University of Southern California Dissertations and Theses

Three essays on economics of early life health in developing countries

(USC Thesis Other)

Three essays on economics of early life health in developing countries

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content THREE ESSAYS ON ECONOMICS OF EARLY LIFE
HEALTH IN DEVELOPING COUNTRIES
by
Riddhi Bhowmick
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)
May 2016
Copyright 2016 Riddhi Bhowmick
Acknowledgments
The last ve years spent as a graduate student in University of Southern California
has been the most fullling, enriching and joyful period of my life. Not only my
knowledge in economics has increased in these years, I feel a lot more evolved as a
human being | intellectually, emotionally and spiritually. USC has given me this
immense opportunity to interact with people from all over the world. This experience
has helped me to appreciate the rich diversity as well as the unity that binds us, the
human race.
Needless to say, I have accrued debts to many in this eventful journey. I would
rst like to thank, my adviser, Professor John Strauss, to whom I owe much of my
intellectual debt. Professor Strauss, himself being a very thorough and intense re-
searcher, expects the same from us. This helped us, who were beginners in research,
to have an idea of the gold-standards for pursuing any research question. His vast
domain knowledge and experience have helped me to grow as a researcher. It has
been a great honor for me to work under his supervision and guidance.
I am thankful to Professor Jerey Nugent, who has always given me encourage-
ment and helped me generously in all my endeavors. His friendliness, love and warmth
have left a great impression on me. His wide array of interests in economics and his in-
sistence on telling a relatable story from any data analysis have beneted my research
a lot. I thank him for meticulously going through the earlier draft of the dissertation
and providing valuable feedback. I am also indebted to Professor Anant Nyshadham,
i
whose help in my initial days of research is of great importance. His enthusiasm for
any research topic was very contagious and motivating. He has shown us what steps
to follow when one has a research question, but does not know how to turn it into a
good research paper.
I would like to thank Professor Tridib Banerjee for readily serving in my disserta-
tion committee. He has provided valuable inputs to my research. My sincere thanks
to Professor Geert Ridder and Jinkook Lee who served in my qualifying committee.
I owe enormous debt to Md Nazmul Ahsan and Rakesh Banerjee who have been
very kind to let me work with them. Our numerous intellectual and lengthy discus-
sions on every conceivable topic under the sky have been most helpful in understand-
ing dierent academic as well as non-academic matters. Their constant company
has made possible the journey in grad school stimulating and refreshing. My special
thanks to my batch mate and dear friend, Bilal Muhammad Khan whose warmth and
wholeheartedness I have always valued, cherished and enjoyed. I am also thankful to
Tushar Bharati for our frequent heart to heart discussions on anything and everything!
I greatly acknowledge the help provided by Morgan Ponder and Young Miller in
dealing with any administrative concerns. They have been very kind and supportive
throughout. I also thank Fatima Perez for her help in performing various job related
tasks.
I express my sincere gratitude to Swami Sarvadevanandaji Maharaj from Vedanta
Society, Hollywood. I have the greatest regard for him as a person who has deep
love, wisdom and sel
essness. His constant encouragement to not limit oneself in the
ii
shallow body-mind complex but to do something greater, and thus transcending this
limitation has helped me to free myself from all the concerns and insecurity during
grad life.
Finally, I thank my parents for everything. When I was growing up, they never
let me feel anything wanting, whether in terms of physical comforts or educational
needs, even with their modest means . My father's sense of honesty and righteousness
have always been an inspiration. And I can not express my feelings, in words for the
countless sacrices, my mother has made through out her life. Even when she is torn
inside because of my staying away from her, she has kept on encouraging me over
phone to do good research, in a loving voice. It's my earnest hope that I can become
the perfect son that they deserve.
iii
Table of Contents
Acknowledgments i
List of Tables vii
List of Figures x
Abstract xi
1 Introduction 1
2 Parental Health, Households, Communities and Fetal Health in
India
1
6
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Background Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.1 Likelihood of Male Birth as an Indicator of Fetal Health Quality . 14
2.2.2 Health Production Functions . . . . . . . . . . . . . . . . . . . . 16
2.3 Conceptual Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.2 Potential Empirical Issues . . . . . . . . . . . . . . . . . . . . . . 22
2.3.3 Estimating Equation . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.4.1 Dependent Variables . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.4.2 Household and Community Variables . . . . . . . . . . . . . . . . 25
1
Jointly written with Md Nazmul Ahsan.
iv
2.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.5.1 Parental Education . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.5.2 Parental Height . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.5.3 Income . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.5.4 Access to Information . . . . . . . . . . . . . . . . . . . . . . . . 31
2.5.5 Community Characteristics . . . . . . . . . . . . . . . . . . . . . 32
2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.7 Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.8 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3 The Unintended Consequences of the Village Midwife Program in
Indonesia
2
50
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.2 Village Midwife Program in Indonesia . . . . . . . . . . . . . . . . . . . 55
3.3 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.4 Data and Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.5 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.5.1 Empirical Challenges . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.5.2 Empirical Framework . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.6.1 Male Births . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.6.2 Birth Weight . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.6.3 Threats to Identication . . . . . . . . . . . . . . . . . . . . . . . 67
3.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.8 Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2
Jointly written with Md Nazmul Ahsan.
v
3.9 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4 Evaluating Impact of Community Health Workers on Maternal
and Child Health Behavior in India 84
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.3 Background: ASHA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.5 Empirical Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.6 Estimation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.6.1 Miscarriage and Birth Time Complications . . . . . . . . . . . . . 96
4.6.2 Usage of Antenatal Care (ANC) Services . . . . . . . . . . . . . . 97
4.6.3 Vaccinations and Vitamin Supplementation . . . . . . . . . . . . 97
4.6.4 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
4.6.5 Incidence of Disease and Prevention . . . . . . . . . . . . . . . . . 99
4.7 Discussion and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.8 Tables and Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.9 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
5 Conclusion 121
Comprehensive bibliography 124
vi
List of Tables
2.1 Fetal Survival Figures for India . . . . . . . . . . . . . . . . . . . . . 36
2.2 Birth Weight and Birth Size Figures for India . . . . . . . . . . . . . 37
2.3 Summary Statistics- Household Variables . . . . . . . . . . . . . . . . 38
2.4 Summary Statistics- Community Characteristics . . . . . . . . . . . . 39
2.5 Fetal Survival Regressions- Rural . . . . . . . . . . . . . . . . . . . . 40
2.6 Fetal Survival Regressions- Urban . . . . . . . . . . . . . . . . . . . . 41
2.7 Birth Weight and Birth Size Regressions- Rural . . . . . . . . . . . . 42
2.8 Birth Weight and Birth Size Regressions- Urban . . . . . . . . . . . . 43
2.9 Eect of Access to Information on Fetal Survival and Birth Size . . . 44
2.10 Community Determinants of Fetal Survival and Birth Size- General
Village Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.11 Community Determinants of Fetal Survival and Birth Size- Health
Facilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.12 Community Determinants of Fetal Survival and Birth Size-
Government Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.13 Community Determinants of Birth Weight . . . . . . . . . . . . . . . 48
2.14 Likelihood of Male Child and Maternal Height Relationship in Other
Countries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.1 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.2 Impacts of The Village Midwife Program on Likelihood of Male Birth 72
vii
3.3 Impacts of The Village Midwife Program on Likelihood of a Male
Child at First Birth . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.4 Impacts of The Village Midwife Program on Likelihood of Male
Birth, by Mother Education . . . . . . . . . . . . . . . . . . . . . . . 74
3.5 Impacts of The Village Midwife Program on Birth Weights (in
Grams), by Gender . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.6 Testing for Selective Fertility . . . . . . . . . . . . . . . . . . . . . . . 76
3.7 Testing for Selection of Mothers: Mother Education . . . . . . . . . . 77
3.8 Testing for Selection of Mothers: Mother Age at Birth . . . . . . . . 78
3.9 Testing for Gender-specic Reporting Bias due to the Village Midwife
Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.10 Impacts of The Village Midwife Program on Likelihood of Male
Birth- Continuous Treatment . . . . . . . . . . . . . . . . . . . . . . 80
3.11 Impacts of The Village Midwife Program on Likelihood of Male Child
at First Birth- Continuous Treatment . . . . . . . . . . . . . . . . . . 81
3.12 Impacts of The Village Midwife Program on Likelihood of Male
Birth, by Mother Education- Continuous Treatment . . . . . . . . . . 82
3.13 Impacts of The Village Midwife Program on Birth Weights (in
Grams), by Gender- Continuous Treatment . . . . . . . . . . . . . . . 83
4.1 Village Level Correlates for ASHA Program . . . . . . . . . . . . . . 105
4.2 High Focus States: Village Level, First Stage regression . . . . . . . . 106
4.3 Low Focus States: Village Level, First Stage regression . . . . . . . . 107
4.4 Impact of ASHA on Miscarriage and Birth Complications . . . . . . . 108
4.5 Impact of Asha on Antenatal Care Usage . . . . . . . . . . . . . . . . 109
4.6 Impact of Asha on Vaccinations . . . . . . . . . . . . . . . . . . . . . 110
viii
4.7 Dierent NRHM provisions in High and Non-Focus states
(Reproduced from Rao (2014) . . . . . . . . . . . . . . . . . . . . . . 111
4.8 Robustness Results (+/-500): Miscarriage and Birth Complications . 112
4.9 Robustness Results (+/-500): Antenatal Care Usage . . . . . . . . . 113
4.10 Robustness Results (+/-500): Vaccinations . . . . . . . . . . . . . . . 114
4.11 Robustness Results (+/-800): Miscarriage and Birth Complications . 115
4.12 Robustness Results (+/-800): Antenatal Care Usage . . . . . . . . . 116
4.13 Robustness Results (+/-800): Vaccinations . . . . . . . . . . . . . . . 117
4.14 Impact of Asha on Incidence of Diarrhea on Children . . . . . . . . . 118
4.15 Impact of Asha on Incidence of Fever and Cough of Children . . . . . 119
4.16 Impact of Asha on Receipt of Drugs against Intestinal Worms . . . . 120
ix
List of Figures
4.1 Probability of receipt of ASHA program and Village population . . . . 103
4.2 Distribution of Villages with respect to Population . . . . . . . . . . . 104
x
Abstract
Early life health is an important determinant of subsequent health and dierent later
life economic and non-economic outcomes. In the three essays of this dissertation, I
provide empirical evidence of how parental human capital and provision of health care
services have benecial impact on the early life health of children from disadvantaged
communities in developing countries. Based on the medical evidence of male fragility
in utero, I have also argued why likelihood of male birth is a potent measure of fetal
survival and should be used in the literature on human capital formation more often.
On a related note, I have pointed out that gender of a child at rst birth should not
be used as a natural experiment as gender-specic survival rate of fetus is strongly
in
uenced by parental socio-economic and health conditions, especially in developing
countries.
In the rst essay (Chapter 2), I examine whether parental human capital can sub-
stitute for inadequate provision of quality health infrastructure in the areas of India
with adverse disease environment. I use the second and third waves of the National
Family Health Survey and restrict the sample to rst born within last ve years from
the survey to minimize recall bias and to avert the possibility of selection into live
births due to sex selective abortions. I nd positive evidence for the hypothesis |
maternal education and stature have strong positive associations with fetal health
in areas with unfavorable disease and health infrastructure conditions. My results
suggest that the intergenerational persistence in poor health status of Indian women
is partly biological and partly due to son preference, given the cultural and economic
xi
context of India.
In the following essay (Chapter 3), I explore the importance of accounting for
selection into live births to examine the impact of in utero shocks on later life out-
comes, especially in a developing country context. For that purpose, I consider the
impact of the Village Midwife Program (VMP) in Indonesia on the likelihood of male
births and birth weights. Using all four waves of the Indonesian Family Life Survey
(IFLS) and a dierence-in-dierence strategy, I nd that the provision of a midwife in
a community increases the probability of a male birth by 4 percentage points, mostly
for mothers with at most primary education. I also nd that provision of midwives
leads to 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.
In the last essay (Chapter 4), I evaluate the impact of the introduction of Com-
munity Health Workers (CHW) in India during 2005-06 on a variety of pregnancy
outcomes as well as antenatal and postnatal investments. Applying Fuzzy Regression
Discontinuity Design (RDD) on the third wave of District Level Household and Fa-
cility Survey (DLHS), I nd that mothers from treatment villages are more likely to
experience less term complications at birth and are somewhat less prone to have mis-
carriages. My results tentatively suggest that the higher receipt of antenatal health
services is one of the pathways for better pregnancy outcomes. I also nd that it
leads to higher rates of vaccination mainly for girl children in the treatment villages.
Overall, my results show a benecial impact of the program, especially for the disad-
xii
vantaged sections of the society like women and people from backward classes.
xiii
Chapter 1
Introduction
Child characteristics at birth, or what might be called birth endowments, have signi-
cant impact on an individual's future economic well-being and the eects may extend
even to future generations (Behrman and Rosenzweig, 2004; Almond et al., 2005;
Bhalotra and Rawlings, 2011; Gluckman and Hanson, 2004). Unsurprisingly, the in
utero environment and nutrition have been found to play important roles in human
capital formation (Almond and Currie, 2011). It is argued that tissues and organs of
the body go through some critical periods of development so that when a stimulus
or insult (like malnutrition) occurs it may have lasting or lifelong eects (Barker,
2000; Bhalotra and Rawlings, 2011). Although, there is a substantial literature which
examines the parental and socio-economic correlates of child health (like height and
survival) in developing countries, the literature on the same is non-existent for birth
endowments. Moreover, studies on the causal impact of maternal nutritional status on
dierent pregnancy outcomes in developing countries are very few. There are reasons
to believe why one can't use estimates from developed countries for that purpose. In
countries like India or Indonesia, inputs in health production functions, like(parental
health status, food and medicine prices, and disease environment) vary considerably
across people with dierent socio-economic characteristics like education, income and
location while better availability of medical safety nets can partially compensate for
such low socio-economic hindrances in developed countries. Consequently, the impact
of maternal health can be more substantial in developing economies by acting as a
1
substitute for poor health infrastructure.
A related question is whether or not formal health care really improves the provi-
sion of early life health inputs, like antenatal care, vaccinations, vitamin supplementa-
tion etc. Although the returns to formal health care are quite substantial (Adhvaryu
and Nyshadham, 2015), high transportation costs and long travel time inhibit ac-
cess to formal health care (Wong et al., 1987; Dor et al., 1987; Mwabu et al., 1993;
Dow, 1999; Adhvaryu and Nyshadham, 2012, 2014). An alternative strategy is to
introduce community health workers who will reach out to the clients themselves.
Although there have been a number of such programs across the developing world,
good scientic studies measuring the impact of such program on public health care
services, while controlling for non-random program placement, are virtually nil. The
only exception is the impact evaluation of the Indonesian Village Midwife program
by Frankenberg and co-authors.
1
In three essays, this dissertation tries to address all
these questions using insights from the biology, epidemiology and economic literature.
In Chapter 2, the following question is asked :| Can parental human capital sub-
stitute for low provision of quality health infrastructure in areas of India with adverse
disease environment? The empirical analysis for the exercise is done by estimating
static fetal health production functions with the second and third rounds of National
Family Health Survey, held in years 1998-99 and 2005-06, respectively. The results
are consistent with the substitution hypothesis. Parents, especially mothers with at
least secondary education from rural areas of states with high infant mortality, are
more likely to have a child who is male, and who is also less likely to die as a neonate,
and have better birth weight and birth size. Among all the measures of parental
1
Please refer to section 4.2.
2
human capital, maternal height is found to be the strongest correlate of fetal health,
across India. 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.
Using insights from the earlier chapter, Chapter 3 examines the extent to which
fetal health improves with a maternal health intervention program. To do that, im-
pact of the Indonesian Village Midwife (Bidan Desa) Program is estimated for the
likelihood of male births and birth weight of male children relative to female children.
The data is taken from all four waves of the Indonesian Family Life Surveys (IFLS).
There are three principal ndings in this chapter. Firstly, 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 for any reproductive age woman. Since this is around 10
percent of mean of the sample, it is a signicant eect. Secondly, mothers with low
education experience the largest increase in male births. Finally, midwife program
placement is associated with lower birth weight for male children, while no change
in the same is observed for female children. Several checks for fertility selection or
change in gender-specic reporting bias due to midwife program render a causal in-
terpretation to these ndings. These results underscore 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.
The nal essay (Chapter 4) evaluates the eects of a community health worker
program in India and tries to nd if provision of health workers has actually improved
maternal conditions during pregnancy. Also, the chapter examines if this program
has been able to improve the provision of critical health inputs during the rst two
3
years of a child like vaccinations, medicines against intestinal worms, vitamin supple-
mentation and the like. The program is called the Accredited Social Health Activist
(ASHA) worker program which was heavily implemented during 2005-09 in 18 Indian
states with weak public health indicators. Further, since villages with a population of
1000 or more were given top priority, a Fuzzy Regression Discontinuity Design is em-
ployed for the evaluation of the program. The data, used in this chapter comes from
the third wave of DLHS. The results show that mothers from villages with ASHA
workers are 9% less likely to experience breech presentation and are tentatively less
prone to miscarriages. The receipt of antenatal health services is also tentatively
higher in these villages and seems to be one of the primary mechanisms, lying behind
this nding. Regarding critical health inputs, treatment villages have higher rates
of vaccination and somewhat greater rates of vitamin supplementation. There is no
strong support for a lower incidence of diarrhea or fever and cough among children
if the village has community health workers. Also, the program has beneted girl
children from backward classes more as far as vaccination is concerned.
The dissertation makes a number of contributions in the early life health litera-
ture. Firstly, it attempts to bridge a gap between the medical literature which links
parental health with child health at birth (birth endowment) and the economic lit-
erature which try to nd the household and community level determinants of child
health outcomes (child height, survival) in developing countries, based on nationally
representative samples. In doing so, it also provides evidence on how a well-intended
health intervention for mothers may improve the immediate health of the mothers,
but may have adverse health eects for the next generation. Further, the dissertation
adds to the evidence for intergenerational transmission of health through nongenomic
channels. Secondly, the dissertation provides a number of novel insights in the gender-
4
study as well as early life studies in developing economies. They are :| (a) Contrary
to what many studies assume, sex of the child is not orthogonal to parental back-
ground (even in the absence of sex-selective abortions) and should not be treated as a
natural experiment. (b) There is an urgent need to test assumptions about dynamics
of selection into live births to draw inferences about gender dynamics from the gen-
der/outcome associations. And, (c) the historical burden of low health endowment
for women is actually greater than expected due to son preference alone; biology also
plays a role. Thirdly, it proposes a couple of new measures of birth endowments,
which should be invaluable to the burgeoning literature on parental responses to
birth endowments and help understand the dynamics of human capital formation.
Finally, the dissertation provides evidence from multiple countries on the ecacy of
community health workers on the health of children and pregnant mothers.
5
Chapter 2
Parental Health, Households, Communities and
Fetal Health in India
1
2.1 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 Md Nazmul Ahsan.
6
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.
7
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)
3
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. 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.
8
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.
9
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
10
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.
11
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
12
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
13
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.
2.2 Background Literature
2.2.1 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
14
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.
15
2.2.2 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.
16
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.
17
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; Wolfe and Behrman, 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 edu-
cation 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 ef-
fects of parental education 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.
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.
18
The specic 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.
19
2.3 Conceptual Issues
2.3.1 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
14
and quality of
children who are in-utero (and thus yet to be born). 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
15
) and fetal
growth and nutrition, measured by birth weight and size at birth. The reason for us-
ing 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.
Similar argument goes for birth weight. Barker (2000) advocates the importance of
more detailed anthropometrics at birth so that more insights can be gained regarding
dierent 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
utility function as an argument. A similar argument applies for including fetal qual-
ity (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.
20
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
) (2.1)
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.
21
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
) (2.2)
where O is the output, P represents a set of prices while the other notations have
already been dened above.
2.3.2 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.
22
2.3.3 Estimating Equation
O
ismy
= +E
0
i

E
+H
0
i

H
+I
0
i

I
+X
0
+
smy
+"
ismy
(2.3)
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.
2.4 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
23
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.
2.4.1 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 2.1, 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 2.2, 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
24
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.
2.4.2 Household and Community Variables
We show the summary statistics for parental education, maternal height and land
holdings in Table 2.3. 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 2.4, 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
25
etc and usually have lower number of beneciaries of dierent government schemes.
2.5 Results
2.5.1 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 2.5
and Table 2.6). 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
26
mentioned in Thomas et al. (1990).
Turning to the birth weight and birth size regressions (Table 2.7 and Table 2.8),
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.
2.5.2 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
27
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 2.14 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.
28
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.
29
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 2.7 and Table 2.8, the impact
of maternal height remains stronger in rural areas. We do not nd any signicant
impact of father's height on birth size.
2.5.3 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.
30
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.
2.5.4 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 2.9, 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.
31
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.
2.5.5 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 2.10, Table 2.11 and Table 2.12 while the same for birth weight are shown
32
together in Table 2.13. 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 2.4 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 2.11). 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).
33
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.
2.6 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
34
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.
35
2.7 Tables
Table 2.1: 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
36
Table 2.2: 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
37
Table 2.3: 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
38
Table 2.4: 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
39
Table 2.5: 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.
40
Table 2.6: 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.
41
Table 2.7: 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.
42
Table 2.8: 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.
43
Table 2.9: 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.
44
Table 2.10: 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.
45
Table 2.11: 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.
46
Table 2.12: 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.
47
Table 2.13: 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.
48
2.8 Appendix
Table 2.14: 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.
49
Chapter 3
The Unintended Consequences of the Village
Midwife Program in Indonesia
1
3.1 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. Firstly, 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 decision 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 Md Nazmul Ahsan.
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.
50
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 situations
of environmental stress, this confers girls a pre-birth survival advantage.
However, there are not too many quality causal studies in developing countries,
which directly link maternal nutritional status to fetal health quality. In this chap-
ter, 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 introduced in
the late eighties to improve the high maternal mortality rates in Indonesia. It is
specically aimed at improving the health of women of reproductive age. Using a dif-
ference in dierence strategy, Frankenberg and Thomas (2001) show that the program
was successful in increasing the BMI (body mass index) of the women of reproduc-
tive age, especially for women who had lower initial BMI
4
. Further, they calculate the
estimated positive impact of gaining a village midwife on birth weight to be 80 grams.
The novelty of this chapter 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
3
These outcomes include live birth, miscarriage, still birth, gestational age and ectopic pregnancy.
4
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.
51
strategy
5
. 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 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, midwife pro-
gram placement is associated with lower birth weight for male children but not for
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 place-
ment 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.
The current chapter makes some key contributions to a number of literature.
Firstly, this chapter makes a basic methodological contribution in early life studies of
developing countries
6
. In order to describe underlying mechanisms linking in utero
shocks and later life outcomes, many studies implicitly assume a monotonic relation-
ship between these shocks and birth endowments without formally testing them
7
.
Our results underscore the foremost need to account for selection into live birth be-
fore interpreting the associations as causal, especially for better understanding the
economics of human capital formation. Our recommendations are in line with similar
studies in the literature on developed countries that the empirical economic modeling
5
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).
6
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).
7
The reader may refer to the studies by Maccini and Yang (2009), Shah and Steinberg (2013)
etc.
52
has to integrate 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 chapter 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, likelihood
of male child being positively associated with maternal nutritional status implies the
estimate of mean dierence in say, child investments between boys and girls may have
a bias. In the absence of son preference, if parents compensate for poor birth endow-
ments, 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 pref-
erence is not strong in Indonesia, this may imply \gender-reversal" in human capital
formation in future as there is ample evidence showing inter generational transmission
of the same
8
. In fact, gender reversal may simply happen if parents try to reinforce
birth endowments. Since sons are born with poor birth endowments (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
8
See the studies by Bhalotra and Rawlings (2013, 2011) and the literature cited there.
53
birth endowments nd parents to engage in a reinforcing behavior (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 could bias down-
ward the estimated eects of any such program which improves maternal health on
a variety of fetal quality measures. To our knowledge, we are the rst to report such
unintended consequences of any maternal health intervention program. We hypoth-
esize 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
9
.
The rest of the chapter 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.
9
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)
54
3.2 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 initia-
tive 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. 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 mid-
wives 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 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 gov-
ernment 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).
55
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).
3.3 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-
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 is ample evidence from epidemiological studies provide
56
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,
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.
However, the impact of maternal health interventions on birth weight and in-
fant mortality is somewhat, better examined. For the Village Midwife Program,
Frankenberg and Thomas (2001) nd birthweights 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 dur-
ing 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
57
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.
3.4 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
58
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
10
.
The attrition of the households due to migration could be a concern, as Thomas
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 3.1 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
10
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.
59
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.
3.5 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.
3.5.1 Empirical Challenges
Endogenous Program Placement| Evaluation of health or for that matter, any in-
tervention is often complicated, as the outcomes of interest are aected by the char-
acteristics of individuals, households and communities. The government may target
a set of population where the program is likely to be most 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
60
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) show that
midwife also provided contraceptives to reproductive age women. Moreover, they
may also provide suggestions regarding family size or birth timing. Therefore, it is
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
61
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. It is important to point out that mortality is an extreme event
11
. Therefore,
it is unlikely that provision of a midwife program has substantially altered the com-
position of parental characteristics due to maternal mortality attrition. 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.
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 program
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
12
. We examine this
possibility by looking at mother education in this chapter. 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
11
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.
12
? report a higher likelihood of male child at rst birth for taller mothers in areas of India,
where infant mortality rates are high.
62
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
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
facility. We additionally include time varying community characteristics like distance
to market, community electricity status, and number of health posts.
3.5.2 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
(3.1)
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
63
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 varying
community observables such as paved road status, urban status, public phone status,
distance to market, 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
X
ijmt
+
3

jt
+
m
+
t
+
j
+"
ijmt
(3.2)

1
is the co-ecient of interest.
Recall that the birth weight information is not available for all births. One pos-
sibility is that the provision of a midwife may have changed gender-specic birth
64
weight 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.
To explore this issue, we regress a linear probability model for child i who is born in
month m and year t and whose mother lives in community j:
bwreport
ijmt
=
1
Treated
jt
+
2
Post
t
Treatedmalec+
3
X
ijmt
+
4

jt
+
m
+malec
j
+malec
t
+
ijmt
(3.3)
bwreport is an indicator if the birth weight information for child i is available;
it takes a value of 1 if the birth weight is reported and 0 otherwise. The coecient
of interest is
2
. 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.
3.6 Results
3.6.1 Male Births
We report the eects of Village Midwife Program on likelihood of male births in Ta-
ble 3.2. 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
65
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 3.3, 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 chapter, 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 3.4, while only considering
births from last 5 years of interview. The rst three columns correspond to the moth-
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 3.10, Table 3.11, and Table 3.12 .
The direction of relationship is unchanged. In fact, the results are usually statistically
more signicant for the continuous treatment exposure.
66
3.6.2 Birth Weight
We show in Table 3.5, the impacts of the program on birthweight. 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
birthweight is negative. The eects vary a little across specications, but are usually
statistically 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 birthweights for female
children.
In Table 3.13, we show the results for continuous treatment measures. Similar to
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 is statistically insignicant.
3.6.3 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.
67
Fertility
We check for selective fertility by using the same regression specication for Table 3.2
with total number of births as the dependent variable. Results are reported in Ta-
ble 3.6. 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 midwife program
can explain maternal education and mother age at birth of the sample. We report
the results in Table 3.7 and Table 3.8. We don't nd evidence of any eect of the
program on either of the mother characteristics.
Gender-specic Reporting Bias
Our nal check concerns any change in gender-specic reporting of births due to the
program. Table 3.9 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.
3.7 Conclusion
In this chapter, 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. Us-
ing four waves of Indonesian Family Life Surveys (IFLS) and employing a Dierence-
68
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 at most primary education. However, we nd there is a decrease in
birth weights for male children because of the program while there is such impact for
female children. 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.
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
69
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-
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.
70
3.8 Tables
Table 3.1: 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
71
Table 3.2: 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* 0.042* 0.042* 0.035* 0.039** 0.039*
(0.022) (0.022) (0.023) (0.020) (0.020) (0.020)
Distance from Health Facility -0.002 -0.001
(0.007) (0.005)
R
2
0.065 0.066 0.066 0.056 0.058 0.058
Observations 8420 8399 8399 11003 10974 10974
Mean of dependent variable 0.515 0.514 0.514 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
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 Malec 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 include 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, and distance to nearest health facility.
72
Table 3.3: 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 0.056 0.054 0.036 0.041 0.039
(0.035) (0.035) (0.035) (0.031) (0.031) (0.031)
Distance from Health Facility -0.006 -0.006
(0.015) (0.010)
R
2
0.113 0.114 0.114 0.102 0.104 0.104
Observations 4584 4579 4579 6142 6135 6135
Mean of dependent variable 0.512 0.512 0.512 0.514 0.513 0.513
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 sample is based on rst birth only. The dependent variable
Malec 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 include 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, and
distance to nearest health facility.
73
Table 3.4: 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* 0.051* 0.051 0.021 0.025 0.023
(0.031) (0.031) (0.031) (0.037) (0.038) (0.038)
Distance from Health Facility -0.000 -0.007
(0.010) (0.011)
R
2
0.119 0.121 0.121 0.130 0.130 0.130
Observations 4598 4585 4585 3822 3814 3814
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 Malec 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 include 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, and distance to nearest
health facility.
74
Table 3.5: 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* -59.937* -55.641 -2.971 -2.157 5.563
(35.519) (36.086) (36.201) (44.254) (45.402) (45.368)
Distance from Health Facility 21.842 29.804
(16.825) (18.860)
R
2
0.166 0.169 0.169 0.160 0.164 0.164
Observations 3712 3711 3711 3466 3464 3464
Mean of dependent variable 3188.31 3188.60 3188.60 3103.17 3102.94 3102.94
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 include 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, and distance to nearest health facility.
75
Table 3.6: Testing for Selective Fertility
Birth Cohort:1989-99 & 2003-07 Birth Cohort: 1987-2007
(1) (2) (3) (4) (5) (6)
Treated(=1) -0.035 -0.048 -0.069 -0.034 -0.054 -0.065
(0.105) (0.105) (0.106) (0.101) (0.100) (0.100)
R
2
0.302 0.307 0.308 0.292 0.298 0.298
Observations 3592 3586 3586 4671 4662 4662
Mean of dependent variable 2.49 2.49 2.49 2.51 2.51 2.51
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 depen-
dent variable is total number of live births. 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 include 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, and distance to nearest health
facility.
76
Table 3.7: 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 0.191 0.184 0.153 0.161 0.161
(0.122) (0.123) (0.122) (0.119) (0.120) (0.118)
Distance from Health Facility -0.016 -0.001
(0.040) (0.037)
R
2
0.454 0.454 0.454 0.438 0.439 0.439
Observations 8420 8399 8399 11003 10974 10974
Mean of dependent variable 6.64 6.64 6.64 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, and distance to nearest health facility.
77
Table 3.8: 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 0.027 0.024 0.014 0.019 0.022
(0.031) (0.031) (0.031) (0.031) (0.032) (0.031)
Distance from Health Facility -0.007 0.007
(0.011) (0.015)
R
2
0.991 0.991 0.991 0.986 0.986 0.986
Observations 8420 8399 8399 11003 10974 10974
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 include 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, and distance to
nearest health facility.
78
Table 3.9: 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)
R
2
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 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 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, and
distance to nearest health facility.
79
3.9 Appendix
Table 3.10: Impacts of The Village Midwife Program on Likelihood of Male
Birth- Continuous Treatment
(1) (2) (3) (4) (5) (6)
Years of Program Exposure 0.004 0.004 0.004 0.003 0.003 0.003
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)
Distance from Health Facility -0.002 -0.002
(0.008) (0.005)
R
2
0.065 0.066 0.066 0.056 0.057 0.057
Observations 8420 8399 8399 11003 10974 10974
Mean of dependent variable 0.515 0.514 0.514 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 Malec 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 include 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, and distance to nearest health facility.
80
Table 3.11: Impacts of The Village Midwife Program on Likelihood of Male
Child at First Birth- Continuous Treatment
(1) (2) (3) (4) (5) (6)
Years of Program 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 -0.007
(0.015) (0.010)
R
2
0.113 0.114 0.114 0.102 0.104 0.104
Observations 4584 4579 4579 6142 6135 6135
Mean of dependent variable 0.512 0.512 0.512 0.514 0.513 0.513
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 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 Malec 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 include 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, and distance to nearest health facility.
81
Table 3.12: Impacts of The Village Midwife Program on Likelihood of Male
Birth, by Mother Education- Continuous Treatment
(1) (2) (3) (4) (5) (6)
Years of Program Exposure 0.008** 0.007** 0.007** -0.001 -0.000 -0.001
(0.003) (0.003) (0.003) (0.004) (0.004) (0.004)
Distance from Health Facility -0.000 -0.008
(0.010) (0.011)
R
2
0.119 0.121 0.121 0.130 0.130 0.130
Observations 4598 4585 4585 3822 3814 3814
Mean of dependent variable 0.513 0.513 0.513 0.516 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 Y
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 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 de-
pendent variable Malec 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 include 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, and distance to nearest health facility.
82
Table 3.13: Impacts of The Village Midwife Program on Birth Weights (in
Grams), by Gender- Continuous Treatment
(1) (2) (3) (4) (5) (6)
Years of Program Exposure -6.323* -5.314 -4.853 2.382 2.844 3.534
(3.519) (3.597) (3.664) (4.175) (4.334) (4.387)
Distance from Health Facility 21.951 31.349
(16.997) (19.048)
R
2
0.166 0.169 0.169 0.160 0.164 0.165
Observations 3712 3711 3711 3466 3464 3464
Mean of dependent variable 3188.31 3188.60 3188.60 3103.17 3102.94 3102.94
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 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 variable YearsofProgramExposure is number of Years midwife is
present in the community in that birth year. The individual controls include 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, and distance to nearest health facility.
83
Chapter 4
Evaluating Impact of Community Health Workers
on Maternal and Child Health Behavior in India
4.1 Introduction
Early life health inputs are of extreme importance for early life outcomes of a child,
such as survival and birth weight (Bharadwaj and Eberhard, 2008; Bharadwaj and
Lakdawala, 2013). There is a huge literature in economics that nd in-utero events
and childhood endowments aect later life health, IQ and labor market outcomes
(Behrman and Rosenzweig, 2004; Almond et al., 2005). Formal health care systems
can play a key role in the provision of these inputs. But access to formal health care
in developing economies is far from universal, and expanding access to formal health
care across more of the population is one of the key agenda in public policy. And
returns to such health care is quite substantial. Adhvaryu and Nyshadham (2015), for
instance, nd that formal sector health care greatly reduces the incidence of fever and
malaria among children who sought treatment for acute illness. More timely receipt
of and better adherence to antimalarial treatment are the mechanisms identied by
them.
Studies have shown that high transportation costs and long travel time inhibit
access to health care (Wong et al., 1987; Dor et al., 1987; Mwabu et al., 1993; Dow,
1999; Adhvaryu and Nyshadham, 2012, 2014). Banerjee and Sachdeva (2014) nd
84
that connecting villages with an all-weather road increases preventive health care us-
age by the residents of the connected villages in India. They nd that women are more
likely to seek antenatal care, to have delivery being conducted by a trained health
personnel and are more likely to use modern contraceptive methods. Households are
more likely to treat water and are more likely to be covered by health insurance.
They show that the increase in preventive health care usage comes not only from
the increase in income or reduction in travel cost but also from increased awareness,
improvement in health care supply and increased social interactions.
An alternative strategy would be introduction of Community Health Workers
(CHW) who will reach out to the potential clients, instead of the latter going to the
formal care. Recently, the United Nations has launched a campaign to help expand
and accelerate CHW programs in ten sub-Saharan African countries. The aim is to
recruit one million community health workers by the end of 2015 so that there is one
worker for every 650 rural inhabitants (Rao, 2014).
The Government of India has also come to recognize the importance of such a
program in recent years in the context of reducing under-ve mortality. The Ac-
credited Social Health Activist(ASHA) worker program is an implementation of this
idea. It is one of the key strategies of India's federally funded `National Rural Health
Mission (NRHM)' with an implementation time line ranging from 2005 to 2012. The
implementation of the program is quite vigorous. Around 700k ASHA workers were
hired within the rst four years of the program across the country with an annual
expense of $325 per worker
1
.
1
For more details, readers may refer to \Four Years of NRHM 2005-09" by Ministry of Health
and Family Welfare, Government of India. It is available at http://nrhm.gov.in/publications/four-
years-of-nrhm-2005-2009.html.
85
In this chapter, I try to measure the impact of ASHA program on a variety of
pregnancy outcomes as well as antenatal and postnatal investments. Following Rao
(2014), I also refer to the initial policy statement of the Indian government in the
implementation of the program. It says every village with a population of 1000, in
18 high-focus Indian states
2
, with weak public health indicators and/or weak infras-
tructure and areas of non-focus states with high concentration of `scheduled tribes',
should receive an ASHA worker. Using the third wave of District Level Household
Facility Survey (DLHS), To evaluate the success of this program, I use a fuzzy RDD
by exploiting the rule-based implementation of the program. To my knowledge, this
is the rst study which evaluates the ASHA program utilizing the population cut o
rule. Rao (2014) does the evaluation of this program for immunization services but
using a dierence in dierence framework between high and non focus states
3
.
My results show mothers from villages with ASHA workers 9% less likely to ex-
perience breech presentation. There is also at least weak support that these mothers
are less likely to have any miscarriage and more likely to have a normal delivery. I
investigate whether or not the higher receipt of antenatal health services is one of the
reasons for such benecial pregnancy outcomes. I nd tentative evidence for this in
the data, i.e., the provision of community health workers is associated with greater
usage of receiving any antenatal care, especially for mothers from backward section of
the community. I also nd higher rates of vaccination and vitamin supplementation,
mainly for girl children in the treatment villages. The results are robust and some-
2
High focus Indian states are Uttar Pradesh, Bihar, Rajasthan, Madhya Pradesh, Orissa, Ut-
taranchal, Jharkhand, Chhattisgarh, Assam, Sikkim, Arunachal Pradesh, Manipur, Meghalaya,
Tripura, Nagaland, Mizoram, Himachal Pradesh and Jammu and Kashmir.
3
In the remaining chapter, I use the term low focus and non focus interchangeably.
86
times, stronger if I restrict the sample, more tightly around the population cut-o of
1000. It appears backward sections are sometimes more impacted by this program.
Finally, I don't nd any consistent or signicant impact of the community workers
on incidence of diarrhea and fever or cough among children.
The chapter is organized as follows. In section 2, I brie
y discuss the literature on
the eects of CHWs on health and health care usage outcomes. This is followed by
a brief description of the program in Section 3. I discuss the data and the empirical
model in the next sections 4, and 5, respectively. I discuss the results in section 6
and conclude in section 7.
4.2 Literature Review
Good scientic studies measuring impact of CHW on public health care services are
not many. A key issue in evaluating such a program is non-random program place-
ment. ASHA program is no exception- it is implemented heavily in those states which
are lagging in public health indicators. One of the earlier papers which address this
issue in the context of community health workers is a study by Frankenberg and
Thomas (2001) in Indonesia. They use data from the Indonesia Family Life Survey
and investigate the impact of a major expansion in access to midwifery services on
health and pregnancy outcomes for women of reproductive age. They nd additions
of village midwives are associated with a signicant increase in body mass index for
women of reproductive age, but not for men or for older women. Especially they
nd percentage of women with a BMI of less than 21 declines from 44% to 41:3% (a
decrease of 6%), while the percentage with BMI of less than 18:5 declines from 12:8%
to 10:9% (a decrease of nearly 15%). The presence of a village midwife during preg-
87
nancy is also associated with increased birth weight by about 80 grams. One of the
biggest advantages of this study is it uses a panel survey. As the authors point out,
detailed community-level data linked to individual-level data are not always sucient
in that the application of methods that control community or individual-specic un-
observables requires repeated observations on health.
Following this study, a number of other studies have been done on the Indonesia
Village Midwife Program on other health outcomes and usage of health services. The
general nding is this program has been quite successful in improving dierent di-
mensions of health behavior. For instance, mothers with lower levels of education are
more likely to receive antenatal care during the rst trimester of pregnancy and are
more likely to receive iron tablets if there is a midwife in the village. Frankenberg et
al. (2009) also show that the reliance of pregnant women on traditional birth atten-
dants for birth delivery goes down. The increased usage of injectable contraceptives
has been observed for the treatment villages while a decreased incidence of oral con-
traceptive and implant use has been observed (Weaver et al., 2013). Frankenberg et
al. (2005) report children exposed to the program were beneted in terms of better
nutrition status (as measured by height for age). Finally, a recent study by Ahsan
and Bhowmick (2015b) reports a counter-intuitive nding. Using all four waves of
the Indonesian Family Life Survey (IFLS) and a dierence-in-dierence strategy, the
paper nds that the provision of a midwife in a community increases the probability
of a male birth by 4 percentage points, for mothers with at most primary educa-
tion. But it also leads to decrease in birth weight for male children, while no change
is observed for female children. The authors argue that it is the increased survival
chances of poor quality fetuses, due to positive nutritional shocks from the program
that explains the paradox.
88
Moving on to evidence from other countries, Basinga et al. (2011) nd an increase
in 23 percentage points in the number of births attended by a professional, based on
a randomized experiment in Rwanda when community health providers were given a
payment for performance for each institutional delivery. Rao (2014) evaluates eects
of ASHA on immunization outcomes in a dierence in dierence framework. Using all
three rounds of data from the District Level Household and Facility Survey (DLHS),
the author nds statistically signicant increases (in the range of 14-22 percentage
points) in the coverage of specic vaccines and the provision of full immunization in
high focus states and a reduction in the incidence of infants with no immunization
of up to 16 percentage-points. The current chapter looks into the same program,
but utilizes a dierent identication strategy and looks into other sets of outcomes
as well.
4.3 Background: ASHA
The concept of Community Health Workers is not new in India. In 1977, Government
of India (GOI) introduced a CHW scheme across the country. The workers were in
initially called Community Health Workers, then came to be known as Village Health
Guides since 1981. The workers were mostly male as men were the main targets of the
family planning program when the program was initiated. The program was marred
with the issues of poor role denition, inferior quality of training of the workers and
suspicion of indulging in quackery. Based on the recommendations of a high level
review committee, GOI stopped funding the scheme starting April 2002 while the
states were free to run the scheme on their own. As of 2003, it is reported that no
89
states were running the VHG scheme
4
.
In recent years, GOI has reemphasized the importance of Community Health
Workers (CHW) in improving maternal and child health by using the strategy more
eectively and intensively than in the past. As is well recognized in the literature of
preventive health care in developing countries, a majority of health problems of rural
populations can be prevented or managed by the people themselves, as long as they
are identied and treated on time (Dupas, 2011). Rao (2014) rightly points out that
the idea of para-health workers and bare-foot doctors is even more attractive if we
recognize the impossibility of bringing high quality doctors and nurses to remote and
rural locations. In such a situation, the new generation of CHW programs, including
the ASHA worker program, emphasizes a framework of structured but rapid training
of the health care volunteers. They are supposed to provide a \well-dened linkage
with the public-health set-up, a standard and well equipped medicinal kit as well
as a strategy of leveraging modern communications technology to make the health
workers more accessible and easy to monitor" (Rao, 2014).
One of the important programs under NRHM is Janani Suraksha Yojana (JSY)
which is a safe motherhood intervention for reducing maternal and neonatal mortality
by \promoting institutional delivery among the poor pregnant women" (\Four Years
of NRHM 2005-09", Ministry of Health and Family Welfare, Government of India).
ASHA workers are entrusted to carry out this program. One important role for a
worker under ASHA is thus to help pregnant mothers avail maternal care services by
paying regular visits, arrange for necessary transport and accompany them to health
4
For more details, the readers may refer to \What Works for Children in South Asia: Community
Health Workers" | UNICEF (2004)
90
facility for delivery. All these measures should have signicant impacts on maternal
health during pregnancy. Bharadwaj and Lakdawala (2013) emphasize the impor-
tance of prenatal maternal inputs during pregnancy with regards to neonatal survival
and birth weight as well as its importance for postnatal care and later life outcomes.
Rao (2014) also lists some other responsibilities of ASHA workers. For instance, the
workers should spread general awareness on the importance of immunization, health,
sanitation and nutrition. They are also supposed to mobilize women, children and
vulnerable populations for dierent monthly health day activities.
The NRHM policy prescribes that the ASHA worker be a literate (with education
at least up to 8th standard) woman resident of the village, preferably in the age group
of 25 to 45 years who is to receive training for 23 days in ve episodes at the sub-
centre. The ASHA worker is not a paid worker but a volunteer who receives small
sums per task performed. For instance, according to the guidelines issued by the
central government, she is to be paid 150 rupees per immunization session organized.
Similarly, she is to be paid per \village health day" or \household toilet promotion"
day organized. For more details about this program and public health set-up in India,
the readers can refer to Rao (2014).
4.4 Data
I use the third wave of District Level Household and Facility Survey (DLHS) for my
entire analysis. This survey is a district level survey conducted between December
2007 and December 2008 by the IIPS, Mumbai, with the focus on providing infor-
mation on use of reproductive and child health care. DLHS 3 conducted separate
interviews at the village, household, and individual ever married woman (15-49 years
91
of age), and individual unmarried" woman (15-24 years of age levels). The village
level questionnaires include information on the availability of the ASHA worker and
other publicly provided healthcare services.
In my analysis, I have included mothers whose last birth occurred after 2004 from
all the states (except the Union Territories). In Table 4.1, I show some correlates of
receiving the ASHA program. It shows clearly that a village is more likely to get the
program if it is in high focus state, has many households, better connectivity and
has no big government health facilities. Since ASHA workers are supposed to work
closely with primary health ocials in the village like Integrated Child Development
Centre (ICDS) activists and Auxiliary Nurse Midwives (they are in charge of Health
Sub Centers), a positive likelihood is found between these programs and availability
of ASHA program in a village.
I reproduce a table from Rao (2014) in the Appendix that summarizes data on all
policies that were introduced as part of the NRHM and how they compare for high
and non-focus Indian states, as measured in 2007-08 (Table 4.7). The main take away
from the table is to show the dierential implementation of the ASHA worker program
in high focus states when compared to non-focus states. The percentage of villages
with an ASHA worker is nearly twice as much in high focus states in comparison with
non-focus states with a higher completion rate of training. Rao (2014) estimates one
ASHA worker per 14,000 people in all high focus states taken together and one ASHA
worker per population of 75,000 people in non focus states during 2005-06. By 2008,
the intensity of ASHA workers increased to one ASHA worker for every 2,900 people
in high focus states and one ASHA worker for every 16,000 people in non-focus states.
92
4.5 Empirical Model
Following Rao (2014), I exploit the initial policy statement of the Indian govern-
ment in the implementation of the program. The Government of India mandated
that an ASHA worker be introduced, in every village with a population of 1000, in
18 high-focus Indian states, which have weak public health indicators and/or weak
infrastructure. It was also decided that areas of non-focus states with high concentra-
tion of `scheduled tribes' would receive ASHA volunteers. While Rao (2014) only uses
the policy variation across states, I utilize the assignment rule for villages, namely
the population cuto of 1000, as of the Indian Population Census, 2001. This rule
provides me the discontinuity to estimate the impact of provision of these community
health workers. As of the date of the DLHS survey, there are many villages with pop-
ulations exceeding 1,000 that are yet to receive ASHA workers in high focus states.
Thus probability of receiving the treatment does not change from 0 to 1 as we cross
the population threshold. Instead there is a smaller jump in the probability and,
therefore, use a fuzzy RD design.
I show the jump in probability in Table 4.2 and in Figure 4.1. Table 4.2 shows
the regression results from regressing the treatment (if the village has ASHA imple-
mented) on polynomial of village population, village level control variables (described
below) and a dummy variable for whether the village population is greater than or
equal to 1,000. Clearly, the likelihood of a village getting the program implemented
increases by 7% when we cross the cuto of 1,000 for high focus states. As expected,
no signicant jump is observed for low focus states (Table 4.3). I show this further in
Figure 4.1 | there is a distinct (although not very large) discontinuous jump in the
non-parametric ts for treatment on population at the cuto, thus further justifying
93
the use of fuzzy RDD in my estimation.
I use the two-stage least squares (2SLS) to estimate the causal impact of ASHA
program on various outcomes by instrumenting the treatment status of a village with
binary variable of whether the village has any habitation with a population exceeding
1000. I estimate the treatment eect using the following specication.
First Stage:
T
vd
=
0
+
1
D
vd
+
2
f(X
vd
) +
0
ivd

6
+
d
+
ivd
(4.1)
Second Stage:
O
ivd
=
0
+
1
T
vd
+
2
f(X
vd
) +
0
ivd

6
+
d
+
ivd
(4.2)
O
ivd
implies an outcome which is related to an individual i, living in village v
located in district d. T
vd
is treatment status
5
of the village, X
vd
is population of
the village as per Indian Census 2001, and D
vd
= I(Population >= 1000). f(X
vd
)
is a fourth order polynomial of the village population. I also include district xed
eects (
d
) to control for unobserved factors at the district level.
ivd
are village,
household and individual controls. Village controls include logarithm of total num-
ber of households in the village, distance dummies to the nearest town, district head
quarter, bus station and railway station, dummy if the main water source is piped,
dummies for dierent drainage and irrigation facilities in the village, if the village is
electried, dummy if the main crop produced in the village is wheat, dummies for
5
A village is dened as treated if it has ASHA worker visiting at the time of interview. This
treatment status will not vary across individuals within a village.
94
dierent health facilities in the village and if the village has an ultrasound facility
in 5 kms. Household controls include top 4 wealth quintiles (lowest wealth quintile
is the omitted category), number of females in the household, dummies for religion
of household head and if the household head belongs to scheduled caste or tribe or
any other backward classes. Finally individual controls include dummies for the re-
spondent's age, dummies for years of education, occupation dummies, husband's age
dummies and his years of education dummies.
It should be noted, however, a massive nationwide road construction project viz.
Pradhan Mantri Gram Sadak Yojana (PMGSY) was also ongoing when ASHA pro-
gram was launched (Banerjee and Sachdeva, 2014). Under this program, the govern-
ment of India mandates to bring all unconnected habitations with a population of at
least 500 within reach to the nearest link road via an all-weather road. It prioritized
the road provision based on population cutos, rst providing access to an all-weather
road to unconnected habitations with a population of 1000 or more followed by un-
connected habitations with a population of 500 or more. To control for this possible
confounding, I use a dummy if the village has all weather roads and its interaction
with the population cut o in each equation.
I also consider another model to examine whether or not the impact of the ASHA
program is dierent across the SC/ST/OBC and rest of the population. To do that, I
examine the co-ecients for treatment status and treatment status interacted by an
indicator for SC/ST/OBC status of an individual. Since these variables are endoge-
nous, I instrument them with I(Population >= 1000) and its interaction with the
indicator for SC/ST/OBC status. The remaining specication of this model is same
as the earlier one.
95
4.6 Estimation Results
4.6.1 Miscarriage and Birth Time Complications
I present the results for the impact of ASHA on miscarriage and birth time complica-
tions in Table 4.4. In the top panel, I show the estimation results for the rst model.
The results from the model that accounts for the possible heterogeneous impact on
the backward classes are shown in the bottom panel. The underlying regression mod-
els include district xed eects, four degree polynomial in village population and a
host of individual, household and village level covariates. While the rst column
refers to all births since 2004, the remaining columns are for the last birth only, be-
cause of data limitations. Overall, the results show an improvement in lowering the
incidence of miscarriages or birth complications. Women in villages with population
exceeding 1; 000 are 7% less likely to have a miscarriage, since 2004 (top panel, second
column). The magnitude is quite large, but is not signicant at conventional levels.
I also nd women in treatment villages are 9% less likely to have experienced breech
presentation, which is statistically signicant at 10% level. I nd lower chances of
high blood pressure or convulsion during delivery and higher incidence of normal de-
livery (as opposed to cesarean section or assisted), but they have large standard errors.
Moving on to the bottom panel, I nd a mother from backward section of the
community is marginally less likely to have miscarriages than a mother from the
non-backward section, although not statistically signicant. However, it appears that
mothers from such backward communities are less impacted by the provision of ASHA
workers in terms of lessening birth complications, as compared to the mothers from
privileged sections. For instance, the incidence of breech presentation has gone down
96
in absolute terms for the mothers from backward sections, but the impact is lower for
them in relative terms. The relevant co-ecients are jointly signicant at 5%. Same
holds true for other measures, but are not statistically signicant at conventional
levels.
4.6.2 Usage of Antenatal Care (ANC) Services
Results for usage of Antenatal Care Services are presented in Table 4.5. I consider
2 outcomes in this connection: if the mother has received any antenatal care during
the pregnancy, and if the delivery did not take place at home. As earlier, top panel
contains the results for the rst model, while the bottom panel pertains the model,
accounting for population heterogeneity. A mother from a village with population of
1; 000 or above, is 12% more likely to receive any antenatal care during the pregnancy
and 11% more likely to not give birth in home environment. Although the magnitudes
are quite large compared to the mean, but are not statistically signicant for large
standard errors.
The results in the bottom panel tentatively suggest that the impact is much
stronger for SC/ST/OBC population. In fact, the magnitudes are around 50% larger,
compared to the mothers from the non-backward sections. However, large standard
errors render these results statistically signicant.
4.6.3 Vaccinations and Vitamin Supplementation
Table 4.6 presents the results on incidence of vaccination and Vitamin A supplemen-
tation. The survey collects this information for last two children, born after 2004. I
97
rst report the likelihood if the child is fully vaccinated
6
. Then I consider the same
variable while using a model where I include another co-variate viz. an indicator if the
mother has shown the vaccination card to the interviewer. The nal two outcomes
are if the child has received any Vitamin A supplementation and if vaccination is
provided against Hepatitis B. I restrict the sample to children who are at least one
year old at the time of survey. As is evident from the table, the incidence rates of
vaccination or vitamin supplementation are a bit higher for the boys as compared to
girls. There is large cross-disciplinary literature on India which has the same nd-
ing. It is argued that high degree of son preference is the underlying reason for the
preferential treatment of boys (Ahsan and Bhowmick, 2015a; Bharadwaj and Lak-
dawala, 2013; Carvalho et al., 2013; Pande, 2003). So I show the results separately
for girl and boy children in the table. I nd a stronger impact of ASHA program
on most of the outcomes for girl children. For instance, a girl from a village with
population 1; 000 and above is around 30% more likely to be fully vaccinated (refer to
the upper panel). When I include the indicator for vaccination card shown, the mag-
nitude goes down marginally to 28% and remains statistically signicant at 10%. I
also nd the program to have positive impact on Vitamin A supplementation or vacci-
nation against Hepatitis B for both girls and boys, though not statistically signicant.
The results in the bottom panel seem to imply that the impact of ASHA program is
greater for children from backward communities. For instance, a SC/ST/OBC girl is
50% more likely to be completely vaccinated in a village with ASHA workers, relative
6
\According to the guidelines developed by the World Health Organization, children are con-
sidered fully vaccinated when they have received a vaccination against tuberculosis (BCG), three
doses of the diphtheria, whooping cough (pertussis), and tetanus (DPT) vaccine; three doses of the
poliomyelitis (polio) vaccine; and one dose of the measles vaccine by the age of 12 months. BCG
should be given at birth or at rst clinical contact, DPT and polio require three vaccinations at
approximately 4, 8, and 12 weeks of age, and measles should be given at or soon after reaching 9
months of age "(NATIONAL FAMILY HEALTH SURVEY (NFHS-3) Report 2005-06).
98
to a girl who is not born to a SC/ST/OBC family (second column of the bottom
panel). A similar result holds true for boys, although not statistically signicant.
4.6.4 Robustness
Following Angrist and Lavy (1999), I report estimates in Table 4.8 - Table 4.13 report
estimates using two discontinuity samples | one for +/-500 and another for +/-800.
They correspond to 40% and 70% of the entire sample. I show the results for the
model, specied by equations (4.1) and (4.2). The direction for most of the estimates
is unchanged while are substantially larger, at times than those for the full sample.
However, the estimates are now less precise as is evident from the huge standard
errors.
4.6.5 Incidence of Disease and Prevention
In this section, I ask the question if the introduction of a Community Health Worker
lowers the incidence of child diseases. More specically, does ASHA reduce the like-
lihood of a child to suer from diarrhea or fever and cough within past two weeks
of interview? I show the results in the Appendix (Table 4.14 and Table 4.15). The
results are shown separately for boys and girls. For each gender, I present the results
separately for three age groups | age less than 6 months, age between 6 and 23
months and age more than or equal to 24 months. For villages with population of
1; 000 or more, I nd a girl child belonging to the age group of 6 months or less is
30% less likely to suer from incidence of diarrhea (second column from the upper
panel of Table 4.14). The same is true if I consider the incidence of fever or cough
(second column from the upper panel of Table 4.15). However, none of these results
are statistically signicant at conventional levels.
99
I also examine if a child is given drugs for intestinal worms (a major cause for
diarrhea) within last 6 months of interview. I report the results in Table 4.16 of the
Appendix. There seems to be a tentative evidence that girl children from the lowest
age group are more likely to receive the drug, although not statistically signicant
7
.
4.7 Discussion and Conclusion
The current chapter estimates the impact of introduction of community health work-
ers in India on a variety of pregnancy outcomes as well as antenatal and postnatal
investments. I use a Fuzzy Regression Discontinuity Design for the evaluation of this
program since there is a population cut-o rule to implement the program. My results
show an improvement in some of the outcomes considered. Mothers from treatment
villages are less likely to experience birth time complications like breech presentation
and are somewhat less prone to miscarriages. I investigate one channel through which
such benecial pregnancy outcomes could be possible, viz. higher receipt of antenatal
health services. Provision of community health workers is tentatively associated with
higher usage of receiving any antenatal care. I also nd higher rates of vaccination
and vitamin supplementation, mainly for girl children in the treatment villages. One
of the interesting ndings is that ASHA workers seem to be especially successful in
reaching out to the backward sections of the community in providing services like
vaccination or antenatal services. The results are robust if I restrict the sample more
tightly around the population cut-o of 1000. I do not nd any consistent or signif-
icant impact of community health workers on incidence of diseases like diarrhea or
fever and cough.
7
Interestingly, a boy child from a treatment village is less likely to receive the drug (refer to the
seventh column of the upper panel) if he belongs to the age category of 6 and 23 months.
100
One of the main data limitations in my analysis is the limited sample frame. The
survey, used in this chapter was carried out within only three to four years of the
inception of ASHA program. Although the Government of India is successful in re-
cruiting a signicant number of workers within a short span of time, the impact of
the program is still not sizable in many domains as is evident from the results in this
chapter. Another limitation of my data is the unavailability of the year in which the
program was initiated in a village. Information on this front could have given some
indication of the extent of temporal impact of the program.
An important area of future research would be to nd out the impact of this
community health worker program in India on more short and long run health out-
comes. Since health as a measure of human capital is highly multi-dimensional,
multiple health indicators are always preferred while investigating human capital for-
mation (Strauss and Thomas, 1998; Ahsan and Bhowmick, 2015a; Bhalotra and Rawl-
ings, 2011). With the exception of the evaluation of a similar program in Indonesia
(Frankenberg and Thomas, 2001, 2000; Frankenberg et al., 2005, 2009; Weaver et al.,
2013; Ahsan and Bhowmick, 2015b), the evidence on the eectiveness of community
health workers on a wide variety of health measures and health investments in devel-
oping countries is scant. Any future survey on this topic should address this issue with
utmost importance. Also, evidence from Indonesia shows an increase in body mass
index for reproductive age women for communities which received a midwife. I could
not examine such an outcome for India due to data limitations. But my results on
lower birth complications or better vaccinations for girl children suggest that women
in Indian villages are better-o with the introduction of community health workers.
Since high son preference is prevalent in large parts of India, one of the primary jobs
101
of the health workers is to increase awareness about benets of having a girl child
8
.
My results indicate these workers have met with modest success in this endeavor. It
would be interesting to see if this program is able to improve calorie intake, labor
force participation as well as mental health and subjective welfare for reproductive
age women. Considering such a wide variety of outcomes would denitely enrich our
understanding of the ecacy of such a large scale government-funded program.
8
Refer to "National Rural Health Mission: Framework for Implementation 2005-2012" by Min-
istry of Health and Family Welfare, Government of India.
102
4.8 Tables and Figures
Figure 4.1: Probability of receipt of ASHA program and Village
population
103
Figure 4.2: Distribution of Villages with respect to Population
104
Table 4.1: Village Level Correlates for ASHA Program
Coef. Std. Err. t p value
High Focus States 0.46151 0.0072 64.08 0
Logarithm of number of Households 0.03288 0.00287 11.46 0
Main water source in Village is piped -0.02045 0.00631 -3.24 0.001
No Bus Station in the Village -0.06182 0.00806 -7.67 0
District Headquarters is more than 15 kms away -0.01405 0.00837 -1.68 0.093
Railway Station is more than 25 kms away -0.05915 0.00609 -9.71 0
Nearest town is more than 5 kms away -0.02458 0.0073 -3.37 0.001
ICDS program in the Village 0.18898 0.01278 14.79 0
A Health Subcenter is in the Village 0.04518 0.00653 6.92 0
A Primary Health Center in the Village -0.04242 0.01108 -3.83 0
A Block Primary Health Center in the Village -0.06824 0.01561 -4.37 0
A Community Health Center in the Village -0.08031 0.018 -4.46 0
District or Government Hospital in the Village -0.05967 0.01566 -3.81 0
Government Dispensary in the Village 0.02531 0.01443 1.75 0.079
Private Clinic in the Village 0.00895 0.00917 0.98 0.329
Private Hospital or Nursing Home in the Village -0.14081 0.01563 -9.01 0
AYUSH Health Facility in the Village -0.02339 0.01128 -2.07 0.038
An Ultrasound facility within 5 kms -0.0039 0.0103 -0.38 0.705
Mean of Dep. 0.603
SD of Dep. 0.489
N 22,499
Robust standard errors are reported (*** p<0.01, ** p<0.05, * p<0.1).
105
Table 4.2: High Focus States: Village Level, First Stage regression
Coef. Std. Err. t p value
I(Population>= 1000) 0.0687*** 0.019 3.630 0.000
District Fixed Eects Yes
Controls Yes
Mean of Dep. 0.740
SD of Dep. 0.439
N 15,044
F 13.21
Robust standard errors are reported (*** p<0.01, ** p<0.05, * p<0.1). The specication
includes district xed eects and village controls, which include a fourth order polynomial
of village population according to 2001 Census, indicator for all-weather roads in the village
and its interaction with an indicator of population exceeding 1000, logarithm of total number
of households in the village, distance dummies to the nearest town, district head quarter,
bus station and railway station, dummy if the main water source is piped, dummies for
dierent drainage and irrigation facilities in the village, if the village is electried, dummy
if the main crop produced in the village is wheat, dummies for dierent health facilities in
the village and if the village has an ultrasound facility in 5 kms.
106
Table 4.3: Low Focus States: Village Level, First Stage regression
Coef. Std. Err. t p value
I(Population>= 1000) -0.0269 0.029 -0.940 0.346
District Fixed Eects Yes
Controls Yes
Mean of Dep. 0.329
SD of Dep. 0.329
N 7,553
F 0.89
Robust standard errors are reported(*** p<0.01, ** p<0.05, * p<0.1). The specication
includes district xed eects and village controls, which include a fourth order polynomial
of village population according to 2001 Census, indicator for all-weather roads in the village
and its interaction with an indicator of population exceeding 1000, logarithm of total number
of households in the village, distance dummies to the nearest town, district head quarter,
bus station and railway station, dummy if the main water source is piped, dummies for
dierent drainage and irrigation facilities in the village, if the village is electried, dummy
if the main crop produced in the village is wheat, dummies for dierent health facilities in
the village and if the village has an ultrasound facility in 5 kms.
107
Table 4.4: Impact of ASHA on Miscarriage and Birth Complications
Miscarriage Breech High BP or Normal
Presentation Convulsion Delivery
Model 1
ASHA program -0.073 -0.086* -0.033 0.024
(0.044) (0.050) (0.055) (0.046)
Mean of Dep. 0.047 0.051 0.055 0.942
N 183192 130682 130678 130682
Model 2
ASHA program -0.0642 -0.129** -0.0574 0.0518
(0.0463) (0.0540) (0.0579) (0.0518)
ASHA program X Backward Class -0.0109 0.0556** 0.0309 -0.0354
(0.0208) (0.0255) (0.0267) (0.0256)
Mean of Dep. 0.047 0.051 0.055 0.942
N 183192 130682 130678 130682
p-value of Joint Signicance 0.2072 0.0276 0.4329 0.3569
Fuzzy RDD is used for estimation and only High Focus States are considered. Standard errors
are clustered at village level (*** p<0.01, ** p<0.05, * p<0.1). The specications include district
xed eects. Village controls include a fourth order polynomial of village population according to
2001 Census, indicator for all-weather roads in the village and its interaction with an indicator of
population exceeding 1000, logarithm of total number of households in the village, distance dummies
to the nearest town, district head quarter, bus station and railway station, dummy if the main water
source is piped, dummies for dierent drainage and irrigation facilities in the village, if the village is
electried, dummy if the main crop produced in the village is wheat, dummies for dierent health
facilities in the village and if the village has an ultrasound facility in 5 kms. Household controls
include top 4 wealth quintiles (lowest wealth quintile is the omitted category), number of females in
the household, dummies for religion of household head and if the household head belongs to scheduled
caste or tribe or any other backward classes. Finally individual controls include dummies for the
respondent's age, dummies for years of education, occupation dummies, husband's age dummies and
his years of education dummies.
108
Table 4.5: Impact of Asha on Antenatal Care Usage
Any Antenatal Care Birth not at
Home
Model 1
ASHA program 0.115 0.108
(0.128) (0.098)
Mean of Dep. 0.635 0.300
N 130686 130680
Model 2
ASHA program 0.0873 0.0733
(0.129) (0.103)
ASHA program X Backward Class 0.0361 0.0448
(0.0482) (0.0454)
Mean of Dep. 0.635 0.300
N 130686 130680
p-value of Joint Signicance 0.4765 0.2995
Fuzzy RDD is used for estimation and only High Focus States are considered. Stan-
dard errors are clustered at village level (*** p<0.01, ** p<0.05, * p<0.1). The
specications include district xed eects. Village controls include a fourth order
polynomial of village population according to 2001 Census, indicator for all-weather
roads in the village and its interaction with an indicator of population exceeding
1000, logarithm of total number of households in the village, distance dummies to
the nearest town, district head quarter, bus station and railway station, dummy if
the main water source is piped, dummies for dierent drainage and irrigation fa-
cilities in the village, if the village is electried, dummy if the main crop produced
in the village is wheat, dummies for dierent health facilities in the village and if
the village has an ultrasound facility in 5 kms. Household controls include top 4
wealth quintiles (lowest wealth quintile is the omitted category), number of females
in the household, dummies for religion of household head and if the household head
belongs to scheduled caste or tribe or any other backward classes. Finally individual
controls include dummies for the respondent's age, dummies for years of education,
occupation dummies, husband's age dummies and his years of education dummies.
109
Table 4.6: Impact of Asha on Vaccinations
Girl Child Boy Child
Complete Vaccination Vitamin A Vaccination against Complete Vaccination Vitamin A Vaccination against
Hepatitis B Hepatitis B
Not Conrolling for Conrolling for Not Conrolling for Conrolling for
Vaccination Card Vaccination Card Vaccination Card Vaccination Card
Model 1
ASHA program 0.310* 0.275* 0.159 0.014 0.226 0.211* 0.023 0.167
(0.186) (0.146) (0.167) (0.106) (0.163) (0.128) (0.158) (0.115)
Mean of Dep. 0.454 0.454 0.484 0.145 0.476 0.476 0.505 0.159
N 57702 57702 57702 57702 62709 62709 62709 62709
Model 2
ASHA program 0.223 0.265* 0.175 0.0351 0.183 0.176 -0.0306 0.138
(0.191) (0.149) (0.172) (0.112) (0.165) (0.128) (0.159) (0.119)
ASHA program X Backward Class 0.111 0.0125 -0.0204 -0.0268 0.0577 0.0465 0.0705 0.0382
(0.0757) (0.0599) (0.0709) (0.0546) (0.0658) (0.0515) (0.0641) (0.0536)
Mean of Dep. 0.454 0.454 0.484 0.145 0.476 0.476 0.505 0.159
N 57702 57702 57702 57702 62709 62709 62709 62709
p-value of Joint Signicance 0.081 0.1545 0.5949 0.8797 0.2626 0.1814 0.5439 0.2582
Fuzzy RDD is used for estimation and only High Focus States are considered. Standard errors are clustered at village level (*** p<0.01, ** p<0.05, * p<0.1). The specications include district xed eects.
Village controls include a fourth order polynomial of village population according to 2001 Census, indicator for all-weather roads in the village and its interaction with an indicator of population exceeding
1000, logarithm of total number of households in the village, distance dummies to the nearest town, district head quarter, bus station and railway station, dummy if the main water source is piped, dummies
for dierent drainage and irrigation facilities in the village, if the village is electried, dummy if the main crop produced in the village is wheat, dummies for dierent health facilities in the village and if
the village has an ultrasound facility in 5 kms. Household controls include top 4 wealth quintiles (lowest wealth quintile is the omitted category), number of females in the household, dummies for religion
of household head and if the household head belongs to scheduled caste or tribe or any other backward classes. Finally individual controls include dummies for the respondent's age, dummies for years of
education, occupation dummies, husband's age dummies and his years of education dummies.
110
4.9 Appendix
Table 4.7: Different NRHM provisions in High and Non-Focus states
(Reproduced from Rao (2014)
High-Focus Non-Focus
% of villages with an ASHA worker 74.02 33.04
Average number of ASHA workers who have completed rst round of training in Sub-Centre area 4.6 1.66
% of Sub-Centers that have received untied funds 76.26 89.14
% of Sub-Centers that have fully utilized untied funds 28.92 50.13
% of Sub-Centre areas that have a Village Health & Sanitation Committee 63.96 84.64
% of Primary Health Centers that have received untied funds 65.34 87.2
% of Primary Health Centers that have fully utilized untied funds 24.26 48.02
% of Community Health Centers that have received untied funds 81.82 90.09
% of Community Health Centers that have fully utilized untied funds 32.62 46.06
% of District Hospitals with a Rogi Kalyan Samiti (RKS) 86.61 93.43
111
Table 4.8: Robustness Results (+/-500): Miscarriage and Birth
Complications
Miscarriage Breech High BP or Birth not at
Presentation Convulsion Home
ASHA program 0.199 -0.404 -0.0910 0.419
(0.452) (0.806) (0.451) (1.075)
N 102,270 55,800 55,796 55,797
Fuzzy RDD is used for estimation and only High Focus States are considered. Stan-
dard errors are clustered at village level (*** p<0.01, ** p<0.05, * p<0.1). Sample
is restricted to villages with population between 500 and 1500. The specications
include district xed eects. Village controls include a fourth order polynomial of
village population according to 2001 Census, indicator for all-weather roads in the
village and its interaction with an indicator of population exceeding 1000, logarithm
of total number of households in the village, distance dummies to the nearest town,
district head quarter, bus station and railway station, dummy if the main water
source is piped, dummies for dierent drainage and irrigation facilities in the vil-
lage, if the village is electried, dummy if the main crop produced in the village
is wheat, dummies for dierent health facilities in the village and if the village has
an ultrasound facility in 5 kms. Household controls include top 4 wealth quintiles
(lowest wealth quintile is the omitted category), number of females in the house-
hold, dummies for religion of household head and if the household head belongs
to scheduled caste or tribe or any other backward classes. Finally individual con-
trols include dummies for the respondent's age, dummies for years of education,
occupation dummies, husband's age dummies and his years of education dummies.
112
Table 4.9: Robustness Results (+/-500): Antenatal Care Usage
Normal Any Antenatal Care
Delivery
ASHA program 0.151 0.855
(0.445) (1.751)
N 55,800 55,799
Fuzzy RDD is used for estimation and only High Focus States
are considered. Standard errors are clustered at village level (***
p<0.01, ** p<0.05, * p<0.1). Sample is restricted to villages with
population between 500 and 1500. The specications include dis-
trict xed eects. Village controls include a fourth order polyno-
mial of village population according to 2001 Census, indicator for
all-weather roads in the village and its interaction with an indicator
of population exceeding 1000, logarithm of total number of house-
holds in the village, distance dummies to the nearest town, district
head quarter, bus station and railway station, dummy if the main
water source is piped, dummies for dierent drainage and irrigation
facilities in the village, if the village is electried, dummy if the main
crop produced in the village is wheat, dummies for dierent health
facilities in the village and if the village has an ultrasound facility
in 5 kms. Household controls include top 4 wealth quintiles (lowest
wealth quintile is the omitted category), number of females in the
household, dummies for religion of household head and if the house-
hold head belongs to scheduled caste or tribe or any other backward
classes. Finally individual controls include dummies for the respon-
dent's age, dummies for years of education, occupation dummies,
husband's age dummies and his years of education dummies.
113
Table 4.10: Robustness Results (+/-500): Vaccinations
Girl Child Boy Child
Complete Vaccination Vitamin A Vaccination against Complete Vaccination Vitamin A Vaccination against
Hepatitis B Hepatitis B
Not Conrolling for Conrolling for Not Conrolling for Conrolling for
Vaccination Card Vaccination Card Vaccination Card Vaccination Card
ASHA program 0.0452 0.0559 -0.000343 -0.0134 0.168 0.200** 0.223* 0.148*
(0.126) (0.0957) (0.126) (0.0866) (0.110) (0.0884) (0.115) (0.0829)
N 24,665 24,665 24,665 24,665 26,415 26,415 26,415 26,415
Fuzzy RDD is used for estimation and only High Focus States are considered. Standard errors are clustered at village level (*** p<0.01, ** p<0.05, * p<0.1). Sample is restricted to villages with population
between 500 and 1500. The specications include district xed eects. Village controls include a fourth order polynomial of village population according to 2001 Census, indicator for all-weather roads in
the village and its interaction with an indicator of population exceeding 1000, logarithm of total number of households in the village, distance dummies to the nearest town, district head quarter, bus station
and railway station, dummy if the main water source is piped, dummies for dierent drainage and irrigation facilities in the village, if the village is electried, dummy if the main crop produced in the village
is wheat, dummies for dierent health facilities in the village and if the village has an ultrasound facility in 5 kms. Household controls include top 4 wealth quintiles (lowest wealth quintile is the omitted
category), number of females in the household, dummies for religion of household head and if the household head belongs to scheduled caste or tribe or any other backward classes. Finally individual controls
include dummies for the respondent's age, dummies for years of education, occupation dummies, husband's age dummies and his years of education dummies.
114
Table 4.11: Robustness Results (+/-800): Miscarriage and Birth
Complications
Miscarriage Breech High BP or Birth not at
Presentation Convulsion Home
ASHA program 0.132 -0.0484 -0.0991 0.359
(0.185) ((0.140)) (0.173) (0.375)
N 158,147 88,764 88,759 88,762
Fuzzy RDD is used for estimation and only High Focus States are considered. Stan-
dard errors are clustered at village level (*** p<0.01, ** p<0.05, * p<0.1). Sample
is restricted to villages with population between 200 and 1800. The specications
include district xed eects. Village controls include a fourth order polynomial of
village population according to 2001 Census, indicator for all-weather roads in the
village and its interaction with an indicator of population exceeding 1000, logarithm
of total number of households in the village, distance dummies to the nearest town,
district head quarter, bus station and railway station, dummy if the main water
source is piped, dummies for dierent drainage and irrigation facilities in the vil-
lage, if the village is electried, dummy if the main crop produced in the village
is wheat, dummies for dierent health facilities in the village and if the village has
an ultrasound facility in 5 kms. Household controls include top 4 wealth quintiles
(lowest wealth quintile is the omitted category), number of females in the house-
hold, dummies for religion of household head and if the household head belongs
to scheduled caste or tribe or any other backward classes. Finally individual con-
trols include dummies for the respondent's age, dummies for years of education,
occupation dummies, husband's age dummies and his years of education dummies.
115
Table 4.12: Robustness Results (+/-800): Antenatal Care Usage
Normal Any Antenatal Care
Delivery
ASHA program -0.0566 0.161
(0.134) (0.356)
N 88,759 88,763
Fuzzy RDD is used for estimation and only High Focus States
are considered. Standard errors are clustered at village level (***
p<0.01, ** p<0.05, * p<0.1). Sample is restricted to villages with
population between 200 and 1800. The specications include dis-
trict xed eects. Village controls include a fourth order polyno-
mial of village population according to 2001 Census, indicator for
all-weather roads in the village and its interaction with an indicator
of population exceeding 1000, logarithm of total number of house-
holds in the village, distance dummies to the nearest town, district
head quarter, bus station and railway station, dummy if the main
water source is piped, dummies for dierent drainage and irrigation
facilities in the village, if the village is electried, dummy if the main
crop produced in the village is wheat, dummies for dierent health
facilities in the village and if the village has an ultrasound facility
in 5 kms. Household controls include top 4 wealth quintiles (lowest
wealth quintile is the omitted category), number of females in the
household, dummies for religion of household head and if the house-
hold head belongs to scheduled caste or tribe or any other backward
classes. Finally individual controls include dummies for the respon-
dent's age, dummies for years of education, occupation dummies,
husband's age dummies and his years of education dummies.
116
Table 4.13: Robustness Results (+/-800): Vaccinations
Girl Child Boy Child
Complete Vaccination Vitamin A Vaccination against Complete Vaccination Vitamin A Vaccination against
Hepatitis B Hepatitis B
Not Conrolling for Conrolling for Not Conrolling for Conrolling for
Vaccination Card Vaccination Card Vaccination Card Vaccination Card
ASHA program 0.386 0.388 -0.560 0.502 0.406 0.206 -0.679 0.757
(0.965) (0.819) (1.143) (0.893) (0.666) (0.468) (0.867) (0.822)
N 39,210 39,210 39,210 39,210 42,279 42,279 42,279 42,279
Fuzzy RDD is used for estimation and only High Focus States are considered. Standard errors are clustered at village level (*** p<0.01, ** p<0.05, * p<0.1). Sample is restricted to villages with population
between 200 and 1800. The specications include district xed eects. Village controls include a fourth order polynomial of village population according to 2001 Census, indicator for all-weather roads in
the village and its interaction with an indicator of population exceeding 1000, logarithm of total number of households in the village, distance dummies to the nearest town, district head quarter, bus station
and railway station, dummy if the main water source is piped, dummies for dierent drainage and irrigation facilities in the village, if the village is electried, dummy if the main crop produced in the village
is wheat, dummies for dierent health facilities in the village and if the village has an ultrasound facility in 5 kms. Household controls include top 4 wealth quintiles (lowest wealth quintile is the omitted
category), number of females in the household, dummies for religion of household head and if the household head belongs to scheduled caste or tribe or any other backward classes. Finally individual controls
include dummies for the respondent's age, dummies for years of education, occupation dummies, husband's age dummies and his years of education dummies.
117
Table 4.14: Impact of Asha on Incidence of Diarrhea on Children
Girl Child Boy Child
Age less than Age between 6 and Age more than Age less than Age between 6 and Age more than
6 months 23 months 24 months 6 months 23 months 24 months
Model 1
ASHA program -0.305 0.056 0.019 0.055 0.058 0.018
(0.301) (0.144) (0.116) (0.212) (0.127) (0.107)
Mean of Dep. 0.124 0.159 0.084 0.134 0.159 0.091
N 10227 28313 38904 11036 30932 42016
Model 2
ASHA program -0.366 0.0551 0.0340 0.0507 -0.00972 0.0333
(0.287) (0.156) (0.120) (0.250) (0.147) (0.104)
ASHA program X Backward Class 0.0969 0.00167 -0.0200 0.00462 0.0803 -0.0209
(0.117) (0.0682) (0.0485) (0.110) (0.0706) (0.0426)
Mean of Dep. 0.124 0.159 0.084 0.134 0.159 0.091
N 10227 28313 38904 11036 30932 42016
p-value of Joint Signicance 0.3151 0.9232 0.9067 0.9626 0.4144 0.8561
Fuzzy RDD is used for estimation and only High Focus States are considered. Standard errors are clustered at village level (*** p<0.01, ** p<0.05, *
p<0.1). The specications include district xed eects. Village controls include a fourth order polynomial of village population according to 2001 Census,
indicator for all-weather roads in the village and its interaction with an indicator of population exceeding 1000, logarithm of total number of households
in the village, distance dummies to the nearest town, district head quarter, bus station and railway station, dummy if the main water source is piped,
dummies for dierent drainage and irrigation facilities in the village, if the village is electried, dummy if the main crop produced in the village is wheat,
dummies for dierent health facilities in the village and if the village has an ultrasound facility in 5 kms. Household controls include top 4 wealth quintiles
(lowest wealth quintile is the omitted category), number of females in the household, dummies for religion of household head and if the household head
belongs to scheduled caste or tribe or any other backward classes. Finally individual controls include dummies for the respondent's age, dummies for years
of education, occupation dummies, husband's age dummies and his years of education dummies.
118
Table 4.15: Impact of Asha on Incidence of Fever and Cough of Children
Girl Child Boy Child
Age less than Age between 6 and Age more than Age less than Age between 6 and Age more than
6 months 23 months 24 months 6 months 23 months 24 months
Model 1
ASHA program -0.432 0.272 0.253 -0.147 0.282 0.1
(0.396) (0.200) (0.194) (0.289) (0.185) (0.163)
Mean of Dep. 0.237 0.310 0.237 0.253 0.323 0.251
N 10227 28313 38904 11036 30932 42016
Model 2
ASHA program -0.385 0.209 0.271 -0.319 0.253 0.101
(0.386) (0.217) (0.201) (0.347) (0.211) (0.162)
ASHA program X Backward Class -0.0764 0.0757 -0.0241 0.208 0.0356 -0.00145
(0.154) (0.0938) (0.0806) (0.150) (0.0973) (0.0684)
Mean of Dep. 0.237 0.310 0.237 0.253 0.323 0.251
N 10227 28313 38904 11036 30932 42016
p-value of Joint Signicance 0.5425 0.2486 0.396 0.3807 0.2318 0.8221
Fuzzy RDD is used for estimation and only High Focus States are considered. Standard errors are clustered at village level (*** p<0.01, ** p<0.05, *
p<0.1). The specications include district xed eects. Village controls include a fourth order polynomial of village population according to 2001 Census,
indicator for all-weather roads in the village and its interaction with an indicator of population exceeding 1000, logarithm of total number of households
in the village, distance dummies to the nearest town, district head quarter, bus station and railway station, dummy if the main water source is piped,
dummies for dierent drainage and irrigation facilities in the village, if the village is electried, dummy if the main crop produced in the village is wheat,
dummies for dierent health facilities in the village and if the village has an ultrasound facility in 5 kms. Household controls include top 4 wealth quintiles
(lowest wealth quintile is the omitted category), number of females in the household, dummies for religion of household head and if the household head
belongs to scheduled caste or tribe or any other backward classes. Finally individual controls include dummies for the respondent's age, dummies for years
of education, occupation dummies, husband's age dummies and his years of education dummies.
119
Table 4.16: Impact of Asha on Receipt of Drugs against Intestinal Worms
Girl Child Boy Child
Age less than Age between 6 and Age more than Age less than Age between 6 and Age more than
6 months 23 months 24 months 6 months 23 months 24 months
Model 1
ASHA program 0.108 0.031 -0.001 -0.029 -0.299* 0.101
(0.087) (0.144) (0.142) (0.060) (0.163) (0.130)
Mean of Dep. 0.025 0.095 0.138 0.025 0.101 0.146
N 19742 18798 38904 21275 20693 42016
Model 2
ASHA program 0.0810 0.134 0.0674 -0.107 -0.261 0.152
(0.0956) (0.154) (0.153) (0.0768) (0.188) (0.132)
ASHA program X Backward Class 0.0354 -0.129* -0.0891 0.0901** -0.0467 -0.0710
(0.0384) (0.0731) (0.0635) (0.0428) (0.0824) (0.0570)
Mean of Dep. 0.025 0.095 0.138 0.025 0.101 0.146
N 19742 18798 38904 21275 20693 42016
p-value of Joint Signicance 0.252 0.2064 0.3711 0.1088 0.1104 0.3154
Fuzzy RDD is used for estimation and only High Focus States are considered. Standard errors are clustered at village level (*** p<0.01, ** p<0.05, *
p<0.1). The specications include district xed eects. Village controls include a fourth order polynomial of village population according to 2001 Census,
indicator for all-weather roads in the village and its interaction with an indicator of population exceeding 1000, logarithm of total number of households
in the village, distance dummies to the nearest town, district head quarter, bus station and railway station, dummy if the main water source is piped,
dummies for dierent drainage and irrigation facilities in the village, if the village is electried, dummy if the main crop produced in the village is wheat,
dummies for dierent health facilities in the village and if the village has an ultrasound facility in 5 kms. Household controls include top 4 wealth quintiles
(lowest wealth quintile is the omitted category), number of females in the household, dummies for religion of household head and if the household head
belongs to scheduled caste or tribe or any other backward classes. Finally individual controls include dummies for the respondent's age, dummies for years
of education, occupation dummies, husband's age dummies and his years of education dummies.
120
Chapter 5
Conclusion
One of the persistent ndings in countries like India and Indonesia is that an impres-
sive economic growth is not accompanied by similar improvements in child health
indicators. According to the CIA World Factbook 2015, the infant mortality rates
1
in 2015 for these two countries are, respectively seven times and four times the rates
in the United States. Early life health being important for subsequent health or later
life outcomes, this dissertation examines how parental human capital and provision of
health care services have benecial impact on the same in these growing, yet under-
developed economies. The ndings from India (Chapter 2) suggest that mothers with
secondary education and above from rural areas of states with high infant mortality
rates, are more likely to have a child who is male, who is also less likely to die as a
neonate, and has 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. A policy angle is also examined by evaluating the im-
pact of a community health worker program in India (Chapter 4). The results show
mothers in rural India are less likely to face certain birth time complications and
somewhat less prone to miscarriages if a health worker is present in the community.
Better usage of antenatal care services seems to be a possible pathway. Since girl
children are usually neglected in large parts of India, the community health workers
1
It is dened as number of deaths of infants under one year old in a given year per 1,000 live
births.
121
are instructed to raise awareness about benets of having a girl child. The disserta-
tion provides evidence that the presence of a community health worker is, actually
associated with greater vaccination rates for a girl, especially if she belongs to the
backward sections of the society. One can conclude from these ndings that health
policies have some potential in improving early life health, especially for the under-
privileged sections of the communities (for instance, girls and people from backward
classes in India); otherwise, parental human capital (health and education) can, at
least partially compensate for poor health infrastructure.
But can a health intervention also entail perverse outcomes? The analysis on
Indonesia hints towards that possibility (Chapter 3). Although presence of a com-
munity health worker improves maternal health (Frankenberg and Thomas, 2000),
it may lead to a deterioration in ospring quality. The results indicate women of
reproductive age are more likely to give birth to a male child after the introduction of
midwife in the community which is expected to improve nutritional status of mothers.
But the program also causes birth weight to go down for a boy child, while no change
is observed for the girl child. The main reason behind this paradoxical impact is due
to the increased chance of survival for marginal fetuses as a result of the program.
Consistent with the ndings from India, this analysis also nds mothers with dis-
advantaged background (as measured by low education) are most impacted by the
program.
There are some interesting implications of the ndings in this dissertation. Firstly,
the dissertation has analyzed the eects of certain health intervention programs on one
aspect of human capital, viz. early life health. However, a holistic welfare evaluation
should further entail examining regular economic outcomes (like wages and educa-
122
tion), health and nutritional status, mental health, and subjective welfare. Secondly,
there is growing evidence that child health intervention program can have substantial
impact on a number of later life economic outcomes. For instance, a school-based de-
worming program in Kenya is associated with increased educational attainment and
greater labor supply with interesting shifts in labor market specialization (Baird et
al., 2011). The results in the dissertation indicate that large scale government funded
health interventions can also improve early life health in developing countries. But
an important question remains:| Will these benets actually translate into greater
economic benets in the future? The answer is not very clear. The analysis in Chap-
ter 3 indicates that more male children with inferior birth weights can be born due
to a maternal health intervention program. If parents try to compensate the initial
inferior endowments of male children by making greater investments in these children,
then a simple mean comparison of such investments between boys and girls may mis-
takenly 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 hu-
man capital formation. Thus children who are born with inferior endowments, may
end up with less education and income and have greater chances of later life health
complications.
123
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," USC-INET Research Paper, 2015, (15-26).
and , \The Unintended Consequences of the Village Midwife Program in In-
donesia," 2015.
124
Almond, Douglas and Bhashkar Mazumder, \Fetal origins and parental re-
sponses," 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.
, 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.
, 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.
Angrist, Joshua D. and Victor Lavy, \Using Maimonides' Rule to Estimate
the Eect of Class Size on Scholastic Achievement," The Quarterly Journal of
Economics, 1999, 114 (2), 533{575.
125
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.
Baird, Sarah, Joan Hamory Hicks, Michael Kremer, and Edward Miguel,
\Worms at work: long-run impacts of child health gains," Berkeley: University of
California at Berkeley, 2011.
Banerjee, Rakesh and Ashish Sachdeva, \Pathways to Preventive Health, Evi-
dence from India's Rural Road Program," 2014.
Barker, DJP, \In utero programming of cardiovascular disease," Theriogenology,
2000, 53 (2), 555{574.
Basinga, Paulin, Paul J Gertler, Agnes Binagwaho, Agnes LB Soucat, Jen-
nifer Sturdy, and Christel MJ Vermeersch, \Eect on maternal and child
health services in Rwanda of payment to primary health-care providers for perfor-
mance: an impact evaluation," The Lancet, 2011, 377 (9775), 1421{1428.
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.
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.
126
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 Juan Eberhard, \Atmospheric air pollution and birth
weight," Available at SSRN 1197443, 2008.
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.
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.
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.
127
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.
Carvalho, Leandro S, Silvia H Barcellos, and Adriana Lleras-Muney, \Child
Gender and Parental Investments in India: Are Boys and Girls Treated Dier-
ently?," American Economic Journal: Applied Economics, 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.
Currie, Janet, \Inequality at Birth: Some Causes and Consequences," American
Economic Review, 2011, 101 (3), 1{22.
, 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.
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.
128
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.
Dupas, Pascaline, \Health behavior in developing countries," Annu. Rev. Econ.,
2011, 3 (1), 425{449.
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.
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, \The eects of access to health services on adult health
status: Evidence from Indonesia's \Midwife in the Village" program," Human Cap-
ital, Reproductive Health Programs, and the Role of Women in the Household and
Broader Economy: Working Paper, 2000, 1.
and , \Women's health and pregnancy outcomes: Do services make a dier-
ence?," 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.
129
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.
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.
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.
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.
130
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.
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.
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.
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.
131
Pande, Rohini P, \Selective gender dierences in childhood nutrition and immu-
nization in rural India: the role of siblings," Demography, 2003, 40 (3), 395{418.
Portner, Claus C, \Sex-selective abortions, fertility, and birth spacing," World
Bank Policy Research Working Paper, 2015, (7189).
Rao, Tanvi, \The Impact of a Community Health Worker Program on Childhood Im-
munization: Evidence from India's' ASHA'Workers," Available at SSRN 2444391,
2014.
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 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.
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.
132
Schultz, T Paul, \Population policies, fertility, women's human capital, and child
quality," Handbook of development economics, 2007, 4, 3249{3303.
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.
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.
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, \Health, nutrition, and economic development," Journal
of economic literature, 1998, pp. 766{817.
133
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.
Thomas, Duncan, Firman Witoelar, Elizabeth Frankenberg, Bondan
Sikoki, John Strauss, Cecep Sumantri, and Wayan Suriastini, \Cutting
the costs of attrition: Results 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.
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.
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.
134
Wolfe, Barbara L and Jere R Behrman, \Women's schooling and children's
health: Are the eects robust with adult sibling control for the women's childhood
background?," Journal of health Economics, 1987, 6 (3), 239{254.
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.
135

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Selection and impacts of early life events on later life outcomes

PDF

Essays on health economics

PDF

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

PDF

Intergenerational transmission of values and behaviors over the family life course

PDF

Essays on health insurance programs and policies

PDF

Intergenerational transfers & human capital investments in children in the era of aging

PDF

Essays in macroeconomics

PDF

Essays in development and experimental economics

PDF

Essays on development economics

PDF

Three essays on behavioral economics approaches to understanding the implications of mental health stigma

PDF

Three essays on cooperation, social interactions, and religion

PDF

Essays on development economics and adolescent behavior

PDF

The determinants and measurement of human capital

PDF

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

PDF

Essays on education and institutions in developing countries

PDF

Essays on development and health economics: social media and education policy

PDF

Essays on education: from Indonesia to Los Angeles, CA

PDF

Essays in labor economics: demographic determinants of labor supply

PDF

Essays in the economics of education and conflict

PDF

Three essays on the evaluation of long-term care insurance policies

Asset Metadata

Creator Bhowmick, Riddhi (author)

Core Title Three essays on economics of early life health in developing countries

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Economics

Publication Date 04/15/2016

Defense Date 03/02/2016

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

Tag antenatal care,community health workers,endowments,fetal selection,gender,Health,human capital,in utero,India,intergenerational transmission,OAI-PMH Harvest,parental response,Pregnancy,regression discontinuity,selection,sex ratio,vaccination

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Strauss, John A. (committee chair), Banerjee, Tridib (committee member), Nugent, Jeff (committee member)

Creator Email bhowmick@usc.edu,rbhowm@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-229359

Unique identifier UC11276483

Identifier etd-BhowmickRi-4260.pdf (filename),usctheses-c40-229359 (legacy record id)

Legacy Identifier etd-BhowmickRi-4260.pdf

Dmrecord 229359

Document Type Dissertation

Format application/pdf (imt)

Rights Bhowmick, Riddhi

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