University of Southern California Dissertations and Theses

Selection and impacts of early life events on later life outcomes

(USC Thesis Other)

Selection and impacts of early life events on later life outcomes

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content SELECTIONANDIMPACTSOFEARLYLIFEEVENTS
ONLATERLIFEOUTCOMES
by
Md Nazmul Ahsan
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulllment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(ECONOMICS)
August 2016
Copyright 2016 Md Nazmul Ahsan
Acknowledgments
This dissertation marks a near end to my student life, albeit only formally. During
this long journey, I have accrued debts to many people who have inspired me toward
education. First and foremost, this dissertation would not have possibly come into
existence without the guidance and support of my adviser John Strauss. John knows
how to bring the best out of a novice young researcher like me. On the one hand,
he did not let me compromise on the quality of my works; on the other hand, he
guided through the dicult process of preparing research papers. His training has
instilled both condence and humility in me when it comes to writing and presenting
my research works.
I am greatly indebted to Anant Nyshadham and Jerey Nugent. Anant Nyshad-
ham has inspired me to think about challenging research questions and encouraged
me to see the bigger picture of answering a question. Jerey Nugent has graciously
provided me with valuable comments and suggestions on various research topics and
encouraged my participation in various workshops organized at USC and elsewhere.
I would like to express my sincere gratitude to Neeraj Sood and Geert Ridder for
serving on my qualifying exam committee and/or dissertation defense committee.
I would like to thank Dr. Nurul Alam and the Matlab HDSS of ICDDR,B for
providing relevant information regarding Matlab and ICDDR,B.
I am profoundly thankful to Rakesh Banerjee and Riddhi Bhowmick for their
friendships and supports during my time at USC. Through numerous discussions
with them Rakesh has helped me sharpen my argument skills, Riddhi has helped me
to get a sense of the greater purpose of life. I am also grateful to all of my rst year
i
cohorts at USC, in particular Robson Morgan and Karrar Hussain for teaming up
with me to pass the rst year core exam. I am very thankful to Arya Gaduh; Arya
has been a great support when it comes to knowing about his native land Indonesia
or Stata coding.
Young Miller and Morgan Ponder have made sure that bureaucratic concerns do
not impede my work and progress; I am also thankful to Fatima Perez and Christopher
Frias for their assistance in various administrative issues. I would like to thank the
Department of Economics and the Dornsife College for ensuring funding throughout
my time in the program.
My journey as a student would have been much less fullling, if I had not met
two of my teachers Tofazzal Hossain and S M Ashiquzzaman. Tofazzal Hossain has
introduced me to the pleasure of acquiring knowledge. S M Ashiquzzaman's principles
of economics class lectures made me fall in love with economics. He has also inspired
me to apply tools of economics in understanding problems beyond the end of textbook
chapters. His encouragement and guidance were critical in my pursuit of higher
education.
I am grateful to parents for their unconditional love and support. I am also thank-
ful to my wife Sharifa Tanjim for sharing responsibilities with me, for her undivided
attention, and care.
ii
Table of Contents
Acknowledgments i
List of Tables vi
List of Figures ix
Abstract x
1 Introduction 1
2 Do Parents Selectively Time Birth Relative to Ramadan?
Evidence from Matlab, Bangladesh 7
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Fetal Origins Hypothesis, Maternal Fasting, and Child Human
Capital Formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 Matlab Family Planning and Child Health . . . . . . . . . . . . . . . 19
2.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.5 Ramadan Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.6 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.7 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.8 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.10 Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.11 Figure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
iii
3 Parental Health, Households, Communities and Fetal Health in
India
1
67
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.2 Background Literature . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.2.1 Likelihood of Male Birth as an Indicator of Fetal Health Quality 75
3.2.2 Health Production Functions . . . . . . . . . . . . . . . . . . . 77
3.3 Conceptual Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.3.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.3.2 Potential Empirical Issues . . . . . . . . . . . . . . . . . . . . . 83
3.3.3 Estimating Equation . . . . . . . . . . . . . . . . . . . . . . . . 84
3.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.4.1 Dependent Variables . . . . . . . . . . . . . . . . . . . . . . . . 85
3.4.2 Household and Community Variables . . . . . . . . . . . . . . . 86
3.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.5.1 Parental Education . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.5.2 Parental Height . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.5.3 Income . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.5.4 Access to Information . . . . . . . . . . . . . . . . . . . . . . . 92
3.5.5 Community Characteristics . . . . . . . . . . . . . . . . . . . . 93
3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
3.7 Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
3.8 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
1
Jointly written with Riddhi Bhowmick.
iv
4 The Unintended Consequences of the Village Midwife Program in
Indonesia
2
111
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
4.2 Village Midwife Program in Indonesia . . . . . . . . . . . . . . . . . . 116
4.3 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
4.4 Data and Measurements . . . . . . . . . . . . . . . . . . . . . . . . . 120
4.5 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
4.5.1 Empirical Challenges . . . . . . . . . . . . . . . . . . . . . . . . 122
4.5.2 Empirical Framework . . . . . . . . . . . . . . . . . . . . . . . 125
4.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
4.6.1 Male Births . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
4.6.2 Birth Weight and Infant Mortality . . . . . . . . . . . . . . . . 128
4.6.3 Threats to Identication . . . . . . . . . . . . . . . . . . . . . . 130
4.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.8 Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
4.9 Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
4.10 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
5 Conclusion 152
Comprehensive bibliography 155
2
Jointly written with Riddhi Bhowmick.
v
List of Tables
2.1 MHSS Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . 52
2.2 IFLS Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.3 Selection in live birth . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.4 Selection in live birth : Muslim Married Women . . . . . . . . . . . . 54
2.5 Family Planning Program and Birth Timing by Ramadan Month
(Muslims): Birth Cohort 1974-95 . . . . . . . . . . . . . . . . . . . . 55
2.6 Family Planning Program and Birth Timing Relative to Ramadan:
Birth Cohort 1974-95 . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.7 Family Planning Program and Birth Timing Relative to Ramadan:
Birth Cohort 1974-95 . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.8 Exposure to Ramadan and Child Height . . . . . . . . . . . . . . . . 57
2.9 Parental SES and Ramadan Exposure for cohort 1974-July,1978 . . . 57
2.10 Exposure to Ramadan and Mother Education: Birth Cohort 1982-1995 58
2.11 Birth Relative to Ramadan and Mother Education . . . . . . . . . . . 59
2.12 Exposure to Ramadan and Mother Education: Birth Cohort 1974-1982 60
2.13 Mother Education and Child Height . . . . . . . . . . . . . . . . . . . 60
2.14 Exposure to Ramadan and Mother Education . . . . . . . . . . . . . 61
2.15 Exposure to Ramadan and Mother Education . . . . . . . . . . . . . 62
A1 Family Planning Program and Birth Timing Relative to Ramadan:
Birth Cohort 1974-95 . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
A2 Ramadan Exposure and Mother Education . . . . . . . . . . . . . . . 65
vi
A3 Birth Relative to Ramadan and Male Fragility . . . . . . . . . . . . . 66
3.1 Fetal Survival Figures for India . . . . . . . . . . . . . . . . . . . . . 97
3.2 Birth Weight and Birth Size Figures for India . . . . . . . . . . . . . 98
3.3 Summary Statistics- Household Variables . . . . . . . . . . . . . . . . 99
3.4 Summary Statistics- Community Characteristics . . . . . . . . . . . . 100
3.5 Fetal Survival Regressions- Rural . . . . . . . . . . . . . . . . . . . . 101
3.6 Fetal Survival Regressions- Urban . . . . . . . . . . . . . . . . . . . . 102
3.7 Birth Weight and Birth Size Regressions- Rural . . . . . . . . . . . . 103
3.8 Birth Weight and Birth Size Regressions- Urban . . . . . . . . . . . . 104
3.9 Eect of Access to Information on Fetal Survival and Birth Size . . . 105
3.10 Community Determinants of Fetal Survival and Birth Size- General
Village Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . 106
3.11 Community Determinants of Fetal Survival and Birth Size- Health
Facilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.12 Community Determinants of Fetal Survival and Birth Size-
Government Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
3.13 Community Determinants of Birth Weight . . . . . . . . . . . . . . . 109
3.14 Likelihood of Male Child and Maternal Height Relationship in Other
Countries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
4.1 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
4.2 Impacts of The Village Midwife Program on Likelihood of Male Birth 136
4.3 Impacts of The Village Midwife Program on Likelihood of a Male
Child at First Birth . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.4 Impacts of The Village Midwife Program on Likelihood of Male
Birth, by Mother Education . . . . . . . . . . . . . . . . . . . . . . . 138
vii
4.5 Impacts of The Village Midwife Program on Birth Weights (in
Grams), by Gender . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
4.6 Testing for Selective Fertility . . . . . . . . . . . . . . . . . . . . . . . 140
4.7 Testing for Selection of Mothers: Mother Education . . . . . . . . . . 141
4.8 Testing for Selection of Mothers: Mother Age at Birth . . . . . . . . 142
4.9 Testing for Gender-specic Reporting Bias due to the Village Midwife
Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
A1 Impacts of The Village Midwife Program on Likelihood of Male Birth
{ Continuous Years of Exposure . . . . . . . . . . . . . . . . . . . . . 144
A2 Impacts of The Village Midwife Program on Likelihood of Male Child
at First Birth { Continuous Years of Exposure . . . . . . . . . . . . . 145
A3 Impacts of The Village Midwife Program on Likelihood of Male
Birth, by Mother Education { Continuous Years of Exposure . . . . . 146
A4 Impacts of The Village Midwife Program on Birth Weights (in
Grams), by Gender { Continuous Years of Exposure . . . . . . . . . . 147
A5 Impacts of The Village Midwife Program on Infant Mortality . . . . . 148
A6 Impacts of The Village Midwife Program on Infant Mortality {
Continuous Years of Exposure . . . . . . . . . . . . . . . . . . . . . . 149
A7 Impacts of The Village Midwife Program on Low Birth Weight . . . . 150
A8 Impacts of The Village Midwife Program on Low Birth Weight {
Continuous Years of Exposure . . . . . . . . . . . . . . . . . . . . . . 151
viii
List of Figures
2.1 Matlab Area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.1 Allocation of Midwives . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
ix
Abstract
The three essays of this dissertation explore fertility selection, selection in pregnancy
outcomes, and implications of maternal health on child health outcomes in developing
countries. A growing number studies document the impact of early life environmental
and nutritional conditions on later life outcomes. Much of the analysis has been
based on in utero period, as critical developments of human life take place in that
period. However, not much is known about selective exposure to those in utero
shocks and selection in pregnancy outcomes due to those shocks. These are important
concerns because the presence of selection may alter interpretations of such impacts
on outcomes as well as may have policy implications for health interventions.
The rst essay (Chapter 2), using data from Bangladesh and Indonesia, shows that
more educated mothers are less likely to overlap their pregnancies with Ramadan. The
prior studies were unable to document any parental selection in in utero exposure to
Ramadan; those studies concluded that maternal fasting during Ramadan|{at the
time of pregnancy|{has negative implications for individual's health and economic
outcomes. In contrast, this essay nds that parental characteristics may aect the
likelihood of in utero Ramadan exposure; it emphasizes that examination of parental
selection in in utero exposure to Ramadan should allow for time varying changes.
This essay shows that ignoring parental selection in in utero exposure to Ramadan
may lead to bias in estimates of in utero exposure to Ramadan on relevant outcomes.
Moreover, this essay documents that an intensive family planning program may help
parents to avoid pregnancies overlapping with Ramadan.
The second essay analyzes whether parental human capital works as complemen-
x
tary or substitute to provision of health facilities in determining fetal health outcomes
in areas with adverse disease environment. This essay uses the second and third waves
of the National Family Health Survey and restricts the sample to rst born within
last ve years from the survey to minimize recall bias and to avoid the possibility of
sex-selective abortions. Along with several other common measures of fetal health
outcomes,this essay also uses the likelihood of a male birth and birth size based on
insights from the epidemiological literature. It nds evidence suggesting parental hu-
man capital may substitute low provision of health facilities in areas with adverse
disease environment. One of the main ndings of this essay is that maternal height
is positively associated with the likelihood of a male birth. This nding implies that
gender dierence in human capital outcomes may persist due to biological reasons
along with son preference in India.
The third essay (Chapter 4) explore whether a maternal health improving inter-
vention aects selection into live birth births and child characteristics or endowments
at birth. For this purpose, this essay analyzes the impact of the Village Midwife
Program (VMP) in Indonesia on the likelihood of a male birth and birth weights
by gender. Using all four waves of the Indonesian Family Life Survey (IFLS) and
a dierence-in-dierence strategy, it nds that the provision of a midwife in a com-
munity increases the probability of a male birth by 4 percentage points, mostly for
mothers with at most primary education. It also nds that provision of midwives
leads to a decrease in birth weight for male children, while no change is observed for
female children. These two results together imply that positive nutrition shock may
lead to an increase in survival of poor quality fetuses. Thus a policy that improves
maternal health can have negative consequences for future human capital formation
and can also make gender-reversal possible.
xi
xii
Chapter 1
Introduction
Economists and social scientists alike are interested in individual's human and health
capital formation because of its implication for individual's economic outcomes and
well-being. Based on insights from bio-medical and epidemiological literature, re-
searchers have endeavored to examine the casual eects of experiencing early life nu-
trition shocks and adverse environment on various later life outcomes (Almond and
Currie, 2011). A growing subset of studies have focused on in utero nutrition and
environment because nutrition and environmental shocks may aect fetus to make
adaptations which are benecial in shorter term survival but have negative health ef-
fects in longer terms (Barker, 2000). Moreover, in utero conditions and environment
may aect fetal survival to live birth and beyond, and birth endowments (Larsen et
al., 2013; Nepomnaschy et al., 2006).
Unlike studies based on animals in which impact of shocks can be purged from
confounding factors by construction of a laboratory experiment set up, studies based
on human beings can be very challenging because individual level characteristics
may correlate with shocks as well as outcomes of interest. Presence or absence of
community level programs such as health facilities and family planning programs
may also interact with parental characteristics in determining exposure to shocks and
pregnancy outcomes. On the one hand, an omission of an interaction of a program
with parental characteristics may lead to falsely concluding that there is no relation
between parental characteristics and exposure to shocks, on the other hand, programs
1
may aect parents from dierent backgrounds dierently, and analyzing them might
provide us important insights about pregnancy outcomes and child birth endowments.
The rst essay analyzes whether provision of a family planning program help par-
ents avert pregnancies overlapping with Ramadan. A growing number of studies nd
that in utero experience of maternal fasting during the time of Ramadan has negative
eects on individual's health, education, and labor market outcomes (Almond and
Mazumder, 2011; Ewijk, 2011; Almond et al., 2014; Majid, 2015; Karimi, 2014; Ewijk
et al., 2013). These studies calculate mean dierence in outcomes between individuals
who are exposed to Ramadan and not exposed to Ramadan in utero for both Mus-
lims and non-Muslims. Under the assumption that parents do not selectively expose
their pregnancies to Ramadan, mean dierences between those individuals exposed
and those not exposed to Ramadan can be attributed to maternal fasting during Ra-
madan. It is important to note that Ramadan is a recurring and a predictable event.
Therefore, intervention such as provision of a family planning program may allow
parents to time birth relative to Ramadan. Using the Matlab Health Socio-economic
survey, I nd that Muslim mothers avoid conceptions during the month and month
before Ramadan. They are also shifting a few months ahead. When similar exercise
is performed with Hindu births, this essay does not nd any eect. Moreover, this
essay compares the height of the Muslim children between those who are exposed and
not exposed to Ramadan both in the family planning program and the non-family
planning program areas, but nds signicant negative eect of exposure to Ramadan
in utero on child height only in the family planning program areas. This essay also
nds that in the family planning program areas more educated Muslim mothers are
less likely to overlap their pregnancies with Ramadan after being exposed to the fam-
ily planning program. This demonstrates that if the placement information of the
family planning program is ignored, it leads to falsely concluding that mother years of
2
education has no association with Ramadan exposure, but when it is taken into con-
sideration, the results show substantial selection in Ramadan exposure. To explore
that negative association between mother years of education and Ramadan exposure
is not limited to the family planning program areas of Matlab, this essay also uses all
four waves of the Indonesian Family Life Survey and nds similar parental selection
in Ramadan exposure. Overall, this study documents two key points. Firstly, it em-
phasizes that parental selection in in utero Ramadan exposure should be examined
allowing for time varying changes; ignoring time varying changes may lead to falsely
concluding that there is no parental selection, and it may also overestimate the in
utero exposure eect of Ramadan. Secondly, it also documents that provision of an
intensive family planning program may help mothers to avoid pregnancies, when it is
the most critical for the fetal health.
The second essay explores whether parental health and human capital may substi-
tute or complement provision of health facilities in areas with adverse disease environ-
ment. For this purpose, this essay uses the second and third rounds of the National
Family Health Survey in India. Endowment at birth or fetal health is an impor-
tant determinant of subsequent health or later-life economic outcomes (Behrman and
Rosenzweig, 2004; Almond et al., 2005; Bhalotra and Rawlings, 2011; Gluckman and
Hanson, 2004). Along with the conventional measures of fetal health and survival
status, we use two new measures|likelihood of a male birth, which re
ects male
fragility in utero, and is a measure of fetal survival, and birth size as a measure of
fetal health. Recent studies in epidemiological studies suggest that male fetuses are
more vulnerable to early-life shocks than female fetuses (Song, 2012; Fukuda et al.,
1998; Sanders and Stoecker, 2015; Shifotoka and Fogarty, 2012; Williams and Gloster,
1992). Moreover, maternal health status matter more for the in utero survival of male
3
children than that of female children Eriksson et al. (2010). Consistent with this ar-
gument and substitution hypothesis, this essay nds that in the rural areas of the
high infant mortality states of India, taller mothers are more likely to give birth to
male children controlling for various individual and state level measures. Moreover,
maternal education and stature is positively associated other measures such as birth
weight, birth size, and lower neo-natal mortality in those areas.
Based on insights from second essay, the third essay examines the impact of an ex-
ogenous improvement in maternal health on the likelihood of a male birth. Although
maternal health has implications for likelihood of male births, the causal evidence
on that topic is scarce. The Indonesian midwife program was designed and imple-
mented to improve the health status of the reproductive age women. Frankenberg
and Thomas (2001) found that the program was eective in improving the health
status of reproductive age women. Using all four waves of the IFLS and a dierence-
in-dierences strategy, this essay nds that the provision of a midwife inadvertently
also increases the probability of a male compared to a female birth. Moreover, it nds
that mothers, who have primary level of education or below, are more likely to give
birth to male children than mothers with higher than primary level of education, after
being exposed to the midwife program. It also examines the impact of the provision
of a midwife program on the child birth weight by gender and nds that provision of
a midwife program leads to lower birth weight for the male children; but it has no
eect on the birth weight of the female children. The results are robust to inclusion
of mother level observables, birth month interacted birth year xed eects, various
others community level time varying observables, and also when the birth sample
is restricted to only rst births. The increase in number of male births and lower
birth weights of the male children due to provision of a midwife program implies that
4
fetuses with poorer health qualities are more likely to be born due to improvement
in maternal health.
Moreover, the last two essays based on India and Indonesia point out that the gen-
der specic selection in live birth can occur purely due to biological reasons. These
two studies on maternal health have implications on methodologies used in gender
studies. It is recommended that mean dierence in gender outcomes should be exam-
ined allowing for selection into live birth. Moreover, parental investment responses
to early-life shock and the impacts of early-life shock on later-life outcomes should be
examined allowing for selection in live birth due to early-life shocks.
This dissertation not only contributes to early life health literature by adding more
evidence but also documents important implications for methodologies used in early
life health literature as well as for health intervention policies in developing countries.
Firstly, it shows that if events that may cause nutritional distress are recurring and
predictable in nature, parental fertility selection relative to these events should be
examined allowing for time varying changes at the community level. Ignoring com-
munity level changes may lead to false conclusions about parental fertility selection
and may produce biased estimates. Secondly, it also applies insights from bio-medical
literature in understanding how a well intentioned maternal health improving inter-
ventions may aect selection into live births as well as can be negatively associated
with birth endowments of the child. Thirdly, it has also methodological implications
for gender related studies. Under the assumption that sex at birth is orthogonal to
parental characteristics, gender dierence in outcomes may be attributed to either
gender of the child or to parental gender specic investment. However, the ndings
from these studies document that gender of the child may not be random at birth
even in the absence of sex-selective abortions purely due to biological reasons. An-
other aspect of this nding is that positive shock in utero may not result in a positive
5
endowment. As a result, studies which examine dynamic complementary and substi-
tution without taking into account selection into live births may produce incorrect
interpretations. Lastly, this dissertation also documents the importance of health and
family planning program interventions in developing countries. Although consider-
able gain can be obtained from providing access to formal health care in developing
countries (Adhvaryu and Nyshadham, 2015), utilization of formal health care may be
limited due to high transportation costs and long travel time (Wong et al., 1987; Dor
et al., 1987; Mwabu et al., 1993; Dow, 1999; Adhvaryu and Nyshadham, 2012, 2014).
One way to increase contraceptive usage and the use of health services would be bring
these services to the doorstep of the household members through community health
workers. Even though in some developing countries have taken such measures, causal
analysis of such programs on various health and other measures is very limited.
6
Chapter 2
Do Parents Selectively Time Birth Relative to
Ramadan? Evidence from Matlab, Bangladesh
2.1 Introduction
More than 1 billion Muslims alive today were exposed to Ramadan in utero (hereafter
in utero exposure to Ramadan would be referred to as \exposure to Ramadan").
1
Ramadan is the 9th month in the Islamic calender. During this month, Muslims are
required to fast from dawn to dusk in each day of the month except some who can seek
exemptions. Although according to Islamic law, pregnant women are exempted from
fasting, in practice however, it has been documented that a majority of the Muslim
pregnant women fast during Ramadan. According to the fetal origins hypothesis,
this may be a serious concern. It is well established that in utero experience of
nutrition shocks and adverse conditions has serious implications for an individual's
health and human capital formation (Almond and Currie, 2011; Currie, 2009).
2
This
has led researchers to explore the consequence of maternal fasting during Ramadan
on her child's various health outcomes as well as economic outcomes. The ndings
so far overwhelmingly corroborate with the earlier evidence based on fetal origins
hypothesis (Almond and Mazumder, 2011; Ewijk, 2011; Almond et al., 2014; Majid,
1
According PEW research center number of Muslim population in 2010 was 1.6 billion. The
period of human gestation if about 9 months long. This implies children born in about 75% Islamic
calender year days will be exposed to Ramadan at some point of pregnancy.
2
Almond and Currie (2011) provides a good review on existing literature on fetal origins hy-
pothesis.
7
2015; Karimi, 2014; Ewijk et al., 2013). These studies nd that exposure to Ramadan
leads to poorer health, education outcomes, and labor market outcomes.
Although in utero experience of maternal fasting during Ramadan has serious
consequences for an individual's human capital formation and well being, not much is
known in regards to what policy might be eective in reducing number of pregnancies
overlapping with Ramadan. Moreover, relatively little is known regarding fertility se-
lection relative to Ramadan. Previous studies have examined selection in Ramadan
exposure by comparing dierent parental socio-economic measures between children
who are exposed to Ramadan and children who are not exposed to Ramadan (Al-
mond and Mazumder, 2011; Ewijk, 2011; Almond et al., 2014; Majid, 2015; Karimi,
2014; Ewijk et al., 2013). Overall, these studies do not nd any dierences in ob-
servable parental characteristics between children who are exposed and not exposed
to Ramadan. Here, it is important to note that Ramadan is a recurring event, and
each year Ramadan starts 11 days ahead from the day it started in the last year. The
use of suciently large number of cohorts allows the disentanglement of the eect of
Ramadan exposure from seasonality.
3
. One potential problem arising from the use
of large numbers of cohorts is that within a geographic unit over time there can be
time varying changes specic to that geographic unit that will not be captured by
the geographic unit xed eects.
4
Some of these changes such as availability of
contraceptive devices may allow parents to time birth relative to Ramadan. More
importantly, parents from dierent SES backgrounds may make use of these changes
dierently in relation to birth relative to Ramadan. As a result, there could be
parental selection in Ramadan exposure because of these changes, and ignoring such
changes may result in falsely concluding that there is no selection concern.
3
It takes Ramadan about 33 years to complete full circuit of western calendar.
4
These changes can be change in prices and community infrastructures.
8
Using the Matlab Health and Socio-Economic Survey (1996), I study whether
provision of an intensive family planning program can help parents to time birth
relative to Ramadan. In Matlab, a region in Bangladesh, the International Center
for Diarrhoeal Disease Research, Bangladesh (ICDDR, B) initiated a family planning
program in some villages (treatment areas) but not in others (control areas). Under
the family planning program, women of reproductive ages, living in the treatment
areas, were given free contraceptives door to door in every two weeks (Joshi and
Schultz, 2013). Another appealing feature of the MHSS is that reliable birth data
can be obtained from the pregnancy histories of women. A demographic Surveillance
System (DSS) has been operating in Matlab from 1966, and it collects information on
vital events like birth, marriage, death, and migration. Reliable birth data is available
from 1974 onward for both the treatment and control villages, which gives us 22 birth
year cohorts to study birth timing relative to Ramadan.
5
Using a dierence-in-dierences strategy, this chapter nds that Muslim mothers
are about ve to six percentage points less likely to give birth eight to nine months
after Ramadan after being exposed to the family planning program. They are also
about eight percentage points more likely to give birth four to seven months after
Ramadan. These two results imply that Muslim mothers increase conceptions ve
to two months ahead of Ramadan but avoid conceptions when Ramadan is either
1 month away or at the time of Ramadan. Using the same strategy on births by
Hindu mothers, this chapter does not nd any change in birth in months relative to
Ramadan on Hindu births. This strengthens the ndings that the intensive family
planning program in Matlab helped mothers selectively time birth relative to Ra-
madan. In addition to these results, using information on the timing of the family
planning program, this chapter nds that in the program areas more educated Muslim
5
Although DSS started from 1966, the data was computerized starting from 1974.
9
mothers are less likely to overlap their pregnancies with Ramadan after the initiation
of the family planning program. Such pattern for more educated Muslim mothers
in the control areas and more educated Hindu mothers in the treatment areas is not
found. This means that children who are exposed and not exposed to Ramadan in the
treatment areas dier in one important observed characteristic|mother education.
To document the consequence of selection, heights of the children|who are under
age ten years during the survey year 1996|are analyzed.
6
Interestingly, a signicant
dierence in child height between Muslim children who are exposed and not exposed
to Ramadan in the control areas is not found. In contrast, in the treatment areas
a statistically signicant dierence in child height between Muslim children who are
exposed and not exposed to Ramadan is found. In the treatment areas, the Muslim
children who are exposed to Ramadan are shorter than the children who are not ex-
posed. This chapter concludes that the dierence in child height between children
exposed and not exposed to Ramadan in the treatment areas could be due to parental
selection in Ramadan exposure.
Moreover, using all four waves of the Indonesian Family Life Survey (IFLS), this
study also documents that more educated Muslim mothers are less likely to overlap
their pregnancies with Ramadan. On other hand, for non-Muslim mothers the like-
lihood of exposure to Ramadan does not vary by mother years of education. This
reinforces the ndings in Matlab, Bangladesh and implies that the selection behavior
is not only limited for a sub-sample of the MHSS, which received the intensive family
planning program. One key distinction between the IFLS data and the MHSS data is
that with the IFLS this chapter does not evaluate a program eect on the selection in
Ramadan exposure but with the MHSS, it evaluates the program eect of the family
planning program on the selection in Ramadan exposure.
6
I only study children under age ten years to reduce the concern of adolescent growth spurt.
10
This chapter makes some key contributions to the literature. First, it shows
parents time birth relative to Ramadan using two rich data sets from two countries.
To the best of my knowledge, this is the rst study that shows timing of birth relative
to Ramadan. Earlier literature (Almond and Mazumder, 2011; Ewijk, 2011; Almond
et al., 2014; Majid, 2015; Karimi, 2014; Ewijk et al., 2013) do not nd any evidence
that parents time birth relative to Ramadan.
Secondly, this chapter shows that an intensive family planning program may help
Muslim mothers to avoid their pregnancies overlapping with Ramadan. In general,
family planning programs are promoted to reduce fertility rate. This study shows an
unintended benet of provision of an intensive family planning program to Muslims
is that it allows to them to avoid birth relative to Ramadan to some extent.
Thirdly, this chapter is also the rst to document the importance of incorporating
time varying changes in examining selection in fertility relative to Ramadan. It
illustrates that ignoring the time varying changes may result in falsely concluding that
there is no selection in Ramadan exposure, whereas incorporating such changes lead to
conclusion that there is. The underlying identication assumption in earlier studies
is that there is no parental selection in Ramadan exposure. Absence of parental
selection in Ramadan exposure implies that higher SES mothers are no less likely to
expose their pregnancies to Ramadan than lower SES mothers. This guarantees that
Ramadan exposure and non-exposure can be considered as a natural experiment, and
the dierences in outcomes between children exposed and not exposed to Ramadan
can be attributed to maternal fasting during Ramadan.
7
. On the other hand, the
failure of the assumption would imply that factors other than Ramadan exposure are
7
Sleep pattern, working hours, food consumption, etc., also change during the Ramadan. Eco-
nomics literature recognize such changes during Ramadan but generally attribute the dierence in
outcomes to exposure to maternal fasting because a majority of Muslim pregnant women fast during
Ramadan, and prolonged fasting may have implications for both maternal and fetal health based on
insights from bio-medical literature.
11
responsible for the observed dierence in outcomes.
Lastly, this study documents how the presence of selection may aect identica-
tion of Ramadan exposure on child health outcomes. It documents that the negative
association between Ramadan exposure and child height|a measure of long term
nutritional status (Strauss and Thomas, 2007)|is more pronounced among the Mus-
lim children residing in the villages which got the intensive family planning program.
Coupled with this observation, this chapter nds that more educated Muslim mothers
residing in the same villages were also less likely to overlap their pregnancies with
Ramadan. This result has implications on the ndings as well as methodologies ap-
plied in the earlier studies. This suggests that mean dierence in SES characteristics
between children exposed and not exposed to Ramadan may not be sucient enough
to examine selection in exposure to a recurring and predictable event like Ramadan.
The earlier studies should also examine parental selection allowing for time varying
changes. It is important to note that this study claims neither that the ndings based
on earlier studies are not valid nor that maternal fasting has no negative implications
for an individual's human capital formation. However, what is emphasized here is
that parental selection in Ramadan exposure can occur due to changes in community
characteristics over time, and ignoring such changes may produce biased estimates.
Unfortunately, some of the studies do not have have information on time varying
changes, but for some studies there are information on time varying changes both at
household as well as community level.
The rest of the chapter is organized as follows. Section 2 discusses the existing
literature on maternal fasting and its implications for child human capital formation.
Section 3 discusses the background of Matlab Family Planning and Child Health
programs. Section 4 discusses the data obtained from the MHSS and IFLS. Section
5 discusses the data on the Islamic calender year and the construction of measures
12
of birth months relative to Ramadan. Section 6 discusses the empirical strategies
used in this chapter to study selective timing of birth relative to Ramadan and also
whether presence of selection may bias exposure eect on health outcomes. Section
7 presents the empirical results and explains the extent to which they dier from
existing literature. Section 8 provides discussion on the results and implications for
existing studies and section 9 provides concluding remarks.
2.2 Fetal Origins Hypothesis, Maternal Fasting, and Child
Human Capital Formation
There is a growing literature on the \fetal origins hypothesis" which links adverse
condition and inadequate nutrition in utero to later life health outcomes. Numerous
evidences show that critical developments take place during these nine months in utero
that are important for future health and well being of an individual (Almond and
Currie, 2011; Currie, 2009). Early studies were based on association between exposure
to shocks and health outcomes. To get causal estimates researchers have exploited
shocks which are exogenous to mothers. Some of these studies used historic events,
such as famine and disease outbreak, to get causal estimates. Unlike these events,
Ramadan is a recurring event, and it causes less severe nutritional compromise. Only
recently, nutrition shock due to maternal fasting during Ramadan has been studied
by economists. Below I discuss the literature on maternal fasting behavior during
Ramadan, fasting eect on maternal health and fetal health and identication issues
in existing studies.
Maternal fasting behavior during Ramadan | Ramadan is a holy month to Mus-
lims, as Prophet Muhammad received his rst revelation in this month. Among the
ve basic pillars|which are regarded as obligations to every Muslim| fasting during
13
Ramadan is one. During the entire month of Ramadan, fasting entails refraining
from eating, drinking, smoking as well as sexual engagement from dawn to dusk. The
fasting hours varies considerably by geographic location and the season Ramadan is
overlapping with. The further the location is away from the equator, the higher the
variation in fasting hours by season. Even though Ramadan is an obligation, it is
exempted for children, breastfeeding and pregnant women, elderly, ill, and travelers.
It is puzzling to observe that pregnant women fast during Ramadan, even though
they are exempted from fasting. Islamic law requires that if women miss fasting during
pregnancy, they should make up for the missed days after delivery. This requirement
as well as guilt and cultural expectations may also deter women from seeking the
exemption (Reeves, 1992; Mirghani et al., 2004; Robinson and Raisler, 2005). Based
on several studies from several countries, Almond and Mazumder (2011) note that
about 70 to 90 percent of pregnant women fast during Ramadan.
Fasting and maternal health during pregnancy | medical research is extremely
insightful in understanding why it is not recommended for pregnant women to fast or
skip meals for prolonged time. Metzger et al. (1982) study eect of fasting on level of
circulating fuels and glucoregulatory hormones among pregnant women in their third
trimester. They nd profound changes in biochemical measures among pregnant
women when fasting was prolonged from 12 hours to 18 hours. These changes are
known as \accelerated starvation". Further, Meis et al. (1984) nd among pregnant
women glucose concentration is signicantly lower after eight hour of fasting during
daytime compared to nighttime. These studies conclude that pregnant women should
not fast or skip breakfast during pregnancy.
Maternal fasting and fetal health | based on Jaddoe and Witteman (2006) study,
Almond and Mazumder (2011) point out two prominent hypotheses which describe
the eect of fasting on fetal health. The rst is \fetal under-nutrition". Nutritional
14
deprivation in utero leads to adaptations which are benecial in short term but causes
lower birth weight. These permanent changes in physiology and metabolism also make
the body more susceptible to heart disease and diabetes. Karimi (2014) provides detail
accounts on why fasting may aect the child height. He describes two phases of fetal
development pre-ossication and post ossication process. Based on the writings
of Gluckman and Hanson (2004) and Cooper et al. (2006), he notes that maternal
nutrition aects both stages of fetal growth. Height also re
ects phenotype and
genotype in
uences (Martorell and Habicht, 1986).
The second hypothesis is that nutritional restrictions hinder the development of
a placental enzyme that is necessary for converting cortisol into inactive cortisone.
Thus, it exposes the fetus to excessive amounts of cortisol. Kapoor et al. (2006) and
Seckl and Holmes (2007) suggest that glucocorticoids such as cortisol in utero leads
to a reprogramming of the hypothalamic pituitary adrenal axis (HPA) which in turn
may aect diabetes chances, cognitive functioning, and high blood pressure. Diken-
soy et al. (2008) reported that at the time of pregnancy fasting during Ramadan is
associated with an increased level of cortisol during pregnancy. Rizzo et al. (1991)
nd association between maternal fasting during pregnancy and cognitive function-
ing. Studies based on animal linked early pregnancy exposure ketone to neurological
impairments (Hunter and Sadler, 1987; Moore et al., 1989; Sheehan et al., 1985).
Epidemiological literature | these ndings based on above mentioned studies pro-
vide foundation for examining the eect of maternal fasting during Ramadan on child
health outcomes. Epidemiological studies nd association between maternal fasting
with reduced fetal breathing, fetal heart rate in utero, and psychomotor development
(Mirghani et al., 2004; DiPietro et al., 2007; Cross et al., 1990). However, Malhotra
et al. (1989) and Mirghani and Hamud (2006)) do not nd any eects on APGAR
scores and birth weight.
15
Almond and Mazumder (2011) note several limitations of the epidemiological stud-
ies which examine the exposure eect of Ramadan on individuals' health outcomes.
First of all, most of those studies are based on a small number of observations. Sec-
ondly, those studies compare the eect on fasters and non-fasters assuming that the
decision to fast is exogenous. Thirdly, those studies do not disentangle the fasting
eect from the seasonality, as they are based on Ramadan overlapping with only one
season.
Almond and Mazumder (2011) and successive studies (Ewijk, 2011; Almond et al.,
2014; Majid, 2015; Karimi, 2014; Ewijk et al., 2013) also discuss that the impact of
Ramadan exposure on outcomes of interest should be analyzed allowing for fertility
selection. However, in terms of analyzing fertility selection relative to Ramadan,
these studies have only emphasized whether or not a pregnancy has overlapped with
Ramadan. It is important to note that these women may not only take a decision
of whether to have a pregnancy overlapping with Ramadan but also breastfeeding
that child in the subsequent Ramadan.
8
This study is the rst to emphasize that the
fertility timing relative to Ramadan may be more complicated than depicted in the
current literature.
Economics literature | the application of Intent To Treat (ITT) analysis distin-
guishes the study of Almond and Mazumder (2011) from the epidemiological studies
on Ramadan. They are also the rst to study the impact of maternal fasting during
Ramadan on child outcomes in the economics literature. Given that human gestation
is about 280 days long and the Islamic calender year is about 354 days long, there
will be some pregnancies which will not be overlapped with Ramadan but some will
be. ITT anaylsis allows them to get rid of the compliance problem related to fasting,
8
Even though pregnant women and breast feeding can seek exemptions, they have to make it up
for their missed fasting later on.
16
and the dierence between children exposed and not exposed to Ramadan can be
analyzed. Under the assumption that parents do not selectively time birth relative to
Ramadan, it gives causal estimates of the impact of experiencing in utero nutrition
shock. Moreover, unlike most of the epidemiological studies, the study of Almond
and Mazumder (2011) is also based on a large number of cohorts.
Almond and Mazumder (2011) use data from Michigan, Iraq and Uganda. They
study the impact of nutrition compromise due to exposure to Ramadan on birth
weight using data from Michingan, and on various forms of disabilities using census
data from Iraq and Uganda. They nd children who were exposed to Ramadan
have lower birth weights and are more likely to be disabled. Following Almond and
Mazumder (2011), Ewijk (2011) nds that exposure to Ramadan increases the chances
of developing health problem such as coronary heart disease and type 2 diabetes.
Majid (2015) nds that exposure to Ramadan leads to fewer hours worked and self-
employment in later life.
9
. Ewijk (2011) and Ewijk et al. (2013) use the wave 3 of the
IFLS. On the other hand, Majid (2015) primarily use the wave 4 of the IFLS. Using
the English register data, Almond et al. (2014) nds that maternal fasting during
Ramadan leads to lower test scores. Karimi (2014) uses 80 Demographic Health
Surveys (DHS) from 35 countries spanning 27 birth cohorts and nds that exposure
to Ramadan in utero leads to shorter stature for children. Ewijk et al. (2013) also
nds negative association between Ramadan exposure and height using the IFLS wave
3. Unlike Karimi (2014), they focus on adult height.
Although the studies in economics literature highlight the importance of identi-
fying impact of maternal fasting during Ramadan on relevant outcomes allowing for
selection, they do not nd any evidence on parental selection in Ramadan exposure.
The key limitation in these studies is that they do not examine parental selection in
9
In this context, self-employment is regarded as a poor labor market outcome.
17
Ramadan exposure allowing for time varying changes. The above mentioned studies
also have data limitations. In their Michigan and Iraq data, Almond and Mazumder
(2011) could not identify the religion of the mother. They used Arab as proxy for
Muslims in their Michigan data. The birth data from Uganda, Iraq, Indonesia and the
DHS data sets were also self-reported. This could be a serious problem as misreported
birth dates will lead to wrong classication of exposure to Ramadan. In comparison
to the some of the past data sets used in this literature, in the MHSS data the religion
of the mother can be clearly identied as well as reliable information on birth date
can be extracted for a large number of cohorts. Unlike the past studies which have
used the IFLS, this chapter uses the birth information from the pregnancy histories
of the mothers from all four waves of the IFLS and restricts the recall period up to
5 year maximum to minimize the recall bias. The benet of using birth data from
pregnancy histories is that it includes all the live births.
Compare to these earlier studies, this chapter documents the eect of an intensive
family planning program on selection in birth timing. The time varying component
under study here is the family planning program in Matlab, as the chapter uses birth
data of the birth cohorts who were born before and after the program. This chapter
illustrates that if the timing and placement information of the family planning pro-
gram are ignored, we might incorrectly conclude that there is no presence of parental
selection in Ramadan exposure. Moreover, this chapter shows that presence of selec-
tion may produce biased estimate of the impact of the exposure to Ramadan. Thus,
this chapter contributes to a set of literature that documents importance of tak-
ing selection into account for causal identication. Thomas (2009), and Brown and
Thomas (2011) document selection in exposure to 1918 in
uenza epidemic in USA.
Prior to their papers, Almond (2006) shows that exposure to In
uenza epidemic have
implications for human capital formation. Brown and Thomas (2011), however, nd
18
that fathers of those who were exposed were from lower SES, were older, had higher
fertility, less likely to be white and less likely to be veteran of WW1. Controlling for
these observables and using similar exposure denition as it was dened by Almond
(2006), they do not nd any eect of In
uenza on child human capital formation
and labor market outcomes. Buckles and Hungerman (2013) document variation in
maternal characteristics with respect to season of birth. They nd that winter births
disproportionately realized by teenagers, unmarried women, less educated mothers,
and white women. Their ndings have implications for using season of birth as an
instrument for schooling in estimating returns to schooling. Prior to their paper, Lam
et al. (1994), using data from USA, show that parents are more likely avoid concep-
tions during extreme summer heat and seasonal variation in births diers between
two races namely the white and the black. Thomas (2009) documents selection in
fertility during famine in Bangladesh. Using the MHSS 1996, he nds that during
the famine in 1974 parents who gave birth were dierent in observable characteris-
tics than who did not. Pitt (1997) describes the importance of taking into account
fertility and mortality selection in estimating determinants of child health. Following
Pitt (1997), Maitra and Pal (2008) document the importance of fertility selection in
estimating relation between birth spacing and child mortality.
2.3 Matlab Family Planning and Child Health
Matlab is a thana (sub-district) in Chandpur District in Bangladesh. It is located
55 kilometers of South-East of Dhaka. The International Center for Diarrhoeal Dis-
ease Research, Bangladesh (ICDDR,B) established a demographic surveillance system
(DSS) in 1966. In the DSS areas, the record of birth, death, and migration (in and
out) are collected from the start of the project. In October 1977, the DSS areas
19
were contracted to 149 villages by excluding 84 villages. The family planning and
health project was launched in 70 villages (treatment areas) and the remaining vil-
lages were comparison areas. No report of using randomization mechanism has been
found (Schultz 2009). Figure 2.1 in appendix shows that the treatment areas were
grouped into clusters.
10
Schultz (2009) argues that the clustering of villages into the
treatment areas retain the spillover eect. Table 3 in appendix presents 1974 census
data which shows that the treatment and the comparison area were very similar ex-
cept for few observable characteristics such as sources of drinking water, number of
cows, and age of both household head and his/her spouse.
Based on several studies, Joshi and Schultz (2013) point out the contraceptives
use went up from 7% to 33 %. Sinha (2005) nds that the family planning program in
Matlab lead to 14% reduction in life time fertility. Barham (2012) also describes the
other treatments added to the treatment areas which are documented in Bhatia et
al. (1980), Phillips et al. (1984) and Koenig et al. (1990). In October 1977, the fam-
ily planning program began in the treatment areas through the provision of modern
contraceptives. From June 1978, pregnant women received tetanus toxoid vaccina-
tion, and also pregnant women in their last trimester pregnancy received iron and
frolic acid tablets. From March 1982, the children aged from 9 months to 59 months
in the treatment area 1 received measles vaccine. This program was expanded to
treatment area 2 on November 1985. From January 1986, DPT, polio, and tuber-
culosis immunization were given to children under age 5. Later in 1986, Vitamin A
supplementation for children under age 5 and nutritional rehabilitation for those who
were nutritionally risky were added to the treatment areas. In appendix section, I
reproduced the table 1 from Barham (2012) which gives a summary of the programs
introduced in the treatment areas and age cohorts the programs have aected.
10
I am thankful to the Matlab HDSS of ICDDR,B for allowing me to use this map.
20
2.4 Data
Matlab Household and Socioeconomic Survey
The rst source of data used in this study is the Matlab Household and Socioeco-
nomic Survey (MHSS) 1996 which was funded by National Institute of Aging and was
collaborative eort of RAND, the Harvard School of Public Health, the University
of Pennsylvania, and the University of Colorado at Boulder. The primary sample
was drawn from a probability sample of 2,883 baris from 7,440 baris in the DSS
1994 sample frame. baris usually consists of cluster of households in close physical
proximity. In all baris, interviews were completed in 2,781 out of 2883 eligible baris.
Within each bari, up to two households were randomly selected. For each baris, one
household was randomly chosen and designated as primary household.
11
If there are
more than two households, the second household was randomly chosen and termed
as secondary household.
12
Otherwise, the second household was designated as the
secondary household. Out of the 2,781 baris, 94 baris were inappropriately inter-
viewed and therefore disregarded from analysis which leaves us with 2,687 baris. Out
of these baris, 656 baris consisted of one household, and rest of them had two or more
households. Ideally, there should be 2,013 households but the survey team could nd
only 1,677 households. The remaining secondary households are purposive sample
based on relationship to the rst household. In this chapter, I limit my analysis only
to primary households.
In the survey mothers were asked about birth dates of each of their children.
Later the birth dates were matched with the DSS data sets for their consistency
and accuracy. Although the DSS started in 1966, the events (i.e. birth, marriage)
11
In the data set, the primary households are denoted as status = 1.
12
In the data set, the secondary households are denoted as status = 2.
21
which took place beginning from 1974 were linked to computerized system of the
DSS. During the data collection process of the MHSS, the consistency and accuracy
of the dates of the events (i.e. birth, marriage) was only checked for the data which
was computerized. Therefore, I have reliable birth dates for 22 birth year cohorts
from 1974 to 1995. All birth dates before 1974 are self reported (Rahman et al.,
1999). There are also some other limitations with this data. I obtained the birth
data from the pregnancy history of the women interviewed in the MHSS 1996. This
is a limitation because I can know about births prior to 1996, only if the women living
in sampled household survived until 1996.
Another limitation of the data is that for some births only month and year of birth
are known, and the birth dates are replaced with zero. It also varies considerably
between the treatment areas and the control areas. There are 2086 births which had
date zeros out of 8573 from 1974 to 1995. Out of the 2086 births, the treatment
areas had 856 births and the control areas had 1230 births.
13
Moreover, among the
births which have birth date information, 11 % of births were reported to take place
on date 1. I drop these births which report date 1 to avoid wrong classication of
births relative to the time of Ramadan they were born. The dierence in probability
of reporting date 1 is not statistically signicant between the treatment and control
areas.
After dropping the unknown birth dates, months and years, the nal data set
consists of 5754 births for birth year cohort from 1974 to 1995. Out of 5754 births,
2638 births are from the treatment areas and 3116 births are from the control areas.
Out of 2638 births in the treatment areas, 2290 were born to Muslim mothers and
348 were born to Hindu mothers. Out of 3509 births in the control areas, 2981 were
13
This could be due to the fact that family planning lead to decrease in total number of births
in the treatment areas. Hence missing births also declined in treatment areas.
22
born to Muslim mothers and 135 were born to Hindu mothers.
For the height analysis, I restrict the sample to the birth cohorts born between
1986 and 1995. Therefore, the oldest child in the height sample was less than 10
years old in 1996. This limits the concern that the children in the sample may have
migrated away as well the concern regarding adolescent growth spurt (Falkner and
Tanner, 1986). In the MHSS 1996, height was not measured for all children but only
2 randomly selected out of all the children present in the household (Rahman et al.,
1999). After limiting sample to children whose birth dates which are not missing or
reported date 1, there are 669 males and 655 females.
To study the association between exposure to Ramadan and child height, I match
the birth dates from the mother's pregnancy with the birth month and year of the
individuals surveyed in the Matlab. I later match anthropometric data for each
individual. I limit this study to only single birth. I found only 28 twins in this data.
In the MHSS, the village level facility information was collected for 140 out of
the 141 villages. Therefore, when the village level time varying characteristics are
included in the regression specication, the sample would be based on 140 villages.
Another concern is the dening for years of education. The years of education
is only known if the household member went to a formal school. An overwhelming
majority of the household members, if educated, go to a formal school. However, some
went to Madrasa, BRAC school or Maktab.
14
Some Madrasa and BRAC schools
also provide educations very similar to formal schools and curriculum approved by
the government. It is not clear whether the household member went to a Madrasa or
BRAC school which provide formal education or not. I dene zero years of education
if the household member went to Madrasa, BRAC and Matab to avoid missing data.
Table 2.1 summarizes the data from Matlab. The mothers on the average have
14
Less than 3.5 percent of all the births can be attributed to women who went to these schools.
23
2.17 and the fathers on the average have 3.28 years of education. The average age
of mothers at the survey is 37.40. About 13 % of the total births were not exposed
to Ramadan at the time of pregnancy and 8 % the total births were born during
Ramadan. This implies 79% of the births were exposed to Ramadan at some point
during the pregnancies.
15
In the sample, 8 % of total births are Hindu births. 52 %
of the total births resulted in male births which is close to the natural sex ratio at
birth.
Indonesian Family Life Survey
In this study, the second source of data comes from all four waves of the Indone-
sian Family Life Survey (IFLS) which were conducted on 1993, 1997, 2000 and 2007
respectively. At the time of the rst survey 1993{94, the sample drawn was repre-
sentative of 83% of the population residing in 13 provinces out of 27 provinces of
Indonesia's. Within each of the 13 the provinces, the enumeration areas (EA) were
randomly selected for inclusion in the nal survey. In the rst wave, 7224 households
were interviewed and detailed individual level information were also collected includ-
ing the age and education of the household members. The later waves sought to
follow up the same household and the re-contact rates of households from rst wave
were 94.4%, 95.1% and 93.6% in the second, third and fourth wave respectively. The
low attrition rate is one of the appealing features of IFLS. The attrition rate is lower
because the migrant households and individuals were carefully tracked in the later
waves. Thomas et al. (2012) point out that the migrant and non-migrant individuals
are dierent in observable characteristics. The design of IFLS to follow up on the
migrant households and individuals reduces the concern of any selection bias due to
15
Some births may still get classied as being exposed. However, past literature have used a very
similar exposure denition.
24
migration and disappearance from the original households.
During each wave of the survey the reproductive age women were asked about
their pregnancy histories. These pregnancy histories includes the birth date, birth
outcome, gender of the child, age of the mother at birth for all the births. All the
information regarding birth dates and outcomes are self-reported. For the purpose of
the analysis of this chapter, I restrict the births which took place within the last ve
years from the survey wave year to reduce the recall bias. This gives us 12,331 births
from all four waves of the IFLS. Out of these births, 1430 births have some missing
birth dates and months. The information on mother education is missing for about
59 births. Moreover, the community information is also missing for about 144 births.
After dropping these births, the nal sample consist 10,753.
16
One key dierence
between this chapter and other studies which used IFLS is that the other studies
do not use the data from the pregnancy histories of women to examine selection in
Ramadan exposure. Ewijk (2011) and Ewijk et al. (2013) use the third wave of the
IFLS and Majid (2015) use the fourth wave of the IFLS. However, the advantage of
using pregnancy history is that pregnancy history contains information on all births.
One advantage of the IFLS over the MHSS is that the selection in maternal mor-
tality is less of a concern. In the MHSS, I have information on pregnancy history for
last 22 years for women who were alive during the survey year. With all four waves
of the IFLS, I can restrict the pregnancy histories of the women for up to ve years
and still manage to get birth information for 16 birth year cohorts.
17
Table 2.2 summarizes the sample from IFLS. The mothers on the average have
7.45 years of education which is far more than the average years of education of
16
Majid (2015), Ewijk (2011) and Ewijk et al. (2013) also take similar approach to missing birth
dates. However, they loose a much higher percentage of sample due to missing birth dates.
17
I have only 16 birth year cohorts due to the fact that IFLS 2 and 3 were conducted in less than
ve years from the previous wave. The birth history was restricted up to the last wave.
25
mothers in MHSS. The mothers on the average are 28.89 years old at the time of
survey. Interestingly, the proportion of children who are not exposed to Ramadan in
utero is 13% which is very similar to MHSS. Moreover, the proportion who are born
during Ramadan is again 0.08%. The non-Muslim accounts for 11% of the births.
About 51.2% of the total births in IFLS resulted in male births.
2.5 Ramadan Measures
For a given year, the dates of the Western calender which overlaps with the Ramadan
Month of the Islamic calender can be matched.
18
From the birth data only birth
date of an individual is known but gestation period is unknown. However, it can be
observed that how many months after Ramadan an individual `i' was born. Generally,
the gestation period for human is about 280 days long. For each date of birth, I
construct century day code (CDC) following Almond and Mazumder (2011). Here,
ramadan
0
is the month of the month of Ramadan and takes a value of 1 if the
individual `i' was born during the month of Ramadan and 0 otherwise. Similarly,
ramadan
1
would be the next month after the month of Ramadan and takes a value
of 1 if the individual `i' was born next 30 days after Ramadan and 0 otherwise. Thus,
ramadan
X
would imply X month after Ramadan and would take a value of 1 if the
individual `i' was born during the X month after Ramadan and 0 otherwise. Based
on the denition of these variables, for X < Y , ramadanXtoY can be constructed
where ramadanXtoY takes a value of 1 if the individual `i' was born exact X to Y
months after Ramadan and 0 otherwise.
Given that the gestation period of human is approximately 9 month long, the
individuals who were born duringramadan
10
andramadan
11
are not exposed to Ra-
18
Following Almond and Mazumder (2011), we construct the Ramadan month from Institute of
Oriental Studies at the University of Zurich using their website www.oriold.uzh.ch/static/hegira.html
26
madan in utero. A variable nonexposed is constructed that takes a value of 1 if an
individual is born 10 to 11 months after Ramadan and 0 otherwise.
2.6 Empirical Strategy
The discussion on empirical strategies is divided in three subsections. First subsec-
tion describes the potential empirical challenges. Second subsection discusses the
empirical frameworks for examining the impact of the family planning on change in
birth time relative to Ramadan and change in parental selection in Ramadan expo-
sure. Third subsection discusses the empirical frameworks for analyzing association
between Ramadan exposure and the child height.
Empirical Challenges
Endogenous program placement | as described earlier, the family planning program
in Matlab was not randomly allocated between the treatment and control areas.
Although the treatment and control areas were very similar in observable character-
istics in 1974 (Barham, 2012; Joshi and Schultz, 2013), they could potentially dier
in unobservable dimensions. In the regression equations village level xed eects are
included which will control for all the village level time invariant unobservables. As
long as the placement of the family planning program are not correlated with time
varying unobserved factors, the dierence-in-dierences strategy will provide us with
unbiased estimate.
Selection in birth | the MHSS birth sample comes from pregnancy histories of the
women. One key issue, in the case of the MHSS, is that because of family planning
program some women may choose to avoid giving birth completely. For example,
27
family planning program may allow some parents to delay births who may not have
done so in the absence of the program. These parents may also not have avoided
birth relative to Ramadan, had they chosen to have given births. Thus, in the data
only those parents who have given birth and avoid birth relative to Ramadan might
be observed.
To address this concern, the proportions of women who had ever given a live
birth between the control and treatment areas are compared in Table 2.3. Panel A of
Table 2.3 shows about 93{94% of the women have at least one birth and women in the
treatment areas are 1% less likely to give birth but the dierence between two areas
is not statistically signicant. Among those who had not given a birth yet (panel B),
the years of education of the women, education of their husbands, their ages at the
time of survey are compared. The married women who had not given a birth yet in
the treatment areas are older, less educated and has husband with less education than
their counterparts in the control areas. However, these dierences are not statistically
signicant. A similar analysis is performed on the Muslim married women who had
not given a birth yet (Table 2.4) and the conclusions remain the same. It is also
important to note that with the MHSS, I analyze birth cohort spanning over 22 years
and in Matlab where eventually almost every woman gives birth to a child. Given
that there exists a large number of birth cohorts and the fact more than 90 percent
of women had given one live birth at some point her reproductive age, selection in
live birth is of less concern in this context.
Maternal mortality and migration attrition | the birth sample in the MHSS comes
from the pregnancy histories of the women observed in Matlab in 1996. The family
planning program and maternal health interventions may aect the mortality of the
women in the treatment areas. It is not clear how this will aect the bias. If women
from the lower SES are more likely to benet from these programs and they are less
28
likely to avoid births relative to Ramadan, then we will observe a downward bias in
the estimation of the program eect. On the other hand, if women from higher SES
are more likely to benet from these programs and they are more likely to avoid birth
relative to Ramadan, we will observe an upward bias. Ronsmans et al. (1997) nd
that maternal mortality from all causes declined in both the treatment and control
areas from 1976 to 1993, but the dierence is not statistically signicant between the
treatment and the control areas.
Migration may aect the results, if the migration pattern diers between the
treatment and control areas because of the program placement. Joshi and Schultz
(2013) point out that the relation between fertility outcomes and female migration
in and out from the Matlab study areas is not statistically signicant. Barham and
Kuhn (2014) nd that male and female out migration was higher in the control areas
between 1983 and 1988. They attribute the lower out migration in the treatment areas
compared to that in the control areas to the health interventions implemented in the
treatment areas during 1983{88 period. The results of female out migration in the
study of Barham and Kuhn (2014) is not robust to inclusion of village xed eects.
As mentioned earlier, village xed eects are included in all regression equations
employed in this chapter. In addition to this, one advantage of this analysis here
is that program eect is studied both for Muslims and non-Muslims. If migration is
responsible for the birth pattern relative to Ramadan, this may show for non-Muslims
as well| as the data for both Muslims and non-Muslims are available from both the
treatment and control areas. Furthermore, Joshi and Schultz (2013) show that pre-
family planning program fertility rate is very similar between the treatment and the
control areas using the MHSS data. Barham (2012) shows that the levels of cognitive
functioning, in the MHSS data, are very similar between the individuals from the
treatment and control areas who were born prior to the health interventions in the
29
treatment areas. In the context of Ramadan exposure, this study also do not nd any
dierence in Ramadan exposure between the two areas prior to the implementation
of the family planning program.
Another concern is that the information on birth location of the births are missing
in the data. In the chapter, it is assumed that the location of the mother at the time of
survey as the birth location of the individuals born in the Matlab. Joshi and Schultz
(2013) point out that female migrations to the control areas from the treatment areas
and vice-versa are very low. Moreover, Joshi and Schultz (2013) and Sinha (2005)
also consider that the location of the mother at the time of the survey as the birth
location of the children born in Matlab.
In the case of the IFLS, selection due to maternal mortality is less of a concern
as the subsequent waves were conducted at sucient short intervals and pregnancy
histories from at most last ve years of each survey wave are used for the analysis
of this chapter. Moreover, selection due to migration is even lesser of a concern as
the survey team carefully followed the migrant households and individuals in the
subsequent waves.
Trend dierence | another key concern in any dierence-in-dierences strategy
is that the treatment and control villages may experience dierent time trends and
establishment of other facilities. The dierence-in-dierences point estimate may cap-
ture the inherent trend dierences between the treatment and control areas. As long
as the trend is similar for both Muslims and Hindus, the point estimates give us the
causal eect of the family planning program. I also control other time varying char-
acteristics such as timing of electricity connection, village tube-well establishment,
satellite clinic establishment, and timing of the family planning program interacted
with the large market.
19
Electricity may allow parents to access information about
19
In MHSS 1996, there is only information whether the village has large market but no information
30
contraceptives from television and radio which in turn may aect the use of contracep-
tives. Satellite health clinic may also provide provide contraceptives or information
on using contraceptives. Proximity of a large market may re
ect the cost of getting
contraceptives. Nearness of tube-well will aect cost of getting water for women.
Birth Timing
To examine the impact of the family planning program in Matlab on the fertility
selection relative to Ramadan, birth data is needed both before and after the initiation
of the program. Recall that reliable birth data from the MHSS is available for 22 birth
year cohorts born from 1974 to 1995 for both the treatment and control areas, and the
family planning was initiated in October, 1977. It could be tempting to think that
a dierence-in-dierences strategy can be applied on Muslim births to get the causal
estimates of the eect of the family planning program on fertility selection relative to
Ramadan controlling for xed seasonal variation in births. However, one key concern
still remains for causal identication. Ramadan completes the full circuit of the
western calender in 33 years, but the MHSS has reliable birth data for only 22 birth
cohorts. As a result, Ramadan is not balanced with respect to seasonality. To analyze
whether this is a concern, a dierence-in-dierence strategy can be applied on the
Hindu births similar to that applied on the Muslim births. The underlying assumption
is that both Muslim and Hindus are aected by residual seasonality, if there is any,
and only Muslims have incentive to respond to the family planning program in timing
birth relative to Ramadan. If the dierence-in-dierences estimate on Hindu births is
not statistically dierent from zero but on Muslim births it statistically dierent from
zero, this would imply that Muslims are indeed timing birth relative to Ramadan and
the dierence-in-dierences estimate provides us a causal estimate.
on from what year it has large market.
31
I estimate the following linear probability model for individual `i' whose mother
lives in village `j' and was born in month m and year t:
R
ijmt
=
1
Post
mt
+
2
Post
mt
Treated
j
+
3
X
ijmt
+
4

jt
+
m
+
t
+
j
+
ijmt
(2.1)
The dependent variable R takes various measures of birth in months from Ra-
madan. These measures take a value of 1 if true and 0 otherwise. For example, if
R represents the variable ramadan8to9, in equation (1) ramadan8to9 would take a
value of 1 if an individual `i' born whose mother lives in j in month m and year t were
born 8 to 9 months after Ramadan and 0 otherwise. It is assumed that the individuals
are born in the same locations their mothers live at the time of the survey.
20
Post
takes a value of 1 from August from 1978 and 0 otherwise. Although the family plan-
ning program started in October, 1977 (Joshi and Schultz, 2013), the female village
workers started working in November, 1977 (Bhatia et al., 1980).
21
Post is identied
because the year xed eects do not absorb the coecient of post, as the program
did not start from the beginning of a year. The Treated variable takes a value of 1
for villages which received the family planning program and 0 otherwise. The inter-
action termPost
mt
Treated
j
captures the eect of contraceptive program on birth
timing. In the regression equation (1), the village xed eect
j
absorbs the main
eect ofTreated as well as controls for the village level unobservables. X
ijmt
is set of
mother level observables like mother education in years and mother age at the time
of survey. The mother ages at the survey are specied as spline functions with knots
at every 5 year interval starting from age 15 to allow
exibly for non-linearities.
jt
20
This is not a serious concern, as Joshi and Schultz (2013) point out migrations from the treat-
ment to the control areas and vice-versa are very low.
21
The start date of the program and the post date is dierent because those who were conceived
before November 1977 were not exposed to the program. Thus post starts from August, 1978 nine
months from the start of the program.
32
is set of village level time varying characteristics such as from what year the village
has tube-well, electricity, satellite health clinics, and post interacted with whether
the village has a large market. I regress the equation (1) by splitting the sample by
Muslim and Hindu and compare the coecient ofPost
mt
Treated
j
for both Muslims
and Hindus. Since only Muslim population is aected by fasting during Ramadan, it
allows us disentangle whether or not the change in birth relative to Ramadan after
the initiation of the intensive family planning program is due to Ramadan or any
other event which may aect both Muslims and Hindus.
22
In a full sample combining
the birth sample of both Hindu and Muslims, I can recover the point estimates for
Muslims and Hindus by interacting Post
mt
Treated
j
and other relevant variables
with Hindu and compare it with Post
mt
Treated
j
. The results will not dier in a
split sample and full sample as long as same xed eects and controls are included in
the regression equations.
The provision of a family planning program may also change the observed charac-
teristics of the children who are exposed to Ramadan and not exposed to Ramadan
in utero. Here, the observed characteristic of interest is mother years of education,
as based on several studies Schultz (2002) point out mother years of schooling has
more eect on child health, schooling, and adult productivity than same years of
father's education. Moreover, there is a substantial literature which documents that
the better mother education is positively associated with better child height and
BMI (Behrman and Deolalikar, 1988; Behrman and Wolfe, 1989, 1984; Strauss and
Thomas, 1995, 1998).
To analyze selection in Ramadan exposure, I limit the sample to birth cohorts born
22
During the month of Ramadan the prices of goods and services and number of national holidays
may change and the mothers may want to time birth based on that. The underlying assumption is
that both Hindus and Muslims are aected by the changes in prices and other events like number
of national holidays.
33
from 1974 to 1982. The family planning program may aect the female education
who in future will be mothers. Thus, limiting the sample to a smaller birth cohorts
reduces the concern that the education of the mothers, who are giving birth from
1974 to 1982, will be aected by family planning program. I divide this sample in
three categories: Muslim births in the control areas, Muslim births in the treatment
areas, and Hindu births in the treatment areas; and I estimate the following linear
probability model for individual `i' whose mother lives in village `j' and was born in
month m and year t:
nonexposed
ijmt
=
1
motherage
ijmt
+
2
mothereduy
ijmt
+
3
Post
mt
mothereduy
ijmt
+

m
+
t
+#
j
+"
ijmt
(2.2)
The variable nonexposed takes a value of 1 if children were not exposed to Ra-
madan during pregnancy and 0 otherwise. The variable mothereduy is measured as
years of education andmotherage is measured as age at the survey.
t
,#
j
and
m
are
year xed eects, village xed eects, and month xed eects respectively. With equa-
tion (2), the consequence of ignoring placement and timing information of the family
planning program can be analyzed. For that, the variablePostmothereduy
ijmt
will
be omitted and equation (2) will be regressed adding the Muslim birth sample from
both the treatment and control areas for the same birth cohorts.
Recall that association between Ramadan exposure and child height is studied for
children who are less than 10 years old. Therefore, the oldest children in the sample
are born in 1986 when the family planning is already in place in the treatment areas.
Therefore, a variant of equation (2) will be used where instead of post treated
j
will
be interacted with mothereduy and Block will interacted with
t
to capture the
34
additional interventions implemented in the treatment areas.
With all four waves of pregnancy history data from the IFLS, whether exposure
to Ramadan varies by mothers' level of educations is examined. For that purpose I
use a variant of equation (2) where instead of village xed eects, I apply community
xed eects. In addition to month and year xed eects, birth date xed eects
are included in the model.
23
The standard errors are clustered at community level
instead of village level. It is important to reiterate that with the MHSS data we are
examining selection in Ramadan exposure due to family planning program but with
the IFLS we are only examining association between Ramadan exposure and mother
education.
In utero Ramadan Exposure and Child Height Correlations
If we nd that parents are selectively timing birth relative to Ramadan, it is important
that we study the eect of maternal fasting on child health controlling for selection.
Although the family planning program is exogenous to households in Matlab, it may
aect the child health through the child quality quantity trade o. Therefore, due
to exclusion restriction violation the family planning program can not be used as
an instrument for birth time relative to Ramadan in child health outcome function.
However, the correlations between Ramadan exposure and child height can be ex-
amined in both the treatment areas and control areas. If a mother with higher SES
avoids pregnancy overlapping with Ramadan, we may nd the correlations between
pregnancy overlapped with Ramadan and child height is negative. On the other
hand, if a mother with higher SES does not time birth relative to Ramadan in control
areas because of the absence of the family planning program, we may not nd any
23
Since the IFLS data is based on self-reported birth dates, this would reduce the concern if there
is any systematic relation between date of birth and Ramadan exposure during pregnancy.
35
correlation between child height and pregnancy overlapped with Ramadan. Under
this scenario we should be worried that perhaps selection is driving the results rather
than maternal fasting during Ramadan.
I regress following equation for individual `i' whose age is `a' living in village j and
born in month `m' :
H
ijma
=
1
Exposed
ijma
+
2
BornduringRamadan
ijma
+
3
Hindu
ijma
+

4
Hindu
ijma
Exposed
ijma
+
5
Hindu
ijma
BornduringRamadan
ijma
+

6
malecHindu
m
+malec
a
+'
j
+
ima
(2.3)
When the outcome of interest is child height (H). Currie and Vogl (2013) argues
that child height is a good proxy for child birth weight. Height represents long run
nutritional status (Strauss and Thomas, 2007). The variable Exposed takes a value
of 1 if the child was born between 1 and 9 months after Ramadan and 0 otherwise.
The variable BornduringRamadan takes a value of 1 if the child was born during
the month of Ramadan and 0 otherwise. BornduringRamadan is dened separately
from Exposed because individuals who were born during Ramadan were partially
exposed to Ramadan in utero. The omitted category in the regression equation (3)
are the individuals who are not exposed to Ramadan.
m
,
a
and'
j
are birth month,
age in months xed eects and village xed eects respectively. The malec takes a
value of 1 if the child is a male and 0 otherwise. The interaction term malec
a
absorbs the main eect of malec. Moreover, it takes into account dierential growth
pattern for male and female children. The standard errors in all regression equations
are clustered at village level.
36
2.7 Results
Family planning program and Birth Timing
To study selective timing of birth, regression of ramadan
X
on post, posttreated
are performed, where X takes the values 0 to 11, controlling for birth month and year
xed eects as well as village xed eects. The results are presented in Table 2.5.
The cohorts under study were born from year 1974 to 1995. It shows women living
in the treatment areas are around 3 percentage points less likely to give birth 8
and 9 months after Ramadan after being exposed to the family planning program.
This demonstrates that women in the treatment areas are avoiding conceptions when
Ramadan is one month away or in the month of Ramadan. On the other hand, the
coecient of posttreated is positive for ramadan
4
to ramadan
7
and except for
ramadan
6
. For ramadan
6
, the coecient of posttreated is negative but close to
zero. For ramadan
0
to ramadan
2
the posttreated coecients are positive but
close to zero. It is important to note that in Table 2.5, the eect of family planning
program is examined on the birth for each separate month relative to Ramadan.
Grouping the months would allow us to gain more statistical power as well as be more
informative. Based on the sign of coecients, I construct variablesramadan4to7 and
ramadan8to9 where for exampleramadan8to9 takes a value of 1 if individual `i' was
born either during ramadan
8
or ramadan
9
and 0 otherwise. Each of ramadan8to9
and ramadan4to7 are then regressed on the same independent variables including
maternal years of education and spline fucntion of mother age at the survey with
several knots. To examine that these results are not driven by any other event other
than Ramadan, similar empirical strategies are applied on same Hindu birth cohorts.
Biological mother xed eects, which absorb the mother time invariant preference for
37
fasting, mother education, mother age at survey, time invariant health and nutritional
status such as height and time invariant asset level, are also applied as a further
robustness check.
In Table 2.6, the impacts of the family planning program on the births during
ramadan8to9 for both Muslims (column 1{3) and Hindus (column 4{6) are presented.
It shows that Muslims women|living in the treatment areas| are 6.2 percentage
points less likely to give birth 8 and 9 months after Ramadan (column 1) following
the family planning program. After the inclusion of other village level time varying
characteristics the magnitudes drops slightly but still remain signicant (column 2).
The inclusion of biological mother xed eect in contrast increases the coecient
magnitude (column 3). On the other hand, for Hindus living in the treatment areas,
no change in birth during ramadan8to9 is found following family planning program
in all three specications (column 4{6).
Similar to Table 2.6, the eect family planning program on birth duringramadan4to7
is analyzed in Table 2.7. Here, it shows that provision of a family planning program
actually increased the Muslim births by 8.6 percentage points during ramadan4to7
(column 1). For Hindus, no change in birth during ramdan4to7 (column 4) is found.
The inclusion of village level time varying characteristics do not change the result
for Muslims (column 2) that much and for Hindus it gives slightly positive estimate
(column 5) but still much less than that for Muslims. Even after the inclusion of
biological mother xed eects (column 3), the coecient magnitude are very similar
for Muslims. For Hindus, however, the coecient magnitude is 0:048. It should be
noted that when biological mother xed eects are applied, the coecient of Post
Treated are only identied within the same mothers who gave birth before and after
the family planning program. The sensitivity of the coecient of PostTreated
for Hindus, when biological mother xed eects are applied, could be due to small
38
number births by the same mother before and after the family planning program. To
examine results are not sensitive to splitting the sample in Hindu and Muslim births,
for the same dependent variables as in Table 2.6 and Table 2.7, Hindu is interacted
with PostTreated and other relevant variables, and regressions are employed on
the full sample. The results are presented in Table A1 which show that the same
results as in Table 2.6 and Table 2.7.
Birth Relative to Ramadan and Child Height
Based on the evidence provided by earlier literature, it becomes important to examine
the eect of exposure to Ramadan on child health. One advantage in this study is
that the association between Ramadan exposure and child health outcomes can be
studied for both the treatment and control areas. As mentioned earlier, child height
is used in this chapter as a measure of child health, as height represents long term
nutritional status (Strauss and Thomas, 2007). The results are presented in Table 2.8.
Pooling the height data of Muslim children from treatment and control areas who are
less than 10 years old at the time of survey, it is found that Ramadan exposure is
associated with about 2.78 centimeter decrease in child height (column 1) and it is
also signicant at the 10 percent level. However, when the analysis is done separately
for control and treatment areas, the results are striking. In control areas (column 2)
exposure is associated with decrease in height by 1.3 centimeter whereas in treatment
areas exposure is associated with decrease in height by 4.42 centimeters (column 3).
To explore the possibility that anything other than Ramadan is responsible for this
association in column (3), the association of Ramadan exposure with child height are
analyzed for both Muslim and Hindu children in treatment ares (column 4). The
coecient ofExposed for Muslims in column (4) is similar to that in column (3) but
39
for Hindus it is positive and joint F-test is statistically insignicant. This implies
the negative association between Ramadan exposure and height is only present for
Muslim children in the treatment areas. These results on height are puzzling. Suppose
we had height data of the children only from the treatment areas and produced the
results in column (3) and (4) in Table 2.8 following the empirical strategy of the
earlier studies (Almond and Mazumder, 2011; Ewijk, 2011; Almond et al., 2014;
Majid, 2015; Karimi, 2014; Ewijk et al., 2013) we might have incorrectly concluded
that exposure to Ramadan leads to shorter stature. However, when we have data
height data from both the treatment and control areas, we can not draw the same
conclusion even after using the same empirical strategies.
One explanation for observing negative eects in the treatment areas but no ef-
fects in control areas is that survival bias could be dierent between the two areas.
If children exposed to Ramadan, who might have been shorter on average, did not
survive till the survey period in the control areas, we might have observed similar re-
lation between exposure to Ramadan and child height. However, statistical dierence
in mortality of children between the treatment and control areas is not found in the
data. Moreover, it should be noted that mortality is also an extreme event.
It could be also possible that that the pregnant women in the treatment areas are
more likely to fast during Ramadan compared to the control areas. It is impossible
to rule out such possibility, as I do not have information on the fasting behaviors of
women residing in the treatment and control areas. However, such dierences might
seem unlikely due to the geographical proximity of both areas.
Apart from these possibilities, we might observe the results presented in column
(3) and (4) of Table 2.8, if there is parental selection in Ramadan exposure. If parents
with higher SES background are less likely to overlap pregnancies with Ramadan,
we might observe the dierence in child health outcomes between children who are
40
exposed and not exposed to Ramadan. Even though the treatment and control areas
are very similar in observable dimensions in 1974, reproductive age women, in the
treatment areas in Matlab, were provided with an intensive family planning program
starting from October,1977. Following the family planning program, mothers from
dierent SES backgrounds may respond dierently in relation to overlapping their
pregnancies with Ramadan. Among the SES characteristics, mother education is
considered to be an important determinant of child health outcome (Schultz, 2002).
The possibility whether more educated mothers in the treatment areas are less likely
to overlap their pregnancies with Ramadan is considered in the next subsection.
Selection in Ramadan Exposure
Recall the child height were studied for children who are less than 10 years old at the
time of survey or possibly birth cohorts born between 1986 and 1995, and the family
planning program was initiated in October, 1977 much before these children were
born. Before examining parental selection in Ramadan exposure for the same birth
cohorts whose heights were studied in Table 2.8, it is important to examine parental
selection in Ramadan exposure before the placement of the family planning program.
It could be possible that Muslim mothers, living in the treatment areas, selectively
overlap their pregnancies to Ramadan even before the start of the family planning
program. If that happens and we also nd parental selection for the birth cohorts who
were born after the program, we might incorrectly attribute the observed selection in
Ramadan exposure to the family planning program. In Table 2.9, the likelihood of
parental selection in Ramadan exposure before the program placement is examined.
It shows that there is no dierence between the treatment and control areas in terms
of Ramadan exposure and parental SES measured by mother years of education,
41
father years of education, and mother age.
Table 2.10 shows association between exposure to Ramadan and mother years of
education in the treatment and control areas for birth cohort born between 1982 and
1995. The oldest individual in this sample is 14 years old and it also encompasses the
birth cohort born between 1986 and 1995. Joshi and Schultz (2013) also study these
births cohorts to analyze the impact of the family planning program on the child
health and education outcomes. The results in column (1) show that in the control
areas there is no dierence in mother years of education between children exposed
and not exposed to Ramadan. On the other hand, the results in column (2) show that
in the treatment areas more educated Muslim mothers are less likely to expose their
children to Ramadan at the time of pregnancy. In contrast to the results in column
(2), the results in column (3) shows that in the treatment areas more educated Hindu
mothers are more likely to overlap pregnancies with Ramadan, but the association is
not statistically signicant.
Table 2.11 documents the consequence of ignoring placement information of a pro-
gram, for instance a family planning program, that can induce selection in Ramadan
exposure. The rst two columns are based on the birth cohorts born between 1982
and 1995 and last two columns are based on the birth cohorts born between 1986 and
1995. Only Muslim births are considered for this analysis. In column (1), treated is
not interacted with mothereduy, and it shows that more years of mother education
has zero or no relation with Ramadan exposure. In contrast, when treated is inter-
acted with mothereduy, the coecient of the interaction term is positive and larger
than that of mothereduy and the joint F test of mothereduy is statically signicant
(p-value= 0.02). For birth cohorts born between 1986 and 1995, the results in column
(3) and (4) re
ect ndings in column (1) and (2) respectively. Given that mothers
in Matlab, on the average has very low level of education, it could be possible that
42
the relations observed in column (3) and column (4) are driven by outliers. To check
that it is not the case, a similar analysis is performed replacing mothereduy with
motheredu wheremotheredu takes a value of 1 if mother has atleast 1 year of formal
education and 0 otherwise. The results, presented in Table A2, are very similar to
that of Table 2.11. Therefore, the conclusions remain the same.
One concern in empirical frameworks presented in Table 2.10 and Table 2.11 is
that the family planning program may aect the female education. Thus, it could be
possible that the family planning is aecting the selection in Ramadan exposure and
mother years of education is an intermediary variable. To address this concern, an
analysis similar to that in Table 2.10 and Table 2.11 is performed restricting the birth
cohorts born between 1974 and 1982. This gives about four and half years of birth
sample before and after the initiation of the family planning program. The results
are presented in Table 2.12. Column (2) shows, no change in selection in Ramadan
exposure post initiation of the family planning program in the control areas which
is as expected, since the control areas did not receive the family planning program.
However, column (3) shows that in the treatment areas more educated mothers are
less likely to expose their children to Ramadan following the family planning program
even after controlling for month xed eects which captures seasonality. However,
there could still be residual seasonality which may not have captured by the month
xed eects. It could be still possible that parents are timing birth against the
any other factor instead of Ramadan.
24
If that is the case, similar selection pattern
would be observed for more educated Hindu mothers living in the treatment areas.
Column (4) in Table 2.12 shows no change in birth pattern for more educated Hindu
mothers. This result validates the ndings that more educated Muslim mothers are
24
For example, parents may face both high temperature and Ramadan at the same time but
parents may want to time conceptions based on temperature rather than Ramadan.
43
less likely to overlap pregnancies overlapping with Ramadan, if they are provided with
a family planning program. With Table 2.12, the consequence of ignoring timing and
placement of family planning program can be analyzed. In column (1), the birth
data from the treatment and control areas are pooled and it shows that there is
no selection. In contrast, column (3) of Table 2.12 shows substantial selection in
Ramadan exposure.
To explore whether or not selection in Ramadan exposure is limited only to Mat-
lab, Bangladesh, data from Indonesia|the IFLS is also used. As discussed in the
data section, previous studies Ewijk (2011); Majid (2015); Ewijk et al. (2013), which
also used the IFLS, did not examine the selection in Ramadan exposure using the
data from pregnancy histories of the reproductive age women. The results are pre-
sented in Table 2.14 and Table 2.15. In column (1) of Table 2.14, the results are
presented for all the births which took place in the last ve years of the survey wave.
The results show that more educated Muslim mothers are less likely to overlap their
pregnancies to Ramadan. However, for non-Muslim mothers, the years of education
has almost zero correlation with being exposed to Ramadan. In column (2), as a ro-
bustness check, the sample is restricted to mothers whose highest education is up to
12 years,
25
and the results do not change. In Table 2.15, we dene mother education
as dummy variable where it takes a value of 1 if educated and 0 otherwise. Again,
results essentially remain the same.
2.8 Discussion
In the previous section three main results are presented. First, it is shown that
the provision of a family planning program leads to a decrease in births eight to nine
25
Some mothers in IFLS who report that they went to college or beyond college but has missing
years of education.
44
months after Ramadan but an increase in births four to seven months after Ramadan.
Given that the length of human gestation is about 9 months, it implies that Muslim
mothers are avoiding conceptions when Ramadan is one month away and in the
month of Ramadan. On the other hand, they are increasing conceptions when it is
two to ve months away. Secondly, it is shown that signicant negative association
between Ramadan exposure and child height only exists for Muslim children in the
treatment areas but not for Muslim and Hindu children living in the control and
treatment areas respectively. Thirdly, it is also shown that in the treatment areas
more educated Muslim mothers are less likely to have pregnancies overlapped with
Ramadan. The second and third results together implies that perhaps the observed
negative relation between Ramadan exposure and the height of the Muslim children
in the treatment areas is probably due to parental selection in Ramadan exposure.
Moreover, it is shown that the lower likelihood between mother years education and
Ramadan exposure is not limited to Matlab, Bangladesh but also is pertinent to
Indonesia.
It is puzzling to observe that mothers are avoiding conceptions when Ramadan is
imminent or taking place and shifting conceptions a few months ahead. If mothers
are at all concerned about avoiding pregnancies with Ramadan, they should choose
a time period which do not overlap with Ramadan. Then why do we also nd that
mothers in the treatment areas are more likely to give birth 4 to 7 months after
Ramadan.
I point out two hypotheses for this behavioral pattern but can not test them
because of the data limitation. Firstly, it could be possible that women, who are
pregnant for longer periods of time, are less likely to be asked to fast during Ramadan.
Thus, if she relaxes the use of contraceptives few months ahead of Ramadan and
possibly get pregnant, she may may seek exemption from fasting. On other hand, if
45
she is pregnant for one month, she might be more likely to be asked to fast during
Ramadan. In the USA, it has been found most pregnancies are not acknowledged
until after the rst month of gestation (Floyd et al., 1999). Unfortunately, we do
not have any information on the fasting behavior of the Muslim women residing in
Matlab at the time of their pregnancies to test this hypothesis. Secondly, it could be
possible that household expenditure for Muslims may increase during the period of
Ramadan to celebrate both Ramadan and Eid al-Fitr. After the end of Ramadan, on
the rst day of the next Islamic month Shawwal, Muslims celebrate Eid al-Fitr which
is one of the major festivals in the Muslim world. It could be possible that Muslim
parents are avoiding conceptions when Ramadan is imminent to avoid pregnancies
at the time when the household face increase in expenditure due to Ramadan and
Eid al-Fitr, and this would be consistent with the ndings in the earlier literature.
For example, Alam and P ortner (2013) nd an increase in the use of contraceptives
at the time of crop loss for agricultural households in Tanzania. Similar to their
ndings, Eckstein et al. (1984) nd that favorable price, weather and wage shock lead
to increase in fertility. Galloway (1987), using historic data from France, shows that
increase in price of wheat lead to a decline in fertility of the urban poor. Therefore,
increase in household expenditure can potentially explain avoidance of conceptions
when Ramadan is imminent. However, it does not properly explain why they are
shifting the conceptions a few months ahead. One possibility is that by the time the
households recover from the phase of elevated expenditure due to Ramadan, Eid al-
Fitr, and Eid al-Adha, if the women gets pregnant she can not avoid her pregnancies
overlapping with the Ramadan in the following year. Recall if a woman conceives in
a narrow period of about 74 days after Ramadan is over, she may completely avoid
pregnancies overlapping with Ramadan.
26
This would explain why they are avoiding
26
Eid al-Adha and Eid al-Fitr are two biggest religious festivals in the Muslim world. Eid al-Adha
46
conceptions when Ramadan is imminent and why they are shifting conceptions a
few months ahead. This hypothesis can not be tested as well because of the data
limitation.
In the birth timing results, it was also found that the results are robust to inclu-
sion of biological mother xed eects. This result has a methodological implication.
Ewijk (2011) and, following Ewijk (2011), Majid (2015) motivate to solve the prob-
lem of selection in Ramadan exposure by controlling biological mother xed eects.
Controlling for biological mother xed eects this chapter nds that time varying
changes in the community aect timing of births. Griliches (1979) in his seminal
paper discusses identication limitation of using biological mother xed eects. He
argues that parents may intervene based on discrepancy in endowments among sib-
lings. Indeed, using data from Colombia, Rosenzweig and Wolpin (1988) nd that
intra-family health heterogeneities exist and parents also responds to those hetero-
geneities. They further show that application of family xed eects are ineective
in addressing such heterogeneities, and such strategy may fail to produce consistent
estimates in the presence of heterogeneities.
The second and third results have implications for empirical methodologies em-
ployed in the existing studies which examine the impact of exposure to Ramadan on
various outcomes. As mentioned earlier, existing studies rely on the assumption that
parents do not time birth relative to Ramadan. Although these studies recognize
that selection in Ramadan exposure can bias the estimates of interests, they do not
examine selection allowing for time varying changes. If the empirical strategies of
the earlier studies were followed and placement information of the family planning
program were ignored, we would have incorrectly concluded there is no selection in
Ramadan exposure (see Table 2.11 column (1) and (3)). Moreover, based on the
takes place about 70 days after Eid al-Fitr.
47
results shown in column (1) in Table 2.8, we would have falsely concluded that the
negative association between Ramadan exposure and the child height as causal. How-
ever, column (2) and (4) of Table 2.11 and column(2) of Table 2.10 show substantial
selection in Ramadan exposure when placement information of the family planning
program were taken in to account. Mother education is regarded as an important de-
terminant of child health and also in Matlab, a positive association between mother
years of education and child height is found (Table 2.13). As a result, it is not sur-
prising to nd that the dierence in height between exposed and non-exposed Muslim
children in the treatment areas are larger than that of in the control areas, once it is
known that in the treatment areas more educated mothers are less likely to expose
their children to Ramadan.
It is important to point out that results on the child height should be interpreted
carefully. In the control areas, selection in Ramadan exposure is not found and the
exposure to Ramadan is not also associated with the child height. However, this
should not be interpreted as|Ramadan exposure has no eect on child height or
health outcomes. The reason is that the coecient of the variable exposed is an ITT
(Intent to Treat) estimate which takes average impact of those who fast and who do
not fast. It could be still possible that there is an eect on child health for those
who fast. Moreover, health has a complex and multidimensional nature (Strauss and
Thomas, 2007). It could be possible that fasting does not aect child height but
aects other relevant health outcomes.
Recall that one limitation in this data set is that only birth dates of the children
born in Matlab are known but their gestation periods are unknown. Therefore, it is
important to evaluate whether premature births of babies can potentially confound
the estimates. For premature and post-term births to explain these results, two
things should also take place. Let's say that the observed decrease in births after
48
8 to 9 months after Ramadan was due to premature births and the conceptions of
those births actually took place after Ramadan was over. To premature births to
explain this pattern, mothers in the treatment areas should give birth prematurely
within 7 months after Ramadan after getting the family planning program. Secondly,
only the Muslim births in the treatment areas should be aected. Chances of both
taking place are very rare. If anything we should observe more premature births
in the control areas than in the treatment areas because the mothers living in the
treatment areas some maternal health care along with the family planning program
that was not available to mothers in the control areas. Therefore, the treatment areas
should have less premature births relative to the control areas and the results should
have, if anything, downward bias. Moreover, if any health shocks aecting mothers
and leading to premature births in the treatment areas, it should aect both Muslims
and Hindus living in the treatment areas. The fact that I do not observe similar
birth pattern of Hindus relative to Ramadan attests to the results that health shocks
leading to premature births are not responsible for this result.
Another concern is that male fragility may explain the decrease in births 8 to
9 months after Ramadan in the treatment areas after the initiation of the family
planning program. Almond and Mazumder (2011) nds exposure to Ramadan in rst
two months of gestation leads to lower male births. In this data, I do not observe such
pattern. The results are presented in appendix Table A3 birth relative to Ramadan
and male birth for the control areas only because I do not observe selection in birth
timing in control areas. Moreover, male fragility also less likely to be an explanation
because the treatment areas received other maternal health interventions which might
increase the chances of male birth. Thus, male fragility can be ruled out as a likely
explanation.
49
2.9 Conclusion
There is little doubt about the welfare implications of the impact of adverse environ-
ments and nutrition shocks in utero on health outcomes. We have seen proliferation
of studies which have documented the consequences of in utero shocks. In the con-
text of nutrition shock due to maternal fasting during Ramadan, prior studies have
treated Ramadan exposure as a natural experiment and attributed the negative eects
of exposure to Ramadan on health and other outcomes to maternal fasting during
Ramadan. Although the prior studies have highlighted the importance of identifying
exposure eect of Ramadan allowing for parental selection in Ramadan exposure, they
do not nd any evidence of such selection. The examination of parental selection in
Ramadan exposure in prior studies is inadequate, as selection in Ramadan exposure
is not examined allowing for time varying changes.
Using two data sets from two countries, this chapter shows that mothers with
more education are less likely to overlap their pregnancies with Ramadan. As mother
education is one of the most important determinants of child health, such selection
in Ramadan exposure can lead to unequal outcomes between exposed children and
non-exposed children. Therefore, comparing the mean outcomes of exposed children
with that of non-exposed children, we may falsely conclude that the eects are due to
maternal fasting during Ramadan. This chapter emphasizes that parental selection
in Ramadan exposure should be examined allowing for time varying changes, as Ra-
madan is a recurring and a predictable event. It also documents that ignoring time
varying changes may lead to falsely concluding that there is no evidence of parental
selection. It further shows presence of parental selection may produce biased esti-
mates of the eects of Ramadan exposure. In the light of these ndings, this chapter
recommends that it is important to examine the nature of selection in Ramadan ex-
50
posure appropriately, and any evaluation of the impact of maternal fasting during
Ramadan on relevant outcomes should be performed allowing for selection.
Moreover, this chapter also documents a benecial aspect of provision of a family
planning program. Although the earlier studies have documented negative conse-
quences of Ramadan exposure on health, labor, and education outcomes, there was
no evidence what kind of policy may help mothers to avoid their pregnancies overlap-
ping with Ramadan. The evidence from Matlab suggests that provision of a family
planning program may allow mothers to time birth relative to Ramadan.
51
2.10 Tables
Table 2.1: MHSS Summary Statistics
Panel A : Parental Characteristics
Variables Mean
Mother years of education 2.17
Father years of education 3.28
Mother age 37.40
Panel B: Birth Characteristics
nonexposed 0.13
Born during Ramadan 0.08
Male 0.52
Hindu 0.08
Notes: This table describes summary characteristics from the MHSS for birth co-
horts born between 1974 and 1995. In panel A summary characteristics of parents
are presented. Motherage is measured at the time of survey. In Panel B Birth char-
acteristics are presented. nonexposed takes a value of 0 if pregnancy overlapped rst,
second and third trimester and 1 otherwise. The trimesters are calculated from the
birth date based on the assumption that pregnancy lasts for 9 months. Born during
Ramadan takes a value of 1 if the individual was born during the month of Ramadan
0 otherwise. The calculation of exposure to Ramadan is based under the assumption
that pregnancy lasts for 9 months.
52
Table 2.2: IFLS Summary Statistics
Panel A : Parental Characteristics
Variables Mean
Mother years of education 7.45
Mother age 28.89
Panel B: Birth Characteristics
nonexposed 0.14
Born during Ramadan 0.08
Male 0.51
Non-Muslim 0.11
Notes: This table describes summary characteristics from the IFLS restricting preg-
nancy history to maximum 5 years. In panel A summary characteristics of parents
are presented. Motherage is measured at the time of survey. In Panel B Birth char-
acteristics are presented. nonexposed takes a value of 0 if pregnancy overlapped rst,
second and third trimester and 1 otherwise. The trimesters are calculated from the
birth date based on the assumption that pregnancy lasts for 9 months. Born during
Ramadan takes a value of 1 if the individual was born during the month of Ramadan
0 otherwise. The calculation of exposure to Ramadan is based under the assumption
that pregnancy lasts for 9 months.
Table 2.3: Selection in live birth
Panel A
VARIABLES Control Treatment Di p-value
Live birth .94 0.93 0.01 0.39
Panel B
Women Age 23.55 27.90 3.35 0.11
Women Yedu 3.89 3.13 0.77 0.41
Husband Yedu 3.06 2.67 0.39 0.67
Notes: This table describes whether there is selection in live birth because of the fam-
ily planning program between the treatment and control areas. The rst two columns
show means in the treatment and control areas. The variable Yedu is education in
years. age is age at survey. The means are calculated using the survey weights. Panel
A compares the probability of at least one birth between the treatment and control
areas. Panel B compares observable characteristics of women who did not give birth
yet between the treatment and control areas . This table is based on data from the
MHSS.
53
Table 2.4: Selection in live birth : Muslim Married Women
Panel A
VARIABLES Control Treatment Di p-value
Live birth 0.95 0.93 0.02 0.19
Panel B
Women Age 24.09 26.99 1.90 0.33
Women Yedu 3.76 3.22 0.49 0.46
Husband Yedu 3.06 2.67 0.39 0.67
Notes: This table describes whether there is selection in live birth because of the
family planning program between the treatment and control areas. The rst two
columns show means in the treatment and control areas. The variable Yedu is ed-
ucation in years. age is age at survey. The means are calculated using the survey
weights. Panel A compares the probability of at least one birth between treatment
and control areas. Panel B compares observable characteristics of women who did
not give birth yet between the treatment and control areas . This table is based on
data from the MHSS.
54
Table 2.5: Family Planning Program and Birth Timing by Ramadan Month (Muslims): Birth Cohort 1974-95
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
post treated 0.016 0.009 0.002 -0.010 0.024 0.042* -0.007 0.024 -0.033* -0.035 -0.008 -0.025
(0.025) (0.021) (0.016) (0.023) (0.022) (0.023) (0.019) (0.017) (0.017) (0.021) (0.022) (0.017)
post 0.260*** 0.171*** 0.178*** 0.158*** 0.072 -0.282*** -0.192*** -0.226*** -0.119*** -0.066* 0.048 -0.005
(0.041) (0.031) (0.038) (0.046) (0.052) (0.047) (0.030) (0.038) (0.036) (0.037) (0.029) (0.030)
R2 0.103 0.092 0.079 0.087 0.062 0.084 0.061 0.083 0.068 0.073 0.084 0.085
Observations 5271 5271 5271 5271 5271 5271 5271 5271 5271 5271 5271 5271
Mean Y 0.086 0.068 0.076 0.088 0.104 0.104 0.093 0.096 0.079 0.075 0.076 0.057
Month FE Y Y Y Y Y Y Y Y Y Y Y Y
Year FE Y Y Y Y Y Y Y Y Y Y Y Y
Village FE Y Y Y Y Y Y Y Y Y Y Y Y
Standard errors are clustered at the village ID level.
*** p<0.01, ** p<0.05, * p<0.1
Notes: The dependent variables are based on birth months from Ramadan. ramadan 0 is the month of Ramadan. ramadan 1 is rst 30 days after the
month of Ramadan, ramadan 2 is the second 30 days after the month of Ramadan and so on. For x=0,1,11 ramadanx takes a value of 1 if the individual
was born x months after Ramadan and 0 otherwise. Post takes a value of 1 from August,1978 and 0 otherwise. ramadan 11 is the last 24 days based
on calender year starting from the month of Ramadan. Post takes a value of 1 if the birth took place from August, 1978 and 0 otherwise. The variable
treated takes a value of 1 if mother lives in the treatment areas and 0 otherwise. The analysis in this table are based on pregnancy histories of the Muslim
mothers only. This table is based on data from the MHSS.
55
Table 2.6: Family Planning Program and Birth Timing Relative to
Ramadan: Birth Cohort 1974-95
Muslim Birth Sample Hindu Birth Sample
VARIABLES ramadan8to9 ramadan8to9 ramadan8to9 ramadan8to9 ramadan8to9 ramadan8to9
(1) (2) (3) (4) (5) (6)
post treated -0.062** -0.057** -0.066* 0.004 0.005 0.067
(0.028) (0.025) (0.037) (0.095) (0.101) (0.208)
post -0.183*** -0.169*** -0.159* 0.020 0.012 0.053
(0.046) (0.048) (0.082) (0.168) (0.167) (0.305)
R2 0.111 0.112 0.427 0.306 0.310 0.534
Observations 5166 5166 5166 466 466 466
Mean Y 0.154 0.154 0.154 0.182 0.182 0.182
Month FE Y Y Y Y Y Y
Year FE Y Y Y Y Y Y
Village FE Y Y Y Y Y Y
Village TVC N Y Y N Y Y
Mother FE N N Y N N Y
Standard errors are clustered at the village ID level.
*** p<0.01, ** p<0.05, * p<0.1
Notes: The dependent variable ramadan8to9 takes a value of 1 if the individual was born 8 to 9 months after
Ramadan and 0 otherwise. Post takes a value of 1 from August, 1978 and 0 otherwise. The variable treated takes a
value of 1 if the village received the family planning program and 0 otherwise. All regressions include spline function
of mother age at 5 year interval from age 15 onward and mother years of education. Village T.V.C are village time
varying characteristics are large market, electricity, tube-well and satellite clinic. Post is interacted with whether
the village has large market. For electricity, tube-well and satellite clinic post takes a of value 1 from the year after
these were established and 0 otherwise. This table is based on data from the MHSS.
Table 2.7: Family Planning Program and Birth Timing Relative to
Ramadan: Birth Cohort 1974-95
Muslim Birth Sample Hindu Birth Sample
VARIABLES ramadan4to7 ramadan4to7 ramadan4to7 ramadan4to7 ramadan4to7 ramadan4to7
(1) (2) (3) (4) (5) (6)
post treated 0.086*** 0.080** 0.082* -0.017 0.026 -0.048
(0.032) (0.033) (0.049) (0.218) (0.212) (0.333)
post -0.627*** -0.623*** -0.611*** -0.681* -0.755** -0.888*
(0.063) (0.062) (0.094) (0.352) (0.366) (0.529)
R2 0.191 0.191 0.497 0.342 0.356 0.550
Observations 5166 5166 5166 466 466 466
Mean Y 0.396 0.396 0.396 0.337 0.337 0.337
Month FE Y Y Y Y Y Y
Year FE Y Y Y Y Y Y
Village FE Y Y Y Y Y Y
Village TVC N Y Y N Y Y
Mother FE N N Y N N Y
Standard errors are clustered at the village ID level.
*** p<0.01, ** p<0.05, * p<0.1
Notes: The dependent variable ramadan4to7 takes a value of 1 if the individual was born 4 to 7 months after
Ramadan and 0 otherwise. Post takes a value of 1 from August, 1978 and 0 otherwise. The variable treated takes
a value of 1, if the village received the family planning program and 0 otherwise. All regressions include spline
function of mother age at 5 year interval from age 15 onward and mother years of education. Village T.V.C are
village level time varying characteristics are large market, electricity, tube-well and satellite clinic. Post is interacted
with whether the village has large market. For electricity, tube-well and satellite clinic post takes a value of 1 from
the year after these were established and 0 otherwise. This table is based on data from the MHSS.
56
Table 2.8: Exposure to Ramadan and Child Height
Muslim Children Only Muslim+Hindu Children
VARIABLES height(C+T) height(C) height(T) height(T)
(1) (2) (3) (4)
BornduringRamadan -1.219 -0.011 0.165 -0.070
(1.496) (2.578) (2.789) (2.296)
exposed -2.782* -1.384 -4.428 -4.516*
(1.490) (1.927) (2.813) (2.294)
Hindu -4.627*
(2.329)
BornduringRamadan Hindu 5.388
(8.008)
exposedHindu 6.797**
(3.106)
R2 0.876 0.889 0.938 0.924
Observations 1174 655 519 611
Mean Y 100.5 100.8 100.1 100.7
Month FE Y Y Y Y
Malec Age in Month FE Y Y Y Y
Village FE Y Y Y Y
F:exposed+exposed Hindu 0.6
P-Value 0.5
Standard errors are clustered at the village ID level.
*** p<0.01, ** p<0.05, * p<0.1
Notes: The height sample is based on children who are less than 120 months old at the time of survey. The
dependent variable height is measured in centimeters. Exposed takes a value of 1 if pregnancy overlapped rst,
second and third trimester and 0 otherwise. The trimesters are calculated from the birth date based on the
assumption that pregnancy lasts for 9 months. BornduringRamadan takes a value of 1 if the individual was born
during Ramadan and 0 otherwise. The omitted category is the pregnancies which did not overlap with Ramadan.
All regression specications include mother age, mother years of education. C and T denotes control are treatment
areas respectively. C+T imply the samples from the control and treatment areas are pooled. This table is based on
data from the MHSS.
Table 2.9: Parental SES and Ramadan Exposure for cohort 1974-July,1978
VARIABLES Control Dierence p-value
motherage -0.002 0.002 0.38
mothereduy 0.001 -.004 0.21
fathereduy 0.001 -.002 0.39
Notes: This table presents coecients of regressions nonexposed on motherage,
fathereduy and mothereduy controlling for month and year xed eects. The col-
umn named Dierence shows dierence in coecient between treatment and control
areas for pre-program birth cohorts. The third column shows p-value of of the dier-
ence. The dependent variable nonexposed takes a value of 0 if pregnancy overlapped
rst, second and third trimester and 1 otherwise. The variable mothereduy and
fathereduy is mother and father education in years respectively. motheage is age
at survey. Post takes a value of 1 from August,1978 and 0 otherwise. This table is
based on data from the MHSS.
57
Table 2.10: Exposure to Ramadan and Mother Education: Birth Cohort
1982-1995
Muslim Birth Sample Hindu Birth Sample
nonexposed(C) nonexposed(T) nonexposed(T)
(1) (2) (3)
motherage 0.001 -0.001 0.002
(0.001) (0.001) (0.005)
mothereduy -0.004 0.006* -0.012
(0.003) (0.003) (0.011)
R2 0.240 0.272 0.668
Observations 1990 1432 215
Mean Y 0.146 0.138 0.153
Month FE Y Y Y
Block Year Y Y Y
Village FE Y Y Y
Standard errors are clustered at the village ID level.
*** p<0.01, ** p<0.05, * p<0.1
Notes: The dependent variablenonexposed takes a value of 0 if pregnancy overlapped rst, second and third trimester
and 1 otherwise. The trimesters are calculated from the birth date based on the assumption that pregnancy lasts
for 9 months. All regression specications include mother age and mother years of education. mother years of
education. C and T denote the control areas and the treatment areas respectively. Column 1 and 2 are Muslims
from the control areas and the treatment areas only. Column 3 is Hindus from the treatment areas. This table is
based on data from the MHSS.
58
Table 2.11: Birth Relative to Ramadan and Mother Education
Muslim Birth Cohort 1982-1995 Muslim Birth Cohort 1986-1995
nonexposed nonexposed nonexposed nonexposed
(1) (2) (3) (4)
motherage 0.000 0.000 0.000 0.000
(0.001) (0.001) (0.001) (0.001)
mothereduy 0.000 -0.005* -0.002 -0.008**
(0.002) (0.003) (0.003) (0.004)
treated mothereduy 0.011*** 0.014***
(0.004) (0.005)
R2 0.248 0.250 0.359 0.362
Observations 3422 3422 2356 2356
Mean Y 0.143 0.143 0.160 0.160
Month FE Y Y Y Y
Block Year FE Y Y Y Y
Village FE Y Y Y Y
F:mothereduy+treatedmothereduy 5.36 2.98
P-Value 0.02 0.09
Standard errors are clustered at the village ID level.
*** p<0.01, ** p<0.05, * p<0.1
Notes: The dependent variablenonexposed takes a value of 0 if pregnancy overlapped rst, second and third trimester
and 1 otherwise. The trimesters are calculated from the birth date based on the assumption that pregnancy lasts
for 9 months. The treated takes a value of 1 if the villages received the family planning program and 0 otherwise.
All regression specications include mother age and mother years of education. The sample includes Muslim births
both from treatment and control areas. This table is based on data from the MHSS.
59
Table 2.12: Exposure to Ramadan and Mother Education: Birth Cohort
1974-1982
Muslim Birth Sample Hindu Birth Sample
VARIABLES nonexposed(C+T) nonexposed(C) nonexposed(T) nonexposed(T)
(1) (2) (3) (4)
motherage 0.000 -0.001 0.002 -0.006
(0.001) (0.001) (0.001) (0.006)
mothereduy 0.003 0.003 -0.003 -0.018*
(0.002) (0.005) (0.003) (0.010)
post -0.045 -0.139** -0.143
(0.049) (0.054) (0.104)
postmothereduy -0.002 0.013** 0.014
(0.007) (0.005) (0.017)
R2 0.402 0.421 0.406 0.684
Observations 2012 1080 932 146
Mean Y 0.117 0.111 0.123 0.185
Month FE Y Y Y Y
Year FE Y Y Y Y
Village FE Y Y Y Y
F:mothereduy+postmothereduy 0.034 6.258 0.065
P-Value 0.855 0.015 0.801
Standard errors are clustered at the village ID level.
*** p<0.01, ** p<0.05, * p<0.1
Notes: The dependent variablenonexposed takes a value of 0 if pregnancy overlapped rst, second and third trimester
and 1 otherwise. The trimesters are calculated from the birth date based on the assumption that pregnancy lasts
for 9 months. The treated takes a value of 1 if the villages received the family planning program and 0 otherwise.
Variables motherage is mother age at survey and mothereduy is mother years of education at survey. Post take a
value of from August, 1978 and 0 otherwise. C and T denote the control areas and the treatment areas respectively.
C+T imply the samples from the treatment and control ares pooled. Column 2 and 3 are based on Muslim sample
from the control areas and treatment areas respectively. Column 4 is based on Hindu sample from the treatment
areas. This table is based on data from the MHSS.
Table 2.13: Mother Education and Child Height
All Control Areas Treatment Areas
(1) (2) (3)
motherage -0.009 0.003 0.014
(0.047) (0.075) (0.070)
mothereduy 0.378*** 0.295* 0.495***
(0.099) (0.158) (0.132)
R2 0.865 0.883 0.921
Observations 1299 688 611
Mean Y 100.7 100.8 100.7
Month FE Y Y Y
Malec Age in Month FE Y Y Y
Village FE Y Y Y
Standard errors are clustered at the village ID level.
*** p<0.01, ** p<0.05, * p<0.1
Notes: The height sample is based on children who are less than 120 months old at the time of survey. The
dependent variable height is measured in centimeters. The sample includes all the children from the treatment and
control areas. All imply sample from both the treatment and control areas are pooled. This table is based on data
from the MHSS.
60
Table 2.14: Exposure to Ramadan and Mother Education
(All Education) (Less than Equal to 12 Years)
VARIABLES nonexposed nonexposed
motherage 0.00125** 0.00134**
(0.000571) (0.000603)
mothereduy 0.00221* 0.00303*
(0.00130) (0.00163)
nonm 0.0412 0.0377
(0.0313) (0.0334)
nonmmothereduy -0.00384 -0.00259
(0.00287) (0.00372)
Constant 0.197*** 0.200***
(0.0324) (0.0340)
Observations 10,753 9,853
R-squared 0.142 0.141
Birthday FE Y Y
Month FE Y Y
Year FE Y Y
commid93 Y Y
F: mothereduy+nonmmothereduy 0.40 0.02
P >F 0.54 0.90
Standard errors clustered at community ID level.
*** p<0.01, ** p<0.05, * p<0.1
Notes: The dependent variablenonexposed takes a value of 0 if pregnancy overlapped rst, second and third trimester
and 1 otherwise. The trimesters are calculated from the birth date based on the assumption that pregnancy lasts
for 9 months. Variablesmotherage is mother age at survey andmothereduy is mother years of education at survey.
The nonm (non-Muslim) takes a value of 1 if the mother is non-Muslim and 0 otherwise. Commid93 represents
original IFLS communities. mothereduy is dened as years of mother education. The results in the table are based
on all four waves of the IFLS restricting pregnancy history to maximum 5 years from the survey year.
61
Table 2.15: Exposure to Ramadan and Mother Education
(Birth cohort-last ve year from survey) (Birth cohort-1989-2007)
VARIABLES nonexposed nonexposed
motherage 0.00130** 0.00137***
(0.000578) (0.000504)
motheredu 0.0423** 0.0323*
(0.0191) (0.0166)
nonm 0.0340 0.0456
(0.0385) (0.0401)
nonmmotheredu -0.0269 -0.0220
(0.0380) (0.0416)
Constant 0.170*** 0.112***
(0.0369) (0.0331)
Observations 10,753 13,351
R-squared 0.142 0.144
Birthday FE Y Y
Month FE Y Y
Year FE Y Y
commid93 Y Y
F: mothereduy+nonmmotheredu 0.21 0.07
P >F 0.64 0.79
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Notes: The dependent variablenonexposed takes a value of 0 if pregnancy overlapped rst, second and third trimester
and 1 otherwise. The trimesters are calculated from the birth date based on the assumption that pregnancy lasts
for 9 months. Variables motherage is mother age at survey and motheredu takes a value of 1, if mother had atleast
1 year of education and 0 otherwise. The nonm (non-Muslim) takes a value of 1 if the mother is non-Muslim and
0 otherwise. Commid93 represents original IFLS communities. motheredu takes a value of 1 if mother is educated
and 0 otherwise. The results in the table are based on all four waves of the IFLS.
62
2.11 Figure
Figure 2.1: Matlab Area
63
Table A1: Family Planning Program and Birth Timing Relative to Ramadan:
Birth Cohort 1974-95
ramadan8to9 ramadan4to7
(1) (2) (3) (4) (5) (6)
posttreated -0.062** -0.057** -0.066* 0.085*** 0.082** 0.083*
(0.028) (0.025) (0.037) (0.032) (0.032) (0.049)
posttreatedHindu 0.076 0.070 0.161 -0.101 -0.100 -0.178
(0.092) (0.096) (0.184) (0.230) (0.228) (0.323)
post -0.180*** -0.165*** -0.158* -0.620*** -0.615*** -0.600***
(0.046) (0.048) (0.083) (0.063) (0.063) (0.093)
postHindu 0.119 0.105 0.133 -0.094 -0.100 -0.277
(0.136) (0.136) (0.226) (0.320) (0.317) (0.435)
R2 0.126 0.128 0.435 0.199 0.199 0.499
Observations 5632 5632 5632 5632 5632 5632
Mean Y 0.156 0.156 0.156 0.391 0.391 0.391
Month FE Y Y Y Y Y Y
Hindu Year FE Y Y Y Y Y Y
Hindu Village FE Y Y Y Y Y Y
Village TVC N N N N N N
Mother FE N N N N N N
F: 0.026 0.020 0.270 0.005 0.006 0.086
P-Value 0.873 0.889 0.604 0.944 0.939 0.770
Standard errors are clustered at the village ID level.
*** p<0.01, ** p<0.05, * p<0.1
Notes: The dependent variable ramadan4to7 takes a value of 1 if the individual was born 4 to 7 months after
Ramadan and 0 otherwise. The dependent variable ramadan8to9 takes a value of 1 if the individual was born 8 to
9 months after Ramadan and 0 otherwise. Post takes a value of 1 from August, 1978 and 0 otherwise. The variable
treated takes a value of 1, if the villages received the family planning program and 0 otherwise. The variable
posttreated is the interaction of post and treated. All regressions include spline function of mother age at 5 year
interval from age 15 onward and mother years of education. Village T.V.C are village time varying characteristics
are large market, electricity, tube-well and satellite clinic. Post is interacted with whether the village has large
market. For electricity, tube-well and satellite clinic post takes a value 1 from the year after these were established
and 0 otherwise. This table is based on data from the MHSS.
64
Table A2: Ramadan Exposure and Mother Education
Muslim Birth Sample Hindu Birth Sample
Birth Cohort 1982-1995 Birth Cohort 1986-1995 Birth Cohort 1982-1995 Birth Cohort 1986-1995
(1) (2) (3) (4) (5) (6) (7) (8)
motherage 0.000 0.000 0.000 0.000 0.006 0.006 0.005 0.005
(0.001) (0.001) (0.001) (0.001) (0.005) (0.005) (0.004) (0.004)
motheredu 0.002 -0.017 -0.007 -0.028 -0.068 -0.081 -0.091 -0.050
(0.013) (0.017) (0.018) (0.025) (0.061) (0.164) (0.074) (0.107)
treated motheredu 0.045* 0.052 0.017 -0.052
(0.024) (0.032) (0.174) (0.132)
R2 0.248 0.249 0.359 0.360 0.633 0.633 0.775 0.776
Observations 3422 3422 2356 2356 317 317 223 223
Mean Y 0.143 0.143 0.160 0.160 0.142 0.142 0.152 0.152
Month FE Y Y Y Y Y Y Y Y
Block Year FE Y Y Y Y Y Y Y Y
Village FE Y Y Y Y Y Y Y Y
F:motheredu+treatedmotheredu 3.02 1.49 1.11 1.34
P-Value 0.08 0.22 0.30 0.25
Standard errors are clustered at the village ID level.
*** p<0.01, ** p<0.05, * p<0.1
Notes: The dependent variablenonexposed takes a value of 0 if pregnancy overlapped rst, second and third trimester
and 1 otherwise. The trimesters are calculated from the birth date based on the assumption that pregnancy lasts
for 9 months. The treated takes a value of 1 if the villages received the family planning program and 0 otherwise.
Variables motherage is mother age at survey and motheredu takes a value of 1 if mother has any formal education
and 0 otherwise. Post takes a value of 1 from August, 1978 and 0 otherwise. treated takes a value of 1 if the areas
received family planing program and 0 otherwise. Column 1 and 2 are Muslims from the control areas and the
treatment areas only. Column 3 is Hindus from the treatment areas. This table is based on data from the MHSS.
65
Table A3: Birth Relative to Ramadan and Male Fragility
(1) (2)
VARIABLES malec malec
(1) (2)
ramadan 0 0.038 0.036
(0.043) (0.042)
ramadan 1 0.005 -0.003
(0.054) (0.056)
ramadan 2 0.017 0.016
(0.049) (0.050)
ramadan 3 0.072 0.067
(0.050) (0.053)
ramadan 4 0.033 0.035
(0.044) (0.045)
ramadan 5 -0.015 -0.018
(0.043) (0.043)
ramadan 6 0.064* 0.068**
(0.034) (0.031)
ramadan 7 0.021 0.019
(0.044) (0.046)
ramadan 8 0.043 0.045
(0.053) (0.053)
ramadan 9 0.045 0.048
(0.039) (0.040)
R2 0.039 0.043
Observations 2981 2937
Mean Y 0.528 0.529
Month FE Y Y
Year FE Y Y
Village FE Y Y
F 0.921 0.818
P>F 0.341 0.369
Standard errors are clustered at the village ID level.
*** p<0.01, ** p<0.05, * p<0.1
Notes: The dependent variable malec takes a value of 1 if the individual born is a male. ramadan
0
is the month of
Ramadan. Each ramadanx x=1-9 takes 30 days from the time Ramadan ends. ramadanx takes a value of 1 if true
and 0 otherwise. Omitted category is the births which were not exposed to Ramadan. Column 1 does not include
any mother level observables. Column 2 include mother age dened as spline function with knots at 5 year interval
from 15 onward and mother years of education.
66
Chapter 3
Parental Health, Households, Communities and
Fetal Health in India
1
3.1 Introduction
The nutritional and survival status of children are considered to be important indi-
cators of standards of living, especially in less developed countries (LDCs). Unlike
the experience of the Western World, economic growth and falling costs of nutrition
or treatment are not accompanied by proportionate improvements in health indica-
tors for many developing countries, including India. India fares poorly in almost all
child health indicators. In 2006, eg, a quarter of child deaths in the world of under 5
years of age took place in India (Subramanian et al., 2009). Other indicators like the
prevalence of stunting, wasting and anemia are also among the highest in the world
(Jayachandran and Pande, 2015).
There is a large cross-disciplinary literature that argues child characteristics at
birth, commonly called birth endowments are extremely important in subsequent
health status, including survival. For example, nutrition studies nd negative rela-
tionships between birth weight and child mortality (Chen et al., 1980). Similarly,
medical studies show higher chances of later life cardiovascular disease and non-insulin
dependent diabetes for people had lower birth weight, thinness and short body length
1
Jointly written with Riddhi Bhowmick.
67
at birth (Barker, 2000). The reason is ascribed to the high sensitivity of in utero
neurological and physiological development to the environment. It is argued that
tissues and organs of the body go through some critical periods of development when
a stimulus or insult (like malnutrition) may have lasting or lifelong eects. Apart
from direct eects on health, there are other important eects of childhood endow-
ments on later life economic outcomes including education and income (Behrman and
Rosenzweig, 2004). There is also mounting evidence which suggest that the eects of
childhood endowments may well extend to next generation of children (Bhalotra and
Rawlings, 2011; Gluckman and Hanson, 2004).
There is another strand of literature that tries to link parental health to child
health at birth (fetal health, henceforth). For India, there are some medical studies
which look into this topic. For instance, Mohanty et al. (2006) link maternal anthro-
pometry to the incidence of low birth weight for 395 singleton pregnancies. Veena
et al. (2004) examine associations between parental health and child's birth weight,
birth length and head circumference. These studies are limited in scope, as they are
mostly based on non-representative local samples and also look into a limited set of
determinants (mostly parental health). On the other hand, there is a large literature
in economics which tries to nd the household and community level determinants
of child health outcomes, such as child height and survival in developing countries,
based on nationally representative sample (Strauss, 1990; Thomas et al., 1990; Lavy
et al., 1996). But very little is understood about the eects of these variables on child
health at birth.
2
In this chapter, we attempt to bridge the gap between these two
sets of literature.
2
A notable exception is Bhalotra and Rawlings (2011) who among other things, nd negative
associations between maternal health and incidence of low birth weight and neonatal mortality across
a number of LDCs.
68
One of the major reasons health as a measure of human capital is distinct and more
complex from measures like education is its multi-dimensional nature (Strauss and
Thomas, 1998). Dierent measures of health aect schooling, wage, later life health
and survival probabilities dierently. So considering multiple health-indicators is al-
ways preferred in any study, investigating human capital formation. In this spirit, we
consider two dimensions of fetal health namely fetal survival to the term and fetal
growth and nutrition. Fetal survival to the term is measured by the likelihood of the
rst child being male and mortality risk within a month of birth. The former measure
is motivated from the large body of evidence from epidemiology and medical studies
nding in utero male survival disadvantage during periods of maternal stress (Song,
2012; Eriksson et al., 2010; Hamoudi and Nobles, 2014). Fetal growth is measured
by birth weight and birth size. In this chapter, we use multiple waves of National
Family Health Survey (NFHS) data and restrict samples to adult mothers (age 18
years or more) with height information who have rst live birth in last 5 years of the
survey.
3
We use maternal (and sometimes also paternal) height as the only health
indicator for parents as they represent long term health status which is unchanged
during conception or pregnancy. Although important in its own right, any short term
health measure for parents (such as anemia status or Body Mass index) is jointly
determined with the fetal health outcomes. Unless these short term health measures
are randomized, one obtains biased estimate of relationship between parental health
and fetal health measures. This is a crucial point which many earlier studies on this
topic have largely ignored (see for example the studies by Gibson and Mace (2003),
Bhalotra and Rawlings (2011)).
3
Also known as Demographic Health Survey.
69
Based on the government report,
4
we divide all the states into high and low infant
mortality regions and run separate regressions for the rural and urban sub samples in
each of these regions. High infant mortality states include Madhya Pradesh (includes
Chhattisgarh), Orissa, Uttar Pradesh (includes Uttarakhand), Assam, Rajasthan,
Bihar (includes Jharkahnd), Haryana, Andhra Pradesh and Gujarat while all the
remaining states belong to low infant mortality regions. Interestingly, high infant
mortality regions include the states of Gujarat, Haryana and Andhra Pradesh which
have relatively high per capita incomes in the country.
We nd parents, especially mothers with at least secondary education are more
likely to have births with better fetal health in rural areas of states with high infant
mortality. The eect of maternal health, as measured by her height, is very strong
and signicant for all measures of fetal health in this region. A 10 centimeter increase
in height of a mother from this region is associated with more than 2% increase in
likelihood of having a male child at rst birth, approximately 2% less likelihood of a
child dying as a neonate, 160 grams increase in birth weight, 7% lower incidence of
low birth weight and 6% increase in likelihood of decent birth size. While parental
education has lower impact for low infant mortality states, maternal height remains
important for most measures of fetal health. Overall, the results are consistent with
the hypothesis that parental human capital can substitute for low provision of qual-
ity health infrastructure in areas suering from adverse disease environment. Income,
access to information and other community characteristics are also associated with
various dimensions of fetal health, although the relationships are not always strong or
consistent. We perform several checks to ensure that the association found for gender
4
Infant Mortality Rate India and States/UT during Eleventh Plan (2006) which is available
for download from the government of India's following website: https://data.gov.in/catalog/infant-
mortality-rate-india-and-statesuts-during-eleventh-plan.
70
of the rst child with maternal characteristics is not driven by sex-selective abortions.
The observed sensitivity of gender of the rst child to parental (and a few com-
munity) characteristics raises a basic methodological issue in the empirical literature
of developing countries in atleast, two important ways. Firstly, a number of papers in
gender-study and demography literature treat gender of the rst child to be random
so that any association between certain outcomes and sex of the rst child can be
interpreted as causal (Rosenblum, 2013; Bhalotra and Cochrane, 2010). Secondly, it
is common in empirical work to treat gender of an individual to be pre-determined
and orthogonal to many other background variables so that it is included either as an
independent control variable or is used to carry out heterogeneity analysis. Similar to
the argument made in Bhalotra and Clarke (2013) for twin births, we argue that these
studies fail to make the distinction between male conception and male birth. Likeli-
hood of a male birth depends both upon the likelihood of male conception and sex
specic survival rates of the fetus during gestation. In a country like India, parental
health status, food and medicine prices, disease environment vary considerably by
socio-economic characteristics like education, income and location. In the developed
countries, however, better availability of medical safety nets can partially compen-
sate for such low socio-economic hindrances.
5
Our results conrm the stronger role
of parental background (especially health) in determining the gender of the rst born
for areas with supposedly worse disease environments and poor quality public health
infrastructure. Thus o-spring sex as a natural experiment may be ill-suited in this
setting.
6
Inclusion of all maternal health and socioeconomic characteristics available
5
Bhalotra and Clarke (2013) express a similar concern in treating twin births as exogenous in
a developing country context. They, for example, nd positive association between twin births and
maternal BMI, height and family income for a series of developing countries.
6
In fact, a recent study by Hamoudi and Nobles (2014) report selection into live birth may
account for a large proportion of the observed association between ospring sex and divorce in the
71
in NFHS, is not sucient to guarantee the conditional exogeneity of sex of the rst
birth outcome. A host of other environmental factors and health practices of the
mother are potentially responsible for a male child at rst birth, exclusion of which
results into biased estimates. In our opinion, instrumenting sex of the ospring is a
better option than looking for these additional controls.
Our study makes some further contributions in the empirical literature on human
capital formation in both developed and developing countries. Firstly, we contribute
to the burgeoning literature on parental responses to birth endowments. In a recent
summary of the literature, Almond and Mazumder (2013) point out the importance
of measures of birth endowment which are good indicators of health and also eas-
ily observable to the parents. They nd birth weight as the main (if not the only)
measure of birth endowment, used in the literature. We would like to argue that
gender of the child and birth size should also be considered as potent early health
indicators, especially for areas where access to modern prenatal care (and hence to
machines for monitoring fetal health as well as sex) is scarce. Needless to say, in-
clusion of these additional endowment measures will be helpful in understanding the
dynamics of human capital formation which is an important area of research in itself.
7
Secondly, we contribute to the small yet important literature that emphasizes po-
tentially large fetal selection eect in many fetal origin studies (Currie, 2011; Sanders
and Stoecker, 2015; Ahsan and Bhowmick, 2015b). Sanders and Stoecker (2015) re-
port around 60% of perinatal deaths in the US is due to observed fetal deaths in 1989.
United States which was previously thought to be due to son preference alone.
7
It includes the issue of \dynamic complementarities" in human capital production function
which means returns to childhood investment are larger if the stock of human capital is higher in
the preceding period. The readers may refer to Heckman (2007) for details on this topic.
72
Based on the observation of male fragility in utero, they argue improvements in fe-
tal health due to Clean Air Act Amendments of 1970 (CAAA) in the United States
resulted in (relatively) more averted male fetal losses which were re
ected in higher
ratio of male live births. The authors provide an estimate of 9900 fetal losses that
could be prevented due to CAAA. Ahsan and Bhowmick (2015b) provide related evi-
dence in a developing country context. Basing on the well known fact that the Village
Midwife Program in Indonesia is eective in raising the health status of reproductive
age women, the authors nd that the provision of a midwife in a community also
increases the probability of a male birth by 4 percentage points, mostly for mothers
with at most primary education. In addition, they report that provision of midwives
leads to decrease in birth weight for male children while no change is observed for
female children. The authors explain this paradoxical result by noting that positive
nutrition shock may lead to increase in survival of poor quality fetuses. The authors
conclude that it is critical to account for selection into live births to examine the
impact of in utero shocks on later life outcomes.
Thirdly, a number of studies indicate how the societies suer at large from un-
equal treatment of girls through generations (Bhalotra and Rawlings, 2011; Osmani
and Sen, 2003). The reason for ill-treatment of girls in India is ascribed to high de-
gree of son preference (Bharadwaj and Lakdawala, 2013). We argue, intergenerational
persistence in poor health status of women in developing countries, can additionally
arise from biological considerations. The argument becomes clear if we extrapolate
our nding that male children are more likely to be born to healthier mothers, to the
past. Since India has experienced reasonable economic growth along with improve-
ments in public health infrastructure over time, greater and greater proportions of
Indian females (who are future mothers) were born to women with relatively poorer
73
health and socio-economic status in the past generations. In other words, the his-
torical burden of low health endowment for Indian women is actually greater than is
expected due to son preference alone | biology is also playing a role.
Fourthly, our nding that parental stock of health (measured by height) aects the
fetal outcomes provides further evidence for intergenerational transmission of health
through nongenomic channels, since adult height is sensitive to childhood environ-
ments (Deaton, 2006). Some recent studies have shown how shorter and longer run
ill-eects of being born in disadvantaged families (eg. born to shorter mothers) and/or
with poor birth endowments (due to early life trauma) can be, at times, mitigated
through investments in maternal education, economic growth, and public health pro-
vision (Bhalotra and Rawlings, 2013; Adhvaryu, 2014; Ahsan and Bhowmick, 2015b).
The studies also report stronger gains for initially relatively disadvantaged children
which is consistent with health production functions, exhibiting diminishing marginal
returns.
Finally, our results are important in the light of some recent policy debates in In-
dia. For example, it is argued any further reductions in infant mortality in countries
like India can be achieved mostly through investments in maternal health, educa-
tion and public health provision (Bhalotra and Rawlings, 2011; Deaton, 2006). The
contribution of this chapter is to investigate this issue by identifying the population
segments where the role of such parental, household and community factors in ascer-
taining fetal survival and quality is highest.
The organization of the chapter is as follows. We summarize the bio-medical and
economic literature on child health and its determinants in section 2. We discuss the
74
conceptual economic framework and estimation in section 3. In section 4, we describe
the data along with summary statistics. We present and discuss the regression results
with suitable robustness checks in section 5. Finally, we conclude in section 6.
3.2 Background Literature
3.2.1 Likelihood of Male Birth as an Indicator of Fetal Health
Quality
Sex ratio at birth (dened as male births, per 100 female births) as a measure of
societal well-being has been proposed in late 90's (Shifotoka and Fogarty, 2012). The
idea rests on two variants of adaptive sex ratio adjustment hypothesis which says
\parents can adjust ospring sex ratio adaptively to achieve optimal survival and re-
productive results" (Song, 2012). The earlier (and more popular) variant is known as
Trivers-Willard (TW) hypothesis (Trivers and Willard, 1973). It predicts mothers are
more likely to give birth to sons in favorable conditions and daughters in unfavorable
conditions. The reason lies in higher resource-sensitivity and variability of reproduc-
tive success of male ospring. Investments in sons yields greater reproductive future
returns than daughters of similar quality if quality is good and vice versa (Song,
2012). The other variant, advanced in (Myers, 1978), postulates mothers are more
likely to give female births in poor conditions not because of greater reproductive
success. This happens because daughters are less nutrition-intensive than boys which
is a great survival quality in tough conditions. Numerous studies have tried to test
the adaptive sex ratio hypothesis for both the explanations. Using National Center
for Health Statistics (NCHS) Vital Statistics Birth Cohort Linked Birth/Infant Death
Data, Almond and Edlund (2007) nd married, educated and younger mothers are
75
more likely to give birth to a male ospring, supporting TW hypothesis. A married
mother who had some college education is approximately 0.80% more likely to bear
a son (or have a son survive to age 1) than an unmarried mother who did not com-
plete high school. The authors argue postnatal mortality is important in shaping TW
pattern. Bhaskar and Gupta (2007) nd evidence in support of TW hypothesis for
South and East India, while they ascribe unbalanced sex ratios in North and West
India due to discrimination against girls. One notable exception to these ndings is a
study by (Orzack et al., 2015), who nd sex-ratio to be male-biased among abnormal
human embryos.
However, causal analysis with social status and health conditions is dicult due
to their non-manipulability in an experimental set-up (Song, 2012). So there is a huge
literature which examines exogenous variation of maternal conditions in the context
of natural experiments. In particular, these studies have focused on the impacts of
intrauterine nutrition on sex ratio along with other infant and adult outcomes. For
instance, Song (2012) nds evidence in favor of adaptive sex ratio hypothesis for
Chinese Great Leap Forward Famine. (Hern andez-Juli an et al., 2014) nd similar
results for 1974 Bangladesh famine. Some other studies which nd positive evidence
include (Fukuda et al., 1998) for earthquake in Japan, Zorn et al (2002) for war in
Slovenia, Catalano et al (2006) for terrorist attack, Sanders and Stoecker (2015) for
air pollution, Williams and Gloster (1992) for food availability, Shifotoka and Fogarty
(2012) for prevalences of HIV and tuberculosis etc.
8
.
8
Studies which link maternal nutrition to sex ratio in a non-experimental set up nd mixed
results(Hern andez-Juli an et al., 2014) These studies include dierent markers of maternal nutrition
including height, weight etc. We discuss more about them in the next subsection.
76
3.2.2 Health Production Functions
Thomas et al. (1991) note the biomedical approach of estimating child health pro-
duction functions typically involves modeling several anthropometric outcomes as
functions of dierent health inputs (like diet, health care usage, parental time alloca-
tion, water and sanitation facilities etc) as well as characteristics of the child, parents
and environment. The functions may vary by child age, gender and innate healthi-
ness. Naturally, knowledge of all the inputs as well as their prices are required for
estimating such production functions. For lack of such huge data, economists resort
to estimation of reduced form health functions, obtained from maximizing household
utility subject to dierent budget constraints and health production functions. We
undertake similar approach in this chapter. Exact details of this approach can be
found under the subsection "Model".
However, a number of medical studies which investigate determinants of fetal
health, more specically sex ratios,
9
have made little distinction between pre-determined
and endogenous inputs. For example, James (1987) cites numerous studies which re-
late maternal smoking, season, maternal diets to sex ratio. Although important in
the production functions of fetal quality and survival, these inputs are jointly deter-
mined with the outcomes of interest. In other words, such studies have ignored the
possibility of economic choices a household makes. Strauss (1990) notes, such studies
don't distinguish between production function and reduced form and thus, OLS esti-
mates produce a hybrid of the two. Consequently, absent suitable instruments, these
estimates are biased and quite misleading.
9
Song (2012) argues omitted variable bias (especially non-inclusion of genetic and environmental
factors) in such observational studies have resulted in inconsistent ndings regarding whether poor
maternal nutritional conditions reduce the proportion of male births among human.
77
The above point is, especially helpful in nding the measure of maternal health,
suitable in such studies. Gibson and Mace (2003) demonstrate a strong association
between the sex of the most recent birth and maternal nutritional status, measured
either by body mass index (BMI) or mid-upper arm muscle area (AMA) (measures
of fat and muscle mass) for southern Ethiopia.
10
In another study, Bhalotra and
Rawlings (2011) consider maternal anemia and Body Mass Index (BMI), along with
maternal height to investigate health of the child. Except for height, all the remaining
measures are short term measures of maternal health. Since these short term mea-
sures of maternal health are not pre-determined, they may have changed at the time
of the survey from the time of reproductive outcome of a baby (live birth, miscarriage,
abortion etc). In light of the discussion above, parental short term conditions (e.g.
weight, body mass) and outcomes related to fetal health and reproduction are jointly
determined by households. Without randomization, their estimated relationship will
be confounded by unobserved determinants and misleading.
11
On the other hand,
parental height is an indicator of their long run nutrition status and as such more
suited for our study. We, thus, consider parental height as indicator of their health
in our analysis.
We restrict our analysis only to rst births. The reason is twofold. Firstly, higher
10
Other such studies which link maternal health to sex ratio include Andersson and Bergstr om
(1998); Stein et al. (2004) etc. Like Gibson and Mace (2003), Andersson and Bergstr om (1998)
found positive association of better maternal health with more male births. But the second study
nds a much weaker support for the adaptive sex ratio hypothesis.
11
An example is more expenditure on food items may result in lesser expenditure on mandatory
antenatal care (due to limited budget constraint). More food consumption increases the BMI of the
parents, reduces incidence of anemia and thus reduces possibility of adverse reproductive outcomes.
On the other hand, less antenatal care increases the chances of miscarriage, still-birth or induced
abortions. If the latter negative eects outweigh the former positive eects, we will falsely conclude
better maternal health is associated with worse reproductive outcomes! Controlling for antenatal
care usage and food expenses is insucient as these are choice variables themselves.
78
order births can be endogenous to the child quality (including gender, general health-
iness, cognitive abilities etc) at rst birth. We have argued earlier why any child
quality is endogenous. Secondly, although important in fetal quality and reproduc-
tion production functions, variables like child gender, maternal age at birth, birth
order, birth intervals etc are not pre-determined. Including them will result in biased
estimates.
12
Theoretically, parental education can aect child health directly as well as through
the choice of inputs into the production function which can be termed as technical
and allocative eciency respectively (Thomas et al., 1991). There is a substantial
literature which shows a positive impact of parental education, especially maternal ed-
ucation on child survival and child height. Using Colombian Census data, Rosenzweig
and Schultz (1982b)) nd maternal education increases chances of child survival by
lowering the cost of using benecial child health technologies through greater ability
of access and utilize new information. But some of the studies argue the correlation
between education and child height is due to some unobserved background variables
including maternal health and there is no direct eect of education on height as such
(Behrman and Wolfe, 1987,?). But Thomas et al. (1991) cite a number of studies
which nd independent eects of parental education on child height. The authors
conclude, in the context of Brazil, the impact of maternal education on child height
can be entirely explained by availability of information, similar to Rosenzweig and
Schultz's nding fo child survival. Some studies nd positive eects of parental ed-
ucation on fetal quality as well. For instance, Almond and Edlund (2007) report
educated mothers are more likely give birth to a male child in the US. The specic
12
For instance, Bhalotra and Rawlings (2011, 2013) include variables like child gender, birth order,
age at birth to examine infant mortality and anthropometric failure. Stein et al. (2004) include birth
spacing to examine relation between maternal health and sex ratio at birth.
79
mechanisms behind this association, however are not fully clear.
Thomas et al. (1990) nd parental height (especially maternal) to have a positive
impact on child survival in Brazil, after controlling for parental education and income.
They hypothesize parental height can be a proxy for genetic factors, human capital
investment as well as family background. Similar ndings have been reported in a
number of later studies.
13
Eriksson et al. (2010) discuss mechanisms which link fetal health to maternal
health conditions. Nutrition of fetus depends on maternal nutrition (includes both
the shorter run aspects like diet and nutritional stores as well as lifetime nutrition,
re
ected in metabolism) along with the size of placenta, re
ecting the ability to
transport nutrients from mother to fetus. The authors treat maternal height as an
indicator of lifetime nutrition and metabolism. They argue, short mothers have lesser
protein synthesis during pregnancy and therefore make a lesser amount of amino acids
available to the fetus. In fact, around one quarter of the variability in the length of
newborn babies is related to maternal protein synthesis, which depends on the size of
the mother's visceral mass, which in turn is linked to her height. A small placental
area, would therefore have more severe eects on fetal nutrition if the mother was
short. Thus, fetal under-nutrition manifests in reduced growth and birth weight at
early life while increasing vulnerability to cardio-vascular and hypertension diseases
in later life.
13
To name a few, Lavy et al. (1996) nd similar results for Ghana, especially for rural areas;
Bhalotra and Rawlings (2011, 2013) present cross-country evidence for the same phenomenon.
80
3.3 Conceptual Issues
3.3.1 Model
We follow the conceptual framework for unitary household behavior developed in
Strauss (1990) and Thomas et al. (1990). Parents try to maximize household utility
by choosing consumption levels of both food and non-food items and quality of chil-
dren who are in-utero (and thus yet to be born)
14
. More specically, we focus on two
dimensions of fetal quality viz. survival to the term (as measured by likelihood of a
male at rst birth, and mortality risk within one month of birth) and fetal growth
and nutrition, measured by birth weight and size at birth
15
. The reason for using
birth size as a measure of fetal growth and nutrition along with the more popular
measure of birth weight because there is no one to one correspondence between the
two measures. Two babies of same size, may face dierent patterns of fetal growth so
that the relative size of dierent organs at birth are dierent across the babies. Simi-
lar argument goes for birth weight. Barker (2000) advocates the importance of more
detailed anthropometrics at birth so that more insights can be gained regarding dier-
ent adaptations a fetus makes, in response to dierent environmental stressors.
16
Since
food consumption may directly aect people's satisfaction, we include it in the util-
ity function as an argument. A similar argument applies for including fetal quality
(which is a strong indicator of future health of an individual) in the utility function.
14
Non-food items may also include number of children. In our study, however, we focus only on
rst births and so we don't consider this dimension anymore.
15
Almond et al. (2011) argue neonatal mortality which is roughly, mortality risk within a month
is commonly linked to the health environment during pregnancy. Postneonatal deaths or deaths
between 28 days and a year after birth are usually due to post-birth negative shocks like infectious
diseases and accidents.
16
He cites studies which nd, eg. people who were thin at birth but with normal birth weight, tend
to be insulin resistant and thus, have higher chances of developing non-insulin dependent diabetes.
81
The constraints which the household faces include the time constraint, usual
household budget constraint and health production functions. The last one cap-
tures the relationship between fetal quality and a number of inputs as described in
equation (1) below. We largely follow the exposition in Strauss and Thomas (1998).
H =H(C;N;T ;X;D;;e
h
) (3.1)
where H represents dierent outputs related to fetal quality; C, N, T represent con-
sumption of food or non-food items, health inputs and leisure respectively and under
the control of parents. We include consumption levels of food (includes smoking
and alcohol consumption) in this function because fetal nutrition depends directly
on maternal nutrition. And non-food items (like books, movies, traveling etc) can
aect maternal stress system which in turn can aect fetal health.
17
Hamoudi and
Nobles (2014) cite many recent studies which indicate women exposed to chronic
hyper-stimulation of stress system are less likely to give a live birth. Examples of the
former could be antenatal care like iron supplementation, tetanus injections etc and
usage of dierent community services. Health is assumed to increase in both health
inputs and leisure, but the sign is negative or positive, depending on the specic items
for food and non-food consumption. The pre-determined inputs include household
level variables,X like parental age, education, health (we consider only height), asset
holding etc and community level variables, D, capturing among other things disease
environment and local public health infrastructure. is the pre-determined com-
ponent of fetal quality due to genetic and environmental conditions and may be at
least, partially known to the parents. e
h
represents the measurement error, unknown
to both the econometrician or the parents.
17
Rosenzweig and Schultz (1982a) regard consumer goods like books as health-neutral. We argue
above why this may not be true for fetal health, in light of new medical evidence.
82
Maximizing the household utility subject to the three constraints yields the re-
duced form equation for any of the measures of fetal quality:
O =O(X;D;P;;e
h
) (3.2)
where O is the output, P represents a set of prices while the other notations have
already been dened above.
3.3.2 Potential Empirical Issues
Strauss (1990) points out the problems of treating several of the observable variables
like parental education and health and community variables as being uncorrelated
with unobserved household level variables. Selective migration and endogenous pro-
gram placement may also result into misleading correlations between community fac-
tors and investments in maternal health during pregnancy or neonatal child care.
One way to circumvent the problem would be to include household xed eects in
the regression equation. Under linear specications, such approach eliminates time
invariant unobserved factor. However, inclusion of household xed eects leads to
dropping nearly ninety percent of our sample and so we refrain from using it. In
addition, the issue of selective migration is not a matter of big concern in our setting
for a couple of reasons. Firstly, migration rate in rural India is very low, partly due to
the existence of caste networks acting as mutual insurance to their members (Munshi
and Rosenzweig, 2009); and secondly, we restrict our sample to rst borns in last
ve years of the survey, thus reducing the possibility of migration to a great extent.
Finally, we use amount of land owned as exogenous. This is again justied in Indian
context, as land ownership stays relatively constant over time in India.
83
3.3.3 Estimating Equation
O
ismy
= +E
0
i

E
+H
0
i

H
+I
0
i

I
+X
0
+
smy
+"
ismy
(3.3)
whereO
ismy
represents the fetal outcome for the rst child of motheri who is born in
states, in monthm and yeary. E;H andI represent vectors for measures of parental
education, parental height and household income respectively; X include covariates
like dummies for mother age, relationship of the mother with the household head,
interview date and religion of the household.
smy
is the state by birth month by
birth year dummies, which proxy for dierent prices, among other things. "
ismy
is the
idiosyncratic error term.
3.4 Data
We use the second and third waves o the National Family Health Survey(NFHS),
conducted in the years 1998-99 and 2005-06 respectively. This repeated cross-section
data set includes socio-demographic background of the household members, along
with complete fertility history, health inputs and birth weight and size information of
the children. Each round includes roughly around 90,000 households, representative
of the state and national level. We restrict our sample only to ever-married women
in the age group 18-49 years with information on height, dropping those who have
twins in the rst birth and who are visitors to the household where interview took
place. Also, the states of Chattisgarh, Jharkhand and Uttaranchal were created after
the 2nd wave. So we code these states the same as they originally belonged to viz.
Madhya Pradesh, Bihar and Uttar Pradesh respectively. We have again restricted
the sample to mothers who have given their rst births at most ve years before the
date of interview so as to minimize recall bias. While the birth weight and birth size
84
information are available for all rst born within last ve years of the survey in the
third wave, it is available only for babies born with last three years in the second
wave. Also, community survey is done only in the second wave and is restricted to
rural areas only, We could not match all the communities to mother level information
in the states of Assam and Arunachal Pradesh. So we drop these two states entirely
for specications where community characteristics are included.
3.4.1 Dependent Variables
We measure fetal survival by two outcome variables. First, a dummy taking the value
of 1 if the rst child is a male, and zero otherwise. The second variable is a dummy,
taking the the value 1 if the child died within a month of birth, and zero otherwise.
In Table 3.1, we present the number of observations, mean and standard deviation
for these variables for the dierent sub samples. Likelihood of a male birth is between
51 and 52 percent across all sub samples which is close to the biological ratio. Rural
sample has signicantly higher rates of neonatal mortality which is driven by the high
infant mortality region (nearly double the rates in rural areas).
Fetal growth and nutrition is measured by birth weight (in kilograms) and a
dummy which takes the value of 1 if the child had a birth size greater or equal to
average, and 0 elsewhere. We construct two additional dummies if the child is of low
birth weight (birth weight less than 2.5 kilos) or of very low birth weight (birth weight
less than 1.5 kilos). In Table 3.2, we nd birth weight information is available for
mostly urban samples- so all the subsequent results related to birth weight have to
be interpreted with some caution. Children born to mothers in rural areas have lower
birth weight on an average- for instance, the incidence of lower birth weights is 23% in
85
the rural areas at the national level as opposed to 20% in the urban areas. The rural-
urban gap is again, wider in states with high rates of infant mortality. Incidence of
very low birth rate is around 1 to 2 percentage points in the overall sample. Children
of healthy birth size are more likely to be born in urban areas while the rural-urban
gap is around 5 percentage points across all sub samples.
3.4.2 Household and Community Variables
We show the summary statistics for parental education, maternal height and land
holdings in Table 3.3. More than fty percent of mothers from the rural areas of
high infant mortality region have no formal school education while the gure drops to
nearly twenty percent for the urban areas. The proportions are signicantly lower in
the low infant mortality rates- twenty ve and eleven percentage respectively. Fathers
are more educated on average than mothers in all sub samples with greater rural dis-
advantage in high infant mortality states. The average mother height in urban areas
is close to 152.5 centimeters while it drops by one centimeter in the rural areas for
the high infant mortality states. Landholdings are higher for a rural household from
high infant mortality region as compared to all other sub samples (1.25 acres vis-a-vis
less than an acre).
In Table 3.4, we present the summary statistics for community characteristics. A
typical village from low infant mortality states has more irrigable land, more likely to
be electried, less likely to have any epidemic in the year, preceding the survey, more
likely to have a private or a visiting doctor and has a greater coverage of government
health facility than a village from high infant mortality states. However, it is farther
from amenities like school, all-weather road, transport facility, private health facility
86
etc and usually have lower number of beneciaries of dierent government schemes.
3.5 Results
3.5.1 Parental Education
We rst discuss the results for fetal survival measures which include likelihood of a
male child at rst birth and if the child at rst birth died within a month (Table 3.5
and Table 3.6). Mother's and father's education are usually both negatively and
signicantly associated with neonatal mortality with a signicant regional variation.
Likelihood of a male child at rst birth has no association with education levels of
parents, consistent with the ndings in the literature (Rosenblum, 2013).
The eects of maternal education are stronger for rural areas, especially in the
states where infant mortality rates are higher. For instance, a child from rural high
infant mortality region whose mother has completed secondary (or higher) education,
is 3 percentage point less likely to die as a neonate than a child with an illiterate
mother from that region. Eects of maternal education are much less (at times in-
signicant) for all other areas. Unlike our ndings, Thomas et al. (1990) report a
somewhat contradictory nding for Brazil where they nd higher impact of mother's
education for poorer regions of Brazil while lower impacts in rural areas.
Similar to mother's education, eects of father education on neonatal mortality
are signicant only in high infant mortality states. Interestingly, the eects are much
stronger for urban areas of this region- the magnitudes are extremely large given the
average mortality rates in these areas. Importance of father's education in child sur-
vival is well-recognized in the medical literature. The readers may refer to the studies
87
mentioned in Thomas et al. (1990).
Turning to the birth weight and birth size regressions (Table 3.7 and Table 3.8),
we again nd positive signicant impact of parental education on birth weight and
birth size.
It seems parents completing only primary education does not seem to have much
impact on measures of either fetal survival or nutrition outcomes. This indicates a
certain minimum requirement of education on the part of parents which can translate
into better health for children.
3.5.2 Parental Height
Maternal height is extremely important for fetal survival, especially for rural areas
and high infant mortality regions. For instance, a 10 centimeter increase in height
of a mother from rural area of high infant mortality region is associated with more
than 2% increase in likelihood of a male child at rst birth and approximately 2%
less likelihood of a child dying as a neonate. When we consider the urban area of this
region, the magnitude remains roughly similar for neonatal mortality, while it drops
sharply and becomes insignicant for the likelihood of a male child. For low infant
mortality regions, the impacts on neonatal mortality are much lower in urban areas-
although the magnitudes are similar across urban and rural areas, they are signicant
only in urban area. We nd no signicant impact of maternal height on the likelihood
of male child for either areas from this region.
Our nding that maternal height can signicantly impact gender of the child is a
88
novel result in the economic literature, especially in the Indian context. As discussed
earlier, medical studies linking maternal health to gender of the child have found
inconsistent results, probably due to failure to account for economic choices a house-
hold makes involving maternal inputs and fertility (like birth order). We circumvent
these problems by restricting our analyses to rst births and through estimation of
reduced form functions. Given the cultural and economic context of India, one may
think that the positive association between maternal height and likelihood of a male
child could be driven by sex selective abortion. Although we can not rule out such
possibility but we do not think sex selective abortion is responsible for observed rela-
tionship between maternal height and likelihood of a male birth for various reasons.
Firstly, there is a huge literature which argue absence of sex selection at rst birth
order in India (Portner, 2015; Rosenblum, 2013). Proportion of males in our sample
is between 51% and 52% which is close to the natural biological ratio. Secondly, ear-
lier studies nd educated women to be more likely to perform sex selective abortions
(Portner, 2015; Bhat and Zavier, 2007). With the exception of mother's height, we
don't nd any other parental or household correlates for male child. As a robustness
check, we use data from Ethiopia to explore whether or not the result is applicable
for countries with low son preference but suer from adverse disease environment and
lack quality health infrastructure. Table 3.14 in the Appendix shows shorter mothers
in the less-advanced South are less likely to give birth to male children. Although
the relationship is not statistically signicant, it is supported by the ndings of Gib-
son and Mace (2003) and Stein et al. (2004). Following Bharadwaj and Lakdawala
(2013), another robustness check could have been to perform the analysis for rst
NFHS wave only when ultrasound (used to identify sex of the fetus) was not available
in many parts of India. Unfortunately, this wave does not contain any parental height
information, which is crucial for our analysis.
89
We nd substantial impact of maternal height on fetal nutrition as measured by
birth weight and size. Unlike fetal survival regressions, this impact is stronger in
urban areas at the national level- the impact of maternal height on incidence of low
birth weight or the likelihood of a child to have decent birth size is nearly double
in urban areas as compared to rural areas. However, this observation does not hold
for birth weight when we focus on the high infant mortality region. The impact of
maternal height on birth weight or incidence of low or very low birth weight is a bit
higher in rural areas for this region. Overall for this region, 10 centimeter increase in
maternal height is associated with 160 grams increase in birth weight, 6% to 7% lower
incidence of low birth weight and 4% to 6% increase in incidence of decent birth size.
Going by the mean levels of these variables, the magnitudes are quite large. While
the direction is negative, we don't nd signicant impact of maternal height because
of extremely low incidence of very low birth weight. Interestingly, the estimated im-
pacts in urban area are roughly similar across high and low infant mortality regions.
Rural areas of low infant mortality regions exhibit a much muted impact of maternal
height on these indicators.
With inclusion of father's height,
18
the impact of maternal height on neonatal
mortality becomes lower than the full sample regressions in the rural areas, but re-
mains similar (and signicant) for urban areas. Father's height has no signicant
impact in either rural or urban India on likelihood of being a male child or neonatal
mortality. Father height has signicant impact on birth weight of the child. In fact,
the impact on birth weight is similar for mother and father's height in rural areas.
18
Father's height is available only in the third wave of NFHS. We don't include the tables for the
regression with father's height in the chapter; they are available upon request.
90
For urban areas, maternal height seems to be a stronger correlate than father's height
for birth weight measures. As we saw earlier in Table 3.7 and Table 3.8, the impact
of maternal height remains stronger in rural areas. We do not nd any signicant
impact of father's height on birth size.
3.5.3 Income
Income, as proxied by quartic root of acres of agricultural land have negative impact
on neonatal mortality in the rural areas. The impacts are similar across the rural
areas of both low and high infant mortality regions. However, the relationship is not
statistically signicant because of large standard errors. Turning to fetal nutrition
measures, land holdings do not seem to have signicant impact on birth weight but
have some impact on birth size. The impact is positive and stronger in rural areas- a
unit increase in land in rural area from either region is associated with 3% increase
in likelihood of a child with healthy birth size. However, the association is negative
in the urban areas, although insignicant or marginally signicant across dierent
sub-samples. Since agricultural land is not a good proxy for income for urban resi-
dents, the results possibly indicate an underlying selection. It is possible that urban
residents with greater agricultural land are more recent migrants from rural areas
and have greater ties with their native villages. They are yet to realize the greater
benets of living in urban areas and it is re
ected in the observed negative association
between land holdings and birth size.
19
In the restricted samples with father height, we could include another measure
of income: a dummy if the household possesses a below-poverty line (BPL) card.
19
In fact, the relationship between neonatal mortality and land holdings across all urban sub
samples is also positive, though not statistically signicant.
91
Unsurprisingly, a family from rural India with a BPL card is 4% less likely to have
a male child at rst birth. This result reverses for urban India, but is statistically
insignicant. Possessing a BPL card is also associated with a signicant negative
impact on neonatal mortality in urban area. It has a signicant positive impact on
birth weight in rural areas, along with land holdings. However, the impact on birth
size is positive but not signicant at conventional levels.
3.5.4 Access to Information
One of the main mechanisms through which maternal education impacts child health
is greater access to information (Thomas et al., 1991). So we investigate if mother's
access to information is important in determining fetal health. We measure this by
an indicator if the mother watches television every week.
20
Since this information
access indicator is endogenous, so we instrument it with number of TV sets in the
community, distance from post oce, distance from phone booth with national calling
facility. Instead of using continuous measures, we use the highest 3 quartiles of these
variables in the regression with the lowest quartiles, serving as omitted categories.
For this analysis, we use only the rural sub sample of second wave of NFHS as the
community level information (used as instruments) is available only for this wave and
only for rural areas.
In Table 3.9, we present the results for high and low infant mortality states sepa-
rately. We don't show the results related to birth weight as the sample size becomes
extremely small and the instruments cease to be strong. We report the impact of
maternal height as well to further check if the eect of maternal height to fetal health
20
We considered two other measures of access to information variables- indicators if the mother
reads newspaper daily and if she listens to radio daily. Unfortunately, we could not nd strong
instruments for these variables.
92
is indeed through biological channel and not through any other channel. It turns out
the eect of maternal height on likelihood of male child, neonatal mortality and like-
lihood of healthy-sized child is roughly unchanged and remains strongly signicant
with the inclusion of access to information. The eect of watching television is posi-
tive (but not always signicant) on all measures for low infant mortality rates and the
magnitudes are substantial- a child from this region whose mother regularly watches
television is 7% less likely to die within a month of birth (statistically signicant at
10%) than the child whose mother watches television less frequently. Although not
individually signicant, the information variable is jointly signicant with maternal
height. For high infant mortality state, this impact is reduced by fty percent. How-
ever, the sign reverses for male child or birth size regressions. Looking at the p-value
for overidentifying restrictions for these two sets of regressions, it appears the instru-
ments have not fully accounted for possible endogeneity. So we should interpret these
two associations with caution.
3.5.5 Community Characteristics
As mentioned earlier, the community information is available only for the rural areas
and for the second wave of NFHS. Because of small sample size, we present the birth
weight results for entire rural India while other measures are shown for high and low
infant mortality states separately. We divide community characteristics into three
broad characteristics, viz. general characteristics, health facilities and government
schemes. Because of signicant correlation within each group of characteristics, we
report the joint signicance of each group in the tables. The results related to fetal
survival and birth size are shown for each category of community features are shown
in Table 3.10, Table 3.11 and Table 3.12 while the same for birth weight are shown
93
together in Table 3.13. Here we should point out that these results should be inter-
preted with caution as we do not account for the possibility of endogenous program
placement (Rosenzweig and Wolpin, 1986). Indeed, a quick glance at Table 3.4 reveals
very similar levels of community infrastructure across high and low infant mortality
regions. This may re
ect higher investment in public infrastructure in these states to
overcome the incidence of poor health in this region.
Turning to the results, community characteristics seem to be strongly associated
with fetal nutrition and not much with the survival measures. General village charac-
teristics do not have strong correlation with either fetal survival or nutrition measures.
Prevalence of agriculture (measured by availability of irrigable land in the village) has
a consistent negative association with any measure of fetal health and across all sub
samples, although not always statistically signicant. For all the remaining charac-
teristics, the direction of association with any health measure is mostly inconsistent
across sub-samples, possibly due to optimally placed public programs. This is most
evident for the availability of health facilities. For example, presence of a village
health guide is associated positively with birth size in high infant mortality states
while the association is negative in the low infant mortality states (Table 3.11). Sim-
ilar inconsistency is found for presence of other health facilities. Further, the impact
of number of beneciaries for dierent government schemes on neonatal mortality
and birth size is strong in high infant mortality states- but the association is again
inconsistent as we compare one scheme with the other. While number of beneciaries
under \Training Rural Youth for Self-Employment" (TRYSEM) is positively associ-
ated with neonatal mortality, the association is negative when we consider another
scheme, viz. "Indira Awas Yojana" (IAY).
94
To sum up this section, we nd the impact of community infrastructure on fetal
health is slightly stronger in high infant mortality states. But drawing any policy
conclusion would be inappropriate as the results are not robust across dierent sub
samples. Any further analysis on this topic should proceed in the framework of
program evaluation. Also, data on quality of infrastructure should be more helpful
than gures on only quantity and accessibility to get a clearer idea of the impact
of public infrastructure on any health measure. For Ghana, Lavy et al. (1996) nd
child services (measured by the weekly hours of availability of child health care)
has a signicant and positive impact on the survival of children. Further, they nd
investment in facility quality, not availability, is more important for child height.
3.6 Conclusion
Despite phenomenal economic growth in last two decades, India ranks nearly at the
bottom in child health indicators. The rates of infant mortality in India are not
only high by global standards, but also have signicant rural-urban divide with large
inter-state variation. Fetal health being important for subsequent health or later life
outcomes, we examine in this chapter whether parental human capital can substitute
for low provision of quality health infrastructure in areas of India with adverse disease
environment. Recognizing the multi-dimensional nature of health, we consider two
dimensions of fetal health namely fetal survival to the term and fetal growth and
nutrition. Along with the common measures of fetal health such as birth weight and
neonatal mortality, we use two less frequent measures of fetal health | likelihood of
a male birth which re
ects male fragility in utero and birth size based on insights
from the bio-medical literature. Noting that rural areas have lower infrastructure and
states with higher infant mortality have worse disease environment, we run separate
95
analysis for each region, controlling for state by month by year eects and various
other variables. We nd parents, especially mothers with at least secondary educa-
tion from rural areas of states with high infant mortality, are more likely to have a
child who is male, less likely to die as a neonate, better birth weight and birth size.
While parental education has lower impact for low infant mortality states, maternal
height remains important for most measures of fetal health. Overall, the results are
consistent with the substitution hypothesis that parental human capital can, at least
partially compensate for poorer health infrastructure. We perform several checks to
ensure the relationship between maternal health and likelihood of a male child is
driven by biological reasons and not by sex selective abortions.
Our nding questions the validity of using gender of the rst child as a natural
experiment or its inclusion as an independent control variable in the Indian con-
text. Another implication is males are more likely to be born to healthy mothers and
children born to healthy mothers are also more likely to be healthy. Thus, intergen-
erational persistence in poor health status of women can arise also due to biological
reasons, along with son preference, given the cultural and economic context of India.
Our results are important in understanding the issues related to parental responses
to birth endowments and dynamics of human capital formation. They also under-
score the need to correct for fetal selection, mostly ignored in fetal origin studies in
developing countries.
96
3.7 Tables
Table 3.1: Fetal Survival Figures for India
Rural Urban
Observations Mean S.D. Observations Mean S.D.
All India
Male 16632 51.92% 0.4996 10,518 51.47% 0.4998
Neonatal Mortality 16,632 5.06% 0.2192 10,518 3.13% 0.1741
High Infant Mortality States
Male 8,790 52.08% 0.4996 4,233 51.05% 0.4999
Neonatal Mortality 8,790 6.37% 0.2442 4,233 3.85% 0.1924
Low Infant Mortality States
Male 7,842 51.73% 0.4997 6,285 51.76% 0.4997
Neonatal Mortality 7,842 3.60% 0.1862 6,285 2.64% 0.1604
97
Table 3.2: Birth Weight and Birth Size Figures for India
Rural Urban
Observations Mean S.D. Observations Mean S.D.
All India
Birth Weight (in kilograms) 4,677 2.7975 0.6822 6,028 2.8341 0.6480
Low BW 4,677 22.54% 0.4179 6,028 19.76% 0.3982
Very Low BW 4,677 2.07% 0.1425 6,028 1.69% 0.1290
Healthy-sized Child 11,789 75.68% 0.4290 8,285 80.28% 0.3979
High Infant Mortality States
Birth Weight (in kilograms) 1,514 2.7660 0.7516 2,125 2.8117 0.6967
Low BW 1,514 25.03% 0.4333 2,125 20.89% 0.4066
Very Low BW 1,514 2.51% 0.1565 2,125 2.21% 0.1471
Healthy-sized Child 6,006 76.59% 0.4235 3,392 81.25% 0.3904
Low Infant Mortality States
Birth Weight (in kilograms) 3,163 2.8125 0.6460 3,903 2.8463 0.6197
Low BW 3,163 21.34% 0.4098 3,903 19.14% 0.3934
Very Low BW 3,163 1.87% 0.1353 3,903 1.41% 0.1179
Healthy-sized Child 5,783 74.74% 0.4346 4,893 79.60% 0.4030
98
Table 3.3: Summary Statistics- Household Variables
High Infant Mortality States Low Infant Mortality States
Rural Urban Rural Urban
Obs Mean s.d. Obs Mean s.d. Obs Mean s.d. Obs Mean s.d.
Mother's Education
No Education 8790 52% 0.50 4233 21% 0.41 7842 25% 0.43 6285 11% 0.32
Primary 8790 16% 0.36 4233 12% 0.32 7842 18% 0.38 6285 10% 0.30
Secondary 8790 28% 0.45 4233 42% 0.49 7842 48% 0.50 6285 52% 0.50
Higher 8790 4% 0.20 4233 25% 0.44 7842 9% 0.28 6285 26% 0.44
Father's Education
No Education 8790 27% 0.44 4233 11% 0.31 7842 17% 0.38 6285 7% 0.25
Primary 8790 16% 0.36 4233 10% 0.30 7842 16% 0.37 6285 10% 0.30
Secondary 8790 43% 0.49 4233 45% 0.50 7842 53% 0.50 6285 52% 0.50
Higher 8790 15% 0.36 4233 33% 0.47 7842 14% 0.35 6285 31% 0.46
Mother's Height (in centimeters) 8790 151.54 5.51 4233 152.39 5.68 7842 152.05 5.59 6285 152.45 5.55
Acres of Agricultural Land 8790 1.25 6.35 4233 0.59 3.23 7842 0.95 4.59 6285 0.67 10.72
99
Table 3.4: Summary Statistics- Community Characteristics
High Infant Mortality States Low Infant Mortality States
Obs Mean s.d. Obs Mean s.d.
General Village Characteristics
Hectares of irrigable land in the village 4678 275.17 355.17 3830 278.44 386.65
Village is electried 4678 79% 0.40 3830 93% 0.25
Village has underground drainage 4678 2% 0.13 3830 2% 0.13
Zero epidemic in last one year 4678 62% 0.49 3830 67% 0.47
Distance from nearest town (in kms) 4678 15.54 13.67 3830 16.01 17.22
Distance from nearest transport facility (in kms) 4678 6.06 11.32 3830 10.79 22.64
Distance from all weather road (in kms) 4678 13.80 29.90 3830 19.79 36.03
Distance from nearest school (in kms) 4678 0.38 3.52 3830 0.50 5.54
Health Facilities
Village has:
Private Doctor 4678 37% 0.48 3830 40% 0.49
Visiting Doctor 4678 29% 0.45 3830 32% 0.47
Village Health Guide 4678 35% 0.48 3830 32% 0.47
Traditional Attendant 4678 57% 0.49 3830 60% 0.49
Mobile Health Unit 4678 13% 0.33 3830 11% 0.31
Number of Health/Family Welfare Camps last year 4678 1.35 3.25 3830 1.04 2.35
Distance from Nearest Government Health facility (in kms) 4678 2.99 5.35 3830 1.84 6.28
Distance from Nearest Private Health facility (in kms) 4678 8.78 12.31 3830 11.94 21.17
Government Schemes
Number of beneciaries
Training Rural Youth for Self-Employment (TRYSEM) 4678 2.19 8.35 3830 3.47 10.06
Employment Guarantee Scheme (EGS) 4678 2.69 13.69 3830 2.43 12.27
Development of Women and Children of Rural India (DWARCA) 4678 5.08 15.65 3830 4.45 13.98
Indira Awas Yojana (IAY) 4678 11.09 18.51 3830 8.22 15.13
100
Table 3.5: Fetal Survival Regressions- Rural
All India High Infant Mortality States Low Infant Mortality States
Male Neonatal Mortality Male Neonatal Mortality Male Neonatal Mortality
Mother's Education
Primary 0.00478 -0.00361 0.00502 -0.00799 0.0102 0.00847
(0.0136) (0.00646) (0.0173) (0.00887) (0.0221) (0.00893)
Secondary -0.00252 -0.0175*** -0.00839 -0.0272*** 0.0122 -0.00345
(0.0126) (0.00540) (0.0163) (0.00755) (0.0201) (0.00737)
Higher 0.00872 -0.0254*** -0.0220 -0.0331** 0.0356 -0.0147
(0.0228) (0.00866) (0.0331) (0.0137) (0.0329) (0.0110)
Father's Education
Primary 0.00780 0.00545 0.0206 0.00960 -0.0185 0.00218
(0.0148) (0.00741) (0.0187) (0.0100) (0.0243) (0.0105)
Secondary 0.0131 -0.00720 0.0330** -0.00995 -0.0216 -0.00323
(0.0129) (0.00629) (0.0160) (0.00826) (0.0218) (0.00918)
Higher 0.00715 -0.0105 0.00631 -0.0131 0.00435 -0.00303
(0.0181) (0.00812) (0.0228) (0.0110) (0.0298) (0.0115)
Mother's Height (in centimeters) 0.000719 -0.00128*** 0.00244** -0.00176*** -0.00147 -0.000769
(0.000834) (0.000385) (0.00110) (0.000550) (0.00128) (0.000517)
Quartic root of Acres of Agricultural Land 0.00585 -0.00472 0.00750 -0.00694 0.00519 -0.00210
(0.0108) (0.00459) (0.0143) (0.00684) (0.0163) (0.00574)
Observations 16632 16,632 8,790 8,790 7,842 7,842
Mean (dep. var.) 0.5192 0.0506 0.5208 0.0637 0.5173 0.0360
s.d. (dep. var.) 0.4996 0.2192 0.4996 0.2442 0.4997 0.1862
p-value of Mother's Education 0.913 0.0029 0.8397 0.0027 0.7588 0.1973
p-value of Father's Education 0.7789 0.1605 0.1448 0.144 0.5397 0.9215
Note: Robust standard errors are reported (*** p<0.01, ** p<0.05, * p<0.1). Other controls include dummies for religion of the household head,
mother age and relationship of the mother with the household head, interview month by year xed eects and state by birth month by birth year
xed eects.
101
Table 3.6: Fetal Survival Regressions- Urban
All India High Infant Mortality States Low Infant Mortality States
Male Neonatal Mortality Male Neonatal Mortality Male Neonatal Mortality
Mother's Education
Primary 0.0181 0.0126 -0.00855 0.0163 0.0378 0.00952
(0.0243) (0.00965) (0.0340) (0.0144) (0.0351) (0.0132)
Secondary -0.00130 0.00168 0.000835 0.0141 4.23e-05 -0.00932
(0.0200) (0.00753) (0.0284) (0.0112) (0.0288) (0.0105)
Higher -0.00439 -0.00720 -0.0428 0.00704 0.0181 -0.0206*
(0.0253) (0.00817) (0.0375) (0.0129) (0.0352) (0.0109)
Father's Education
Primary -0.0260 -0.00671 0.0101 -0.00434 -0.0595 -0.000356
(0.0280) (0.0106) (0.0404) (0.0171) (0.0394) (0.0133)
Secondary -0.0176 -0.0121 -0.00409 -0.0374*** -0.0380 0.0124
(0.0242) (0.00919) (0.0342) (0.0138) (0.0348) (0.0122)
Higher -0.0165 -0.0145 0.0208 -0.0397*** -0.0512 0.00967
(0.0281) (0.00991) (0.0415) (0.0154) (0.0392) (0.0130)
Mother's Height (in centimeters) -0.000306 -0.00128*** 0.000336 -0.00188*** -0.000824 -0.000834*
(0.00111) (0.000395) (0.00170) (0.000670) (0.00147) (0.000481)
Quartic root of Acres of Agricultural Land -0.000438 0.00204 0.00890 0.00440 -0.00448 0.000654
(0.0153) (0.00528) (0.0227) (0.00894) (0.0210) (0.00630)
Observations 10,518 10,518 4,233 4,233 6,285 6,285
Mean (dep. var.) 0.5147 0.0313 0.5105 0.0385 0.5176 0.0264
s.d. (dep. var.) 0.4998 0.1741 0.4999 0.1924 0.4997 0.1604
p-value of Mother's Education 0.8062 0.1243 0.5196 0.502 0.4972 0.0252
p-value of Father's Education 0.829 0.5127 0.8208 0.0189 0.4498 0.4911
Note: Robust standard errors are reported (*** p<0.01, ** p<0.05, * p<0.1). Other controls include dummies for religion of the household head,
mother age and relationship of the mother with the household head, interview month by year xed eects and state by birth month by birth year
xed eects.
102
Table 3.7: Birth Weight and Birth Size Regressions- Rural
All India High Infant Mortality States Low Infant Mortality States
Birth Weight Low BW Very Low BW Healthy-sized Child Birth Weight Low BW Very Low BW Healthy-sized Child Birth Weight Low BW Very Low BW Healthy-sized Child
Mother's Education
Primary 0.0718 -0.0214 -0.00571 -0.00938 0.269*** -0.0786 -0.0246 0.0179 -0.0424 0.00671 0.00516 -0.0430*
(0.0556) (0.0339) (0.0125) (0.0145) (0.0997) (0.0569) (0.0230) (0.0181) (0.0681) (0.0438) (0.0148) (0.0240)
Secondary 0.0967** -0.0610** -0.00941 0.0170 0.279*** -0.109** -0.0217 0.0276* -0.0254 -0.0293 0.00228 -0.00208
(0.0489) (0.0297) (0.0105) (0.0126) (0.0788) (0.0486) (0.0173) (0.0165) (0.0637) (0.0389) (0.0131) (0.0203)
Higher 0.168*** -0.104** -0.0163 0.0672*** 0.358*** -0.124* -0.0507** 0.0613* 0.0293 -0.0716 0.00298 0.0549*
(0.0636) (0.0406) (0.0133) (0.0224) (0.124) (0.0726) (0.0236) (0.0332) (0.0773) (0.0506) (0.0166) (0.0318)
Father's Education
Primary -0.0576 0.0453 0.0115 -0.00873 -0.177 0.0333 0.0335 -0.0141 0.0141 0.0525 0.00528 0.00354
(0.0638) (0.0406) (0.0142) (0.0159) (0.124) (0.0719) (0.0269) (0.0198) (0.0735) (0.0504) (0.0174) (0.0265)
Secondary -0.00917 0.00396 0.0131 -0.0124 -0.0912 0.00722 0.00123 -0.0322* 0.0541 3.67e-05 0.0177 0.0186
(0.0568) (0.0354) (0.0121) (0.0138) (0.101) (0.0593) (0.0173) (0.0168) (0.0688) (0.0448) (0.0168) (0.0237)
Higher 0.0244 -0.00544 0.0110 0.0332* -0.124 -0.00909 -0.00679 0.00330 0.119 -0.0132 0.0196 0.0802***
(0.0667) (0.0426) (0.0142) (0.0188) (0.125) (0.0718) (0.0225) (0.0239) (0.0796) (0.0535) (0.0188) (0.0309)
Mother's Height (in centimeters) 0.0113*** -0.00430*** -0.000789 0.00297*** 0.0164*** -0.00933*** -0.00153 0.00390*** 0.00994*** -0.00230 -0.000370 0.00185
(0.00258) (0.00160) (0.000546) (0.000845) (0.00550) (0.00318) (0.00120) (0.00113) (0.00290) (0.00188) (0.000602) (0.00127)
Quartic root of Acres of Agricultural Land 0.0305 -0.0226 -0.00135 0.0314*** 0.00523 -0.0127 0.00528 0.0270** 0.0468 -0.0278 -0.00353 0.0364**
(0.0272) (0.0170) (0.00636) (0.00926) (0.0524) (0.0293) (0.0111) (0.0122) (0.0317) (0.0212) (0.00782) (0.0144)
Observations 4,677 4,677 4,677 11,789 1,514 1,514 1,514 6,006 3,163 3,163 3,163 5,783
Mean (dep. var.) 2.7975 0.2254 0.0207 0.7568 2.7660 0.2503 0.0251 0.7659 2.8125 0.2134 0.0187 0.7474
s.d. (dep. var.) 0.6822 0.4179 0.1425 0.4290 0.7516 0.4333 0.1565 0.4235 0.6460 0.4098 0.1353 0.4346
p-value of Mother's Education 0.0723 0.0606 0.681 0.0088 0.0013 0.1415 0.1922 0.2167 0.6463 0.3933 0.9869 0.0213
p-value of Father's Education 0.5586 0.5026 0.7548 0.0161 0.5373 0.9305 0.4681 0.1003 0.3309 0.4596 0.4778 0.0252
Note: Robust standard errors are reported (*** p<0.01, ** p<0.05, * p<0.1). Birth Weight is expressed in kilo grams. Low BW and Very Low BW are indicators of Birth Weight less than 2.5 kilos and 1.5 kilos respectively. Other controls include
dummies for religion of the household head, mother age and relationship of the mother with the household head, interview month by year xed eects and state by birth month by birth year xed eects.
103
Table 3.8: Birth Weight and Birth Size Regressions- Urban
All India High Infant Mortality States Low Infant Mortality States
Birth Weight Low BW Very Low BW Healthy-sized Child Birth Weight Low BW Very Low BW Healthy-sized Child Birth Weight Low BW Very Low BW Healthy-sized Child
Mother's Education
Primary 0.0663 -0.0265 -0.0149 -0.00824 0.125 -0.0352 -0.0206 -0.0288 0.0599 -0.0194 -0.0126 0.0148
(0.0614) (0.0380) (0.0129) (0.0236) (0.113) (0.0609) (0.0201) (0.0324) (0.0718) (0.0495) (0.0171) (0.0346)
Secondary -0.00521 -0.0180 -0.0113 0.0124 -0.0863 -0.0333 -0.00693 0.0135 0.0640 -0.00913 -0.0151 0.0165
(0.0500) (0.0306) (0.0110) (0.0187) (0.0826) (0.0459) (0.0171) (0.0255) (0.0623) (0.0411) (0.0146) (0.0280)
Higher 0.0513 -0.0646* -0.0185 0.0295 -0.0761 -0.0543 -0.0180 0.0402 0.142** -0.0677 -0.0200 0.0206
(0.0556) (0.0342) (0.0121) (0.0230) (0.0940) (0.0541) (0.0184) (0.0325) (0.0681) (0.0447) (0.0160) (0.0331)
Father's Education
Primary 0.0358 0.0268 -0.00634 0.00191 0.115 0.0245 -0.0139 0.0333 -0.00671 0.0237 -0.00191 -0.0282
(0.0615) (0.0430) (0.0128) (0.0268) (0.119) (0.0761) (0.0238) (0.0364) (0.0720) (0.0526) (0.0150) (0.0393)
Secondary 0.0830 -0.0179 -0.00327 0.0564** 0.128 -0.00589 0.00266 0.0364 0.0448 -0.0294 -0.00385 0.0626*
(0.0528) (0.0367) (0.0101) (0.0229) (0.0910) (0.0595) (0.0178) (0.0315) (0.0650) (0.0467) (0.0125) (0.0334)
Higher 0.142** -0.0575 -0.000975 0.0947*** 0.146 -0.0250 0.00773 0.0714* 0.126* -0.0750 -0.00286 0.103***
(0.0569) (0.0391) (0.0114) (0.0258) (0.0990) (0.0646) (0.0199) (0.0371) (0.0694) (0.0492) (0.0141) (0.0365)
Mother's Height (in centimeters) 0.0155*** -0.00692*** -0.000978** 0.00553*** 0.0160*** -0.00609*** -0.000772 0.00566*** 0.0149*** -0.00723*** -0.00104* 0.00515***
(0.00208) (0.00127) (0.000425) (0.000971) (0.00394) (0.00219) (0.000668) (0.00143) (0.00239) (0.00157) (0.000541) (0.00133)
Quartic root of Acres of Agricultural Land 0.00358 -0.000567 0.00321 -0.0228* 0.0549 -0.00449 0.00485 -0.0154 -0.0362 0.00454 0.00345 -0.0298*
(0.0227) (0.0141) (0.00452) (0.0118) (0.0375) (0.0237) (0.00667) (0.0168) (0.0283) (0.0177) (0.00607) (0.0165)
Observations 6,028 6,028 6,028 8,285 2,125 2,125 2,125 3,392 3,903 3,903 3,903 4,893
Mean (dep. var.) 2.8341 0.1976 0.0169 0.8028 2.8117 0.2089 0.0221 0.8125 2.8463 0.1914 0.0141 0.7960
s.d. (dep. var.) 0.6480 0.3982 0.1290 0.3979 0.6967 0.4066 0.1471 0.3904 0.6197 0.3934 0.1179 0.4030
p-value of Mother's Education 0.1152 0.0606 0.4271 0.4274 0.1309 0.7863 0.4679 0.2904 0.0826 0.0625 0.6598 0.9371
p-value of Father's Education 0.0321 0.0354 0.9357 0.0001 0.5181 0.8622 0.6798 0.2478 0.0336 0.0396 0.9887 0.0003
Note: Robust standard errors are reported (*** p<0.01, ** p<0.05, * p<0.1). Birth Weight is expressed in kilo grams. Low BW and Very Low BW are indicators of Birth Weight less than 2.5 kilos and 1.5 kilos respectively. Other controls include
dummies for religion of the household head, mother age and relationship of the mother with the household head, interview month by year xed eects and state by birth month by birth year xed eects.
104
Table 3.9: Effect of Access to Information on Fetal Survival and Birth Size
High Infant Mortality States Low Infant Mortality States
Male Neonatal Mortality Healthy-sized Child Male Neonatal Mortality Healthy-sized Child
Mother's Height (in centimeters) 0.00294** -0.00201*** 0.00415** -0.000936 -0.000300 -0.00165
(0.00141) (0.000723) (0.00179) (0.00162) (0.000637) (0.00189)
Watches Television* -0.0270 -0.0342 -0.0718 0.0277 -0.0702* 0.108
(0.0875) (0.0424) (0.106) (0.0919) (0.0385) (0.127)
Observations 4,678 4,678 2,181 3,830 3,830 1,983
Mean (dep. var.) 0.5286 0.0648 0.7464 0.5253 0.0386 0.7262
s.d. (dep. var.) 0.4992 0.2461 0.4351 0.4994 0.1928 0.4460
p-value for Joint Signicance 0.1125 0.0149 0.0595 0.8276 0.1177 0.5134
First Stage: F 17.66 17.66 10.94 13.03 13.03 5.66
p-value for Overidentifying Restrictions 0.098 0.8702 0.1417 0.8906 0.4671 0.5028
* Instrumented by top three quartiles of number of TV sets in the community, distance from post oce and distance from phone booth with national
calling facility.
Note: Standard errors are clustered at the village level. (*** p<0.01, ** p<0.05, * p<0.1). Other controls include dummies for religion of the
household head, mother age and relationship of the mother with the household head, interview month by year xed eects and state by birth month
by birth year xed eects.
105
Table 3.10: Community Determinants of Fetal Survival and Birth Size- General Village Characteristics
High Infant Mortality States Low Infant Mortality States
Male Neonatal Mortality Healthy-sized Child Male Neonatal Mortality Healthy-sized Child
Quartic root of hectares of irrigable land in the village -0.00697 0.00108 -0.00191 -0.00125 -0.00132 -0.0316***
(0.00606) (0.00295) (0.00774) (0.00667) (0.00239) (0.00919)
Village is electried -0.0238 -0.0140 0.00607 -0.0182 -0.0230 0.0335
(0.0248) (0.0128) (0.0321) (0.0447) (0.0189) (0.0623)
Village has underground drainage -0.0334 -0.000358 0.187** -0.0137 -0.0451** -0.00468
(0.0650) (0.0272) (0.0747) (0.0801) (0.0178) (0.0821)
Zero epidemic in last one year -0.0272 -0.0117 0.0301 0.00294 -0.00709 0.0265
(0.0182) (0.00972) (0.0232) (0.0246) (0.00910) (0.0317)
Distance from nearest town (in kms) -0.000158 0.000183 0.000298 -0.000821 -2.95e-05 -0.00140
(0.000660) (0.000345) (0.000795) (0.000731) (0.000286) (0.00101)
Distance from nearest transport facility (in kms) 0.000244 5.47e-05 0.000646 0.000735 -1.99e-05 0.00243***
(0.000804) (0.000358) (0.000925) (0.000625) (0.000221) (0.000586)
Distance from all weather road (in kms) 9.67e-05 -0.000237* 0.000425 -6.50e-05 6.04e-05 -0.000193
(0.000274) (0.000126) (0.000395) (0.000353) (0.000118) (0.000427)
Distance from nearest school (in kms) 0.00302*** 0.000251 0.00679 -0.00182 -0.00177* 0.00101
(0.00112) (0.000822) (0.00649) (0.00236) (0.00107) (0.00293)
Observations 4,678 4,678 2,181 3,830 3,830 1,983
Mean (dep. var.) 0.5286 0.0648 0.7464 0.5253 0.0386 0.7262
s.d. (dep. var.) 0.4992 0.2461 0.4351 0.4994 0.1928 0.4460
p-value for Joint Signicance 0.0887 0.3849 0.1615 0.9136 0.2015 0.0000
Note: Standard errors are clustered at the village level. (*** p<0.01, ** p<0.05, * p<0.1). Only rural sample of wave 2 is used while dropping Assam and Arunachal Pradesh from
the sample. Other individual controls include dummies for religion of the household head, mother age and relationship of the mother with the household head. The specications
also include interview month by year xed eects, state by birth month by birth year xed eects, 3 dummies for population size in the village and 4 dummies for type of main
irrigation in the village.
106
Table 3.11: Community Determinants of Fetal Survival and Birth Size- Health Facilities
High Infant Mortality States Low Infant Mortality States
Male Neonatal Mortality Healthy-sized Child Male Neonatal Mortality Healthy-sized Child
Village has:
Private Doctor -0.0222 -0.00997 0.0220 -0.0289 -0.00853 -0.0515*
(0.0197) (0.0104) (0.0248) (0.0252) (0.00815) (0.0309)
Visiting Doctor -0.0254 0.00931 0.0487** -0.0203 -0.00198 0.0246
(0.0188) (0.00971) (0.0243) (0.0243) (0.00808) (0.0312)
Village Health Guide 0.0159 -0.0161* -0.0529** -0.0458* 0.00758 0.0768**
(0.0185) (0.00968) (0.0257) (0.0254) (0.00895) (0.0315)
Traditional Attendant 0.00703 0.00837 -0.0586** 0.0281 0.00241 0.0270
(0.0175) (0.00926) (0.0229) (0.0213) (0.00740) (0.0291)
Mobile Health Unit -0.0144 0.0129 0.121*** 0.0234 0.00743 -0.0696*
(0.0250) (0.0117) (0.0311) (0.0301) (0.0115) (0.0385)
Number of Health/Family Welfare Camps last year 0.00228 -0.000419 0.00264 0.00305 0.000372 0.00463
(0.00259) (0.00112) (0.00322) (0.00381) (0.00177) (0.00530)
Distance from Nearest Government Health facility (in kms) -0.000368 0.000570 -0.000680 0.00178 5.67e-05 -0.000942
(0.00146) (0.000675) (0.00106) (0.00228) (0.000834) (0.00274)
Distance from Nearest Private Health facility (in kms) 0.000175 0.000111 0.00111 -0.000707 -0.000254 0.000348
(0.000734) (0.000356) (0.000994) (0.000664) (0.000253) (0.000843)
Observations 4,678 4,678 2,181 3,830 3,830 1,983
Mean (dep. var.) 0.5286 0.0648 0.7464 0.5253 0.0386 0.7262
s.d. (dep. var.) 0.4992 0.2461 0.4351 0.4994 0.1928 0.4460
p-value for Joint Signicance 0.6823 0.4611 0.0001 0.2212 0.9089 0.0385
Note: Standard errors are clustered at the village level (*** p<0.01, ** p<0.05, * p<0.1). Only rural sample of wave 2 is used while dropping Assam and Arunachal Pradesh from
the sample. Other individual controls include dummies for religion of the household head, mother age and relationship of the mother with the household head. The specications
also include interview month by year xed eects, state by birth month by birth year xed eects, 3 dummies for population size in the village and 4 dummies for type of main
irrigation in the village.
107
Table 3.12: Community Determinants of Fetal Survival and Birth Size- Government Schemes
High Infant Mortality States Low Infant Mortality States
Male Neonatal Mortality Healthy-sized Child Male Neonatal Mortality Healthy-sized Child
Number of beneciaries
Training Rural Youth for Self-Employment (TRYSEM) -4.30e-05 -0.00101** -0.00101 0.000556 -0.000169 0.000996
(0.00111) (0.000412) (0.00143) (0.00135) (0.000395) (0.00146)
Employment Guarantee Scheme (EGS) -0.000132 0.000392 0.00238*** 0.000116 8.43e-05 -0.00111
(0.000665) (0.000333) (0.000789) (0.000829) (0.000256) (0.00121)
Development of Women and Children of Rural India (DWARCA) 0.000497 -0.000189 -0.00102 -0.000237 0.000174 -0.000608
(0.000568) (0.000260) (0.000674) (0.000863) (0.000292) (0.000996)
Indira Awas Yojana (IAY) -0.000217 0.000431* -0.00104* 0.000290 0.000105 -0.000468
(0.000525) (0.000253) (0.000550) (0.000753) (0.000278) (0.000970)
Observations 4,678 4,678 2,181 3,830 3,830 1,983
Mean (dep. var.) 0.5286 0.0648 0.7464 0.5253 0.0386 0.7262
s.d. (dep. var.) 0.4992 0.2461 0.4351 0.4994 0.1928 0.4460
p-value for Joint Signicance 0.9186 0.0614 0.0104 0.972 0.9225 0.7271
Note: Standard errors are clustered at the village level (*** p<0.01, ** p<0.05, * p<0.1). Only rural sample of wave 2 is used while dropping Assam and Arunachal Pradesh from
the sample. Other individual controls include dummies for religion of the household head, mother age and relationship of the mother with the household head. The specications
also include interview month by year xed eects, state by birth month by birth year xed eects, 3 dummies for population size in the village and 4 dummies for type of main
irrigation in the village.
108
Table 3.13: Community Determinants of Birth Weight
Birth Weight Low BW Very Low BW
General Village Characteristics
Quartic root of hectares of irrigable land in the village -0.0385* 0.0190 0.00228
(0.0219) (0.0134) (0.00432)
Village is electried 0.124 -0.0487 0.0423
(0.172) (0.125) (0.0330)
Village has underground drainage -0.118 -0.0243 -0.0144
(0.233) (0.146) (0.0259)
Zero epidemic in last one year -0.0335 0.0270 -0.00355
(0.0677) (0.0435) (0.0139)
Distance from nearest town (in kms) 0.00186 -0.000874 -0.000907*
(0.00247) (0.00160) (0.000466)
Distance from nearest transport facility (in kms) 0.00408* -0.00240* -0.000257
(0.00237) (0.00127) (0.000334)
Distance from all weather road (in kms) -0.000424 0.000130 -3.89e-05
(0.000918) (0.000642) (0.000196)
Distance from nearest school (in kms) -0.0156 -0.00313 0.00257
(0.0170) (0.00670) (0.00182)
Health Facilities
Village has:
Private Doctor -0.131* 0.109** -0.00505
(0.0711) (0.0445) (0.0145)
Visiting Doctor -0.0869 0.0673 0.00156
(0.0634) (0.0415) (0.0121)
Village Health Guide 0.0543 -0.0274 0.00789
(0.0740) (0.0488) (0.0128)
Traditional Attendant 0.204*** -0.0801* -0.0221*
(0.0654) (0.0424) (0.0127)
Mobile Health Unit 0.00692 -0.0234 -0.0278
(0.0915) (0.0561) (0.0174)
Number of Health/Family Welfare Camps last year -0.00554 0.00242 0.00134
(0.0120) (0.00565) (0.00232)
Distance from Nearest Government Health facility (in kms) 0.0150 0.00283 -0.00219
(0.0103) (0.00556) (0.00142)
Distance from Nearest Private Health facility (in kms) -3.38e-06 -0.000373 0.000256
(0.00286) (0.00167) (0.000363)
Government Schemes
Number of beneciaries
Training Rural Youth for Self-Employment (TRYSEM) 0.000682 -0.00253 -0.000461
(0.00353) (0.00225) (0.000739)
Employment Guarantee Scheme (EGS) -0.00107 -0.000623 6.32e-05
(0.00225) (0.00144) (0.000388)
Development of Women and Children of Rural India (DWARCA) 0.000991 0.00130 0.000220
(0.00234) (0.00158) (0.000509)
Indira Awas Yojana (IAY) -0.00145 0.00148 -0.000172
(0.00196) (0.00137) (0.000295)
Observations 1,337 1,337 1,337
Mean (dep. var.) 2.7810 0.2289 0.0202
s.d. (dep. var.) 0.6878 0.4203 0.1407
p-value for Joint Signicance: General Village Characteristics 0.2709 0.3603 0.6215
p-value for Joint Signicance: Health Facilities 0.0223 0.0306 0.3396
p-value for Joint Signicance: Government Schemes 0.9257 0.5701 0.897
Note: Standard errors are clustered at the village level (*** p<0.01, ** p<0.05, * p<0.1). Only rural sample of wave 2 is used while
dropping Assam and Arunachal Pradesh from the sample. Other individual controls include dummies for religion of the household
head, mother age and relationship of the mother with the household head. The specications also include interview month by year
xed eects, state by birth month by birth year xed eects, 3 dummies for population size in the village and 4 dummies for type
of main irrigation in the village.
109
3.8 Appendix
Table 3.14: Likelihood of Male Child and Maternal Height Relationship in
Other Countries
Ethiopia
South Rest of Ethiopia
Mother's Height less than Median Height -0.0301 0.00607
(0.0278) (0.0253)
Observations 2110 2995
Mean (dep. var.) 0.5128 0.5142
s.d. (dep. var.) 0.5000 0.4999
Note: Robust standard error in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Other
controls include 3 dummies for mother education with zero or missing education being the
omitted category, mother age and age square, dummy if the residence is in urban location,
interview month by year xed eects and region by birth year by birth month xed eects.
110
Chapter 4
The Unintended Consequences of the Village
Midwife Program in Indonesia
1
4.1 Introduction
It is well documented that in utero environment and nutrition have important im-
plications for human and health capital formation of an individual (see Almond and
Currie (2011) for a review). Exogenous environmental and nutritional shocks to moth-
ers are used as identication strategies to get the relevant causal estimates. However,
less is known if the composition of the quality of children changes at birth because of
selection in live birth, due to those shocks.
2
This issue is of interest on its own because
of two reasons: rstly, this has direct implications on human capital formation due
to in utero shocks; secondly, this also has serious implications for behavioral models
which examine parental response to in utero shocks. Economists are long interested
in whether the investment decisions by parents are compensatory or complementary
to such shocks (Almond and Mazumder, 2013). If changes in quality of children at
birth occur due to these shocks, it would imply that selection in live birth due to these
shocks should be directly incorporated in the associated empirical modeling exercise.
1
Jointly written with Riddhi Bhowmick.
2
The quality of children at birth can be thought as their birth endowments. Studies mostly
consider birth endowment as uni dimensional and take birth weight as a key measure (Almond and
Mazumder, 2013). Ahsan and Bhowmick (2015a) discuss some other measures of birth endowments
to be birth size as well as gender of the child.
111
There is a substantial biological, medical, and epidemiological literature which nd
strong associations between maternal health (both short and long term nutritional
status) and pregnancy outcomes.
3
Eriksson et al. (2010) point out that maternal
health and nutrition matter more for male than female births. Given the weight of
placenta, the authors argue, boys tend to be bigger at birth than girls. As a result,
the boy's placentas are more ecient but may have less reserve capacity. In situa-
tions of environmental stress, this confers girls a pre-birth survival advantage. Not
surprisingly, a report by MacDorman and Kirmeyer (2007) nds that fetal mortality
for male fetuses are 7 percentage higher than female fetuses in 2005 based on data
from USA.
4
Moreover, it has been widely noted that males experience more neo-natal
mortality than females (Bhaumik et al., 2004), and health environment during preg-
nancy largely in
uences survival during the neo-natal period and beyond (Almond et
al., 2011).
Despite it's signicance, there are not many quality causal studies in developing
countries which directly link maternal nutritional status to fetal health quality. In
this paper, we try to provide one of the rst causal estimates of the same. For this
purpose, we consider the Indonesian Midwife (Bidan Desa) program which was intro-
duced in the late eighties to improve the high maternal mortality rates in Indonesia.
It is specically aimed at improving the health of reproductive age women. Using
a dierence in dierence strategy, Frankenberg and Thomas (2001) show that the
program was successful in increasing the BMI (body mass index) of the women of re-
productive age, especially for women who had lower initial BMI.
5
Furthermore, they
3
These outcomes include live birth, miscarriage, still birth, gestational age and ectopic pregnancy.
4
Using data from USA, a similar report (MacDorman et al., 2007) nds that male fetuses are 9
percentage higher than female fetuses.
5
It should be noted that BMI is a non-linear measure of health status. An increase in BMI for
those who have low BMI implies an improvement in health status.
112
calculate the estimated positive impact of gaining a village midwife on birth weight
to be 80 grams.
6
The novelty of this paper is that it tests the epidemiological insight on male
fragility in utero with the help of the provision of a Village Midwife Program. More
specically, we calculate the causal eects of Bidan Desa on the likelihood of male
births, along with other birth outcomes of male children relative to female children
by using the timing of the midwife placement in the community as our identication
strategy.
7
Using a dierence-in-dierence strategy and all four waves of the Indone-
sian Family Life Surveys (IFLS), we document three main results. Firstly, we nd
that placement of a midwife in a community leads to an increase in the likelihood of
a male birth by 4 percentage points at any birth order. The results are even stronger
for rst order births. Secondly, the increase in male births due to midwife program
is more pronounced among mothers who have lower education. Thirdly, the midwife
program placement is associated with lower birth weight for male children but not
for female children. We also nd, following the midwife program, male children ex-
perienced higher infant mortality than female children. We do not nd any fertility
selection or any change in gender-specic reporting bias due to midwife program.
Thus, the eect of midwife program placement on male births and birth outcomes
can be regarded as causal. Based on these three results, we conclude that in utero
shocks may not have same the eect across the socio-economic status and may alter
the composition of quality of children born.
6
They estimate the short run eect of a village midwife program, while this paper also considers
long term eect.
7
We only consider "reduced form" eect because the program may have spill over eect on other
family members including adult males (see Frankenberg and Thomas (2001).
113
The study makes some key contributions to a number of literature. Firstly, this
paper makes a basic methodological contribution in early life studies of developing
countries.
8
In order to describe underlying mechanisms linking in utero shocks and
later life outcomes, many studies implicitly assume a monotonic relationship between
these shocks and birth endowments without formally testing them.
9
Our results un-
derscore the foremost need to account for selection into live birth before interpreting
the associations as causal, especially for better understanding the economics of hu-
man capital formation. Our recommendations are in line with similar studies in the
literature on developed countries that the empirical economic modeling has to inte-
grate evidence and techniques from other elds like reproductive biology, sampling
design, demography and the like.
Secondly, our ndings also contribute to the literature of gender discrimination.
Under the assumption that sex of the child is random at birth, the mean dierence in
outcome between boys and girls can be attributed to gender discrimination (Hamoudi
and Nobles, 2014; Schultz, 2007; Rosenblum, 2013). For such studies, researchers typ-
ically rely on important, yet untested assumptions about dynamics of selection into
live birth to draw inferences about gender dynamics from the gender/outcome asso-
ciations (Hamoudi and Nobles, 2014). The results in our paper add to the growing
evidence that sex of the o-spring at birth is not random, especially in developing
countries where socio-economic background can aect the sex specic survival rates
of the fetus during gestation (Ahsan and Bhowmick, 2015a). Moreover, the likelihood
of a male child being positively associated with maternal nutritional status implies
8
In the context of the US, selection into live birth may account for a large proportion of the
observed association between ospring sex and divorce which was earlier thought to be due to son
preference (Hamoudi and Nobles, 2014).
9
The reader may refer to the studies by Maccini and Yang (2009), Shah and Steinberg (2013)
etc.
114
the estimate of mean dierence | for example, child investments between boys and
girls | may have a bias. In the absence of son preference, if parents compensate for
poor birth endowments, then our results imply the direction of bias will be positive.
There is another interesting (and paradoxical) implication of our nding that more
male children are born to less-educated mothers due to the Village Midwife Program.
Since son preference is not strong in Indonesia, this may imply \gender-reversal" in
human capital formation in future as there is ample evidence showing inter gener-
ational transmission of the same.
10
In fact, gender reversal may simply happen if
parents try to reinforce birth endowments. Since sons are born with poor birth en-
dowments (as measured by birth weight), parents may invest less in them in absence
of any son preference. This is a serious possibility as most of the recent studies on
parental response to birth endowments nd parents to engage in a reinforcing behav-
ior (see Almond and Mazumder (2013) for a review).
Lastly, this study provides evidence on how a well-intended health intervention
for mothers may entail adverse health outcomes for the o-springs. One usually ex-
pects betterment in maternal health will help improve child health in all dimensions.
Almond et al. (2011) raise the possibility of a paradoxical situation. They argue the
composition of births could change as a result of the same forces which improve fetal
health. Because of improved in utero environment, marginal fetuses have improved
survival chances and thus will have negative compositional consequences on dierent
birth outcomes (in our case, birthweight for male children). This the estimated ef-
fects of any such program which improves maternal health on a variety of fetal quality
measures could bias downward. To our knowledge, we are the rst to report such un-
intended consequences of any maternal health intervention program. We hypothesize
10
See the studies by Bhalotra and Rawlings (2013, 2011) and the literature cited there.
115
such perverse consequences should be more common in developing countries as there
are fewer coping mechanisms for smoothing fertility and health behavior, as compared
to developed countries
11
.
The rest of the paper is organized as follows. Section 2 gives a background on
the Village Midwife Program in Indonesia. Section 3 provides a literature review.
Section 4 discusses the survey data and measurements we use. Section 5 sets out the
empirical strategy. Section 6 discusses the results. Section 7 is concluding discussion.
4.2 Village Midwife Program in Indonesia
The primary objective behind introduction of the Village Midwife Program in the
late 80s, as stated by Indonesian Ministry of Health (DepKes), was to improve ma-
ternal health with a special emphasis on reducing maternal mortality in rural areas
(Frankenberg and Thomas, 2001). According to a World Bank report, the initiative
expanded rapidly throughout Indonesia, from nearly 5,000 workers in 1987 to 80,000
in 2009 (Weaver et al., 2013). Weaver et al. (2013) report the percentage of IFLS
communities with a village midwife increased from 9.6 to 46.3 percent between 1993
and 1997 which rose to around 60 percent in 2007. Figure 4.1 shows the number of
IFLS villages that have gained a midwife from 1989 to 2007.
At the beginning, village midwives were typically recruited from nursing programs
and received one additional year of training on midwifery. Later, this was changed
to require that village midwives attend a three-year midwifery academy. The village
midwives were largely in their early twenties and single at the entry level and were
usually placed in their province of origin. Initially midwives' practices were stationed
11
Indeed, a number of studies on developed countries which looked for such paradoxical program
impacts of in utero exogenous shocks, did not nd any (Almond et al., 2011, 2009)
116
at village delivery post; home of the village leader acted as the delivery post if no such
station existed in the village (Weaver et al., 2013). After assignment to a community,
village midwives were guaranteed a government salary for at least three years. They
should engage in public practice during normal working hours, while they can have a
private practice after hours. The goal was to help the midwives sustain their practice
without a government salary, once the government contract ends (Frankenberg and
Thomas, 2001).
The primary goals of a village midwife include aecting reproductive health of
women by providing a variety of health and family planning services. She should
work with traditional birth attendants, and act as a link to formal health care delivery
systems (eg by referring complicated obstetric cases to health centers and hospitals).
Unlike the formal health delivery systems, a midwife should pro-actively seek out
for patients and visit their homes. As is evident from a number of studies, she acts
as a general health resource in a community | advising dierent health promoting
behaviors including sanitation and nutrition, dispensing medications, immunizations,
well-child care, and a variety of acutecare services such as sick-patient visits, the
administering of antibiotics, and attending to wounds etc (Frankenberg and Thomas,
2001; Weaver et al., 2013).
4.3 Literature Review
A number of studies examine the impact of Village Midwife Program on a variety of
health outcomes. Frankenberg and Thomas (2001) nd communities receiving village
midwives between 1993 and 1997 are associated with a signicant increase in body
mass index in 1997 relative to 1993 for women of reproductive age. Other successes in-
117
clude higher usage of antenatal care during rst trimester of pregnancy among women
with lower levels of education, increases in receipt of iron tablets, and less reliance on
traditional birth attendants for birth delivery (Frankenberg et al., 2009), increased
usage of injectable contraceptives while decreased incidence of oral contraceptive and
implant use (Weaver et al., 2013). Children exposed to the program were also bene-
ted in terms of better nutrition status (as measured by height for age) (Frankenberg
et al., 2005).
The literature on eects of maternal health intervention programs on sex ratio
(as measured by number of males per 100 females) is scant, especially in developing
countries. In particular, we are not aware of any such study on Village Midwife Pro-
gram. But in general, there are ample evidences from epidemiological studies provide
which show exogenous variations of maternal conditions can impact the likelihood of
a male birth. In particular, these studies have focused on the impacts of intrauterine
nutrition on sex ratio (as measured by number of males per 100 females) along with
other infant and adult outcomes. For instance, Song (2012) nds evidence in favor of
adaptive sex ratio hypothesis for Chinese Great Leap Forward Famine. Some other
studies which nd positive evidence include (Fukuda et al., 1998) for earthquake in
Japan, Zorn et al (2002) for war in Slovenia, Catalano et al (2006) for terrorist attack,
Valente (2015) for civil war in Nepal, Sanders and Stoecker (2015) for air pollution,
Williams and Gloster (1992) for food availability, Shifotoka and Fogarty (2012) for
prevalences of HIV and tuberculosis etc.
Unlike the impact maternal health interventions on sex ratio, the impact of ma-
ternal health interventions on birth weight and infant mortality is somewhat better
examined. For the Village Midwife Program, Frankenberg and Thomas (2001) nd
118
birth weights are greater in communities after introduction of midwives than before.
In the context of developed countries, there are a number of studies in medicine and
economics which examine impacts of dierent kinds of interventions on birth weight
and neonatal mortality. For instance, Almond et al. (2011) nd pregnancies exposed
to Food Stamp Program in US three months prior to birth yielded deliveries with
increased birth weight, with the largest gains at the lowest birth weights. Pollution
and temperature are also found to aect these outcomes. Currie et al. (2009) nd
a one unit change in mean carbon monoxide during the last trimester of pregnancy
increases the risk of low birth weight by 8 percent, and a one unit change in mean
carbon monoxide during the rst two weeks after birth also increases the risk of infant
mortality by 2.5 percent relative to baseline levels. Similarly, Desch^ enes et al. (2009)
report exposure to extreme temperature in utero associated with lower birth weight,
especially in 2nd and 3rd gestational trimesters. Both of these studies are based on
the US. There is also a growing literature which nd similar evidence from a number
of developing countries (Burgess et al., 2013; Arceo et al., 2015).
In spirit, our study is closest to the study by Hern andez-Juli an et al. (2014).
Using the Bangladesh famine of 1974 as a natural expeiment, they nd women who
were pregnant during the famine were less likely to have male children and children
exposed in utero to the most severe period of the famine were more likely to die as a
neonate. Conditional on being a live birth, the male children are also more likely to
die as a neonate if exposed to famine in utero. We complement this study by focusing
on a specic positive intervention on maternal health, viz. provision of midwives.
119
4.4 Data and Measurements
We use all four waves of the Indonesian Family Life Survey (IFLS) which were con-
ducted in 1993, 1997, 2000 and 2007 respectively. At the time of the rst survey 1993{
94, the sample drawn was representative of 83% of the population residing in 13 out of
27 provinces in Indonesia. Within each of these 13 provinces, the enumeration areas
(EA) were randomly selected for inclusion in the nal survey. In the rst wave, 7,224
households were interviewed and detailed individual level information were collected,
including the age and education of the household members. The later waves sought
to follow up on the same household and the re-contact rates of households from rst
wave were 94.4%, 95.1% and 93.6% in the second, third and fourth waves respectively.
For the purpose of this study, we use data from two sources in the survey: 1)
the pregnancy histories of the women of reproductive age which include information
on birth date, birth outcome, gender of the child, age of the mother, birth weight
report, child alive or not, if not alive{age of death, birth weight of each of the births
the women ever had 2) community level information on various community level
infrastructures as well as information about midwife from the village head. The
placement-time information of the midwife is only available from the second wave of
the IFLS.
One limitation of the data is that the placement-time information of the midwife
is only available for communities which were randomly selected in the rst wave.
12
The attrition of the households due to migration could be a concern, as Thomas
12
312 communities/enumeration areas were randomly selected for the rst wave of the IFLS.
However, some of the sampled households and respondents later moved to other communities.
120
et al. (2012) show that the migrants could be dierent from non-migrants in terms
of observable characteristics. However, Weaver et al. (2013) argue that conditional
on select individual and community characteristics, the receipt of the midwife of a
community is not signicantly related to woman's migration out of the study com-
munities or loss to follow up over study period. We discuss more about this issue in
the empirical strategy.
The main analysis of this study is based on the birth sample obtained from last
ve year pregnancy histories of the ever married women in the sampled households.
If the follow up wave took place within ve year, we restrict the pregnancy history
up to the last wave year to avoid double counting. Since the fourth wave took place
7 years after the third wave, we also separately consider birth sample obtained from
last seven years for continuity. Table 4.1 presents the summary information of the
birth sample for pregnancy histories restricted up to last ve years. It shows that
about 51 percentage of all the live births is male. The women in the sample on an
average has more than primary education, as the mean years of education is 6.64.
The mean age at birth is 31 years. The mean birth weight is 3,163 grams. The birth
weight information is not available for all the births, which is another limitation of
the data.
4.5 Empirical Strategy
The empirical strategy section is divided in two subsections. The rst subsection
describes the empirical challenges, and the second subsection describes the empirical
framework for analyzing the impact of the midwife program on probability of a male
birth, and birth weights by gender.
121
4.5.1 Empirical Challenges
Endogenous Program Placement | Evaluation of a health or | for that matter |
any intervention is often complicated, as the outcomes of interest are aected by the
characteristics of individuals, households, and communities. A government may tar-
get a set of population where the program is likely to be more successful, or on a set
of population which is badly in need of such programs. Under these circumstances,
an evaluation of a government initiated program is complex, as the programs are not
randomly allocated.
As discussed earlier, the midwife program was initiated to lower the rate of ma-
ternal mortality. Therefore, the placement of the program could depend on the
community level observable and unobservable characteristics. In fact, Frankenberg
and Thomas (2001) show a community's levels of poverty and remoteness in
uenced
whether it received a village midwife. For causal identication of the eect of midwife
program, we exploit the variation in timing of the midwife program and include com-
munity xed eects in all regression equations. The application of community xed
eects absorbs the community specic observable as well as unboservable character-
istics. On the other hand, the timing variation allows us to compare the outcomes of
interest for the same community before and after the program.
Selective fertility, Migration and Mortality Attrition | One concern in evaluating
the impact of a health intervention program on pregnancy outcomes is that the health
intervention may aect fertility selection. Frankenberg and Thomas (2001) document
that midwife also provided contraceptives to reproductive age women. Moreover, they
may also provide suggestions regarding family size or birth timing. Therefore, it is
122
necessary to examine whether the placement of the midwife program has changed the
observed parental characteristics of the birth sample or changed the parental fertility
behavior.
As far as migration is concerned, calculating program eects is dicult even in a
randomized control design study if the migration pattern diers between the treat-
ment and control groups. In our set up also, the migration pattern between areas with
midwife and areas without a midwife can dier. However, Weaver et al. (2013) do
not nd any signicant association between migration and midwife placement after
controlling for individual and other community level observables.
Another important concern is that the provision of a midwife program may aect
the maternal mortality rate in the program communities. Recall that the pregnancy
outcomes are obtained from the pregnancy histories of women respondents of the sam-
pled households, only if the women are alive during the survey year. If the provision of
a midwife aects the mortality rates of women dierently for dierent socio-economic
characteristics (SES), the observed parental characteristics of the birth sample will
change. Change maternal mortality rate is unlikely to aect change in sample char-
acteristics, as maternal mortality is an extreme event, and therefore, such change is
going to very small.
13
Moreover, the subsequent waves of the IFLS were conducted
in short intervals. The use of short recall period of pregnancy histories reduces any
concern that provision of a midwife may have aected the parental characteristics of
the observed birth sample in the program areas.
13
Based on several studies, Frankenberg and Thomas (2001) document that Indonesia experienced
390 to 650 maternal deaths per 100,000 live births in early '90s.
123
To examine whether parental characteristics of the birth sample have changed
due to the provision of a midwife, we analyze the following variables with the pro-
gram placement: maternal years of education, mother age at birth, and number of
total live births. There is a large literature which documents better educated and
healthier mothers tend to have healthier children (Rosenzweig and Schultz, 1982b;
Thomas et al., 1990, 1991; Ahsan and Bhowmick, 2015a). If such mothers are more
likely to give birth after the provision of a midwife in their communities, we may
observe an increase in the probability of a male birth only because of the composition
of mothers while the program had no real health eect on mothers.
14
We examine
this possibility by looking at mother education in this paper. Similarly, change in
age at birth may capture complications regarding pregnancies at certain age, along
with other unobservable dimensions of the mothers. Moreover, number of live births
may re
ect change in parental change in quality vs quantity trade o. It should be
noted that any change in birth outcome, attributed to change in fertility behavior of
certain sub-sections due to the program still counts as a program eect | we only
have to be careful regarding the mechanisms of impact.
Trend dierence | As in any Dierence-In-Dierence (DID) research design, a
valid concern is that treatment and control groups may face dierent time trends.
Often the government provides multiple facilities or \treatments" to specic commu-
nities. In that case, DID may overestimate the impact of a program if the placement
of the other programs are not taken into account. The richness of the IFLS data
allows us to consider various time varying community characteristics, which may
correlate with the provision of a midwife in a community and may also aect the
14
Ahsan and Bhowmick (2015a) report a higher likelihood of male child at rst birth for taller
mothers in areas of India, where infant mortality rates are high.
124
pregnancy outcomes of the mother. Following Frankenberg and Thomas (2001), we
include these time varying community level variables in our estimating equations:
paved road status, urban status, public telephone, and distance to nearest health fa-
cility. We additionally include time varying community characteristics like distance to
market, community electricity status, distance to district capital center, and number
of health posts.
4.5.2 Empirical Framework
To examine the impact of the midwife program on likelihood of a male birth, we
estimate the following linear probability model for child i who is born in month m
and year t and whose mother lives in community j:
malec
ijmt
=
1
Treated
jt
+
2
X
ijmt
+
3

jt
+
m

t
+
j
+
ijmt
(4.1)
The dependent variable malec takes a value of 1 if the child is male and 0 other-
wise.
1
, the co-ecient for Treated captures the impact of the midwife program on
the probability of a male birth. We have considered two variants for Treated. In our
primary regressions, this variable takes a value of 1 if the community has a midwife in
the year of birth of the child, else takes a value of 0. In another variant, it is dened
as years of exposure of the community to the program, so that the heterogeneous
impact due to program intensity can be captured. The interaction term,
m

t
represents birth month by birth year eects, which controls for seasonality, monthly
prices, and dierent other time varying observables and unobservables. The inclusion
of
j
controls for community level xed observables and unobservables. X
ijmt
is a set
of mother level observables such as mother education and mother age at the time of
survey. These variables are dened as spline functions.
jt
includes set of time vary-
125
ing community observables such as paved road status, urban status, public phone
status, distance to market, distance to district capital center, number of health posts,
community electricity status and distance to nearest health facility. The distance to
nearest health facility and provision of a midwife program may highly correlate with
each other, therefore we regress equation (1) by both excluding and including the
distance to nearest health facility.
To examine the impact of the midwife program on birth weight, we estimate the
following regression equation for childi who is born in monthm and yeart and whose
mother lives in community j:
birthweight
ijmt
=
1
Treated
jt
+
2
Post
t
Treatedmalec +
3
X
ijmt
+

4

jt
+
m
+malec
j
+malec
t
+
ijmt
(4.2)

2
is the co-ecient of interest. In a variant model, we examine the impact of
the midwife program on low birth weight, where low birth weight is dened by births
which are less than 2500 grams.
Moreover, we analyze whether males are more likely to experience infant mortal-
ity compared to females following the midwife program. If poor quality male fetuses
are more likely to survive till live birth, they might experience more infant mortality.
The dependent variable takes a value of 1 if the child experience mortality before age
1 and 0 otherwise.
Recall that the birth weight information is not available for all births. One possi-
bility is that the provision of a midwife may have changed gender-specic birth weight
126
reporting. In that case, the mechanism of impact of the midwife program on birth
weight is change in reporting and not any biological change among mothers. If
2
is signicantly dierent from zero, then we should be worried that the dierence in
association between provision of a midwife program and birth weight, across gender
could be driven by reporting bias.
4.6 Results
4.6.1 Male Births
We report the eects of Village Midwife Program on likelihood of male births in Ta-
ble 4.2. In the rst three columns, we consider all births in last 5 years from the date
of interview, while we consider all births between 1987 and 2007 in the remaining
columns. In each of these two sets of columns, we rst include birth month-by-birth
year xed eects and individual controls, then add community controls, and nally
add distance from nearest health facility. As expected from our discussion at the
beginning, the impact is quite positive and signicant. The impact is quite large |
a mother residing in a community with village midwife is 4 percentage points more
likely to give birth to a male child. The results are extremely stable across all the
specications and samples. We repeat the same exercise in Table 4.3, but for rst
births only. The point estimates are usually larger, while the statistical signicance
is lost due to smaller sample size. Similar to the ndings of this paper, Sanders and
Stoecker (2015) nd that the Clean Air Act Amendments in 1970 in USA led to a
similar increase in likelihood of a live birth being a male in the aected counties.
Next we show the results by mother education in Table 4.4, while only considering
births from last 5 years of interview. The rst three columns correspond to the moth-
127
ers with education level of primary or less while the last three columns correspond
to mothers with higher levels of education. We nd mothers who are primary or less
educated are more heavily impacted due to the program in terms of likelihood of male
birth. These mothers are 5 percentage points more likely to give birth to a child due
to the program. The impact is more than halved (and also insignicant) for mothers
who are high educated.
We also repeat the exercises above, but for a continuous measure of treatment
which in this context is the years of exposure to the program. This is dened as the
number of years a community has received a midwife, till the year of birth of the
child. The results are reported in the appendix, Table A1, Table A2, and Table A3.
The direction of relationship is unchanged. In fact, the results are usually statistically
more signicant for the continuous treatment exposure.
4.6.2 Birth Weight and Infant Mortality
We show in Table 4.5, the impacts of the program on birth weight. The rst three
columns contain the results for male children while the nal three columns have the
same for female children. For male children, we nd the eect of the program on birth
weight is negative. The eects vary a little across specications, but are usually sta-
tistically signicant at 10 percent levels. The results, combined with the results from
earlier tables imply increase in male children have largely come about from increased
likelihood of survival of marginal fetuses which are of lower quality. The program, on
the other hand, does not seem to impact the birth weights for female children.
In Table A4, we show the results for continuous treatment measures. Similar to
128
binary treatment specications, the program eect on birth weight for male children
is negative and statistically signicant at 10 percentage points. There is a slightly
positive eect for female children, but it is statistically insignicant.
Similar to the results presented above, we nd that males are more likely to be
born with low birth weights compared to females following a midwife program (Ta-
ble A7 and Table A8). Although these coecients are not statistically signicant,
they are consistent with the ndings presented in Table 4.5 and Table A4.
While the current study nds negative eect of a midwife program on birth weights
for male children, Frankenberg and Thomas (2001) nd positive eect of the program
on birth weights. This study diers from Frankenberg and Thomas (2001) in two
key ways. Firstly, Frankenberg and Thomas (2001) analyze short run eect of the
program; in contrast, our paper considers the longer term eects for the same. In-
deed, we nd that longer exposure to the midwife program leads to an increase in
the likelihood of a male birth (Table A1 and Table A2) and negatively aects birth
weight of the male children (Table A4). Secondly, Frankenberg and Thomas (2001)
do not explore the impact of the midwife program on birth weight by gender.
We also consider whether males are more likely to experience infant mortality if a
community receives a midwife. We don't nd any male-specic dierential in infant
mortality if we consider the binary treatment measure (Table A5). However, the
likelihood of a child to die as an infant, following the placement of a midwife in the
community is higher for a boy than a girl (Table A6). This is again consistent with
our earlier ndings in Table A1, where we have shown that years of exposure to the
Midwife program to be positively associated with likelihood of male birth.
129
4.6.3 Threats to Identication
Motivated by the ndings of Ahsan (2015), Brown and Thomas (2011), and Buckles
and Hungerman (2013), we check for selective fertility behavior as well as self-selection
of mothers to the program. If babies born to mothers during the program are born to
dierent families than those in absence of the same, then the mechanisms identied
behind the observed eects of the program would be misleading. We also check if
there is any change in gender-specic reporting due to the program in a similar spirit.
Fertility
We check for selective fertility by using the same regression specication for Table 4.2
with total number of births in a community as the dependent variable.
15
Results
are reported in Table 4.6. First three columns consider births in last 5 years from
the date of interview while all births between 1987 and 2007 are considered in the
remaining columns. The results indicate no strong evidence of selective fertility, and
it remains true in most rigorous of specications.
Mother Characteristics
Similar to the approach in Buckles and Hungerman (2013), we also test for parental
selection into the program. For that purpose, we check if presence of a midwife
program can explain maternal education and mother age at birth of the sample. We
report the results in Table 4.7 and Table 4.8. We don't nd evidence of any eect
of the program on either of the mother characteristics. Not only the coecients are
statistically insignicant, but also their magnitudes are very small.
15
Mother education and mother age are averaged within a community.
130
Gender-specic Reporting Bias
Our nal check concerns any change in gender-specic reporting of births due to the
program. Table 4.9 reports the ndings. Although, there appears to be a general
increase in reporting due to the program, but we don't nd the increase to be gender-
specic. In other words, the reporting has increased uniformly across both gender.
4.7 Conclusion
In this paper, we estimate the impacts of the Village Midwife Program in Indonesia
on two measures of birth endowment- likelihood of a male birth and birth weight.
Using all four waves of the Indonesian Family Life Surveys (IFLS) and employing
a Dierence-In-Dierence empirical strategy, we show that a woman of reproductive
age is 4 percentage points more likely to give a male birth due to the program which
is nearly 10 percent of the mean. This is consistent with the epidemiological studies
which nd improved nutritional status of mothers leads to higher survival chances of
male fetuses to term. It is well-documented that this program is successful in raising
the body mass index of reproductive age women. The impact we estimate is largely
driven by mothers with primary education and below. However, we nd there is a de-
crease in birth weights associated with the program, while there is no such impact for
female children. Moreover, infant mortality for males are higher compared to females
following the program. We conclude that increased chance of survival for marginal
fetuses is the main reason for this paradoxical impact. The observed relationships
could be interpreted as causal as our estimates are robust to several strategies which
attempt to correct for a number of unobserved factors, driving access to midwives
and birth outcomes together.
131
Our results have interesting implications for economic and health policy. Several
studies nd girls to be particularly vulnerable to
uctuations in economic and envi-
ronmental conditions (Maccini and Yang, 2009; Rose, 1999), possibly due to gender
bias and imperfect consumption smoothing, so that interventions like social insurance
schemes, public health investments, or food security programs can act as eective
shields for this subgroup with respect to health consequences. Our nding that more
male children with inferior birth weights may be born due to such program brings
two opposing consequences, depending on parental responses to initial birth endow-
ments, if degree of son preference is not high (true for Indonesia). If parents try to
compensate the initial inferior endowments of male children, then a simple mean com-
parison of parental investments between boys and girls may mistakenly come across
as gender bias towards boys. On the other hand, if parents try to reinforce the birth
endowments which seems to be more evident in the empirical literature (Almond and
Mazumder, 2013), then a gender-reversal may ensue in human capital formation.
One key point from our results is that interventions with the potential to improve
maternal health before or during pregnancy can have serious distributional impacts
for live births which may lead to erroneous conclusions. It is important to carry out
investigations on dynamics of selection into live birth, in general and sex-specic live
births, in particular for any line of literature, related to human capital formation, de-
mography, or early life studies. Insights from reproductive biology and epidemiology
should be incorporated into the standard behavioral modeling structure of economists
as well as in sampling and survey designs.
An analysis of the overall welfare impact of any such intervention should include
regular economic outcomes (like consumption, income etc), along with health and nu-
132
tritional status, mental health and subjective welfare (Thomas et al., 1990; Adhvaryu
et al., 2014). We analyze the eect of the Village Midwife program on mainly one
aspect of human capital formation which is health status at birth. Future research
on the eectiveness of this program should focus on these other important outcomes
to arrive at a more holistic conclusion about it's ecacy.
133
4.8 Figures
Figure 4.1: Allocation of Midwives
134
4.9 Tables
Table 4.1: Summary Statistics
Mean S.D. N
Male Child 0.515 0.500 8420
Mother Years of Education 6.641 3.577 8420
Mother Age at Birth 31.17 6.678 8420
Birth Weight (in kilos) 3.163 0.592 6251
135
Table 4.2: Impacts of The Village Midwife Program on Likelihood of Male
Birth
Birth Cohort:1989-99 & 2003-07 Birth Cohort: 1987-2007
(1) (2) (3) (4) (5) (6)
Treated(=1) 0.040* 0.044* 0.043* 0.035* 0.040** 0.040**
(0.022) (0.022) (0.023) (0.020) (0.020) (0.020)
Distance from Health Facility -0.001 -0.001
(0.007) (0.005)
R2 0.065 0.066 0.066 0.056 0.058 0.058
Observations 8420 8394 8394 11003 10969 10969
Mean of Dependent Variable 0.515 0.515 0.515 0.517 0.517 0.517
Birth Month Birth Year FE Y Y Y Y Y Y
Individual Controls Y Y Y Y Y Y
Community Controls N Y Y N Y N
Original IFLS Community Y Y Y Y Y Y
Note: Standard errors are clustered at the community level (*** p<0.01, ** p<0.05, * p<0.1).
The rst three columns report all births from last 5 years while the last three columns consider
all births from 1987 to 2007. The dependent variable MaleChild takes a value of 1, if the child
is male, and 0 otherwise. The variable Treated takes a value of 1, if midwife is present during
the birth year of the child, and 0 otherwise. The individual controls includes mother years of
education (splines with knots at 6,9 and 12) and mother age at survey (splines with knots at
20, 25, 30, 35, 40 and 45). Community controls include time varying changes at the community
level: paved road status, electricity status, number of health posts, urban status, public phone
status, distance to market, distance to district capital center, and distance to nearest health
facility.
136
Table 4.3: Impacts of The Village Midwife Program on Likelihood of a
Male Child at First Birth
Birth Cohort:1989-99 & 2003-07 Birth Cohort: 1987-2007
(1) (2) (3) (4) (5) (6)
Treated(=1) 0.053 0.058* 0.056 0.036 0.044 0.042
(0.035) (0.035) (0.035) (0.031) (0.031) (0.031)
Distance From Health Facility -0.006 -0.006
(0.014) (0.009)
R2 0.113 0.114 0.114 0.102 0.104 0.104
Observations 4584 4578 4578 6142 6134 6134
Mean of Dependent Variable 0.512 0.512 0.512 0.514 0.514 0.514
Birth Month Year FE Y Y Y Y Y Y
Individual Controls Y Y Y Y Y Y
Community Controls N Y Y N Y N
Original IFLS Community Y Y Y Y Y Y
Note: Standard errors are clustered at the community level (*** p<0.01, ** p<0.05, * p<0.1).
The rst three columns report all births from last 5 years while the last three columns consider
all births from 1987 to 2007. The sample is based on rst birth only. The dependent variable
MaleChild takes a value of 1, if the child is male, and 0 otherwise. The variable Treated takes
a value of 1, if a midwife is present in the community during the birth year of the child, and
0 otherwise. The individual controls mother years of education (splines with knots at 6,9 and
12) and mother age at survey (splines with knots at 20, 25, 30, 35, 40 and 45). Community
controls include time varying changes at the community level: paved road status, electricity
status, number of health posts, urban status, public phone status, distance to market, distance
to district capital center, and distance to nearest health facility.
137
Table 4.4: Impacts of The Village Midwife Program on Likelihood of Male
Birth, by Mother Education
Primary and Below Primary Above
(1) (2) (3) (4) (5) (6)
Treated(=1) 0.052* 0.052* 0.052* 0.021 0.026 0.024
(0.031) (0.031) (0.031) (0.037) (0.038) (0.038)
Distance From Health Facility 0.000 -0.006
(0.010) (0.011)
R2 0.119 0.121 0.121 0.130 0.130 0.131
Observations 4598 4581 4581 3822 3813 3813
Mean of Dependent Variable 0.513 0.513 0.513 0.516 0.517 0.517
Birth Month Year FE Y Y Y Y Y Y
Individual Controls Y Y Y Y Y Y
Community Controls N Y Y N Y Y
Original IFLS Community Y Y Y Y Y Y
Note: Standard errors are clustered at the community level (*** p<0.01, ** p<0.05, * p<0.1).
The rst three columns report the results for mothers with education primary and less while
the last three columns report the same for the mothers with education more than primary. The
dependent variable MaleChild takes a value of 1, if the child is male, and 0 otherwise. The
variableTreated takes a value of 1, if a midwife is present in the community during the birth year
of the child, and 0 otherwise. The individual controls mother years of education, and mother
age at survey (splines with knots at 20, 25, 30, 35, 40 and 45). Community controls include
time varying changes at the community level: paved road status, electricity status, number of
health posts, urban status, public phone status, distance to market, distance to district capital
center, and distance to nearest health facility.
138
Table 4.5: Impacts of The Village Midwife Program on Birth Weights (in
Grams), by Gender
Male Birth Weight Female Birth Weight
(1) (2) (3) (4) (5) (6)
Treated(=1) -68.706* -68.682* -63.646* -2.971 -0.446 8.837
(35.519) (35.331) (35.455) (44.254) (45.463) (45.311)
Distance From Health Facility 23.452 33.586*
(17.040) (19.006)
R2 0.166 0.168 0.169 0.160 0.164 0.165
Observations 3712 3709 3709 3466 3462 3462
Mean of Dependent Variable 3188.31 3188.43 3188.43 3103.17 3102.80 3102.80
Birth Month FE Y Y Y Y Y Y
Birth Year FE Y Y Y Y Y Y
Individual Controls Y Y Y Y Y Y
Community Controls N Y Y N Y Y
Original IFLS Community Y Y Y Y Y Y
Note: Standard errors are clustered at the community level (*** p<0.01, ** p<0.05, * p<0.1).
The rst three columns report the results for male children while the last three columns do
the same for female children. The dependent variable birthweight is measured in grams. The
variable Treated takes a value of 1, if a midwife is present in the community during the birth
year of the child, and 0 otherwise. The individual controls mother years of education (splines
with knots at 6,9 and 12) and mother age at survey (splines with knots at 20, 25, 30, 35, 40
and 45). Community controls include time varying changes at the community level: paved road
status, electricity status, number of health posts, urban status, public phone status, distance to
market, distance to district capital center, and distance to nearest health facility.
139
Table 4.6: Testing for Selective Fertility
Birth Cohort:1989-99 & 2003-07 Birth Cohort: 1987-2007
(1) (2) (3) (4) (5) (6)
Treated(=1) -0.052 -0.072 -0.091 -0.052 -0.073 -0.086
(0.104) (0.103) (0.103) (0.100) (0.100) (0.100)
R2 0.300 0.303 0.304 0.289 0.291 0.292
Observations 3671 3661 3661 4820 4807 4807
Mean of Dependent Variable 2.50 2.50 2.50 2.52 2.52 2.52
Birth Year FE Y Y Y Y Y Y
Individual Controls Y Y Y Y Y Y
Community Controls N N N N N N
Original IFLS Community
Note: Standard errors are clustered at the level (*** p<0.01, ** p<0.05, * p<0.1). The de-
pendent variable is total number of live births. The variable Treated takes a value of 1, if a
midwife is present in the community during the birth year of the child, and 0 otherwise. The
individual controls mother years of education (splines with knots at 6,9 and 12) and mother age
at survey (splines with knots at 20, 25, 30, 35, 40 and 45). Community controls include time
varying changes at the community level: paved road status, electricity status, number of health
posts, urban status, public phone status, distance to market, distance to district capital center,
and distance to nearest health facility.
140
Table 4.7: Testing for Selection of Mothers: Mother Education
Birth Cohort:1989-99 & 2003-07 Birth Cohort: 1987-2007
(1) (2) (3) (4) (5) (6)
Treated(=1) 0.176 0.192 0.185 0.153 0.162 0.161
(0.122) (0.122) (0.121) (0.119) (0.119) (0.117)
Distance From Health Facility -0.016 -0.001
(0.040) (0.037)
R2 0.454 0.456 0.456 0.438 0.440 0.440
Observations 8420 8394 8394 11003 10969 10969
Mean of Dependent Variable 6.64 6.65 6.65 6.60 6.61 6.61
Birth Month Year FE Y Y Y Y Y Y
Individual Controls Y Y Y Y Y Y
Community Controls N Y Y N Y N
Original IFLS Community Y Y Y Y Y Y
Note: Standard errors are clustered at the community level (*** p<0.01, ** p<0.05,
* p<0.1). The dependent variable is mother years of education. The variable
Treated takes a value of 1, if a midwife is present in the community during the
birth year of the child, and 0 otherwise. The individual controls include mother age
at survey (splines with knots at 20, 25, 30, 35, 40 and 45). Community controls
include time varying changes at the community level: paved road status, electric-
ity status, number of health posts, urban status, public phone status, distance to
market, distance to district capital center, and distance to nearest health facility.
141
Table 4.8: Testing for Selection of Mothers: Mother Age at Birth
Birth Cohort:1989-99 & 2003-07 Birth Cohort: 1987-2007
(1) (2) (3) (4) (5) (6)
Treated(=1) 0.021 0.030 0.028 0.014 0.022 0.026
(0.031) (0.030) (0.031) (0.031) (0.032) (0.031)
Distance From Health Facility -0.006 0.007
(0.011) (0.015)
R2 0.991 0.991 0.991 0.986 0.986 0.986
Observations 8420 8394 8394 11003 10969 10969
Mean of Dependent Variable 31.17 31.16 31.16 32.23 32.22 32.22
Birth Month Year FE Y Y Y Y Y Y
Individual Controls Y Y Y Y Y Y
Community Controls N Y Y N Y N
Original IFLS Community Y Y Y Y Y Y
Note: Standard errors are clustered at the level (*** p<0.01, ** p<0.05, * p<0.1).
The dependent variable is mother age at the time of birth of her child. The variable
Treated takes a value of 1, if a midwife is present in the community during the birth
year of the child, and 0 otherwise. The individual controls mother years of education
(splines with knots at 6,9 and 12) and mother age at survey (splines with knots at
20, 25, 30, 35, 40 and 45). Community controls include time varying changes at
the community level: paved road status, electricity status, number of health posts,
urban status, public phone status, distance to market, distance to district capital
center, and distance to nearest health facility.
142
Table 4.9: Testing for Gender-specific Reporting Bias due to the Village
Midwife Program
(1) (2) (3)
Treated(=1) 0.047* 0.042 0.036
(0.028) (0.027) (0.027)
Male Child Treated(=1) -0.001 -0.003 -0.003
(0.032) (0.032) (0.032)
R2 0.448 0.448 0.449
Observations 9460 9437 9437
Mean of Dependent Variable 0.76 0.76 0.76
Birth Month FE Y Y Y
Male Child Birth Year FE Y Y Y
Individual Controls Y Y Y
Community Controls N Y Y
Male Child Original IFLS Community Y Y Y
Note: Standard errors are clustered at the community level (*** p<0.01,
** p<0.05, * p<0.1). The dependent variable bwreport takes a value of
1, if the birth weight is reported, and 0 otherwise. The variableTreated
takes a value of 1, if a midwife is present in the community during the
birth year of the child, and 0 otherwise. The individual controls mother
years of education (splines with knots at 6,9 and 12) and mother age at
survey (splines with knots at 20, 25, 30, 35, 40 and 45). Community con-
trols include time varying changes at the community level: paved road
status, electricity status, number of health posts, urban status, public
phone status, distance to market, distance to district capital center, and
distance to nearest health facility.
143
4.10 Appendix
Table A1: Impacts of The Village Midwife Program on Likelihood of Male
Birth { Continuous Years of Exposure
(1) (2) (3) (4) (5) (6)
Years of Exposure 0.004 0.004* 0.004 0.003 0.004 0.004
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)
Distance from Health Facility -0.002 -0.001
(0.007) (0.005)
R2 0.065 0.066 0.066 0.056 0.058 0.058
Controlservations 8420 8394 8394 11003 10969 10969
Mean of Dependent Variable 0.515 0.515 0.515 0.517 0.517 0.517
Birth Month Year FE Y Y Y Y Y Y
Individual Controls Y Y Y Y Y Y
Community Controls N Y Y N Y N
Commid93 Y Y Y Y Y Y
Note: Standard errors are clustered at the community level (*** p<0.01, ** p<0.05, * p<0.1).
The rst three columns report all births from last 5 years while the last three columns consider
all births from 1987 to 2007. The dependent variable MaleChild takes a value of 1, if the
child is male, and 0 otherwise. The variable YearsofProgramExposure is number of Years
midwife is present in the community in that birth year. The individual controls mother years
of education (splines with knots at 6,9 and 12) and mother age at survey (splines with knots at
20, 25, 30, 35, 40 and 45). Community controls include time varying changes at the community
level: paved road status, electricity status, number of health posts, urban status, public phone
status, distance to market, distance to sub-district, and distance to nearest health facility.
144
Table A2: Impacts of The Village Midwife Program on Likelihood of Male
Child at First Birth { Continuous Years of Exposure
Birth Cohort:1989-99 & 2003-07 Birth Cohort: 1987-2007
(1) (2) (3) (4) (5) (6)
Years of Exposure 0.006* 0.006* 0.006* 0.004 0.004 0.004
(0.004) (0.004) (0.004) (0.003) (0.003) (0.003)
Distance from Health Facility -0.006 -0.006
(0.014) (0.010)
R2 0.113 0.114 0.114 0.102 0.104 0.104
Controlservations 4584 4578 4578 6142 6134 6134
Mean of Dependent Variable 0.512 0.512 0.512 0.514 0.514 0.514
Birth Month Year FE Y Y Y Y Y Y
Individual Controls Y Y Y Y Y Y
Community Controls N Y Y N Y N
Commid93 Y Y Y Y Y Y
Note: Standard errors are clustered at the level (*** p<0.01, ** p<0.05, * p<0.1). The rst
three columns report all births from last 5 years while the last three columns consider all births
from 1987 to 2007. The dependent variable MaleChild takes a value of 1, if the child is male,
and 0 otherwise. The variableYearsofProgramExposure is number of Years midwife is present
in the community in that birth year. The individual controls mother years of education (splines
with knots at 6,9 and 12) and mother age at survey (splines with knots at 20, 25, 30, 35, 40
and 45). Community controls include time varying changes at the community level: paved road
status, electricity status, number of health posts, urban status, public phone status, distance to
market, distance to sub-district, and distance to nearest health facility.
145
Table A3: Impacts of The Village Midwife Program on Likelihood of Male
Birth, by Mother Education { Continuous Years of Exposure
Primary and Below Primary Above
(1) (2) (3) (4) (5) (6)
Years of Exposure 0.008** 0.007** 0.007** -0.001 -0.000 -0.000
(0.003) (0.003) (0.003) (0.004) (0.004) (0.004)
Distance from Health Facility 0.000 -0.008
(0.010) (0.011)
R2 0.119 0.121 0.121 0.130 0.130 0.130
Controlservations 4598 4581 4581 3822 3813 3813
Mean of Dependent Variable 0.513 0.513 0.513 0.516 0.517 0.517
Birth Month Year FE Y Y Y Y Y Y
Individual Controls Y Y Y Y Y Y
Community Controls N Y Y N Y Y
Commid93 Y Y Y Y Y Y
Note: Standard errors are clustered at the level (*** p<0.01, ** p<0.05, * p<0.1). The rst
three columns report the results for mothers with education primary and less while the last three
columns report the same for the mothers with education more than primary. The dependent
variable MaleChild takes a value of 1, if the child is male, and 0 otherwise. The variable
YearsofProgramExposure is number of Years midwife is present in the community in that
birth year. The individual controls mother years of education and mother age at survey (splines
with knots at 20, 25, 30, 35, 40 and 45). Community controls include time varying changes at
the community level: paved road status, electricity status, number of health posts, urban status,
public phone status, distance to market, distance to sub-district, and distance to nearest health
facility.
146
Table A4: Impacts of The Village Midwife Program on Birth Weights (in
Grams), by Gender { Continuous Years of Exposure
Male Birth Weight Female Birth Weight
(1) (2) (3) (4) (5) (6)
Years of Exposure -6.323* -6.070* -5.541 2.382 2.998 3.837
(3.519) (3.573) (3.637) (4.175) (4.296) (4.342)
Distance from Health Facility 23.662 35.212*
(17.210) (19.150)
R2 0.166 0.168 0.169 0.160 0.164 0.165
Controlservations 3712 3709 3709 3466 3462 3462
Mean of Dependent Variable 3188.31 3188.43 3188.43 3103.17 3102.80 3102.80
Birth Month FE Y Y Y Y Y Y
Birth Year FE Y Y Y Y Y Y
Individual Controls Y Y Y Y Y Y
Community Controls N Y Y N Y Y
Commid93 Y Y Y Y Y Y
Note: Standard errors are clustered at the level (*** p<0.01, ** p<0.05, * p<0.1). The rst
three columns report the results for male children while the last three columns do the same for
female children. The variableYearsofProgramExposure is number of Years midwife is present
in the community in that birth year. The individual controls mother years of education (splines
with knots at 6,9 and 12) and mother age at survey (splines with knots at 20, 25, 30, 35, 40
and 45). Community controls include time varying changes at the community level: paved road
status, electricity status, number of health posts, urban status, public phone status, distance to
market, distance to sub-district, and distance to nearest health facility.
147
Table A5: Impacts of The Village Midwife Program on Infant Mortality
Birth Cohort:1989-99 & 2003-07
(1) (2) (3)
Treated(=1) 0.010 0.010 0.009
(0.014) (0.014) (0.014)
Male Child Treated(=1) -0.002 -0.001 -0.001
(0.017) (0.017) (0.017)
R2 0.097 0.098 0.098
Controlservations 8420 8394 8394
Mean of Dependent Variable 0.03 0.03 0.03
Birth Month FE Y Y Y
Male Child Birth Year FE Y Y Y
Individual Controls Y Y Y
Community Controls N Y Y
Male Child Commid93 Y Y Y
Note: Standard errors are clustered at the community level (*** p<0.01, ** p<0.05, * p<0.1).
The dependent variable takes a value of 1 if the child experienced mortality within 1 year of
his/her life, and 0 otherwise. The variable Treated takes a value of 1, if a midwife is present
in the community during the birth year of the child, and 0 otherwise. The individual controls
mother years of education (splines with knots at 6,9 and 12) and mother age at survey (splines
with knots at 20, 25, 30, 35, 40 and 45). Community controls include time varying changes at
the community level: paved road status, electricity status, number of health posts, urban status,
public phone status, distance to market, distance to sub-district, and distance to nearest health
facility.
148
Table A6: Impacts of The Village Midwife Program on Infant Mortality {
Continuous Years of Exposure
Birth Cohort:1989-99 & 2003-07
(1) (2) (3)
Years of Exposure -0.002* -0.003* -0.003*
(0.001) (0.001) (0.001)
Male Child Years of Exposure 0.003* 0.003* 0.003*
(0.001) (0.001) (0.001)
R2 0.097 0.098 0.098
Controlservations 8420 8394 8394
Mean of Dependent Variable 0.03 0.03 0.03
Birth Month FE Y Y Y
Male Child Birth Year FE Y Y Y
Individual Controls Y Y Y
Community Controls N Y Y
Male Child Commid93 Y Y Y
Note: Standard errors are clustered at the community level (*** p<0.01, ** p<0.05, * p<0.1).
The dependent variable takes a value of 1 if the child experienced mortality within 1 year of
his/her life, and 0 otherwise. The variable YearsofProgramExposure is number of Years
midwife is present in the community in that birth year. The individual controls mother years
of education (splines with knots at 6,9 and 12) and mother age at survey (splines with knots at
20, 25, 30, 35, 40 and 45). Community controls include time varying changes at the community
level: paved road status, electricity status, number of health posts, urban status, public phone
status, distance to market, distance to sub-district, and distance to nearest health facility.
149
Table A7: Impacts of The Village Midwife Program on Low Birth Weight
Birth Cohort:1989-99 & 2003-07
(1) (2) (3)
Treated(=1) 0.020 0.020 0.019
(0.025) (0.025) (0.025)
Male Child Treated(=1) 0.009 0.009 0.009
(0.029) (0.029) (0.029)
R2 0.122 0.124 0.124
Controlservations 7253 7246 7246
Mean of Dependent Variable 0.08 0.08 0.08
Birth Month FE Y Y Y
Male Child Birth Year FE Y Y Y
Individual Controls Y Y Y
Community Controls N Y Y
Male Child Commid93 Y Y Y
Note: Standard errors are clustered at the community level (*** p<0.01, ** p<0.05, * p<0.1).
The dependent variable low birth weight is measured takes a value of 1 if the child has birth
weight less 2500 grams and 0 otherwise. The variable Treated takes a value of 1, if a midwife
is present in the community during the birth year of the child, and 0 otherwise. The individual
controls mother years of education (splines with knots at 6,9 and 12) and mother age at survey
(splines with knots at 20, 25, 30, 35, 40 and 45). Community controls include time varying
changes at the community level: paved road status, electricity status, number of health posts,
urban status, public phone status, distance to market, distance to sub-district, and distance to
nearest health facility.
150
Table A8: Impacts of The Village Midwife Program on Low Birth Weight {
Continuous Years of Exposure
Birth Cohort:1989-99 & 2003-07
(1) (2) (3)
Years of Exposure -0.002 -0.003 -0.003
(0.002) (0.002) (0.002)
Male Child Years of Exposure 0.003 0.003 0.003
(0.003) (0.003) (0.003)
R2 0.122 0.124 0.124
Controlservations 7253 7246 7246
Mean of Dependent Variable 0.08 0.08 0.08
Birth Month FE Y Y Y
Male Child Birth Year FE Y Y Y
Individual Controls Y Y Y
Community Controls N Y Y
Male Child Commid93 Y Y Y
Note: Standard errors are clustered at the community level (*** p<0.01, ** p<0.05, * p<0.1).
The dependent variable low birth weight is measured takes a value of 1 if the child has birth
weight less 2500 grams, and 0 otherwise. The variable YearsofProgramExposure is number
of Years midwife is present in the community in that birth year. The individual controls mother
years of education (splines with knots at 6,9 and 12) and mother age at survey (splines with
knots at 20, 25, 30, 35, 40 and 45). Community controls include time varying changes at the
community level: paved road status, electricity status, number of health posts, urban status,
public phone status, distance to market, distance to sub-district, and distance to nearest health
facility.
151
Chapter 5
Conclusion
Health and environment during early life|in particular in utero period|play an
important role in shaping human capital of an individual. In the context of this
literature, the three essays of this dissertation show that in some cases impact assess-
ment of in utero shocks on certain outcomes of interest should be done allowing for
fertility selection or selection in pregnancy outcomes. Moreover, the essays also point
out importance of health and family planning program interventions in developing
countries in improving child health outcomes. On the one hand, it calls for more
investment in maternal health where adverse disease environment is a great concern,
on the other hand, it shows that a maternal health improving intervention may pro-
duce low quality o-spring, and therefore, maternal health interventions should be
followed by more investments in ospring health.
Using data from Bangladesh, Chapter 2 shows that provision of an intensive family
planning program may allow mothers to avoid pregnancies overlapping with Ramadan
to an extent. Given the large size of Muslim population in the world and the fact
that Ramadan is recurring in nature, Chapter 2 provides an evidence of an eective
intervention that may alleviate the in utero exposure to Ramadan. Moreover, it shows
ignoring an intervention such as a family planning program falsely lead to concluding
there does not exist parental selection in in utero exposure to Ramadan. It further
shows that ignoring such selection may lead to bias in estimating the impact of in
utero exposure to Ramadan on child height. These ndings have implications for
152
studies which regard in utero to Ramadan as natural experiment.
Using data from India, Chapter 3 shows that parental health and human capital
are important determinants of fetal health in the areas of adverse disease environment
and low provision of health facilities. It further documents selection into live births
in those areas; it nds that taller mothers are more likely to give birth to a male
child. Chapter 3 explains that this is possible because height is a marker of long term
nutritional status, and mothers who are taller are possibly better suited to protect a
male fetus in utero because males are more fragile in nature in utero. The ndings
of this chapter calls more health interventions where adverse disease environment is
a great concern. Moreover, it shows that investment in girls' health may produce
immense benet for future, as they will be mothers in future.
Chapter 4 documents the unintended eects of a maternal health improving in-
tervention. Medical literature documents not all pregnancies result in live births.
Moreover, maternal health status and health behavior might be an important de-
terminant of fetal survival. Using data from Indonesia, Chapter 4 shows that the
midwife program, which was targeted toward improving health status of reproductive
age women, lead to an increase in the likelihood of a male birth. It further shows
that such increase was more concentrated among women are from lower SES group.
Moreover, it also nds that the provision of a midwife program is associated with
much lower for birth weight for males but not for females. These ndings suggest
that impact of in utero shocks should be examined allowing for selection into live
births. Moreover, health interventions that improve maternal health should also take
measures that can improve fetal health quality to compensate for the selection eect.
Chapter 3 and Chapter 4 also point out that child sex may not be random at
birth as it has been assumed in some studies, and parental characteristics and health
behaviors might be an important determinant of child sex at birth purely due to
153
biological reasons.
154
Comprehensive bibliography
Adhvaryu, Achyuta, \Learning, misallocation, and technology adoption: Evidence
from new malaria therapy in Tanzania," The Review of economic studies, 2014,
p. rdu020.
and Anant Nyshadham, \Endowments at Birth and Parents' Investments in
Children," The Economic Journal, 2014.
and , \Return to Treatment in the Formal Health Care Sector: Evidence from
Tanzania," American Economic Journal: Economic Policy, 2015, 7 (3), 29{57.
, James Fenske, and Anant Nyshadham, \Early life circumstance and adult
mental health," University of Oxford, Department of Economics Working Papers,
2014, 698.
Adhvaryu, Achyuta R and Anant Nyshadham, \Schooling, child labor, and
the returns to healthcare in Tanzania," Journal of Human Resources, 2012, 47 (2),
364{396.
Ahsan, Md Nazmul, \Do Parents Selectively Time Birth Relative to Ramadan?
Evidence from Matlab, Bangladesh," 2015.
and Riddhi Bhowmick, \Parental Health, Households, Communities and Fetal
Health in India," 2015.
and , \The Unintended Consequences of the Village Midwife Program in In-
donesia," 2015.
155
Alam, Shamma Adeeb and Claus C P ortner, \Income Shocks, Contraceptive
Use, and Timing of Fertility," Technical Report, Working paper. Seattle, WA:
Seattle University, Seattle 2013.
Almond, Douglas, \Is the 1918 In
uenza Pandemic Over? LongTerm Eects of In
Utero In
uenza Exposure in the Post1940 U.S. Population," Journal of Political
Economy, August 2006, 114 (4), 672{712.
and Bhashkar Mazumder, \Health Capital and the Prenatal Environment: The
Eect of Ramadan Observance During Pregnancy," American Economic Journal:
Applied Economics, October 2011, 3 (4), 56{85.
and , \Fetal origins and parental responses," 2013.
and Janet Currie, \Killing me softly: The fetal origins hypothesis," The journal
of economic perspectives: a journal of the American Economic Association, 2011,
25 (3), 153.
and Lena Edlund, \Trivers{Willard at birth and one year: evidence from US
natality data 1983{2001," Proceedings of the Royal Society of London B: Biological
Sciences, 2007, 274 (1624), 2491{2496.
, Bhashkar Mazumder, and Reyn van Ewijk, \In Utero Ramadan Exposure
and Children's Academic Performance," The Economic Journal, September 2014,
pp. n/a{n/a.
, Hilary W Hoynes, and Diane Whitmore Schanzenbach, \Inside the war on
poverty: The impact of food stamps on birth outcomes," The Review of Economics
and Statistics, 2011, 93 (2), 387{403.
156
, Kenneth Y. Chay, and David S. Lee, \The Costs of Low Birth Weight," The
Quarterly Journal of Economics, 2005, 120 (3), 1031{1083.
, Lena Edlund, and Marten Palme, \Chernobyl's Subclinical Legacy: Prenatal
Exposure to Radioactive Fallout and School Outcomes in Sweden," The Quarterly
Journal of Economics, 2009, 124 (4), 1729{1772.
Andersson, Roland and Staan Bergstr om, \Is maternal malnutrition associ-
ated with a low sex ratio at birth?," Human Biology, 1998, pp. 1101{1106.
Arceo, Eva, Rema Hanna, and Paulina Oliva, \Does the eect of pollution on
infant mortality dier between developing and developed countries? Evidence from
Mexico City," The Economic Journal, 2015.
Barham, Tania, \The Eects of Child Health Interventions on Later Life Cognitive
Functioning: Evidence from Matlab in Bangladesh," American Economic Journal:
Applied Economics, 2012, 4 (1), 245{73.
and Randall Kuhn, \Staying for Benets.," Journal of Human Resources, 2014,
49 (4).
Barker, DJP, \In utero programming of cardiovascular disease," Theriogenology,
2000, 53 (2), 555{574.
Behrman, Jere and Anil Deolalikar, \Health and nutrition," in Hollis Chenery
and Srinivasan T.N., eds., Handbook of Development Economics, 1 ed., Vol. 1,
Elsevier, 1988, pp. 631{711.
Behrman, Jere R. and Barbara L. Wolfe, \More evidence on nutrition demand,"
Journal of Development Economics, January 1984, 14 (1), 105{128.
157
Behrman, Jere R and Barbara L Wolfe, \How does mother's schooling aect
family health, nutrition, medical care usage, and household sanitation?," Journal
of econometrics, 1987, 36 (1), 185{204.
Behrman, Jere R. and Barbara L. Wolfe, \Does More Schooling Make Women
Better Nourished and Healthier? Adult Sibling Random and Fixed Eects Esti-
mates for Nicaragua," The Journal of Human Resources, October 1989, 24 (4),
644{663.
Behrman, Jere R and Mark R Rosenzweig, \Returns to birthweight," Review
of Economics and statistics, 2004, 86 (2), 586{601.
Bhalotra, Sonia and Damian Clarke, \Child Quantity versus Quality: Are Twin
Births Exogenous?," 2013.
and Samantha B Rawlings, \Intergenerational persistence in health in devel-
oping countries: The penalty of gender inequality?," Journal of Public Economics,
2011, 95 (3), 286{299.
and Samantha Rawlings, \Gradients of the intergenerational transmission of
health in developing countries," Review of Economics and Statistics, 2013, 95 (02),
660{672.
Bhalotra, Sonia R and Tom Cochrane, \Where have all the young girls gone?
Identication of sex selection in India," 2010.
Bharadwaj, Prashant and Leah K Lakdawala, \Discrimination Begins in the
Womb.," Journal of Human Resources, 2013, 48 (1).
Bhaskar, V and Bishnupriya Gupta, \India's missing girls: biology, customs, and
economic development," Oxford Review of Economic Policy, 2007, 23 (2), 221{238.
158
Bhat, PN Mari and AJ Francis Zavier, \Factors in
uencing the use of prenatal
diagnostic techniques and the sex ratio at birth in India," Economic and Political
Weekly, 2007, pp. 2292{2303.
Bhatia, Shushum, W Henry Mosley, Abu SG Faruque, and Jyotsnamoy
Chakraborty, \The Matlab family planning-health services project," Studies in
family planning, 1980, pp. 202{212.
Bhaumik, Urmi, Iain Aitken, Ichiro Kawachi, Steve Ringer, John Orav,
and Ellice Lieberman, \Narrowing of sex dierences in infant mortality in Mas-
sachusetts," Journal of perinatology, 2004, 24 (2), 94{99.
Brown, Ryan and Duncan Thomas, \On the long term eects of the 1918 US
in
uenza pandemic," 2011.
Buckles, Kasey S and Daniel M Hungerman, \Season of birth and later out-
comes: Old questions, new answers," Review of Economics and Statistics, 2013, 95
(3), 711{724.
Burgess, Robin, Olivier Deschenes, Dave Donaldson, and Michael Green-
stone, \The unequal eects of weather and climate change: Evidence from mor-
tality in india," Unpublished working paper, 2013.
Chen, Lincoln C, AKMA Chowdhury, and Sandra L Human, \Anthropo-
metric assessment of energy-protein malnutrition and subsequent risk of mortality
among preschool aged children.," The American journal of clinical nutrition, 1980,
33 (8), 1836{1845.
Cooper, Cyrus, Avan A Sayer, Elaine Margaret Dennison, P Gluckman,
M Hanson et al., \The developmental environment: clinical perspectives on
159
eects on the musculoskeletal system.," Developmental origins of health and disease,
2006, pp. 392{405.
Cross, JH, J Eminson, and BA Wharton, \Ramadan and birth weight at full
term in Asian Moslem pregnant women in Birmingham.," Archives of disease in
childhood, 1990, 65 (10 Spec No), 1053{1056.
Currie, Janet, \Healthy, wealthy, and wise: Is there a causal relationship between
child health and human capital development?," Journal of Economic Literature,
2009, 47 (1), 87{122.
, \Inequality at Birth: Some Causes and Consequences," American Economic Re-
view, 2011, 101 (3), 1{22.
and Tom Vogl, \Early-Life Health and Adult Circumstance in Developing Coun-
tries," Annual Review of Economics, 2013, 5 (1), 1{36.
, Matthew Neidell, and Johannes F Schmieder, \Air pollution and infant
health: Lessons from New Jersey," Journal of health economics, 2009, 28 (3), 688{
703.
Deaton, Angus, \Global patterns of income and health: facts, interpretations, and
policies," Technical Report, National Bureau of Economic Research 2006.
Desch^ enes, Olivier, Michael Greenstone, and Jonathan Guryan, \Climate
change and birth weight," The American Economic Review, 2009, pp. 211{217.
Dikensoy, Ebru, Ozcan Balat, Bahar Cebesoy, Ayhan Ozkur, Hulya Ci-
cek, and Gunay Can, \The eect of Ramadan fasting on maternal serum lipids,
cortisol levels and fetal development," Archives of Gynecology and Obstetrics, May
2008, 279 (2), 119{123.
160
DiPietro, Janet A., Marc H. Bornstein, Chun-Shin Hahn, Kathleen Costi-
gan, and Aristide Achy-Brou, \Fetal Heart Rate and Variability: Stability and
Prediction to Developmental Outcomes in Early Childhood," Child Development,
November 2007, 78 (6), 1788{1798.
Dor, Avi, Paul Gertler, and Jacques Van Der Gaag, \Non-price rationing
and the choice of medical care providers in rural Cote d'Ivoire," Journal of health
economics, 1987, 6 (4), 291{304.
Dow, William H, \Flexible discrete choice demand models consistent with util-
ity maximization: an application to health care demand," American Journal of
Agricultural Economics, 1999, pp. 680{685.
Eckstein, Zvi, T Paul Schultz, and Kenneth I Wolpin, \Short-run
uctuations
in fertility and mortality in pre-industrial Sweden," European Economic Review,
1984, 26 (3), 295{317.
Eriksson, Johan G, Eero Kajantie, Clive Osmond, Kent Thornburg, and
David JP Barker, \Boys live dangerously in the womb," American Journal of
Human Biology, 2010, 22 (3), 330{335.
Ewijk, Reyn, \Long-term health eects on the next generation of Ramadan fasting
during pregnancy," Journal of Health Economics, December 2011, 30 (6), 1246{
1260.
Falkner, Frank and James Mourilyan Tanner, Methodology Ecological, Genetic,
and Nutritional Eects on Growth, Vol. 3, Springer, 1986.
161
Floyd, R. Louise, Pierre Decou
e, and Daniel W Hungerford, \Alcohol
use prior to pregnancy recognition1," American Journal of Preventive Medicine,
August 1999, 17 (2), 101{107.
Frankenberg, Elizabeth, Alison Buttenheim, Bondan Sikoki, and Wayan
Suriastini, \Do women increase their use of reproductive health care when it
becomes more available? Evidence from Indonesia," Studies in family planning,
2009, 40 (1), 27.
and Duncan Thomas, \Women's health and pregnancy outcomes: Do services
make a dierence?," Demography, 2001, 38 (2), 253{265.
, Wayan Suriastini, and Duncan Thomas, \Can expanding access to basic
healthcare improve children's health status? Lessons from Indonesia's `midwife in
the village'programme," Population Studies, 2005, 59 (1), 5{19.
Fukuda, Misao, Kyomi Fukuda, Takashi Shimizu, and Henrik Mller, \De-
cline in sex ratio at birth after Kobe earthquake.," Human reproduction, 1998, 13
(8), 2321{2322.
Galloway, Patrick R, \Dierentials in demographic responses to annual price vari-
ations in pre-revolutionary France," European Journal of Population/Revue eu-
rop eenne de D emographie, 1987, 2 (3-4), 269{305.
Gibson, Mhairi A and Ruth Mace, \Strong mothers bear more sons in rural
Ethiopia," Proceedings of the Royal Society of London B: Biological Sciences, 2003,
270 (Suppl 1), S108{S109.
Gluckman, Peter D and Mark A Hanson, The fetal matrix: evolution, develop-
ment and disease, Cambridge University Press, 2004.
162
Griliches, Zvi, \Sibling models and data in economics: Beginnings of a survey,"
The Journal of Political Economy, 1979, pp. S37{S64.
Hamoudi, Amar and Jenna Nobles, \Do Daughters Really Cause Divorce?
Stress, Pregnancy, and Family Composition," Demography, 2014, 51 (4), 1423{
1449.
Heckman, James J, \The economics, technology, and neuroscience of human capa-
bility formation," Proceedings of the national Academy of Sciences, 2007, 104 (33),
13250{13255.
Hern andez-Juli an, Rey, Hani Mansour, and Christina Peters, \The eects of
intrauterine malnutrition on birth and fertility outcomes: evidence from the 1974
Bangladesh famine," Demography, 2014, 51 (5), 1775{1796.
Hunter, E. S. and T. W. Sadler, \D-({)-beta-hydroxybutyrate-induced eects on
mouse embryos in vitro," Teratology, October 1987, 36 (2), 259{264.
Jaddoe, Vincent W. V. and Jacqueline C. M. Witteman, \Hypotheses on
the Fetal Origins of Adult Diseases: Contributions of Epidemiological Studies,"
European Journal of Epidemiology, February 2006, 21 (2), 91{102.
James, William H, \The human sex ratio. Part 1: A review of the literature,"
Human Biology, 1987, pp. 721{752.
Jayachandran, Seema and Rohini Pande, \Why Are Indian Children So Short?,"
Technical Report, National Bureau of Economic Research 2015.
Joshi, Shareen and T Paul Schultz, \Family planning and women's and children's
health: Long-term consequences of an outreach program in Matlab, Bangladesh,"
Demography, 2013, 50 (1), 149{180.
163
Kapoor, Amita, Elizabeth Dunn, Alice Kostaki, Marcus H. Andrews, and
Stephen G. Matthews, \Fetal programming of hypothalamo-pituitary-adrenal
function: prenatal stress and glucocorticoids," The Journal of Physiology, April
2006, 572 (1), 31{44.
Karimi, Seyed M, \The Eect of Malnutrition In Utero on Height: Ramadan as a
Natural Experiment," 2014.
Koenig, Michael Alan, Mehrab Ali Khan, Bogdan Wojtyniak, John D
Clemens, Jyotsnamoy Chakraborty, Vincent Fauveau, James F Phillips,
Jalaluddin Akbar, and Uday S Barua, \Impact of measles vaccination on
childhood mortality in rural Bangladesh.," Bulletin of the World Health Organiza-
tion, 1990, 68 (4), 441.
Lam, David A, Jerey A Miron, and Ann Riley, \Modeling seasonality in
fecundability, conceptions, and births," Demography, 1994, 31 (2), 321{346.
Larsen, Elisabeth Clare, Ole Bjarne Christiansen, Astrid Marie Kolte,
and Nick Macklon, \New insights into mechanisms behind miscarriage," BMC
medicine, 2013, 11 (1), 1.
Lavy, Victor, John Strauss, Duncan Thomas, and Philippe De Vreyer,
\Quality of health care, survival and health outcomes in Ghana," Journal of health
economics, 1996, 15 (3), 333{357.
Maccini, Sharon and Dean Yang, \Under the Weather: Health, Schooling, and
Economic Consequences of Early-Life Rainfall," The American Economic Review,
2009, 99 (3), 1006.
164
MacDorman, Marian F and Sharon Kirmeyer, \Fetal and perinatal mortality,
United States, 2005," National vital statistics reports, 2007, 56 (3).
, Donna L Hoyert, Joyce A Martin, Martha L Munson, and Brady E
Hamilton, \Fetal and Perinatal Mortality, United States, 2003," National Vital
Statistics Reports, 2007, 55 (6).
Maitra, Pushkar and Sarmistha Pal, \Birth spacing, fertility selection and child
survival: Analysis using a correlated hazard model," Journal of Health Economics,
May 2008, 27 (3), 690{705.
Majid, Muhammad Farhan, \The Persistent E ects of in Utero Nutrition Shocks
Over the Life Cycle: Evidence From Ramadan Fasting," Journal of Development
Economics, November 2015, 117, 48{57.
Malhotra, Ashok, P. H. Scott, J. Scott, H. Gee, and B. A. Wharton,
\Metabolic changes in Asian Muslim pregnant mothers observing the Ramadan
fast in Britain," British Journal of Nutrition, May 1989, 61 (03), 663{672.
Martorell, Reynaldo and Jean-Pierre Habicht, \Growth in early childhood in
developing countries.," 1986.
Meis, Paul J., James C. Rose, and Melissa Swain, \Pregnancy Alters Diurnal
Variation of Plasma Glucose Concentration," Chronobiology International, January
1984, 1 (2), 145{149.
Metzger, BoydE, RitaA Vileisis, Veronica Ravnikar, and Norbert Freinkel,
\"ACCELERATED STARVATION" AND THE SKIPPED BREAKFAST IN
LATE NORMAL PREGNANCY," The Lancet, March 1982, 319 (8272), 588{592.
165
Mirghani, Hisham M and Omer A Hamud, \The eect of maternal diet re-
striction on pregnancy outcome.," American journal of perinatology, 2006, 23 (1),
21{24.
Mirghani, Hisham M., Sarath Weerasinghe, Shamsaa Al-Awar, Lolowa Ab-
dulla, and Mutairu Ezimokhai, \The Eect of Intermittent Maternal Fasting
on Computerized Fetal Heart Tracing," Journal of Perinatology, November 2004,
25 (2), 90{92.
Mohanty, C, Rajniti Prasad, A Srikanth Reddy, Jayant K Ghosh,
TB Singh, and BK Das, \Maternal anthropometry as predictors of low birth
weight," Journal of Tropical Pediatrics, 2006, 52 (1), 24{29.
Moore, D. C. P., M. Stanisstreet, and C. A. Clarke, \Morphological and
physiological eects of -hydroxybutyrate on rat embryos grown in vitro at dierent
stages," Teratology, September 1989, 40 (3), 237{251.
Munshi, Kaivan and Mark Rosenzweig, \Why is mobility in India so low? Social
insurance, inequality, and growth," Technical Report, National Bureau of Economic
Research 2009.
Mwabu, Germano, Martha Ainsworth, and Andrew Nyamete, \Quality of
medical care and choice of medical treatment in Kenya: an empirical analysis,"
Journal of Human Resources, 1993, pp. 838{862.
Myers, Judith H, \Sex ratio adjustment under food stress: maximization of quality
or numbers of ospring?," American Naturalist, 1978, pp. 381{388.
Nepomnaschy, Pablo A, Kathleen B Welch, Daniel S McConnell, Bobbi S
Low, Beverly I Strassmann, and Barry G England, \Cortisol levels and very
166
early pregnancy loss in humans," Proceedings of the National Academy of Sciences
of the United States of America, 2006, 103 (10), 3938{3942.
Orzack, Steven Hecht, J William Stubbleeld, Viatcheslav R Akmaev,
Pere Colls, Santiago Munn e, Thomas Scholl, David Steinsaltz, and
James E Zuckerman, \The human sex ratio from conception to birth," Pro-
ceedings of the National Academy of Sciences, 2015, 112 (16), E2102{E2111.
Osmani, Siddiq and Amartya Sen, \The hidden penalties of gender inequality:
fetal origins of ill-health," Economics & Human Biology, 2003, 1 (1), 105{121.
Phillips, James F., Ruth Simmons, J. Chakraborty, and A. I. Chowdhury,
\Integrating Health Services into an MCH-FP Program: Lessons from Matlab,
Bangladesh," Studies in Family Planning, July 1984, 15 (4), 153{161.
Pitt, Mark M., \Estimating the Determinants of Child Health When Fertility and
Mortality Are Selective," The Journal of Human Resources, January 1997, 32 (1),
129{158.
Portner, Claus C, \Sex-selective abortions, fertility, and birth spacing," World
Bank Policy Research Working Paper, 2015, (7189).
Rahman, Omar, Jane Menken, Andrew Foster, and Paul Gertler, \Matlab
[Bangladesh] Health and Socioeconomic Survey (MHSS), 1996," Ann Arbor, Michi-
gan: Inter-university Consortium for Political and Social Research (ICPSR 2705),
1999, 13.
Reeves, Jane, \Pregnancy and fasting during Ramadan.," BMJ: British Medical
Journal, 1992, 304 (6830), 843.
167
Rizzo, Thomas, Boyd E Metzger, William J Burns, and Kayreen Burns,
\Correlations between antepartum maternal metabolism and intelligence of o-
spring," New England Journal of Medicine, 1991, 325 (13), 911{916.
Robinson, Trinka and Jeanne Raisler, \Each one is a doctor for herself": Ra-
madan fasting among pregnant Muslim women in the United States," Ethn Dis,
2005, 15 (1 Suppl 1), S1{99.
Ronsmans, Carine, Anne Marie Vanneste, Jyotsnamoy Chakraborty, and
Jeroen Van Ginneken, \Decline in maternal mortality in Matlab, Bangladesh:
a cautionary tale," The Lancet, 1997, 350 (9094), 1810{1814.
Rose, Elaina, \Consumption smoothing and excess female mortality in rural India,"
Review of Economics and statistics, 1999, 81 (1), 41{49.
Rosenblum, Daniel, \The eect of fertility decisions on excess female mortality in
India," Journal of Population Economics, 2013, 26 (1), 147{180.
Rosenzweig, Mark R and Kenneth I Wolpin, \Evaluating the eects of op-
timally distributed public programs: Child health and family planning interven-
tions," The American Economic Review, 1986, pp. 470{482.
and , \Heterogeneity, intrafamily distribution, and child health," Journal of
Human Resources, 1988, pp. 437{461.
and T Paul Schultz, \The behavior of mothers as inputs to child health: the
determinants of birth weight, gestation, and rate of fetal growth," in \Economic
aspects of health," University of Chicago Press, 1982, pp. 53{92.
and , \Child mortality and fertility in Colombia: individual and community
eects," Health policy and education, 1982, 2 (3), 305{348.
168
Sanders, Nicholas J and Charles Stoecker, \Where have all the young men
gone? Using sex ratios to measure fetal death rates," Journal of health economics,
2015, 41, 30{45.
Schultz, T. Paul, \Why Governments Should Invest More to Educate Girls," World
Development, February 2002, 30 (2), 207{225.
Schultz, T Paul, \Population policies, fertility, women's human capital, and child
quality," Handbook of development economics, 2007, 4, 3249{3303.
, \How Does Family Planning Promote Development? Evidence from a Social Ex-
periment in Matlab, Bangladesh|1977{1996," Yale University, Economic Growth
Center, New Haven, Conn, 2009.
Seckl, Jonathan R and Megan C Holmes, \Mechanisms of disease: glucocorti-
coids, their placental metabolism and fetal'programming'of adult pathophysiology,"
Nature clinical practice Endocrinology & metabolism, 2007, 3 (6), 479{488.
Shah, Manisha and Bryce Millett Steinberg, \Drought of opportunities: con-
temporaneous and long term impacts of rainfall shocks on human capital," Tech-
nical Report, National Bureau of Economic Research 2013.
Sheehan, E. A., F. Beck, C. A. Clarke, and M. Stanisstreet, \Eects of beta-
hydroxybutyrate on rat embryos grown in culture," Experientia, February 1985, 41
(2), 273{275.
Shifotoka, Anna and Andrew W Fogarty, \Is the human sex ratio at birth
a sentinel health indicator that is associated with HIV/AIDS and tuberculosis
prevalences in the 21st century?," Journal of epidemiology and community health,
2012, pp. jech{2011.
169
Sinha, Nistha, \Fertility, child work, and schooling consequences of family planning
programs: Evidence from an experiment in rural Bangladesh," Economic Develop-
ment and Cultural Change, 2005, 54 (1), 97{128.
Song, Shige, \Does famine in
uence sex ratio at birth? Evidence from the 1959{
1961 Great Leap Forward Famine in China," Proceedings of the Royal Society of
London B: Biological Sciences, 2012, p. rspb20120320.
Stein, Aryeh D, Paul G Barnett, and Daniel W Sellen, \Maternal undernu-
trition and the sex ratio at birth in Ethiopia: evidence from a national sample,"
Proceedings of the Royal Society of London B: Biological Sciences, 2004, 271 (Suppl
3), S37{S39.
Strauss, John, \Households, communities, and preschool children's nutrition out-
comes: Evidence from rural C^ ote d'Ivoire," Economic Development and Cultural
Change, 1990, pp. 231{261.
and Duncan Thomas, \Human resources: empirical modeling of household and
family decisions.," 1995.
and , \Health, nutrition, and economic development," Journal of economic
literature, 1998, pp. 766{817.
and , \Health over the life course," Handbook of development economics, 2007,
4, 3375{3474.
Subramanian, SV, Leland K Ackerson, George Davey Smith, and Neetu A
John, \Association of maternal height with child mortality, anthropometric failure,
and anemia in India," Jama, 2009, 301 (16), 1691{1701.
170
Thomas, Duncan, \The causal eect of health on social and economic prosperity:
Methods and ndings," Health Economic Growth Framework Paper, 2009.
, Firman Witoelar, Elizabeth Frankenberg, Bondan Sikoki, John Strauss,
Cecep Sumantri, and Wayan Suriastini, \Cutting the costs of attrition: Re-
sults from the Indonesia Family Life Survey," Journal of Development Economics,
2012, 98 (1), 108{123.
, John Strauss, and Maria-Helena Henriques, \Child survival, height for age
and household characteristics in Brazil," Journal of Development Economics, 1990,
33 (2), 197{234.
, , and , \How does mother's education aect child height?," Journal of human
resources, 1991, pp. 183{211.
Trivers, Robert L and Dan E Willard, \Natural selection of parental ability to
vary the sex ratio of ospring," Science, 1973, 179 (4068), 90{92.
Valente, Christine, \Civil con
ict, gender-specic fetal loss, and selection: A new
test of the Trivers{Willard hypothesis," Journal of health economics, 2015, 39,
31{50.
van Ewijk, Reyn J. G., Rebecca C. Painter, and Tessa J. Roseboom, \As-
sociations of Prenatal Exposure to Ramadan with Small Stature and Thinness in
Adulthood: Results From a Large Indonesian Population{Based Study," American
Journal of Epidemiology, April 2013, 177 (8), 729{736.
Veena, SR, K Kumaran, MN Swarnagowri, MN Jayakumar, SD Leary,
CE Stein, VA Cox, and CHD Fall, \Intergenerational eects on size at birth
in South India," Paediatric and perinatal epidemiology, 2004, 18 (5), 361{370.
171
Weaver, Emily H, Elizabeth Frankenberg, Bruce J Fried, Duncan Thomas,
Stephanie B Wheeler, and John E Paul, \Eect of Village Midwife Program
on Contraceptive Prevalence and Method Choice in Indonesia," Studies in family
planning, 2013, 44 (4), 389{409.
Williams, Robert J and Susan P Gloster, \Human sex ratio as it relates to
caloric availability," Social biology, 1992, 39 (3-4), 285{291.
Wong, Emelita L, Barry M Popkin, David K Guilkey, and John S Akin,
\Accessibility, quality of care and prenatal care use in the Philippines," Social
science & medicine, 1987, 24 (11), 927{944.
172

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Three essays on economics of early life health in developing countries

PDF

The determinants and measurement of human capital

PDF

Three essays on health & aging

PDF

Essays on development and health economics

PDF

Essays on health insurance programs and policies

PDF

Essays on health economics

PDF

Long-term impacts of childhood adversity on health and human capital

PDF

Three essays on health economics

PDF

Selected papers on methods for evaluating the impact of marijuana use on BMI and other risk factors for metabolic syndrome

PDF

Three essays on cooperation, social interactions, and religion

PDF

Three essays on social policy: institutional development, and subjective well-being as a cause and consequence of labor market outcomes

PDF

Essays on education and institutions in developing countries

PDF

The impact of Patient-Centered Medical Home on a managed Medicaid plan

PDF

Essays on development economics and adolescent behavior

PDF

Discriminating changes in health using patient-reported outcomes

PDF

Investigating daily effects of activity space-based built environment exposures on physical activity behaviors in Hispanic women during pregnancy and early postpartum

PDF

Essays on macroeconomics of health and labor

PDF

Essays on health and aging with focus on the spillover of human capital

PDF

Essays in political economy and mechanism design

PDF

Burden of illness in hemophilia A: taking the patient’s perspective

Asset Metadata

Creator Ahsan, Md Nazmul (author)

Core Title Selection and impacts of early life events on later life outcomes

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Economics

Publication Date 06/16/2016

Defense Date 04/18/2016

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag early life,gender,human capital formation,maternal health,OAI-PMH Harvest,selection,sex ratio

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Strauss, John A. (committee chair), Nugent, Jeffrey (committee member), Sood, Neeraj (committee member)

Creator Email mdnazmua@usc.edu,nzmlahsan@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-252414

Unique identifier UC11280648

Identifier etd-AhsanMdNaz-4436.pdf (filename),usctheses-c40-252414 (legacy record id)

Legacy Identifier etd-AhsanMdNaz-4436.pdf

Dmrecord 252414

Document Type Dissertation

Format application/pdf (imt)

Rights Ahsan, Md Nazmul

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA