009 LIBRARY
Michigan State
University

 

This is to certify that the
dissertation entitled

ESSAYS ON EVALUATION OF THE EFFECTIVENESS OF
PUBLIC HEALTH PROGRAMS ON CHILD’S WELL BEING:
EVIDENCE FROM INDONESIA FAMILY LIFE SURVEY

presented by

ELAN SATRIAWAN

has been accepted towards fulfillment
of the requirements for the

Ph D degree In Agricultural Economics

 

 

BMW Bw"\

Richard H. Bernsten

 

Major Professor’s Signature
5 I 5/1 0 01

Date

MSU is an Affirmative Action/Equal Opportunity Employer

..-.-.---.-.-.-.-.-.-.-.-.---.-.—-—.c-.—.-.—------.-o-—o----A.r---.—.- _.-.-.-.-.-.-._.-.-.-.-.-.-.-.-,-._.-.-,-.-._

 

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BERECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5108 ICIProj/AccarPreleIRC/Datoouejndd

ESSAYS ON EVALUATION OF THE EFFECTIVENESS OF PUBLIC HEALTH
PROGRAMS ON CHILD’S WELL-BEING: EVIDENCE FROM INDONESIA
FAMILY LIFE SURVEY

By

Elan Satriawan

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirement
for the degree of

DOCTOR OF PHILOSOPHY
Agricultural Economics

2009

ABSTRACT
ESSAYS ON EVALUATION OF THE EFFECTIVENESS OF PUBLIC HEALTH

PROGRAMS ON CHILD’S WELL-BEING: EVIDENCE FROM INDONESIA
FAMILY LIFE SURVEY

By
Elan Satriawan

This dissertation consists to two essays. The first essay evaluates the
impact of childhood nutritional status and the presence of a public health
program on subsequent child schooling in Indonesia during 19903. We estimate
dynamic relationship of childhood nutrition and subsequent child schooling in
which we carefully address the potential correlation between childhood nutrition
and important but unobserved factors such as child innate healthiness and
parents’ taste toward child quality. We find that reducing incidence of poor
childhood nutrition reduces also the probability of delayed enrollment, but not
the probability of repeating a grade. More importantly, the estimated effects
when taking into account the endogeneity of childhood nutrition are 5-7 times
stronger than when ignoring the endogeneity of childhood nutrition.

The effect of childhood nutrition on subsequent child schooling is even
higher if child has access to public health facilities. Looking particularly at
presence of midwife, we find that the presence of midwife magnifying the effect
of childhood nutritional status on subsequent child schooling. This result
suggests that the exposure to midwife during early childhood improved child

nutritional status that in turn helped child schooling.

The second essay evaluates the effectiveness of a supplementary feeding
program for maintaining child nutritional status through the period of the 1997-
1998 economic crises in Indonesia. We exploit heterogeneity in program
exposure to evaluate the program effect. The use of program heterogeneity has at
least two advantages for identifying the effect of Indonesia’s supplemental
feeding program. First use of program exposure allows us to estimate an effect
even with low variation of program distribution across targeted subjects. Second,
use of the program allows us to avoid the strong assumption that all targeted
children experienced homogenous exposure to the program.

We show that although the program tended to be universally distributed
during 1998—2000, the distribution of program length varied across communities.
We also provide insight into the government’s implicit allocation rule when
determining program length. Findings on the effect of program exposure show
that the program improved the nutritional status of children 12 to 23 months of
age during the period of economic crisis. Our findings also suggest that the
program helped children with severe malnutrition problems.

Our results also highlight the potential benefit of using heterogeneity in
program exposure when evaluating a universally targeted program. The use of
program intensity is not just helpful to capture program effects as it can also be
used to in conjunction with community fixed effects in a way that avoids bias

from endogenous program placement.

ACKNOWLEDGEMENTS

Al hamdu lillahi rabbi l-‘alamin, All praise is to Allah, the Lord of the
universe, for all of His blessings that makes all of these things happen and allows
me to complete this dissertation.

I would like to express my sincere gratitude to all persons that help to
make the completion of this task possible. First, I am deeply indebted to
Professor John Giles, my dissertation supervisor for all his guidance and very
kind support during the writing process of the dissertation. I have been greatly
fortunate to have him as advisor that treats me more as a colleague rather than
just student supervisee. I would also thank to my major Professor Richard
Bernsten for his support and guidance I have received since I joined the program
in 2003 until now. I also owe my gratitude to my external committee member,
Professor John Strauss of University of Southern California who always provides
me with promptly advises when I ask him. He has set a high standard for his
students and his sharp and critical advises have contributed a lot in shaping my
dissertation and the way I think about empirical work as it is now. I also want to
thank to my other committee members Professor Eric Crawford and John Staatz
for their support and guidance during my study time and dissertation writing

process.

iv

I am also indebted to some of friends and colleagues with whom I
intellectually discussed and debated many issues in the development areas.
Particularly I would like to thank to Firman Witolear, (late) Lesiba Bopape, Lilian
Kirimi, Denis N izalov, Shauffique Shidique, Merry Matenge, Anthony Chapoto,
Elliot Mehgenye, Abdoul Kareem Murekezi, Kirimi Sindi, Vandana Yadav,
Athur Mabiso, and Sarrna Aralas.

I want to also thank to Indonesian and Malaysian community for their
warm reception during our stay in East Lansing, particularly to Firman Witoelar
& Nina, Mas Diwan & Mbak Susi, Pak and Bu Ronnie, Haryono and Ria, Dwi
Agus, Reza and Nana, Shaufique and Nita, Azril and Bismi and other
community members that I can not mention one by one here.

Special thank also goes to member of Indonesia Islamic Study Group: Mas
Eka, Mas Ahmad, Mas Endra, Mas Danet, Kang Dena, Bang Torkis, Ristanto,
Didi and their respective family. They have become my family during my stay in
US and with them I manage to maintain my insanity and to deal with stress I
obtained from life as student.

Now it is turn to express my gratitude to my family back home. I deeply
believe that of the main ingredients of this achievement are unconditional love
and blessing from my parents and parents in-law. To my parents, Ibu and Bapak,
and my parents in—law Ibu and (late) Bapak Ronnie, I would never be able to pay

back all of those that you have given to me. Only Allah will grant you with much

better ones in this world and afterlife. Also to my brother, sister and sister in-law
and their family—Mas Ari & Mbak Miska, Mbak Esti & Mas Iprin, and Mb Ria &
Mas Mirza— sincere love and attentions you dedicate to me can lift-up my spirit
and motivation to get the job done as soon as I can.

Most importantly, none of this would have been possible without the role
of my wife and my children. My wife, Rennta Chrisdiana (Awie) has been
someone who is always there for me, provided me with the genuine love,
bottomless patience and unconditional support. My children, Lila, Haikal, Raisa
and Khalif, they have become a place where I can melt all depressed and
frustrated feelings, made my days become bright again after thousands time of
feeling down. I should say that completing this study and getting the degree is
not only my personal achievement. It is theirs too. It is my and their

achievement.

vi

TABLE OF CONTENT

LIST OF TABLES ............................................................................... viii
LIST OF FIGURES ............................................................................... xi
CHAPTER 1
INTRODUCTION .................................................................................. 1
CHAPTER 2

EVALUATING THE IMPACT OF EARLY CHILDHOOD NUTRITION AND
AVAILABILITY OF HEALTH SERVICE PROVIDERS ON SUBSEQUENT CHILD
SCHOOLING: EVIDENCE FROM INDONESIA FAMILY LIFE SURVEY

......................................................................................................... 7
2.1. Introduction ............................................................................... 7
2.2. Analytical Framework ................................................................. 15
2.3. Empirical Model and Identification ............................................... 19
2.4. Data and Sample Setting ............................................................... 26
2.5. Results ........................................................... '. .......................... 29

2.5.1. Impact of early childhood nutritional status on child schooling ...29
2.5.2. Impact of presence of midwife during early childhood on schooling
.................................................................................................................. 35

2.6. Conclusions .............................................................................. 40

CHAPTER 3

EVALUATING THE IMPACT OF EARLY CHILDHOOD NUTRITION AND
AVAILABILITY OF HEALTH SERVICE PROVIDERS ON SUBSEQUENT CHILD
SCHOOLING: EVIDENCE FROM INDONESIA FAMILY LIFE SURVEY

......................................................................................................... 56
3.1. Introduction ............................................................................. 56
3.2. Previous Evaluation Studies on Indonesia Public Health Program ....... 60
3.3. Analytical Framework ................................................................ 64
3.4. Nutritional intervention for children during economic crisis .............. 66
3.5. Data and variables measurement .................................................. 70
3.6. Empirical Identification and Specification ...................................... 75
3.7. Results and Analysis ................................................................. 79

3.7.1. Determinants of Program Duration ....................................... 79

3.7.2. The effect of supplementary feeding program .......................... 82

3.8. Summary and Conclusions .......................................................... 86
BIBLIOGRAPHY ............................................................................... 116

vii

LIST OF TABLES

. Table 2.1. Descriptive statistics of main variables ....................................... 43

Table 2.2. Impact of childhood nutrition on probability of delayed enrollment,
OLS ......................................................................................... 44

Table 2.3. Impact of childhood nutrition on probability of repeating grade, OLS
........................................................................................ 45

Table 2.4. Impact of childhood nutrition on probability of delayed enrollment:
2SLS ........................................................................................ 46

Table 2.5. Impact of childhood nutrition on probability of repeating grade: ZSLS
................................................................................................ 47

Table 2.6. Heterogeneity impact of childhood nutrition on probability of delayed
enrollment across gender, 2SLS-LIML ............................................. 48

Table 2.7. Heterogeneity impact of childhood nutrition on probability of
repeating grade across gender, ZSLS-LIML ...................................... 49

Table 2.8. Impact of midwife exposure on probability of delayed enrollment:
2SLS-LIML ................................................................................ 50

Table 2.8A. Partial effect of childhood nutrition on probability of delayed
enrollment conditional on the presence of midwife ........................... 51

Table 2.9. Impact of midwife exposure on probability of repeated grade: 2SLS-

LIML ....................................................................................... 52
Table A2.1. First Stage Regression: Height-for-age z-score ........................... 54
Table A2.2. First Stage Regression: being non-stunted (Yes=1) ...................... 55

Table 3.1. Distribution of PMT Coverage Across Communities, 1998-2000 ....... 89

Table 3.2. Standardized Height-for-Age and Incidence of Stunting: Children
between 6 and 59 months in 1997 8: 2000 ......................................... 90

viii

Table 3.3. Mean Child Exposure to Supplementary Feeding Program, Children 6-
59 Months in 2000 ...................................................................... 91

Table 3.4. Descriptive Statistics ............................................................... 92

Table 3.5. Determinants of Program Length across Communities 1997-2000
(pooled OLS) ............................................................................. 93

Table 3.6. The Effect of PMT on Height-for-Age of Children 6-59 Months: Binary
Program Indicator ...................................................................... 94

Table 3.7. The Effect of PMT on Probability of Stunting Among Children 6-59 of
Age: Binary Program Indicator ...................................................... 95

Table 3.8. The Effect of PMT on Height-for-Age of Children 6-59 Months:
Proportion of Age Exposed to the Program ...................................... 96

Table 3.9. The Effect of PMT on the Probability of Stunting Among Children 6-59
Months: Proportion of Age Exposed to the Program .......................... 97

Table A3.7.2.1. Impact of PMT on height-for—age of children 6-11 months: binary
program variable ...................................................................... 103

Table A3.7.2.2. Impact of PMT on height-for-age of children 12-23 months: binary
program variable ...................................................................... 104

Table A3.7.2.3. Impact of PMT on height-for-age of children 24-59 months: binary
program variable ...................................................................... 105

Table A3.7.2.4. Impact of PMT on probability of children (6-11 months) being
stunted: binary program variable ................................................ 106

Table A3.7.2.5. Impact of PMT on probability of children (12-23 months) being
stunted: binary program variable ................................................ 107

Table A3.7.2.6. Impact of PMT on probability of children (24-59 months) being
stunted: binary program variable ................................................ 108

Table A3.7.2.7. Impact of PMT on height-for-age of children 6-11 months:
proportion of child age exposed to the program .............................. 109

ix

Table A3.7.2.8. Impact of PMT on height-for-age of children 12-23 months:
proportion of child age exposed to the program .............................. 110

Table A3.7.2.9. Impact of PMT on height-for-age of children 24-59 months:
proportion of child age exposed to the program .............................. 111

Table A3.7.2.10. Impact of PMT on probability of children (6-11 months) being
stunted: proportion of child age exposed to the program ................. 112

Table A3.7.2.11. Impact of PMT on probability of children (12-23 months) being
stunted: proportion of child age exposed to the program ................. 113

Table A3.7.2.12. Impact of PMT on probability of children (24-59 months) being
stunted: proportion of child age exposed to the program ................. 114

LIST OF FIGURES

Figure 3.1. CDF of 1997 Community Average Height-for-Age by Location (Rural
vs Urban) of Community ............................................................ 98

Figure 3.2. CDF of 1997 Stunted Share of Children in Communities by
Community Location (Urban vs Rural) .......................................... 99

Figure 3.3. CDF of 1997 Average of Household per Capita Expenditure by
Community Location ............................................................... 100

Figure 3.4. CDF of Length of Program in Month by Community Location (Urban
vs Rural), 1998-2000 .................................................................. 101

xi

CHAPTER 1

INTRODUCTION

Governments in developing countries are spending increasing sums of
money to invest in human resources and in improving the quality of life of their
populations. Like public investments for infrastructure like the building roads
and bridges, schools, and hospitals, public investment in human resources are
now also supported through a rationale of providing public goods. Increasing
investment in human development is reflected in devotion of resources to recruit
teachers, medical doctors, midwifes, nurses as well as creating curriculum,
providing medicine to public health providers, and funding supplementary
feeding programs.

One irony in the use of such public resources is that governments neither
know whether these investments any effect on the welfare of the targeted people
as intended, nor what the magnitudes of these effects might be. The situation
becomes worse as the governments of developing countries often scale up the
public program without the sufficient and credible knowledge of program
effectiveness. We might argue that there would be a portion of the investment
that would benefit the poor and the needy. Without sufficient knowledge on how
the program might affect the people, this portion might be minimal and thus the

investment may not be socially efficient.

Designing and implementing a credible program evaluation is always not
an easy task. Other than the approaches that that have been developed recently,
they require data that are not widely available, including multiple cross-section
data, panel data, and data from randomized experiments. Therefore the absence
of credible impact evaluation for public programs in developing countries
reflects not only the absence of the skill to design and execute, but also the
scarcity of the resources.

Appropriate data are becoming more available, particularly survey data,
and some new evaluation methods are readily used now. Survey data are not
specifically designed for the purpose of conducting impact evaluation, but rapid
development of methods to implement impact evaluation has been empowering
researchers to utilize it.1 An alternative approach which is becoming common
involves use of randomized experiments. The approach has some advantageous
over the ones that use survey data particularly in dealing with selection bias.
However it also has some disadvantages: first, randomized evaluations require a
lot of resources if implemented in a large scale, and second, they often face
ethical issues when it the experiments relate with interventions to the poor.2 The
two essays on impact evaluation in this dissertation use survey data and focus on
evaluating policy interventions designed to improve the health and nutritional

status of children.

 

1 See Ravallion (2007) for detail review on various methods of impact evaluation.
2 See Duflo et a1 (2007) for motivation, toolkits and reviews of randomized experiments.

The importance of nutritional status and its economic implications have
been well-documented in the development literature. For the early age groups,
better nutrition for children is associated with lower risk in child mortality and
morbidity. Better nutrition at young ages also explains, in part, higher
productivity during later stages of growth. Studies have also found the
association between adult health and earnings [see for example Martolell and
Arroyave (1988), Behrman and Deolalikar (1988), Behrman (1996), and Strauss
and Thomas (1998)].

With a focus on developing economies, much of the attention in this
literature have been focused on the relation between the weather-related and
other shocks with the health and nutritional outcomes of individuals. Many
papers also have been written in this literature on how households might
respond to the shocks and on the effectiveness of the adopted shock-coping
mechanisms. A main result from these studies show that not all households,
particularly those that face liquidity constraints, can smooth consumption and
thus can maintain their health and nutritional status (see the survey in Dercon,
1993)

Nevertheless some important questions have received little attention in
this literature. For instance, while the current literature mainly focused on short-
run and transitory effect of the shocks, the question of whether and how shocks
may have long-run and permanent effects is relatively unanswered. In addition,

inadequate attention has been made to answer how heterogeneity in public

program ’intensity’ (which may include quality, duration or exposure) affects the
impact on individuals’ health.3 In addition, researchers and policy analysts have
only recently paid attention to the longer-run relationship between early
childhood nutrition with child other human capital aspects such as subsequent
academic achievement. There remains much to be done in these areas.

Our purpose here is to address such unanswered issues to complement
current literature. We divide this study into two essays. The first one, in chapter
2, examines the impact of preschool-age child nutritional status on child’s
subsequent academic attainment. In addition, taking into account the availability
and the quality of local health programs, the essay will evaluate the effect of
government health policy on academic attainment. Evidence of the positive
impact of early childhood nutrition on educational achievement would
document the importance of nutritional investment during childhood and
provide a justification for government intervention such as supplementary
feeding program. In addition, the evidence on the impact of public health
programs, if any, on health status of children during early age would provide
support for government policies providing this type of public investment.

The next essay, in chapter 3, evaluates the impact of a supplementary food
program for preschool age children in Indonesia on child nutritional status by

making use of heterogeneity in the duration of program exposure. The analysis

 

3 King and Behrman (2008) discuss the use of timing and duration of exposure on evaluation of
social programs and how it might generate different impacts than measurement with simple
binary program exposure.

makes use of heterogeneity program duration to evaluate the impact on child’s
nutritional status and thus makes an important contribution to the program
evaluation literature.

This study uses data from all publicly available waves of Indonesia Family
Life Survey (IFLS) up to now: 1993, 1997 and 2000. The first (1993) and second
(1997) wave of the surveys were carried out by RAND Corp of Santa Monica and
Demographic Institute. While the third (2000) wave was conducted by RAND
Corp and Center for Population and Policy Studies, Gadjah Mada University.
IFLS is panel / longitudinal data that collects rich information on the socio-
economic characteristics of individuals, households and communities where they
resided as well as public facilities they utilized. This brings us one advantage to
link household and community-facilities surveys enabling us to establish
connection between the outcomes or impact we expect to see at the individual
level with the policy change that is observed as changes in community and
facilities where these individuals lived.

Our general findings from first essay show that childhood nutritional
status was important for subsequent schooling outcomes, and differed in impact
for boys and girls. The results also imply that previous studies that do not taking
into account potential source of bias suffer underestimate the effect. In addition,
we find that presence of a village midwife in the community in fact magnifies the
positive effect of childhood nutrition on child schooling implying that a village

midwife complements nutritional status in early childhood.

Findings from second essay indicate that a supplemental feeding program
was effective but that its impact was limited only to infants under 24 months.
Children from older age groups were not affected by the program. We also find
that the exposure to the program reduced the likelihood of stunting within the

same age group.

CHAPTER 2
EVALUATING THE IMPACT OF EARLY CHILDHOOD NUTRITION AND
AVAILABILITY OF HEALTH SERVICE PROVIDERS ON SUBSEQUENT

CHILD SCHOOLING: EVIDENCE FROM INDONESIA FAMILY LIFE
SURVEY

2.1. Introduction

Due to the perceived importance of both health status and education of
children for both current well-being and their future productivity as adults,
much attention in both the research and policy communities has focused on early
childhood nutritional status and the enrollment of children in school. With
strong interest in these areas, studies documenting correlations between these
dimensions of child human capital and subsequent well-being as adults have
multiplied in recent years.4 In addition, empirical research has also attempted to
identify intermediate factors affecting the relationship between child health
schooling, and thus also affecting the outcomes of children as adults. The general
assumption about the direction of influence between child health and
educational investments is that child outcomes in school are more favorable with
improvements in early childhood nutrition.

While the findings from a substantial body of existing research suggest
that child nutrition is important for child schooling outcomes, many of these

studies suffer from serious bias and fail to establish a causal relationship between

 

4 For most recent review in related studies see Glewwe and Miguel (2007) and Strauss and
Thomas (2007).

child nutritional status and child schooling.5 The most important source of bias
stems from a failure to take into account the fact that both child schooling and
child health status reflect household decisions. Many studies estimate the effect
of early childhood nutrition on subsequent child schooling outcomes assuming
that there is no correlation between childhood nutrition and important
unobserved factors such as child innate ability, parent preferences toward child
quality, or parent gender preferences.

Although estimation controlling for this source of endogeneity may still
be sensitive to underlying household behavioral assumptions and the nature of
unobserved heterogeneity, research by Behrman and Lavy (1998), has shown that
estimates of the effect of childhood nutrition on child schooling are biased
downward when early childhood nutrition is treated as exogenous. Behrman
and Lavy (1998) further demonstrate that if estimation assumes that (i) child
health is correlated with unobserved individual, household and community level
heterogeneity such as genetic endowment, home study environment, or
availability of education facilities, and (ii) that if there are no unobserved inputs
into child cognitive development and prices can be used as instruments, then the
impact of health status on educational outcomes is three to seven times as large
as those when ignoring endogeneity of child health. The bias is even larger when

the second assumption is dropped.

 

5See the reviews in Pollit (1990), Behrman (1996) and Behrman and Lavy (1998).

Studies using only cross-section data frequently suffer from an additional
source of bias. Typically they estimate current period child nutritional status on
contemporaneous child schooling or use recall methods to measure past
childhood nutrition and estimate an effect on current period child schooling.
While it is difficult to argue that the parameters estimated from the former
approach can be used to establish a causal relationship between health and
schooling outcomes within the same period, parameters estimated using
retrospective information are likely to suffer from recall bias. Once we recognize
these concerns, it is difficult to imagine that causality between childhood
nutrition and child schooling can be established using cross-sectional survey
data.

To date there are four significant studies exploiting panel data which
examine the relationship between nutritional status and child schooling and also
address the methodological concerns noted above.6 Alderman et al (2001), use
panel data from rural Pakistan and find that child nutritional status affected
school enrollment, and that the impact was greater for girls than for boys. Their
preferred estimate employs a dynamic model and uses price shocks at the time

when children were of 5 years old as instrumental variables.7 Their results show

 

6See Alderman, Behrman, Lavy and Menon (2001); Alderman, Hoddinott, and Kinsey (2006);
Glewwe, Jacoby and King (2001), and Ghuman, Behrman, Gultiano and King (2006).

7The choice of price shocks as instruments avoids the strong identifying assumption that there is
no correlation between child height-for-age up to age two and after two years of age, as in
Glewwe et al (2001). Still, one might be concerned about the timing of the price shocks used. As
they also note, price shocks at age 5 might not adequately capture health status of children when
of preschool age. Strauss and Thomas (2007) also note the potential for long-term effects of shocks

that when one controls for endogeneity in child nutritional status, its effect was
three times more important for enrollment than when the model was estimated
without controlling for endogeneity in childhood nutrition.

Findings from Ghuman et al (2006) demonstrate the importance of pre-
school nutritional status (using height-for-age z-scores and hemoglobin levels)
for child enrollment in first grade. Using similar assumptions as Alderman et al
(2001), they instrument endogenous child nutritional status with characteristics
of day care centers in villages where children lived prior to elementary age. They
find that childhood nutrition has a significant and positive impact on school
enrollment, but in contrast with Alderman et a1, their instrumental variables
estimates suggest upward bias in parameters produced by OLS.8

Another strategy is used both in Alderman et al (2006) and Glewwe et a1
(2001). These two papers share a similar assumption that child nutritional status
is correlated with two important (unobserved) factors that also affect child
schooling performance: (i) the home environment provided by a parent and
affecting school performance; and (ii) the child’s health endowment, which affects
how a child performs relative to others in school. To deal with these

unobservables, they combine household (maternal) fixed-effect and instrumental

 

to affect current period household welfare, which would further complicate estimation of the
childhood effect. We consider this issue further in our empirical discussion of our instrumental
variables below.

8The inconsistency might be caused by weak instruments bias. As indicated in their results, F-
statistics for their excluded instrumental variable instrumental variable is only significant at 10%
level and it is not clear whether the results are robust to weak instruments bias.

10

variables estimation techniques. Alderman et a1 (2006) use negative shocks (from
war and drought) that affected children of preschool age to instrument for child
nutritional status. They find that better preschool nutritional status is associated
with more completed years of schooling. Glewwe et al (2001) use height-for-age
for older siblings and differenced month of birth dummy variables as
instrumental variables. They find that undernourished children entered school
later and performed more poorly on cognitive achievement tests relative to
better-nourished children.

This paper has two objectives. First, we add to the existing literature
examining the relationship between early childhood nutrition and subsequent
child schooling using a unique dataset collected in Indonesia. In common with
the four studies discussed above, we use an instrumental variable technique to
estimate the effect of childhood nutrition on the probability of delayed
enrollment or repeated grade. We use rainfall shocks occurring in utero for
sampled children along with other household and community variables to
identify children’s nutritional status during pre-school age. We argue that shocks
experienced prior to birth are relevant for determining both height-for-age z-
scores and stunting as measures of long-term malnutrition. Contemporaneous
shocks, such as price shocks, are unlikely to generate an appreciable effect on
long-term measures of nutritional status such as height-for-age z—scores.

In addition, as health and nutritional status may as well be affected by

government policy as well as parental choice and health shocks affecting the

11

innate healthiness of children (Glewwe 2005), we evaluate how exposure to
community-based health service providers during early childhood affected the
influence of health status on educational outcomes. In particular, we examine
whether exposure to village midwives alleviates the effect of low early
nutritional status in early childhood on subsequent school enrollment, or
alternatively, whether presence of midwives complements the benefits of early
nutritional status for school enrollment. At present, there are few studies that
directly link past experience of malnutrition, exposure to public health programs
and subsequent socio-economic outcomes.

Results of our analysis have important policy implications regardless of
the estimated effect. Presence of a village midwife may reduce the effects of
negative shocks to early childhood nutrition on subsequent school outcomes and
assist with recovery from the effects of shocks to health status during early
childhood. Alternatively, if midwives simply complement the effects of good
health status, then they may still be playing an important role in maintaining
health status and facilitating school enrollment, but this result would underline
the importance of finding other means to support early childhood nutritional
status. We use presence of a village midwife in the community when children
were of pre-school age to represent child exposure to community-based health
services, and interact this indicator with our measure of childhood nutrition. We
then include community dummy variables in this intent-to-treat approach to

control for features of the community correlated with placement of a midwife.

12

This approach allows us to identify how presence of a midwife influences the
effect of early childhood nutrition on subsequent enrollment while avoiding bias
from endogenous placement of a midwife in the village.

Our analysis examines another dimension along which village midwives
may play an important role for influencing outcomes. Earlier studies
demonstrated the important role played by village midwives in improving the
health of prime age women (Frankenberg and Thomas 2001) and of pre-school
age children (Frankenberg et a1 2005).9 We argue that if there is link from
childhood nutrition to child schooling, then the presence of a public health
program that improved child health could also have an important impact on
child-schooling.

Our analyses make use of panel data from three waves Indonesia Family
Life Survey (IFLS). This ongoing survey provides a rich source of information on
individuals and households, as well as their access to facilities and the
characteristics of the communities where they reside. In particular, we will link
the childhood nutritional status of children between 6-59 months in 1993 with
their schooling in 1997 (for an older group, who were 3 to 4 years old in 1993)
and in 2000 (for a younger group, who were up to 2 years of age in 1993). We
further link sampled children with presence of a midwife in the community

where they lived in 1993 to evaluate how exposure to a village midwife prior to

 

9Extensive discussions of the village midwife program can also be found in Frankenberg and
Thomas (2001)

13

five years of age interacted with early childhood nutrition influenced school
enrollment. In order to identify nutritional status, we also exploit historical
rainfall data (Kirono, 2000, and Kirono et al, 1999) from the month and year
when the sampled children were conceived.

Our results show that childhood nutritional status reduced the probability
of delayed enrollment, but not the probability of repeating a grade. The effect of
childhood nutrition on subsequent schooling is greater for boys than for girls.
Our preferred specifications produce estimates that are 5 to 7 times larger than
when estimates ignore the endogeneity of childhood nutrition. In addition, we
find that presence of a village midwife in the community in fact magnifies the
effect childhood nutrition on child schooling implying that a village midwife
complements nutritional status in early childhood. We also perform simulations
to show the likely effects of increasing the share of communities with midwives.
The result shows further improvement in the role of child health and nutrition in
improving child schooling. This implies that exposure to village midwife,
particularly during childhood, might be used as policy instrument to reduce gap
in child schooling through improvement of child health and nutrition.

This paper thus contributes to the literature on impacts of early childhood
nutrition in two significant ways. First, this study adds to existing studies which
control for the endogeneity of childhood nutrition when estimating its impact on
child schooling. In particular, we utilize exogenous rainfall shocks as source of

identification for nutritional status during childhood. Second, we examine how

14

placing health service providers in Indonesian communities affects the
relationship between early childhood nutritional status and subsequent
schooling outcomes. We thus estimate whether the importance of early
childhood nutrition status for subsequent school enrollment is affected by
presence of community based health service providers. Our results provide an
indication of how presence of health providers may interact with child health
status to improve schooling outcomes.

The remaining of the paper is organized as follows. In section 2, we
provide a simple framework to explain analytically how childhood nutrition may
affect child schooling conditional on other variables. We also show how
government policy may determine child schooling through a change in health,
health environment and health prices. In section 3, we propose an empirical
model and strategy to identify the effect of early childhood nutrition on
schooling, and then extend our discussion to examine how presence of a village
midwife interacts with early childhood nutrition to affect child schooling. We
next discuss data, the community setting, and concerns with the data in section 4.

In section 5, we present and discuss our results and conclude in section 6.
2.2. Analytical Framework

This section presents a simple analytical framework to model the

relationship between past child nutritional status, availability of a health service

15

provider and child schooling.10 We adopt a two-period analytical framework
developed by Glewwe (2005, 2007). The first period is a preschool stage in which
children are younger than 5 years of age. Clinical nutritionists argue that this
period, and in particular the period between 6 and 24 months, is when
nutritional interventions and supplemental feeding are likely to have the greatest
impact on subsequent child biological and cognitive development. If parents and
the health service provider have knowledge of this relationship, we expect
greater investment in child health during the period. The second period occurs
when children are of primary school age, or age 5 and older. During this period,
and conditional on child health and nutritional status, parents and government
invest in child education, and achievements in schooling may be related to child
nutritional status in the first period.

We start with production function for child schooling in period two which
is specified as a function of child and parent characteristics during both periods

one and two:

S2 = A(HI,H2,EII,E12,¢, SC) (1)

where S is a child schooling outcome, H represents health and nutritional status

during period i (i=1,2), El is (parental) education input at period i (i=1,2), (II is

(unobserved) innate child ability (e.g., intelligence, motivation), and SC are

 

10Our discussion in this section draws inspiration from Glewwe (2005).

16

school characteristics. Equation (1) focuses on assessing the role of child
nutritional status on child schooling while holding other factors constant. This
relationship is expressed as a structural equation since it only includes variables
that measure direct effects of each right hand side variable on child education.

Academic achievement, as highlighted by (1), is important but not our
main object of analysis. It is difficult to estimate because child academic
attainment and some other factors such as El are endogenous as they are under
the control of parents and thus reflect parental preferences toward children’s
education as well as health.‘1 School characteristic, SC, are also potentially
endogenous since parents can choose the school their children attend and the
government can decide by how much to invest in school quality. In addition,
other important factors, such as child innate ability and school inputs, are
unobserved. Our objective here is to evaluate the effect of child nutritional status
and health policy on child schooling using conditional demand functions for
child education. By estimating conditional demand, we avoid some of the
complications arising when estimating the effect of childhood nutrition on child
schooling using the schooling production function described above.

We derive the conditional demand function for child education by first

substituting the endogenous independent variables, other than nutritional status,

 

‘1 Assuming that household face resources constraint, allocating resources toward educational
inputs should affect those toward children’s health inputs.

l7

with relevant exogenous variables. The reduced-forms for the education inputs

El in each period 1' (i=1,2) are specified as:12

E1l = ei,(Y,Med,Fed,PS,0',a;PH,,HE,,2',77) (2)

1312 = ei, (H, , Y, Med, Fed, PS, a,a;PH,,HE,,r,n) (3)

where Y is parental income, Med and Fed are mother and father education,
respectively, P3 are prices of schooling and educational inputs, ois parental
preference toward child schooling, PH are prices for health in each period 1'
(i=1,2), HE is health environment, 2' is parents’ preferences toward child health,
and 17 is the innate healthiness of the child.

We next substitute equations (2) and (3) into equation (1) to yield the

conditional demand function for child schooling as:

S2 = aCD(H,,HZ,Y,Med,Fed,PS,SC,0',a;PH,,PH2,HE,,HE2,r,n) (4)

 

‘2 As discussed in Glewwe (2005), one complication arises when specifying the reduced form for
child educational inputs: it is reasonable that child health and nutritional status may enter into
parent decisions about the education inputs of their children. This creates a problem particularly
since current health status is endogenous. To avoid this complication, researchers sometimes use
previous period health status in current period reduced form education demand functions to
avoid this complication.

18

This function is still not fully in reduced form because it includes current
nutritional status as an endogenous variable in addition to exogenous variables.
This specification picks up the direct impact of changes in health status (as in
equation 1) as well as an indirect effect when change in health affects other
variables before the impact on change in child schooling, S. For example part of
the impact of good health in period 1 operates indirectly through parental
education input in period 2, as parents respond to health status by increasing or
decreasing their inputs into child education.

In addition, equation (4) can also be used to identify the impact of
government policies in health and education on child schooling. The impact of
health policy can be characterized as a change in either health prices (PH) or
health environment (HE). In particular, the impact of health policy is channeled
through two pathways. First, a health pathway includes the direct impact of
child health on children cognitive development through equation (1), and
indirect impacts of (lagged) child health on parental education inputs through
equations (2) and (3). Secondly, there is a reallocation pathway through which
substitution and income effects of PH and HE influence parental education

through equations (2) and (3).

2.3. Empirical Model and Identification
We first focus on estimating the effect of childhood nutritional status on

subsequent child schooling. Our empirical model of the conditional demand for

19

child schooling is represented by a dynamic model of the impact of nutrition on

subsequent schooling:13

52: = a + ,6le 1‘5,er + Y'ch + Alzzc + V2: (5)

where S is child schooling, H is nutritional status in period one when children
are 5 years old or younger, X is vector of household characteristics (including,
among others, income, mother’s education and father’s education), P is a vector
of prices, Z is a vector of community characteristics, which might have an effect
on child school enrollment, and v is a disturbance term. Numbers in subscript
indicate period of realization for each variable. While letters in subscript, i, h and
c, each indicate individual, household or community level variables.

As discussed in the previous section, current health status is supposed to
be included in equation 5. Our primary interest in equation 5, however, is to
measure the impact of early childhood nutritional status on subsequent
schooling outcomes. To do so, we need to exclude the current health status from
the schooling outcomes equation as we assume that the impact of past nutritional

status on current schooling is through current health status. Otherwise the

 

13 For our conditional demand for child schooling to be consistent with equation 4, we should
include current nutritional status in the empirical model shown in equation 5. As discussed by
Behrman and Lavy (1998) and Alderman et al (2001) the coefficient on current health status in
relation to child schooling is difficult to interpret as the estimated impact of current health on
current child schooling. We therefore dropped current health status from right hand side of
equation 5 and focus on estimating the impact of previous period (period 1) nutritional status on
current (period 2) child schooling. Our identification strategy must be robust to potential biases
introduced by unobserved nutritional status.

20

impact of past nutritional status on schooling will be absorbed by current health
status. In addition, including current health status in the schooling outcomes
equation introduces another endogeneity problem when estimating equation 5.
Estimating equation 5 with OLS is likely to produce bias in parameter [3 as
nutritional health status in period 1 is correlated with unobserved time-invariant
innate child health as well as parent preferences toward child health captured in
v. Some studies such as Alderman et a1 (2006), Glewwe and King (2001) combine
maternal-household fixed-effects with instrumental variable techniques to
address unobserved heterogeneity at parent (household) and individual levels.
Alternatively, as discussed in Glewwe et al (2001) and Alderman et a1 (2001), if
we can find shocks (price or weather) that (i) are of sufficient magnitude to affect
child stature but not their siblings, (ii) vary sufficiently across households (or
even individuals), we can use such shocks to identify childhood height-for-age as
it addresses unobserved heterogeneity at both individual and household levels.
We follow an identification strategy similar to that of Alderman et a1
(2001) and Alderman et al (2009), but differ with them in our choice of the timing
of shocks during childhood used to identify childhood nutrition. We argue that
price shocks that were measured contemporaneously with early child nutrition
indicator may not be sufficient for explaining the child height -which is used to
measure early childhood malnutrition. This is because any type of shock, will
likely take considerable time to be reflected in child height. Alderman et al (2001)

acknowledge this concern although argue that as long as such

21

(contemporaneous) price shocks are still partially correlated with childhood
nutritional status, we can still use them to identify endogenous childhood
nutrition.

We make use of long-term historical rainfall data, and use in utero rainfall
shocks, particularly those during the mother’s second and third trimester of
pregnancy, and interact them with child age (in months) when height was
measured. The interaction between in utero rainfall shocks and child age is
employed to improve the instruments’ power in identifying child nutritional
status by exploiting the length of time from the occurrence of the shocks to the
time when child height was measured in 1993. The identifying assumption is that
in utero rainfall shocks and time to the period when height was measured have
no impact on subsequent child schooling except through childhood nutrition. In
addition, we also include the value of total household assets, the height of father
and mother and interactions of number of posyandu and child age, all measured
in period 1, as additional excluded instruments.

There are a few potential concerns with these instruments. First, past
rainfall shocks might have had large enough magnitudes to have long term
impacts on household assets or consumption which thus directly affects child
schooling. Alternatively, prior rainfall shocks might have led to disasters, such as
floods that adversely affected long term household access to education facilities
and thus also to later child schooling. Such concerns, if not addressed, would

cast doubt on the validity of the instruments. To deal with concerns about affects

22

on wealth, we control for per-capita expenditure measure and include a village
dummy in the second stage regression so the effect of past rainfall shocks on
education, through household wealth, would be indirect and operating through
these variables. Similarly, one might be concerned that parents’ height (from
period 1) should not be in excluded instruments as they might affect subsequent
child education. But we argue that the effect of parents’ height on child
education would be conditional on child health. Thus, including a child height
indicator in the second stage regression should take care of this concern.

In addition, as the model suggests, prices and some community level
variables also determine child schooling outcomes. So we include in the equation
some food prices and community-level fixed effects. We argue that after
implementing this procedure, H 1 i is no longer correlated with omitted variables
in error term and thus [i is unbiased estimator of the impact of childhood
nutritional status on primary school enrollment.

We then seek to evaluate the effect of presence of midwife in the
community during childhood in the same schooling outcomes conditional on
past nutritional status. Our approach is to estimate intent-to-treat effect of the
village midwife. The important role of midwives has been identified for several
outcomes such as womens health (Frankenberg and Thomas 2001) and young
child nutritional status (Frankenberg et al 2005). Nevertheless none of previous
studies have examined how exposure to community-based health service

providers, such as midwives, might interact with early childhood nutritional

23

status to affect child schooling outcomes. We attempt to establish a potential link
between exposure to a village midwife when children are still young (under 5
years old) and subsequent enrollment when of school age. Specifically, we are
interested in how exposure to a midwife interacts with nutritional health status
in early childhood. Exposure to midwife might reduce the negative effect of past
malnutrition on subsequent school enrollment if village midwives provide health
services that compensate for in utero shocks affecting nutritional health status.
Alternatively, we may find that presence of a midwife reinforces the benefits of
better early childhood nutritional status. We thus want to estimate the following

equafion:

52: : a+T+fllHli 'l'flzllfidlgc3 +£le *Mdl): +8er +732: +V21 (6)

where MM” is an indicator for a presence of midwife in the community where a
child resided in period 1, or when they were of pre-school age (in 1993). We
estimate equation (2) using instrumental variables methods by employing shocks
at the early age of life as instruments for early childhood nutritional status. In
addition to those we discussed when estimating equation 5, one concern is that
presence of midwife in period 1 is likely to be correlated with some omitted

variable, such as availability of education facilities within the community, that

24

might affect subsequent school enrollment.“ If this is true, ignoring such
correlation will lead to bias in the parameter of interest. We thus include a fixed

community effect, jIc , and rewrite (6) as below:

$2: =a+T+IBIHII +fl2Hli *Mdl? +6'x2r'l’7'821‘l'flc +V2: (7)

Note that when we include in to control for (potential) endogenous midwife
placement, the midwife dummy can no longer be included directly as it will be
perfectly collinear with the village fixed effect. By controlling for community
fixed-effects, we control for endogenous placement of midwives by 1993.
Community fixed-effects are also useful as they control for unobserved
elementary school characteristics as we assume that the sampled children would
go to the closest school to the community.15 In this way we assume those schools’
characteristics are fixed across periods.16

The exclusion of a midwife variable due to inclusion of a community
fixed-effect means that we will not be able to estimate the direct effect of midwife
on schooling outcomes. Fortunately that is not our main interest. Instead we
would like to see how exposure to a midwife affects the schooling outcomes

through early childhood nutritional status. To do so, we will focus on the

 

1“ For example, Pitt et a1 (1993) find that the presence of one type of facilities was correlated with
another type of facilities.

15 Although IFLS interviews elementary schools, it is not possible to link every child in the
sample to the school he or she attended without making arbitrary assumptions.

1" The use of community fixed-effect here also implies that we should focus only on the children
that lived in panel communities during the observed periods.

25

interaction between early childhood height-for-age and dummy variable for the
presence of a midwife in community during period when child was in pre-school

age.

2.4. Data and sample setting

This study uses three waves of panel data from Indonesia Family Life
Survey (IFLS) 1993, 1997 and 2000. IFLS is a panel survey which collected very
rich socio-economic information on many aspects of individual lives and
households as well about characteristics of communities where those individuals
and households resided. The detail description about sampling in each of the
three survey waves is provided in Strauss et a1 (2004), Frankenberg and Thomas
(2000), and Frankenberg et al (1995), respectively.

Our sample in this study includes children who were between 6 and 59
month in 1993 (born between 1988 and 1993) and have their height (or length)
measured in the 1993 survey. We then link the 1993 measured nutritional status
for children who were born between 1988 and 1990 (with their enrollment status
in 1997 and for those who were born from 1991 to 1993 with their primary school
enrollment in 2000 then pool those two data files. In this set up, we thus examine
the impact of early childhood nutritional status on schooling of children who
were between 7-9 years of age in 1997 and 2000. The IFLS data also allow us to
identify how midwife exposure, when the children were 5 years old or younger,

interacts with early childhood nutrition in affecting enrollment. As mentioned

26

above, we expect that the exposure to midwife health services during such an
early age might either reduce the negative effect of malnutrition during
childhood, or alternatively complement the health endowment of children who
were not suffering from malnutrition.

Historical rainfall data are obtained from Kirono (2000) and Kirono et a1
(1999) which collect rainfall from 62 weather stations across Indonesia from 1960
to 1999. For our purpose, we use calculate shocks using rainfall data from the
entire period spanning 1960 to 1993, and then calculate shocks for the period
when our sampled children were born. From the data, we generate monthly
rainfall shocks which we define as monthly rainfall deviations from long-term
monthly rainfall trends and presented as monthly shocks.

The summary statistics of key variables are in table 1 below. All variables
are measured in period 2 (1997 and 2000 surveys, pooled) except for two main
variables of interest, nutritional status and presence of midwife in the
community, and instrumental variables that are measured in period 1 (1993
survey). We look at two child schooling measures, whether child experienced
delayed enrollment when they entered primary school and whether they ever
repeated a grade, both are expressed as binary dummy variables.17 There are 18
and 10 percent of children who experienced delayed enrollment and repeated

classes respectively.

 

‘7 For Indonesia, the two indicators appear to be important measures for primary education
particularly after reaching near universal enrollment for primary education.

27

For measures of nutritional status (measured by height-for-age) we use
(continuous) height-for-age z-score and (binary) whether a child was non-
stunted which is a measure of child malnutrition.18 The health literature has
suggested height (or length for baby and infant) is an indicator with less
measurement error for child health and nutritional status relative to the other
health measures.19 More importantly, the use of height-for-age will fit with the
focus of this study that investigates the long-term relationship between early
childhood nutrition to subsequent child schooling. The 1993 IFLS data show that
children younger than 5 of age have heights that are on average 1.39 standard
deviations lower than those of children with similar age and gender in US.
Meanwhile nearly 29 percent of children in the 1993 sample were exposed to
village midwives.

We also control for other covariates that may affect the household decision
to send children to elementary school and in part might represent parent
preferences for child education, the home study environment and the
intrahousehold allocation process. We control for parents’ education, age of
household head, household composition, and per capita expenditure. In
addition, as the model suggests, we also include current price for some food
items (rice, sugar, cooking oil and condensed milk). We also include a time

dummy to control for any secular trend across two different periods of life (1 for

 

13 Stunted is defined when child height-for-age z-score is less than negative 2.

19 In IFLS, height and weight are measured by special trained health workers with regularly
calibrated health equipment. For this reason, we believe that measurement error is negligible in
this case.

28

school age period, 0 otherwise). The mean and standard deviation for each of

these variables are presented in table 1.

2.5. Results

In this section we present and discuss the results. We first focus on the
impact of childhood nutritional status on subsequent child schooling outcomes.
The results, consistent with some previous studies, show that childhood
nutrition matter in determining subsequent child schooling. Our preferred
estimates also indicate stronger effect of childhood nutrition on subsequent
schooling implying biased results when correlation between nutritional status
and omitted variables captured in error term are ignored. We then try to seek
potential effect of the presence of village midwife in the community where those
children resided when they were in pre-school age. In particular, we are
interested in whether such exposure might reduce the negative effects of

malnutrition which occurred earlier in childhood.

2.5.1. Impact of early childhood nutritional status on child schooling

We estimate equation 5 and firstly ignore the correlation of childhood
nutritional status with unobserved heterogeneity such as child health
endowment and parent preferences for child quality. The result in table 2 and
shows that improved childhood nutrition lowers the probability of a child

experiencing delayed enrollment. The size of the coefficient when not controlling

29

for community fixed-effect suggests that an increase in child height-for—age by
one standard deviation lowered the probability of delayed enrollment by 3.2
percent. When controlling for community fixed-effects, the effect became slightly
stronger, where an increase in child height-for-age by one standard deviation
reduced the probability of delayed enrollment by 3.5 percent. The effect of other
covariates appears to be consistent with literature on determinants of child
schooling. Increasing parent, and in particular father's education, reduces the
probability of delayed enrollment. Per capita expenditure is also significant and
has a negative sign as one would expect. Household composition variables are
also important for child schooling but with different signs of influence on child
schooling. Number of children (age 6-14 years old) residingin the household has
the disadvantaged of delaying primary school enrollment. This might imply that
there is competition among for household educational resources. In addition,
number of male adults (age 15-59 years old) in the family reduces the probability
of a child experiencing delayed enrollment, but this occurs only when we are not
controlling for community fixed-effects.

Using a dummy indicator for stunting provides further support for our
findings. Suffering from stunting makes the probability of delayed enrolment
increase by 8 or 9 percent depending on whether we control for community
fixed-effects. Other covariates show similar direction of influence on child
delayed enrollment compared to when we do not control for the community

fixed-effect.

3O

Table 3 shows result from estimating childhood nutrition on probability of
repeating grade using a model that fails to control for endogenous health status.
Estimation results for each outcome using different specifications (with and
without community fixed-effects) show insignificant and inconsistent sign of
childhood nutrition effect on probability of grade repetition. Some other
covariates however remain significant with consistent sign as in the previous
estimation.

Although the sign of the parameters of interest appear to be consistent
with theoretical model, the estimations using treating health status as exogenous
will be biased for the reasons we discussed earlier. Behrman and Lavy (1998)
note that the direction of the bias from this na'l've model, whether upward or
downward, depend on the nature of intrahousehold allocation process. To
address this issue, we adopt instrumental variable technique in estimating the
effect of early childhood nutrition on child schooling.

We first examine effects of early childhood nutrition on probability of
delayed enrollment and present results in Table 4. We first look at the bottom
panel of table 4 where some statistical test results for instrumental variables are
provided. The tests are conducted both for measures of childhood nutritional
status, height-for-age and stunting status, and each are conducted for
specifications with and without community fixed-effects. As we see, the F-test for
excluded instruments for both endogenous regressors, height-for-age and

stunting status, produce a significant statistic at the 1 percent confidence level.

31

Recent econometrics literature on instrumental variables suggests that these test
statistics are not sufficient. Weak instrument bias may be present when there is
non-zero but small correlation between endogenous regressors and excluded
instruments although F-statistics of the excluded instruments are significant.20

We thus perform additional tests which include: (i) Kleibergen-Paap rk
LM statistics, to test the relevance of the excluded instruments on the
endogenous regressors (under Ho: equation is underidentified), and (ii) Hansen]
statistics which test over-identifying restriction (under joint null hypothesis that
the excluded instruments are valid). These tests show that the excluded
instruments are strong and valid for the endogenous regressors whether or not
we control for community fixed-effect in the equation. We also present the first
stage regression results in the appendices.

Results from estimating the impact of childhood nutrition on child
schooling using ZSLS are in top panel of table 4 (delayed enrollment) and table 5
(grade repetition). We start with the effect of childhood nutrition on delayed
enrollment in table 4. In terms of the direction of the influence, the effects of
childhood nutrition from 2SLS estimation are mostly consistent with results
shown in table 2. With few exceptions, other covariates are also consistent with
those in table 2. Controlling for endogeneity in childhood nutritional status

makes the effect of early childhood nutrition on probability of delayed

 

2" see Wooldidge (2002) for theoretical implications for this problem and Bhaum et al (2003, 2007)
for practical strategy to deal with this.

32

enrollment stronger, implying downward bias in the parameter estimated using
OLS.

Without controlling for community fixed-effects, an increase in childhood
height-for-age by 1 standard deviation lowered probability of delayed
enrollment by 9.7 percent (column 1). Non-stunted children are 31.3 percent less
likely to experience delayed enrollment relative to stunted children (column 3).
However results from this specification may still be biased if there is omitted
heterogeneity at the village level, including school characteristics that are likely
correlated with child schooling. It is also possible that school access and quality
differ across cohorts, particularly as the 1997/1998 economic crisis may have led
to deterioration in the quality of schools. To handle these issues, we include
community fixed-effects in the IV model.

As shown column 2 of table 4, the effect of controlling for community
fixed-effects makes the effect of childhood nutrition on child schooling even
stronger. An increase in child height-for-age in period 1 by 1 standard deviation
reduced the probability of delayed enrollment by 15.8 percent. In addition, non-
stunted children are more likely to avoid delayed enrollment relative to stunted
child by about 54 percent. As we have addressed most of potential problems,
these results, we argue, are relatively unbiased compared to those from models
failing to control for endogeneity of childhood nutritional status and community

fixed-effects.

33

When using grade repetition as the school outcome, we also find a
negative sign of the effect childhood nutritional status on probability of
repeating grade (table 5), we fail to reject the null hypotheses that the effect of
childhood nutrition equals to zero. As this is consistent with the result from OLS
estimation in table 3, one possible explanation why we found no childhood
nutrition effect is that grade repetition is rare (the mean sample value of grade
repetition is 9.6 percent). This probably occurs because we use young school age
children who were 7-9 years old by the time we observe their schooling outcome
and therefore the occurrence of grade repetition was not as high as for older
children.

We next examine whether the effect of childhood nutrition on subsequent
child schooling differs between boys and girls. We focus on delayed enrollment
outcomes as we do not see significant effect of childhood nutrition on grade
repetition. The results are presented in table 6.

As we can see in the bottom panel of table 6, F-statistics of excluded
instruments are lower than 10 although they still maintain significance at the 1
percent level. To address concern over the possibility of weak instruments, we
estimate the effect of childhood nutrition on delayed enrollment based on gender
using instrumental variable-limited information maximum likelihood (IV-LIML).
Baum et al (2007) show that IV-LIML estimation is robust in the presence of weak

instrument.

34

Result in table 6 shows that childhood nutrition mattered for boys more
than girls. Particularly 1 standard deviation increase in height-for-age z-score
lowers probability of delayed enrollment by 21.3 percent for boys compared to
12.4 percent for girls. Using an extreme measure of malnutrition, non-stunted
boys have 66.1 percent chance of enrolling on time relative to stunted boys. For
girls, being non-stunted improves probability of enrolling on time by 42.7
percent relative to stunted girls.

These results therefore suggest the importance of childhood nutrition for
child school enrollment decisions when they about to reach school age, but not
for child grade repetition. These are consistent with previous findings and
therefore reinforce the need for investment and intervention to prevent
malnutrition at a very young age. Consistent with previous studies (see
Alderman 2001, Behrman dan Lavy 1998, Glewwe et al 2000, and Glewwe and
King 2001), the results also show that the estimated effect generated by OLS may
suffer a substantial bias and which may mislead policy makers in addressing the

problems related to early childhood malnutrition.

2.5.2. Impact of presence of midwife during early childhood on schooling
Knowing the importance of early childhood health and nutritional status

in lowering on child schooling, we then asked whether the presence of a midwife

during such critical period of age helped children when they reached school-age.

The link that we try to establish between past exposure to public health program

35

and later schooling outcomes is built on the previous findings that the presence
of village midwife increased health of young children (Frankenberg et a1 2005).
Therefore if the presence of village midwife improved child health (as measured
by height-for-age), then we may expect that such a program may also yield
improvements in child schooling outcomes conditional on improvements in child
health.

Identifying an effect of the village midwife is not straightforward. One
might believe that presence of a midwife in period one (when child is in early
age) is exogenous for education measured in period 2 (when child is in school
age). However, as shown in Pitt et al (2003), placement of a public program is
likely to be correlated with the presence of other programs that already exist
(and remained) in the community. In our case, placement of public programs
might be correlated with availability or quality of school or other education
programs which in turn might affect parent decisions on investment in child
education.21 We thus argue that controlling for community fixed-effects is
important in this case and this should address correlation between the placement
of a midwife with time-invariant omitted heterogeneity including the presence of
other public programs in the community.

As we previously mentioned, however, there is a cost of including

community fixed-effect. As we seek to evaluate the intent-to-treat effect of

 

21Using Indonesia SUSENAS data, Pitt et a1 (2001) show that the placement of public programs
was correlated with the presence of another program that already existed and the effect of the
program is sensitive to whether or not one controls for this correlation.

36

presence of midwife, the inclusion of community fixed-effect will absorb all fixed
community level effects including presence of a midwife. We thus identify the
midwife effect by calculating the partial effect of childhood nutritional status
when presence of midwife is explicitly controlled for in the equation and
compare it with the one from equation without control of presence of midwife.
The difference is attributed to the presence of midwife in the community in
period 1. If presence of midwife indeed helps schooling of children given their
childhood nutritional status, then we will see that the presence of midwife will
increase the partial effect of childhood nutrition on subsequent schooling
outcomes.

We focus on the specifications that control for community fixed-effect in
column 2 and 4 of table 8. As we expect, the midwife variable is dropped by the
inclusion of community variable and the sign of childhood nutrition (for both
height-for-age z-score and non-stunted status) are negative. The effect of
childhood nutrition is significant at the 1 percent level where the size of the
effects are -0.155 (height-for-age z-score) and -0.507 (dummy for non-stunted).
Interactions between childhood nutritional status (both for height-for-age z-score
and dummy for non-stunted) and midwife are insignificant.

In the presence of an interaction term, however, we should not only use
individual statistical test to evaluate the partial effect of parameter of interest.
Instead, as noted in Wooldridge (2003), we need to conduct joint significant test

for height-for-age and interaction of height-for-age and midwife. The F-statistics

37

show that childhood nutrition variables and interactions with midwife are
significant at 1 percent level as shown in bottom panel of column 2 and 4 of table
8.

We thus calculate the partial effect of childhood nutrition in the presence
of a midwife and present the result in table 8A. We first look at the effect by
using height-for-age z-score as a measure of nutritional status. We find that the
partial effect of height-for-age (at midwife sample mean, 0.287) is -0.166 which
says that an increase in one standard deviation of child height lowered the
probability of delayed enrollment by 16.6 percent which is higher than the effect
when no midwife at all (15.5 percent).

Based on the estimated parameter in table 8, we also predict the effect of
child height in cases where presence of midwife in the communities increased to
50 and 75 percent. When probability of a midwife in the community is increased
to 50 percent, a one standard deviation increase in childhood nutritional status
lowers the probability of delayed enrollment by 17.4 percent. While if presence of
midwife is expanded further to 75 percent, the effect of childhood nutrition
becomes even stronger where an increase in one standard deviation of child
height lowers the probability of delayed enrollment by 18.4 percent.

When using stunting status (column 2 of table 8A), the partial effect of
childhood nutrition in the presence of midwife (sample mean=0.287) is -0.553.
This suggests that by being not stunted during childhood, the probability a child

enrolls in school on time is 55.7 percent higher than if he (or she) suffered

38

stunting during childhood. Recall that when we do not explicitly control for
presence of a midwife, eliminating stunting during early childhood could reduce
the probability of delayed enrollment by 50.7 percent relative to those who were
stunted during childhood. Predicted effects of being not-stunted during
childhood on probability of delayed enrollment by hypothetically increasing the
probability of a midwife in the community to 50 and 75 percent are consistent
with our calculation using height-for-age z-score as measure of childhood
nutrition. When the presence of a midwife is increased to 50 (75) percent, the
importance of not being stunted during childhood is even higher. N on-stunted
children are 38.7 (62.7) percent more likely to be enrolled in school at age 7, and
this is 58.7(62.7) percent higher than if they had experienced stunting during
early childhood.

We next turn to table 9 where we estimate the effect childhood nutrition in
the presence of midwife on probability of repeating a grade. As in table 8, we
again focus on the specifications that include community fixed-effects to control
for potential endogenous placement of a village midwife. We first find that
childhood nutritional status does not have an effect on probability of repeating a
grade as none of coefficients for height-for-age z-score and dummy for non-
stunted are significant. The join significant test for childhood nutrition (for both
height-for-age and dummy of non-stunted) and its interactions with dummy for

midwife presence indicate that they jointly are not significantly different from

39

zero. This is consistent with result in table 5 which shows no significant effect of
childhood nutrition on probability of repeating a grade.

Coefficients for interaction between childhood nutrition and dummy for
presence of midwife, however, are negative and significant at 5 and 10 percent
level. Coefficients for interaction between dummy for presence of midwife and
height—for-age z-score is -0.073 (column 2), while for its interaction with dummy
for not-stunted is 0.333 (column 4). What do these coefficients suggest? As the
effect of childhood nutritional status on probability of repeating a grade is
statistically zero (as also indicated in table 5), the significant interaction
coefficients (between childhood nutrition and midwife) indicate the effect of
presence of a midwife. It suggests that an increase in one standard deviation in
child height among children exposed to a midwife will reduce the probability of
repeating a grade by 7.3 percent relative to those who live in community without
village midwife (column 2 of table 9). It also that being well nourished in a
community with village midwife lowers the probability of repeating a grade by

33.3 relative to other children in the sample area.

2.6. Conclusions

This study evaluates the impact of childhood nutritional status and the
presence of a public health program on subsequent child schooling in Indonesia
during 19905. Guided by the model, we estimate dynamic relationship of

childhood nutrition and subsequent child schooling in which we carefully

40

address the potential correlation between childhood nutrition and important but
unobserved factors such as child innate healthiness and parents’ taste toward
child quality. In addition, with access to long-term historical rainfall data, we
create rainfall shocks during conception period and use them to identify
childhood nutritional status. By controlling for endogenous nutritional status, we
find that reducing incidence of poor childhood nutrition reduces also the
probability of delayed enrollment, but not the probability of repeating a grade.
More importantly, the estimated effects when taking into account the
endogeneity of childhood nutrition are 5-7 times stronger than when ignoring
the endogeneity of childhood nutrition.

The effect of childhood nutrition on subsequent child schooling is even
higher if child has access to public health facilities. Looking particularly at
presence of midwife, we find that the presence of midwife magnifying the effect
of childhood nutritional status on subsequent child schooling. This result
suggests positive effect of presence of midwife on child schooling.

What does this result imply? From the policy perspective, this result
reinforces the importance of investment on nutritional status of children at very
young ages, even just after conception and before birth. One effective channel of
health investment, as this study suggests, can be through local public health
facilities such as village midwife. From a methodological perspective, the results

presented in this study highlight the importance of taking into account the

41

endogeneity in childhood nutrition and other estimation issues when estimating

its effect on subsequent child schooling.

42

Table 2.1. Descriptive statistics of main variables

 

 

Variable Mean Std. Dev.
Delayed enrollment (Yes=1) 0.146 0.353
Ever repeated grade (Y es=1) 0.096 0.295
Heigh-for-age z-score (haz), lagged -1.390 1.442
Presence of village midwife, lagged 0.287 0.453
Mother education (years) 4.205 5.505
Father education (years) 5.057 5.782
Household head age (years) 42.285 9.911
# of children (5-14 yo) in household 1.960 0.945
# of female adult (15-59 yo) in household 1.382 0.706
# of male adult (15-59 yo) in household 1.284 0.801
Per capita expenditure (log) 11.992 0.712
Price of rice (log) 7.414 0.387
Price of sugar (log) 7.813 0.375
Price of cooking oil (log) 7.846 0.376
Price of condensed milk (log) 7.781 0.503

 

Notes: lagged variables are measured in period 1 (1993 survey).

43

Table 2.2. Impact of childhood nutrition on probability of delayed enrollment, OLS

Dep. Var: delayed enrollment

Measure of childhood nutrition

 

 

Height-for age Non-stunted
(Y es=1) z—score (Yes=1)
1 2 3 4
Childhood nutrition -0.032*** -0.035*** -0.090*** -0.083***
(0.006) (0.007) (0.018) (0.020)
Time dummy 0.087“ 0096* 0077* 0090*
(0.042) (0.049) (0.042) (0049)
Mother education -0.003* -0.001 -0.003* -0.001
(0.002) (0.002) (0.002) (0.002)
Father education -0.005*** -0.004** -0.005*** -0.004**
(0.002) (0.002) (0.002) (0.002)
Age of head of household 0.001 0.001 0.001 0.001
(0.001) (0.001) (0.001) (0.001)
# 6-14 yo children in HH 0030*“ 0037*“ 0031*“ 0039*“
(0.009) (0.010) (0.009) (0.010)
# female adults in HH 0.000 0.007 -0.002 0.006
(0.012) (0.014) (0.012) (0.014)
# male adults in HH -0.027** -0.015 -0.027** -0.016
(0.011) (0.012) (0.011) (0.012)
Per-capita expenditure -0.040*** -0.031** -0.042*** -0.034**
(0.012) (0.015) (0.012) (0.015)
Price of rice 0.017 -0.02 0.016 -0.027
(0.030) (0.051) (0.030) (0.051)
Price of sugar 0.008 0.065 0.011 0.063
(0.059) (0.073) (0.059) (0.073)
Price of cooking oil 0.016 0.006 0.016 0.002
(0.021) (0.030) (0.021) (0.030)
Price of condensed milk -0.082** -0.138*** -0.080** -0.133***
(0.035) (0.050) (0.035) (0.050)
Constant 0.822“ 0.989* 0.838“ 1.103“
(0.413) (0.515) (0.413) (0.515)
Community fixed-effect N 0 Yes N 0 Yes
R-squared 0.06 0.25 0.06 0.25
Observations 1944 1944 1944 1944

 

Notes: Dependent variable is whether child experienced delayed enrollment (yes=1). Robust
standard error is in parenthesis. (***), (**), (*) respectively indicate significant at 1, 5 and 10

percent level.

44

Table 2.3. Impact of childhood nutrition on probability of repeating grade, OLS

 

Dep. Var: Repeated grade Measure of childhood nutrition

 

 

 

Height-for age Non-stunted
(Yes=1) z-score (Yes=1)
1 2 3 4
Childhood nutrition -0.004 0.001 -0.005 0.005
(0.005) (0.006) (0.015) (0.016)
Time dummy 0.043 0.049 0.041 0.049
(0.036) (0.043) (0.036) (0.043)
Mother education -0.003** -0.004** -0.003** -0.004**
(0.002) (0.002) (0.002) (0.002)
Father education -0.003** -0.002 -0.003** -0.002
(0.002) (0.002) (0.002) (0.002)
Age of head of household 0.000 0.000 0.000 0.000
(0.001) (0.001) (0.001) (0.001)
# 6-14 yo children in HH 0.006 0.009 0.006 0.01
(0.007) (0.008) (0.007) (0.008)
# female adults in HH 0.001 0.007 0.001 0.007
(0.010) (0.011) (0.010) (0.011)
# male adults in HH -0.003 0.003 -0.003 0.003
(0.009) (0.010) (0.009) (0.010)
Per-capita expenditure -0.019** -0.012 -0.020** -0.012
(0.008) (0.010) (0.008) (0.010)
Price of rice -0.026 0.024 -0.027 0.025
(0.020) (0.030) (0.020) (0.031)
Price of sugar 0.004 0.049 0.005 0.049
(0.058) (0.075) (0.058) (0.075)
Price of cooking oil -0.012 -0.028 -0.012 -0.028
(0.021) (0.029) (0.021) (0.029)
Price of condensed milk 0.025 -0.039 0.025 -0.039
(0.022) (0.036) (0.022) (0.036)
Constant 0.391 0.183 0.397 0.18
(0.355) (0.455) (0.355) (0.456)
Community fixed-effect No Yes No Yes
R-squared 0.02 0.20 0.02 0.21
Observations 1944 1944 1944 1944

 

Notes: Dependent variable is dummy for child experienced repeated grade (yes=1). Robust
standard error is in parenthesis. (***), (**), (*) respectively indicate significant at 1, 5 and 10
percent level.

45

Table 2.4. Impact of childhood nutrition on probability of delayed enrollment: 2SLS

 

Dep. Var: delayed enrollment

Measure of childhood nutrition

 

 

Height-for age Non-stunted
(Yes=1) z-score (Y es=1)
1 2 3 4
Childhood nutrition -0.097*** -0.158*** -0.313*** -0.539***
(0.023) (0.030) (0.077) (0.108)
Time dummy 0.140*** 0.166*** 0115*“ 0.154"’*“r
(0.046) (0.058) (0.045) (0.059)
Mother education -0.003 0.000 -0.002 0.000
(0.002) (0.002) (0.002) (0.002)
Father education -0.005*** -0.003 -0.005*** -0.004*
(0.002) (0.002) (0.002) (0.002)
Age of head of household 0.001 0.001 0.001 0.002
(0.001) (0.001) (0.001) (0.001)
# 6-14 yo children in HH 0.022“ 0.026“ 0.024“ 0.030“
(0.010) (0.012) (0.010) (0.013)
# female adults in HH 0.009 0.016 0.005 0.009
(0.014) (0.015) - (0.013) (0.016)
# male adults in HH -0.030** -0.018 -0.030** -0.026*
(0.012) (0.013) (0.012) (0.015)
Per-capita expenditure -0.021 -0.011 -0.024* -0.022
(0.014) (0.017) (0.014) (0.017)
Price of rice 0.011 -0.046 0.000 -0.09
(0.033) (0.065) (0.033) (0.067)
Price of sugar -0.025 0.056 -0.023 0.038
(0.062) (0.081) (0.061) (0.085)
Price of cooking oil 0.02 0.043 0.025 0.034
(0.022) (0.034) (0.023) (0.035)
Price of condensed milk -0090** -0.159*** -0.086** -0.143**
(0.037) (0.058) (0.037) (0.060)
Community fixed-effect No Yes No Yes
F-stat on the excluded instrument 23.46 18.85 19.25 13.56
(p—value) (0.000) (0.000) (0.000) (0.000)
Kleibergen-Paap rk LM statistic 122.39 92.71 111.19 73.68
(p-value) (0.000) (0.000) (0.000) (0.000)
Hansen J stat 3.89 5.74 4.21 4.53
(p-value) (0.566) (0.361) (0.518) (0.477)
Observations 1910 1890 1910 1890

 

Notes: Dependent variable is lagged height-for-age z-score. Robust standard error is in
parenthesis. (***), (**), (*) respectively indicate significant at 1, 5 and 10 percent level. Excluded
IVs are: interaction rainfall shocks during second and third trimester of in utero period and child
age (in months), total household assets (log), height of father and mother (cm), age of children (in
months) and interaction between number of posyandu in the village and child age, all are in
period 1. First-stage regression for lagged height-for-age and stunted are in table A1 and A2.

46

Table 2.5. Impact of childhood nutrition on probability of repeating grade: 2SLS

 

Dep. Var: Repeated grade

Measure of childhood nutrition

 

 

 

Height-for age Non-stunted
(Yes=1) z-score (Yes=1)
1 2 3 4
Childhood nutrition -0.026 -0.015 -0.091 -0.043
(0.019) (0.022) (0.061) (0.076)
Time dummy 0.052 0.055 0.046 0.053
(0.039) (0.045) (0.038) (0.044)
Mother education -0.003* -0.004** -0.003 -0.004**
(0.002) (0.002) (0.002) (0.002)
Father education -0.003** -0.002 -0.003** -0.002
(0.001) (0.002) (0.001) (0.002)
Age of head of household 0.000 0.000 0.000 0.000
(0.001) (0.001) (0.001) (0.001)
# 6-14 yo children in HH 0.002 0.007 0.002 0.008
(0.008) (0.009) (0.008) (0.009)
# female adults in HH 0.004 0.006 0.004 0.005
(0.011) (0.012) (0.011) (0.012)
# male adults in HH -0.004 0.002 -0.004 0.001
(0.009) (0.010) (0.009) (0.010)
Per-capita expenditure -0.013 -0.009 -0.013 -0.01
(0.010) (0.011) (0.010) (0.011)
Price of rice -0.02 0.028 -0.024 0.024
(0.026) (0.040) (0.026) (0.040)
Price of sugar -0.003 0.044 -0.003 0.043
(0.060) (0.077) (0.059) (0.077)
Price of cooking oil -0.011 -0.026 -0.01 -0.027
(0.022) (0.030) (0.022) (0.030)
Price of condensed milk 0.023 -0.04 0.024 -0.039
(0.022) (0.036) (0.022) (0.036)
Community fixed-effect No Yes No Yes
F-stat on the excluded instrument 23.46 18.85 19.25 13.56
(p—value) (0.000) (0.000) (0.000) (0.000)
Kleibergen-Paap rk LM statistic 122.39 92.71 111.19 73.68
(p-value) (0.000) (0.000) (0.000) (0.000)
Hansen J stat 2.80 4.46 2.43 4.91
(p-value) (0.730) (0.485) (0.788) (0.427)
Observations 1910 1890 1910 1890

 

Notes: Dependent variable is (lagged) dummy for child was stunted during childhood. Robust
standard errors are in parenthesis. (***), (**), (*) respectively indicate significant at 1, 5 and 10
percent level. Excluded IV s are: interaction rainfall shocks during second and third trimester of in
utero period and child age (in months), total household assets (log), height of father and mother
(cm), age of children (in months) and interaction between number of posyandu in the village and
child age, all are in period 1. First-stage regression for lagged height-for-age and stunted are in

table A1 and A2.

47

Table 2.6. Heterogeneity impact of childhood nutrition on probability of delayed
enrollment across gender, ZSLS-LIML

Dep. Var: delayed enrollment Measure of childhood nutrition

 

 

 

Height-for age N on-stunted
(Yes=1) z-score (Yes=1)
Boys Girls Boys Girls
Childhood nutrition -0.213*** -0.124*** -0.661*** -0.427***
(0.053) (0.043) (0.229) (0.145)
Time dummy 0.153 0.157" 0.11 0.166“
(0.107) (0.065) (0.107) (0.068)
Mother education 0.001 -0.003 0.004 -0.003
(0.004) (0.003) (0.005) (0.003)
Father education -0.004 -0.004 -0.005 -0.004
(0.004) (0.003) (0.004) (0.003)
Age of head of household 0.001 0.000 0004* 0.001
(0.002) (0.001) (0.002) (0.001)
# 6-14 yo children in HH 0.024 0033* 0.031 0.026
(0.020) (0.017) (0.021) (0.019)
# female adults in HH 0.028 0.031 -0.01 0.034
(0.029) (0.021) (0.029) (0.023)
# male adults in HH -0.026 -0.012 -0.032 -0024
(0.022) (0.020) (0.024) (0.021)
Per-capita expenditure -0.011 -0.006 -0.015 -0.033
(0.027) (0.022) (0.029) (0.020)
Price of rice -0.112 -0.088 -0.142 -0.149
(0.100) (0.095) (0.106) (0.092)
Price of sugar 0.105 0.115 0.093 0.114
(0.128) (0.118) (0.129) (0.126)
Price of cooking oil 0.111* 0.016 0.047 0.034
(0.066) (0.047) (0.068) (0.046)
Price of condensed milk -0.173* -0.151* -0.137 -0.168*
(0.103) (0.085) (0.103) (0.087)
Community fixed-effect Yes Yes Yes Yes
F-stat on the excluded instrument 9.34 19.31 5.77 6.28
(p-value) (0.000) (0.000) (0.000) (0.000)
Kleibergen-Paap rk LM statistic 45.70 45.98 41.73 32.74
(p-value) (0.000) (0.000) (0.000) (0.000)
Hansen J stat 5.26 2.30 6.96 1.77
(p—value) (0.385) (0.806) (0.224) (0.881)
Observations 914 878 914 878

 

Notes: Dependent variable is whether child experienced delayed primary school enrollment
(yes=1). Estimation uses limited information maximum likelihood which is robust to potentially
weak instruments. (***), (**), (*) respectively indicate significant at 1, 5 and 10 percent level.
Excluded IVs are: interaction rainfall shocks during second and third trimester of in utero period
and child age (in months), total household assets (log), height of father and mother (cm), age of
children (in months) and interaction between number of posyandu in the village and child age,
all are in period 1. First-stage regression is in table A3 and A4.

48

Table 2.7. Heterogeneity impact of childhood nutrition on probability of repeating grade

 

 

 

across gender, ZSLS-LIML
Dep. Var: delayed enrollment Measure of childhood nutrition
Height-for age N on-stunted
z-score (Yes=1)
Boys Girls Boys Girls
Childhood nutrition 0.011 -0.041 -0.008 0.126
(0.034) (0.029) (0.102) (0.101)
Time dummy 0.082 0.061 0.087 0.062
(0.078) (0.057) (0.077) (0.057)
Mother education -0.006** -0.003 -0.006** -0.003
(0.003) (0.003) (0.003) (0.003)
Father education -0.001 -0.002 -0.001 -0.002
(0.003) (0.002) (0.003) (0.002)
Age of head of household -0.002 0.001 -0.002 0.001
(0.001) (0.001) (0.001) (0.001)
# 6-14 yo children in HH 0.024 -0.005 0.023 -0.007
(0.016) (0.012) (0.016) (0.013)
# female adults in HH 0.005 -0.006 0.007 -0.005
(0.021) (0.018) (0.021) (0.018)
# male adults in HH 0.011 -0.012 0.011 -0.015
(0.018) (0.015) (0.018) (0.016)
Per-capita expenditure -0.029* 0.005 -0.029* -0.004
(0.017) (0.018) (0.017) (0.016)
Price of rice -0.001 0.067 -0.003 0.048
(0.047) (0.056) (0.048) (0.060)
Price of sugar 0.105 -0.094 0.103 -0.096
(0.121) (0099) (0.120) (0.102)
Price of cooking oil -0.078 0.019 -0.074 0.023
(0.058) (0.040) (0.055) (0.039)
Price of condensed milk -0.078 0.028 -0.078 0.025
(0.064) (0.050) (0.064) (0.051)
Community fixed-effect Yes Yes Yes Yes
F-stat on the excluded instrument 9.34 9.31 5.77 6.28
(p-value) (0.000) (0.000) (0.000) (0.000)
Kleibergen-Paap rk LM statistic 45.70 45.98 41.73 32.74
(p-value) (0.000) (0.000) (0.000) (0.000)
Hansen J stat 5.14 0.69 5.18 1.13
(p-value) (0.399) (0.984) (0.394) (0.951)
Observations 914 878 914 878

 

Notes: Dependent variable is whether child experienced grade repetition (yes=1). Estimation uses
limited information maximum likelihood which is robust to potentially weak instruments. (***),
(**), (*) respectively indicate significant at 1, 5 and 10 percent level. Excluded IV 5 are: interaction
rainfall shocks during second and third trimester of in utero period and child age (in months),
total household assets (log), height of father and mother (cm), age of children (in months) and
interaction between number of posyandu in the village and child age, all are in period 1. First-
stage regression is in table A3 and A4.

49

Table 2.8. Impact of midwife exposure on probability of delayed enrollment:
ZSLS-LIML
Dep. Var: delayed enrollment

 

Measure of childhood nutrition

 

 

Height-for age Non-stunted
z-score (Yes=1)
1 2 3 4
Childhood nutrition -0.076*** -0.155*** -0.247*** -0.507***
(0.026) (0.039) (0.093) (0.150)
Childhood Nutrition*Midwife -0.052 -0.038 -0.188 -0.160
(0.046) (0.058) (0.157) (0.224)
Midwife -0.055 0.147
(0.062) (0.051)
Community fixed-effect No Yes No Yes
F-test for variables of interest
(p-value)
Childhood nutrition and 15.30 27.67 15.51 24.61
Childhood nutrition*midwife (0.001) (0.000) (0.001) (0.000)
Midwife and 2.10
Childhood nutrition*midwife 1.71 (0.425) (0.349)
F—stat on the excluded instrument 14.15 10.07 11.73 8.11
(p-value) (0.000) (0.000) (0.000) (0.000)
Kleibergen-Paap rk LM statistic 135.42 82.43 105.04 62.05
(p-value) (0.000) (0.000) (0.000) (0.000)
Hansen J stat 8.95 9.90 7.59 8.46
(p-value) (0.399) (0.449) (0.669) (0.584)
Observations 1910 1890 1910 1890

 

Notes: Dependent variable is whether child experienced delayed primary school enrollment
(yes=1). Other covariates in each specifications (but not displayed here): time dummy, height of
parents, age of household head, number of children as well as male and female adults in
household, per-capita expenditure, and food prices. Estimation uses limited information
maximum likelihood which is robust to potentially weak instruments. (***), (**), (*) respectively
indicate significant at 1, 5 and 10 percent level. Excluded IVs in specifications in column 1 & 3 are
interaction rainfall shocks during second and third trimester of in utero period and child age (in
months), total household assets (log), height of father and mother (cm), age of children (in
months) and interaction between number of posyandu in the village and child age, all are in
period 1. For specifications in column 2 and 4 are those as for column 1 and 3 plus their
interaction with dummy variable for presence of midwife in period 1. F-test on the excluded
instruments for interaction between childhood nutrition and presence of midwife for each
specification in column 1,2,3,4 are respectively 6.83, 5.69, 5.26, and 4.46.

50

Table 2.8A. Partial effect of childhood nutrition on probability of delayed
enrollment conditional on the presence of midwife
Proportion of presence of midwife Height-for-age Non-stunted

 

 

 

z-score (yes=1)
At sample mean (28.7%) -0.166*** -0.553***
[0.033] [0.121]
Simulations:
Increase presence of midwife to 50% -0.174*** -0.587***
(0.033) (0.118)
Increase presence of midwife to 75% -0.184*** -0.627***
(0.039L (0.138)

 

Notes: Robust standard errors are in parentheses. Calculations are based on the
parameter in table 8 column 2 and 4. (***), (**), (*) respectively indicate significant at 1, 5
and 10 percent level.

51

Table 2.9. Impact of midwife exposure on probability of repeated grade: 2SLS-LIML

 

Dep. Var: delayed enrollment

Measure of childhood nutrition

 

 

Height-for age Non-stunted
(Yes=1) z-score (Yes=1)
1 2 3 4
Childhood nutrition -0.004 0.02 -0.003 0.138
(0.022) (0.029) (0.075) (0.108)
Childhood nutrition*Midwife -0.061* -0.073* -0.211* -0.333**
(0.037) (0.043) (0.127) (0.162)
Midwife -0.072 -0.058
(0.051) (0.043)
Community fixed-effect No Yes No Yes
F-test for variables of interest
(p-value)
Childhood nutrition and 3.74 2.97 3.65 4.26
Childhood nutrition*midwife (0.154) (0.226) (0.161) (0.119)
Midwife and 2.98
Childhood nutrition*midwife 2.91 (0.234) (0.225)
F-stat on the excluded instrument 14.15 10.07 11.73 8.11
(p-value) (0.000) (0.000) (0.000) (0.000)
Kleibergen-Paap rk LM statistic 135.42 82.43 105.04 62.05
(p-value) (0.000) (0.000) (0.000) (0.000)
Hansen] stat 14.98 13.37 14.29 12.41
(p-value) (0.133) (0.185) (0.162) (0.259)
Observations 1910 1890 1910 1890

 

Notes: Dependent variable is whether child experienced grade repetition (yes=1). Other

covariates in each specifications (but not displayed here): time dummy, height of parents, age of
household head, number of children as well as male and female adults in household, per-capita
expenditure, and food prices. Estimation uses limited information maximum likelihood which is
robust to potentially weak instruments. (***), (**), (*) respectively indicate significant at 1, 5 and
10 percent level. Excluded N5 in specifications in column 1 & 3 are interaction rainfall shocks
during second and third trimester of in utero period and child age (in months), total household
assets (log), height of father and mother (cm), age of children (in months) and interaction
between number of posyandu in the village and child age, all are in period 1. For specifications in
column 2 and 4 are those as for column 1 and 3 plus their interaction with dummy variable for

presence of midwife in period 1. F-test on the excluded instruments for interaction between
childhood nutrition and presence of midwife for each specification in column 1,2,3,4 are

respectively 6.83, 5.69, 5.26, and 4.46

52

APPENDICES

53

Table A2.1. First Stage Regression: Height-for-age z-score

 

Time dummy

Mother education

Father education

Household head age

# of older children, 6-14 yo

# of adult female

# of adult male

PCE (log)

price of rice (log)

price of sugar (log)

price of cooking oil (log)

price of condensed milk (log)

3rd trimester rainfall shock*Age in period 1 (x1000)
2nd trimester rainfall shock*Age in period 1 (x1000)
Assets in period 1 (log)

# of posyandu in period 1*age in period 1 (x1000)

Constant

0.395
(0.165)
0.010
(0.008)
0.002
(0.008)
-0.002
(0.004)
-0129
(0.034)
0.114
(0.054)
-0053
(0.044)
0.233
(0.048)
0.057
(0.103)
-0414
(0.199)
0.076
(0.105)
-0114
(0.124)
0.033
(0.011)
-0041
(0.010)
0.068
(0.020)
-0025
(0.065)
-1.762
(1.631)

 

Note: Robust standard errors are in parantheses.

54

Table A2.2. First Stage Regression: being non-stunted (Y es=1)

 

Time dummy

Mother education
Father education
Household head age

# of older children, 6-14 yo
# of adult female

# of adult male

PCE (log)

price of rice (log)

price of sugar (log)

price of cooking oil (log)

price of condensed milk (log)

3rd trimester rainfall shock*Age in period 1 (x1000)

2nd trimester rainfall shock*Age in period 1 (x1000)

Assets in period 1 (log)

# of posyandu in period 1*age in period 1 (x1000)

Constant

0.051
(0.06)
0.006
(0.00)
0.001
(0.00)
-0001
(0.00)
0.035
(0.01)
0.023
0.000
-002
(0.014)
0.06
(0.016)
-002
(0.043)
-012
(0.075)
0.04
(0.034)
-002
(0.039)
-0012
(0.000)
-001
(0.000)
0.026
(0.006)
0.003
(0.000)
0.62
(0.56)

 

Note: Robust standard errors are in parantheses.

55

CHAPTER 3

AN EVALUATION OF A SUPPLEMENTARY FEEDING PROGRAM IN
INDONESIA: IDENTIFICATION USING OF HETEROGENEITY IN
PROGRAM EXPOSURE

3.1. Introduction

A growing number of studies have established a link between
malnutrition during early childhood and slower physical growth, delayed motor
development, lower IQ, and low educational achievement. A related literature
has demonstrated an association between improved health and nutritional status
during first few years of life with later life outcomes that affect health, education
and productivity, including such outcomes as mortality, chronic disease,
educational attainment, and returns to labor.22 In recognition of the
consequences of malnutrition, governments in many countries have established
programs aiming to reduce or to prevent malnutrition during infancy and early
chfldhood.

A set of public health programs introduced by the Indonesian government
and focusing on infant and maternal health fits within this general policy focus
found in many developing countries. Basic health programs -- both preventive

and curative — have been delivered to community through village midwives and

 

22(e.g., Alderman et al 2001; Alderman, Behrman and Hoddinot, 2005; Alderman, Hoddinot and
Kinsey 2006; Glewwe, Jacoby and King 2001 ; Glewwe and King 2001 ; Maluccio et al 2005;
Martorell 1999; and Strauss and Thomas, 1998) .

56

integrated child health services clinics (P03 Pelayamm Terpadu, Posyandu).23 When
Indonesia was buffeted by economic crisis during 1997 and 1998, the
Government of Indonesia (GOI) launched a supplementary feeding program,
Program Makamm Tambahrm (PMT), as part of its broader group of social safety
net programs. The specific aim of the feeding program was to help pregnant
women and children under five years of age cope with adverse effects of the
crisis. The potential importance of the program is evident from the scale of its
rollout from 1998 to 2000.

In this study, we evaluate the effect of a supplementary feeding program
on child health and nutritional status in Indonesia from 1998 to 2000. We
estimate the ’intent-to-treat’ effect of the program by quantifying the effect of
program exposure within the community on targeted individuals regardless of
actual participation in the program.24 The data used for the study come from
two rounds of panel data from Indonesia Family Life Survey (IFLS-2 and IFLS-3)
covering pre- (1997) and post-crisis (2000) periods. The IFLS includes detailed

information on the implementation of Social Safety Net programs, including the

 

23The village midwife program was designed to address basic health issues particularly among
reproductive age women. Some studies have evaluated the program impact, finding that the
placement of a village midwife improved the health of women between 15-49 years old and
children between 1-4 years old (see Frankenberg and Thomas, 2001 and Frankenberg, Suriastini
and Thomas, 2004). The establishment of Posyandu was aimed, in particular, at maintaining the
health and nutritional status of children under 5 year of age through growth monitoring and by
providing basic preventive health services and nutritional supplementation.

24 The use of an intent-to-treat approach in evaluating the effect of a program has an important
advantage relative to the use of a participation indicator: this approach avoids the complication
raised by the fact that participation by individuals is not exogenous.

57

PMT. The rich household and community panel surveys allow us to contribute to
the current literature on program evaluation in two important ways.

First, this is the first study that uses a national panel survey to evaluate
the main child nutrition support program implemented by the Indonesia’s
government during the 1997-98 crisis period. The focus on crisis years should be
of interest to researchers as well as policy makers since some studies have shown
that there was no significant decline in Indonesian children’s health and
nutritional status during the crisis period (Block et a1, 2005, Frankenberg, Beegle
and Thomas, 1999, and Strauss et. a1. 2004).25 A natural question from these
findings is whether public programs aiming to support child health and
nutritional status helped to prevent declines in health status from occurring.

In addition, using Indonesia for an evaluation of a nutritional program
has important short- and long-term policy relevance. While evidence shows an
improvement in child health, one indicator of extreme malnutrition, the level of
stunting, was comparable in Indonesia in 2000 to many low-income countries in
sub-Saharan African. Identifying the effects of the program on child nutritional

status promises to provide important guidance as to whether public

 

25 Strauss et a1 (2004) provide evidence that lack of change to child health in two IFLS-based
studies (Frankenberg et al, 1999 and Strauss et a1, 2004) may be due to the fact that survey timing
did not coincide with negative shocks to child health. IFL52+ in 1998 may have occurred too early
to detect a negative shock to child health, and IFLS 3 in 2000 may have been too late to capture it.
In addition, they note that while there was no strong evidence of a decline in child health during
the crisis period, this could also be driven by such factors as drought in the pre-crisis period, or
birth or mortality selection.

58

interventions at a time of crisis may assist in reducing malnutrition among
children during and after financial crises.

A second contribution of the paper is methodological. To date, the vast
majority of studies in the program evaluation literature make use of a single
binary indicator for program exposure. The use of this type of program variable
is practical yet may not be able to capture program impact when at least one of
two conditions apply: (i) the program is universally distributed, or (ii) the
targeted communities receive heterogeneous exposure. The implementation of
PMT was characterized by both of these features. In order to evaluate impact in
light of this implementation approach, we make use of the detailed information
on program implementation available in IFLS-3 and then exploit the
heterogeneity in program exposure across communities to estimate the program
effect.

Our findings indicate that the program was effective but its impact was
limited to children between 12 and 23 months of age. Children from younger and
older age groups were not affected by the program. Depending on the
specifications used, the program increased the standardized height-for-age of
children 12 to 23 months old by 0.48 to 0.55 standard deviations. We also find
that the exposure to the program reduced the likelihood of stunting within the
same age group by 15 percent by 2000.

This paper is organized as follows. In section 2, we highlight some

findings from previous studies on program evaluation, particularly those

59

focusing on Indonesia, and discuss the contribution of this research to the
existing literature. Section 3 provides a description about the program and how it
was distributed. Section 4 provides an analytical framework, which is then
followed by discussions of the data source in section 5. We discuss our empirical
strategy and the estimation issues we face in section 6. In Section 7, we present

and discuss the results, and provide conclusions in section 8.

3.2. Previous Evaluation Studies on Indonesia Public Health Program

Despite the potential importance of supplementary feeding interventions,
there are relatively few efforts to evaluate the effectiveness of these programs in
developing countries. In Indonesia, there is only one study evaluating the
effectiveness of the post-crisis supplementary feeding program. Sandjaja et a1
(2001) find little evidence that the program improved nutritional status of
targeted children, yet they suggest that the program helped avert a decline of
nutritional status among children between 6 and 23 months of age.

The Sandjaja study, however, made use of a small sample size and thus
we should be concerned about whether results are generalizable. In addition, the
Sandjaja study demonstrated that the program was targeted toward less healthy
children, but it fails to address the potential bias to their evaluation from
endogenous program placement. In particular, the matching method employed
does not control for unobserved heterogeneity likely to be correlated with both

the outcome and likelihood of treatment. Moreover, because they were only able

60

to control for post-intervention characteristics of children, the matching
procedure is also likely to select a control group that is not necessarily
comparable with the treated group before the intervention took place. Other
studies have faced similar problems: data are inadequate for evaluation and
often cannot control for program placement.

Other studies evaluating public health programs in Indonesia have
implemented more appropriate strategies to deal with potential bias in
estimating program effects (Frankenberg 1995, Frankenberg and Thomas 2001,
Frankenberg, Suriastini and Thomas 2005, and Gertler and Molyneaux 1994).
With access to more comprehensive and much larger data sets, they control for
endogenous program placement and estimate program effects by implementing
regional fixed-effect or difference-in-difference approaches. Our discussion of
evaluation strategies and results focus on the best of these approaches, found in
the two most recent papers above.26

In order to identify effects of placing a midwife within a community,
Frankenberg and Thomas (2001) first compare health status of women in primary
age prior to the introduction of a midwife in a community with the health of the
same individuals after the program. They argue that the program effect may be
contaminated by two sources of unobserved heterogeneity: individual and

community. To deal with unobserved heterogeneity at the individual level, they

 

26For additional reviews, see Strauss and Thomas (1995) for review of Frankenberg (1995) and
Gertler and Molyneaux (1994).

61

estimate the effect of a midwife on change in health status. First-differencing
health status eliminates time-invariant forms of individual heterogeneity. They
then include a community fixed-effect to sweep out any time-invariant
unobserved heterogeneity at the community level that may be correlated with
placement of midwives. In addition, they note that the program effect may leak
to some non-targeted groups. To control for such spillovers, they include more
control groups such as primary age males and older males and females in the
regression specification. Their findings indicate that the midwife program
increases body mass index of women in reproductive age by 0.20 more than of
both older and reproductive age men. Their exercise also suggests that the
program has spillover benefits for older women.

Meanwhile, Frankenberg et al (2005) argue that comparing health status of
children residing in a community with a midwife with that of their counterparts
living in a community without a midwife is difficult to interpret due to
confounding impact of the selective assignment of midwives. In order to
measure the midwife effect, they compare the height-for-age of young children
cohort up to 4 years of age, who were exposed to a midwife, with height for age
of older age cohorts who lived in the same community but who were not
exposed to a midwife when young. Their findings show that children who were
fully exposed to a midwife during early childhood had significantly better
height-for-age scores than older children who lived in the same community but

were not exposed to the program.

62

While the empirical strategy in these two studies may adequately address
bias from endogenous program placement, identification rests on a questionable
assumption. The use of simple binary program variable implies that the program
was implemented homogenously in terms of level and duration across
communities. In reality many public health programs are implemented with
various types of subprograms or heterogeneous in intensity and exposure (e.g.,
in terms of duration, supplies, training and additional support). An analysis
relying on a simple binary indicator may either overstate or understate the
effectiveness of the program.27

In sum, only a few studies evaluating public health programs in Indonesia
use methodologically sound approaches and data. Other evaluations have relied
on limited information regarding program implementation and the distribution
of other programs, they often have relied on poor identification strategies, and
assumed that programs were distributed with homogenous exposure across
regions. Our study addresses these limitations by taking into account the
heterogeneity in program exposure and distribution of other similar programs,
and, at the same time, controlling for the effects of non-random program

placement. We utilize detailed information on implementation and program

 

27 The use of program intensity as a measure of program exposure was first introduced in the
literature by Rosenzweig and Wolpin (1986), but use of such measures have only grown recently
(see Dulfo 2001, and more recently Armecin eta12006, Behrman eta12004, Gertler 2004, and
Yamauchi 2005). Gertler (2004) provides direct comparison of program effects from the two types
of program variables, single binary and multiple binary program variables, and shows that both
types of program indicators yield significant parameters for program effects. Using multiple
dummies for length of program exposure, however, generates more detailed and insightful
parameters for characterizing program effects.

63

features to characterize heterogeneity in program exposure, and then exploit this

heterogeneity to quantify the effect of the supplementary feeding program.

3.3. Analytical Framework

We follow a simple framework introduced by Rosenzweig and Wolpin
(1986) to take into account heterogeneity in program exposure when modeling
the effect of the program on child nutritional status. We assume that preferences
of household members are inter-temporally separable and that in the current
period the household maximizes a quasi-concave utility function over goods,

services and the health status of children:

(1) mailXUKXCHfiZ’)

where X is a vector of consumption goods and services consumed by child 1', H is
the health or nutritional status of child 1', and Z is a vector of household
characteristics. The production of health (or nutritional status), H, is

characterized by the following production function:

<2) H" = h(N’,X",Z",7t’)

in which, child health is a function of per child health inputs N, household

characteristics Z, and community characteristics it -both observed and

64

unobserved. The household also faces a budget constraint in which total
consumption of goods and services as well as health inputs cannot exceed total

income:

(3) Y’=p.X’+(p.-S.)N’

where Y is total income, p is price of goods and services X, p, and s respectively
are price and subsidy of health input N.

Solving the optimization problem in equation (1) conditional on equation
(2) and (3) yields a reduced-form health production function for each individual

within the household:

(4) Hi=h(px’pn’S:\/9Yinui)

Our interest here is to identify the impact of public nutrition program on child

nutritional status. Using equation (4) we predict program impact as follows:

(5) 311'“ _ aH’ 3N’ + 3H' ax’ + 021’
63;, 6N’ 633, EX' as; as],

 

 

 

Equation 5 tells that the effect of the program (subsidy) on health or nutritional
status can be decomposed into three components. First is the subsidy effect of

nutritional inputs through the change in demand for nutrition inputs (a price

65

effect). The second component is a change in child nutritional status of program
(subsidy) through relocation of resources within household.28
The third term is bias which is introduced if the size of the subsidy, s, is

affected by unobserved characteristics of children or a community ,u . The sign of

the bias is ambiguous. It is negative when the government or program
distributor follows a compensatory principle, or when a program is distributed
more to less-endowed areas, which will thus lead to downward bias in estimates
of the true program effect. Conversely, if program is placed more intensively in
better endowed areas, the sign of bias becomes positive and the estimated
program effect overstates the true effect. Thus unless program 5 allocation is

independent of unobserved heterogeneity ,u , an estimation of the program effect

must control for these forms of unobserved heterogeneity.

3.4. Nutritional intervention for children during economic crisis.
The goal of supplementary feeding program was to maintain and to
improve the nutritional status of children under five, particularly those who

were susceptible to the effects of the 1997-98 economic crisis.29 The program was

 

28For example, some research in the literature has found that nutritional intervention programs
may be ineffective if parents relocated some nutritional resources away from treated children to
other household members.

29 A monthly supplementary feeding program took place prior to the crisis as part of a public
health program. In contrast to post-crisis supplementary feeding program, the pre-crisis program
was much less intense and distributed solely through Village Integrated Health Post (Pos
Pelayamm Terpadu, Posyandu). Under this program, food supplements were delivered along with
some basic health services to preschool age children and pregnant women residing in the
community.

66

targeted through a two stage process. First, central and regional governments
decided which communities or villages would receive the program. The
targeting decision at this stage involved two steps: first, it determined which
communities or villages received the program, and second, it decided the length
of exposure and intensity of the program in each village. Once a community was
assigned to receive the program and program duration was determined, funding
was distributed through a public health clinic, Pusat Kesehatan Masyarakat
(Puskesmas), and a list of children eligible to receive supplementary food was
then prepared by the village midwife in consultation with other village leaders,
which may have included the village head, women activists, puskesmas staff and
family planning field staff.30

As with other health programs during 1997/ 98 economic crisis, program
supervision was the responsibility of the village midwife. If a midwife was not
available in the community, puskesmas staff assumed responsibility instead.31
Descriptive information from the IFLS show that in 62 percent of communities,
the village midwife administered the program, and in majority of the rest (35

percent of sampled communities), the program was supervised by a member of

 

3'0 IFLS data show that although the village midwife was mainly responsible for administering the
program, others were also involved in preparing program's recipient list, including the village
head or other officials, family planning workers, Puskesmas staff, and community activists.

31 Village midwives are health workers that are trained to provide basic health services in
communities or villages. Their work is coordinated and supervised by the head of Puskesmas
whose scope of services include one kecamatan (an administrative area intermediate in size
between the district and the village).

67

the puskesmas staff.32 The program was implemented and supplemental food
distributed to the community through several different providers. From the IFLS
community and facility surveys, we know that the program was implemented by
Posyandu in the majority of sampled communities. Other than the Posyandu,
village midwifes and village women’s association (Program Kesejahteraan
Keluarga, PKK) also played a role in delivering nutritional supplements to
members of the community.33

The program targeted poor children between 6 and 59 months of age and
pregnant women. The intervention specifically divided targeted children into
three sub-groups: (i) infants (6-11 months), (ii) young children (12-23 months),
and (iii) children (24-59 months).34 The majority of communities and villages that
received the program during the 1998-2000 period served all of these sub-groups.
The community surveys also indicate that in a small portion of communities,
non-targeted individuals also received some benefits, including children 5-14
years old, women in reproductive age, the elderly and in some cases, even adult

males.

 

32 The fact that the program was handled by the midwife may raise an issue in identifying the
effect of the program. Other health services provided by the midwife might contaminate the
estimated the effect of supplementary feeding program. We discuss a strategy to deal with this
issue later in empirical section.

33 Despite the significant role for supplementary food program delivery, some services in
Posyandu experienced a decline during period 1997-2000. Strauss et a1 (2004) find that provision
of oralit in 2000 decreased by 9.4 percent while child growth monitoring services were 14.1
percent lower relative to 1997.

34 In addition to children under 5, the program also targeted pregnant women, particularly those
women from poor households. Some services provided to this sub-group included pregnancy
monitoring and provision of pregnancy vitamins.

68

Across three sub-groups of children under five years of age, the program
differed in terms of quantity, content of diet and the frequency of feedings
provided to targeted groups. For infants (6-11 months), supplemental diets were
provided in the form of soft meals to supplement breast milk. The nutritional
composition per 100 grams of food included 360-430 kcal of energy and 10-15
grams of protein, and this was provided 3 to 4 times a day for 180 consecutive
days. Young children (12-23 months) and children between 24-59 months
received a locally prepared snack with a nutritional composition that included
energy (360-430 kcal) and protein (9-11 grams) for 90 consecutive days. The
difference in the program between the two older groups was that for younger
children, a snack was provided three to four times a day, and for older children it
was only provided once a week.

The program was introduced in early 1998 but not all communities
received it at the same time. The program coverage, as indicated in Table 1, was
low at the beginning of the crisis (the end of 1997 and early 1998) because the
crisis was not anticipated in the 1997-1998 fiscal budget.35 It was only during the
following fiscal year (1998-1999) that coverage of the program expanded above

70 percent, and then further expanded to over 90 percent in 1999-2000, before it

 

35 The national budget plan (fiscal year) in Indonesia prior to year of 2001 ran from April 1 to
March 31. After 2001, the government of Indonesia adjusted change the fiscal year to correspond
to the calendar year.

69

declined to 82 percent during the first half of fiscal year 2000-2001.36 Overall,
during the 1998-2000 period, most of the sampled communities (94.7 percent)

were exposed to the program.

3.5. Data and variables measurement

The data used in this study come from two rounds (1997 and 2000) of
Indonesia Family Life Survey (IFLS), which covers periods before and after the
1997-1998 Asian Financial Crisis and implementation of child supplementary
food program. The IFLS is a panel survey that collects data on various aspects of
individual, household and family livelihood. The survey also collects
information about facilities and conditions in the communities where households
and individuals reside, including important characteristics of the socio-economic
environment, physical infrastructure, and health and education facilities.37 In this
study we link comrnunity-level data on implementation of the supplementary
feeding program since the beginning of the crisis in 1998 with individual-level
data on health and nutritional status of children.

Our sample for this study includes all children who were between 6 and
59 months in 1997 and 2000, and lived in IFLS sampled communities (2170 and

2618, respectively). The average age of these children in 1997 and 2000 was 34.12

 

36 Pritchett et a1 (2002) mention that the initial fiscal year 1998-1999 did not include a post for
emergency programs to broadly cope with negative impact of crisis. It was only July 1998 that the
ongoing budget was revised and thus contained an item for Social Safety Net Program.

37 Sampling and survey methods for IFLSZ and IFLSB can be found in Frankenberg, Beegle and
Thomas (1999) and Strauss et al (2004b), respectively.

70

and 32.67 months, respectively. Information from the community and facility
survey of the IFLS indicates that fewer than 4 percent of the sampled
communities provided program benefits to non-targeted groups (non-pregnant
women, children 5-14 years old and other adults). For this reason, we believe that
the proportion of non-targeted children who were exposed to the program is
negligible.

We use child standardized height-for-age as our measure of nutritional
status. The health literature suggests that, in terms of measurement error, this is a
less problematic indicator for child health and nutritional status than other
alternative health measures.38 The use of a standardized height-for-age measure
to represent health and nutritional status outcomes is appropriate as we believe
that the impact of crisis, if any, should be evident using a measure that captures
longer-term impacts.

In order to focus on more extreme effects, we also examine the proportion
of children who suffer from stunting, or who have standardized height-for-age
measures of less than negative 2. This measure represents deficit in standardized
height-for-age (less than or equal to -2) and often refers in medical literature as
shortness. We calculate standardized height-for-age using a STATA ado program
called zanthro (Vidmar et a1, 2004). In zan thro, standardized height-for-age are

calculated by comparing the actual child height in the IFLS sample with those

 

33 In IFLS, height and weight are measured by trained health workers with regularly calibrated
health equipment. We believe that enumeration protocols helped to limit measurement error for
this variable.

71

from a US reference population found in the 2000 Center for Disease Control and
Prevention (CDC) Growth Charts.

In Table 2, we show the mean child standardized of height-for-age and
proportion of stunting in 1997 and 2000 and change in these statistics over the
two survey rounds. We observe a positive but insignificant change in child
standardized height-for-age from 1997 to 2000 among children between 6 and 59
months of age. A stronger improvement in nutritional status was found in a
significant reduction in stunting.39 The reduction in stunting, however, is only
significant for the older groups (12-23 months and 24-59 months). Specifically,
the proportion of 12-23 months and 24-59 months children who were stunted
decreased by 8.3 and 2.8 percent, respectively, from 1997 to 2000. We found no
change in the stunted share of children from 6-11 months of age. This descriptive
underscores findings documented by Strauss et al (2004), which suggested that
nutritional status of children 6-59 months of age did not deteriorate during the
crisis period.

As discussed above, a simple binary program variable to measure
program impact may be inadequate for two reasons. First, as shown in Table 1,
the program was found in majority of communities, and use of a program

dummy (with 1 indicating access to the program, and 0 otherwise) may not

 

39A child exhibits stunting if his or her standardized height for age is less than or equal to
negative two.

72

guarantee sufficient variation between those with access to the program and
those who lack it.

In addition, the use of a binary program exposure variable in an
evaluation exercise implicitly assumes that program was implemented with
homogenous exposure on its targeted subjects within and across communities. If
a program was implemented with different exposure across communities, the
use of a binary program exposure dummy may lead us to underestimate the
program impact. Sandjaja et al (2001) and the IFLS community surveys
document differences in content, quality and duration of the program across
communities.

To better identify program impact, we also use an alternative measure
better suited to capturing heterogeneity in program exposure across targeted
individuals. Specifically, we use proportion of life in which a child was exposed
to the program. The measure is generated by combining the age of children who
lived in the treatment villages and length of time that the program existed in the
community. The prOgram length is generated by using information on the start

and end dates of the program in each community. 40,41

 

40 In IFLS3, community informant was asked about details of program implementation, including
the beginning and the ending date of the program, the kinds of food given to the recipients, the
program administrator, and other information.

41One might use program length itself as other alternative measure. However there are two issues
when estimating program effect using this measure. First, it assigns the same length of exposure
value to each child living in the community regardless of age. Thus, this measure fails to pick up
variation in program exposure among targeted children living within similar community.
Second, program length is very likely to be endogenous as the government might systematically
distribute or phase in the program based on some criteria that were not observed by program
evaluator.

73

We follow an approach first introduced by Rosenzweig and Wolpin (1986)
to combine information on variation in child age and the timing of program

introduction to construct an index measure of program exposure. In particular,

let (ts a 1,, 9 1),) represent date of the survey, date in which the program is first

introduced in community v, and birth date of a child, respectively. We assume
that the program will have different exposure across children depending on the
age of children and the duration of the program in the community. For example,

a child born before the introduction of the program will be exposed for ts-tp

years. Letting P M T): represent program exposure of child 1' in community I)

who was age a at the survey date, where a=ts-tb, we calculate the program

exposure variable as:

PMT): =0 if the program does not exist in the community 0 when the
survey was held,

=((t_, —tp)/a) if I, 2 tp 2 t), in community 1),
= 1 if t, 21), 21,, in community 0.
This variable thus measures the proportion of a child’s life with exposure
to the program and varies between 0 and 1 and this varies across children. ”1”
indicates that a child has been exposed for his or her entire life, or 100 percent

exposure. This occurs when a child was born after the introduction of the

program and when the program was still active at the time of the last survey

 

74

(2000). Therefore, only children who were born in 1998 or afterward, or those
who were between 6 to 35 months old, could have full program exposure. In
contrast, ”0" indicates that a child was not exposed at all to the program. This
value is assigned to children who lived in a community without the program in
1997 and 2000 or those who lived in community without the program but were 5
years old or older in 1998. Values between 0 and 1 reflect children who were
exposed for some part of their life, but were born at least six months before the
introduction the program. Table 3 shows mean program exposure for each of the
targeted groups which are also disaggregated into various sub-groups for the
2000 sample.

We also control for other covariates to isolate the effect of the program on
child nutritional status. These covariates cover biological, socio-economic and
demographic characteristics that are likely to be correlated with child nutritional
status. Other than child age and gender, we control for mother’s height and
education, and per capita expenditure as well as additional community
characteristics (see Table 4 for a list of community characteristics). Table 4 shows
descriptive statistics for important household and community level

characteristics.

3.6. Empirical Identification and Specification
We start with standard econometric specification to estimate the effect of

supplementary feeding program on child nutritional status:

75

HM = a0 + a1]; + azPMTv" + ,BPMTf * T, + 8M (6)
where H is the health outcome variable which is the nutritional status (using
anthropometric measures) of child 1' living in community v at time t, T is dummy
variable for time period, and PMT is a program exposure variable (which may be
binary or continuous in alternative specifications).

Our parameter of interest in equation 6, [3, is a difference-in-difference
estimator which measures the impact of the program on the treatment group.
When using a binary program variable, the treatment group is comprised of the
communities and villages that received the program. The control group is all
children residing in a village or community without the program. Therefore [3
measures change of nutritional status among children under 5 who lived in
treatment villages relative to those who lived in control villages. Because nearly
all communities were exposed to the program for some amount of time after 1998

and none of communities in 1997 received it, [3 essentially compares nutritional

status of children in 2000 relative to 1997.

Other parameters in (6) are also of potential interest. al measures any
common change in child nutritional status that is not associated with the
program. One of the effects captured by a] will be the effect of the crisis, which
we expect to be negative. Meanwhile a2 will pick up features of communities

that may affect nutritional status independent of the program itself.

76

The estimated program effect from (6), however, is likely to be biased.
Program allocation may be correlated with omitted characteristics of
communities receiving the program. Even worse, many of these omitted
variables are unobserved by the analyst. To deal with this issue, one approach is
to control for unobserved heterogeneity by including community fixed-effect.
This procedure implicitly assumes that unobserved factors that are correlated
with the program are all at the community level and are time-invariant. In
addition we include household characteristics, Z", and time-varying community
characteristics, Z“, to increase the efficiency of the parameter of interest.
Including community fixed-effect and some community and household-level

covariates, suggests the following model:

HM = a0 + a,T, + ,BPMT,“ * T, + 6,Z,.’,’,, + 5,25, + p, + 5,, (7)

Controlling for a community fixed-effect in this model, will control for
PM T placement and other time-invariant community characteristics. We control
for placement through the fixed effect, and then focus our attention on [3, our
main parameter of interest.

The use of binary program variable in our case may not accurately capture
program effect. This is because the program, as previously shown, was
heterogeneously exposed to the communities. Simplifying and treating it as it
was homogenously distributed across communities implies that all targeted

children lived in the communities with program would be fully exposed to the

program irrespective the length of program has been in the community and the

77

age of children. The estimated parameter of interest when using binary program
variable provides an upper bound program effect.

We then propose an alternative measure of program variable. Information
on the starting and ending dates of the program together with child’s birth place
and birth date are used to calculate the proportion of child age exposed to the
program and to identify how exposure to the program affects nutritional status
of children. We expect, ceteris paribus, that the higher the proportion of child’s
age exposed to the program, the greater the improvement in nutritional status
and the less likely that targeted children showed stunting in 2000.

The estimation procedure when using proportion of child’s age that is
exposed to the program as program variable should principally be similar with
the estimation using binary program variable. The only difference is that the
program variable is no longer a simple binary program indicator, but instead
ranges from 0 to 1. In addition, this alternative program variable has the benefit
of varying by child rather than community. This alternative program variable
provides us with a strategy for identifying program effects while also controlling
for community-level unobserved heterogeneity.

After controlling for (potential) endogenous program placement, a time
dummy (to control for common change in program outcome across periods), Zh
and 26, our identifying assumption in estimating the program effect is that
unobserved factors captured in the error term 8 are uncorrelated with our

program effect variable. Given that assumption holds, [3 thus indicates the

78

effectiveness of the program. Prior to estimating equation 7, we also investigate
what factors affect allocation of program and program intensity across
communities. To do so, we estimate determinants of program duration at the

community level.

3.7. Results and Analysis
3.7.1. Determinants of Program Duration

As noted above, the program was found in almost all communities during
the period from 1998 to 2000, but we observe variation in program length across
communities. In this section, we ask what factors, if any, affect variation in
program exposure across communities. We estimate the program length on all
1997 covariates, including health status, socio-economic status, health and
physical infrastructure as well as some variables that proxy remoteness of the
community using pooled OLS. We also include district fixed-effect to control for
district-level unobserved heterogeneity that may have affected timing of
program implementation.

Estimation results are shown in Table 5. Without measures of community
nutritional status in the specification (model 1 of table 5), three variables
assOciated with remoteness of the community appear to be statistically different
from zero. In addition, joint significance tests on the group of variables
associated with remoteness confirm the importance of these factors for allocation

of the program. The signs on coefficients of two variables, distance from district

79

capital and availability of public transportation, imply that more remote
communities were exposed to the program for a longer period of time. The
coefficient on another indicator, whether the community is urban, indicates that
the urban community tends to receive longer program exposure relative to rural
communities between 1998 and 2000.

If program allocation during 1998-2000 was based entirely on 1997
characteristics as modeled here, the finding that urban communities received
longer treatment would be inconsistent with the fact that nutritional status in
urban communities in 1997 was better than in rural areas (see figure 3.1 and 3.2).
However the allocation procedure might also be based on the community
characteristics in 1998 and 1999, or changes from 1997, which we cannot observe
in the IFLS.

We then include two different measures of nutritional status in model 2
and 3, one is 1997 average height-for-age of children under 5 and the other is
proportion of children living in the community in 1997 who were stunted. Our
three measure of remoteness have consistent signs and significance when we
include community average indicators of child health. The coefficient on 1997
average height-for-age has a negative sign, which implies that communities with
healthier children tended to receive the program for a shorter amount of time
than communities where average nutritional status was worse, but the coefficient

is not significantly different from zero. When using the share of children who

80

were stunted in the community, however, we find a strong and significant
predictor of program duration.

Descriptive evidence shown in Figures 3.1, 3.2, 3.3 and 3.4 provides
additional information on how program was distributed. Figure 3.1 plots the
average height-for-age distribution for urban and rural communities in 1997. It
shows that over most of the distribution, average child height-for-age is greater
in urban communities. There appear to be crossing points of the two curves in
the lower and upper tail when average height-for-age z-scores are -2.2 and -0.2,
respectively, and this indicates that in the lower tail average health status may be
worse for some urban communities. Figure 3.2 enlarges the lower tail of figure
3.1 and it shows, consistent with figure 3.1, that proportion of stunted children
was higher in rural communities than urban ones except for upper part of the
distribution. Using per capita expenditure, we also find that urban households
have a higher per capita expenditure in 1997 than rural households (Figure 3.4).

Given these distributions of child nutritional status measures and per
capita expenditure in 1997, one might expect that rural communities would be
exposed to the program for a longer period. But figure 3.4, supports our finding
in table 5, and shows that up to the 20 month point, urban communities were
exposed to the program for more time during the 1998-2000 period. The
exception is probably in the upper part of distribution, to the right side of the
crossing point at 20 months, where the distribution of program length was

essentially equal. These findings imply that the program might be properly

81

distributed only to communities with severe malnutrition problems. In
communities with less severe nutritional problems, the program seemed to be
biased toward urban rather than rural communities.

The observable characteristics may not include enough information to
conclude that the program was properly targeted. There may be other
information that cannot be observed and controlled for with our data here that
explain why program length distribution tended to be urban biased. For
example, Frankenberg et a1 (1999) use data that capture changes in people’s well-
being at the beginning of the crisis and find that urban communities experienced
larger increases in the proportion of households living below the poverty line.42
Yet they also find that during 1997-1998, there was little evidence showing that
child health and nutritional status had deteriorated significantly, and this is
consistent with evidence cited by Block et a1 (2004) and Strauss et al (2004). They
do not show how changes in child health and nutritional status vary between

urban and rural communities.

3.7.2. The effect of supplementary feeding program.
In order to control for the different benefits and protocols across the three
targeted age groups, we perform our analyses separately for children who are 6-

11 months, 12-23 months and 24-59 months. Table 6 presents estimation results

 

42 Frankenberg et a1 (1999) use IFLSZ and IFLSZ+ which cover the period before and the
beginning of the crisis -up to November 1998.

82

for the effect of the program on child nutritional status using a binary indicator
for presence of the program in 1998 and 2000.43 We first present pooled OLS and
then include community fixed-effect models to control for unobserved
community characteristics systematically related to program implementation.
The results shown in Table 6 suggest that the sign for estimated effects of the
program on child standardized height-for-age, except for infant 6-11 months, are
negative, but not statistically significant. The inclusion of community fixed-
effect, not community characteristics, largely increases the coefficient of program
effect as well as R-squared and therefore F-test score, but the program effect
remains statistically insignificant. With almost all sampled communities
receiving the program during 1998-2000, and very little variation in program
exposure, it is not surprising that we find an insignificant program effect. In
addition, the coefficient on the year dummy is negative, an indication of how the
1997/ 98 economic crisis may have affected child nutritional status, but the effect
is statistically insignificant.

Next, we estimate models examining program impact on probability of
stunting (height-for-age is less than or equals to -2) in 2000 using the binary
program variable. The signs on the parameter of interest are consistent with
initial expectation except for the group of children 24-59 months. The program

only appeared to be effective in lowering the probability of stunting in 2000

 

43'Table 6 presents a summary of results, the full set of parameter estimates are shown in
Appendix Tables A7.2.1 and A723.

83

among infant 6-11 months when we control for unobserved heterogeneity using
with community fixed effects. In particular, community fixed-effect estimates
(model 3 of Table 7) indicate that exposure to the program reduced the
probability of stunting in 2000 by 33 percent. Meanwhile when the community
fixed-effect estimates include time-varying community characteristics in the
specification, the exposure to the program lowers the probability of being
stunted in 2000 by 39 percent. However this estimated effect is likely to
overestimate the true effect since binary program variable when program exists
(PMT=1) implies a full exposure to the program for all targeted children living in
the community with the program.

We next use fraction of child age exposed to the program to capture
heterogeneity in program exposure to identify the program effect. We argue that
this program variable measures program effect more accurately as this measure
takes into account the heterogeneity of program exposure across communities
and variation in potential exposure among targeted children. Results of
estimated effects on height for age are shown in Table 8, and the full estimation
results are presented in Tables A3.7.2.7 and A3.7.2.9. Estimation using pooled
OLS (with and without time-varying community characteristics in the
specifications) yields parameters of interest with mixed signs, but all are
statistically insignificant (model 1 and 2 of table 8). When including community
fixed-effects, our estimation produces positive and significant program effects on

height-for-age of children 12-23 months of age (panel B of table 8). In particular,

84

our results suggest that exposure to the program increased child standardized
height-for-age between 0.48 and 0.55 standard deviations, depending on whether
' or not time-varying community characteristics are included. The estimated
effects are statistically significant at the 5 percent level. The estimated effect of
the program on children 24-59 months is negative and statistically insignificant,
which is also what we found when using the binary program indicator. Our
findings for infants between 6 and 11 months are negative and insignificant
when including community fixed-effects and this is inconsistent with findings
using the binary program variable. These findings, however, make sense since
estimation using binary program variable is very likely to overestimate the
program effect since it implicitly assumes that all targeted individuals were fully
exposed during all their life span in the exposed communities.

We next use exposure to the program to evaluate the effectiveness of the
program on reducing the probability of children being stunted in 2000 relative to
1997. We find that the program helped children with severe malnutrition
problems during the 1997-2000 period, but only the children in 12 to 23 month
age group. In particular, exposure to the program reduced the probability of
being stunted in 2000 by about 15 percent, and this effect is statistically
significant at the 10 percent level. Unlike in previous estimations, the inclusion of
community characteristicsconly had a slight impact on the magnitude of the
effect. On the other hand, we also found that the program did not reduce

stunting in children in the 6-11 month and 24-59 month age groups.

85

We draw additional comfort in our results because other parameter
estimates, such as coefficients on child age and parent's human capital
(education and health), are consistent with findings elsewhere in the economics
literature (tables A3.7.2.1-A3.7.2.12). In particular, age (in months) exhibits a non-
linear relationship with health and nutritional status of children. There is a
negative relationship only until children reach 24 months and then this
relationship is positive as children age beyond 24 months. Education of parents,
particularly mother’s education, is also important for improving nutritional
status of children, but only for older children (24-59 months). Although we see a
positive association between father’s education and the nutritional status of
children, the relationship is not statistically different from zero. On the other
hand, parent’s height (in cm), which proxies for parent well-being when in early
childhood and cognitive capacity, is positively associated with child nutritional
status. The effect is strong across models and specifications regardless type of

program variable used in the model.

3.8. Summary and Conclusions

This paper evaluates the effectiveness of a supplementary feeding
program for maintaining child nutritional status through the period of the 1997-
1998 economic crises in Indonesia. Since the program was nearly universal in its
implementation, use of a simple binary indicator to measure program exposure

may not allow us to accurately capture the program effect. With access to

86

detailed information on program implementation, we exploit heterogeneity in
program exposure to evaluate the program effect.

The use of program heterogeneity has at least two advantages for
identifying the effect of Indonesia’s supplemental feeding program. First use of
program exposure allows us to estimate an effect even with low variation of
program distribution across targeted subjects. Second, use of the program allows
us to avoid the strong assumption that all targeted children experienced
homogenous exposure to the program and therefore allows us to capture the
program effect more accurately than when using a binary program indicator.

We show that although the program tended to be universally distributed
during 1998-2000, the distribution of program length varied across communities.
We also provide insight into the government’s implicit allocation rule when
determining program length. Community-level average child health outcomes
are important factors influencing duration of exposure to the problem.

Findings on the effect of program exposure show that the program
improved the nutritional status of children 12 to 23 months of age during the
period of economic crisis, and in particular, improved standardized height-for-
age by an average of 0.48 to 0.55 standard deviations. Given the increase of
standardized height-for—age for all sampled children between 1997 and 2000 was
only 0.04, this implies that the crisis would have otherwise led to a negative

impact on child nutritional status if the program was not launched. Our findings

87

also suggest that the program helped children with severe malnutrition
problems, but again the impact was limited to those who were 12-23 months.

Putting this finding together with results from targeting regression
suggests that program could have been more effective and had significant impact
on broader targeted groups if it was targeted more properly to communities with
worse health problem. Since we know, for example, that the nutritional status of
rural children was on average lower than their urban counterparts in 1997, the
program would have probably had more impact if it was allocated more to rural
communities.

Our results also highlight the potential benefit of using heterogeneity in
program exposure when evaluating a universally targeted program. The use of
program intensity is not just helpful to capture program effects as it can also be
used to in conjunction with community fixed effects in a way that avoids bias
from endogenous program placement. Such benefits suggest that future surveys

continue to collect information on program implementation.

88

Table 3.1. Distribution of PMT Coverage Across Communities, 1998-2000
Total Urban Rural

 

Communities with PMT from 1998-2000 (%) 94.68 56.48 38.21

In the 1998/1999 fiscal year (%) 71.10 43.52 27.57
In the 1999/ 2000 fiscal year (%) 92.69 56.81 35.88
In the 2000/ 2001 fiscal year* (%) 82.72 49.17 33.55

 

Note: Number of communities in the sample is 301. (*) indicates the fiscal year
ended in December.

89

Table 3.2. Standardized Height-for-Age and Incidence of Stunting:
Children between 6 and 59 months in 1997 8: 2000

 

 

 

 

 

1997 2000 Diff

6-11 Months

Mean -0.96 -0.90 0.06
(0.12) (0.07) (0.13)

Stunted (%) 20.29 16.67 -3.62
(0.03) (0.02) (0.03)

Observations 207 318

12-23 Months

Mean -1.66 -1.62 0.04
(0.06) (0.05) (0.08)

Stunted (%) 36.32 28.05 -8.28**
(0.02) (0.02) (0.03)

Observations 468 574

24-59 months

Mean -1.60 -1.56 0.04
(0.03) (0.03) (0.04)

Stunted (%) 35.12 32.33 -2.79*
(0.01) (0.01) (0.02)

Observations 1726

1495

 

Notes: Calculated from IFLS data. Standard errors are in parentheses. (*) is
significant at 10%; (**) is significant at 5%; (***) is significant at 1 %.

90

Table 3.3. Mean Child Exposure to Supplementary Feeding Program, Children 6-

 

 

 

59 Months in 2000.
All groups Gender Community
Boy Girl Urban Rural
6-11 months 0.34 0.34 0.33 0.36 0.32
Observations 4788 271 254 230 295
12-23 months 0.26 0.27 0.24 0.30 0.23
Observations 1576 526 516 474 568
24-59 months 0.14 0.14 0.14 0.14 0.14
Observations 3221 1625 1596 1364 1857

 

Notes: Calculated from IFLSB data.

91

Table 3.4. Descriptive Statistics

 

 

Variable Mean Std. Dev.
Mother's education 6.38 6.15
Father’s education 6.65 4.13
Mother's height 150.27 5.13
Father’s height 160.77 6.24
Gender of HH head (male=1) 0.88 0.32
Main activity of HH (farm=1) 0.40 0.49
HH access to private toilet (Y es=1) 0.57 0.49
HH access to sanitation (Yes=1) 0.19 0.39
Per-capita expenditure (’0005 RP) 215.2 728.3
Type of community (urban=1), C 0.43 0.50
Distance of com. to bus station (km), C 4.23 7.09
Prop. of land with technical irigation, C 0.08 0.18
Asphalt road in the com (Y es=1), C 0.76 0.43
Village has sewerage (Yes=1), C 0.54 0.50
Village has piped water (Yes=1), C 0.56 0.50

 

Notes: Calculated from IFLS3 data (C), Indicates Community-Level Variables

92

Table 3.5. Determinants of Program Length across Communities 1997-2000(pooled OLS)

 

 

 

 

 

 

Model (1) (2) (4)
Health Status
Average Height-for-Age - -0.87 -
(0.90)
Proportion of Children Stunted (% - - 4.61*
(2.64)
Sociol-economic status
Average Per-Capita Expenditure -0.91 -0.7 -0.47
(1.16) (1.19) (1.20)
Farm Household Share (%) -3.01 -2.72 -2.73
(2.77) (2.80) (2.76)
Share of Households with Male Head (%) 1.43 1.28 1.35
(2.12) (2.10) (2.10)
Physical Infrastructure
Availability of Asphalt Road (Yes=1) 2.1 2.02 2.07
(1.77) (1.76) (1.76)
Proportion of Land with Semi-Tech Irrigation (%) -4.25 -4.18 -4.7
(4.02) (3.96) (3.88)
Access to piped water (yes=1) 0.12 0.15 0.15
(1.39) (1.38) (1.38)
Remoteness of communiy
Distance to District Capital (km) 004* 004* 004*
(0.02) (0.02) (0.02)
Distance to Bus Station (km) -O.14 -0.14 -0.15
(0.10) (0.10) (0.10)
Public Transportation in Community (Yes=1) -4.30*** -4.30*** -4.39***
(1.53) (1.52) (1.52)
Urban Community (Yes=1) 2.61* 2.75* 2.78*
(1.56) (1.56) (1.55)
District Fixed-Effect Yes Yes Yes
R-squared 0.24 0.25 0.25
F-test (economic status) (p-value) 0.79 0.59 0.53
(0.50) (0.62) (0.66)
F-test (infrastructure) (p-value) 0.80 0.78 0.90
(0.49) (0.51) (0.44)
F-test (remoteness of community) (p-value) 3.18 3.18 3.26
(0.01) (0.01) (0.01)

 

Notes: Dependent variable is number of months the program was in the community between
1998-2000. Independent variables are community averages of 1997 covariates. Standard errors are

in parentheses. (*), (**) and (***) indicate significance respectively at 10%, 5% and 1%.

93

Table 3.6. The Effect of PMT on Height-for-Age of Children 6-59 Months: Binary
Program Indicator

 

 

Pooled Community
OLS Fixed-Effects
(I) <2) (3) (4)
A. 6-11 Months
Year Dummy (2000=1) -0.22 -0.301 -0.41 -0.516
(0.395) (0.397) (0.702) (0.707)
Program Since ’98 (Yes=1)*Year 0.248 0.365 0.51 0.62
(0.384) (0.387) (0.711) (0.723)
Community Characteristics Included No Yes No Yes
R-squared 0.14 0.16 0.57 0.58
B. 12-23 Months
Year Dummy (2000=1) 0.027 -0.02 0.09 0.099
(0.230) (0.234) (0.239) (0.256)
Program Since ’98 (Y es=1)*Year -0.108 -0.052 -0.073 -0.06
(0.225) (0.230) (0.265) (0.280)
Community Characteristics Included No Yes No Yes
R-squared 0.08 0.08 0.40 0.40
B. 24-59 months
Year Dummy (2000=1) 0.074 0.102 0.114 0.126
(0.136) (0.137) (0.186) (0.186)
Program Since ’98 (Y es=1)*Year -0.135 -0.157 -0.138 -0.153
‘ (0.133) (0.134) (0.190) (0.191)
Community Characteristics Included No Yes No Yes
R-squared 0.15 0.15 0.27 0.27

 

Note: The dependent variable is the child height for age z-score, and the program effect is a
dummy for whether the community had a supplemental feeding program interacted with a
dummy for year=2000. Covariates include (but not reported in this table) mother education and
height, male head of household, farm household, per capita expenditure, household access to
private toilet, sanitation and free health services, distance of village to bus station, community’ 5
access to sewerage and piped water, number of village midwife and posyandu in the community,
and type of community. Robust standard errors are in parentheses. (*), (**) and (***) indicate
significance respectively at 10%, 5% and 1 %.

94

Table 3.7. The Effect of PMT on Probability of Stunting Among Children 6-59 of Age:
Binary Program Indicator

 

 

Pooled Community
OLS Fixed-effects
(1) (2) (3) (4)
A. 6-11 Months
Year Dummy (2000=1) 0.07 0.085 0.223 0.277
(0.107) (0.108) (0.199) (0.204)
Program Since ’98 (Yes=1)*Year -0.124 -0.144 -0.332 -0.399
(0.104) (0.105) (0.196)* (0.202)**
Community Characteristics No Yes No Yes
R-squared 0.10 0.11 0.53 0.54
B. 12-23 Months
Year Dummy (2000=1) -0.076 -0.058 0.02 0.044
(0.087) (0.088) (0.149) (0.152)
Program Since ’98 (Yes=1)*Year 0.019 -0.007 -0.099 -0.123
(0.085) (0.087) (0.154) (0.158)
Community Characteristics No Yes No Yes
R-Squared 0.08 0.09 0.38 0.38
B. 24-59 Months
Year Dummy (2000=1) -0.008 -0.014 -0.037 -0.039
(0.053) (0.054) (0.071) (0.071)
Program Since ’98 (Yes=1)*Year 0.013 0.019 0.045 0.049
(0.052) (0.053) (0.072) (0.072)
Community Characteristics No Yes N 0 Yes
R-squared 0.11 0.12 0.22 0.22

 

Note: There are 4788 observations. The dependent is an indicator for whether the child has a
height for age 2 score less than -2, and the program effect is a dummy for whether the community
had a supplemental feeding program interacted with a dummy for year=2000. Covariates
include (but not reported in this table) mother education and height, male head of household,
farm household, per capita expenditure, household access to private toilet, sanitation and free
health services, distance of village to bus station, community’s access to sewerage and piped
water, number of village midwife and posyandu in the community, and type of community.
Robust standard errors are in parentheses. (*), (**) and (***) indicate significance respectively at
10%, 5% and 1%.

95

Table 3.8. The Effect of PMT on Height-for-Age of Children 6-59 Months:
Proportion of Age Exposed to the Program

 

 

Pooled Community
OLS Fixed-Effects
41) (2) (3) (4)
A. 6-11 Months
Year Dummy (2000=1) 0.044 0.118 -0.036 -0.052
(0.172) (0.173) (0.343) (0.358)
Age Exposed to the Program*Year -0.048 -0.128 0.219 0.24
(0.187) (0.190) (0.477) (0.501)
Community Characteristics No Yes No Yes
R-Squared 0.14 0.16 0.57 0.58
B. 12-23 Months
Year Dummy (2000=1) -0.093 -0.089 -0.213 -0.227
(0.104) (0.105) (0.145) (0.152)
Age Exposed to the Program *Year 0.035 0.04 0.482 0.552
(0.125) (0.126) (0.229)“ (0.237)**
Community Characteristics No Yes No Yes
R-Squared 0.08 0.09 0.41 0.41
B. 24-59 Months
Year Dummy (2000=1) -0.023 -0.009 0.009 0.002
(0.058) (0.058) (0.071) (0.073)
Age Exposed to the Program *Year -0.12 -0.143 -0.095 -0.072
(0.117) (0.118) (0.171) (0.174)
Community Characteristics No Yes No Yes
R-Squared 0.15 0.15 0.27 0.27

 

Note: Dependent variable is child height for age z-score. Covariates include (but not reported in
this table) mother education and height, male head of household, farm household, per capita
expenditure, household access to private toilet, sanitation and free health services, distance of
village to bus station, community’s access to sewerage and piped water, number of village
midwife and posyandu in the community, and type of community. Robust standard errors are in
parentheses. (*), (**) and (***) indicate significance respectively at 10%, 5% and 1%.

96

Table 3.9. The Effect of PMT on the Probability of Stunting Among Children 6-59
Months: Proportion of Age Exposed to the Program

 

 

Pooled Community
OLS Fixed-Effects
(1) (2) (3) (4)
A. 6-11 Months
Year Dummy (2000=1) -0.052 -0.06 -0.064 -0.061
(0.047) (0.047) (0.100) (0.103)
Age Exposed to the Program *Year 0.006 0.012 -0.061 -0.08
(0.051) (0.052) (0.130) (0.134)
Community Characteristics N 0 Yes No Yes
R-Squared 0.09 0.11 0.52 0.53
B. 12-23 Months
Year Dummy (2000=1) -0.072 -0.079 0.003 0.006
(0.039)* (0.040)** (0.053) (0.056)
Age Exposed to the Program *Year 0.028 0.028 -0.154 -0.155
(0.047) (0.048) (0.082)* (0.084)*
Community Characteristics No Yes No Yes
R-squared 0.08 0.09 0.38 0.38
B. 24-59 Months
Year Dummy (2000=1) -0.009 -0.01 -0.014 -0.01
(0.023) (0.023) (0.027) (0.028)
Age Exposed to the Program *Year 0.049 0.053 0.075 0.062
(0.046) (0.046) (0.068) (0.069)
Community Characteristics No Yes No Yes
R-Squared 0.11 0.12 0.22 0.22

 

Note: There are 4788 observations. The dependent is an indicator for whether the child has a
height for age 2 score less than -2. Covariates include (but not reported in this table) mother
education and height, male head of household, farm household, per capita expenditure,
household access to private toilet, sanitation and free health services, distance of village to bus
station, community’s access to sewerage and piped water, number of village midwife and
posyandu in the community, and type of community. Robust standard errors are in parentheses.
(*), (**) and (***) indicate significance respectively at 10%, 5% and 1%.

97

Cumulative Probability

 

 

 

I I l l
-3 -2 —1 0
Average height-for-age in community, 1997

A

 

 

——— Rural Urban

 

 

 

Figure 3.1. CDF of 1997 Community Average Height-for-Age by Location (Rural
vs Urban) of Community

98

 

 

Cumulative Probability

 

 

 

0-
I T I I l T
0 .2 .4 .6 .8 1
Proportion of stunted children in community, 1997

 

 

 

 

— — - Rural Urban

 

Figure 3.2. CDF of 1997 Stunted Share of Children in Communities by
Community Location (Urban vs Rural).

99

 

Cumulative Probability

 

 

 

I I I l
10 11 12 13 14
Community average household per capita expenditure, 1997

Urban

 

 

 

——— Rural

 

 

Figure 3.3. CDF of1997 Average of Household per Capita Expenditure by
Community Location.

100

 

Cumulative Probability

 

 

 

I I I I I
O 10 20 30 40
Length of Program (in months), 1998-2000

 

 

— — - Rural Urban

 

 

 

Figure 3.4. CDF of Length of Program in Month by Community Location (Urban
vs Rural), 1998-2000.

101

APPENDICES

102

Table A3.7.2.1. Impact of PMT on height-for-age of children 6-11 months: binary
program variable

 

 

OLS Fixed-effect
Year (2000=1) -0.22 -0.301 -0.41 -0.516
(0.395) (0.397) (0.702) (0.707)
Program since ‘98 (Yes=1)*Year 0.248 0.365 0.51 0.62
(0.384) (0.387) (0.711) (0.723)
Gender (boy=1) 0.103 0.112 0.001 0.002
(0.124) (0.124) (0.198) (0.208)
Age -0.127 -0.126 -0.171 -0.171
(0.036)*** (0.035)*“‘ (0.057)*** (0.058)***
Mother education (years) 0.001 0.001 -0.024 -0.024
(0.015) (0.015) (0.023) (0.024)
Father education (years) 0.008 0.01 0.014 0.012
(0.015) (0.015) (0.022) (0.022)
Mother height (cm) 0.061 0.059 0.058 0.055
(0.012)*” (0.012)*** (0.019)“ (0.019)*“
Father height (cm) 0.047 0.049 0.07 0.069
(0.012)*** (0.012)*‘”‘ (0.019)*** (0.019)***
Head of HH (male=1) -0.027 -0.047 0.272 0.273
(0.222) (0.221) (0.444) (0.463)
Number of children 5-14 yo in HH 0.022 0.019 0.019 0.029
(0.062) (0.062) (0.089) (0.091)
HH owns private toilet (Yes=1) 0.127 0.087 0.167 0.179
(0.136) (0.136) (0.225) (0.233)
HH has sanitation (Yes=1) 0.021 -0.048 0.247 0.31
(0.176) (0.198) (0.403) (0.462)
Per capita income (In) 0.071 0.094 0.039 0.037
(0.094) (0.094) (0.134) (0.138)
Type of community (urban=1), C - 0.336 - 0.065
(0.169)** (0.468)
Distance to district capital (km), C - 0 - -0.002
(0.003) (0.006)
Distance to bus station (km), C - 0.031 - 0.009
(0.011)*** (0.022)
Land w/ technical irrigation (%), C - -0.001 - 0.275
(0.333) (0.840)
Comm. has asphalt road (Yes=1), C - -0.251 - 0.001
(0.163) (0.393)
Comm. has sewerage system (Yes=1), C - 0.075 - -0.294
(0.140) (0.343)
Comm. has piped water (Y es=1), C - -0.009 - -0.127
(0.152) (0.369)
Constant -17.655 -18.02 -20.318 -19.585
(2.558)*** (2.558)“* (3.966)*** (4.050)***
R-squared 0.14 0.16 0.57 0.58

 

Notes: n is 525. (C) indicates community-level variables. Robust standard errors are in
parentheses. (*), (**) and (***) indicate significance respectively at 10%, 5%; and 1 %.

103

Table A3.7.2.2. Impact of PMT on height-for-age of children 12-23 months: binary
program variable

 

 

 

OLS Fixed-effect
Year (2000=1) 0.027 -0.02 0.09 0.099
(0.230) (0.234) (0.239) (0.256)
Program since ’98 (Yes=1)*Year -0.108 -0.052 -0.073 -0.06
(0.225) (0.230) (0.265) (0.280)
Gender (boy=1) -0.136 -0.138 -0.211 -0.206
(0.075)* (0.075)* (0.089)“ (0090)“
Age -0.043 -0.043 -0.033 -0.034
(0.011)*** (0.011)*** (0.012)*** (0.012)***
Mother education (years) 0.005 0.005 -0.008 -0.008
(0.009) (0.009) (0.009) (0.009)
Father education (years) 0.003 0.002 0.001 0.001
(0.009) (0.009) (0.010) (0.010)
Mother height (cm) 0.023 0.022 0.028 0.028
(0.007)*** (0.007)*** (0.008)*** (0.008)***
Father height (cm) 0.031 0.031 0.037 0.037
(0.008)*** (0.008)*** (0.009)*** (0.009)***
Head of HH (male=1) -0.013 -0.002 -0.147 -0.153
(0.123) (0.123) (0.135) (0.138)
Number of children 5-14 yo in HH 0.006 0.011 0.028 0.027
(0.037) (0.038) (0.045) (0.045)
HH owns private toilet (Yes=1) 0.106 0.104 0.12 0.118
(0.081) (0.081) (0.106) (0.107)
HH has sanitation (Yes=1) 0.164 0.141 0.076 0.053
(0.099)* (0.114) (0.161) (0.171)
Per capita income (In) 0.157 0.161 0.105 0.101
(0.061)*** (0.061)*** (0.073) (0.073)
Type of community (urban=1), C - 0.13 - 0.159
(0.104) (0.212)
Distance to district capital (km), C - -0.001 - 0
(0.002) (0.003)
Distance to bus station (km), C - 0.01 - 0.016
(0.006) (0.012)
Land w/ technical irrigation (%), C - 0.393 - -0.28
(0.210)* (0.429)
Comm. has asphalt road (Yes=1), C - -0.118 - 0.006
(0.103) (0.189)
Comm. has sewerage system (Yes=1), C - -0.08 - -0.081
(0.086) (0.165)
Comm. has piped water (Yes=1), C ‘ - 0.077 - 0.005
(0.093) (0.161)
Constant -11.754 -11.677 -12.728 -12.817
(1.696)*** (1.708)*** (1.964)*** (1.991 )***
R-squared 0.08 0.08 0.40 0.40

 

Notes: n is 1042. (C) indicates community-level variables. Robust standard errors are in
parentheses. (*), (**) and (***) indicate significance respectively at 10%, 5% ; and 1%.

104

 

Table A3.7.2.3. Impact of PMT on height-for-age of children 24-59 months: binary
program variable

 

 

 

OLS Fixed-effect
Year (2000=1) 0.074 0.102 0.114 0.126
(0.136) (0.137) (0.186) (0.186)
Program since ’98 (Yes=1)*Year -0.135 -0.157 -0.138 -0.153
(0.133) (0.134) (0.190) (0.191)
Gender (boy=1) -0.01 -0.008 -0.001 -0.002
(0.040) (0.040) (0.041) (0.041)
Age 0.005 0.005 0.004 0.004
(0.002)*** (0.002)*** (0.002)" (0002)“
Mother education (years) 0.017 0.016 0.011 0.011
(0.005)*** (0.005)*** (0.005)“r (0005)“
Father education (years) -0.005 -0.006 -0.003 -0.003
(0.005) (0.005) (0.005) (0.005)
Mother height (cm) 0.054 0.053 0.052 0.053
(0.004)*** (0.004)*** (0.004)*** (0.004)***
Father height (cm) 0.032 0.032 0.032 0.033
(0.004)*** (0.004)*** (0.004)*** (0.004)***
Head of HH (male=1) 0.016 0.02 -0.031 -0.023
(0.067) (0.067) (0.068) (0.069)
Number of children 5-14 yo in HH -0.043 -0.041 -0.019 -0.019
(0.019)*‘Ir (0.019)" (0.021) (0.021)
HH owns private toilet (Yes=1) 0.101 0.087 0.074 0.068
(0.044)“r (0.044)“ (0.048) (0.049)
HH has sanitation (Yes=1) 0.319 0.252 0.311 0.281
(0.055)*** (0.064)*** (0.071)*** (0.075)*“r
Per capita income (In) 0.158 0.155 0.156 0.153
(0.031)*** (0.031)*** (0.035)*** (0.035)"*
Type of community (urban=1), C - 0.138 - -0.037
(0.058)" (0.092)
Distance to district capital (km), C - -0.001 - -0.002
(0.001) (0.001)
Distance to bus station (km), C - 0.002 - 0.002
(0.003) (0.005)
Land w/ technical irrigation (%), C - 0.018 - -0.168
(0.114) (0.182)
Comm. has asphalt road (Yes=1), C - -0.019 - 0.183
(0.053) (0.086)“
Comm. has sewerage system (Yes=1), C - -0.054 - 0.012
(0.047) (0.076)
Comm. has piped water (Y es=1), C - 0.005 - -0.002
(0.049) (0.077)
Constant -17.011 -16.763 -16.716 -16.869
(0.848)*** (0.854)*** (0.933)*** (0.952)***
R-squared 0.15 0.15 0.27 0.27

 

Notes: n is 3221. (C) indicates community-level variables. Robust standard errors are in
parentheses. (*), (**) and (***) indicate significance respectively at 10%, 5%; and 1 %

105

Table A3.7.2.4. Impact of PMT on probability of children (6-11 months) being stunted:
binary program variable

 

 

 

OLS Fixed-effect
Year (2000=1) 0.07 0.085 0.223 0.277
(0.107) (0.108) (0.199) (0.204)
Program since '98 (Yes=1)*Year -0.124 -0.144 -0.332 -0.399
(0.104) (0.105) (0.196)* (0.202)**
Gender (boy=1) -0.025 -0.025 0.021 0.025
(0.033) (0.034) (0.054) (0.056)
Age 0.024 0.023 0.019 0.021
(0.010)“ (0.010)“ (0.017) (0.017)
Mother education (years) 0.003 0.003 0.01 0.01
(0.004) (0.004) (0.007) (0.007)
Father education (years) 0.000 0.000 -0.001 -0.001
(0.004) (0.004) (0.006) (0.006)
Mother height (cm) -0.011 -0.011 -0.01 -0.009
(0.003)*** (0.003)*** (0.005)* (0.005)*
Father height (cm) -0.014 -0.014 -0.021 -0.02
(0.003)*** (0.003)*** (0.005)*“ (0.005)*“
Head of HH (male=1) 0.016 0.014 0.003 -0.005
(0.060) (0.060) (0.116) (0.120)
Number of children 5-14 yo in HH 0.009 0.010 0.008 0.006
(0.017) (0.017) (0.025) (0.025)
HH owns private toilet (Yes=1) -0.046 -0.04 -0.075 -0.074
(0.037) (0.037) (0.058) (0.060)
HH has sanitation (Yes=1) 0.031 0.023 0.019 -0.011
(0.048) (0.054) (0.109) (0.131)
Per capita income (In) -0.029 -0.034 -0.033 -0.039
(0.025) (0.026) (0.039) (0.040)
Type of community (urban=1), C - -0.025 - -0.004
(0.046) (0.128)
Distance to district capital (km), C - 0.000 - 0.001
(0.001) (0.002)
Distance to bus station (km), C - -0.005 - -0.002
(0.003)* (0.006)
Land w/ technical irrigation (%), C - -0.061 - -0.359
(0.091) (0.222)
Comm. has asphalt road (Yes=1), C - 0.033 - 0.001
(0.044) (0.115)
Comm. has sewerage system (Yes=1), C - -0.042 - 0.044
(0.038) (0.098)
Comm. has piped water (Y es=1), C - 0.032 - -0.014
(0.041) (0.088)
Constant 4.119 4.196 5.338 5.103
(0.690)*** (0.696)*** (1.145)*** (1.189)***
R-squared 0.10 0.11 0.53 0.54

 

Notes: 11 is 525. (C) indicates community-level variables. Robust standard errors are in
parentheses. (*), (**) and (***) indicate significance respectively at 10%, 5% ; and 1%.

106

Table A3.7.2.5. Impact of PMT on probability of children (12-23 months) being stunted:

binary program variable

 

Year (2000=1)

Program since ’98 (Yes=1)*Year
Gender (boy=1)

Age

Mother education (years)

Father education (years)

Mother height (cm)

Father height (cm)

Head of HH (male=1)

Number of children 5-14 yo in HH
HH owns private toilet (Yes=1)
HH has sanitation (Y es=1)

Per capita income (In)

Type of community (urban=1), C
Distance to district capital (km), C
Distance to bus station (km), C
Land w/ technical irrigation (%), C
Comm. has asphalt road (Yes=1), C
Comm. has sewerage system (Y es=1), C
Comm. has piped water (Yes=1), C

Constant

R-squared

 

OLS Fixed-effect
-0.076 -0.058 0.02 0.044
(0.087) (0.088) (0.149) (0.152)
0.019 -0.007 -0.099 -0.123
(0.085) (0.087) (0.154) (0.158)
0.042 0.041 0.071 0.075
(0.028) (0.028) (0.034)“ (0.035)“
0.007 0.008 0.002 0.001
(0.004)* (0.004)* (0.005) (0.005)
0.000 0.000 0.004 0.005
(0.003) (0.003) (0.004) (0.004)
0.000 0.001 0.001 0.002
(0.003) (0.003) (0.004) (0.004)
-0.01 -0.01 -0.013 -0.013
(0.003)*** (0.003)*** (0.003)*** (0.003)***
-0.01 -0.011 -0.012 -0.012
(0.003)*** (0.003)*** (0.003)*** (0.003)***
-0.013 -0.015 0.068 0.069
(0.047) (0.047) (0.058) (0.059)
-0.003 -0.005 -0.014 -0.014
(0.014) (0.014) (0.017) (0.017)
-0.053 -0.05 -0.048 -0.043
(0.031)* (0.031) (0.041) (0.042)
-0.111 -0.078 0.009 0.023
(0.038)*** (0.043)* (0.061) (0.065)
-0.041 -0.045 -0.053 -0.052
(0.023)* (0.023)* (0.029)* (0.029)*
- -0.036 - 0.009
(0.039) (0.076)
- 0.000 - 0.001
(0.001) (0.001)
- -0.004 - -0.001
(0.002)* (0.005)
- -0.128 - 0.001
(0.080) (0.143)
- 0.034 - -0.089
(0.039) (0.073)
- -0.034 - 0.031
(0.033) (0.058)
- -0.04 - -0.062
(0.035) (0.066)
3.996 4.072 4.787 4.833
(0.641)*** (0.645)*** (0.744)*** (0.750)“
0.08 0.09 0.38 0.38

 

Notes: n is 1042. (C) indicates community-level variables. Robust standard errors are in

parentheses. (*), (**) and (***) indicate significance respectively at 10%, 5% ; and 1 %.

107

Table A3.7.2.6. Impact of PMT on probability of children (24-59 months) being stunted:
binary program variable

 

 

 

OLS Fixed-effect
Year (2000=1) -0.008 -0.014 -0.037 -0.039
(0.053) (0.054) (0.071) (0.071)
Program since ’98 (Yes=1)*Year 0.013 0.019 0.045 0.049
(0.052) (0.053) (0.072) (0.072)
Gender (boy=1) 0.013 0.013 0.004 0.005
(0.016) (0.016) (0.016) (0.016)
Age -0.003 -0.003 -0.003 -0.003
(0.001)*** (0.001)*** (0.001)*** (0.001)***
Mother education (years) -0.005 -0.005 -0.005 -0.005
(0.002)*** (0.002)*** (0.002)” (0.002)“r
Father education (years) 0.001 0.001 0.001 0.001
(0.002) (0.002) (0.002) (0.002)
Mother height (cm) -0.018 -0.017 -0.017 -0.017
(0.002)*** (0.002)*** (0.002)*** (0.002)***
Father height (cm) -0.01 -0.01 -0.01 -0.011
(0.002)*** (0.002)*** (0.002)*** (0.002)***
Head of HH (male=1) 0.000 -0.002 0.012 0.011
(0.026) (0.026) (0.028) (0.028)
Number of children 5-14 yo in HH 0.019 0.017 0.014 0.014
(0.008)“ (0.008)“ (0.008)* (0.008)*
HH owns private toilet (Yes=1) -0.027 -0.025 -0.025 -0.025
(0.017) (0.017) (0.020) (0.020)
HH has sanitation (Yes=1) -0.091 -0.077 -0.08 -0.082
(0.022)*** (0.025)*** (0.027)*“ (0.029)*"
Per capita income (In) -0.053 -0.051 -0.057 -0.057
(0.012)*** (0.012)*** (0.013)*** (0.013)***
Type of community (urban=1), C - -0.041 - 0.022
(0.023)* (0.038)
Distance to district capital (km), C - 0 - 0
(0.000) (0.001)
Distance to bus station (km), C - -0.001 - 0
(0.001) (0.002)
Land w/ technical irrigation (%), C - -0.06 - -0.032
(0.045) (0.069)
Comm. has asphalt road (Yes=1), C - -0.016 - -0.068
(0.021) (0.035)*
Comm. has sewerage system (Yes=1), C - 0.028 - -0.008
(0.018) (0.029)
Comm. has piped water (Yes=1), C - 0.004 - 0.003
(0.019) (0.029)
Constant 5.401 5.305 5.325 5.394
(0.333)*** (0.336)*“ (0.366)*** (0.373)***
R-squared 0.11 0.12 0.22 0.22

 

Notes: 11 is 3221. (C) indicates community-level variables. Robust standard errors are in
parentheses. (*), (**) and (***) indicate significance respectively at 10%, 5% ; and 1%

108

Table A3.7.2.7. Impact of PMT on height-for-age of children 6—11 months: proportion of
child age exposed to the program

 

 

 

OLS Fixed-effect
Year (2000=1) 0.044 0.118 -0.036 -0.052
(0.172) (0.173) (0.343) (0.358)
Age exposed to the program*Year -0.048 -0.128 0.219 0.24
(0.187) (0.190) (0.477) (0.501)
Gender (boy=1) 0.107 0.118 -0.01 -0.009
(0.124) (0.124) (0.198) (0.208)
Age -0.129 -0.129 -0.172 -0.172
(0.036)” (0.035)*" (0.057)*** (0.058)***
Mother education (years) 0.001 0.001 -0.025 -0.024
(0.015) (0.015) (0.022) (0.024)
Father education (years) 0.007 0.009 0.013 0.011
(0.015) (0.015) (0.022) (0.022)
Mother height (cm) 0.061 0.059 0.059 0.057
(0.012)*** (0.012)*** (0.019)*** (0.019)***
Father height (cm) 0.047 0.048 0.07 0.069
(0.012)*** (0.012)*** (0.019)*** (0.019)***
Head of HH (male=1) -0.041 -0.072 0.278 0.276
(0.222) (0.221) (0.443) (0.461)
Number of children 5-14 yo in HH 0.024 0.022 0.021 0.03
(0.062) (0.062) (0.089) (0.091)
HH owns private toilet (Yes=1) 0.113 0.061 0.174 0.185
(0.136) (0.137) (0.224) (0.232)
HH has sanitation (Y es=1) 0.027 -0.043 0.238 0.303
(0.177) (0.198) (0.395) (0.453)
Per capita income (In) 0.075 0.100 0.037 0.034
(0.094) (0.094) (0.134) (0.138)
Type of community (urban=1), C - 0.362 - 0.084
(0.170)“Ir (0.462)
Distance to district capital (km), C - 0.000 - -0.002
(0.003) (0.006)
Distance to bus station (km), C - 0.029 - 0.007
(0.011)*** (0.022)
Land w/ technical irrigation (%), C - -0.019 - 0.229
(0.335) (0.849)
Comm. has asphalt road (Yes=1), C - -0.265 - -0.023
(0.164) (0.385)
Comm. has sewerage system (Y es=1), C - 0.07 - -0.267
(0.140) (0.341)
Comm. has piped water (Yes=1), C - -0.021 - -0.168
(0.152) (0.358)
Constant -17.656 -17.979 -20.413 -19.717
(2.559)*** (2.558)*** (3.929)*** (4.035)***
R-squared 0.14 0.16 0.57 0.58

 

Notes: n is 525. (C) indicates community-level variables. Robust standard errors are in
parentheses. (*), (**) and (***) indicate significance respectively at 10%, 5%; and 1 %.

109

Table A3.7.2.8. Impact of PMT on height-for-age of children 12-23 months: proportion of
child age exposed to the program

 

 

 

OLS Fixed-effect
Year (2000=1) -0.093 -0.089 -0.213 -0.227
(0.104) (0.105) (0.145) (0.152)
Age exposed to the program*Year 0.035 0.04 0.482 0.552
(0.125) (0.126) (0.229)“ (0.237)**
Gender (boy=1) -0.137 -0.139 -0.223 -0.219
(0.074)* (0.074)* (0.088)“ (0.089)**
Age -0.042 -0.043 -0.029 -0.029
(0.011)*** (0.011)*** (0.012)“ (0012)“
Mother education (years) 0.005 0.005 -0.009 -0.009
(0.009) (0.009) (0.009) (0009)
Father education (years) 0.003 0.002 0.000 0.000
(0.009) (0.009) (0.010) (0.010)
Mother height (cm) 0.023 0.022 0.027 0.027
(0.007)*** (0.007)*** (0.008)*** (0.008)***
Father height (cm) 0.031 0.031 0.035 0.035
(0.008)*** (0.008)*** (0.009)*** (0009)“
Head of HH (male=1) -0.011 -0.001 -0.158 -0.167
(0.123) (0.123) (0.135) (0.138)
Number of children 5-14 yo in HH 0.007 0.011 0.033 0.031
(0.037) (0.038) (0.044) (0.045)
HH owns private toilet (Yes=1) 0.107 0.105 0.091 0.09
(0.081) (0.081) (0.105) (0.105)
HH has sanitation (Yes=1) 0.168 0.146 0.098 0.099
(0.100)* (0.114) (0.159) (0.167)
Per capita income (In) 0.157 0.161 0.115 0.114
(0.061)** (0.061)*** (0.072) (0.072)
Type of community (urban=1), C - 0.125 - 0.093
(0.104) (0.214)
Distance to district capital (km), C - -0.001 - 0.001
(0.002) (0.003)
Distance to bus station (km), C - 0.011 - 0.018
(0.006)* (0.012)
Land w/ technical irrigation (%), C - 0.396 - -0.36
(0.211)* (0.426)
Comm. has asphalt road (Y es=1), C - -0.116 - 0.07
(0.103) (0.187)
Comm. has sewerage system (Yes=1), C - -0.079 - -0.075
(0.086) (0.163)
Comm. has piped water (Y es=1), C - 0.079 - -0.03
(0.093) (0.162)
Constant -11.671 -11.627 -12.423 -12.555
(1.698)*** (1.710)*** (1.957)*** (1.982)“*
R-squared 0.08 0.09 0.41 0.41

 

Notes: n is 1042. (C) indicates community-level variables. Robust standard errors are in
parentheses. (*), (**) and (***) indicate significance respectively at 10%, 5% ; and 1%.

110

Table A3.7.2.9. Impact of PMT on height-for-age of children 24-59 months: proportion of
child age exposed to the program

 

 

 

OLS Fixed-effect
Year (2000=1) -0.023 -0.009 0.009 0.002
(0.058) (0.058) (0.071) (0.073)
Age exposed to the program*Year -0.12 -0.143 -0.095 -0.072
(0.117) (0.118) (0.171) (0.174)
Gender (boy=1) -0.01 -0.008 -0.001 -0.003
(0.040) (0.040) (0.041) (0.041)
Age 0.005 0.005 0.004 0.004
(0.002)“ (0.002)“ (0.002)* (0.002)*
Mother education (years) 0.017 0.017 0.011 0.011
(0.005)*" (0.005)*“ (0.005)*“r (0005)“
Father education (years) -0.005 -0.006 -0.003 -0.003
(0.005) (0.005) (0.005) (0.005)
Mother height (cm) 0.054 0.053 0.052 0.053
(0.004)*** (0.004)*** (0.004)*** (0.004)***r
Father height (cm) 0.032 0.032 0.032 0.033
(0.004)*** (0.004)*** (0.004)*** (0.004)***
Head of HH (male=1) 0.018 0.023 -0.03 -0.022
(0.066) (0.067) (0.068) (0.069)
Number of children 5-14 yo in I-IH -0.043 -0.04 -0.019 -0.019
(0.019)“ (0.019)" (0.021) (0.021)
HH owns private toilet (Yes=1) 0.104 0.09 0.074 0.068
(0.044)“ (0.044)“r (0.048) (0.049)
HH has sanitation (Y es=1) 0.318 0.25 0.312 0.281
(0.055)*** (0.064)*** (0.071)*** (0.075)*“
Per capita income (In) 0.157 0.155 0.156 0.153
(0.031)*** (0.031)*** (0.035)*“ (0.035)*“r
Type of community (urban=1), C - 0.14 - -0.034
(0.058)" (0.093)
Distance to district capital (km), C - -0.001 - -0.002
(0.001) (0.001)
Distance to bus station (km), C - 0.003 - 0.003
(0.003) (0.005)
Land w/ technical irrigation (%), C - 0.01 - -0.161
(0.115) (0.182)
Comm. has asphalt road (Yes=1), C - -0.021 - 0.178
(0.053) (0.086)“
Comm. has sewerage system (Yes=1), C - -0.056 - 0.011
(0.047) (0.076)
Comm. has piped water (Yes=1), C - 0.007 - 0.005
(0.049) (0.078)
Constant -17.016 -16.775 -16.706 -16.862
(0.848)*** (0.854)*** (0.933)*** (0.952)“*
R-squared 0.15 0.15 0.27 0.27

 

Notes: 11 is 3221. (C) indicates community-level variables. Robust standard errors are in
parentheses. (*), (**) and (***) indicate significance respectively at 10%, 5%; and 1 %

lll

Table A3.7.2.10. Impact of PMT on probability of children (6—11 months) being stunted:
proportion of child age exposed to the program

 

 

 

OLS Fixed-effect
Year (2000=1) -0.052 -0.06 -0.064 -0.061
(0.047) (0.047) (0.100) (0.103)
Age exposed to the program*Year 0.006 0.012 -0.061 -0.08
(0.051) (0.052) (0.130) (0.134)
Gender (boy=1) -0.027 -0.027 0.026 0.03
(0.033) (0.034) (0.054) (0.056)
Age 0.024 0.024 0.019 0.021
(0.010)“ (0.010)“ (0.017) (0.017)
Mother education (years) 0.003 0.003 0.01 0.011
(0.004) (0.004) (0.007) (0.007)
Father education (years) 0.000 0.000 -0.001 0.000
(0.004) (0.004) (0.006) (0.006)
Mother height (cm) -0.011 -0.011 -0.011 ~0.01
(0.003)*** (0.003)*** (0.005)“ (0.005)*
Father height (cm) -0.013 -0.013 -0.021 -0.02
(0.003)*** (0.003)*** (0.005)*“ (0.006)*“
Head of HH (male=1) 0.022 0.021 0.004 -0.002
(0.060) (0.060) (0.115) (0.118)
Number of children 5-14 yo in HH 0.008 0.009 0.006 0.005
(0.017) (0.017) (0.026) (0.026)
HH owns private toilet (Yes=1) -0.041 -0.033 -0.077 -0.075
(0.037) (0.037) (0.059) (0.060)
HH has sanitation (Yes=1) 0.03 0.023 0.018 -0.01
(0.048) (0.054) (0.108) (0.129)
Per capita income (In) -0.03 -0.036 —0.032 -0.037
(0.025) (0.026) (0.039) (0.040)
Type of community (urban=1), C - -0.03 - -0.019
(0.046) (0.129)
Distance to district capital (km), C - 0 - 0
(0.001) (0.002)
Distance to bus station (km), C - -0.005 - -0.001
(0.003) (0.006)
Land w/ technical irrigation (%), C - -0.061 - -0.312
(0.091) (0.225)
Comm. has asphalt road (Yes=1), C - 0.035 - 0.02
(0.045) (0.115)
Comm. has sewerage system (Yes=1), C - -0.041 - 0.026
(0.038) (0.098)
Comm. has piped water (Yes=1), C - 0.037 - 0.012
(0.041) (0.090)
Constant 4.117 4.178 5.392 5.167
(0.691 )*** (0.697)*** (1 .140)*** (1 .192)***
R-squared 0.09 0.11 0.52 0.53

 

Notes: n is 525. (C) indicates community-level variables. Robust standard errors are in
parentheses. (*), (**) and (***) indicate significance respectively at 10%, 5% ,' and 1%.

112

Table A3.7.2.11. Impact of PMT on probability of children (12-23 months) being stunted:
proportion of child age exposed to the program

 

 

 

OLS Fixed-effect
Year (2000=1) -0.072 -0.079 0.003 0.006
(0.039)* (0.040)** (0.053) (0.056)
Age exposed to the program*Year 0.028 0.028 -0.154 -0.155
(0.047) (0.048) (0.082)* (0.084)*
Gender (boy=1) 0.041 0.041 0.074 0.077
(0.028) (0.028) (0.034)“ (0035)"
Age 0.008 0.008 0.001 0.000
(0.004)* (0.004)* (0.005) (0.005)
Mother education (years) 0.000 0.000 0.005 0.005
(0.003) (0.003) (0.004) (0.004)
Father education (years) 0.000 0.001 0.001 0.001
(0.003) (0.003) (0.004) (0.004)
Mother height (cm) 7 -0.010 -0.010 -0.013 -0.013
(0.003)*** (0.003)*** (0.003)*** (0.003)***
Father height (cm) -0.011 -0.011 -0.011 -0.011
(0.003)*** (0.003)*** (0.003)*** (0.003)***
Head of HH (male=1) -0.012 -0.015 0.071 0.071
(0.047) (0.047) (0.058) (0.059)
Number of children 5-14 yo in HH -0.003 -0.005 -0.015 -0.015
(0.014) (0.014) (0.017) (0.017)
HH owns private toilet (Yes=1) -0.054 -0.050 -0.041 -0.038
(0.031)* (0.031) (0.041) (0.042)
HH has sanitation (Yes=1) -0.110 -0.075 0.006 0.014
(0.038)*** (0.043)* (0.061) (0.064)
Per capita income (In) -0.042 -0.045 -0.055 -0.055
(0.023)* (0.023)* (0.029)* (0.029)*
Type of community (urban=1), C - -0.039 - 0.025
(0.039) (0.075)
Distance to district capital (km), C - 0.000 - 0.000
(0.001) (0.001)
Distance to bus station (km), C - -0.004 - -0.001
(0.002) (0.005)
Land w/ technical irrigation (%), C - -0.126 - 0.035
(0.080) (0.144)
Comm. has asphalt road (Y es=1), C - 0.035 - -0.104
(0.039) (0.074)
Comm. has sewerage system (Yes=1), C - -0.034 - 0.030
(0.033) (0.058)
Comm. has piped water (Yes=1), C - -0.040 - -0.048
(0.035) (0.066)
Constant 4.017 4.099 4.714 4.774
(0.642)*** (0.646)*** (0.742)*** (0.747)***
R-squared 0.08 0.09 0.38 0.38

 

Notes: 11 is 1042. (C) indicates community-level variables. Robust standard errors are in
parentheses. (*), (**) and (***) indicate significance respectively at 10%, 5%; and 1 %.

113

Table A3.7.2.12. Impact of PMT on probability of children (24-59 months) being stunted:
proportion of child age exposed to the program

 

 

 

OLS Fixed-effect
Year (2000=1) -0.009 -0.01 -0.014 -0.01
(0.023) (0.023) (0.027) (0.028)
Age exposed to the program*Year 0.049 0.053 0.075 0.062
(0.046) (0.046) (0.068) (0.069)
Gender (boy=1) 0.013 0.013 0.004 0.005
(0.016) (0.016) (0.016) (0.016)
Age -0.003 -0.003 -0.003 -0.003
(0.001)*** (0.001)*** (0.001)*** (0.001)***
Mother education (years) -0.005 -0.005 -0.005 -0.005
(0.002)*** (0.002)*** (0.002)" (0002)“
Father education (years) 0.001 0.001 0.001 0.001
(0.002) (0.002) (0.002) (0.002)
Mother height (cm) -0.018 -0.018 -0.017 -0.017
(0.002)*** (0.002)*** (0.002)*** (0.002)***
Father height (cm) -0.01 -0.01 -0.01 -0.011
(0.002)*** (0.002)*** (0.002)*** (0.002)***
Head of HH (male=1) -0.001 -0.003 0.012 0.011
(0.026) (0.026) (0.028) (0.028)
Number of children 5-14 yo in HH 0.019 0.017 0.014 0.014
(0.008)“ (0.008)*“r (0.008)* (0.008)*
HH owns private toilet (Yes=1) -0.027 -0.025 -0.025 -0.025
(0.017) (0.017) (0.020) (0.020)
HH has sanitation (Yes=1) -0.09 -0.076 -0.081 -0.081
(0.022)*** (0.025)*** (0.027)*** (0.029)*“
Per capita income (In) -0.053 -0.051 -0.057 -0.057
(0.012)*** (0.012)*** (0.013)*** (0.013)***
Type of community (urban=1), C - -0.043 - 0.018
(0.023)* (0.038)
Distance to district capital (km), C - 0.000 - 0.000
(0.000) (0.001)
Distance to bus station (km), C - -0.001 - -0.001
(0.001) (0.002)
Land w/ technical irrigation (%), C - -0.056 - -0.034
(0.045) (0.069)
Comm. has asphalt road (Yes=1), C - -0.015 - -0.064
(0.021) (0.035)*
Comm. has sewerage system (Yes=1), C - 0.029 - -0.007
(0.018) (0.029)
Comm. has piped water (Yes=1), C - 0.004 - 0.001
(0.019) (0.029)
Constant 5.405 5.309 5.324 5.39
(0.333)*** (0.336)*** (0.366)*** (0.373)***
R-squared 0.11 0.12 0.22 0.22

 

Notes: n is 3221. (C) indicates community-level variables. Robust standard errors are in
parentheses. (*), (**) and (***) indicate significance respectively at 10%, 5% ; and 1%.

114

BIBLIOGRAPHY

115

BIBLIOGRAPHY

Alderman, Harold, Behrman, Jere R., Lavy, Victor and Menon, Rekha. (2001)
Child Health and school enrollment: A longitudinal analysis. Journal of
Human Resources, 31 (1):185-205.

Alderman, Harold and Behrman, Jere R. (2006). Reducing the incidence of low
birth weight in low-income countries has substantial economic benefits. World
Bank Research Observer, 21 (1).25-48.

Alderman, Harold, Hoddinott, John and Kinsey, William. (2006). Long-term
consequences of early childhood malnutrition. Oxford Economic Papers,
58(3):450-474.

Alderman, Harold, Hoogeven, Hans, and Rossi, Mariacristina (2009). Preschool
Nutrition and Subsequent Schooling Attainment. Economic Development and
Cultural Change, 57: 239-360

Armecin, Graeme, Behrman, Jere R., Duazo, Paulita, Ghuman, Sharon, Gultiano,
Socorro, King, Elizabeth and Lee, Nannette. (2006). Early Childhood
Development Through an Integrated Program: Evidence from the
Philippines. World Bank Policy Research Working Paper No.

Baum, C.F., Schaffer, ME, and Stillman, S. (2003). Instrumental Variables and
GMM: Estimation and testing. The Stata Ioumal, 3(1):1-31.

Baum, C.F., Schaffer, ME, and Stillman, S. (2007). Enhanced routines for
instrumental variables / GMM estimation and testing. The Stata Journal,
7(4):465-506.

Behrman, Jere (1996). The impact of health and nutrition on education. World
Bank Research Observer, 11(1):23-38.

Behrman, J., Cheng, Y. and Todd, PE. (2004). Evaluating Preschool Programs
When Length of Exposure to the Program Varies: A Nonparametric
Approach. Review of Economics and Statistics, 86/ 1, 108-132.

Behrman, Jere R., and Lavy, Victor. (1998). Child Health and Schooling Achievement:
Association, Causali ly and Household Allocations. Philadelphia: University of
Pennsylvania. Mimeo.

Behrman, Jere R. and Rosenzweig, Mark R. (2004). Returns to birthweight. Review
of Economics and Statistics, 86(2):586-601.

116

Behrman, Jere R., Hoddinott, John, Maluccio, John, Soler-Hampejsek, Erica,
Behrman, Emily L., Martorell, Reynaldo, Ramirez-Zea, Manuel and Stein,
Aryeh. (2006). Watt determines adult cognitive skills? Impacts of pre-schooling,
schooling and post-schoolin g experiences in Guatemala. Department of Economics,
University of Pennsylvania, Mimeo.

Frankenber, Elizabeth and Karoly, Lynn. (1995). The 1993 Indonesia Family Life
Survey: Overview and field report. Santa Monica: RAND.

Duflo, E., (2001). Schooling and Labor Market Consequences of School
Construction in Indonesia: Evidence from an Unusual Policy Experiment, ,
American Economic Review, 91 / 4, 795-813

Duflo, E., Glennerster, R., Kremer, M., (2007). Using Randomization in
Development Economics Research: A Tool Kit. In Schultz, P., and Strauss, J.,
eds, Handbook of Development Economics, Volume 4, Amsterdam: North Holland
Press.

Frankenberg, Elizabeth, (1995). The Effects of Access to Health Care on Infant
Mortality in Indonesia. Health Transition Review, 5, 143-163.

Frankenberg, E., Thomas, D. and Beegle, K. (1999). The Real Cost of Indonesia’s
Economic Crisis: Preliminary Findings from the Indonesia Family Life
Surveys. Labor and Population Program Working Papers Series 99-04, RAND,
Santa-Monica.

Frankenberg, Elizabeth and Thomas, Duncan. (2000). The Indonesia Family Life
Survey (IFLS): Study design and Results from waves 1 and 2. Santa Monica:
RAND.

F rankenberg, Elizabeth and Thomas, Duncan (2001). Women’s health and
pregnancy outcomes: Do services make a difference. Demography, 38(2):253-
265.

Frankenberg, Elizabeth, Suriastini, Wayan and Thomas, Duncan (2005). Can
expanding access to basic healthcare improve children’s health status?

Lessons from Indonesia’s ’midwife in the village’ program. Population Studies,
59(1):5-19.

Gertler, P., (2004). Do Conditional Cash Transfers Improve Child Health?

Evidence from PROGRESSA’s Control Randomized Experiment. American
Economic Review, 94 / 2: 336-341.

117

Gertler, P., and J.W. Molyneaux, (1994). How Economic Development and
Family Planning Programs Combined to Reduce Indonesian Fertility.
Demography, 31 / 1, 33-63.

Ghuman, Sharon, Behrman, Jere, and Gultiano, Socorro, (2006). Children’s
Nutrition, School Quality and Primary School Enrollment in the Philippines.
I CSEAD Working Paper Series Vol. 2006-24.

Glewwe, Paul (2007). The impact of child health and nutrition on education in
developing countries: Theory, econometric issues and recent empirical
evidence. Food and Nutrition Bulletion, 26(2): S235-S250.

Glewwe, Paul and King, Elizabeth (2001). The impact of early childhood
nutritional status on cognitive development: Does the timing of malnutrition
matter? The World Bank Economic Review, 15(1):81-113.

Glewwe, Paul and Miguel, Edward (2007). The impact of child health and
nutrition on education in less developed countries. In T. Paul Schultz and
John Strauss, eds., Handbook of Development Economics, Volume 4, Amsterdam:
North Holland Press.

Glewwe, Paul, Jacoby, Hanan and King, Elizabeth. (2001) Early childhood
nutrition and academic achievement: a longitudinal analysis. Journal of Public
Economics, 81(3):345-368.

Glewwe, Paul, Jacoby, Hanan (1995). An economics analysis of delayed primary
school enrollment in low income country: The role of early childhood
nutrition. The Review of Economics and Statistics, 77(1):156-169.

Jayne, T., Strauss, J., Yamano, T. and Molla, D. (2002). Targeting of Food Aid in
Rural Ethiopia: Chronic Need or Inertia? Journal of Development Economics,
68:247-88.

King, Elizabeth and Behrman, Jere (2009). Timing and Duration of Exposure in
Evaluations of Social Programs. World Bank Research Observer, 24(1):55-82

Kirono, D.G.C. (2000). Indonesian seasonal rainfall variability, links to El Nino
Sou them Oscillation and Agricultural Impacts, Unpublished PhD Dissertation.
Monash University, Australia.

Kirono, D.G.C ., Tapper, N]. and McBride, IL. (1999). Documenting Indonesian
rainfall in the 1997-1998 El Nino event. Physical Geography, 20: 422-435.

118

Maluccio, J.A., J. Hoddinott, J.R. Behrman, R. Martorell, A. R. Quisumbing, and
AD. Stein (2006). The Impact of an Experimental Nutritional Intervention in
Childhood on Education among Guatemalan Adults. F CND Discussion Papers
No 207, IFPRI, Washington DC.

Maluccio, John, Hoddinott, John, Behrman, Jere R., Quisumbing, Agnes,
Martorell, Reynaldo and Stein, Aryeh D. (2006). The impact of nutrition
during early childhood on education among Guatemalan adults. IFPRI,
Washington, DC, Mimeo.

Martorell, R. (1999). The Nature of Child Malnutrition and Its Long—Term
Implications. Food and Nutrition Bulletin 20: 288-292.

Martorell, Reynaldo (1993). Enhancing Human Potential in Guatemalan Adults
Through Improved Nutrition in Early Childhood, Nutrition Today: 6-13.

Pitt, M., Rosenzweig, M. and Gibbons, D. (1993). The determinants and
consequences of the placement of government programs in Indonesia. The
World Bank Economic Review, 72319-348.

Pollitt, Ernesto, (1990). Malnutrition and Infection in the Classroom. Paris: UNESCO.

Pollitt, Ernesto, Gorman, Kathleen, Engel, Patrice, Martorell, Reynaldo and
Reynaldo Rivera, Reynaldo. (1993) Early supplementary feeding and cognition,
Monographs of the Society for Research in Child Development, Serial No.
235, Vol. 58, No. 7.

Ravallion, M., (2007). Evaluating Anti-Poverty Program. In T.Paul Schultz and
John Strauss, eds, Handbook of Development Economics, Volume 4, Amsterdam:
North Holland Press.

Rosenzweig, M. and K. Wolpin, (1986). Evaluating the Effects of Optimally
Distributed Public Programs. American Economic Review, 76:470-482.

Sandjaja, Sihadi, S. Mulyati, Arnelia, M. Saidin, Herudarini, Suhartato, Y.
Widodo, A. Atrnawikarta, Sudjasmin, Sudikno, I. Zuraedah, (2001). Studi
Dampak Program Makanan Tambahan terhadap Status Gizi dan Kesehatan Bayi dan
Anak. Unpublished Report, PPPG 8: Bappenas.

Strauss, John, Beegle, Kathleen, Dwiyanto, Agus, Herawati, Yulia, Sikoki,
Bondan, and Witoelar, Firman (2004). The Third Wave of the Indonesia Family
Life Survey: Overview and Field Report. RAND Corp., Santa Monica

119

Strauss, John, Beegle, Kathleen, Dwiyanto, Agus, Herawati, Yulia, Pattinasarany,
Daan, Satriawan, Elan, Sikoki, Bondan, Sukamdi and Witoelar, Firman (2004).
Indonesian Living Standards Before and After the Financial Crisis. RAND and
Institute for Southeast Asian Studies (ISEAS), Santa Monica CA and
Singapore.

Strauss, John, and Thomas, Duncan (1995). Human resources: Empirical
modeling of household and family decisions. In J.R. Behrman and TN.
Srinivasan, eds, Handbook of Development Economics, Volume 3A, Amsterdam:
North Holland Press.

Strauss, John, and Thomas, Duncan (1998). Health, nutrition and economic
development. Journal of Economic Literature, 36(2):766-817.

Strauss, John, and Thomas, Duncan. (2007). Health over the life course. In T.Paul
Schultz and John Strauss, eds, Handbook of Development Economics, Volume 4,
Amsterdam: North Holland Press.

Wooldridge, Jeffrey (2002). Econometric Analysis of Cross Section and Panel Data.
MIT Press

Wooldridge, Jeffrey (2003). Introductory Econometrics: A Modern Approach.
Thomson-South Western.

Yamauchi, Chikako (2005). Evaluating Poverty Alleviation through Microcredit:

Methodological and Empirical Evidence from Indonesia. Department of
Economics, UCLA, unpublished paper.

120