes (SCs) and scheduled tribes (STs) have significantly higher infant mortality than other backward castes (OBC) and forward castes. The differences are large; for instance, controlling for other factors, SC/STs have, on average, an infant mortality rate that i s 50% higher that o f forward castes. This suggests the importance o f targeting socially disadvantaged groups; since a number o f STs live in ST habitations, these habitations could be targeted for health interventions. 2.55 Within households, higher birtkorder females face significantly higher mortality risk than lower birtkorder males and females. The higher mortality o f higher- order females typically arises due to their neglect by parents, typically inthe provision o f nutrition and medical care. These results indicate the im ortance o f targeting health interventions to older (higher order) girls within households. 8 2.56 The lack o f access to toilets has a significant association with increased infant mortality. Households without any access to toilets experience 16% higher infant mortality than those with some type o f toilet access. Surprisingly, piped water access i s observed to have no significant (independent) association with the probability o f an infant 27 O f course, the inverse association between the presence o f ICDS angunwadi centers in a village and child mortality rates may reflect the fact that anganwadi centers, at least at the time o f the NFHS-1 survey (1992- 93), were located in more affluent communities that happened to have lower child malnutrition and mortality rates. However, this possibility appears unlikely, given the design o f the program. ** Since the dependent variable is dichotomous (viz., whether or not a child dies within 12 months o f its birth),the model has been estimated by the maximumlikelihood probit method. 29 In the probit model, we have included an explanatory variable - predicted household consumption expenditure per capita - to proxy household living standards. The NFHS-2 is a rich data set, but it has the limitation that it does not contain informatbn on income or expenditure, both o f which are widely used as measures o f household welfare. Using data on household energy sources, land ownership, and household demographic characteristics (all o f which are available in the NFHS2), we have predicted monthly consumption expenditure per capita for each o f the NFHS-2 households on the basis o f an econometric relationship between actual monthly consumption expenditure and energy sources, land assets and demographic variables that was estimated with unit record data from the National Sample Survey (NSS) 55`h round data (1999-2000). The distribution o f predicted monthly consumption expenditure per capita in the NFHS-2 sample was observed to be very similar to that in the NSS 55thround sample. 30 Rahman et al. (2003) also find that parity and birth spacing have strong associations with child mortality in West Bengal, Bihar, Uttar Pradesh and Rajasthan. This means that family planning interventions that result in fewer and better-spaced births will also reduce child mortality. 25 death, probably reflectingthe fact that it i s highly correlated with toilet access.31Another infrastructural variable - the availability o f electricity - i s associated with the reduction o f infant mortality by 22 deaths per 1,000 live births (about 34%). Interestingly, however, irregularity o f electricity supply in the village reverses the inverse association by 10 infant deaths (14%), indicating that while the availability o f electricity i s associated with a reduction in infant mortality, the association is much stronger for regular electricity supply. Yet another infiastructural variable - the percentage o f villages in a district that are connected by a pucca (sealed) road - also i s significantly associated with infant mortality decline. A one-percentage point increase in this variable is associated with a reductionin infant mortality o f about 0.2%. 2.57 Aninteresting - andunusual- finding fiom the analysis is that household living standards, as proxied by the log o f household consumption expenditure per capita, have a perverse positive association with infant mortality, once there is control for schooling and other characteristics o f the household.32Likewise, log o f per capita state gross domestic product, used as a proxy for the level o f development in a state, also has a perverse positive association with infant mortality when there i s control for infrastructure variables and public spending on health. It is difficult to explain these counter-intuitive results, but one possibility i s that the typically-observed inverse association between income and infant mortality in fact reflects the association between infant mortality and parental schooling, infrastructure (e.g., electricity availability), and housing quality. Once there i s control for these variables, the association between infant mortality and income or living standards becomes insignificant or even negative. 2.58 Public spending on health and family welfare per capita i s observed to lave a significant inverse association with infant mortality, with the negative association being slightly weaker among the poor states.33 2.59 Tetanus immunization and antenatal care coverage also have sizeable associations with reduced infant mortality.34 The risk o f infant death i s lower by 28% 3 1A study by Jalan and Ravallion (2003) that attempted to estimate the impact o f piped water access on diarrhea prevalence and duration among children under five in rural India, using a propensity score- matching methodology, found that while there were significant health gains overall from access to piped water, there appeared to be no gains for the poorest 40% o f children. 32 Following convention, the consumption expenditure, state government health expenditure and per capita gross state domestic product variables are transformed to natural logs before including them in the probit model. However, the use o f the original variables, as opposed to the transformed natural logs, does not change the basic results reported here. 33 A s an alternative, h e probit equation was estimated with a set o f 23 state dummy variables, which replaced all the state-level variables (e.g., log o f gross state domestic product per capita, log o f state govemment health expenditure per capita, and the latter variable interacted with a dummy variable for the poor states.) Although the full set o f state dummy variables was significant at the 5% level, the explanatory power o f the regression, as measured by a pseudo R-squared measure, only increased from 0.07 to 0.08 with the substitution o f the state fixed effects for the three state-level variables. This suggests that the state fixed-effects model is not a superior model to the one reported here in terms o f goodness-of-fit. 34 Tetanus immunization and antenatal care are generally household choice variables and therefore inappropriate candidates to be explanatory variables in an infant mortality regression. However, in the Indian context, both o f these variables are heavily (ifnot exclusively) influenced by availability (or supply) considerations. The most important reason why a large number o f pregnant women do not obtain tetanus 26 when a mother has been immunized against tetanus and by 15% when a mother has received some form o f antenatal care during her pregnancy. After controlling for tetanus immunization and antenatal care, professional medical attendance at the time o f delivery does not appear to be significantly associated with infant mortality. 35 2.60 As inother studies from around the world, maternal schooling is observed to be significantly and inversely associated with infant mortality. The estimates suggest that each additional schooling year o f a mother i s associated with a reduction in infant mortality o f about 4%. Maternal schooling has a much stronger association with infant mortality reduction than schooling o f the household head.36 2.61 There have been relatively few studies o f the correlates o f infant mortality in India, especially using recent unit record data. A study by Hughes and Dunleavy (2000) using the NFHS-1 data (for 1992-93) focused largely on the association between household environmental variables, especially the use o f clean cooking fuels, and infant mortality. The study observed a significant association betweeninfant mortality and the use o f kerosene or bottled natural gas (as opposed to cow dung or wood) for cooking. Many o f Hughes and Dunleavy's other results are similar to those reported here. Bhargava (2003) also uses the earlier NFHS-1 data, but only for Uttar Pradesh. His results too are broadly comparable to those reported here using the NFHS-2 data for all o f India. Bhargava focuses on the differential association between the presence o f older boys and girls in the household and the mortality risk of children, and finds that the presence o f an older girl i s associated with a larger reduction in infant mortality o f a child. Finally, Arulampalam and Bhalotra (2002) use the more recent NFHS-2 data, but only for three states (Uttar Pradesh, West Bengal and Kerala), and focus on the association between the death o f a previous child in a household and the mortality risk faced by the index child. They find that the death o f the previous child i s associated with a significantly enhanced risk of mortality for a subsequent child. None o f the three studies includes as wide a range o f policy variables as that included in this analysis; mr do they undertake any simulations o f the likely impact o f different policy instruments on mortality outcomes. shots or any form of antenatal care is that such care is simply not readily available in most villages. W e therefore implicitly treat both o f the variables as "access" - not utilization - variables. 3 5The lack o f significance o f this variable does not necessarily mean that medical attendance at birth is not important. The result most likely reflects the fact that antenatal care and medical attendance birth are strongly correlated with each other, resulting inthe influence o f medical attendance at birthbeing picked up by the antenatal care variable. 36 Note that the association between infant mortality and maternal schooling implied by the estimated equation may not reflect causality, since there is no control for unobserved endowments in the relationship. For instance, Behrman and Rosenzweig (2002) find that a positive significant association o f mother's schooling on child schooling in cross-sectional estimates becomes significantly negative if data on identical twins are used to control for unobserved genetic and family background endowments, perhaps because more-schooled women - holding constant ability - spend more time in the labor force and less in child care. The available data do not permit extensive exploration o f such possibilities. 27 J. Simulationsto 2015 2.62 Based on the multivariate probit model estimated above, we have undertaken simulations o f the infant mortality rate for the poor and the nonpoor states under different intervention scenarios.37 Since the explanatory power of the estimated probit model i s low, the simulations discussed below should be treated as indicative o f possible trends inthe future - not as definitive prediction^.^^ The simulation results are shown in Table II.4.39Ifthe poor states were simply brought up to the national averages interms of coverage o f sanitation, road access, electricity, antenatal care, tetanus immunization, female schooling, and public spending on health and family welfare per capita, the cumulative reduction in the infant mortality rate would be o f the order o f about 11.8 infant deaths per 1,000 live births (or 16%). If the magnitude o f the proposed interventions were scaled up so as to bring the poor states to the mean level o f the nom poor states, the cumulative reduction inthe infant mortality rate would be greater - about 36.5 infant deaths per 1,000 live births (or 48%). It i s thus clear that a combination of interventions can have a strong association with infant mortality reduction in the poor states. Table 11.4: Projected decline in the infant mortality rate with various interventions in the poor and non-poor states Bringing thepoorhon-poor states to the level ofi Type of Intervention Non-poor National states` 50% 100% average average coverage coverage No access to toilet (%) -0.8 -2.4 -3.5 -10.1 (-1.1) (-7.7) Access to regular electricity supply (%) -1.2 -2.5 -4.8 -15.7 (-2.4) (-13.3) Access to irregular electricity supply -1.o -3.6 -4.4 -10.5 (%) (-0.8) (-6.91 Female schooling (years) -1.9 -5.8 -10.6a -14.3 (-4.8)a (-8.5)b Government expenditure on health and -3.2 -10.9 family welfare per capita (Rs.) Percent o f villages in district connected by apucca road (%) -1.5 -4.7 -9.6 (-4.9) 37 A s noted in chapter 1, the simulations undertaken throughout this report are based on inferences o f one way causality from the independent variables to the dependent variables (viz., infant mortality, child malnutrition, school enrollment, etc.). 38 The pseudo Rsquared measure for the estimated probit model is 0.07, suggesting that the included explanatory variables `explain' only 7% o f the cross-sectional variation in infant mortality. It is not uncommon, however, for econometric models estimated with cross-sectional data to have l o w explanatory power, in large part due to the importance o f unobserved individual and household heterogeneity and idiosyncratic `shocks' in determining household outcomes. The explanatory power o f regression models typically declines with sample size. 39 The simulations were done by multiplying desired changes in the values o f the right-hand side variables to the marginal effects reported in Annex Table 1. The resulting changes in the probability o f an infant death, when multiplied by 1,000, can be interpreted as the projected declines in the infant mortality rate from the base year. 28 Table 11.4: Projected decline inthe infant mortality rate with various interventions in the poor and non-poor states Bringing thepoorhon-poor states to the level of Type of Intervention Non-poor National states' 50% 100% average average coverage coverage Coverage o f ietanus toxoid -1.1 -3.4 -5.6 immunization for mother (%) (-2.2) Antenatal care coverage (%) -1.1 -3.2 -4.5 (-1.3) Notes: Figures inparentheses refer to the mortality reductionobtained inthe non-poor states. 'The assumed coverage levelinthis case is 6.5 years o f schooling. bTheassumed coverage level inthis case is 8 years o fschooling. Empty cells indicate no significant reductionininfant mortality duetothe specific interventionbeing considered. 2.63 Figure 11-30shows the projected decline in the infant mortality rate in the poor states with all the seven interventions noted above being pursued simultaneously and gradually to 2015. The scope and magnitude o f the assumed interventions, which are shown inTable 11.5, are merely meant to illustrate the likely reduction ininfant mortality under one possible scenario. There is obviously no suggestion that the assumed interventions will indeed take place, and, even if they do, whether the interventions will proceed at the pace assumed inTable 11.5. 2.64 While each o f the interventions i s observed to contribute to the reduction in infant mortality, it is obvious from the figure that two interventions - additional public spending on health and family welfare and increased schooling o f mothers - are most strongly associated with infant mortality reduction. With the full 'package' o f interventions, the nfant mortality rate in the poor states is projected to decline by 46 infant deaths per 1,000 live births by 2015 - to a level o f 29 infant deaths per 1,000 live births. This would bejust below the MD goal o f 31 deaths per 1,000 live births for the poor states. This suggests that while attaining the infant mortality MDG will be challenging in the poor states o f the country, it will not be impossible, as long several mortality-reducing interventions are pursued simultaneously. Table 11.5: Assumptions about various interventions to reduce the infant mortality rate inthe poor states, 1998-99 to 2015 Starting Assumed Ending value in change per value in Intervention 1998-99 vear 2015 Female schooling (years) 2.7 0.3 7.8 Per capita government expenditure on health and family welfare (Rs.) 95 4% 185 Population coverage o f regular electricity supply (%) 27.7 1%point 44.7 Population coverage o f tetanus toxoid immunization for pregnant women (YO) 70.0 1% point 87.0 Coverage o f antenatal care (%) 55.5 1% point 72.5 29 Table 11.5: Assumptions about various interventions to reduce the infant mortalitv rate inthe Door states. 1998-99 to 2015 Starting Assumed Ending value in change per value in Intervention 1998-99 year 2015 Village access topucca roads(%) 59.5 ]%point 76.5 Populationwith no accessto toilets (%) 76.5 -2% points 42.5 2.65 The finding on the FigureII. Projecteddecline inthe infant mortality rateinthe poor 29: statesunderdifferentinterventionscenarios,1998-2015 positive association (graphshows cumulative effectof each additionalintervention) between increased govem- s c as mWo z- mz mo *o 2 8 S ~ % " = " " z Y o o o o o e o o o o o e e o o " " " " " " N ment health expenditure O 0 and infant mortality needs -' 5 two qualifications. First, -10 -10 merely increasing health -1s - I S spending will not be -20 -20 enough; the composition, -25 -2s -30 -."Increasingreal Expandingtetanuscoverage govt healthexp. per capita -30 quality 9 and effectiveness .35 -3s o f public speding i s as im- -40 -40 portant as raising its quan -45 -45 tity. This is especially true -50 -50 o f the poorer states in India, which are plagued with the most serious problems o f governance and service delivery in the health (and other) sectors. Serious attention needs to be paid to making government health services work in these states, especially for the poor. As the World Development Report 2004 points out, this i s a complex and difficult task that entails creation o f the right institutions and incentives in the system to improve service delivery, such as devolving responsibility for service delivery to local governments and communities (see Box 11.2), contracting out certain types o f service delivery to the nongovernment sector, and empowering consumers to demand better services from government health facilities. 2.66 Second, the finding that public spending on health is associated with infant mortality reduction inthe poor states i s o f little use to policy makers. Identifying the type o f health interventions on which to spend resources in the poor states i s o f much greater policy interest. The evidence in this report, as well as the results o f previous studies, suggest that a package consisting o f expanded child and maternal immunization, antenatal care coverage, nutritional supplementation (including promotion o f exclusive breast-feeding), and home-based neonatal services (including treatment o f pneumonia) (see Box 11.1below) is likely to be a high-impact intervention strategy (Jones et al. 2003). 30 Box 11.1: Home-Based Neonatal Care: Results from a Field Trial inRural Maharashtra Nearly twethirds o f infant deaths inIndia occur inthe first month o f birth. Thus, substantial reductions in infant mortality can take place only with a reduction in neonatal mortality. It i s believed that the large majority o f neonates in India die due to sepsis (typically, septicaemia, meningitis, and pneumonia). While the most efficient way to treat illneonates i s to admit them to hospitals, specialized hospital care i s either inaccessible to the lural population or prohibitively expensive. Itwas inresponse to this need that Bang and his colleagues developed a package ofhome- based neonatal care, including the management o f sepsis, and tested it ina three-year field trial from 1995 to 1998 in 39 villages inGadchiroli district inMaharasthra. Gadchiroli is a `backward' district inwhich rice cultivationandforestry are the main sources of income andinfrastructure (viz., roads, communications, school facilities, and public health facilities) i s poor. The team identified 39 intervention villages, and introduced neonatal care to these villages in a sequential manner from April 1995 to March 1998. Inthe first year, female village health workers listedpregnant women in the village, collected data via home visits inthe third trimester, observed labor and neonates at birth, and visited the home frequently for 28 days after birth to weigh the child and observe neonatal morbidity. Data from the first year were used to plan Wher interventions. Inthe second year o f the study, the female health workers were trained in and began providing home-based management o f neonatal illnesses. The care typically consisted o f clinical diagnosis o f sepsis and subsequent treatment with an injection o f Gentamicin and Co-Trimoxazole syrup. In the third year o f the intervention, health education o f mothers and grandmothers about the care o f pregnant women and neonates was added to the program. By the third year o f the intervention, 93% o f neonates in the treatment area were receiving home-basedcare. The team had selected 47 control villages inan adjacent area o f the same district, and had collected baseline data inboth treatment and control villages. An evaluation done after the thirdyear o f the intervention indicated net reductions o f 62%, 46% and 71% in neonatal, infant and perinatal mortality rates inthe intervention area as compared to the control area. Inabsolute terms, the infant mortality rate inthe treatment area fell from 75.5 deaths per 1,000 live births in the baseline period (1993-95) to 38.8 in 1997-98, while it merely declined from 77.1 to 74.9 inthe control villages over the same period. Case fatality in neonatal sepsis declined from 16.6% before the intervention to 2.8% after treatment by the village health workers. The cost of home-based neonatal care worked out to about US$5.30 per neonate, o f which $3.80 was the recurrent cost. The Bang experiment shows that it is possible to halve the infant mortality rate in populations with poor economic and nutritional status and with low female literacy by providing inexpensive health education andhome-basedneonatal care. Source: Banget al. (1999). 31 Box 11.2: I s Decentralization of Health Services Associated with Child Mortality? Well before 1992, the states of Maharashtra andGujarat hadreassignednumerous activities, including the administration o f primary health care, to elected rural bodies at the district level (the zilZa parishads). These bodies were also provided with the flexibility to recruit certain classes o f employees (typically, low level). Even though higher-level employees were state employees on deputation to the local bodies, they were administratively under the control o f the ziZZa parishad. Members o f Parliament and state legislatures were kept out o f the membership o f the local bodies. Finally, funds for administering health care activities were directly transferredto these bodies. Karnataka was another state that introduced decentralization in the mid-1990s. As in Maharashtra and Gujarat, entire departments (including primary education and health care) and associated expenditures were transferred to local bodies at the village and district levels. In Karnataka, unlike as in Maharashtra and Gujarat, planning departments were also m v e d under the control o f the local bodies. A state finance commission oversaw the division o f resources at the state level between the state govemment andthe local bodies. West Bengal was yet another state which made some progress toward decentralized governance before 1992. West Bengal village panchayats were actively involved in executing a number of programs such as mass literacy campaigns, irrigation schemes, and employment generation. The passage o f the historic 73rd and 74th constitutional amendments - viz., the local government or Panchayati Raj Act o f 1992 - by the Indian Parliament gave fresh impetus to the decentralization efforts o f various states. The Act gave control to elected village and urban councils (panchayati raj institutions or PRIs) over a wide range o f social and developmental activities o f governments, including education, health care, nutrition, and safe drinking water and sanitation. According to the Amendment, PRI members were to be elected (with guaranteed representation o f women and minorities), and PRIs were to be funded by `block grants' from the state and central governmental budgets as well as from local taxes which they have the authority to levy. While all states have passed conformity legislation in line with the Constitutional changes o f 1992, and most have held elections and set up state fmance commissions for devising a framework for devolving funds to local bodies, some states, such as Kerala and Madhya Pradesh, have gone further and moved entire departments to the control of the local rural bodies and involved them in the development planning process. Kerala has moved toward making large transfers o f funds to local bodies, while panchayats in Madhya Pradesh have the right to demand funds for schools and hire their own teachers. An econometric study o f over 1,500 villages, based on a nationwide survey of 33,230 households in 1994, found that after controlling for mean household income, income inequality, access to transport and availability o f health infrastructure, under-five mortality rates in villages were significantly lower in states where health services had been significantly decentralized (viz., Gujarat, Maharashtra, Kerala, Kamataka and West Bengal) relative to states where there had been no such move (Mahal et al. 2001). Source:Mahalet al. (2001). 32 Annex 11.1:Infant Mortality, Government Health Expenditure, and Per Capita Income across States, 1980-99 2.67 Several studies have tried to estimate the relationship between health outcomes and public spending on health using cross-country data (Anand and Ravallion 1993, Filmer and Pritchett 1999, Gupta et al. 1999, Bidani and Ravallion 1997, Rajkumar and Swaroop 2002). These studies have reported mixed results. Filmer and Pritchett find insignificant association between public spending and infant mortality. However, Gupta et al. find public spending on health to have strong associations with infant and child mortality reduction. As Gupta et al. disaggregate mortality by poverty status, they are able to estimate the differential association o f public spending with the mortality rates o f the poor and the nonpoor. They find that the association between child deaths and public spending on health is twice as strong for the poor as for the nonpoor. Anand and Ravallion and Bidani and Ravallion also obtain a strong inverse association between infant mortality and public expenditure on health. Rajkumar and Swaroop interacts public spending on health with a governance index and finds that public spending on health i s significantly associated with child and infant mortality reduction, but only for countries with good governance (as measured by a corruption index and a variable reflecting the 'quality o f the bureaucracy'). 2.68 Virtually all o f these studies that have explored the association between health outcomes and public spending have used cross-country data (either for a single year or over several periods); we are not aware o f any study that has used sub-national data to analyze this relationship. Cross-country analyses are plagued by issues o f comparability o f expenditure data across countries; for instance, there are large variations in what constitutes health expenditures and in the level o f government - local, state or federal - that i s responsible for public spending on health. In contrast, there is greater comparability inpublic spending on health across states within a single country. 2.69 We have used twenty years o f state-level data on infant mortality rates (IMRs), real gross state domestic product per capita (GSDP), and real public spending on health to analyze the association between infant mortality on the one hand and public spending and economic growth on the ~ther.~';~ 40 Data on government health expenditure were obtained f o m detailed budget demand documents o f individual states. Expenditure on health includes spending on public health; urban and rural health services; medical education, training and research; general administration; and family welfare (i.e., population programs, including maternal and child health). It includes expenditure incurred by a state out o f its own revenues as well as central government health allocations to that state. 4 1A state-level regression o f infant mortality on public health spending is predicated on the assumption that public spending is not endogenously distributed across states (Rosenzweig and Wolpin 1986). Such an assumption may not always hold, especially if government health expenditures are allocated across states based on health conditions. However, the inclusion o f state fixed effects in the regression effectively controls for endogenous program placement. W e allow for a central planner to distribute public spending on health across states according to an unobserved health attribute; in other words, states that have innately high mortality may spend more on health than those having innately l o w mortality. Since the fixed-effects 33 2.70 The results o f this analysis are reported in Annex Table 11.1 below. OLS estimates (shown in columns 1 and 2) indicate a very strong association between infant mortality and government health expenditure even after control for per capita GSDP and a time trend, with a one percent increase in government health spending per capita being associated with a one percent decline in infant mortality. However, when there i s control for unobserved heterogeneity across states - in the form o f time-invariant state fixed effects - in the model (columns 35), the magnitude o f the association declines sharply and, inthe case o f the regression with a time trend (column 5), becomes insignificant. 2.71 Unfortunately, time-series data on other covariates o f infant mortality, such as adult female literacy, are simply not available at the state level for the period 1980-99. To rule out the possibility that the association between infant mortality and government health expenditure per capita could be due to the omission o f female literacy from the model, we have calculated state-level averages o f the adult female literacy rate from four rounds o f the National Sample Survey - for 1983-84, 1987-88, 1993-94 and 1999-2000. The inclusion o f adult female literacy means that the infant mortality regression can only be estimated for these four years, which reduces the sample size from 278 observations to merely 56. 2.72 To explore whether the association between infant mortality and public spending varies across different levels o f per capita income, we have interacted GSDP per capita with government health expenditure as well as with adult female literacy inthe infant mortality regression. The results (shown in columns 7 and 8 o f Annex Table 11.1) indicate that the inverse association between infant mortality and per capita public spending becomes weaker - while that between infant mortality and adult female literacy becomes stronger - at higher levels o f per capita GSDP. Indeed, the estimated coefficient on government health expenditure per capita is negative only at very low levels o f per capita GSDP (about Rs. 6,050 in 1993-94 prices). While two-thirds o f the states fell under this threshold for the earliest year o f the sample (1980), fewer than 10% o f the states do so for the latest year (1999). 2.73 On the other hand, the coefficient on adult female literacy in the estimated infant mortality equation becomes negative only beyond a threshold per capita GSDP level o f Rs. 10,500 (again 1993-94 prices). This means that while government health expenditure per capita is associated with infant mortality for very poor states, adult female literacy i s not. On the other hand, at higher levels o f per capita income, there i s an association o f infant mortality with adult female literacy but not with public spending on health. 2.74 Finally, we have attempted to estimate a nonparametric relationship among the three variables for which continuous time-series data are available over the period 1980- 99 (viz., infant mortality, government health expenditure per capita, and GSDP per capita). The nonparametric regression allows for complete flexibility in the estimated relationship and includes single-year dummy variables and state fixed effects. It also modelpurges the unobserved(time-invariant) health attribute from the model, it effectively addresses the endogenousprogramplacementproblem. 34 allows us to calculate the 95% confidence interval o f the estimated elasticity at varying levels o f per capita GSDP. The implied elasticities o f infant mortality with respect b government health expenditure per capita are shown in Annex Figure 11.1. The results indicate that (i) the association between infant Annex Fig11.1:Estimatedelasticityofinfantmortalitywith respect to res1 g w ' t health mortality and public expenditureacross states, 1980-99, by levelofstate GDP per capita(nonparameterkernel estimateswith control for state fired eflects snd annuslyear dummies) spending on health per 8 0 8.2 8.4 8.6 8.8 9.0 9.2 9.4 9.6 9.8 10.0 capita is significant and 0.W negative at all levels o f per capita GSDP income, and (ii) associationweakens the considerably (even though remaining significantly different from zero) at higher levels o f state GDP $'$% B upper bound of 95% confidenceinterval per capita. /% Esumated slasucity v lower bound o f 95% confidenceintewd 2.75 Thus, there i s 4.40 J Naturallog ofreal per capita GSDP certainly some evidence o f an inverse association between infant mortality and public spending on health, but it i s not very robust to altemtive specifications. The most general econometric specifications seem to suggest that public spending on health matters, but only for the very poorest states. At the same time, the finding that adult female literacy does not matter for these states-but does for the better-off states - i s troubling. 2.76 There i s some evidence inthe existing literature to indicate that public spending on health matters more to health outcomes for the poor than the nonpoor. For instance, Bidani and Ravallion (1997) consistently find an impact o f public spending for the poor but not for the nonpoor. Likewise, Gupta et al. (1999) find a stronger inverse association between child deaths and public spendingfor the poor than for the nonpoor. The finding that the inverse association between infant mortality and real government health expenditure per capita i s stronger for poor states relative to nonpoor states may seem counter- intuitive, given the large observed inefficiencies and leakages in government health spending in the poor (and poorly-governed) states o f India. However, the association between infant mortality and public spending may be stronger for the poor than the nonpoor states for two reasons. First, the better-off states with low infant mortality have a larger proportion o f ifant deaths occurring during the neonatal period. On the other hand, the typical (and relatively inexpensive) child survival interventions have a greater impact on bringing down deaths in the post-neonatal than in the neonatal period. Consequently, it i s much easier to bring down infant mortality rates in the poor, high-mortality states with proven, low-cost, post-neonatal interventions, such as child immunizations and oral rehydration therapy. In the better-off (and low-mortality) states, however, neonatal deaths constitute a larger proportion o f total infant deaths. Reduction o f neo-natal deaths typically requires more expensive interventions, such as 35 professionally-attended deliveries or deliveries in institutions as well as post-delivery and emergency hospital-based care.42 2.77 Second, the smaller health impact o f government health expenditure in the better-off states may reflect the increasing role o f the private sector in providing health inputs at higher income levels. Inthe poor states, not only is the private sector in health more limited in scope, but complementary household inputs into health production, such as nutrition and good hygiene, are also typically lacking. As a result, the public sector and public spending on health assume far more important roles in determining health outcomes than in the nonpoor states. However, as per capita incomes in a state increase, nutrition and hygiene improve, as does the supply o f private health providers, all of which diminishthe relative importance o f government health expenditure. 2.78 However, these results should not be construed as a license to simply increase government health expenditure in the poor states, because they do not consider the counter-factual - viz., the improvements in health outcomes that would occur if the quality o f government health expenditures were simultaneously improved via greater accountability in service delivery and better governance. All that these results indicate i s the need for a state-stratified mortality-reduction approach, with the appropriate infant mortality-reducing interventions, such as expanded immunization, antenatal care, nutritional supplementation (including promotion o f exclusive breast-feeding), and home- based neonatal services, being targeted to the poor states via an effective and accountability-based delivery system (see Jones et al. 2003). 42Although neo-natal mortality reduction typically requires hospital-based care, it i s possible to provide a relatively inexpensive package o f home-based neonatal services, as shown by a highly-successful field trial inMaharashtra in 1995-98(see Box 11.1for a detailed description o fthe intervention). I 37 3. CHILD MALNUTRITION 3.1 Reduction in child malnutrition is another MDG related to an improvement in child we1fa1-e.~~Child malnutrition significantly increases the risk o f infant and child death, with some estimates suggesting that child malnutrition i s responsible for half or more o f child deaths in the developing world.44 The NFHS-2 data analyzed in the previous chapter showed a strong relationship between under-five child mortality rates and child underweight rates across the various regions o f India (see Figure 11.22 in Chapter 2, Section G). There i s also a large body o f evidence from around the world relating under-nutition in childhood to lower levels o f school performance, cognitive development, health, and, ultimately, to lower levels o f labor productivity in adulthood. Thus, the economic and human costs of child malnutrition in India are likely to be very high.45 3.2 The millennium development goal i s to reduce the percentage o f underweight children by one-half between 1990 and 2015.46 For India, this would imply a reduction inthe childunderweight rate from 54.8% in 1990to 27.4% in2015.47 A. Patterns and Trends 3.3 Levels. Childmal-nutrition rates inIndia are extraordinarily high. The NFHS-2, which i s the most recent household survey containing information on child nutrition, indicates that Early one-half o f children aged 0-35 months are underweight or stunted,48 which translates into ?proximately 37 million malnourished children. About 18-23 percent o f children are severely underweight or stunted in the sense o f being more than three standard deviations below the relevant NCHS standards. This suggests that Indian 43 Another important form of malnutrition that is not pursued in this report i s inadequate consumption of micronutrients, such as Vitamin A, iron and iodine. 44 For instance, based on worldwide evidence, Pelletier and Frongillo (2002) estimate that a 5 percentage point reduction in the prevalence of low weight-for-age could reduce child mortality by about 30% and under-5 mortality by 13%. 45 The World Bank (1998) suggests that the cost o fundernutritionin India i s at least US$10 billion annually interms of lost productivity, morbidity andmortality. 46 While the nutrition MDG i s based on the weight-for-age indicator, it should be recognized that there are other important indicators of child malnutrition. Weight-for-age i s an indicator ofboth short- and long-term malnutrition. Height-for-age or stunting i s a better indicator o f long-term (cumulative) malnutrition, while weight-for-height or wasting is generally considered the appropriate indicator for tracking short-term fluctuations innutritional status. 47 The most reliable estimate is available from the NFHS-I for 1992-93. The rate shown for 1990 is projected from the change observedbetween 1992-93 and 1998-99. 48 As in the literature, a child is considered underweight when his or her weight-for-age i s more than two standard deviations below the NCHS reference weight. A child i s stunted when his or her height-for-age is more than two standard deviations below the NCHS reference. Severe underweight and stunting occur when the relevant nutrition indicator is more than three standarddeviations below the NCHS reference. 38 children suffer from short-term, acute food deficits (as reflected in low weight-for-age) as well as from longer-term, chronic under-nutrition (as manifested in high rates o f stunting).49 3.4 Trends. Both the Figure111.1: Percentof childrenaged 0-35 monthswho are Underweight,1998. NFHS-1, conducted in 1992-93, 99, and annual%change in this rate between 1992-93 and 1998-99, by state and the NFHS-2, conducted in =Child underweight rate (%), 1998-99 -Annual rate of change (%), 1992-98 1998-99, obtained information on child anthropometry. A E 50 comparison o f the estimates o f c9 45 these two surveys indicates a 2E 40 modest decline o f about 11% *E 35 (from a rate o f 52.7% to 47%) 30 2 25 during the 6-year period - s 20 amounting to an annual rate o f decline o f 1.9%. In contrast, underweight rates in neighboring Bangladesh fell from 68% in 1992 to 51% in 2000 - an annual rate o f decline o f 3.6% (World Bank 2003). In Vietnam, the child underweight rate fell from 49% in 1993 to 36% in 1998 - an annual rate o f decline o f 6.1% (World Bank 1999)! India thus appears to be an under-performer inreducing child malnutrition during the 1990s. 3.5 Interstate Variations. An average child underweight Figure III.2: Child underweightrate MDGsby state, 2015 (Yoof children 0-35 monthsunderweight) rate o f 47% masks wide MUDW variations in child malnutrition across states. Child underweight rates vary from a low o f 24-28% in the Northeastern states and Kerala to 51-55% in the states o f Bihar, Rajasthan, Uttar Pradesh, Madhya Pradesh and Orissa (Figure 111.1). Likewise, the decline in child under-weight rates over time has also varied greatly across states. In Punjab, for nstance, the child underweight rate fell at an annual rate o f 7.6% between 1992-93 and 1998-99, while Rajasthan saw an increase o f 2% per annum in the child underweight rate during the same period. Based on 1992-93 and 1998-99 values, Figure 111.2 shows the 2015 MD goals for each o f the states. 49 Wasting rates (Le., low weight-for-height) are significantly lower thanunderweight or stunting rates, but this is typically the case inmost low-incomecountries. 39 3.6 Although the proportion o f children aged Figure III.3: Change in absolutenumber of underweightchildren 0-35 monthsof age between 1992-93 & 19998-99, by state ('000 children) - 0-3 years inIndia who were r- underweight declined from 52.7% 1992-93 to 47% in 1998-99, the absolute number o f underweight children hardly changed over this period because o f a large increase in the population o f children aged 0-3. Indeed, only about a quarter million fewer Indian children aged 0-3 were underweight in 1998- 99 as compared to 1992-93. There were wide variations 2 - in the absolute decline in 1 - the number of underweight 3d E b children across states. .c O - E ** 15 ' 30 35 40 4% 50 b 55 60 65 Tamil Nadu and Andhra P -1 0 Pradesh each saw declines 9 .2- 3L.a U ? - * o r * 4 in the absolute number of .E p -3 underweight children o f 38 0 ' - 4 - ** L about 0.5 million, but . 5 - Rajasthan and Bihar saw an a -6 increase o f 0.5 million each -7- b in the number o f -8 - underweight 0-3 year olds (Figure 111.3). Uttar Pradesh and Madhya Pradesh each recorded an increase o f about 0.2 millionunderweight children aged 0-3 duringthe same period. 3.7 Figure 111.4 suggests that there i s no Figure 1II.S: Child underweightrate ( O h )and gross state domestic systematic association product per capita across states, 1998-99 between the initial level o f - 55 - child malnutrition in 1992- c .9 Maharashtra 93 and the rate o f decline in 50- 0 malnutrition between 1992- j Gujarat 45- e 93 and 1998-99. Nagaland, c ? which had a child 2 40- Andhra Pradesh VI e underweight rate o f only 2c 3 5 - 27.7% in 1992-93, -cI c experienced an annual rate 5::3 0 - Kerala e o f decline o f 2.3% over the following 6 years - the 25 1 I. same rate o f decline 40 experienced by Bihar, which began with a much higher child underweight rate o f 62.6% in1992-93. 3.8 I s child malnutrition related to living standards? The cross-state data suggest an inverse, albeit not perfect, association be-tween the child under-weight rate and gross state domestic product per capita (Figure 111.5). That Kerala emerges as a positive outlier-having a much lower child underweight rate than would be suggested by its per capita income-is no big suprise, but the fact that Gujarat and Maharashtra have significantly higher child underweight rates relative to their per capita income is surprising. On the other hand, Andhra Pradesh, which i s a middle-income state, has a lower child underweight rate than would be predicted by its gross state domestic product per capita. This suggests that cultural and social-not just economic-factors have an important role to play in determining child malnutrition inIndia. B. PublicSpending onNutrition 3.9 Much o f the public spending on child nutrition in India takes place on the Integrated Child Development Services program. This program consists o f anganwadi centers (AWCs) in each village, typically staffed by a village woman with 5-8 years o f schooling and an assistant. The anganwadi worker receives a cash income o f Rs. 1,000 per month to provide growth monitoring, food supplementation, and pre-school education to targeted children aged (16 years in the villag. Although the program covers all the villages in the country, recent surveys from a few states suggest that relatively few (only about 10-30%) children aged 0-6 years in states such as Uttar Pradesh, Madhya Pradesh and Rajasthan regularly attend the A W C in their community (Heywood 2003). This may be because the amount o f food supplementation provided to children i s meager or irregular or both. While the Central government pays for the salaries o f the anganwadi worker and assistant, the individual states are responsible for lifting the food grains from the stocks o f the Food Corporation o f India and Figure 111.6: Governmentexpenditure on ICDS (child nutrition) program (excluding training) per child aged 0-6 years, 1999-2000 (nominal Rs.) paying for the cost o f transporting and distributing these food grains to the AWCs. This i s the component o f the ICDS that i s typically under- funded (World Bank 1998, 2001). 0 100 ZOO 300 400 500 600 700 800 41 3.10 Figure111.6 shows the total amount (excluding training) spent by various 65- states in the country on the ICDS --60- program. The d * amounts are expressed in terms o f spending per child aged 0-6 years, since that is the target group o f the ICDS." Two observations can be made from this figure; first, the amounts spent by most states are low -typically below Rs. 200 per child per annum. Second, there are large inter-state disparities in 10 - * 93 to 1998-99 spending on nutrition. The poor, high-malnutrition a$ 0 states, such as Bhar, Uttar P -55 " 0: i 5 ;O ;5 1; l i 5 l;O I;, 2bO 2i5 Pradesh, Madhya Pradesh e 0 &a .- 3 4 0 - and Rajasthan, spend only *. 1 ') P r . E 5 Rs. 30-50 per child, while .-C I -20 Gujarat, Punjab, and Haryana spend Rs. 90-100. -5 b z Tamil Nadu's expenditure i s about Rs. 170, while -30] -40 spending in the YOchange in gov't exp on the ICDS program (nutrition) per child 0-6 Northeastern states i s above Rs. 500. Figure 111.9: Changes in child underweight rates and in real per capita gross state domestic product across states, 1992-93 to 1998-99 3.11 I s there an association between child underweight rates and per- child spending on the ICDS? Pooled data on 14 states for two years - 1992- * 93 and 1998-99 - suggest an inverse association (Figure 111.7). However, since there i s no control for 10 15 20 25 30 3 5 40 45 % change in realper capita GSDP (1993-94 Rs.) 50 Even though the target group i s 0-6 years, the ICDS has historically focused on children aged 3-6 years. The focus on 3-6 year olds i s a major design flaw o f the ICDS, since malnutrition typically sets in much earlier in childhood. Experience from other countries, as well as from other nutritional interventions in India (e.g., Tamil Nadu Integrated Nutrition Project, discussed in B o x III.l), shown that child has malnutrition can be addressed much better by targeting nutritional supplementation to children in the younger age groups. 42 other variables, such as per capita income, the association does not necessarily indicate a positive effect o f public spending on nutrition. Indeed, there does not appear to be any association between changes in the level o f per-child spending on ICDS and changes in the child underweightrate (Figure 111.8). 3.12 However, the data do suggest an inverse association between changes in the child underweight rate and changes in gross state domestic product per capita (Figure 111.9). States that experienced greater growth in real gross state domestic product per capita between 1992-93 and 1998-99 experienced larger declines in the child underweightrate duringthe same period. C. IntrastateVariations 3.13 Appendix Table 8 presents data on the child underweight rate in 1998-99 for different regions in the country. The child underweight rate ranges from a low o f 11% inthe Hills region of Assam to a highof 60% inChhattisgarh. More than a third of the regions have child mderweight rates o f 50% or greater, confirming the ubiquity o f child malnutrition inthe country. 3.14 What i s even more worrying i s that more than a quarter o f the regions in the country for which underweight data are available experienced an increase in child underweight rates between 1992-93 and 1998-99 (Appendix Table 8). These regions are scattered around the country - in the poor states (Jharkhand, Bihar, Orissa, and Uttar Pradesh) but also in the more prosperous states, such as Gujarat, Maharashtra, and Haryana. D. Concentrationof ChildMalnutrition 3.15 Child malnutrition in India is not as heavily concentrated geographically as, say, infant deaths. For instance, the four states - Uttar Figure 111.10: Contribution of 20 states to the national number of underweight 0-35 montholds, 1998-99 Pradesh, Madhya Pradesh, 100 91 94 90 --- Bihar and Rajasthan - account for 51% o f all infant deaths inthe country, but for 43% o f all underweight children under the age of 3 (Figure 111.10). Nevertheless, in absolute terms, these are large fix-ther, one finds that child ''It i s not possible to obtain reliable district-level estimates o f child malnutrition in India. Regional-level estimates are more reliable, given the sample size o f h e NFHS-2 survey. Data from the NFHS-2 are not available for a few ofthe regions. 44 perception that they are producing insufficient quantities o f milk due to their poor nutrition and heavy workload. Premature introduction o f foods other than breast milk greatly ncreases the risk o f infection in the small infant, and this sets in motion the process o f malnutrition. It also puts the infant at greater risk o f malnutrition, since weaning diets are often inadequate in India. Supplementary feeding begins with a thin gruel o f rice, often heavily diluted with water and with some vegetables or legumes added as a relish depending on season and availability, but generally in very small quantities. The consequent low energy density o f this weaning food leads to a Educed intake o f calories and protein, and is an important cause o f growth faltering during the weaning period, from six months to two years o f age. The NFHS-2 data indicate that nearly 49% o f children aged 0-4 months and 58% o f children aged 0-6 months are not exclusively breastfed (Le., supplementary feeding is introduced), which i s not in line with the recommendations o f WHO and UNICEF that exclusive breastfeeding continue for the first six months o f a child's life. 3.19 Infections. Illness and infection, especially diarrheal infections, are strongly associated with child malnutrition. Infections Educe the ability o f the body to absorb critical nutrients from food, which in turn leads to malnutrition.53 The NFHS-2 data indicate that an average infant begins suffering from diarrheal diseases very early in his or her life; by the age o f 6 months, he or she has already experienced an average o f 2.2 diarrheal episodes, and by the age of 12 months, 5.2 Figure 111.13: Cumulabve numberof diarrheal infectionsexperiencedby infants,by age (months), 1998-99 illness episodes (Figure 6, 111.13). Diarrhea i s even 6 more prevalent among 5 - - Bihar, Madhya Pradesh,Onssa, 5 infants in the poor states o f Bihar, Madhya Pradesh, - Rajasthan,Una1Pradesh All India 4 4 ` Uttar Pradesh, Orissa and --Other States 3 . 3 Rajasthan. The NFHS-2 data also show that children 2 . 2 who have suffered a diarrheal episode are 15% 1 more likely to be o r . 0 underweight than children 0 I 2 3 4 5 6 7 8 9 10 11 12 who have not experienced Age (months) diarrhea.54 3.20 Maternal Weight and Low Birth Weight. For a large number o f Indian children, malnutrition begins very early in life - when they are born with low birth 53 The association reflects two-way causality; while infections lead to malnutrition, malnourished children are more susceptible to infections. 54 It is not just diarrhea that reduces nutrient absorption o f the body; even repeated bouts o f fever (indicating infection) and acute respiratory infections can slow down weight gain and lead to malnutrition. It is unlikely however that a single bout o f diarrhea, fever or cough could lead to severe malnutrition. What is more likely is that the children who report being illduring the two-week reference period are the ones who repeatedly come down with diarrhea, fever or cough. It i s this chronic fever and cough that is associated with a higher risk o f malnutrition. Indeed, malnutrition also increases a child's susceptibility to infections, and so the `causality' can go inthe reverse direction. 45 weight. As Figure 111.14 shows, children born with a weight of less than 2.5 kgs are at significantly greater risk of subsequent malnutritionthan childrenwhose birth weight is above 2.5 kgs. Nationally, about 22% of births classify as low birth weights, with wide variations across states. In Uttar Pradesh and Madhya Pradesh, more than a third of all children weigh less than 2.5 kgs at birth. Even in Kerala, the proportion of low birth weights is as highas 18%. 3.21 Low birth weight in turn is determined by a number of factors, but important among them is maternal nutrition. Malnourished or Figure111.14: Percentof childrenunder3 who are underweight,by mother`s low-weight mothers are a, weight and birth weight of child, 1998-99 more likely to give birth to low-weight babies, which 61 implies that children o f low-weight mothers are 46 47 more likely to be 36 28 malnourishedthan children 26 of heavier mothers. Figure 111.14 shows this trend; nearly three-quarters of I I 35-39 I 40-49 I >=SO C2,500 gmsI 2,5OOA2,999 >=3,000 gms children under 3 born to I mothers whose weight is Mother`sweight (kgs.) Birthweight(pms.) less than 35 kgs. are likely to be mderweight.This ratio drops to just over a quarter for mothersweighing 50 kgs or more. F. Socioeconomic Correlates of ChildMalnutrition 3.22 Living Standards. There i s remarkably little variation in child under-weight rates across economic groups. - - Indeed, the Figure 111.15: Child underweight rates ( O h )by per capita expenditure prevalence of child 6o quintile, 1998-99 malnutrition is virtually t iFemales HMales OPoor states nother states identical across the bottom 55 N four consumptionquintiles, 5o with only the top quintile (representing the richest 45 20% o f individuals) 4o showing significantly lower child underweight rates 35 (Figure 111.15).55 The fact that nearly a third of the top 30 consumption quintile in the 2s country - a group that is Poorest Second Fourth Richest ~~ likely to have good Yer capita expenditure quintile "Consumptionquintilesarebasedonpredictedhouseholdconsumptionexpenditurepercapita,the estimation of which is described in footnote 24 in chapter 2. 46 economic access to food - Figure 111.16: Child underweight rates ( O h )of various social groups in poor and other states, 1998-99 i s malnourished suggests that cultural and social 61Poor States =Other States l A l l States factors have an important 55 role to play in determining child malnutrition in India. 50 This is also consistent with the evidence, cited earlier, 45 that the child underweight rate is relatively high in 40 prosperous states like Gujarat and Maharashtra. 35 Scheduled Tribe ScheduledCaste Other Backward Caste Fawardcaste 3.23 Social Groups. The NFHS-2 data also Figure 111.17: Child (0-3 years) underweight rates (%) by sex and by 60 1 F S mother's schooling, 1998-99 indicate somewhat higher child mderweight rates for scheduled castes and tribes and other backward castes UFemales .Males relative to the forward castes (Figure 111.16). Scheduled castes and other backward castes appear to be at a greater disadvantage relative to the forward castes in the poor states than inthe nonpoor states. None 1-5years 6-8years 9-lo years 11-12years >12years Mother's schooling 3.24 Maternal Schooling. There i s a large literature from around the world that documents the many benefits o f maternal schooling for child outcomes - infant and child mortality, child nutrition, and child schooling. The NFHS-2data show a sharp decline in the incidence o f child malnutrition with mother's schooling (Figure 111.17). Children o f mothers with no schooling are nearly two times more likely to be underweight than children o f mothers with more than 8 years o f schooling. Children o f mothers with 6 8 years o f sclmoling are more than two times more likely Figure111.18: Childunderweightrates(YO),by birth order and sex, 1998-99 Birthorder to be underweight than 4 & above children o f mothers with 55 55 more than 12 years o f schooling. Interestingly, unlike the case o f infant mortality, the gender 45 disparity in child underweight rates does not 4o seem to narrow with mother's schooling. 35 47 3.25 Birth Order. As with infant mortality, the birth order of a child has a significant association with the probability o f him or her being underweight. The disparity across birth order is greater for girls than for boys. For instance, while the underweight rate i s 42% for first-born girls under 3, it is as high as 58% for birth order four or higher girls (Figure 111.18). 3.26 Infrastructure. Given that repeated bouts o f infection, especially gastro- intestinal and diarrheal, are an important reason for child malnutrition in India, access to safe drinking water and sanitation can improve child nutrition by reducing a child's exposure to water- and vector-borne diseases. Figure III.19: Chid Underweight rates(Yo),by infrastructure access, 199b99 Likewise, access to a- 57 BPoor states .Other states electricity can also improve SS- 53 nutritional status by improving the hygiene, cooking and health 45. practices in the household and in the community. Rural roads enable easier Is access to markets and m-- Notavailable I Irregular I Regular None I health workers and thereby Yes Village electricity supply Pipd water access 3.27 The NFHS-2data show significant associationsbetween child underweight rates and the infiastructural variables (Figure 111.19). The child underweight rate i s about a third lower in villages having regular electricity supply than in villages with no electricity. Likewise, access to piped water and some toilet facilities is associated with significantly lower child underweight rates, although piped water access appears to be less important to child malnutrition in the non-poor states. G. Multivariate Analysis of Child Malnutrition 3.28 To examine the likelihood o f the various states in India attaining the child underweight MD goal, we have estimated a multivariate model o f child underweight rates, usingthe NFHS-2 unit record data (at the child The multivariate model has the advantage of controlling for several variables that may be simultaneously associated with child malnutrition. The estimation results are reported inAnnex Table 2, while only the broadfindingsof the empirical analysis are discussed here. 3.29 The multivariate model confirms most o f the bivariate relationships discussed earlier. Older children are observed to have a higher risk o f malnutrition, but at a decreasing rate, such that the risk o f malnutrition peaks at age 24 months. Mother's age reduces the probability o f a child being underweight, but at a decreasing rate. Controlling for other factors, children belonging to scheduled castes, scheduled tribes and other 56 Since the dependent variable in the model is a dichotomous variable (Le., whether or not a child is underweight), the modelhasbeen estimatedby the maximumlikelihoodprobit method. 48 backward castes are significantly more likely to be underweight than children belonging to forward castes. As with infant mortality, girls per se are not at significantly greater risk o f malnutrition than boys, but sex interacts with birth order such that higher birth order girls have a significantly greater risk o f beingunderweight than higher order boys. 3.30 Infrastructure generally has strong associations with child malnutrition. Children in households having no access to a toilet are, on average, 8.6% more likely to be underweight than children in households having access to a toilet. Surprisingly, piped water access has no significant (independent) association with the probability o f malnutrition, probably reflecting the fact that it i s highly correlated with toilet access. As in the case infant mortality, access to electricity is strongly associated with child malnutrition, but the association i s much weaker with irregular electricity supply. Finally, proximity to sealed (pucca) roads i s also associated with a sharp reduction in child underweight rates. 3.3 1 Other variables that are significantly associated with child malnutrition are the mother's schooling, whether there was a medical professional at the time o f the child's birth, log o f predicted monthly consumption expenditure per capita (proxying for the household's income and living standard), and log o f government expenditure on nutritional programs per child aged 0-6 years in the child's state o f residence. The significant association between government nutritional expenditures per child and child malnutrition i s supported by the finding, discussed in Section Ibelow, that the risk o f a child being underweight i s inversely associated with the presence o f an Integrated Child Development Services (ICDS) center in the child's village o f residence. Much o f the government nutritional expenditure in India i s on the ICDS program. The empirical results further indicate that the (inverse) association between child malnutrition and levels o f government nutritional expenditure i s stronger inthe poor states than inthe nom poor states. However, neither public spending onhealth and family welfare per capita nor per capita GDP in the child's state o f residence has a significant association with child underweight rates.57 H. Simulationsto 2015 3.32 Based on the multivariate probit model estimated earlier, we have undertaken simulations o f the child underweight rate for the poor and the nonpoor states under different intervention scenarios. These are shown in Table 111.1. If the poor states were simply brought up to the national average interms o f coverage o f sanitation, road access, electricity, medical attention at time o f delivery, female schooling, household income (consumption) and public spendingon nutritionper child, the cumulative reduction inthe child underweight rate would be o f the order o f about 8 percentage points (or 15%). Ifthe 57 As an altemative, the probit equation was estimated with a set o f 23 state dummy variables, which replaced all the state-level variables (e.g., log o f gross state domestic product per capita, l o g o f state govemment nutrition per child, interaction between the log o f govemment nutrition expenditure per child and a dummy variable for the poor states, and log o f health expenditure per capita). Although the full set o f state dummy variables was significant at the 5% level, the explanatory power o f the regression, as measured by a pseudo R-squared measure, actually declined from 0.172 to 0.170 with the substitution o f the state fixed effects for the four state-level variables. This suggests that the state fixed-effects model is not a superior model to the one reported here interms o f goodness-of-fit. 49 magnitude of the proposed interventions were scaled up so as to bring the poor states to the average level prevailing in the non-poor states, the cumulative reduction in the child underweight rate would be 21 percentage points or 38%. Table 111.1: Projected decline in child underweight rates (percentage points) with various interventions inthe poor and norrpoor states Bringing thepoor states to the level of Non-poor National states' 50% 100% Type oflntervention average average coverage coverage -0.5 -1.4 -2.0 -5.7 No access to toilet (%) (-0.6) (-4.3) -0.4 -0.8 -1.6 -5.3 Access to regular electricity supply (%) (-0.8) (-4.5) Access to irregular electricity supply -0.3 -1.1 -1.3 -3.1 ("h) (-0.2) (-2.0) -1.3 -3.9 -5.8 -8.0 Female schooling (years) (-1.8)a (-4.1)b Monthly consumption expenditure per -1.o -2.9 capita (Rs.) Government expenditure on nutrition -2.9 -6.9 per child aged 0-6 years (Rs.) Percent o f villages in district connected -1.O -2.9 -6.2 by apucca road (%) (-3.3) Medical attention at time o f child's -0.3 -1.1 -0.5 -2.1 delivery (%) (-1.O) Notes: Figures inparentheses refer to the underweightreduction obtained inthe nonpoor states. 'The assumed coverage level inthis case is 6.5 years of schooling. the assumed coverage level inthis case is 8 years o f schooling. Emptycells indicate no substantial reductionininfant mortality due to the specific intervention being considered. 3.33 Below, we use the probit results discussed above to simulate cumulative changes in the child underweight rate in the poor states from 2000 to 2015 based on certain intervention scenarios. The nature and magnitude o f the interventions are detailed in Table 111.2. As noted already in chapter 2 the scope and magnitude o f the assumed interventions are only meant to illustrate the likely reduction in child malnutrition under one possible scenario. It is obviously not possible to predict whether the assumed interventions will indeed take place, and, even if they do, whether they will proceed as the pace assumed in Table 111.2. Table 111.2: Assumptions about various interventions to reduce the child underweightrateinthe poor states, 1998-99 to 2015 Starting Assumed Ending value in change per value in Intervention 1998-99 year 2015 Female schooling (years) 2.7 0.3 7.8 Per child govemment expenditure on nutrition programs (Rs.) 51 4% 98 Consumption expenditure per capita (Rs.) 422 3% 698 50 Table 111.2: Assumptions about various interventions to reduce the childunderweight rateinthe poor states,1998-99 to 2015 Starting Assumed Ending value in changeper value in Intervention 1998-99 year 2015 Population coverageofregular electricity supply e?) 21.7 1%point 44.1 Population coverage of professionally assisteddeliveries (%) 32.3 1.5%points 57.8 Village access topucca roads(%) 59.5 1%point 76.5 Populationwith no accessto toilets (%) 76.5 -2%point 42.5 3.34 Figure 111.20 shows the projected changes in the child underweight rate in the poor states when the seven interventions shown in Table 111.2 are pursued simultaneously. It i s obvious that, while each o f the interventions contributes to the reduction in child malnutrition, the ones that are associated with the largest declines in child malnutrition are increased per-child public spending on nutritional programs, increases in household consumption expenditure per capita, and expansion o f adult female schooling. What is encouraging i s that, together, the seven interventions are associated with a reduction o f 25 percentage points in the child underweight rate in the poor states - enough for them to reach their MD goal o f 27.3% o f children being underweight. This suggests that while attaining the child nutrition MDG will be challenging in the poor states, it i s clearly feasible with a package o f interventions that include economic growth, increased public spending on nutritional programs, improved physical infrastructure (electricity, sanitation access, medical attendance at birth and rural roads), and expansions in female schooling. FigureIII.20: Projecteddeclinein % of children0-3 who are underweightin the Door states. 1998-2015. underdifferentinterventionscenarios 3.35 The caveats noted (graphshowscumulativeeffect ofeachadditionalintervention) in chapter 2 apply here as m h 0 h 0 : h 5 5 5 5 - N 8 s 3 0 "0 S N 0 N N 0 N N N well. The fact that public 0 spending on nutritional 5 programs i s associated with -10 lower rates o f child malnutrition does not mean -15 that increasing government -20 nutritional expenditure i s -25 sufficient. There i s considerable evidence that -30 nutritional programs in -35 India, such as the ICDS, National Mid-Day Meal Program, and the various micronutrient programs, have poor coverage, targeting and implementation. The ICDS, for example, mostly focuses on children aged 36 years, but the consensus among nutritionists i s that it i s critical for direct nutritional interventions to reach 6-24 month olds and pregnant women to prevent 51 malnutrition (World Bank 1998, 2001).58 Further, the ICDS anganwadi center health worker - one per center - is typically over-burdened, as she has to manage pre-school education, supplementary feeding, and outreach activities. Another problem with the ICDS program i s the frequent disruptions in food supplies that take place at the anganwadi center. The responsibility for the food component o f the program lies with the state governments, which typically under-finance this component owing to cost and logistical difficulties. One evaluation o f the ICDS found that disruptions in food distribution were very common at most anganwadi centers, with the average center going without any food rations for 64 days per year (out o f an intended 300 feeding days) (National Institute o f Public Cooperation and Child Development 1992). 3.36 But it is also the case that spending on direct nutritional programs is very low in India; it amounts to only 0.19 percent o f GNP. Incontrast, neighboring Sri Lanka spends about one percent o f GNP on direct nutritional programs (World Bank 1998). A major World Bank report on malnutrition argues that, given the magnitude o f the malnutrition crisis, India should be prepared to spend a minimum o f 0.5 per cent o f GNP on direct nutrition programs - more than double what it currently spends (World Bank 1998). It goes without saying that the scaling-up o f direct nutritional interventions would have to go hand in hand with revamped design, greater devolution, and better implementation o f such interventions. The highly-successful Tamil Nadu Integrated Nutrition Program (TINP), which has been in operation for more than twenty years (and preceded the ICDS) in the state o f Tamil Nadu, offers important lessons for the design of direct nutritional interventions (see Box 111.1). I.ICDSandChildMalnutrition 3.37 Since the Integrated Child Figure111.21: Percent of childrenunder 4 years who are underweight,by sex and presence of ICDSanganwadi center in village, 1992-93 Development Scheme 58 (ICDS) accounts for much RNoanganwadicentermvillage OAnganwadi centerin village o f the public spending on 56 55 nutrition in India, it may be instructive to analyze the 54 impact o f that scheme on child nutritional outcomes. 52 Unfortunately, such an 5o evaluation i s stymied by the lack o f availability o f 48 relevant data and by the fact that, by now, the 46 scheme extends to virtually Males Females Bothsexes every village in the country. As such, there are no `control' villages tht offer a counterfactual - viz., the prevalence o f child malnutrition inthe absence o f the program. 58 There is also some anecdotal evidence that the ICDS has done better (in terms o f addressing child malnutrition) in states such as Tamil Nadu and Rajasthan which have better targeted the program to younger age groups. 52 3.38 However, at the time the NFHS-1 data were collected in 1992-93, ICDS did not fully cover all the villages in the country. Fortunately, the NFHS-1 village/community questionnaire obtained information on whether a sampled village had an ICDS anganwadi center (AWC).59 By merging the household information on child anthropometry and the village information on the existence o f an AWC, it i s possible to examine how the location o f an A W C in a village is associated with child malnutrition levels inthat village. 3.39 The NFHS-1 data indicate that 34.5% o f villages in the NFHS sample had an AWC. Overall, the data show the child underweight rate (for children under 4) to be somewhat lower in the villages having an AWC than in village not having one (51% versus 55%).60 However, upon disaggregating the numbers by sex, it is found that the presence o f an ICDS angawadi center i s associated with a much larger reduction in malnutrition for boys than for girls (Figure 111.21). 3.40 Since it is important to control for other variables, such as household living standards and maternal schooling, in comparing the child underweight rate across A W C and non-AWC villages, we have estimated a multivariate probit model o f child malnutrition with the NFHS-1 data, exactly along the lines o f the model estimated with the 1998-99 NFHS-2 data (and shown in Annex Table 2) but with the addition o f an explanatory variable indicating the presence o f an ICDS anganwadi center in a child's village o f residence. The empirical results, shown in Annex Table 3, confirm the pattern observed in Figure 111.21; the presence o f an ICDS anganwadi center in a village i s associated with a reduction o f about 5% inthe child underweightrate, but this association i s observed only for boys. There i s no significant association o f an ICDS anganwadi center with the premlence o f malnutrition among girls aged 0-3. This surprising finding could reflect that parents tend to selectively bring their boys, but not their girls, for supplementary feeding at the center. Or it could indicate that anganwadi workers or helpers provide a larger allocation o f food to boys than girls. This i s an issue that merits further exploration. s9 This information was not collected in the NFHS-2 data, presumably because ICDS coverage (of villages) was near universal by the time o f that survey. 6o Note that in the NFHS- 1 (1992-93), anthropometric data were obtained for children under the age o f 4 years, while the cut-off was 3 years in the case o f the NFHS-2 (1998-99) data. 53 Box 111.1: The TamilNaduIntegratedNutritionProject The Tamil Nadu Integrated Nutrition Project, which was started in 1980 by the state government o f Tamil Nadu, was one o f the first projects in the world to make large-scale use o f growth monitoring of children aged 436 months old as a means to target the neediest children and monitor their progress. TINP has been hailed as the most successful nutrition intervention program in India (and probably the world) (Berg 1987). It began as an area-targeted (to the rural areas o f six districts having the lowest caloric consumption in the state but subsequently extended to the entire state), age-targeted (concentrating exclusively on children 636 months of age, who accounted for 90 per cent o f the pre-school deaths in the state, and pregnant and lactating women), and need-targeted program. The latter was achieved by monitoring the weights o f all children 636 months old in the project villages, and enrolling only those children whose weight gain over a certain period fell below standard in a 90-day supplementation program that includeddaily feeding and counseling o f mothers. Since the children were on the supplementation program only for the duration of time their weight gain was below standard, the project was basically seen as a short-term intervention that sought to reduce long-term dependence o f beneficiaries on public assistance. To this day, TINP relies heavily on local nutrition workers, working in conjunction with local women's and girls' groups. The groups are taught behavior-change strategies. They learn to promote birth weight recording, regular monthly weighing, and spot feeding, while participating in community assessment, analysis, and problem- solving. TINP links the delivery o f health and nutrition services. Children who do not respond to the nutrition supplementation are provided health services, which include check-ups and referrals, treatment o f diarrhea, deworming, and immunization. These services are also available to pregnant and lactating women. In addition, the program includes intensive counseling o f mothers in nutrition andhygieneeducation. Evaluation studies o f the TINP have indicated significant effects o f the program. Severe malnutrition fell significantly, by 44 percent between 1992 to 1997, although moderate malnutrition was still quite widespread. Beneficiary children were able to maintain their weight advantage for two years or longer after they completed the program, indicating bng-term effects. Costs per beneficiary were lower than for less targeted nutrition programs. Indeed, one study estimated that the annual recurrent costs o f the TINP were less than one-half o f the ICDS program operating in Tamil Nadu, whle it had an impact on severe malnutrition that was two times as much as the ICDS. The cost difference betweenthe TINP and the ICDS arises almost entirely from the fact that the ICDS is a mass feeding program, while the TI" i s highly-selective supplementary feeding program. Source: Dapice (1987), Berg (1987), Chatterjee (1996). 54 4. PRIMARY SCHOOLING 4.1 Universal primary enrollment i s another one o f the MDGs related to child welfare. The millennium development goal i s to ensure that, by 2015, all children are in school, the net primary enrollment ratio i s loo%, and that all the pupils entering grade 1 are retained until grade 5 (typically the last year o f primary school). 4.2 The numerous benefits o f schooling are well known and have been widely discussed in the literature on economic development. Schooling i s one o f the most powerhl instrumeds for reducing poverty, unemployment and inequality; improving health and nutrition; and promoting sustained, human development-led growth. It i s also self-perpetuating across generations, with educated parents much more likely to provide schooling to their children. Both the pecuniary and nonpecuniary returns from schooling have been well-documented in the literature for several countries, including India. FigurelV.1: Grossprimaryenrollment rates, India, 1950-51 to 1999-2000 InA A. Overall Trends 4.3 Levels and Trends. India has made rapid strides in schooling during the last 4-5 decades. The gross primary enrollment rate, which was only 43% in 1950-51, reached 100% by 1990-91, and has fallen slightly since then(FigureIV.1).61 1950-51 1960-61 1970-71 1980-81 1990-91 1995-96 1998-99 1999-2000 4.4 As in many other countries, gross enrollment rates obtained from school administrative data in India end to overstate the actual enrollment rate.62 Household survey data show much smaller net, gross and age-specific enrollments. The most recent source o f nationally-representative data on enrollments i s the 55`h round o f the National Sample Survey (NSS) mdertaken in 1999-2000. These data suggest that only 78% o f children aged 611 years were attending school in 1999-2000. At 52.5%, the net primary ~~ 6 'Note that "primary" schooling refers to grades >5 in this chapter. The term "lower primary" i s sometimes used in India to denote grades 1-5, while "upper primary" refers to grades 68."Elementary" education refers to grades 1-8. 62 Note that age-specific or net primary enrollment rates are not available from the school-based administrative data. 55 enrollment rate was significantly lower.63 There are many reasons for the discrepancy between household surveybased and school administrative records-based enrollment rates, including the fact that household surveys typically obtain information on school attendance during a short reference period (the question to households typically is: "is your child currently attending school?"), while administrative data refer to students actually enrolled in the registers o f the school at the beginning o f the school year. In addition, gross enrollment rates from administrative records are very sensitive to incorrect estimates o f the population of schoogaged Figure IV.2: Grossprimary enrollmentrate,by state, 1999-2000 children. Finally, there are incentives for school Arunachal Xleghalaya \I P administrators and district Maharashtra officials to overstate the Ravmthan number o f enrolled students, since many types .-......_"" o f government education W Bengal _ _ " ' ' " 'i_"'" ""' ' j " '"'~ ", .Idil ""'"" ' I ! expenditure allocations to Mantpur Nagaland HP schools and districts are Kerala often based on the number ..I., 1.1 o f enrolled students.64 Punab B?har 1&Kashmir UGoa P 4.5 Interstate Ea 70 80 w IW 110 I20 I30 140 Variations. An average gross primary enrollment rate o f 95% for the country in 1999-2000 for the country masks wide variations across states. Gross primary enrollment rates vary from a low o f 65% in Uttar Pradesh to a high o f 139% in Sikkim (Figure IV.2). Some o f the numbers seem counter-intuitive, such as the relatively large rate for Rajasthan and the low rates for Kerala and Punjab, but these likely reflect differences in the extent o f p-ivate-school enrollment across states. SchooLbased enrollment FigureN.3:Publicspendingon elementaryeducation per childaged6-14 years data often do not include andper studentenrolledinelementaryschool,1998-99 students attending private &zg hnachalPradest schools. HimachalPradest Naealanc Me&aIay, Go, B. Economic Growth, Gujara hbNpUr PublicSpending, and Keralr TamilNadr T.,"r>r ..lr-.l i School Enrollments Assam Maharashtr, Harvana 5 4.6 Patterns of g public expenditure on J m u & Kashmr "K%. MadhvaPradRh- Onssa elementary education As A?$:% Figure IV.3 shows, states B l h WestBengal spend very different " w g : g g 63The gross primary enrollment rate for 1999-2000 was 61%. 64 Even so, the discrepancy between the gross primary enrollment rate o f 95%, as reported by school administrative data, and 61%, as calculated from household surveys, is overly large. It is unlikely that it can be accounted for by students who are enrolled in school but not attending. 56 amounts o n elementary education. In 1998-1999, Sikkim, Mizoram, Arunachal Pradesh and Himachal Pradesh spent upwards o f Rs. 3,000 per child o f elementary school age (i.e., 6-14 years). At the other end, Bihar, Uttar Pradesh, Andhra Pradesh and West Bengal spent less than Rs. 1,000 per child 614. Thus, the states having l o w school enrollment or attendance rates are also the ones that spend less per child and per enrolled child. Note that some o f the inter-state variation in the discrepancy between the amount o f public expenditure per child and that per enrolled child arises due to variations in the proportion o f children attending private schools. InGoa, Kerala, Punjab and Haryana, for instance, private schooling i s more common among children 614, with the result that public spending per child enrolled in public school i s significantly larger than public spending per child aged 6- 14. 4.7 Figure IV.4 shows the increase in public Figure N.4: Annual % growth in elementary enrollments,populationaged 6-14, and real governmentexpenditureon elementaryeducationper chlld aged6-14, by state, spending on education over 1980-99 .annual growth in aggregatepublic spendingon elementaryeducation ~ time and across states. In annual growth (%) m realpublic spendingper child 6-14 oannualelementaryenrollmentgrowth(%) most states, real growth in mannualgrowth in numberof children aged 6-14(%) aggregate public spending on elementary education increased impressively between 1980-81and 1999- 2000 - exceeding the growth in the elementary schookaged population (ages 6-14) as well in elementary enrollments. As a result, real public spendingper child 6-14 increasedinvirtually every state - at annual rates o f 2% or more. 4.8 An important issue that arises is the allocation o f public spending on education across levels. Since the unit cost of university education i s much greater than that o f elementary and secondary schooling, the share of universities and higher educational institutions in public spending on education typically exceeds their share of the total student population in most countries. This i s true of India as well, but the mismatch Figure IV. 5: Distributionof enrolledstudents and of publicspending on educationin two states, by level, 1998- 99 AndhraPradesh HimaehalPradeih 0Elcmenmry0Secondary0Unwcrrlry QElcmenury DSesoodary ClUnivciiiry IW% IWh 50% 50% 40"h 40% 3Ph 30% 20% 20% 10% IVh 0% Students Public spending Public rpcndlng 57 between student and spending shares varies a great deal across states. An example of a state where the mismatch i s large i s Andhra Pradesh. The elementary education sector accounts for 63% o f all students in the state but fir only 50% o f all public spending on education (Figure IV.5). In contrast, the higher schooling sector in Andhra Pradesh commands 20% o f all public resources going into education but accounts for only 4% o f the students. At the other extreme is Himachal Pradesh, where the shares o f the elementary education sector in both the number o f students and in total public spending on education are approximately similar. 4.9 It may be instructive to see how inter-state changes in public spending on elementary education per child are correlated with changes in enrollment rates over time. An advantage of the administrative data on enrollments is that they are available at the state level in India going back a number o f years. In order to analyze the association between economic growth, public spending on elementary education, and gross primary enrollment rates across states and over time, we have merged the state-level data on gross primary enrollment rates over the period 1980-99 Figure N.6:Gross lowerprimary enrollmentratesand real public with state-level data on real spendingon elementaryschoolingper child 6-14, across states, 1980-99 gross state domestic 5.1- product per capita (GSDP) 3* 0 5.0- 1- and real public spending on 4.9- 4.8- elementary education per -4I54.7- child over the same 4.6' period.65 'g4.5. 6E ....'....?. -;:* .. 4.4- * 4 4.10 A plot o f the data -gpe4.3. 4.2. suggests a positive associa- -1 4.1. tion between the gross 4.04 1 primary enrollment rate and 5.0 5.5 6.0 6.5 7.0 1.5 8.0 real government expendi- Ln real gov't exp on elementaryschoolingper child 6-15 ture on elementary schooling per child aged 614 years (Figure IV.6). However, since there i s no control for other variables that may also influence school enrollments, it i s difficult to make much o f the association observed inthe figure. 4.11 Another manner inwhich to view the data is to examine the inter-state increases inelementary enrollments over the period 1980-81to 1999-2000 relative to the growth o f per child public spending on elementary schooling over the same period. This is shown in Figure IV.7. The figure suggests that over this 20-year period, some poor, low-enrollment states, such as Madhya Pradesh and Rajasthan, were able to enroll large numbers o f children in elementary school with relatively small increases in public sperding on elementary education. But other poor states, such as Uttar Pradesh, Bihar and Orissa, 65 Data on government elementary education expenditure were obtained from detailed budget demand documents o f individual states. Although government expenditure on primary education is o f interest in this paper, state government expenditure in India are available only for the elementaiy level, comprising the lower primary (grades L5) and upper primary (grades 68) levels. Note that the state government expenditure data used in this paper includes expenditure incurred by a state government out o f its own revenues as well as central government allocations to that state. 58 were able to increase elementary enrollments by only half as much for every additional (public) Rupee devoted to elementary schooling. The differences observed inFigure IV.7 could, o f course, Eflect differences in the quality o f schooling across states (as i s most likely the case with Kerala, which had the lowest number o f additional students in elementary school in relation to increased Figure IV.7: Inter-state increasesin elementaryschool(grades 1-8) expenditures), but they enrollments and in real government expenditure on elementary c education,1980-99 could also reflect 9000 differences in the marginal 8000 cost of expanding 7000 enrollments. In Madhya 6000 Pradesh, the launch o f the 5000 Education Guarantee 4000 3ooo Scheme (see Box IV.l), 2ooo which expanded the looo number o f schools in the rural areas o f the state at relatively low cost, may have been responsible for the l a r g increase in enrollments relative to changes in government elementary education expenditure per child. 4.12 FigureIV.8 shows the additional numbers o f Figure IV.8: Additional number of primary school(grades 1-5) students students enrolled at the enrolledper (1993-94) Rupeeincreasein gross statedomesticproduct per lower primary level capita,1980-99 3000 between 1980 and 1999 in several states relative to the 2500 real increase in gross state 2000 domestic product per capita 1500 over the same period. The poor states o f Bihar, loo0 Madhya Pradesh, Uttar 500 Pradesh, Orissa and Rajasthan rank among the 0 states that were able to obtain the largest increases @ in primary enrollments for given ncreases in (real) gross state domestic product per capita. Of course, these data simply reflect the fact that these poor states have experienced very little economic growth, but nevertheless have been able to increase primary enrollments significantly duringthe last two decades. 4.13 In Annex IV.l, an attempt is made to estimate the relationship between gross primary enrollment rates and government expenditure on elementary schooling per child 6- 14 more formally and with controls for other variables. While the results o f the analysis are not unambiguously clear, there i s evidence o f a significant positive association 59 between the gross primary enrollment rate and per-child government expenditure on elementary education across states, even after controlling for per capita income and adult female literacy. In addition, depending upon different specifications, the positive association between the gross primary enrollment rate and government elementary school expenditure i s observed to be stronger for the very poor states than for the nonpoor states. 4.14 The finding that the positive association between enrollments and government expenditure on education i s stronger in the poor states finds some support from another study on India. Lanjouw and Ravallion (1999) find tht the poor inIndia typically benefit more than proportionately at the margin when there i s an overall expansion in primary school enrollments. This happens because poor students are almost always the last to be enrolled, and the better-off are typically already in school. As a result, government educational expenditures that expand schooling access are generally well-targeted to the poor. 4.15 The econometric results presented in Annex IV.l also indicate a strong positive association between the gross primary enrollment rate and per capita GSDP, with the association becoming weaker at higher per capita income levels. Surprisingly, however, when there is control for per capita GSDP and per-child public spending on elementary education, adult female literacy inot significantly associated with the gross primary enrollment rate. Figure IV.9: School and primary school attendance rates, by age, 1999-2000 90 C. HouseholdSurvey- BasedEnrollment 80 . Estimates 70 60 . 4.16 As noted earlier, \ 50. I/ gross enrollment rates \ obtained from school 40 ./ \ administrative records 30 1 differ significantly from 20. - `\. \ \ household survey-based ,,,. Attendingany level estimates o f enrollment - -Attendingprimaw b-. 66 In what follows, the terms `attendance' and `enrollment' are used interchangeably. 60 4.17 Trends. Despite the grim situation, there Figure IV.10: Percent change in age-specific school attendance and net primary attendancerates, 1993-94 to 1999-2000 was some progress since the early 1990s. The 50th 15 11 round o f the NSS I O (undertaken in 1993-94) 5 indicates that there was an 0 increase o f about 7-10% in 5 the age-specific enrollment -10 8%change inage-specific attendance rate rate for ages 611 between -15 0%change innet primaiyattendance rate -20 1993-94 and 1999-2000 -25 (Figure IV.10). -30 Additionally, the net -35 J -3 I primary attendance rate for 6 7 8 9 I O 11 12 13 14 15 16 I? 18 Age (yea=) 6-11 year olds also increased, albeit by not very much, over the same period. The decline inthe net primary attendance rate observed for 13-15 year olds i s a positive development insofar as it reflects a reduction in the rate o f over-age enrollment inprimary schools. Figure N.11: Age-specific and net primary school attendancerates for 6-11 year olds, by state, 1999-2000 .Net primatyattendancerate Difference behveennet primary& age-specific(611) attendance rate fimachal Praderh 4.18 Inter-State Krma Differ-ences. There are TamilSikkim \id" Xdgalmd h t i\Zlzorama h r i i b large dif-ferences across Tnpw states in the primary blabGoa attendance rate (Figure IV.11). Attendance rates for the age group 611 exceed Jamnu h h aAjsama t a k 90% in 9 states - Kerala, WestBengal onssl Tamil Nadu, Makarashtra, Ra,aSthan UnarPradesh Madhya Pradesh Goa, Himachal Pradesh, Arunacha'PradeSh Bihar and the states of the 20 30 40 50 60 70 80 90 100 Northeast. At the other end, the primary attendance rates are only 75% or lower in Bihar, Orissa, Rajasthan, Uttar Pradesh, and Madhya Pradesh. With only 53% of children aged 611 attending school, Biharranks as the poorest-performing state inthe country. 4.19 Figure IV.11 also shows the very large disparity between the age-specific and the net primary attendance rate in all the states. Even in the states having high overall primary attendance rates, net primary attendance rates are significantly lower. For instance, Himachal Pradesh and Kerala, which have more than 95% o f children aged 6- 11 attending school, have a net primary attendance rate o f only 68% and 61%, respectively. In Bihar, a mere 28% of children aged 6-11 attend primary school! The pervasively low net primary attendance rate (relative to the age-specific rate) i s largely the result of late entry in primary school. For instance, the NSS 55th round data indicate hat, in the country as a whole, 25% o f 7-year old, 22% o f 8-year old, and 15% o f 9-year old school attendees were actually attending pre-primary (instead o f primary) school in 1999-2000. 61 4.20 Figure IV.12 shows the changes across states in Figure IV.12: Percentage of children aged 6-11 years attending school, by state, ,on, 1993-94 and 1999-2000 the age-specific enrollment 1993-94 01999-2000 rate for 6-11 year olds on I" between 1993-94 and 1999- 80 2000. In most o f the poor 70 states, such as Rajasthan, 60 Madhya Pradesh, Uttar Pradesh and Orissa, there 50 was a large increase in the 40 .J@@h4$$$@8 %$$J u proportion o f children aged 8 However,attending was an 6- 11 J*@ 7 Bihar school. exception, with a small decline inthe proportion attending school between 1993-94 and 1999-2000. 4.21 Intra-State Differences. Regional differences in the net primary school attendance rate are shown in Appendix Table 9.67 Six regions inthe country - largely in Bihar, Jharkhand, Orissa and Manipur - had a net primary attendance rate o f less than 40% in 1999-2000. Sixteen regions (about one-fifth o f all the regions) had a net primary attendancerate o f less than 50%. On the other hand, no region enjoyed a net rate o f more than 76%. Evenmore discouraging is the fact that 29 regions (or more than a third o f the regions in the country) either did not see an improvement intheir net primary attendance rate between 1993-94 and 1999-2000, or saw it decline (Appendix Table 9). Nineregions saw their net primary attendance rate increaseby 40% or more during the same period. D. Concentration ofOut-of-School Children Aged 6-11 4.22 According to the NSS 55* round data, there Figure IV.13: Contributionof 17 states to the national number of 6-11 were nearly 30 million out- year olds out of school, 1999-2000 I00- 94 97 of-school children aged 6- 90-- 11 inIndia in 1999-2000. If the goal of policy is to reduce the absolute number o f out-ofkchool children, it i s important to analyze the 0Cumulativeshareinnationwidenumberofout-of-school6-11yearolds 0Shareinnationwidenumberofout-of-school6-1Iyearolds distribution o f these children across states and sub-national units. The : : : : ! I NSS data indicate that nearly half o f all out-of- "".""<&@"."" 8 ""-8 &J/d.@&&J I 2 Schoolingyears ofhighest-educatedadult female inhousehold difference. In households where the adult female has 1-5 years o f schooling, the primary completion rate i s 66%, but it increases to more than 80% with more than 8 years o f schooling. G. Infrastructureand Schooling 4.32 The NSS data also suggest that access to infrastructure is associated with higher primary school attendance and completion rates. Districts having greater electricity coverage not only have higher age-specific, gross and net primary attendance rates but also higher primary completion rates (Figure IV.19). The evidence i s somewhat more mixed with respect to access to roads. Primary school attendance does not vary much with better road access. However, better road access does seem to improve primary completion rates (Figure IV.20). 'OThe NSS data do not permit identification o f the mother o f each child in the sample (unless the sample is restricted to biological children o f the household head). W e have therefore used the schooling years o f the highest-educatedadult female in a household as a proxy for maternal schooling. 65 H. Teachers andPrimary CompletionRates Figure IV.19: School attendancerates (ages 6-11) and primary completion rate for 12-year olds, by access to electricity, 1999-2000 OAge-specificattendancerate(Oh) 4.33 Since the pupil- OGrossprimaryattendancerate(%) teacher UNetprimaryanendancerate(%) ratio i s often OPrimarycompletionratefor 12-year olds (%) 87 considered an indicator o f school quality, one might 80 expect this ratio to be 70 associated with primary 60 completion rates. However, the NSS 55th round data do 50 now show any significant 40 association between the two 30 variables. While the 0-25 26-50 51-75 >75 primary completion rate for Yo of populationindistrict havingelectricity access 12-year olds falls as the Figure IV.20: School attendancerates (ages 6-11) and primary number o f pupils per completionrate for 12-yearolds, by access topucca roads, 1999-2000 teacher goes from below 40 Age-specificattendancerate(%) 85- Gross primaryattendancerate (oh) 85 to between 41 and 50, it is Net primaryattendancerate (%) 80. Primarycompletionratefor 12-yearolds (%) greatest in districts where 75 76 -r I, the pupil-teacher ratio is 75. above 50. The lack o f 70. association may reflect 65. other factors at work, such 60. 57 as teacher absenteeism or 55. the socioeconomic 50. background o f students. - 45 0-40 41-65 66-90 >90 4.34 One major YOofvillages indistricthavingaccesstoapucca road program launched by the Government o f India to address quality ksues was Operation Blackboard. Under this program, the Government provided a second teacher to all single- teacher primary schools and a teaching-learning equipment packet to all primary schools. Between 1987, when the program began, to 1994, when all the originally-targeted schools had been served, Operation Blackboard accounted for over half o f central government spending on elementary education and extended to virtually every district o f the country. On average, Operation Blackboard added two teachers for every 1,000 primary-school-aged children and increased the stock o f primary school teachers by 10%. 4.35 A rigorous econometric evaluation o f Operation Blackboard, using multiple rounds o f the National Sample Surveys, a census o f school resources (the All-India Educational Survey), and Operation Blackboard administrative data, found that, despite substantial misallocation o f Operation Blackboard teachers by state and local governments, the program did reduce the prevalence o f single-teacher schools and increase the number o f teachers per school (Chin 2002). The study also concluded that the teacher component o f Operation Blackboard significantly raised primary school completion and literacy rates for girls, but there was no such effect for boys. Girls' primary school completion increased by 3-4 percentage points, and girls' literacy 66 increased by 2-3 percentage points. Thus, Operation Blackboard not only induced illiterate girls who otherwise would never have attended school to attend, but also induced literate girls who otherwise would have dropped out to stay in school. Thus, the study suggests that improvements in school quality - in particular, the availability o f additional teachers inschools - may be a viable way o f retaining girls inprimary school. 4.36 Teacher Pay. To what extent does teacher pay influence teacher performance and student achievement? Using data from a sample survey o f 902 students and 172 teachers across 20 government- funded and 10 private schools inIndia, Kingdon and Teal (2002) explore the evidence for the payment o f performance-related pay and the extent to which such pay structures impact on student achievement. They find that, even after controlling for student ability, parental background, and indicators o f teacher and school quality, students in private schools get significantly better academic results (as measured by tests o f numeracy and literacy) by relating teacher pay to achievement. However, this association i s not observed for govemment schools, reflecting the fact that since govemment teaching jobs in India are typically permanent contracts with virtually no power o f dismissal, higher wages have little influence on teacher motivation inthe public sector. On the other hand, the flexibility that the private education sector has to set wages and dismiss lax teachers allows school managers to use teacher wages as incentives to enhance teacher performance. 4.37 Teacher Absenteeism Since retention o f students in primary schools depends upon the type and caliber o f instruction they receive, primary completion rates are likely to be more sensitive to teacher absenteeism than enrollment rates. Ina study o f schools in Uttar Pradesh, Dreze and Gazdar (1997) argue that ". .. the specific problem o f teacher absenteeism and shirking ... i s by far the most important issue o f education policy in Uttar Pradesh today." Dreze and Gazdar came to this conclusion after finding that two- thirds o f the teachers were absent from their positions during announced visits to 16 schools in the state. A study by PROBE (1999), based on visit to over 200 primary schools across the country, found that the head teacher (or principal) was absent innearly a third o f the schools on the day o f the sumy team's visit. A recent national survey focusing on service delivery inprimary health centers and primary schools found that, on average, one-quarter o f India's primary school teachers were not present at the schools where they are supposed to teach on any given day'l - rates that are highincomparison 7' The absence rate is the percentage o f staff that is supposed to be present but is not o n the day o f an unannounced visit. It includes both authorized and unauthorized absences. In fact, most teacher absences in schools may be authorized absences (Howes and Murgai 2004). For example, the PROBE study observed that o f the 200 days o f potential teacher attendance during a year, teachers had valid reasons to be absent from school for 50 days (or 25%): 20 days o f holidays and permitted leave, 21 days o f non-teaching duties (including deputation and in-service training), and 9 days for collection o f salaries (PROBE 1999). Thus, the high absence rate likely reflects generous leave terms and significant non-teaching duties assigned to teachers. Whether teacher absences are authorized or unauthorized, they reflect the general lack o f accountability o f the public school system to the students they are supposed to serve. There is evidence that the amount o f academic instructional time in a school influences student outcomes, especially in government schools. The study by Kingdon and Teal (2002) mentioned above found that a one percent increase in the minutes o f academic instruction per week in a government school increased student test scores in that school by almost 0.5 percent. This would seem to suggest that teacher absences, whether authorizedor unauthorized, are likely to have a large deleterious impact on student achievement. 67 to those observed in other countries (World Bank 2003). The problem o f absenteeism appears to be much worse inthe poorer states, with 39% o f primary school teachers being absent on any given day inBihar (Chaudhury et al. 2003). 4.38 One reason for the high rates o f teacher absenteeism may be extensive involvement o f teachers inpolitics. Kingdon and Muzammil (2003) cite the case o f Uttar Pradesh, where there has been significant penetration by teachers into state politics, in large part due to a constitutional provision for their reserved representation in the upper house o f the state legi~lature.~~ This has led to significant lobbying and union activity by the teachers and has earned them substantial salary gains and other monetary benefits from the state government. The salary gains have crowded out public spending on other items o f expenditure, such as instructional materials and supplies and scholarship^.^^ In addition, the increased participation o f teachers inregional and local politics has diverted them from their teaching responsibilities, and may in part be the reason for the high rate o f absenteeism. Of course, the problem o f teacher absenteeism goes far beyond the politicization o f teachers; it reflects a general lack o f accountability o f teachers and schools - indeed, the entire educational system - to community. students and the Figure 11.21: Rankingof statesinthe performance of government schooleducation services,2002 22 4.39 In contrast to Chaudhury et al. (2003), a national survey conducted by the Public Affairs Centre, which ranked the performance o f government school education services on a number o f indicators, such as access to a government primary school *@.& @+""&JNd8@&+$P &- within a kilometer from the / / place o f residence, propor- tion o f households sending their children to government primary schools, and user satisfaction with the behavior o f teachers and with the quality o f school facilities, found that the poor states (with the exception o f Orissa) did not quite rank at the bottom o f all the states inthe country. The Northeastern states, Punjab, and Haryana ranked lower than Bihar, Madhya Pradesh, and Uttar Pradesh in terms o f the quality o f their public school education services (Figure 11.21) (PAC 2002).74 '*Government school teachers are barred from contesting for elections to the lower house o f the state legislature. However, teachers in government-aided, private schools are not similarly debarred from elections. 73 I tis estimated that a Rupee spent on providing a full packet o f instructional materials in schools improves test scores by 14times as much as a Rupee spent on teacher salaries and 12 times as much as a Rupee spent on school facility improvement. 74 This is in contrast to the finding o f the same (PAC) survey that the poor states ranked in the bottom half o f all the states inthe country interms o f the quality o ftheir public health services. 68 4.40 Reducing teacher absenteeism and making schools accountable to students and the community is no simple task, however. As the World Development Report 2004 points out, it requires broad-ranging institutional reform, incorporating, among other things, empowerment o f citizens and communities who can holdthe state accountable for performance, devolution o f administrative and financial powers to communities, greater autonomy to schools, involvement o f parents in school management, and ensuring the motivation o f front-line workers. A good example o f an intervention that seeks to motivate front-line providers to improve the delivery o f services i s the Learning Guarantee Program, launched inthe poor and backwarddistricts o f Kamataka by the state government and the Azim Premji Foundation. The program aims to improve access and learning outcomes by introducing competition among government schools for performance-related awards (See Box IV.2 for a description o f the program). I.MultivariateAnalysisofPrimaryAttendanceandCompletion 4.41 To examine the likelihood o f the various states in India attaining the child schooling MD goals, we have estimated a multivariate model o f school attendance and primary attendance for 6- 11year olds and primary completion for 12-year olds, usingthe NSS 55' round unit record data (at the child The multivariate model has the advantage o f controlling for several variables that are simultaneously associated with child schooling. The estimation results are reported in Annex Tables 4 and 5, while only the broad findings o f the empirical analysis are discussed here. 4.42 The multivariate model confirms many o f the bivariate relationships shown in the graphs above. While the probability o f attending school or primary school increases with age, girls at each age have a significantly lower probability o f school and primary school attendame. Further, the gender disparity in attendance rates appears to increase with age. Since primary completion is considered only for 12-year olds, age comparisons inprimary completion are not relevant. However, the results do suggest the importance of gender, with 12-year old girls being nearly 4% less likely to complete primary school than 12-year old boys. 4.43 The schooling o f the highest-educated adult male or female in a household i s significantly associated with school attendance and completion - more so than the schooling o f the household head or the head's spouse. Adult female schooling typically has a smaller association than adult male schooling with school and primary school attendance, but the opposite i s true with respect to primary completion. 4.44 Evenafter controlling for log o f monthly consumption expenditure per capita, historically-disadvantaged social groups, such as scheduled castes, scheduled tribes and other backward castes, have significantly lower attendance and completion probabilities than forward castes, with scheduled tribes being generally the worst-off group. Likewise, Muslimsconsistently have lower attendance and completionrates than nonMuslims. The differences across social and religious groups are large; for instance, children aged 6- 11 75 Since the dependent variables in all three models are dichotomous(Le., assuming a value of zero or one), the modelshavebeenestimatedby the maximumlikelihoodprobitmethod. 69 belonging to scheduled tribes have a 16% lower probability o f attending school than children not belonging to scheduled tribes, and Muslim children aged 12 years have a 21% lower probability of having completed primary school than norrMuslimchildren 4.45 A household's living standards, as proxied by the log of its monthly consumption expenditure per capita, are strongly associated with school attendance and primary completion. Likewise, the log o f gross domestic product per capita in an individual's state o f residence has significant positive associations with all three measures of school enrollment and completion. Even after control for gross state domestic product per capita, the log of government expenditure on elementary schooling per child 6 1 4 years ina state is strongly associated with primary school attendance and primary completion butnot with overall school attendance.76 4.46 We have merged data on the number of cognizable kidnappings o f women and girls per capita in a district with the NSS household survey data to examine the association betweencrime against women and the school attendance and completion rates of girls. The empirical results suggest that crime against women is associated with a significant decline in female school attendance, female primary school attendance, and primary completion rates. The magnitude o f the estimated coefficients are, however, not overly large. 4.47 Infrastructure has predictable associations with school attendance and completion, with access to electricity having tk strongest associations. Access to roads i s not significantly associated with school and primary school attendance among 611 year olds, but it i s significantly associatedwith the probability o f children completing primary school. 4.48 An interesting question is the extent to which the quantity and quality o f school infrastructure ina community are associatedwith school attendance and completion rates. The availability o f primary schools per 1,000 children aged 6-11 in a district - an indicator o f schooling quantity - i s strongly associated with school and primary school attendance among 611 year olds, but it has no significant association with primary completion rates. On the other hand, lowering the pupil teacher ratio at the primary level ina district - an indicator of increasedschooling quality - is associatedwith higher rates o f school attendance and primary completion. This suggests that school attendance i s currently constrained by the availability o f primary schools,77 while both school 76 As an alternative, state fixed-effects models were estimated, wherein a set o f 23 state dummy variables replaced the two state-level variables (viz., log o f gross state domestic product per capita and log o f state government elementary expenditure per child 614). Although the state dummy variables were significant as a whole at the 5% level, the explanatory power o f the regressions, as measured by a pseudo R-squared measure, did not increase much with the substitution o f the state fixed effects for the state-level variables. The pseudo R-squared measures increased from 0.20 to 0.21, 0.06 to 0.07, and 0.09 to 0.13 in the case o f school enrollment, primary enrollment and primary completion, respectively. This suggests that the state fixed effects models are not superior to the models reported here in terms o f goodness-of-fit. 77 Naturally, given the linear prediction, this result would hold only up to some limit. 70 attendance and primary completion would likely benefit from school quality improvements inthe form o f a reduction o f the pupilteacher r a t i ~ . ~ ' , ~ ' J. Simulationsto 2015 4.49 Based on the multivariate probit models discussed above, we have undertaken simulations o f the school attendance, primary school attendance and the primary completion rate for the poor states under different intervention scenarios. These are shown in Table IV.l below. If the poor states were simply brought up to the national average in terms o f the nine interventions considered - viz., better road and electricity access, increased adult male and female schooling, household income (consumption) growth, growth in public spending on elementary education per child, reduced crimes against women, expanded number o f primary schools per 1,000 children, and reduction in the pupil teacher ratio at the primary level - the cumulative increases in the school and 6.9 percentage points (or lo%, 14.3%, and 12.4%), respectively. If the magnitude o f attendance, primary school attendance, and primary completion rates would be 7.3, 7.1, the interventions were to be scaled up, so as to bring the poor states to the average level o f the nonpoor states, the cumulative increase in school enrollment and completion would be significantly larger. The school attendance, primary school attendance, and primary completion rates in the poor states would increase by 20.7, 19.8, and 20.6 percentage points (or 28.4%, 39.8%, and 38.3%), respectively. The largest increases in school attendance and primary completion are obtained with improved living standards, expanded access to electricity, and increased government expenditure on elementary schooling. '*Lanjouw and Ravallion (1999) find that government educational expenditures that expand access are better targeted to poor people than resources that exclusively raise quality, since poor students are almost always the last to be enrolled and the better-off are typically already in school. A study for Kenya that looked explicitly at quantity (number o f schools) and quality (teacher-pupil ratio) improvements also found that an expansion in the number o f school facilities increased the enrollment o f children in the poorest expenditure quintiles but had no impact on the enrollment o f children in the top quintiles. On the other hand, an improvement in the teacher-pupil ratio increased the enrollment rate o f children in the top quintiles, but actually reduced the enrollment o f children in the poor quintiles (Deolalikar 1998). 79 Another study by Duflo (2001) found that the construction o f 61,000 primary schools by the Indonesian Government between 1973 and 1978 - one o f the largest school construction programs on record -led to a large increase in schooling. O n average, children aged 36 in 1974 received 0.12-0.19 more years o f schooling for each school constructedper 1,000 children in their region o f birth. 71 Table IV.l: Projected increase in school attendance and primary completion rates (percentage points) with various interventions inthe poor states Bringingpoor states to the level of the Bringingpoor states to level of the national average non-poor states ' average Primary Primary Primary School school Primary School school completio Intervention attendance attendance completion attendance attendance n Male schooling (years) 0.5 0.3 0.1 1.5 1.o 0.9 Female schooling (years) 0.6 0.2 0.6 1.8 0.6 2.1 Monthly per capita consumption expenditure 3.4 3.6 2.9 10.6 11.1 9.0 Annual government expenditure on elementary education per child 6-14years 0.5 1.3 1.4 3.6 Number o f cognizable kidnappings o f women and girls per 100,000 population in district 0.0 0.0 0.0 0.1 0.1 0.3 % of villages in district connected by pucca road 0.3 0.7 Access to electricity indistrict (%) 1.6 1.1 1.2 4.9 3.7 3.4 Number o f primary (grades 1-5) schools per 1,000 children aged 6-11 in district (x 1000) 0.9 0.6 1.4 0.9 Pupil teacher ratio in primary schools (grades 1-5) in district (x 1000) 0.4 0.7 0.4 0.4 1.o 0.6 Sum o f all interventions 7.3 7.1 6.7 20.7 19.8 20.6 4.50 We have also undertaken simulations o f the increase in the school attendance and the net primary school attendance rate in the poor states o f the country under the assumption that eight interventions are pursued simultaneously and gradually every year up to 2015. The nature and magnitude o f these assumed interventions are shown inTable IV.3. As noted in earlier chapters, the policy scenario assumed for the simulations is hypothetical. FigureIV.22 shows changes inthe projected school attendance rate for 6- 11 year olds in the poor states to 2015, under the assumption that all of the interventions are pursued simultaneously. The school attendance rate is projected to increase by 31.5 percentage points by 2015 - enough to attain universal primary enrollment. Most o f the interventions, with the exception o f a reduction inthe incidence o f crimes against women 72 and a reduction in the pupil-teacher ratio in primary schools (both o f which have statistically significant but numerically small associations), are associated with appreciable increases inthe school attendance rate. Table IV.2: Assumptions about various interventions to increase the school attendance and the net primary school attendance rate inthe poor states. 1999-2000to 2015 Starting Assumed Ending value in change per value in Intervention 1999-2000 vear 2015 Adult male schooling (years) 4.5 0.25 8.5 Adult female schooling (years) 2.0 0.3 7.8 Government expenditure on elementary education per child aged 6- 14 (Rs.) 955 4% 1,789 Consumption expenditure per capita (Rs.) 409 3% 656 Population coverage of electricity (%) 42.5 l%point 58.5 Crime against women (no. o f female kidnappings and rapes per 100,000 pop.) 1.65 -0.05 0.85 No. of primary schools per 1,000 children aged 6-1 1 5.1 0.2 8.3 Pupil teacher ratio in primary schools 91 -1.o 75 4.51 However, the situation is very different FigureIV.22: Increase in projected % of children aged 6-11 attendingschool with respect to the net in the poor states, 1999-2015, under differentinterventionscenarios (graphshows cumulativeeffectof eachadditionalintervention) primary attendance rate. Figure IV.23, which shows the projected increases in the net primary attendance rate in the poor states, shows that rate increasing by only about 27 percentage points by 2015 - well short o f the 50 percentage points needed to attain the MD goal. The o m - N e .o c m m h 5 5 00 5 v) 5 0 ? - " e " ) - - - - - ~ 0 55 0 5 0 00 00 0 0 00 0 ~ 5 0 0 0 5 relative ranking o f different " " N " " " " " N interventions inraising the net primary attendance rate is similar to that o f interventions to raise the overall school attendance rate, with the exception o f per-child government expenditure on elementary schooling, which has a significant association with the net primary attendance rate but not with the overall school attendance rate. 4.52 These results suggest that while it may be possible to get all children aged 6 11 in the poor states into school with concerted action on many fronts, raising the net primary enrollment or attendance rate to 100% by 2015 will be extremely challenging. Of course, the relative importance o f ensuring that all children aged 611 are in school versus ensuring that all children aged 611 are attending primary school is ckbatable. 73 Certainly, for the poor states o f India, even ensuring that all children aged 611 are in school (irrespective o f level) will be a worthwhile and laudatory achievement. 4.53 The findings o f a comparable analysis o f Figure IV.23: Increasein projected % of children aged 6-11 attendingprimary school in the poor states, 1999-2015,under different intervention scenarios primary completion rates in (graph shows cumulative effect o f each additional intervention) Increase in netprimary enrollment rote needed to attain universalnetprimary enrollment rote the poor states o f the I country are depicted in Figure IV.24. The 45 interventions considered for primary completion are largely the same as those for primary enrollment, with two exceptions. First, as the number o f primary schools per 1,000 children aged 6-11 is not significantly associated with primary completion, it is dropped from the projections shown in Figure IV.25. Second, as the availability o f village roads i s significant in explaining primary completion (but not primary enrollment), it i s included inthe projections. We assume that the percentage o f villages connected to a pucca road inthe poor states increases FigureIV.24: Increase inprojected primary completion rate (%) in the poor states, 1999-2015, under different intervention scenarios by one percentage point _ _ - _ _ _ _(graph_shows_cumulative_effect_ofeach_additional_intervention)- - _ _ - _ - - - - - - _ - - - _ _ - _ - - - _ _ _ _ _ _ _ _ _ - _ _ _ _ each year - from 60% in V 1999 to 76% in 2015. Int.nrmtinn Increase in primary completion rare needed to attain MDG -Reducing - the primary pupil teacher ratio -Reducing crime against women ImDmvine road access -Growth 4.54 As observed in inhh cons exp per cap -Increasing adult female schooling -Increasing Figure IV.24, the net _- adult male schooling Expandingelecmciq access // primary completion rate is -Increasing gov'texp on elementary schooling per child 6-14 projected to increase by about 29 percentage points by 2015, which is well short o f the 46 percentage point increase that would be needed to attain the MD goal o f 100% primary completion, but nevertheless represents significant progress. The largest inprovements in the completion rate occur with an ncrease in living standards, followed by an expansion o f female adult schooling and public spending on elementary education per child 614. Increases in adult male schooling and improved electricity coverage also are associated with smaller, but significant, associations with the primary completion rate. 4.55 The main implication of the findings presented in this chapter is that attaining the MDGs relating to universal net primary enrollment and 100% primary school completion will be extremely challenging in the poor states. Nevertheless, large gains in 74 both o f these indicators are possible, given economic growth, improved infrastructure, and increasedpublic spending on elementary education. Inaddition, the same package o f interventions should be successful in getting all children aged &11 years attending school. 4.56 Why do the simulations suggest attainment o f the educationrelatedMDGs to be a more challenging task than the attainment o f the infant mortality and child underweight-related MDGs? Inpart, this reflects the large discrepancy between the goals and the current status o f the various MD indicators in the poor states o f the country. The net primary enrollment rate in the poor states was only about 50% in 1999-2000. To attain universal net primary enrollment, the poor states in the country would have to double the net primary enrollment rate by 2015. In contrast, the infant mortality-related MDG calls for a reduction o f about 60% in infant mortality in the poor states (from a level o f 76 infant deaths per 1,000 live births in 1998-99 to a level o f 31 by 2015). Likewise, the underweight-related MDG calls for the proportion o f underweight children inthe poor states to be reducedby about 46% by 2015 - from a level of 51% in 1998-99 to 27% by 2015. Thus, the absolute nature o f the educationrelated MDGs (viz., the target being unrelated to the starting point) creates a large challenge for the poor states in the country that have very low rates o f net primary enrollment and primary completion. 4.57 As part o f its commitment to universalizing access to elementary schooling and ensuring completion o f elementary schooling by 2010, the Government o f India has launched the National Program o f Universal Elementary Education (UEE), known in Hindias Sawa ShikshaAbhiyan (SSA). (See Box IV.3 for a description ofthe SSA.) The Program not only i s consistent with the Millennium Development Goals (MDGs) for education, but goes beyond it in covering 8 years o f primary schooling, under a much tighter time frame. Among the quantitative goals o f the SSA are: all children to complete five years o f primary schooling by 2007; all children to complete eight years o f schooling by 2010; no gender and social disparities inprimary schooling by 2007 and inelementary schooling by 2010; and universal retention o f children in elementary schools by 2010. The SSA is a centrally sponsored scheme (CSS) whereby the Union Government provides incremental resources ina cost-sharing arrangement with the states: 85:15 inthe 9* Plan (FY 2001), 75:25 in the loth Plan (FY 2002-06), and 50:50 in the llth (FY Plan 2007-10). The SSA anticipates that Rs. 600 billion o f additional resources will be required from the budgets o f the central and state governments over the next 10 years to attain the SSA goals. 4.58 The simulations shown in Figure IV.22-IV.24 suggest that it will be very difficult to meet the government's goal o f having all children inIndia enrolled inprimary schools and completing the full course o f primary schooling by 2007. The goal o f getting all children aged 6-14 years to complete eight years o f schooling by 2010 would appear to be even more challenging. 75 Box IV.l: The EducationGuaranteeScheme of MadhyaPradesh Madhya Pradesh i s one o f the poorest states in India. It also has one o f the lowest levels o f school enrollment inthe country. Inorder to remedy this situation and as part o f its wide-ranging decentralization program, the government o f Madhya Pradesh began the Education Guarantee Scheme in 1997 to ensure access to primary schooling to every child in the State. The EGS represented a community-centered and rights-based approach to universalizing primary education in a quick, time-bound manner. Under the EGS, the Government gave a guarantee to provide a primary school facility to any habitation which did not have such a facility within a distance o f a kilometer. Further, the state government was obligated to provide the school within a period o f 90 days after receiving the demand for it from the local community. The EGS represents a three-way partnership among the state government, local governments (typically the village council or panchuyat), and the community. The community raises the demand, identifies a local resident who couldbe the teacher inthe new school, and provides the space for a school facility. The local government $anchuyat) appoints the teacher and oversees the functioning o f the school. The state government supports the school through a grant to cover the teacher's salary, arranges for the training o f the teacher, and provides teacher-learning material and other inputs for quality. Within a year o f its introduction, the EGS created a primary school facility inevery habitation o f the State. Indeed, during 1997 - the first year o f the scheme's operation - there were 40 new primary schools that opened every day, suggesting the enormous unmet demand for schooling that existed in the state. By August 1998, Madhya Pradesh had a primary school facility in every habitation. Thus, within a short period o f 18 months, the state eliminated the historical backlog o f schooling - at one-third the usual cost o f establishing new school facilities. The EGS did not merely provide physical access; more importantly, it provided social equity, with the new school facilities being largely demanded to bring primary schooling to scheduled tribes and girl children. The EGS has become a national model for a community-based approach to universalizing primary schooling in India. While there have been no rigorous evaluations o fthe EGS, some studies suggest that schooling access for the poorest 40% o f households increased more rapidly in Madhya Pradesh between 1992-93 and 1998-99 ("before EGS" and "after EGS" years for which national survey data are available) than in the rest o f India (McCarten and Vyasulu 2003). An econometric evaluation o f the central government's District Primary EducationProgram (DPEP-I) also found that Madhya Pradesh was the only state (of a total o f 7 states considered) which showed significant impacts o f DPEP-I on school attendance rates among 6-10 year olds (Jalan and Glinskaya 2003). 76 ~ Box IV.2: The Learning Guarantee Program in Karnataka In 2002, the Azim Premji Foundation, in collaboration with the Government o f Karnataka, launched the Learning Guarantee Program, which is aimed at guaranteeing the leaming o f children in government schools. All government higher and lower primary schools (approximately 10,000) situated inthe 43 blocks o f the northeastern districts o f Karnataka State (identified as most backward) were eligible to participate in this competition. Between 2002-03 and 2004-05, when the program will end, participating schools will have three opportunities to enter the competition and improve their capabilities. The program is essentially a competition, with rewards and recognition schemes for teachers and children o f schools which demonstrate an improvement in student learning. To encourage schools to both expand access and improve quality o f learning, the program rewards schools on the basis o f leaming outcomes in the school's habitation (not the school's population). If a minimum o f 60% o f all children in the school's habitation demonstrate, in an independent assessment, that they have acquired 90% o f the prescribed competencies in mathematics and Kannada (the regional language o f Kamataka), the school would receive an award o f Rs. 5,000. With 70% o f the children in the habitation acquiring the expected learning outcomes, the reward would be Rs. 10,000. Finally, a school that demonstrates 80% o f children inthe habitation achieving the expected leamingoutcomes qualifies for the highest rewardo fRs. 20,000. Thus, a school can win a maximum o f Rs. 60,000 inthe three rounds o f the competition. Inthe first year ofthe program, nearly 1,000 schools offered themselves for evaluation. Many more are expected to opt into the program inthe second andthirdyears. The Learning Guarantee Program i s an attempt to encourage government schools to deliver learning outcomes and to make them accountable for the services they deliver. It i s expected that the LGP will enable primary stakeholders (parents and the community) to put pressure on their children's schools to become a certified Learning Guarantee school. It i s also expected that `Leaming Guarantee' schools will motivate out-of-school children to enroll in school and encourage the children in school to continue (and not drop out). By fostering healthy competition among all communities, the program i s likely to both improve access and quality. Source: Azim PremjiFoundation(2002). 77 Box IV.3: TheSarva ShikshaAbhiyan: A Programfor UniversalElementaryEducationinIndia In accordance with the constitutional commitment to ensure free and compulsory education for all children up to the age o f 14 years, provision o f universal elementary education has been a salient feature o f national policy since independence. This resolve has been spelt out emphatically in the National Policy since independence (NPE), 1986 and the Program of Action (POA) 1992. A number of schemes and Programs were launched inpursuance ofthe emphasis embodied inthe NPE and the POA. These included the scheme of Operation Blackboard (OB); Non Formal Education (NFE); Teacher Education (TE); Mahila Samakhya (MS); State specific Basic Education Projects like the Andhra Pradesh Primary Education Project (APPEP); Bihar Education Project (BEP), Lok Jumbish (LJP) in Rajasthan; National Program o f Nutritional Support to Primary Education (MDM); andthe District PrimaryEducation Program (DPEP). The Sawa Shiksha Abhiyan (SSA) is a historic stride towards achieving the long cherished goal o f Universalization o f Elementary Education (UEE) through a time-bound integrated approach, inpartnershipwith States. SSA, which promises to changethe face ofthe elementary educationsector o f the country, aims to provide useful and quality elementary education to all children aged 614 years by 2010. The SSA i s an effort to recognize the need for improving the performance o f the school system and to provide community-owned quality elementary education in the mission mode. It also envisages bridgingo f gender and social gaps. Objectives o f the SSA All children in school, Education Guarantee Centre, Alternative School, `Back to School' camp by 2003; All children complete five years ofprimary schoolingby 2007; All children complete eight years o f schoolingby 2010; Focus on elementary education o f satisfactory quality with emphasis on education for life; Bridge all gender and social category gaps at primary stage by 2007 and at elementary education level by 2010; Universal retentionby 2010. Structure for Implementation The Central and State governments will together implement the SSA in partnership with the local governments and the community. To signify the national priority for elementary education, a National SSA Mission i s being established with the Prime Minister as the Chairperson and the Union Minister o f Human Resource Development as the Vice Chairperson. States have been requested to establish State level Implementation Society for UEE under the Chairmanship o f Chief Minister Education Minister. This has already been done inmany States. (continued on next page) 78 Box IV.3 (cont'd): The Sawa ShikshaAbhiyan: A Program for UniversalElementary Education inIndia The SSA will not disturb existing structures in States and districts but would only try to bring convergence in all these efforts. Efforts will be made to ensure that there is functional decentralization down to the school level in order to improve community participation. Besides recognizing PRIdTribal Councils in Scheduled Areas, including the Gram Sabha, the States would be encouraged to enlarge the accountability framework by involving NGOs, teacher, activists, and women's organizations. Coverage and Period The SSA will cover the entire expanse o f the country before March 2002 andthe duration o f the Programin every district will depend upon the District Elementary Education Plan (DPEP) preparedby it as per its specific needs. However, the upper limit for the Program period has been fixed as ten years, i.e., up to 2010. Strategies Central to SSA Institutional reforms - As part o f the SSA, institutional reforms in the States will be carried out. The state will have to make an objective assessment o f their prevalent education system including educational administration, achievement levels in schools, fmancial issues, decentralization and community ownership, review o f state Education Act, rationalization o f teacher deployment and recruitment o f teachers, monitoring and evaluation, education o f girls, SC/ST and disadvantaged groups, policy regarding private schools, and ECCE. Many States have already impkmented institutional reforms to improve the delivery system for elementary education. Sustainable Financing - The SSA i s based on the premise that financing o f elementary education interventions has to be sustainable. This calls for a long-term perspective on fmancial partnership between the Central andthe State governments. Community ownership - The Program calls for community ownership o f schookbased interventions through effective decentralization. This will be augmented by involvement o f women's groups, VEC members, and members o f PanchayatiRaj institutions. Institutional capacity building - The SSA conceives a major capacity building role for nationaL and state-level institutions llke NIEPA, NCERT, NCTE, SCERT, and SIEMAT. Improvement in quality requires a sustainable support system o f resource persons. Improving mainstream educational administration - The Program will have a community-based monitoring system. The Educational Management Information System (EMSI) will correlate school level data with community-based information from micro planning and surveys. Besides this, every school will have a notice board showing all the grants receivedby the school and other details. (continued on next Daze) 79 Box IV.3 (cont'd): TheSawa Shiksha Abhiyan: A Program for UniversalElementary Educationin India Habitation as a unit ofplanning - The SSA works on a community-based approach to planning with habitation as a unit o f planning. Habitation plans will be the basis for formulating district plans. Accountability to community - SSA envisages cooperation between teachers, parents and PRIs, as well as accountability andtransparency. Education of girls - Education o f girls, especially those belonging to the scheduled castes and scheduled tribes, will be one o f the principal concerns o f the SSA. Focus on special groups - There will be a focus on the educationparticipation o f children from scheduled castes and tribes, religious and linguistic minorities, disadvantaged groups, and disabled children. Pre-Projectphase - SSA will commence throughout the country with a well-planned pre- project phase that provides for a large number o f interventions for capacity development to improve the delivery andmonitoring system. Thrust on quality- SSA lays a special thrust on makingeducation at the elementary level useful and relevant for children by improving the curriculum, child-centered activities, and effective teaching methods. Role of teachers - SSA recognizes the critical role o f teachers and advocates a focus on their development needs. Setting-up o f BRCICRC, recruitment of qualified teachers, opportunities for teacher development through participation in curriculum-related material development, focus on classroom process, and exposure visits for teachers are all designed to develop the human resources o f teachers. District Elementary Education Plans - As per the SSA framework, each district will prepare a District Elementary Education Plan, reflecting all the investments being made in the education sector, with a holistic and convergent approach. Components of SSA The components of the SSA include appointment o f teachers, teacher training, qualitative improvement o f elementary education, provision o f teaching leaming materials, establishment of Block andCluster Resource Centers for academic support, construction o f classrooms and school buildings, establishment o f education guarantee centers, integrated education o f the disabled, and distance education. Source: Governmentof India (2002). 80 Annex IV.l: GrossPrimaryEnrollmentRates, Government ElementaryEducation Expenditure,and Per CapitaIncome across States, 1980-99 4.59 We use twenty years o f state-level data on gross primary enrollment rates (GPERs), real gross state domestic product per capita (GSDP), and real public spending on elementary education per child 6- 14years to analyze the association between the gross primary enrollment rate on the one hand and public spending on elementary education and economic growth on the other.80381 4.60 The results o f this analysis are reported in Annex Table IV.l below. OLS estimates indicate a very strong positive association between the gross primary enrollment rate and government expenditure on elementary education per child even after control for per capita GSDP, with a one percent increase in government elementary education sperding per child being associated with a 0.17 percent increase in the gross primary enrollment rate. Surprisingly, when control for unobserved heterogeneity across states - in the form o f time-invariant state fixed effects - i s introduced in the model, the magnitude o f the positive association increases (from 0.17 to 0.23). The inclusion o f state fixed effects, however, rendersthe association between the gross primary enrollment rate and per capita GSDP insignificant, indicating that state heterogeneity i s positively associated with GSDP. An interaction between the log of government health expenditure and the log o f per capita GSDP is negative and significant in the fixed-effects model, indicating that the positive association between the gross primary enrollment rate and public spending on education i s weaker at higher levels o f per capita state income. It is interesting that these results mirror those obtained for infant mortality (and reported in Annex Kl),also usingcross-state, time-series data over the same period. 4.61 Unfortunately, time-series data on other covariates o f gross enrollment, such as female adult literacy, are simply not available at the state level for the period 1980-99. To rule out the possibility that the significant association between the gross primary enrollment rate and government educational expenditure per child could be due to the omission o f female adult literacy from the model, we have calculated state-level averages o f the adult female literacy rate from four rounds o f the National Sample Survey - for 1983-84, 1987-88, 1993-94 and 1999-2000. The inclusion o f female literacy means that Data on government elementary education expenditures were obtained from detailed budget demand documents o f individual states. The expenditure variable includes expenditure incurred by a state out o f its own revenues as well as central govemment educational allocations to that state. A state-level regression o f the gross primary enrollment rate on govemment elementary education spending i s based on the assumption that public spending is not endogenously distributed across states (Rosenzweig and Wolpin 1986). Such an assumption may not always hold, especially if government educational expenditures are allocated across states based on literacy conditions. However, the inclusion o f state fixed effects in the regression effectively controls for endogenous program placement. W e allow for a central planner to distribute public spending on elementary education across states according to an unobserved educationiliteracy attribute; in other words, states that have innately l o w literacy rates may spend more on elementary education than those having innately high literacy. Since the fixed-effects model purges the unobserved (time -invariant) literacy attribute from the model, it effectively addresses the endogenous program placement problem. 81 the gross primary enrollment rate regression can only be estimated for these four years, which reduces the sample size from 266 to merely 56. However, with one exception, the results continue to indicate a strong positive association between the gross primary enrollment rate and public spending on education and a negative and significant coefficient on the interaction between public spending and per capita GSDP. 4.62 The finding that there i s a positive association between the gross primary enrollment rate and real government elementary education expenditure per child should not come as a surprise. Additional expenditures increase school capacity (in the form o f both classrooms and teachers), and to the extent that enrollments are capacity- constrained, additional expenditures will allow more children to be enrolled in a larger number o f schools and classrooms. 4.63 What may be more surprising i s that the observed positive association between public spending on elementary education and the gross primary enrollment rate i s stronger in the poor states than in the nonpoor states. This may simply reflect the fact that schools and school spaces are more limited in the poor states than in the nonpoor states, so that an expansion in school capacity results in a greater increase in enrollments inthe former than inthe latter. As noted earlier inthis chapter, there is some support for such a finding from another study on India. Lanjouw and Ravallion (1999) find that the poor in India typically benefit more than proportionately at the margin when there i s an overall expansion in primary school enrollments. This happens because poor students are almost always the last to be enrolled, and the better-off are typically alreadyin school. As a result, Lanjouw and Ravallion (1999) argue that government educational expenditures that expand schooling access are generally well-targeted to the poor.82 4.64 Surprisingly, the econometric results presented in Annex IV.l do not indicate a strong association between the gross primary enrollment rate and adult female literacy rates in a state, once there is control for both per capita GSDP and per-child education spending. 4.65 It is important, however, to interpret these results in the context o f the discussion in section H o f this chapter. The results do not consider the counter-factual - viz., the improvements in educational outcomes that would occur if the quality o f government education expenditures were simultaneously improved via greater accountability in service delivery and better governance. All that these results indicate i s the urgent need for investments in additional schools, classrooms, and teachers, to be targeted to the low-enrollment states via an effective and accountability-based delivery system. 82 A study for Kenya that looked explicitly at quantity (number o f schools) and quality (teacher-pupil ratio) improvements also found that an expansion in the number o f school facilities increased the enrollment o f children in the poorest expenditure quintiles but had virtually n o impact on the enrollment o f children inthe top quintiles. O n the other hand, the same study found h a t an improvement in the teacher-pupil ratio increased the enrollment rate o f children in the top quintiles, but actually reduced the enrollment o f children inthe poor quintiles (Deolalikar 1998). 82 Annex Table IV.l: Fixed-effects, log-linear regression of the gross primary enrollment rate, pooled data for 1980-1999 across Indian states OLS Fixed-effects Fixed-effects with adult female literacy* Independent t- t- variable Coeff. t-ratio Coeff. t-ratio Coeff. ratio Coeff. t-ratio Coeff. ratio Lnreal gov't elemen. educ. per child 6-14 (GI 0.173 4.90 0.226 5.67 2.211 5.23 2.490 2.02 1.344 0.87 Lnreal gross state domestic product per capita (Y) 0.229 6.74 -0.034 -0.47 1.605 4.53 1.951 1.90 1.971 1.93 LnG x LnY -0.225 4.71 -0.268 -1.95 -0.139 0.81 Lnadult female literacy (F) 0.175 0.86 2.160 1.33 LnFx LnY -0.238 1.23 Time trend (t) -0.001 -0.46 0.003 0.96 -0.004 1.04 -0.007 -0.60 -0.005 0.44 Intercept 1.504 5.61 3.452 5.33 -10.946 3.51 -13.896 -1.52 -13.671 1.51 Number o f observations 252 252 266 56 56 F-test for model 35.91 37.72 38 2.98 2.83 R-squared 0.422 0.808 0.8247 0.7971 0.8057 F-test for state effects 36.00 40.216 4.663 4.465 Notes: All regressions (with the exception o f those marked *) use pooleddata across 14 states and 20 years. Dependent variable is log o f the gross primary enrollment rate. Data are obtained from various issues o f the Selected Educational Statistics, state governmentbudget demand documents, and CSO statistics on GSDP o f states. All values are in constant 1993-94 Rupees (using statespecific GSDP deflators). Standard errors are corrected for heteroscedasticityusing the Huber-white method. Figures in bold are statistically significant at the 10% or lower level. *Data on female literacy are available only for 1983-84, 1987-88, 1993-94 and 1999-2000, so these regressions are restricted to these four years. 83 5. GENDERDISPARITYIN SCHOOLING 5.1 As inother parts ofthe world, there are large disparities inthe economic, social and schooling opportunities available to men and women in India. Gender disparities begin early in life, with female infants having a lower chance o f survival than male infants, owing largely to parental neglect of female infants. The discrimination against the female continues as the child grows older -first inthe form of smaller rations of food and nutrition (especially for higher-order females) and later in the form of fewer schooling opportunities relative to boys. As women enter the labor force, they face discrimination inthe labor market as well, primarily inthe form o f lower wages. 5.2 This paper focuses on gender disparities in FigureV.1: Ratioof female to malegross primary enrollment rate, 1950-51to 1999-2000 schooling opportunities. 85 The MD goal i s to eliminate gender disparities 75 in schooling, such that the ratio of girls to boys 65 enrolled at all schooling 55 levels, but particularly at the primary and secondary 45 levels, is 100%. 35 A. Trends and Patterns 5.3 Levels and FigureV.2: Ratio of femaleto malegross primaryenrollmentrate, by state, Trends. Schookbased 1999-2000 administrative data suggest Punjab Ha ani that India has made SliVkllll Kerala impressive gains in Delhi reducing the male-female gap in the gross primary enrollment rate in the last fifty years (Figure V.1). The ratio o f the female to male gross primary _ _M_ P- enrollment rate nearly onssa KaShm*r doubled from a level of Blhm U P Rajafthan 41% in 1950-51 to 81% in 50 60 70 80 90 100 1993-94, where it has 84 stayed since.83 5.4 Interstate Variations. There are large FigureV.3: Ratio of females to males enrolled in primary (grades 1-5) school, 1980-81 and 1999-2000, by state interstate variations in the =Ratio offemalesto males enrolled, 1980-81 extent o f gender disparity m R a t i o of females to males enrolled, 1999-00 -% change in femaleimaleratio. 1980-99 in schooling (Figure V.2). I00 IO0 90 90 The school administrative 80 80 data indicate that gender 9 70 70 disparity i s greatest in 2 60 6o 0 50 50 I Bihar, Uttar Pradesh and 11 40 40 Rajasthan, where the gross LL 30 30 primary enrollment rate for 20 20 10 10 females i s about two-thirds 0 0 or less than that for males. On the other hand, there i s parity or near-parity between male and female gross primary enrollment rates inPunjab, Haryana, Sikkim and Kerala.84 5.5 How has gender disparity in primary FigureV.4: Percent changein the ratio of females to males enrolled in primary schools, 1980-81 to 1999-2000 enrollments changed over A time across states? Figure V.3, which shows the ratio o f females to males enrolled in primary school in 1980-81 and 1999-2000, indicates wide variation in performance across states. The largest relative gains for girls occurred in Haryana, where the ratio o f 5 1 ' " A ' ' " ' A ' kmales to males enrolled 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 1M1 Ratio o f females to malesinprimaryschool in 1980-81 (%) in primary school almost doubled ocer the 19-year period. Bihar, Karnataka and West Bengal also recorded large increases inthe ratio o f female s to males inprimary school. At the other extreme, Orissa, Uttar Pradesh, and Kerala experienced small relative declines in the female-male ratio 83 As inthe previous chapter, primary schooling here refers to grades 1-5. 84 The observation that Punjab and Haryana have higher rates o f female than male primary enrollment deserves an explanation. These states actually have a very different form o f gender discrimination in schooling. Since these are rich states, both boys and girls get enrolled in school. However, since there is a growing private sector in these states, parents tend to enroll their male children in private schools while enrolling their daughters in government schools. Since school-based administrative data typically only cover enrollment in government institutions, the boys who are shifted out o f the government schools are not counted while the girls attending government schools are counted. This gives a misleading picture o f girls having higher enrollment rates than boys. 85 (although it i s important to note that the female-male ratio o f primary school students was already over 95% inKerala in 1980-81). Ingeneral, states that had low ratios o f female to male primary students in 1980-81 experienced somewhat larger percentage increases in the ratio ktween 1980-81 and 1999-2000 (with Uttar Pradesh and Orissa being major exceptions) (Figure V.4). B. Economic Growth, PublicSpending, and Gender Disparity 5.6 It is not possible to disaggregate government Figure V.5: Relationship betweenthe ratio of females to males in primary schooland real public spendingon elementary education per child 6-14, expenditure on elementary 14 statesin 1980-81 and 1999-2000 education by gender. 1001 4 However, it i s possible to see how gender disparity in primary enrollment i s associated with both public spending on elementary education and per capita 40 income across states. As in 'z 30 the previous chapter, we have done this by merging . . . . . . . . . . . . . 200 300 400 500 600 700 800 900 1,000 1,100 1,200 1,300 1,400 1,500 state-level data on the ratio Realpublicspendingon elementaryeducationper child6-14 (1993-94 Rs.) o f females to males in ; 14 states in 1980-81 and 1999-2000 suggestsspend morethaton stateselementarythat 2a9 0 : :: education per child aged 6- 6 80- 14 tend to have higher 70- ** ratios o f kmales to males 5+5 60- 4 in primary school. This, in 50- * turn, implies that, at the . + margin, public spending on ,: 4o b elementary education i s 2 t 30- associated with more 20 -9 86 capita income. States having a higher GDP per capita are observed to have higher ratios o f &males to males in primary school. There are several reasons for expecting gender equality, at least in primary schooling, to improve with both public spending and per capita income. Ifparents are more likely to send their male than female children to school in the face o f capacity (school, class-room, and teacher) comstraints, additional public spending that would result in a larger supply o f schools, classrooms and teachers would benefit primary schookaged girls to a greater extent than boys. Likewise, increases in income per capita and living standards could relax household budget constraints that typically tendto work against the girl child's school attendance. C. Household Survey- Based Estimates of FigureV.8: Ratio of femaleto male students attending primary and secondaryschool, by age, 1999-2000 Gender Disparity 90 85 5.8 As noted in the 80 previous chapter, gross 75 enrollment rates obtained 70 from school administrative 65 60 records differ significantly from household survey 55 50 based estimates o f enroll- 45 40 ment (more appropriately, attendance). Data from the 35 55* round o f the NSS 6 7 8 9 10 1 1 12 13 14 15 16 17 18 Age (vean) _ . (undertaken in 1999-2000) show that the female disadvantage in schooling opportunities starts as early as age 6 and continues all the way until age 18 (and beyond) (Figure V.7). However, the female disadvantage manifests itself much more strongly after age 13, as evidenced by the consistently falling ratio o f Emale to male students in primary and secondary school (Figure V.8). 5.9 Inter-State DiffeYences. There are large differences across states in the ratio o f female to male students at the primary and secondary level (Figure V.9). The Northeastern states, Kerala, and Himachal Pradesh generally have the highest ratios o f female to male students, while Madhya Pradesh, Uttar Pradesh, Bihar and Rajasthn have 87 the lowest ratios. InBihar and Rajasthan, there are only slightly more than half as many female students as male students at the primary and secondary levels. 5.10 Trends. There Figure V.9: Ratio of girls to boys in primary and secondaryschools (%), was, however, some by state, 1999-2000 improve- ment in the relative position o f &males in the latter half of the 1990s. The 50th and 55'h And rounds o f the NSS indicate that, between 1993-94 and 1999-2000, the fkmale school attendance rate increased more than the Nagaland lammu Goa male attendance rate for Madh a Radesh T"pUa every single age group. The Ut& Pradesh RaJaslha" B k disparity between the 40 50 60 70 80 90 100 increases fir males and Emales was greatest for older children, typically aged 13 and above (Figure V.10). 5.11 Intra-State Figure V.10: % change in age- & gender-specific school attendance rates, 1993-94 to 1999-2000 Differences. Regional 2o differences in the ratio o f .Males OFemales female to male students in primary and secondary school are presented in lo Appendix Table 9. As many as seven regions in 5 the country - in Bihar, Madhya Pradesh, Rajasthan 0 and Gujarat - had female students constituting less 3 -5 than 60% o f male students in primary and secondary schools. h o t h e r 18 regions had a ratio o f female to male students o f less than 75%. This means that, in nearly a third o f the regions in the country, female students constitute fewer than three-quarters o f male students at the primary and secondary levels. 5.12 Even more discorcerting is the finding that, between 1993-94 and 1999-2000, as many as 30 regions (more than a third) experienced a decline in the ratio o f female to male students. Surprisingly, these regions are located in such states as Tamil Nadu, Andhra Pradesh, Karnataka, Maharashtra and even Kerala (Appendix Table 9). D. Socioeconomic DifferencesinGender Disparity 5.13 There are some large, and sometimes unexpected, socio-economic pat-tems in the relative positiono f girls inschooling opportunities. As would be expected, the ratio o f 88 female to male students enrolled at the primary and secondary levels is much lower for the socially disadvantaged groups, like scheduled tribes, scheduled castes and other backward castes, than for the mainstream social groups (e.g., forward castes) (Figure V.11). However, the pattern across economic groups is unexpected. The bottom consumption quintile has the highest ratio o f female to male students among all consumption quintiles, while the second quintile has the lowest ratio (Figure V.11). The differences across the top three quintiles are relatively small. 5.14 Gender disparity in schooling opportunities for children has a predictable pattern across adult schooling levels in the Figure V.11: Ratioof girls to boysinprimary andsecondaryschools("A), household. by socialgroup and consumtionquintile, 1999-2000 The ratio o f female to male students i s 80 greater in households where the highest-educated male 75 or female has some schooling (as opposed to no schooling) (Figure V.12). 70 The data also seem to suggest that the relative to a largerOfextent with adult position girls improves 65 female schooling in the I I I household than with adult Social group Pe: capita consimption erpinditure quiniile male schooling. For instance, in househoIds where the highest-schooled adult female has 6 or more years o f schooling, the ratio o f female to male students i s 89%; in contrast, in households where the highest-schooled adult male has 6 or more years o f schooling, the ratio i s only 77%. E. Infrastructureand Gender Disparity 5.15 The NSS data also suggest that access to infrastructure i s associated with lower levels o f gender disparity in schooling. For instance, in Figure V.12: Ratio of girls to boysinprimary and secondaryschools ("A), districts where more than by adult male and female schoolingin household, 1999-2000 three-quarters o f the 5o population has access to 85 electricity, the ratio o f kmale to male students is 80%, while in those 75 districts where electricity coverage i s 25% or less, the 70 ratio is only 61% (Figure 65 V.13). 63 5.16 Likewise, better access to roads i s also 89 associated with lower levels o f gender disparity, although the difference in the ratio o f female to male students across districts with poor Figure V.13: Ratio ofgirlsto boysinprimary and secondaryschools("h), byinfrastructure availability, 1999-2000 and good road infrastructure is not as large 85 as that between districts 80 with poor and good electricity coverage (Figure 75 V.13). 70 5.17 How does gender 65 disparity in enrollments 6o vary with school infrastructure? The data do 55 not show any relationship between the ratio o f female I % ofpopulation with access to electricity in distnctI % of villages ~ndistnct connected to pucca road 1 to male students in primary and secondary schools on the one hand and the number o f primary and secondary schools per 1,000 children aged 6-18 on the other (Figure V.14). However, a clear relationship is observed between government expenditure on elementary and secondary education per child aged 6- FigureV.14: Ratioof girlsto boys inprimary and secondaryschools( O h ) , 18 and the ratio o f female by schoolavailability and by governmentexpenditureon education,1999- 2000 to male students (Figure 85 83 V.14). In states where per- child public spending is 8o less than Rs. 1,000 per 75 annum, the ratio o f female to male students i s 65%. 70 The ratio increases consistently with greater 65 spending, and reaches a I I I 6o level o f 83% in states c Rs. 1,000 Rs. 1000 - Rs. 1,500 - > Rs. 1,999 which spend Rs. 2,000 or I I I 1,&9 1,999 No. of elementary and secondary schools per 1,000 Government expenditure on elementary and secondary more per annum on children aged 6-18 education per child aged 6-18 elementary and secondary education per child 6- 18. F. Multivariate Analysis of Gender Disparity 5.18 To examine the likelihood of the various states in India attaining the gender disparity MD goal, we have estimated a multivariate model o f school attendance separately for boys and girls aged 6-18 years, using the NSS 55th round unit record data (at the child The multivariate model has the advantage o f controlling for several variables that may be simultaneously associated with child schooling. The estimation 86 A single equation with separate intercept and slope female dummy variables was estimated. Since the dependent variable is a dichotomous variable (Le., whether or not a child is in school), the model has been estimated by the maximumlikelihood probit method. 90 results are reported in Annex Table 6, while only the key findings o f the empirical analysis are discussed here. 5.19 The multivariate model indicates a significant positive association between household living standards (as measured by household consumption expenditure per capita) and school attendance, with the association being significantly smaller for girls than for boys.87 Controlling for household living standards, the level o f economic development in a state (as proxied by state per capita GDP) i s inversely associated with female school attendance (and has no significant association with male attendance). 5.20 Public spending on elementary and secondary education per child aged 618 years i s significantly (positively) associated with school attendance, but there i s no significant difference in this association across males and females. Surprisingly, however, the composition of public spending across schooling levels is associated with gender disparity, with a higher share o f secondary schooling in total public spending on education being associated with a higher ratio o f female to male enrollment.88 Such a finding would be consistent with additional public spending on secondary education drawing more girls than boys into school, at least in comparison to additional public spendingon primary education 5.21 The results also indicate that access to electricity i s associated with a reduction o f gender disparity, since it has a stronger association with female than male school attendance. On the other hand, access to roads has a smaller association with female attendance, probably reflecting the fact that, even if roads are available, parents are less likely to send their daughters to schools far away from their village o f residence. 5.22 The findings on the association between adult male and female schooling on the one hand and male and female school attendance on the other hand are o f interest. While there i s no significant gender difference in the association between adult male schooling in a household and child school attendance, a gender difference is observed for adult kmale schooling. The latter clearly has a much stronger (positive) association with 87 These results do not necessarily contradict those reported in Section B o f this chapter. There, it was observed that increases in state per capita income were associated with an increased ratio o f females to males in primary school. Since there was no control for other variables, such as maternal schooling, the positive association could have reflected the fact that higher income levels are associated with improved maternal schooling, which in turn i s strongly associated with higher female relative to male child enrollments. The household analysis undertaken here controls for other variables. In addition, the dependent variable in this analysis is enrollment at the primary and secondary levels (not just the primary level, as was the case in Section B). 88A s an alternative, a state fixed-effects model was estimated, in which a full set o f state dummy variables replaced the six state-level variables (viz., log o f gross state domestic product per capita, log o f state government elementary and secondary expenditure per child 618, the ratio o f government expenditure on secondary schooling to that on elementary and secondary schooling, and interactions o f these three variables with sex o f a child). Although the state dummy variables were significant as a whole at the 5% level, the explanatory power o f the regression, as measured by a pseudo R-squared measure, did not change at all with the substitution o f the state dummy variables for the state-level variables (the pseudo R-squared measure remained at 0.27). This suggests that the state fixed effects model is not superior to the model reported here interms o f goodness-of-fit. 91 female than with male school attendance in the household. These results add to the significant amount o f evidence from around the world that adult female schooling improves the relative position o f girls within a household. 5.23 The results also show that the number o f elementary and secondary school per 1,000 children aged 6-18 i s associated more strongly with female than with male school attendance. Such a finding is consistent with female schooling being relatively more capacity-constrained than male schooling.89 On the other hand, there is no significant gender differential in the (inverse) association between pupil-teacher ratios and school attendance. 5.24 Finally, as would be expected, crime against women (as measured by the number o f cognizable kidnappings o f women and girls per capita) has a significant, inverse association with the school attendance o f females aged 6- 18 years. G. Simulations 5.25 Based on the multivariate analysis reported above, we have undertaken simulations o f the school attendance rate for males and females aged 618 years in the poor states under different intervention scenarios. The nature and magnitude o f these hypothetical interventions are shown in Table V.l. Figure V.15 shows the change in the projected male- female difference in the school attendance rate for 6- 18 year olds in the poor states from 1999 to 2015, under the assumption that all o f these interventions are pursued simultaneously. The gender gap in the school attendance rate i s projected to decline by about 3.6 percentage points (from about 13 percentage points in 1999 to 9.4 percentage points in 2015) - less than a third o f the distance to the MD goal o f a zero gender gap. Table V.l: Assumptions about various interventions to reduce gender disparity in primary and secondary school enrollment rate in the poor states, 1999-2000 to 2015 Starting Assumed Ending value in change per value in Intervention 1999-2000 year 2015 Adult male schooling (years) 4.5 0.25 8.5 Adult female schooling (years) 2.0 0.3 7.8 89 In other words, scarcity of schools in a community is more likely to constrain girls than boys from attending school. 90 As noted in chapter 4, an econometric study by Chin (2002) for India, using several rounds o f the NSS data, concludes that the teacher component of Operation Blackboard - a centrally-sponsored scheme launchedin 1987 that provided a secondteacher to single-teacher primary schools and apacket o f teaching- learning materials to all primary schools - significantly raised primary school completion and literacy rates for girls, but not for boys. Chin's results do not contradict those presentedhere, since she i s concerned with primary completion rates o f boys versus girls while we are concerned with the overall enrollment o f boys and girls aged 618. In addition, Chin analyzes the impact o f additional teachers per school, as provided by Operation Blackboard-not the impact of raising the pupil-teacher ratio. 92 Table V.l: Assumptions about various interventions to reduce gender disparity in primary and secondary school enrollment rate in the poor states, 1999-2000to 2015 Starting Assumed Ending value in change per value in Intervention 1999-2000 year 2015 Share o f secondary education intotal government expenditure on education (%) 36 1% 52 Consumption expenditure per capita (Rs.) 409 3% 656 Population coverage of electricity (%) 42.5 l%point 58.5 % ofvillages indistrict havingaccess to a pucca road(%) 59.5 1%point 76.5 Crime against women (no. o f female kidnappings and rapes per 100,000 pop.) 1.65 -0.05 0.85 No. o fprimary schools per 1,000 children aged 6-11 5.1 0.2 8.3 Pupil teacher ratio inprimary schools 91 -1.o 75 5.26 Figure V.15 shows the contribution o f each o f the proposed interventions to the change in gender disparity from the base year. Improved living standards (proxied by increases in both household and community consumption expenditure per capita) and greater road access are actually associated with increased gender disparity in attendance, as both o f these variables have a stronger positive association with male than with female attendance. The other five interventions are sssociated with reduced gender disparity. Gender disparity in school attendance is strongly associated with an expansion o f adult female schooling, with each year o f adult female schooling being associated with a reduction o f 8% in the gender gap in enrollment. An expansion in the share o f secondary education in public spending on education and an increase in the number o f elementary and secondary schools per child aged 6- 18 are also associated with appreciable reductions inthe gender gap inschool attendame. 5.27 It is interesting to Figure V.15: Projected changesin male-female difference (%age points) in school note that an expansion in attendance rate of children aged 6-18 in the poor states, 1999-2015, under different intervention scenarios(graph shows cumulative effectof each additional intervention) the number o f schools per 1,000 children is associated 9-1 t 9 not only with a reduction o f 6-I 6 the gender disparity in 3 3 school attendance but also 0 0 with an increase in overall school attendance (as -Expanding road access Decline in malerfemale discussed in chapter 4). -Expanding the share ofsecondalyeducationin govt exp on educ. -Expanding adult femaleschooling n needed Io MDG Further, as noted earlier in -'Expanding the numberof primaryschoolsper '000 children6-11 -Reducing crime againstwomen -Emandine electricitvcoveraee chapter 4, other evidence from India suggests that, at the margin, the poor benefit more than proportionately from an expansion in overall enrollments, since they are typically the last students to be 93 enrolled (Lanjouw a d Ravallion 1999). Thus, investing in additional schools would appear to be an important investmentfrom a number o f perspectives. In2002 alone, the Government o f India, under the universal primary schooling initiative, Sawa Shiksha Abhiyan (see Box IV.3), had sanctioned an additional 10,700 primary and elementary schools and an additional 62,000 Education Guarantee Scheme centers to be opened around the country (The Hindu, August 05,2002). 5.28 One issue that the simulation analysis in unable to highlight i s the role o f incentives in narrowing gender disparities. Since gender disparities in schooling outcomes are largely caused by parental discrimination against the girl child, public policies that increase the parental incentive to invest in girls, such as tuition waivers for girls and female stipends and scholarships, are likely to work well in narrowing the gender gap. An example of a policy intervention that has worked remarkably well in narrowing gender disparities in secondary school enrollments i s the Female Secondary School Stipend program in Bangladesh (see Box V.1). This program seeks to not only increase the enrollment o f girls at the secondary level butto also ensure that most o f them are retained until graduation from secondary school. Many states in India also subsidize the schooling o f girls invarious ways. 5.29 Finally, it is important to reiterate the message o f earlier chapters - viz., a simple expansion of school capacity or introduction of a female scholarship program will not work unless it i s accompanied by broad-ranging institutional reform to reduce teacher absenteeism and to make schools and school managers more accountable to students and the community (World Bank 2003). The reform will require, among other things, empowering communities and parents who can hold the state accountable for school performance, decentralizing educational decisionmaking, giving greater autonomy to schools, involving parents in school management, and motivating front- line workers via a systemo f incentives and disciplinary action. 94 Box V.1: BangladeshFemaleSecondary SchoolStipendProgram The attendance o f girls relative to boys in secondary schools started to grow at a record rate in Bangladesh after the government decided to exempt fees and give cash incentives to girl students, under the Rmale Secondary School Stipend (FSSS) program launched by the Government o f Bangladesh in 1994, with assistance from the World Bank, Asian Development Bank, and the Government o f Norway. (Unlike primary school, which i s free, secondary schooling requires pyment o f tuition fees in Bangladesh. In addition, households have to incur all other costs, such as transportation, books, uniforms, school supplies, and examination fees.) The program has been successfbl in its twin objectives o f increasingthe number o f girl students entering secondary school as well as keepingthem in school until graduation. With this program, Bangladesh has become a pioneer in increasing female secondary enrollments and in narrowing gender disparities at the secondary level among the nations o f South Asia. Although the project was initially implemented in 118 thanas in 1994, it was later expanded to all rural thanas in the country and converted to a national female secondary stipend program. Under the program, stipends covering full tuition, examination costs, and an increasing proportion o f school fees, textbooks, school supplies, uniforms, shoes, transport and kerosene (for lamps) are available to girls as they progress from Grades 6 to 10. The coverage o f other costs rises with grade because extra incentives are needed inthe upper grades to reduce highdropout rates. The project is also simultaneously increasing the number o f teachers - especially female teachers - in secondary school; providing occupational skills training to girls who are about to graduate; makingschools more attractive to provide a healthier and safer setting for girls; and strengthening government institutions for secondary education. 95 6. HUNGERPOVERTY 6.1 While the debate on poverty in India goes back several decades, it has focused almost exclusively on changes inthe 'headcount' ratio o f consumption poverty - viz., the proportion o f the population having monthly per capita consumption expenditure that is lower than the poverty line. Inthis report, we focus on hunger-poverty, as measured by calorie deficiency - the inability to consume the energy (calories) required by the body. We use NSS consumption survey data from the 50th and 55th rounds to calculate mean daily calorie intake per person, using food-to-calorie conversion factors obtained from Gopalan et al. (2000) and NSSO (1996b, 2001b). The calculation o f calorie norms or requirements i s much more complicated, and has been the subject o f a contentious debate. For instance, the average calorie norm o f 2,110 calories per capita per day prescribed by the FA0 for South Asia (FA0 1996) i s much lower than the 2,400 calorie norm that has beentypically used inIndia (Meenakshi and Vishwanathan 2003). We have used the normative age-specific calorie requirements recommended by the Planning Commission Task Force (GO1 1979), but with additional adjustments for heavy, moderate and sedentary work for individuals aged 19 years and older based on their occupations and operations (tasks performed). Operational and occupational data were obtained from the employment modules o f the NSS 50* and 55* rounds (see Dubey 2002). 6.2 The MDGs call for a halving o f hunger-poverty between 1990 and 2015. For India, this would mean bringing down the headcount ratio o f calorie Figure VI.1: Percentof populationthat is caloriedeficient, by state, 1999-2000 h deficiency from 62.2% in 1990 to "'E: 31.1% in2015.'l A P Goa T"*lrra Tam%% A. Trends and Patterns MahWaShm west!engal Karnataka B ! h 6.3 Levels and Trends. AuKRala lnma Hunger is pervasive in India. In MP AmchalP 1999-2000, for instance, more than %z hhzoram one-half (53%) o f India's h q a b Mmpw Nagaland population consumed fewer Hmachal P UP calories than it required. This is Ra,asthan J&Kashrmr nearly double the estimated zn M 42 50 a 70 so 91 The most reliable estimates are available from the 50th and 55'h rounds o f the NSS for 1993-94 and 1999-2000, respectively. The rate shown for 1990 i s projected from the change observed between 1993-94 and 1999-2000. 96 national incidence of consumption poverty o f 22-26%. 92993 6.4 There was, however, a substantial decline in the incidence o f hunger-poverty during the 1990s. In 1993-94, as much as 60% o f the Indian population did not consume the calories it required. The decline from 60% to 53% by 1999-2000 represents an annual decline of about 2% inthe proportion o f the population consuming insufficient calories.94 6.5 Interstate Variations. As with other indicators, there are large interstate variations in the extent o f hunger-poverty (Figure VI.l). Assam tops the list, with a staggering 78% o f the population being calorie deficient, followed by many o f the other Northeastern states. Goa and Andhra Pradesh also rank among the five states having the largest proportion o f calorie-deficient population. At the other extreme, Jammu & Kashmir, Rajasthan and Uttar Pradesh have the lowest rate o f calorie deficiency (30- 38%). 6.6 These numbers underscore the difference between the concepts o f consumption and hunger-poverty. Some of the states that rank low on consumption poverty, like Gujarat, rank high on hunger-poverty, while states that rank high on consumption poverty, like Uttar Pradesh, have low levels o f hunger-poverty. This suggests that reduction o f (consumption) poverty i s not synonymous with eradication o f hunger or an improvement in food security. The latter may depend upon such factors as local availability -o- f food, agricultural productivity, and food tastes and preferences o f the population. '' Figure VI.1: Percent change in the proportion of population that was calorie- deficient, 1993-94 to 1999-2000, by state 6.7 The rate at which Mizoram I -20 Manipur I hunger-poverty declined U P duringthe 1990s also varies Arunachal P M P Maharashtra significantly across states SkkU7l Nagaland (Figure VI.2). Uttar Onssa Rajasthan Kerala Pradesh as well as many of HimachalP Tnpura the Northeastern states, All India Meghala a such as Mizoram, Manipur, Karnataza Bihar TamilNadu Arunachal Pradesh and Goa I h l a b Nagaland, experienced , 2 ~, A P ; ~ large declines (greater than 10%) in hunger-poverty Haryana between 1993-94 and 1999- -25 -20 -15 -10 5 0 5 I O ~ 92 The official (Planning Commission) estimate o f poverty incidence for 1999-2000 was 26%, while Deaton and Dreze have estimated poverty incidence to be about 22%, using the same `official' poverty line but different inthe 55` round o fthe NSS. See Deaton andDreze (2002). Rrice deflators and adjustments for changes in the expenditure data collection methods introduced 93 Meenakshi and Vishwanathan (2003) also calculate `headcount' rates o f calorie poverty, but use fixed calorie norms o f 1800,2200 and 2400 calories per person per day. 94 According to official Planning Commission estimates, the headcount ratio o f consumption poverty declined from about 36% to 26% over the same period. However, using an altemative methodology, Deaton and Dreze report the decline to be from 29% to 22% (Deaton and Dreze 2002). Thus, the annual rate o f decline in consumption poverty was much greater (4.7 to 5.5%) than that in hunger-poverty. 95 See Behrman and Deolalikar (1988) for a discussion o f these concepts. 97 2000, while four states - Haryana, Assam, West Bengal and Jammu & Kashmir - actually experienced an increase inthe incidence o f hunger. 6.8 Intra-State Differences. Even within states, there are large differences in the incidence of calorie deficiency. For instance, Appendix Table 9 shows that the incidence o f hunger-poverty in 1999-2000 varied from 48% to 75% even within the state o f Gujarat. Surprisingly, the lowest rate o f hunger-poverty inthat state was found inthe arid areas o f the state, while the highest rate was inthe southem plains. As many as 8 out o f a total o f 78 regions in the country had 70% or more o f their population consuming insufficient calories. 6.9 Appendix Table 9 also reports regional data on the incidence o f hunger-poverty for an earlier year 0993-94). The table shows wide variation in the performance o f different regions inreducing calorie deficiency. A few regions instates as varied as Uttar Pradesh, Maharashtra and Rajasthan were able to reduce the incidence o f calorie deficiency by more than 40% during those six years. At the other extreme, four regions experienced an increase o f more than 25% incalorie deficiency over the same period. B. Socioeconomic Variations inCalorie Deficiency dence o f calorie deficiency across economic groups 80 - B (Figure VI.3). An over- 70- 64 whelming 82% o f the bot- 61 6 0 - 58 tom consumption quintile i s 51 52 49 calorie-deficient, as c o m 40- 40 pared to only 28% o f the top consumption quintile. 30 28 However, the differences in "- calorie deficiency across l o - social groups are signif% o-. cantly smaller. While STs Bottom Second Third Fomh Top ST SC OBC Forward Per capitaexpenditurequintile Socialgroup castes appear to have a higher rate o f calorie deficiency than Other Backward Castes. 6.11 Figure VI.4 shows the relationship between living standards, predicted calorie intake and predicted calorie requirements.96 At low levels o f monthly consumption expenditure per capita (Rs. 450 or lower), predicted calorie requirements exceed calorie intakes, resulting in high levels o f calorie deficiency. With increasing income, predicted calorie intakes increase rapidly - more rapidly than predicted calorie requirements - and this progressively lowers the predicted risk o f hunger-poverty. Figure VI.4 also shows that predicted calorie requirements are relatively low (but not lower than predicted calorie 96 Both calorie requirements and intake have been predicted on the basis o f a regression that includes the log o fmonthly per capita consumption expenditure and its squared term. 98 intake) for the very poor, Figure VI.4: Predicted calorieintake and requirementsper capita per day, 1999-2000 but plateau rather early at about 2,100-2,200 calories 3.5001 perpersonper day. 3,000 Predictcdc 2,500 6.12 The NSS data also 2,000 show a strong association between adult schooling 1,500 levels and calorie 1,000 deficiency (Figure VI.5). 500 Individuals in households " . where the highest-educated 0 500 IO00 IS00 2000 2500 adult male or female has Monthly per cap cons. exp (Rs) fewer than five years o f FigureVI.5: Percentageof populationthat i s caloriedeficient, schooling have the highest by adult male and female schoolingin household, 1999-2000 risk of calorie deficiency (57-59%). The risk declines 651 Q O 0 1 - 4 0 5 - 8 09-10 011-12 Q>12 60 rapidly with more schooling. No significant 55 gender differences in the 50 association between schooling and hunger- 45 poverty are observed. 40 C. Role of Infrastructure 35 30 6.13 Surprisingly, the Male education Female education data do not show much o f an association between hunger-poverty and electricity access (Figure VI.6). Access to safe drinking water i s also not observed to have a strong association with calorie deficiency. But better FigureVI.6: Percentageof populationthat is caloriedeficient, village access to pucca by infrastructure access, 1999-2000 (sealed) roads i s associated 6c 60 n with lower rates o f calorie r deficiency. In districts 55 where 75- 100% o f villages are connected by pucca 5c road, the incidence o f hunger-poverty i s 49%, as 45 compared to a levelof 60% in districts where none of the villages are connected. 4c No I Yes 0-49 I 50-74 I 75-100 0 ðer householdhas ?4ofhouseholds in distnct having % ofvillages in distnct connectedto a pucca 6.14 Figure VI.7 shows accessto electncity access 10safe dnnking water - the incidence o f calorie deficiency by various agricultural indicators. As would be expected, the risk o f hunger-poverty declines significantly with farm size; individuals residing in households cultivating 1.5 hectares or more o f land have calorie deficiency 99 rates that are only about two-thirds o f those for individuals residing in households with no land. In addition, access o f a household to irrigated land and food availability in a district are associated with a reduction in the risk o f calorie deficiency substantially. For instance, the incidence o f calorie deficiency i s as high Figure VI.7: Percentageof population that is calorie deficient, as 58% in households with by variousagriculturalindicators, 1999-2000 no irrigated land, while it i s 59 59 0"