Full text of Working Papers (Federal Reserve Bank of Chicago) : Prenatal Sex Selection and Girls' Well-Being : Evidence From India, Working Paper 2010-11

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

Federal Reserve Bank of Chicago

Prenatal Sex Selection and Girls’
Well‐Being: Evidence from India
Luojia Hu and Analía Schlosser

REVISED
October 2014
WP 2010-11

Prenatal Sex Selection and Girls’ Well‐Being: Evidence from India

Luojia Hu
Economic Research
Federal Reserve Bank of Chicago
230 S. LaSalle Street
Chicago, IL 60604
Phone: (312) 322 8122
e‐mail: lhu@frbchi.org

Analia Schlosser
The Eitan Berglas School of Economics
Tel Aviv University
P.O.B. 39040
Ramat Aviv, Tel Aviv, 69978
ISRAEL
Phone:+972‐54‐4902414
e‐mail: analias@post.tau.ac.il

Prenatal Sex Selection and Girls’ Well‐Being: Evidence from India1
Luojia Hu
Federal Reserve Bank of Chicago

Analía Schlosser
Tel Aviv University

October , 2014

ABSTRACT
We study the impacts of prenatal sex selection on girls’ well‐being in India. We show that
high sex ratios at birth reflect the practice of prenatal sex selection and apply a triple
difference strategy to examine whether changes in health outcomes of girls relative to
boys within states and over time are systematically associated with changes in sex‐ratios
at birth. We find that an increase in prenatal sex selection leads to a reduction in girls’
malnutrition, in particular, underweight and wasting. We further explore various
underlying channels linking between prenatal sex selection and girls’ outcomes.

Keywords: Son preference, prenatal sex selection, ultrasound, sex ratio at birth, gender
discrimination, child health.
JEL codes: J13, J16, I1, O12

1

We thank Angus Deaton, Prashant Bharadwaj, Seema Jayachandran, Elaine Liu, Nancy Qian,
Steven Pischke, two anonymous referees, and seminar participants at the Federal Reserve Bank of
Chicago, Princeton labor lunch, University of Lausanne, University of Zurich, Ben Gurion University,
Bar Ilan University, Wellesley College, UIUC, 2011 SOLE meetings, CESifo Venice Workshop, CEMFI,
Norwegian School of Economics, and NBER Summer Institute for helpful comments. We also thank
Yotam Shem Tov for excellent research assistance. The opinions expressed here are those of the
authors and not necessarily those of the Federal Reserve Bank of Chicago or the Federal Reserve
System. Schlosser thanks the Pinhas Sapir Center for research support.

I.

INTRODUCTION

Son preference in India and other Asian countries has been documented extensively. A large
number of studies have shown that parental preference for boys is manifested in gender
differences in the intrahousehold allocation of resources and medical care, which result in gender
differentials in child outcomes such as nutrition, morbidity and mortality and, in extreme cases,
female infanticide (see, for example, Arnold et al., 1998; Deaton, 1997; George, 1997; Miller,
1981; Kishor, 1993; Rose, 1999; Rosenzweig and Schultz, 1982; Pande, 2003; and Barcellos et al.,
2010). The diffusion of prenatal sex determination technologies since the 1980s has provided
parents with an alternative way to achieve the preferred sex composition of children. Indeed,
starting from the late 1980s, there has been a steep increase in male to female ratios at birth,
which has been attributed to the increasing practice of sex selective abortion. A growing number
of studies have examined the factors affecting the prevalence of prenatal sex selection and have
assessed its effect on sex ratios at birth.1 Little is known, however, about the impact of this
phenomenon on the well‐being of the girls who are born.
Prenatal sex selection may affect girls’ outcomes through different channels. First, girls might
be more likely to be born into households that have weaker son preference so that they would be
more likely to be wanted.2 Second, with access to prenatal sex selection, parents may be less likely
to rely on fertility stopping rules to achieve their desired sex composition of children so that girls
might be born into smaller families and receive a larger share of household inputs. Third, girls’
outcomes might change if the characteristics and living environment of households that do not
practice prenatal sex selection are different from those that practice it. Finally, prenatal sex
selection may affect girls’ outcomes if an increase in the scarcity of girls changes women’s value
in labor and marriage markets subsequently affecting parental incentives to invest in girls.
In this paper, we study the impact of prenatal sex selection on the well‐being of girls in India
and investigate some of the different channels through which prenatal sex selection may affect
girls’ outcomes. India is an interesting environment for studying the impact of prenatal sex
selection due to its large regional variations in parental preference for sons and the striking
differences in the use of prenatal sex selection technologies over time and across regions. Using
a comprehensive data set on child outcomes and household characteristics, we apply a triple‐

1

We review this literature in the next section.
This is a similar rationale to that proposed by Donohue and Levitt (2001) and Ananat et al. (2009) in the
context of abortion legalization in the US.

2

1

differences approach to examine whether changes in the nutritional status and mortality rates of
girls relative to boys within states and over time are systematically associated with changes in the
prevalence of prenatal sex selection. In the absence of a direct measure of parental access to
prenatal sex selection, we use the ratio of males to females at birth (MFR) in the child’s state and
birth cohort and provide evidence showing that increases in MFR at birth reflect the practice of
prenatal sex selection starting from the late 1980s. We then analyze whether changes in the
outcomes of girls within states and over time in comparison with boys can be attributed to
changes in MFR at birth. We then turn to explore the various channels that might link between
prenatal sex selection and girls’ outcomes by analyzing whether changes in MFR within states and
over time appear to be associated with changes in household characteristics, fertility,
breastfeeding practices, and son preference.
Our paper is related to a limited number of recent studies, most of them conducted
concurrently with our own, that examine the effects of prenatal sex selection on girls’ outcomes.
Shepherd (2008) compares between families with high versus low likelihood of using prenatal sex
selection in India and finds inconclusive evidence for a link between prenatal sex selection and a
decline in female child’s mortality. Lin et al. (2009) exploit variation in the use of prenatal sex
selection in Taiwan over time and birth order generated by the legalization of abortion and find a
positive association between prenatal sex selection and female survival. Almond et al. (2010) use
variation in the diffusion of ultrasound technology across regions in China and find that prenatal
sex determination is associated with an increase in girls’ neonatal mortality but do not find any
changes in postnatal investments in girls as manifested in breastfeeding duration and vaccination
rates.3
Our paper adds to this limited number of studies in several aspects. First, compared to some
other countries, India’s high degree of heterogeneity in female discrimination and the differences
in the practice of prenatal sex selection across regions makes it an optimal case for studying the
effects of prenatal sex selection. This is also due to India’s large variation in family size across
households, which provides an opportunity to examine the extent and impact of prenatal sex
selection across different parities and study the effects of prenatal sex selection on fertility
decisions. Second, we conduct a more comprehensive analysis on the impacts of prenatal sex
3

Other recent related studies are Portner (2010), which looks at the determinants of sex‐selective abortion
in India though not their effect on outcomes, and Bhardadwaj and Nelson (2010) which examine gender
differences in prenatal investment in countries with strong son preference, including India, although they
do not estimate the impact of sex‐selective abortion.

2

selection by examining both mortality and nutritional outcomes. Our analysis of nutritional status
provides important findings on intermediate outcomes that are prevalent in a high proportion of
Indian children and have vital life‐long consequences for human capital development and well‐
being (see, e.g. Currie, 2009 and Case and Paxson, 2010). Third, we go a step further and explore
different channels through which prenatal sex selection affects child outcomes such as selection,
family size, preferences, and breastfeeding duration. Finally, our empirical approach, which is
based on triple‐differences models, allows us to control for several confounding factors such as
regional variation in son preference and differential trends across regions, providing a powerful
way to reject alternative explanations for the observed results.
Our results show that an increase in the practice of prenatal sex selection leads to a reduction
in the prevalence of malnutrition (in particular, underweight and wasting) among girls. Our results
are robust to the inclusion of state‐year fixed effects suggesting that our findings are unlikely to
be driven by state specific time varying unobserved factors associated with changes in the
incidence of prenatal sex‐selection. Our results are also robust to several checks that assess the
likelihood of possible biases due to state‐specific differential trends by gender associated with
prenatal sex selection.
Additional results show no evidence that prenatal sex selection leads to a selection of girls
into households with more resources since we do not observe a differential improvement in
household socio‐economic characteristics of girls. However, we find that girls are more likely to
be born in families with weaker son preferences. We also find evidence of a larger reduction in
family size and an increase in breastfeeding duration for girls. On the other hand, prenatal sex
selection does not appear to be associated with a reduction in girls’ mortality or a regional decline
in reported preferences for sons.
The rest of the paper is organized as follows: In the next section, we review the literature and
describe the institutional background of unbalanced sex ratios and prenatal sex selection in India.
Section III discusses the conceptual framework for analyzing the effects of prenatal sex selection
on child outcomes. Section IV describes the data. Section V lays out the empirical strategy and
Section VI presents the results. Finally, Section VII concludes.

II.

BACKGROUND AND INSTITUTIONAL FRAMEWORK

Imbalanced sex ratios have been documented in India as early as the 19th century and throughout
the 20th century (see Visaria, 1971; Miller, 1981, 1984; Dyson and Moore, 1983; and Sen, 1992,
3

2003 who introduced the concept of “Missing Women”). Until the late 1970s, gender imbalances
were mostly manifested at older ages and not particularly at birth and were attributed to excess
female mortality due to maltreatment and neglect and in extreme cases to female infanticide
(see, for example, Dreze and Sen, 1997 and Das Gupta, 1987).4
Several studies have documented sharp increases in male to female ratios at birth since the
late 1980s, especially in northern and western states, which are regions historically known for
strong son preference and gender discrimination (see, for example, Das Gupta and Bhat, 1997;
Arnold et al., 2002; Bhat, 2002; Bhaskar and Gupta, 2007; and Retherford and Roy, 2003). These
studies have shown that the increase in sex ratios at birth is concurrent with the spread of
prenatal sex determination technologies, thus suggesting that sex selective abortion is likely to be
a major contributing factor.5 Bhalotra and Cochrane (2010) estimate that about 0.48 million girls
per year were selectively aborted in India during 1995‐2005, which represents 6.2 percent of all
potential female births. Estimates for northern and western regions are considerably higher. For
example, Kulkarni (2007) estimates that out of 168,997 expected female births in Punjab in 2001,
19 percent (31,648) went missing.
Sex selective abortion requires two steps: sex determination of the fetus and an abortion.
Abortion was legalized in India under the Medical Termination of Pregnancy Act (MTP) in 1972.
Despite its legalization, most abortions are still practiced in unofficial and non‐regulated facilities.6
Access to abortion services is not difficult in India, even in the remotest areas of the country
(Duggal, 2004). Its costs vary by region, type of facility, method, and gestation period ranging
from US$4.5 to US$16.5 (Ravindran, 2002; Sundar, 2003).
Sex determination during pregnancy was first made possible in the late 1970s by the use of
amniocentesis (Jefferey et al., 1984) and became more accessible with the introduction of
4

Other possible explanations attributed to the lower number of females include census underenumeration
of girls or different patterns of age misreporting by sex. While these factors might explain part of the gender
imbalances, Visaria (1971), Miller (1981, 1984), and other researchers show that the main cause for
imbalanced sex ratios until the late 1970s was sex differentials in mortality.
5
Changes in the enumeration of girls or misreporting of age are unlikely to explain this upward trend since
similar increases in sex ratios at birth are also observed among Indian populations living in the US, Canada,
and the UK where birth registration is nearly complete and accurate (see Abrevaya, 2009; Almond and
Edlund, 2008; Almond et al., 2009; and Dubuc and Coleman, 2007). We provide in the next section further
evidence which suggests that prenatal sex selection is the main cause for the upward trend in sex ratios at
birth.
6
It is difficult to obtain precise figures on abortion rates. Using indirect estimation techniques based on 1991
birth rates and population counts, Chhabra and Nuna (1993) assess that about 6.7 million abortions are
carried out every year, with the majority of them taking place in informal facilities. A similar estimate (6.4
million) is provided by Duggal and Ramachadran (2004).

4

ultrasound technologies in the 1980s. In subsequent years, a continued decline in desired fertility
coupled with a slower decline in the total number of desired sons increased the pressure to have
sons at lower parities, thus raising the demand for prenatal sex selection (Das Gupta and Bhat,
1997). At the same time, economic development and trade liberalization accelerated the supply
of prenatal sex determination technologies. Following the policy reforms of the 1990s,
multinationals such as General Electric started manufacturing ultrasound machines in India
improving access to prenatal sex determination and increasing the practice of prenatal sex
selection throughout the country (Geroge, 2006). The diffusion process took place from urban to
rural areas and from households of high socioeconomic status to those of low socioeconomic
status (Khanna, 1997).7
In an attempt to eliminate the practice of sex selective abortion and in response to increasing
public pressure from several NGOs and women’s organizations, the government of India passed
the Prenatal Diagnostic Techniques Regulations and Misuse Act (PNDT Act) in 1994, thus making
it illegal to use ultrasound or amniocentesis in order to determine the sex of a fetus. However,
this legislation proved to be ineffective and the practice of sex selective abortion continued to
spread (see e.g., George, 2002 and Kishwar, 1995). In subsequent years, several private
ultrasound clinics opened, mobile clinics and portable ultrasound machines made prenatal sex
selection accessible in small towns and rural areas (Krugman, 1998), and most of the sex selective
abortions moved to underground clinics (Everett, 1998).
The publication of sex ratio figures from the 2001 Census revealed a further increase in MFR
at ages 0‐6, which led to increased public pressure on the government to enforce and expand the
legal power of the PNDT Act.8 Recent reports indicate some improvement in the enforcement of
the Act. However, sex selective abortion is still being practiced extensively and enforcement of
the law appears to be difficult, if not impossible, to achieve (see e.g., Subramanian and Selvaraj,
2009; Portner, 2010).

7

Ultrasound costs vary widely across regions and between formal and informal facilities. Arnold et al. (2002)
provide an average estimate of US$10‐20.
8
In 2001, the Centre for Enquiry into Health (CEHAT) and the Mahila Sarvangeen Utkarsh Mandal (MASUM)
filed a public interest litigation against the Union of India and all the state governments for the non‐
implementation of the PNDT Act and for the inclusion of all emerging technologies that could be used for
sex selective abortion. As a result, the Indian Government decreed in 2001 that manufacturers of
ultrasound equipment could sell the machines only to registered clinics, which must maintain strict records
of their use. For a description of the PNDT Act and its enforcement, see Mallik (2003) and George (2002).

5

III.

CONCEPTUAL FRAMEWORK

There are various channels through which parental preference for sons can affect girls’ outcomes
in the absence of a technology that allows them to realize their preferences for sex composition.
First, parents can allocate more resources to children that provide them with higher utility (i.e.
boys) as long as boys and girls are substitutes in parents’ utility function (Rosenzweig and Schultz,
1982) or when capital markets are imperfect (Becker and Tomes, 1979 and 1986).9
Parents can also attempt to adjust excess fertility and undesired sex composition of children
through selective allocation of resources that results in the mortality of the less‐desired children
(Simmons et al., 1982). Alternatively, if parents continue to bear children until they achieve their
desired number of sons, girls will have, on average, a larger number of siblings than boys
(Yamaguchi, 1989). Therefore, even if parents treat boys and girls equally within a household, girls
can be expected to grow up in larger families with fewer resources per child (Jensen, 2003). In
addition, parental preference for boys is likely to affect birth spacing (Ben‐Porath and Welch,
1976) and duration of breastfeeding (Jayachandran and Kuziemko, 2009) since parents may wish
to have a child sooner following the birth of a daughter, especially if they are approaching the end
of the mother’s fecund years. Both factors are known to affect children’s health.
Increased access to prenatal sex selection may improve the outcomes of girls who are born
through various channels. First, girls may be more likely to be born into families that have a lower
disutility from girls and therefore may suffer less from discrimination in resource allocation within
the household. As suggested by Goodkind (1996), prenatal sex selection may substitute for
postnatal sex discrimination. Second, prenatal sex selection allows parents to achieve their
preferred number of boys and girls without the need of having additional children.10

9

Differences in parents’ utility from boys and girls might be rooted in taste differences or in differences in
the costs and benefits of raising boys versus girls (Ben‐Porath and Welch, 1976). Arnold et al. (1998) provide
a review of the literature and highlight three main channels that affect parental preferences for sons. The
first is economic utility due to differences in labor market productivity between boys and girls, security in
the case of illness, support in old age and inheritance rules. A second channel is related to social utility,
which provides families with more sons a higher status and additional sources of income in the form of
dowry payments. A third channel is religious utility, which is derived from the performance by sons of
important religious functions, especially upon the death of his parents. Girls, in contrast, are seen as an
economic drain on family resources because of the dowry system and the high cost of weddings.
10
On the other hand, the ability to choose a child’s gender increases the expected utility of children and
might therefore increase the likelihood of having additional children. Family size might also increase with
access to prenatal sex selection if the net cost of raising boys is lower than the cost of raising girls (Ben‐
Porath and Welch, 1976) or if boys yield higher utility than girls and there is substitution between utility
from children and ordinary consumption or between utility from boys and utility from girls. Therefore, the

6

Consequently, family size will not necessarily be larger for girls. Prenatal sex selection might also
improve girls’ outcomes if parents of girls internalize the potential increase in the value of women
as they become scarcer in marriage and labor markets.11 Although it can also be the case that as
girls become scarcer, parents are more likely to want girls but do not need to invest so much in
their quality. In addition, prenatal sex determination might have a negative impact on girls’
outcomes if parents reduce prenatal investment in female fetuses in response to prenatal sex
determination rather than having an abortion (see, for example, Bharadwaj and Nelson, 2010 and
Almond et al., 2010). Finally, girls’ outcomes might also change (improve or worsen) through
selection if parents who practice prenatal sex selection have different characteristics than parents
who do not. Taken together, these various channels imply that the sign of the overall impact of
prenatal sex determination on the outcomes of girls who are born cannot be determined a priori
and remains an empirical question.

IV.

DATA

The data for our empirical analysis are taken from two major data sources. We first use population
counts from the Indian census to examine long term changes in MFR. We then use microdata from
the National Family Health Survey (NFHS) to link between changes in MFR and prenatal sex
selection and study how changes in MFR affect girls’ outcomes.
The NFHS is a large‐scale, multi‐round survey conducted in a representative sample of women
and their families throughout India.12 Three rounds of the survey were conducted in 1992‐3, 1998‐
9 and 2005‐6. Each round covered approximately 90,000 households which contained more than
500,000 individuals and was designed to provide state‐level and national‐level estimates. The
survey includes detailed information on the demographic and socioeconomic background of the
household members, as well as additional modules designed to investigate health, fertility and
mortality. One major advantage of these data is that they record complete birth histories

impacts of prenatal sex selection on family size are ambiguous. Nevertheless, the gender gap in family size
is expected to narrow or even reverse.
11
This and the first channel imply that girls are becoming “more wanted”. Although the first channel works
through selection across households with different preferences, whereas this channel works through a
general equilibrium effect changing preferences over time for all households, even those who do not
practice prenatal sex selection.
12
More precisely, it is a representative sample of women of reproductive age (e.g. ever married women at
ages 13‐49 in the first round, ever married women at ages 15‐49 in the second and all women at ages 15‐
49 in the third round). The NFHS was designed along the lines of the Demographic and Health Survey (DHS)
that has been conducted in many developing countries since the 1980s.

7

(including deaths and children living outside the household) of a large number of women over a
long period, making it feasible to compute MFR at birth by state and cohort.13
We pooled the three survey rounds and selected only households with ever married mothers
aged between 15 and 49 in order to obtain a consistent sample across the three rounds. Table 1
reports summary statistics for boys and girls included in our main samples (columns 1, 2, 4, and
5) and differences by gender (columns 3 and 6). The sample reported in columns 1‐3 (nutrition
sample) includes the youngest two children aged less than three of ever‐married women sampled
in one of the three NFHS rounds who have valid anthropometric data.14 There are 76,500 children
(36,940 girls and 39,560 boys) who satisfy these criteria. The sample reported in columns 4‐6
(mortality sample) includes all children of ever‐married women born within the 10 years
preceding each survey round. This adds up to 357,809 children (172,472 girls and 185,337 boys).
Household characteristics reported in the table are used as control variables in the empirical
analysis.
Most children (about 75 percent) in both samples live in rural areas. About half of the children
have mothers with no formal education and about 30 percent have fathers with no formal
education. Mothers’ age at first birth is relatively low at 19, on average. At the time of the survey,
mothers in the nutrition sample were 25 years old, on average, and they had an average of 3
children. Mothers in the mortality sample are 29 years old on average at the time of the survey
and have an average of about 4 children.15
Columns 3 and 6 show that girls tend to be born into more disadvantaged families than boys.
Their families have lower wealth levels, lower parental education, and a lower degree of exposure
to mass media.16 Another fact worth noting is that girls appear to have more siblings than boys in

13

Note that MFR estimates reported in many studies are based on census data which record only children
who are alive and live in the household at the time of the census. Therefore, they are affected by differential
mortality and living arrangements.
14
The first round of the NFHS collected anthropometric data for the youngest two children in the household
who were under four years of age. The second round restricted the anthropometric data collection to the
youngest two children in the household under the age of three, while the third round extended the data
collection to all children in the household under five years of age. To be consistent across survey rounds,
we restrict the sample to the youngest two children under three years of age. In practice, our results are
insensitive to these restrictions.
15
Note that the average number of children in our samples does not represent completed fertility as most
women are still in their fertile years.
16
The index for media exposure is defined by the sum of indicators for exposure to TV, radio and
newspapers or magazines. Each indicator receives a value of one if the mother reported exposure of at least
once a week or almost every day. The wealth index is a constructed index provided in the NFHS data. The

8

the mortality sample (columns 4‐6), which is consistent with parental stopping rules in fertility
behavior and son preference. There is no similar difference in the nutrition sample (columns 1‐3),
probably due to the fact that a large proportion of children in the sample comes from households
with incomplete fertility. Nevertheless, we see in both samples that mothers of girls are more
likely to report that they want to have another child. Differences in this variable are larger in the
nutrition sample since mothers are more likely to have uncompleted fertility. We will return to
the differences in household characteristics and family size in section VI.2 when we discuss
possible mechanisms underlying our main results.

V.

EMPIRICAL STRATEGY

The research question we seek to address in this paper is: what is the causal effect of prenatal sex
selection on the outcomes of girls who are born? The ideal experiment we would like to study
would be based on random assignment of access to ultrasound or any technology that facilitates
prenatal sex selection. But this is not feasible. Moreover, we do not observe whether individual
households practice sex selective abortion. Our empirical strategy is therefore based on using
aggregate male‐female ratios at birth (MFR), which are a consequence of households behavior,
to proxy for the prevalence of prenatal sex selection and exploit the variation in MFR across states
and over time to estimate the impacts on girls’ outcomes while trying best to isolate other
confounding factors.
In subsection V.1 we show how male‐female ratios at birth (MFR) vary across time and states and
demonstrate that increases in MFR at birth are a consequence of prenatal sex selection. In
subsection V.2 we explain how we use this variation to study the impacts of prenatal sex selection
on girls’ outcomes.

V.1. Incidence of prenatal sex selection and male‐female ratios at birth (MFR)
V.1.a. Long run evidence of state‐level changes in MFR
As mentioned above, one potential limitation of this analysis is that we do not directly observe
the practice of sex selective abortion. However, we do observe its consequences, primarily the
abnormal sex ratio at birth. Figure 1 shows state variation in MFR at age 0 as reported in census

index is based on household assets and housing characteristics and denotes the wealth quintile of the
household relative to all households sampled in the same survey round.

9

records from 1961 to 2001.17 Until the 1980s, MFR at birth did not exceed the normal ranges of
1.03‐1.07 found in various large‐scale studies (for example, Visaria, 1971 and Jacobsen et al.,
1999). Increases in MFR at birth become evident after 1981 and, to a greater extent, after 1991,
which overlaps with the introduction of ultrasound technology in India.
Interestingly, there are large variations in MFR (both in their levels and in their growth rates)
even across those states that appear to have a strong preference for boys. To illustrate this point,
Appendix Table A1 reports indicators of son preference and selected characteristics by state
based on tabulations from the first round of the NFHS and per capita net state domestic product
in 1990.18 States are grouped by region.
The largest increases in sex ratios at birth are found in the northern and western states, which
are characterized by a strong degree of son preference. In Punjab, for example, while MFR was
within the normal range between 1961 and 1981, it increased dramatically between 1981 and
2001 from 1.06 to 1.29. Punjab also appears to have strong son preference as manifested by the
ideal sex ratio reported by mothers (1.48 or 1.47, depending on the measure) and the proportion
of mothers who desire a larger number of sons than daughters (0.48).
While strong son preference is found in states with upward trends in MFR, we also observe
that in several states in the northeast, central, and eastern regions with similar strong preferences
for sons there was either no increase in MFR or only a mild one. In Madhya Pradesh, for example,
mothers reported an ideal sex ratio of 1.55 but sex ratios at birth remained close to natural levels
(MFR of 1.06 in 2001). Southern states are usually characterized by a low degree of son preference
and stable sex ratios at birth.
This suggests that strong son preference cannot alone explain state variation in MFR since
there are several states with strong son preference that have not shown any significant increase
in MFR. Indeed, a factor that distinguishes between states with similar son preference but
different incidence in prenatal sex‐selection is economic development. Northern and western
states are generally more economically developed than states in the northeast, central and

17

MFR from census records number of children aged 0 to 11 months who are alive on census date. So, they
also reflect gender differentials in infant mortality rates and living arrangements. In the next paragraphs we
refer to MFR at age 0 as MFR at birth since, as we will show later, changes in mortality and living
arrangements by gender had only a negligible contribution to the overall trends in this variable.
18
For details about the construction of son preference indicators see Data Appendix.

10

eastern regions.19 On the other hand, development and income levels in many of the northern
and western states are comparable to those in southern states where sex ratios have remained
balanced.
To more systematically examine the relationship between changes in MFR, income and son
preferences, we estimated a set of state‐level regression models of MFR on state income per
capita and son preference. We use MFR data from the 1991 and 2001 census to capture long‐
term changes, and state income per capita from 1990 and son preferences from the first round of
the NFHS survey (1992) as baseline levels of income and son preference. First, we regressed the
state‐level MFR on a post dummy for year 2001, baseline state income per capita (in thousand
Rs.) and average state son preferences and their interactions with the post dummy. 20 In columns
1 and 2 of Table 2 we report estimates from models that include income or son preferences
separately. In column 3 we report estimates from a model that includes both variables. The results
show that that the MFR was higher in states that were richer or had stronger son preferences and
that MFR changes were also steeper in those states. Moreover, column 3 shows that associations
between changes in MFR and income per capita and son preferences are larger while controlling
for both factors at the same time. In column 4, we report estimates from a model that includes
also interactions between son preference and income. The results show that the states that
experience largest increase of MFR are those with both high income and strong son preference.
Overall, evidence above suggests that two factors which characterize states with an increasing
MFR are a strong preference for boys and a higher degree of development and modernization (in
combination). The fact that there are states with a strong preferences for boys (eastern region)
and states with high levels of development (southern region) that have not exhibited significant
increases in MFR provide us with a heterogeneous group of states that are comparable to states
with increasing MFR across different dimensions.

19

Appendix Table A1 also presents some economic and demographic indicators by state for the years 1991‐
2. As can be seen, northern and western states have a higher wealth index, income per capita, share of
households with electricity, and a higher degree of exposure to mass media (TV).
20
The son preference measure used here is the ratio of ideal number of boys to ideal number of girls
reported by mothers (averaged across households in the state). See data appendix for more information
about the definition of the variable.

11

V.1.b. Individual Level Evidence
Figure 1 and Table 2 show a high variation in MFR across states and over time and its
relationship with state‐level income and son preferences. But is the increase in MFR directly
related to the practice of prenatal sex selection? To examine this question, we use the individual
level data from the NFHS to test whether the propensity of giving birth to a boy is higher among
families who might feel a stronger pressure to have a son and whether this propensity increased
when prenatal sex determination became feasible.
In Table 3, we report the likelihood of a male birth at parity N (two or three) as a function of
the sex composition of the older siblings who were alive at the time of conception using a linear
probability model. We examine two samples: children born between 1975 and 1989 and children
born from 1990 onwards. This split is meant to proxy for the availability of ultrasound
technology.21 As seen in column 1, the probability of a male birth during the pre‐ultrasound period
did not vary significantly across households according to the sex composition of their previous
children. In contrast, column 3 shows that during the post‐ultrasound period this probability was
significantly higher for households that had only girls. Estimates of the differential probabilities
are virtually unchanged after adding controls for household characteristics (columns 2 and 4).
Moreover, at parity 3, the differential probabilities (relative to the reference group two sons) are
larger and statistically significant for families with no sons than for families with 1 son after the
1990s (columns 3 and 4 in row 5 show that we can reject equality of the coefficients on no sons
and 1 son at the conventional level). These results suggest that families have strong preferences
for at least one boy, and not necessarily an aversion of girls.
In Figure 2 we plot the differential probability of a male birth by year comparing between
households with no boys among previous children and the rest, by parity. In each case, we also
stratify the sample by household income and son preferences, looking separately at those who
live in states with stronger son preferences (state son preference in first round of NFHS above
median) and are relatively wealthy (household wealth index in the fourth and fifth quintile) and
at other households. We smooth estimates using a 7‐year moving average.

21

The sample split at 1990 is used for illustration purposes. Similar but smaller estimates for the post‐
ultrasound period are obtained when we use 1985 as a cutoff year and higher estimates are obtained if we
split the sample at 1995.

12

Panel A of the figure shows that until the mid‐1980s there were no differences in the
probability of a male birth by sex composition of previous children at parity two. Moreover, there
were no differences between richer households from strong son preference states and the rest.
There is an increase over time, especially after the 1990s, in the differential probability of a male
birth at parity 2 for households with no boys versus the rest. This increase is steeper among
wealthier households who live in states with stronger son preferences.
Panel B shows that the differential probability of a male birth at parity three did not change
much for households who live in states with weaker son preferences or have lower income. In
contrast, we see a sharp increase in the differential probability of a male birth among those who
live in states with stronger son preferences and are relatively wealthier that starts in the early
1990s.22
Evidence presented in Table 3 and Figure 2 shows that the likelihood of a male birth increased
significantly in the 1990s among households who presumably had a stronger desire for a son and
had more resources. This suggests that increases in MFR at birth are likely to be a result of access
to prenatal sex selection. In order to further examine the link between son preference and sex
ratios at birth, additional information on ultrasound use and abortions is needed. Information on
the incidence of abortions is difficult to obtain since a large share of abortions take place in
unofficial and non‐regulated facilities. In addition, abortion is usually misreported, especially if it
is carried out for sex selection purposes. Nevertheless we can use information available in the
survey about ultrasound use.23
Columns 1 and 2 of Table 4 report the differential likelihood that a mother performed an
ultrasound test during a pregnancy of parity N (two or three) as a function of the sex composition
of her N‐1 previous children who were alive at time of conception. As seen in the table, mothers
with no living sons are more likely to perform an ultrasound test in pregnancies at parities two
and three. We further examine the link between ultrasound use and prenatal sex selection by

22

In Appendix Figure A.1 we plot the F‐statistics for Chow tests for a structural break for each of the years
in each series by parity. The results of these tests show no evidence for a break in the 1980s for any of the
subsamples. For high income households in son preference states (solid lines), we see evidence for a
structural break in 1990 (parity 2) and in 1995 (parity 3). In contrast, there is no evidence for such break for
all other households (dotted lines).
23
The second and third round of the NFHS survey asked mothers whether they performed an ultrasound
test for each of their births during the three (in round 2) or five (in round 3) years prior to the survey. The
second round asks about ultrasound use among women who received at least one prenatal checkup during
pregnancy while the third round asks about ultrasound use for each pregnancy. To avoid selection issues
we use information on ultrasound use from the third round only and limit the analysis to that sample.

13

looking at the association between a male birth and the sex composition of older children among
mothers who reported conducting an ultrasound test during pregnancy versus mothers who
reported no ultrasound use. It is worth noting that most women who do an ultrasound do not
practice sex selective abortion. In addition, some women who practice sex selective abortion,
especially those with more years of schooling, do not report doing an ultrasound as they are aware
of the banning of ultrasound for sex selection. Nevertheless, differences in the differential
probability of a male birth between the “ultrasound” and “no ultrasound” sample add supporting
evidence for the practice of sex selective abortion.
Estimates in columns 3 and 4 show that for the sample of women who report ultrasound use,
the probability that a birth is a boy is substantially higher (8 to 10 percentage points) for those
who previously had no sons than the rest. In contrast, there is no clear relationship between sex
composition of previous children and probability of a male birth for households who did not
perform an ultrasound test. The differential probability of a male birth (reported in columns 5 and
6) is small, not statistically significant and has inconsistent signs across parity 2 and 3. Estimates
of the “ultrasound” and “no ultrasound” samples are significantly different at the 10 percent level.
In summary, the evidence presented in this subsection suggests that increases in MFR at birth
are likely to be induced by the increasing practice of prenatal sex selection rather than by
biological or environmental factors. We therefore use MFR at the state level as a proxy for the
practice of prenatal sex selection.

V.2. Effect of Prenatal Sex Selection
We exploit variation in the timing and extent of prenatal sex‐selection across states (proxied
by changes in MFR) and examine whether the observed variation is systematically associated with
changes in girls’ outcomes relative to boys. To take advantage of the year‐to‐year variation in sex
ratios at birth we use MFR constructed from the NFHS based on the complete birth histories
reported by mothers. Another advantage of using the NFHS data as opposed to the census data
is that we observed all children ever born (and not only those who survive until age one).24 On the
other hand, in the survey data, the state‐level MFR for a single year can be quite noisy (especially
for smaller states). 25 We therefore use a smoothed version of MFR which is computed as a 7‐year
24

In practice, as we show later, the mortality adjustment matters little for our results.
While the result of such measurement error is formally ambiguous in a multivariate regression (see e.g.
Wooldridge 2002), it tends to bias regression coefficients in the direction of 0, which would work against us
finding significant effects of the MFR.

25

14

moving average of the ratio of the number of male births to female births by year and state based
on the pooled data of the three rounds of the NFHS survey. We also use an alternative measure
of MFR based sex ratios from the census where we interpolate sex ratios for inter‐census years
using a fourth order polynomial function that fits the NFHS data. Finally, we use a third measure
where we re‐estimate the function adjusting the census MFR for infant mortality.26 The data
appendix describes how we construct these three measures of MFR in detail.
Figure 3 plots the smoothed MFR from the NFHS, the census MFR with the interpolated
function for inter‐census years and the function that adjusts for infant mortality. The population
size at age 0 in 1991 is reported in parenthesis. Overall, we see a large variation in the trends of
sex ratios across states. For example, while MFR increased steeply in Punjab, Haryana and Delhi,
there was almost no change in Kerala or Nagaland. In other states, like Tamil Nadu, MFR was
relatively stable until 2000 and started to increase only then. The figure also shows that trends in
MFR series computed from NFHS or Census are similar, though there are some disparities
between both series, especially in smaller states, probably due to sampling error in the NFHS. In
addition, adjustments for infant mortality do not seem to have an important effect. In any case,
as we show later, our results are very similar irrespective of the MFR series we use.
Using these alternative MFR measures, we estimate the following equation:

(1) yist   s 0   s1 femalei   t 0   t1 femalei  xi '    0 MFR st 

1

 MFR st * femalei    ist

where yist is the outcome of child i in state s born in year t,  s 0 and  s1 are vectors of gender‐
specific state fixed effects,  t 0 and  t1 are vectors of gender‐specific year‐of‐birth fixed effects
and xi is a vector of individual characteristics that include indicators for twin birth, residence in
an urban area, religion, mother’s and father’s level of education, mother’s age (grouped), wealth
quintiles, mass media exposure and mother’s age at first birth. MFRst is the Male‐Female Ratio
at birth for the cohort born in year t in state s.  ist is the error term, which is composed of a state‐

26

We also experimented with lower and higher order polynomials and the results (not reported here but
available upon request) are largely similar.

15

specific random element that allows for any type of correlation within observations of the same
state across time and an individual random element.27
The parameter of interest is

1 , which captures the effect of increases in prenatal sex

selection (proxied by increases in MFR) on changes in girls’ outcomes. This approach is essentially
a continuous version of a triple‐difference estimation strategy. This strategy has the advantage
that it allows us to control for state‐level fixed factors that differentially affect boys and girls (for
example, the degree of discrimination against girls in a state). We can also control for state‐time‐
varying factors that affect boys and girls similarly within each state and cohort and which might
be correlated with changes in MFR over time (for example, improvement in access to health care
facilities). In addition, we control for differential trends in boys’ and girls’ outcomes at the national
level in a very flexible way by including gender‐specific year of birth fixed effects.
Identification relies on the assumption that changes in MFR within a state over time are
unrelated to other unobserved factors that could differentially affect male and female outcomes.
In this regard, it is important to note that we always include in our model the main effect of MFR,
which gives an indicator of the effect of unobserved time varying factors at the state level
correlated with MFR, on boys’ outcomes. As we show later, estimates of MFR are never significant
and have inconsistent signs across outcomes, reducing one of the major concerns regarding
omitted variable bias. Our results are also robust to replacing the main MFR effect with a set of
full interactions between state‐ and year‐of‐birth fixed effects, suggesting that state‐time varying
factors are unlikely to confound our findings. We further assess the plausibility of the main
identifying assumption in Appendix B, where we also consider more general specifications that
allow for gender‐specific coefficients in all covariates and control for time‐varying state
characteristics interacted with gender, thus also allowing them to have a differential effect by
gender.
A possible concern regarding identification could arise if increases in MFR within a state over
time were related to increasing discriminatory preferences against girls rather than increasing
access to prenatal sex selection technology. While we cannot rule out this alternative explanation,
we note that in this case, our estimates would be biased against finding any improvement in girls’
outcomes and would therefore provide a lower bound of the effects of prenatal sex selection. On

27

In all the specifications, we use the national sampling weights provided in the surveys and cluster standard
errors by state. As we show in the next section our results are not sensitive to the use of this specific
weighting scheme.

16

the other hand, note that our estimates capture the overall effect of an increase in prenatal sex
selection on girls’ outcomes. This includes also any possible general equilibrium effect resulting
from parents internalizing future changes in women’s value due to an increase in the scarcity of
girls. We will turn back to this issue in Section VII below when we examine the possible channels
linking between prenatal sex selection and girls’ outcomes.
We do not include household fixed effects in the estimating equations. This is due to several
reasons: First, identification in a household fixed‐effect model would rely on a comparison
between siblings of the opposite sex. However, households with children of opposite sex cannot
be viewed as a randomly selected sample since the sex composition of children is affected by
parental stopping rules in fertility behavior and prenatal sex selection. In addition, given that our
main sample includes children born within three years prior to the survey date, a household fixed‐
effects model would generate estimates for a selected sample of households (i.e. those which had
two births within the last three years). Note also that girls’ outcomes might change due to
selection into different households or a differential reduction in family size. We are therefore
interested in the well‐being of girls across all families and not necessarily within the same
household.

VI.

EMPIRICAL RESULTS

1.

MAIN RESULTS ON NUTRITIONAL OUTCOMES

Girls’ well‐being can be analyzed across two dimensions: the allocation of household inputs such
as material resources, time, health care and feeding practices or children’s outcomes such as
nutrition, morbidity and mortality. We focus our analysis on children’s nutrition and mortality
since they embed information on various types of parental input. These outcomes are measured
more easily than most others and are therefore less likely to suffer from measurement error or
recall bias. We do not examine morbidity since it is more likely to suffer from bias due to gender
differences in diagnosis and recall.28
We measure children’s nutritional status by anthropometric indicators based on height,
weight and age. In particular, we consider three indicators of malnutrition: stunting, underweight
and wasting. All three are defined based on z‐scores, which are computed by subtracting the
median and dividing by the standard deviation of a reference population of the same age and
28

For example, Timaeous et al. (1998) report sex bias in reporting and recalling episodes of disease and sex
differentials in childhood risks of illness and illnesses severity perception.

17

gender. Specifically, a child is considered stunted if his/her height‐for‐age is at least 2 standard
deviations below the median of the reference population (or the associated z‐score is smaller than
‐2). An underweight child has a weight‐for‐age at least 2 standard deviations below the median,
and a wasted child has a weight‐for‐height at least 2 standard deviations below the median.
The three indicators capture malnutrition from different perspectives. Stunting reflects long‐
term malnutrition or cumulative nutrition from conception and is also affected by recurrent or
chronic illnesses. Wasting measures acute malnutrition and represents the failure to receive
adequate nutrition in the period immediately preceding the survey and may be the result of
inadequate food intake or a recent episode of illness leading to weight loss. An important feature
of the wasting indicator is that it does not depend on the accuracy of age reporting. On the other
hand, it is more sensitive to seasonal shocks. Underweight is a composite index of chronic or acute
malnutrition. Note that z‐scores are normalized by gender and age so that they take into account
that boys and girls may follow different growth trajectories.29 Our analysis uses z‐scores based
on the US National Center for Health Statistics (NCHS) standard, which was the most commonly
used measure until 2006.30 About 18 percent of the children aged 0 to 35 months included in our
sample have missing values in at least one of the anthropometric indicators. Nevertheless, we do
not find any significant gender differences in the likelihood of having a missing value in these
indicators. Moreover, we do not find any associations between state variation in MFR and the
likelihood of missing anthropometric data for girls or boys (see Appendix Table A2). 31
A methodological challenge in the analysis of nutritional outcomes on the basis of
anthropometric data is the availability of these data only for surviving children. Therefore, if
prenatal sex selection has any impact on mortality, our analysis of nutrition will be based on a
selective sample. For example, if increases in MFR are associated with a reduction in female child
mortality, they might lead to an increase in the proportion of girls who are close to a survival
29

In an alternative specification, we included age‐in‐months and its interaction with female (instead of
child’s year‐of‐birth) in order to control for any differences in child outcomes that could result from
differences in the age distribution of boys and girls. The results are virtually identical to those presented
here.
30
A new international reference population was published by the World Health Organization in 2006. As
we show in Table A2, our results are not sensitive to the specific reference chart used to define z‐scores.
31
Height was not measured in the first round of the NFHS in Andhra Pradesh, Himachal Pradesh, Madhya
Pradesh, Tamil Nadu and West Bengal because height measuring boards were not available at that time
(IIPS, 1995). Nevertheless, our results for weight are highly consistent with the results for height and our
main results are unchanged when we limit the sample to states that have anthropometric data in the three
survey rounds. We therefore believe that the lack of height data in round 1 for some states is unlikely to
bias the main results.

18

threshold, thus attenuating the estimated effect of MFR on nutritional status. As shown below,
the effects on mortality are negligible or not significant, which mitigates the concern about
selectivity issues in the nutritional status results.
Table 5 reports estimates for MFR and female*MFR from a linear probability model for the
likelihood of being underweight, wasted, or stunted.32 Estimates from a basic model that includes
no covariates, except for a female dummy, and gender‐specific state and year‐of‐birth fixed
effects are reported in columns 3 and 4. Columns 5 and 6 report estimates for the full model
specified in equation (1) which controls also for the household characteristics reported in Table
1. The key parameter of interest is the coefficient of the interaction term female*MFR (column
6), which is negative for all three outcomes and, except for stunting, is statistically significant.
Note that the coefficients for the main effect of MFR (column 5) have inconsistent signs across
outcomes and are never significant, suggesting that changes in MFR are unrelated to changes in
the nutritional outcomes of boys. This last finding is important since it suggests that increases in
MFR at the state level are unlikely to be associated with other changes in unobservables that
affect children’s nutritional status. It is also worth noting that our estimates for the interaction
between MFR and female are largely unchanged after the addition of household covariates (see
columns 4 and 6) minimizing some concerns of omitted variable bias.33
Note that the key female*MFR coefficient reflects the differential change in girls’ outcomes
that results from a one‐unit increase in MFR. For example, the estimate for underweight from
column 6 indicates that a one standard deviation increase in MFR (0.07 points) is associated with
a 4 percentage point reduction in the proportion of girls who are underweight, which is a
reduction of 8 percent relative to the mean outcome for girls. Taking Punjab as an example, this
would mean that the increase in MFR of 0.2 observed between the first and third round of the
NFHS is associated with a 10 percentage point reduction in the proportion of girls who are
underweight.
Columns 7 and 8 of Table 5 report estimates for the models that use MFR constructed from
census data and inter‐census interpolation and columns 9 and 10 report estimates from models
that use the same data but adjust census numbers for infant mortality. Estimates are highly similar

32

Marginal effects from logit models provide similar results.
However, if there is selection in which families have girls (as a result of prenatal sex selection), these
covariates would be endogenous. This is why we report results with and without covariates. Nevertheless,
as we show later, we find no evidence on selection based on these observable characteristics.

33

19

to the results based on NFHS data and suggest that girls’ chances to be underweight or wasted
are lower in states where prenatal sex selection increased. In all models estimates of the
interaction term between MFR and female for stunted are negative and similar in magnitude to
the estimates for wasted but they are less precise.34 However, in contrast to underweight and
wasted, the main MFR effect is positive (although not significant). Therefore, our conclusions for
the effects of prenatal sex selection on children nutritional status apply to a larger extent to
weight and weight for height.35
We also conducted additional analyses to test the sensitivity of our results by using alternative
definitions for nutritional outcomes, alternative model specifications, and alternative weighting
schemes. The analysis is described in Appendix B and the results are reported in Appendix Tables
A3 and A4. Overall, we find our main results are robust under those different choices.
In addition, we estimated Difference‐in‐Differences models for boys and girls separately. The
results (reported in Appendix Table A5) are consistent with our triple difference in difference
models in that both sets of results show relative improvement of girls’ outcomes compared to
boys. Moreover, the DID results show that the relative improvements of girls outcomes are
actually absolute improvements, rather than merely reflecting less deterioration of girls’
outcomes compared to boys.
So far our findings are consistent with the hypothesis that girls outcomes improved due to the
increased use of prenatal sex selection. While an accurate measure of the latter is hard to come
by, we explore some limited information on ultrasound use in the data to test whether girls in
families that are more likely to perform ultrasound test experienced a greater improvement in
health. The third round of the NFHS asked women whether they have undergone an ultrasound
test in each of the pregnancies that took place within the five years preceding the survey. We use
this information to predict women’s likelihood of doing an ultrasound test for the whole sample

34

One possible explanation is that there is more measurement error in height than in weight for young
children given that height is measured lying on a board for children under the age of 2. For example, in our
sample, the standard deviation of the z‐score for height is 1.68, compared to 1.33 and 1.22 of the z‐scores
for weight and for weight‐for‐height, respectively. The conventional ranges are 1.10 to 1.30 for height, 1.00
to 1.20 for weight and 0.85 to 1.10 for weight for height (see WHO 1997).
35
Note that the three nutritional outcomes are not independent of each other. For example, weight‐for‐
height is mechanically related to weight‐for‐age and height‐for‐age. So in our case, if we see a large effect
on “underweight” (based on weight‐for‐age) and a moderate effect on “wasting” (based on weight‐for‐
height), then it should not be surprising to also see a somewhat limited effect on “stunting” (based on
height‐for‐age ).

20

based on women’s predetermined characteristics.36 Next, we examine changes in the frequency
of female births and health outcomes within states and over time by mothers’ predicted likelihood
of ultrasound use. The idea is to examine how outcomes change within households who have a
similar likelihood of doing an ultrasound test as the technology spreads over time.
We performed two exercises and report the results in Table 6.37 Column 1 of Panel A
reports estimates of a model where we regress an indicator for a female birth on the probability
of ultrasound use, year of birth and interactions between both variables, controlling for state fixed
effects. We see that the interaction term between the probability of ultrasound and year of birth
is negative, suggesting that the proportion of female births declines more over time among
households that have a higher probability of performing an ultrasound test.
In columns 2 through 4 we regress the nutritional outcomes on the probability of
ultrasound use, year of birth and gender and the interactions between these variables including
also a gender‐specific state fixed effect. The triple interaction term between female, probability
of ultrasound and year of birth is negative for all three outcomes and statistically significant for
underweight and stunted, suggesting that there is a larger reduction in the proportion of girls who
are malnourished over time among households who are more likely to perform an ultrasound
test.
Panel B of the table report results based on a subsample of children whose mothers are
unlikely to have undergone an ultrasound test (those with a predicted probability below the 25
percentile – 0.078). For this subsample we checked whether the probability of a female birth
changes over time by regressing a female dummy on year of birth and state fixed effects. As seen
in column 1 of Panel B, there is no change over time in the probability of female births in this
subsample. In columns 2 through 4 of Panel B we checked whether there was a change in the
nutritional outcomes for these girls by regressing the nutritional outcomes on year of birth and
its interaction with female while controlling for gender specific state fixed effects. The results
show no reduction in the proportion of girls who are wasted. Results for underweight and stunted
show a declined over time for both boys and girls, but the improvement seen among girls is less
36

We predict this probability using the following covariates: residence in an urban area, religion, mother’s
and father’s level of education, mother’s age, wealth quintiles, index of mass media exposure, mother’s age
at first birth and state fixed effects. We then use the estimated model to predict the probability of
ultrasound test for all women in the three survey rounds.
37
We computed bootstrapped standard errors to account for the fact that the predicted probability of
ultrasound use is estimated. The adjustment is done using the procedure described by Petrin and Train
(2003).

21

pronounced as evidenced by the positive estimate of the interaction term between year of birth
and female. Overall we see no evidence for a differential improvement in girls’ outcomes in
households who are unlikely to perform an ultrasound test.
Taken together, results presented in this table are consistent with our previous findings
and suggest that girls’ outcomes improved as a consequence of the diffusion of prenatal sex
selection. In addition, it seems that an increase in girls’ scarcity does not affect health investments
in girls for households who are unlikely to practice sex selective abortion; a point that we discuss
further below.

2.

MECHANISMS

Broadly speaking, there are two ways in which access to prenatal sex selection can lead to changes
in outcomes: selection and treatment. The selection effect arises from girls being born into
families with different characteristics and son preferences that result in changes in the average
outcomes of the girl population. The treatment effect, on the other hand, comes from changes in
the behavior of a given household, in response to access to prenatal sex selection (or changes in
MFR).
Within this framework, we examine four specific channels through which access to
ultrasound technology can affect health outcomes ‐‐ family characteristics, family size,
wantedness of a child’s gender, and breastfeeding practices. The first channel is a selection effect
while the latter three can be either selection, treatment, or both.
The improvement in girls’ nutritional status could be the result of selection, such that girls
are born into families with better endowments. It could also be explained by changes in family
size. Girls could be increasingly born into smaller families because families with weaker son
preferences might also desire smaller size (selection) or because families became less reliant on
stopping rules in fertility behavior (treatment). Similarly, girls could become more wanted due to
selection into families with weaker son preferences, or due to changes in preferences for all
parents as they internalize the change in the future value of women as they become a scarce
commodity in marriage and labor markets, even in families that do not practice prenatal sex
selection. Finally, girls could be breastfed for longer due to selection into different households or
due to a change in mothers’ behavior as a response to changes in desired number of children or
internalization of changes in MFR.

22

2.a. Are girls born into better‐endowed families?
In order to assess the selection effect, we examine whether family characteristics of girls (such as
parental education, mother’s age at first birth and household wealth status) improved relative to
boys in states where MFR has increased. This is done by estimating a triple‐differences equation
in which we regress each of the family characteristics on MFR and female*MFR in a model that
controls for state fixed effects, year of birth fixed effects and their interactions with a female
indicator.
Table 7 reports estimates for the main MFR effect and the interaction between MFR and
female. The table also reports gender differences in household characteristics (column 3).
Estimates of the main MFR effect reported in column 4 (rows 1‐6) suggest that states with
increasing MFR experienced improvement in some family characteristics, in particular, an
increase in the level of parental education and mother’s age at first birth and a decline in the
likelihood of living in a rural area. This is consistent with the fact that the practice of prenatal sex
selection is related to economic development. On the other hand, there is no association between
MFR and maternal age or the household wealth index.
Estimates for the interaction between MFR and female reported in column 5 (rows 1‐6) are
all small, not significant and have inconsistent signs across the different variables. These results
suggest that there is no differential improvement in household characteristics among families
with girls relative to families with boys and therefore the improvement in girls’ nutritional status
is unlikely to be explained by the fact that girls are born into “better” families.
The positive association between MFR and some household characteristics raises some
concerns regarding our main results for girls’ nutritional outcomes if an improvement in
household characteristics has a differential effect on the nutritional status of boys and girls.
Nevertheless, as reported in Table A4 and discussed in the previous section, our main results were
virtually identical to those obtained from an enriched model in which we allow for a differential
effect of household characteristics by gender. This suggests that the improvement in parental
education or other household characteristics did not have a differential effect on girls’ nutritional
status.

2.b. Are girls born into smaller families?
As noted above, a direct consequence of parental stopping rules in fertility behavior and son
preference is that girls are more likely to be born in larger families. With access to prenatal sex
23

selection, parents of girls would not necessarily need to have more children in order to attain the
desired sex composition. Family size could also decrease through selection, if parents who do not
practice sex selective abortion have, in general, fewer children. As a result of a decrease in family
size, girls might fare better due to an increase in family resources per child.
We examine whether family size has differentially changed for girls relative to boys in regions
with increasing MFR. To this end, we regress family size on MFR and female*MFR. The model also
controls for state fixed effects, year of birth effects and their interactions with a female indicator.
Row 7 of Table 7 reports the coefficients of MFR and MFR interacted with female. The
estimate reported in column 4 shows a negative association between MFR and family size which
may be due to various factors. One possibility is reverse causality. That is, a desire to have fewer
children may increase the demand for sons at lower parities, thus increasing the incidence of
prenatal sex selection. The association between MFR and family size could also be due to
unobserved factors that affect both MFR and fertility. For example, economic development is
generally associated with a reduction in fertility and an increase in access to ultrasound
technology. A third explanation is a causal link from access to prenatal sex selection to family size.
Parents with access to prenatal sex selection are more likely to attain their preferred sex
composition of children without the need of having additional children. Note, however, that
reverse causality and omitted factors are expected to produce a similar association between MFR
and family size for boys and girls while the last one is expected to have a differential effect by
gender with larger reductions for girls. As seen in column 5, the estimate for the interaction
between female and MFR is indeed negative (‐0.923) and marginally significant (s.e.=0.534).
Moreover, estimates reported in the last row of the table show a significant larger reduction in
the average number of older siblings for girls relative to boys (estimate=‐0.980; s.e.=0.557) which
is explained by the reduction in the probability that girls are born at higher parities. These results
confirm two expected consequences of prenatal sex selection: a larger reduction in family size for
girls relative to boys and a decline in the proportion of girls born at higher parities. Both factors
are likely to increase the amount of household resources available for girls.
To further assess the impacts of prenatal sex selection on family size, we also examined
changes in number of children over time by households’ probability of ultrasound test using the
same two‐stage procedure applied for the nutritional outcomes. Results reported in Appendix
Table A6 are in line with our findings using MFR.

24

2.c. Are girls more likely to be wanted?
An additional expected consequence of prenatal sex selection is that girls would be more likely to
be born in families that want them. Girls might also be more likely to be wanted if their parents
internalize the potential change in the value of women as they become a scarce commodity even
in families that do not practice prenatal sex selection. A unique feature of our study is that we are
able to assess these channels by examining the association between MFR and parental
preferences for sons.
To investigate whether girls are differentially more likely to be born into families with
weaker son preference in places with increasing MFR, we regress the various measures of son
preference on MFR and MFR*female along with other controls. Note that we cannot establish a
causal relationship between MFR and son preference since the link between the two can go in
both directions. In addition, there may be other unobserved factors that affect both
simultaneously. Nevertheless, it is still interesting to examine the association between MFR and
son preference, in particular, the differential association by child gender as any bias related to
reverse causality or omitted variables is expected to affect boys and girls similarly.
We use three indices of son preference: two alternative measures based on the ratio of
the ideal number of sons to the ideal number of daughters reported by mothers and an indicator
for whether the mother reports a desire to have a larger number of sons than daughters. We also
define a birth‐specific measure capturing the “wantedness” of the child gender. Specifically, we
define an indicator for a birth being of wanted gender if the mother’s desired number of children
of the child’s gender is greater than the actual number of older siblings of his/her gender who are
alive at the time of the child’s conception.
We report in Table 8 estimates for MFR and its interaction with female for the sample of
children aged 0 to 35 months.38 Results show that for the three measures of son preference (rows
1‐3), the coefficient of MFR is positive and the interaction term between MFR and female is
negative. This suggests that in places with increasing MFR, boys are more likely to be born into
families with stronger son preferences while girls are more likely born into families with weaker
son preferences (relative to boys).39 The last row of the table reports the coefficient for wanted

38

The sample size is larger than our nutrition sample since we do not restrict it to children who have valid
anthropometric data.
39
To address the issue of reverse causality, we also analyze a separate sample of younger women aged 15
to 20 and examine the relationship between their reported son preference and the MFR of children born
5‐10 years prior to the survey date. Presumably reverse causality would be less relevant in these cases. We

25

gender. Both the main MFR effect and its interaction with female are positive (although they are
not statistically significant), and they are consistent with other estimates reported in this table.
Taken together, our analysis on wantedness, suggests that that children born in places
with an increasing use of SSA are more likely to satisfy parental preferences in terms of gender
composition.

2.d. Are girls receiving better parental care?
An improvement in girls’ nutritional status may be a direct consequence of better care and
treatment in the family since girls are more likely to be wanted when prenatal sex selection is
available. Parental treatment is difficult to assess since we do not directly observe parents
behavior. Nevertheless, we can examine mothers’ reports on breastfeeding duration as one
possible indicator for child care and treatment.40
Medical and public health research has found that breastfeeding has large benefits for
children’s health, especially in environments with poor sanitary conditions (see, for example, The
World Health Organization, 2000). The WHO recommends that breastfeeding in developing
countries should continue, in combination with supplementary foods, up to the child’s second
birthday or beyond (WHO, 1991).
Although breastfeeding is nearly universal in India, girls tend to be breastfed for a shorter
period than boys. Parents with son preference may breastfeed their sons for a longer period than
their daughters either due to direct discriminatory treatment or passively through stopping rules
in fertility behavior as mothers stop breastfeeding their daughters sooner in order to conceive
again quickly (Jayachandran and Kuziemko, 2010). With access to prenatal sex selection girls’
breastfeeding duration might increase either due to lower parental discrimination in child care

look at two different samples: ever married women aged 15‐20 and ever married women aged 15‐20 who
have no children (we cannot look at single women because son preference was asked only to married
women). The results reported in Appendix Table A7 show that son preference changed little in states with
increasing sex ratios.
40
Another potential measure of parental input is vaccination. However, as discussed in Barcellos et al.
(2010), evidence from the NFHS data on gender differential in vaccination rates has been mixed. The results
vary dramatically depending on whether vaccination is measured based on information recorded on
vaccination cards or based on mothers’ reports. In this context, selection bias could be a big concern since
very few mothers (only about 30%) had a vaccination card and they are also more likely to have a card for
boys than for girls. An additional problem with vaccination data from the NFHS is an apparent decline in the
record of immunization information in the third round of the NFHS in some states due to logistic and
methodological issues, especially among children with no vaccination card (see Chandran et al. 2011). For
these reasons, we do not focus on this outcome.

26

and treatment or due to a decline in parental pressure to conceive a son after the birth of a
daughter.
We examine the effect of prenatal sex selection on girls’ breastfeeding duration by estimating
equation (1) with the dependent variables being indicators of whether the child was breastfed for
at least 12, 18 or 24 months. In order to take into account the possibility of right‐censoring in
duration, the indicators are defined conditional on children being at least 12, 18 or 24 months
old, respectively, at the time of the survey.
We estimate the model using our main sample (the youngest two children born within 3 years
prior to each survey round). Results are reported in Table 9. The differences in outcome means
reported in column 3 show a disadvantage for girls in breastfeeding duration that widens with
age. For example, girls are one percentage point less likely than boys to be breastfed for at least
12 months, about 4 percentage points less likely to be breastfed for at least 18 months and almost
6 percentage points less likely to be breastfed for at least 24 months.
Estimates for the interaction between MFR and female suggest some increase in duration of
girls’ breastfeeding at long durations (24 months or longer) in states with higher increases in MFR.
These results suggest that breastfeeding practices may be another channel through which
prenatal sex selection improves girls’ nutritional outcomes. The increase in breastfeeding
duration is also consistent with our results that show an increase in the proportion of girls who
are wanted and a larger reduction of family size for girls. As discussed earlier, these two factors
are likely to explain gender difference in breastfeeding duration.

3.

MORTALITY

Gender preferences and discrimination may also affect child mortality. Indeed, a large literature
has documented excess female mortality which becomes evident after the age of one month. In
this section, we examine the effect of prenatal sex selection on children survival probabilities. We
focus on the following set of binary outcomes: surviving beyond the age of 1 month, 1 year and
5 years.41 For each outcome, we estimate a linear probability model with a specification similar
to that of equation (1).42 We consider two samples: the first includes the last two children born
within 3 years prior to each survey and parallels the sample selection criteria used for the

41

To deal with censoring, the indicators are only defined for children who would be “old enough” (i.e., one
month, one year and five years old, respectively) at the time of the survey.
42
Results from logit models (not reported here) provide similar findings.

27

nutritional outcomes and the second includes all children born within 10 years prior to each
survey.
The results are reported in Table 10. Overall, we do not find significant changes in female
survival probabilities in states with a high incidence of prenatal sex selection. The coefficients on
the key interaction term MFR*female (reported in columns 6 and 11) are small in all outcomes
and samples and none of them is statistically significant.
These results are somewhat at odds with our previous findings on nutritional outcomes, family
size and breastfeeding duration. Particularly puzzling is why we find a differential improvement
in female nutritional status but do not see any significant increase in female survival probabilities.
One possible explanation is that families that substitute between prenatal and postnatal female
discrimination are not the same families characterized by excess female child mortality. A second
possible explanation is that differential recall (by gender) of deaths and measurement error may
be biasing the estimated MFR effects towards zero. 43 By comparison, the analysis of nutrition
outcomes of living children (born within 3 years prior to the survey) is based on anthropometric
measures observed at the time of the survey and is thus less likely to be affected by these biases.
A third possible explanation is that while the nutritional status of surviving girls has improved the
magnitude of the improvement is still not large enough to increase the likelihood of survival for
marginal girls. Medical research suggests that the leading contributors to child mortality are
respiratory ailments and infectious and gastro‐intestinal diseases. Malnutrition, while often
underlying and exacerbating these diseases, is not by itself a fatal factor, except in severe or
extreme cases.44 It is also possible that other types of parental investment, such as preventative
care, have a more important impact on reducing diseases and thus mortality.

VII.

CONCLUSION

In this paper, we study the impact of prenatal sex selection on the well‐being of girls in India,
which may come about through selection or/and treatment effect.

43

However, as we show in Appendix C, we find no evidence for differential changes in underreporting (or
recalls) of female deaths in the NFHS or any association between underreporting and sex selection.
44
Our indicators of malnutrition (i.e. anthropometric measures that are two standard deviations below the
median of the reference population) are often thought to reflect mild cases. Indeed, when examining
comparable samples (children aged 0‐35 months), we find only a moderate correlation between the
malnutrition indicators and mortality at the state level in each of the survey rounds. Our results are
consistent with those of Hill and Upchurch (1995) who report a lack of association between female mortality
disadvantage and nutritional disadvantage across several developing countries.

28

We explore these issues using data from the National Family Household Survey. We proxy
parental access to prenatal sex selection using the ratio of male to female births in the year and
state in which the child was born and provide evidence showing that starting in the 1990s high
sex ratios at birth reflect the practice of prenatal sex selection. We then analyze whether changes
in girls’ outcomes within states and over time in comparison to boys are associated with changes
in sex ratios at birth.
We find that an increase in the practice of prenatal sex selection leads to a differential
reduction in the incidence of malnutrition (underweight and wasting) among surviving girls. We
find no association between increases in prenatal sex selection and boys’ outcomes suggesting
that our findings for girls are unlikely to be driven by unobserved state‐time‐varying factors
associated with the increasing practice of prenatal sex selection. Our results are highly robust to
several checks that assess the likelihood of possible biases due to state‐specific differential trends
by gender associated with prenatal sex selection.
An exploration of the channels linking between prenatal sex selection and girls’ outcomes
shows no evidence for a selection of girls into families with different observable socio‐
demographic characteristics. However, consistent with some of the expected consequences of
prenatal sex selection, we find an increase in the proportion of girls born in families with weaker
son preferences. In addition, we find evidence of a larger reduction in family size for girls and a
larger decline in the average number of older siblings for girls, which is explained by the reduction
in the probability that girls are born at higher parities. We also find some suggestive evidence of
better treatment of girls as reflected in breastfeeding duration. On the other hand, we do not find
an increase in girls’ survival probabilities or a reduction in son preference as reported by mothers
in states with an increasing incidence of prenatal sex selection.

29

REFERENCES
Abrevaya, J. 2009. “Are There Missing Girls in the United States? Evidence from Birth Data.”
American Economic Journal: Applied Economics, Vol. 1, pp. 1‐34.
Almond, Douglas and Lena Edlund. 2008. “Son‐Biased Sex Ratios in the 2000 United States
Census.” Proceedings of the National Academy of Sciences, of the United States of
America, Vol. 105 No. 15, pp. 5681‐5682.
Almond, Douglas, Lena Edlund, and Kevin Milligan. 2009. “Son Preference and the Persistence of
Culture: Evidence from Asian Immigrants to Canada.” National Bureau of Economic Research,
Working Paper No. 15391.
Almond, Douglas, Hongbin Li, and Lingsheng Meng. 2010. “Son Preference and Early Childhood
Investments in China”, unpublished manuscript.
Ananat, Elizabeth O., Jonathan Gruber, B. Levine, and Douglas Staiger. 2009. “Abortion and
selection.” The Review of Economics and Statistics, Vol. 91. No. 1. pp. 124–136.
Arnold, Fred, Sunita Kishor, and T.K. Roy. 2002. “Sex‐Selective Abortions in India.” Population and
Development Review, Vol. 28 no. 4, pp. 759‐785.
Arnold, Fred, Minja Kim Choe, and T.K. Roy. 1998. “Son Preference, the Family‐ Building Process
and Child Mortality in India.” Population Studies, Vol. 52, No. 3, pp. 301‐315.
Anderson Siwan and Debraj Ray. 2012. “The Age Distribution of Missing Women in India.”
Economic and Political Weekly, Vol. 47, No. 47‐48, pp. 87
Barcellos, Silvia, Leandro Carvalho and Adriana Lleras‐Muney. 2010. “Child Gender and Parental
Investment in India: Are Boys and Girls Treated Differently?.” Working paper.
Becker, Gary and Nigel Tomes. 1979. “An Equilibrium Theory of the Distribution of Income and
Intergenerational Mobility.” Journal of Political Economy, Vol. 87, pp. 1153‐89.
Becker, Gary and Nigel Tomes. 1986. “Human Capital and the Rise and Fall of Families.” Journal of
Labor Economics, Vol. 4, pp. S1‐S39.
Ben‐Porath, Yoram and Finis Welch. 1976. “Do Sex Preferences Matter?” Quarterly Journal of
Economics, Vol. 90 No. 2, pp. 285‐307.
Bhalotra, Sonia and Tom Cochrane. 2010. “Where Have All the Young Girls Gone? Identification
of Sex Selection in India.” IZA Discussion Paper No. 5381.
Bharadwaj, Prashant and Leah K. Nelson. 2010. “Discrimination Begins In the Womb: Evidence Of
Sex‐Selective Prenatal Investments.” unpublished manuscript, UCSD.
Bhaskar, Venkataraman, and Gupta Bishnupriya. 2007. “India’s development in the era of
growth.” Oxford Review of Economic Policy, Vol 23, No. 2, pp. 135‐142.
Bhat, Mari P. N. 2002. “On the Trail of ‘Missing’ Indian Females – II: Illusion and Reality”, Economic
and Political Weekly, Vol. 37. No. 52, pp. 5244‐5263.
Case Anne and Christina Paxson. 2010. “Causes and Consequences of Early Life Health”,
Demography, Vol. 47S, pp. S65‐S85.
Chhabra, Rami, and Sheel C. Nuna. 1993. “Abortion in India: an overview.” New Delhi: Virendra
Printers.
Chandran S. Anil, Ramachandra Srikrishna and Sekhar P. Satya. 2011. “National Family Health
Survey‐3 Reported Low Full‐Immunization Coverage Rates in Andhra Pradesh, India: Who
is to be Blamed?” Journal of Public Health, pp. 1‐7.
30

Currie, J. 2009. “Healthy, wealthy and wise: socioeconomic status, poor health in childhood, and
human capital development”, Journal of Economic Literature, Vol. 47 No. 1, pp. 87–122.
Das, Gupta, Monica. 1987. “Selective Discrimination Against Female Children in Rural Punjab.”
Population and Development Review, Vol. 13 No. 1, pp. 77‐100.
Das, Gupta M., and P. N. Mari Bhat. 1997. “Fertility Decline and the Increased Manifestation of
Sex Bias in India.” Population Studies Vol. 51 No. 3, pp. 307‐315.
Deaton, Angus. 1997. “The Analysis of Household Surveys: A Microeconometric Approach to
Development Policy.” Baltimore and London: John Hopkins University Press.
Donohue, John J., and Levitt Steven. 2001. “The Impact of Legalized Abortion on Crime.” Quarterly
Journal of Economics, Vol. 116, no. 2, pp. 379‐420.
Dreze, Jean, Sen Amartya. 1997. “Indian Development: Selected Regional Perspectives.” Oxford:
Oxford University Press.
Dubuc, Sylvie, and Coleman David. 2007. “An Increase in the Sex Ratio of Births to India‐Born
Mothers in England and Wales: Evidence for Sex‐Selective Abortion.” Population and
Development Review, Vol. 33, No. 2, pp. 383‐400.
Duggal, Ravi 2004. “The Political Economy of Abortion in India: Cost and Expenditure Patterns”,
Reproductive Health Matters Vol. 12 No. 24, pp. 130‐137.
Duggal, Ravi, and Ramachandran Vimala. 2004. “The Abortion Assessment Project – India: Key
Findings and Recommendations.” Reproductive Health Matters, Vol. 12, pp. 122‐129.
Dyson, Tim, and Mick Moore. 1983. “On Kinship Structure, Female Autonomy, and Demographic
Behavior in India.” Population and Development Review 9 (1): 35‐60.
Everett, Jana 1998. “Indian Feminists Debate the Efficacy of Policy Reform: The Maharashtra Ban
on Sex‐Determination Tests”, Social Politics, Col. 5 No. 3, pp. 314‐337.
George, Sabu. M. 1997. “Female Infanticide in Tamil Nadu, India: From Recognition Back to
Denial?.” Reproductive Health Matters, Vol. 10, pp.124‐31.
George, Sabu M. 2002. “Sex Selection/Determination in India: Contemporary Developments.”
Reproductive Health Matters, Vol. 10 No. 19, pp. 190‐192.
George, Sabu M. 2006. “Sex Ratio in India.“ The Lancet, Vol. 367, p. 1725.
Goodkind, Daniel. 1996. "On Substituting Sex Preference Strategies in East Asia: Does Prenatal Sex
Selection Reduces Postnatal Discrimination?” Population and Development Review, Vol.
22, No. 1, pp. 111‐125.
Griliches, Zvi and Hausman, Jerry A.. 1986. “Errors in Variables in Panel Data”, Journal of
Econometrics, Vol. 31 Issue 1.
Hill, Kenneth, and Dawn M. Upchurch. 1995. “Gender Differences in Child Health: Evidence from
the Demographic and Health Surveys.” Population and Development Review, Vol. 21 No.
1, pp. 127‐151.
Hu, Luojia and Analia Schlosser. 2012. “Trends in Prenatal Sex Selection and Girls’ Nutritional
Status in India,” CESifo Economic Studies, Vol. 58, No. 2, pp.348‐372.
IIPS (International Institute for Population Sciences). 1995. National Family Health Survey (MCH
and Family Planning): India 1992–93. Bombay: International Institute for Population
Sciences.

31

Jacobsen, R., Møller, H., and Mouritsen A. 1999. “Natural Variation in the Human Sex Ratio.”
Human Reproduction, Vol. 14 No. 12, pp. 3120‐3125.
Jayachandran, Seema, and Ilyana Kuziemko. 2009. “Why Do Mothers Breastfeed Girls Less than
Boys? Evidence and Implications for Child Health in India.” NBER Working Paper No.
15041.
Jefferey, Roger, Jefferey Patricia, and Andrew Lyon. 1984. “Female Infanticide and
Amniocentesis.” Social Science and Medicine, Vol. 19 No. 11, pp. 1207‐1212.
Jensen, Robert. 2003. “Equal Treatment Unequal Outcomes? Generating Sex Inequality Through
Fertility Behavior” Mimeo, Harvard University.
Khanna, Sunil K. 1997. “Traditions and reproductive technology in an urbanizing north Indian
village.” Social Science & Medicine, Vol 44, No. 2, pp. 171‐180.
Kishor, Sunita. 1993. “May God Give Sons to All: Gender and Child Mortality in India.”
American Sociological Review, Vol. 58, No. 2, pp. 247‐265.
Kishor, S. 1993. “May God Give Sons to Us All: Gender and Child Mortality in India”, American
Sociological Review, Vol. 58 No. 2, pp. 247‐265.
Kishwar, M. 1995. “When Daughters are Unwanted: Sex Determination Tests in India.” Manushi
86: 15‐22.
Krugman, Andrea. 1998. “Being Female Can Be Fatal: an Examination of India’s Ban on Pre‐Natal
Gender Testing.” Cardozo Journal of International and Comparative Law, Issue 1, pp. 215‐
238.
Kulkarni, P. M. 2007. “Estimation of Missing Girls at Birth and Juvenile Ages in India”. Paper
commissioned by the United Nations Population Fund, UNFP.
Lin Ming‐Jen, Jin‐Tan Liu, and Nancy Qian. 2009. “More Missing Women, Fewer Girls Dying: The
Impact of Abortion on Sex Ratios at Birth and Excess Female Mortality in Taiwan”, NBER
Working Paper w14541
Mallik, Rupsa. 2003. “‘Negative Choice’ Sex Determination and Sex Selective Abortion in India.”
Abortion Assessment Project – India. Center for Enquiry into Health and Allied Themes
(CEHAT), Survey No. 2804 & 2805. Maharashtra, India.
Mahapatra, Prasanta (2010) “An Overview of the Sample Registration System in India”, Prince
Mahidol Award Conference and Global Health Information Forum.
Miller, Barbara. 1981. “The endangered sex: neglect of female children in rural north India.”
Ithaca: Cornell University Press. [re‐published with a new chapter, "The Girl Child in India
in the Year 2000," pp. 197‐217, Oxford University Press, 1997].
Miller, Barbara. 1984. “Daughter Neglect, Women's Work and Kinship: Pakistan and Bangladesh
Compared.” Medical Anthropology Vol. 8, No. 2, pp. 109‐126.
Pande, Rohini P. August 2003. “Selective Gender Differentials in Childhood Nutrition and
Immunization in Rural India: The Role of Siblings.” Demography Vol. 40, No. 3, pp 395‐
418.
Petrin, Amil and Train Kenneth 2003. “Omitted Product Attributes in Discrete Choice Models”,
NBER Working Paper 9452.
Portner, Claus C. 2010. “Sex Selective Abortion, Fertility and Birth Spacing.” mimeo, University of
Washington.
32

Ravindran, T.K.S. 2002. “Gender Gaps in Research on Abortion in India: A Critical Review of
Selected Studies”. The Gender and Reproductive Health Research Initiative. CREA. New
Delhi.
Retherford, Robert D., and T. K. Roy. 2003. “Factors affecting sex‐selective abortion in India and
17 major states.” National Family Health Survey Subject Report, No. 21.
Rose, Elaina. 1999. “Consumption Smoothing and Excess of Female Mortality in Rural India.” The
review of Economics and Statistics, Vol. 81 No. 1, pp. 41‐49.
Rosenzweig, Mark R., and Paul Schultz. 1982. “Market Opportunities, Genetic Endowments, and
Intrafamily Resource Distribution: Child Survival in Rural India.” The American Economic
Review, Vol. 72, No. 4, pp. 803‐815.
Saikia, Nadita, Domantas Jasilionis, Faujdar Ram, and Vladimir M. Shkolnikov (2011), “Trends and
Geographic Differentials in Mortality Under 60 in India”, Population Studies, Vol. 65, No
1, pp. 73‐89.
Shepherd, Mary Elizabeth. 2008. “The Selective Abortion in India: The Impact on Child Mortality.”
Cambria Press.
Sen, Amartya. 1992. “Missing Women.” BMJ, vol. 304, pp. 586‐587.
Sen, Amartya. 2003. “Missing Women – Revisited.”, BMJ, vol. 327, pp. 1297‐1299.
Simmons, George B., Celeste Smucker, Stan Bernstein, and Eric Jensen. 1982. “Post‐Neonatal
Mortality in Rural India: Implications of an Economic Model.” Demography, Vol. 19 No. 3,
pp. 371‐389.
Subramanian, S. V., and S. Selvaraj. 2009. “Social Analysis of Sex Imbalance in India: Before and
After the Pre‐Natal Diagnostic Techniques (PNDT) Act.” J Epidemiol Community Health
Vol. 63, pp. 245‐252.
Sundar, Ramamani, 2003. “Abortion Costs and Financing a Review”. Abortion Assessment Project
– India. Centre for Enquiry into Health and Allied Themes, Mumbai.
Timaeus, Ian, Harries Katie, Fairbairn Francesco. 1998. “Can use of health care explain sex
differentials in child mortality in the developing world.” in UNITED NATIONS (Ed.), Too
young to die: genes or gender? United Nations Population Division, New York.
Visaria, Pravin M. 1971. “The Sex‐ratio of the Population of India.” Census of India 1961, Volume
1: Monograph 10. New Delhi: Ministry of Home Affairs.
Wooldridge, Jeffrey 2002. “Economtric Analysis of Cross Section and Panel Data.” Cambridge, MA:
MIT Press.
World Health Organization. 2000. “Effect of Breastfeeding on Infant and Child Mortality Due to
Infectious Diseases in Less Developed Countries: A Pooled Analysis.” Lancet, Vol. 355, No.
9202, pp. 451‐455.
Yamaguchi, Kazuo. 1989. “A Formal Theory for Male‐Preferring Stopping Rules of Childbearing :
Sex Differences in Birth Order and in the Number of Siblings.” Demography, Vol. 26, No.
3, pp. 451‐465.

33

Figure 1: Male to Female Ratio (MFR) by State - Census data
1961

1971

1991

2001

1981

1.20 − 1.30
1.10 − 1.20
1.05 − 1.10
1.02 − 1.05
1.00 − 1.02
0.98 − 1.00
0.90 − 0.98
0.00 − 0.90
No data

Notes: The figure plots estimates of the Male-to-Female Ratio by state based on census counts of children alive at age 0 (0 to 11
months) for the different years.

−.05

−.05

0

0

.05

.05

.1

.1

Figure 2: Differential Probability of a Male Birth by Year of Birth

1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006
Year
High income households in son preference states
Other households

(a) Parity 2

1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006
Year
High income households in son preference states
Other households

(b) Parity 3

Notes: The figure plots estimates for the differential probability of a male birth by year - (7-year moving average) comparing
between households with no boys among previous children and the rest (panel a for parity two and panel b for parity three). The
full line plots the differential probability for households in the 4th or 5th wealth quintile who live in states with son preferences
above median (according to first round of NFHS). See data appendix for the definition of son preference measure. The dotted line
plots the differential probability for all other households.

Figure 3: MFR By State
Arunachal Pradesh (22,679)

Assam (550,480)

Bihar (1,681,867)

Delhi (202,423)

Goa (18,711)

Gujarat (892,560)

Haryana (418,590)

Himachal Pradesh (121,171)

Jammu & Kashmir (.)

Karnataka (887,193)

Kerala (493,390)

Madhya Pradesh (1,848,996)

Maharashtra (1,767,409)

Manipur (29,782)

Meghalaya (53,313)

Mizoram (14,410)

Nagaland (23,792)

Orissa (669,370)

Punjab (431,370)

Rajasthan (1,268,112)

Sikkim (9,951)

Tamil Nadu (844,730)

Tripura (60,737)

.8

1

1.2 1.4

.8

1

1.2 1.4

.8

1

1.2 1.4

.8

1

1.2 1.4

.8

1

1.2 1.4

.8

1

1.2 1.4

Andhra Pradesh (1,286,860)

1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005

West Bengal (1,476,420)

.8

1

1.2 1.4

Uttar Pradesh (3,393,325)

1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005

NFHS

Interpolat.

Interpolat. adj IM

Census

Notes: The figure plots the Male-to- Female Ratio by state based on alternative data sources. The solid line (NFHS) is based
on a 7-year moving average of the ratio of the number of male births to female births from the NFHS data, the dashed line
(Interpolat.) interpolates the Census MFR for inter-census years, where the interpolation is based on a fourth order polynomial
function estimated from the NFHS data. The dotted line (interpolate. adj. IM) is based on census MFR adjusted for infant
mortality and inter-census years interpolated using NFHS data. The dots (Census) denote the MFR computed from census data
(number of children aged 0-11 months at the census date).

Table 1. Summary Statistics
Nutrition sample
Girls
Boys
Difference
(1)
(2)
(3)

Girls
(4)

Urban

0.236

0.237

0.232

0.235

Index of mass media exposure

0.799

0.824

0.738

0.755

Wealth index

2.82

2.86

2.74

2.77

Mother's Age

25.2

25.3

28.7

28.7

Mother's age at 1st birth

19.0

19.0

18.7

18.7

Mother's education
No education

0.541

0.538

0.613

0.607

Primary school

0.158

0.148

0.146

0.147

Secondary school

0.247

0.256

0.200

0.203

Higher

0.053

0.057

0.040

0.042

Missing

0.001

0.001

0.001

0.001

Father's education
No education

0.300

0.291

0.341

0.339

Primary school

0.193

0.189

0.205

0.201

Secondary school

0.381

0.389

0.346

0.350

Higher

0.119

0.124

0.101

0.103

Missing

0.006

0.007

0.006

0.007

0.792

0.792

0.791

0.794

Muslim

0.158

0.156

0.161

0.156

Other religion

0.049

0.052

0.047

0.049

Missing

0.001

0.001

0.001

0.001

Number of children in the family

2.920

2.927

3.861

3.736

Mother wants another child

0.486

0.363

0.292

0.211

Sample size

36,940

39,560

172,472

185,337

Religion
Hindu

‐0.002
(0.003)
‐0.025
(0.010)
‐0.041
(0.009)
‐0.098
(0.024)
0.000
(0.025)
0.004
(0.004)
0.009
(0.003)
‐0.010
(0.005)
‐0.004
(0.002)
0.000
(0.000)
0.009
(0.003)
0.004
(0.005)
‐0.008
(0.006)
‐0.005
(0.003)
‐0.001
(0.001)
0.001
(0.002)
0.002
(0.003)
‐0.003
(0.002)
0.000
(0.000)
‐0.007
(0.009)
0.123
(0.009)
76,500

Mortality sample
Boys
Difference
(5)
(6)
‐0.003
(0.002)
‐0.017
(0.005)
‐0.030
(0.009)
‐0.047
(0.019)
‐0.023
(0.013)
0.006
(0.003)
‐0.001
(0.002)
‐0.004
(0.002)
‐0.002
(0.001)
0.000
(0.000)
0.002
(0.002)
0.003
(0.003)
‐0.004
(0.002)
‐0.002
(0.002)
0.000
(0.000)
‐0.003
(0.002)
0.005
(0.001)
‐0.002
(0.002)
0.000
(0.000)
0.125
(0.015)
0.081
(0.006)
357,809

Notes: The table reports summary statistics for boys and girls (cols. 1,2,4, and 5) included in the analysis samples and differences between
the characteristics of girls and boys (cols. 3 and 6). Standard errors of the differences clustered at the state level are reported in parenthesis.
The samples pool rounds 1, 2, and 3 of the NFHS. The nutrition sample reported in columns 1‐3 includes the last two children under three
years of age of ever married women with valid anthropometric data. The mortality sample reported in columns 4‐6 includes all children
born within the last 10 years preceding the survey date of ever married women. Observations are weighted using national‐level weights.

Table 2. Predictors of State Changes in MFR
(1)
(2)

(3)

(4)

Income per capita

0.999
(0.456)

1.209
(0.369)

‐5.519
(4.141)

Income per capita x 2001

1.080
(0.340)

1.193
(0.331)

‐4.073
(3.026)

2001

‐1.890
(1.860)

‐3.604
(4.411)

‐12.65
(4.927)

10.89
(13.094)

Ideal MFR

11.84
(2.976)

13.99
(2.889)

‐7.887
(13.493)

Ideal MFR x 2001

5.394
(3.512)

7.520
(3.319)

‐9.607
(9.442)

Income per capita x Ideal MFR

4.943
(3.136)

Income per capita x Ideal MFR x 2001

3.870
(2.257)

Number of observations
R‐sq

52
0.330

52
0.254

52
0.576

52
0.632

Notes: The Table reports regression estimates where the dependent variable is the state MFR (in
1991 or 2001). The explanatory variables are income per capita in 1990 (in thousand Rs), ideal MFR
(averaged at the state level), a dummy for year 2001 and interactions between income or ideal MFR
and year 2001 dummy. The model in column (4) includes also interactions between income per
capita and ideal MFR. See data appendix for definition of ideal MFR. Standard errors clustered at
the state level are reported in parenthesis.

Table 3. Differential Probability of a Male Birth at Parity N
as a Function of the Sex Composition of Older Siblings
Born between 1975‐1989
Born after 1989
No controls
(1)

Full controls
(2)

No controls
(3)

Full controls
(4)

No sons

‐0.005
(0.003)

‐0.005
(0.003)

0.019
(0.007)

0.020
(0.007)

Sample Size

45,198

45,198

74,422

74,422

No sons

‐0.005
(0.011)

‐0.006
(0.011)

0.039
(0.012)

0.036
(0.012)

1 son

0.005
(0.007)

0.005
(0.007)

0.012
(0.009)

0.011
(0.009)

p‐value for test (No sons=1 son)

0.268

0.206

0.003

0.004

30,804

30,804

45,200

55,289

A. Parity 2 (omitted category=Boy)

B. Parity 3 (omitted category=Boy‐Boy)

Sample Size

Notes: The table reports the differential probability of a male birth at parity 2 (panel A) and parity 3 (panel B) as a
function of the sex composition of older siblings. The samples include all children born in the 15 years prior to each
survey date. Estimates reported in columns 1‐2 are for children born between 1975 and 1989. Estimates reported in
columns 3‐4 are for children born in 1990 or afterwards. Regression estimates reported in columns 2 and 4 are
from models that control also for twin status, mother's age, mother's education, mother's age at first birth,
indicators for mother's religion, father's education, mother's mass media exposure, wealth, and rural/urban status.
Observations are weighted using national‐level weights. Standard errors clustered at the state level are reported in
parenthesis. Sample sizes are reported in italics .

Table 4. Sex‐Ratios and Ultrasound Use
Ultrasound Test
Full Sample

Pregnancy outcome = boy
Ultrasound Sample
No Ultrasound Sample

No controls
(1)

Full controls
(2)

No controls
(3)

Full controls
(4)

No controls
(5)

Full controls
(6)

No sons

0.025
(0.011)

0.019
(0.008)

0.077
(0.016)

0.078
(0.017)

0.025
(0.017)

0.025
(0.016)

Sample Size

14,048

14,048

5,098

5,098

8,950

8,950

No sons

0.111
(0.027)

0.081
(0.022)

0.124
(0.056)

0.103
(0.055)

‐0.010
(0.026)

‐0.013
(0.026)

1 son

0.030
(0.015)

0.019
(0.015)

0.034
(0.068)

0.025
(0.067)

0.008
(0.033)

0.007
(0.033)

p‐value for test (No sons=1 son)

0.000

0.000

0.001

0.006

0.218

0.217

Sample Size

8,069

8,069

1,922

1,922

6,147

6,147

A. Parity 2 (omitted category=1 son)

B. Parity 3 (omitted category=2 sons)

Notes: Columns 1 and 2 report the differential likelihood that a mother performs an ultrasound test during pregnancy as a function of the sex
composition of her older children. Columns 3 and 4 report the differential likelihood of a male birth as a function of the sex composition of the older
children among mothers who reported to have performed an ultrasound test during that pregnancy. Columns 5 and 6 report the differential likelihood of
a male birth as a function of the sex composition of the older children among mothers who reported that they did not performed an ultrasound test
during that pregnancy. Panels A and B report estimates for parities 2 and 3, respectively. The omitted category in all regressions is one son for parity two
and two sons for parity three. Regression estimates reported in columns 2, 4 and 6 are from models that control also for twin status, mother's age,
mother's education, mother's age at first birth, indicators for mother's religion, father's education, mother's mass media exposure, wealth, rural/urban
status, and year of birth and state fixed effects. Observations are weighted using national‐level weights. Standard errors clustered at the state level are
reported in parenthesis. Sample sizes are reported in italics .

Table 5. Effects on Nutritional Status of Children
MFR from NFHS

Basic specification

Outcome

Sample
size
(1)

Outcome
mean
(2)

MFR
(3)

Female
x MFR
(4)

Full controls
MFR
(5)

Female
x MFR
(6)

MFR from Census and
inter‐census
interpolation

MFR from Census
adjusted for infant
mortality and inter‐
census interpolation

MFR
(7)

Female
x MFR
(8)

MFR
(9)

Female
x MFR
(10)

Underweight

76,314

0.485

‐0.162
(0.336)

‐0.535
(0.160)

‐0.095
(0.318)

‐0.536
(0.132)

‐0.025
(0.249)

‐0.431
(0.135)

‐0.024
(0.253)

‐0.426
(0.135)

Wasted

69,784

0.179

‐0.050
(0.284)

‐0.247
(0.109)

‐0.029
(0.275)

‐0.250
(0.104)

‐0.122
(0.220)

‐0.301
(0.124)

‐0.117
(0.218)

‐0.292
(0.126)

Stunted

69,571

0.433

0.316
(0.230)

‐0.287
(0.180)

0.386
(0.236)

‐0.293
(0.164)

0.385
(0.292)

‐0.280
(0.181)

0.350
(0.293)

‐0.265
(0.182)

Notes: Sample sizes are reported in column 1. Means of the dependent variables are reported in Column 2. Columns 3 and 4 report regression estimates for
MFR and MFR interacted with a female dummy from a linear probability model that controls for state, survey round, year of birth fixed effects and their
interactions with gender. Columns 5 through 10 report regression estimates from models that control also for twin status, mother's age, mother's education,
mother's age at first birth, indicators for mother's religion, father's education, mother's mass media exposure, wealth, and rural/urban status. Estimates in
columns 3 though 6 come from models that use MFR based on the NFHS data as the main explanatory variable. Estimates in columns 7 and 8 come from
models that use MFR constructed from census data and inter‐census interpolation and estimates in columns 9 and 10 come from models that use the same
data but adjust census numbers for infant mortality. See data Appendix for details about the construction of the MFR series. The sample includes children
aged 0 to 35 months born in the last two births of ever married women sampled in rounds 1‐3 of the NFHS surveys. Sample sizes are smaller for wasted and
stunted because height measures were not taken in round 1 for the following 5 states: Andhra Pradesh, Himachal Pradesh, Madhya Pradesh, Tamil Nadu and
West Bengal. Observations are weighted using national‐level weights. Standard errors clustered at the state level are reported in parenthesis.

Table 6. Outcomes by Probability of Ultrasound Use
Female
Underweight
(1)
(2)
A. Full Sample

Wasted
(3)

Stunted
(4)

Prob. of ultrasound use

5.251
(1.505)

‐10.184
(4.857)

‐1.191
(4.304)

‐21.513
(6.876)

Year of birth

‐0.000
(0.000)

‐0.010
(0.001)

‐0.001
(0.002)

‐0.013
(0.002)

Prob. of ultrasound use x Year of birth

‐0.003
(0.001)

0.005
(0.002)

0.001
(0.002)

0.011
(0.003)

Female x prob. of ultrasound

16.392
(6.024)

4.818
(5.836)

13.363
(5.789)

Female x Year of birth

0.004
(0.002)

0.003
(0.002)

0.002
(0.002)

Female x prob. of ultrasound x year of birth

‐0.008
(0.003)

‐0.002
(0.003)

‐0.007
(0.003)

76,167

69,654

69,444

Number of observations

97,347

B. Low probability of ultrasound sample (25 percentile ‐ prob.< 0.078)
Year of Birth

‐0.00038
(0.00041)

Female x Year of birth

Number of observations

24,630

‐0.007
(0.001)

‐0.001
(0.001)

‐0.012
(0.002)

0.004
(0.002)

0.003
(0.002)

0.002
(0.002)

18,306

16,981

16,941

Notes: The table reports the probability of a female birth (column 1) and nutritional outcomes (columns 2
through 4) by the estimated probability of ultrasound use and additional covariates. Panel A uses the full
sample of children aged 0 to 35 months born in the last two births of ever married women sampled in rounds
1‐3 of the NFHS surveys. Panel B includes only children born in households with a predicted probability of
ultrasound use below the 25 percentile (prob.<0.078). Estimates reported in column 1 come from models
that control also for state fixed effects. Estimates reported in columns 2‐4 come from models that control for
state fixed effects interacted with gender. Observations are weighted using national‐level weights.
Bootstrapped standard errors clustered at the state level are reported in parenthesis.

Outcome

Table 7. Family Characteristics of Girls vs. Boys
Sample
Outcome
Females
size
mean
‐Males
(1)
(2)
(3)

Mother's age

76,500

25.3

MFR
(4)

Female
x MFR
(5)

‐0.098

2.952

‐1.817

(0.024)

(2.268)

(1.703)

Mother's age at 1st birth

76,500

19.0

0.000
(0.025)

4.863
(1.704)

0.785
(0.873)

Mother's education

76,394

3.62

‐0.118
(0.042)

6.489
(1.769)

0.243
(1.419)

Father's education

76,032

6.15

‐0.137

4.538

0.682

(0.051)

(1.215)

(2.275)

Wealth index

76,500

2.84

‐0.041
(0.009)

‐0.268
(0.623)

‐0.039
(0.522)

Rural

76,500

0.763

0.002
(0.003)

‐0.300
(0.121)

0.158
(0.159)

Number of children

76,500

2.92

‐0.007
(0.009)

‐1.421
(0.672)

‐0.923
(0.534)

Number of older siblings

76,500

1.82

‐0.026
(0.009)

‐1.172
(0.593)

‐0.980
(0.557)

Notes: Column 2 reports means of the dependent variables and column 3 reports the female‐male
differential. Columns 4 and 5 report regression estimates for MFR and MFR interacted with a female
dummy from a model that includes gender specific state and year of birth fixed effects. The sample
includes children aged 0 to 35 months born in the last two births of ever married women sampled in
rounds 1‐3 of the NFHS surveys. Observations are weighted using national‐level weights. Standard
errors clustered at the state level are reported in parenthesis.

Table 8. MFR and Son Preference

Outcome

Sample
size
(1)

Outcome
mean
(2)

Females
‐Males
(3)

MFR
(4)

Female
x MFR
(5)

Ideal MFR I

82,040

1.37

‐0.090
(0.009)

0.413
(0.208)

‐0.284
(0.155)

Ideal MFR II

82,040

1.35

‐0.076
(0.009)

0.387
(0.201)

‐0.343
(0.155)

Wants more sons
than daughters

85,504

0.36

‐0.063
(0.005)

0.115
(0.144)

‐0.216
(0.121)

Child of wanted
gender

85,686

0.71

‐0.149
(0.009)

0.242
(0.136)

0.143
(0.091)

Notes: The dependent variables in rows 1 and 2 are based on the ratio of the ideal number of
sons to the ideal number of daughters reported by mothers. The dependent variable in row 3
is an indicator for whether the mother reports a desire to have a larger number of sons than
daughters. The dependent variable in row 4 is an indicator for whether the child is of wanted
gender which is constructed by comparing between mother's responses to ideal number of
sons and daughters and sex composition of older siblings. See data appendix for more details
about the construction of these variables. Means of the dependent variables are reported in
Column 2. Column 3 reports differences in means of the dependent variables between
females and males. Columns 4 and 5 report regression estimates from a model that includes
state and year of birth fixed effects and their interactions with female and controls also for
twin status, mother's age, mother's education, mother's age at first birth, indicators for
mother's religion, father's education, mother's mass media exposure, wealth, and rural/urban
status. The sample includes children aged 0 to 35 months born in the last two births of ever
married women sampled in rounds 1‐3 of the NFHS surveys. Observations are weighted using
national‐level weights. Standard errors clustered at the state level are reported in
parenthesis.

Outcome
Breastfed for:

Table 9. Breastfeeding Duration
Sample
Outcome
Females
size
mean
‐Males
(1)
(2)
(3)

MFR
(4)

Female
x MFR
(5)

At least 12 months
(age>=12 months)

59,670

0.881

‐0.011
(0.004)

0.093
(0.132)

‐0.058
(0.133)

At least 18 months
(age>=18 months)

43,295

0.707

‐0.038
(0.007)

0.128
(0.190)

‐0.206
(0.230)

At least 24 months
(age>=24 months)

29,537

0.544

‐0.056
(0.007)

0.037
(0.186)

0.391
(0.155)

Notes: Means of the dependent variables are reported in Column 2. Column 3 reports differences in means of the
dependent variables between females and males. Columns 4 and 5 report regression estimates for MFR and MFR
interacted with female from a linear probability model that controls for state and year of birth fixed effects and their
interactions with female and the set of covariates specified in Table 5. In addition, the models control for age in
months and age in months interacted with a female dummy. The minimum age restriction used in each of the
regressions is specified in the first column of the table. Observations are weighted using national‐level weights.
Standard errors clustered at the state level are reported in parenthesis.

Table 10. Effects on Survival Probability

Outcome

Sample
(1)

Children Born within 36 months preceding survey
Sample
Outcome
Females
Female
size
mean
‐Males
x MFR
MFR
(2)
(3)
(4)
(5)
(6)

Children Born within 120 months preceding survey
Sample
Outcome
Females
Female
size
mean
‐Males
x MFR
MFR
(7)
(8)
(9)
(10)
(11)

Survived beyond first month

Age ≥
1 month

98,922

0.958

0.007
(0.002)

‐0.010
(0.031)

0.024
(0.047)

356,361

0.952

0.007
(0.001)

‐0.001
(0.031)

0.019
(0.034)

Survived beyond 12 months

Age ≥
12 months

63,961

0.936

0.004
(0.002)

0.028
(0.063)

0.029
(0.098)

323,916

0.925

0.003
(0.002)

0.010
(0.058)

0.003
(0.039)

Survived beyond age 5

Age ≥
60 months

‐‐‐

‐‐‐

‐‐‐

‐‐‐

‐‐‐

188,924

0.892

‐0.004
(0.005)

‐0.062
(0.080)

‐0.031
(0.067)

Notes: The sample for columns 2‐6 includes all children aged 0 through 35 months born in the last two births to ever married women sampled in rounds 1‐3 of the NFHS surveys. The
sample for columns 7‐11 includes all children born in the last ten years prior to survey date to ever married women sampled in rounds 1‐3 of the NFHS surveys. The minimum age restriction
used for each outcome is specified in column 1. Columns 3 and 8 report means of the dependent variables and columns 4 and 9 report the female‐male differential in the dependent
variables. Columns 5 and 6 and columns 10 and 11 report regression estimates for MFR and MFR interacted with a female dummy from a linear probability model that includes gender
specific state and year of birth fixed effects and control also for the covariates specified in table 5. Observations are weighted using national‐level weights. Standard errors clustered at the
state level are reported in parenthesis.

Appendix A – Data description
Definitions of MFR
MFR from NFHS data
We use MFR constructed from the NFHS based on the complete birth histories reported
by mothers of children born in the last 15 years. We focus on birth events that took place
in the last 15 years because a large proportion of children aged 15 and above do not live
in the household so misreporting of births is more likely for older children. In addition,
Anderson and Ray (2012) find that under reporting of deaths, and female deaths in
particular, is more severe for ages above 15. We pool the three survey rounds and
summarize the number of female and male births by state and year of birth. We define
MFR as the total number of male births divided by the total number of female births in a
given state and a year. To reduce volatility due to sampling error, we use a smoothed
version of MFR which is computed as a 7‐year moving average of the ratio of the number
of male births to female births by year and state. Note that pooling data from the three
survey rounds dilutes a bit the problem of recall bias since we count children born in the
same year who are at different ages at the survey date. Smoothing the data reduces also
possible errors of age misreporting and age heaping. We assessed the sensitivity of our
results to the specific smooth window in rows 10 and 11 of Appendix Table A4. Estimates
are highly consistent with our main results. As an additional robustness check, we also
computed MFR using only children born in the last 10 years (instead of 15). Results (not
reported here but available upon request) are virtually unchanged.

MFR from Census
We also construct an alternative measure of MFR that incorporates the Census data for
the years 1981, 1991, 2001, and interpolates inter‐census years using NFHS data. For each
state, we specify a fourth order polynomial function to interpolate the MFR for inter‐
census years. We estimate the parameters of the function by running a constrained
nonlinear regression of the MFR from the NFHS with the constraint that the fitted curve
goes through the Census MFR in the 3 census years. We weight each cell by the number
of observations in the NFHS data. We also experimented using a higher or lower order of
polynomial to interpolate the inter‐census years. Results (not shown here but available
upon request) are highly consistent with our main results.

1

MFR from Census adjusted for infant mortality
Because census data reports only population counts of children who are alive at the
enumeration day, we also re‐estimate the polynomial function adjusting the census MFR
for infant mortality. For this purpose, we compiled infant mortality state data from
publications of the Sample Registration System (SRS) which is the main data source for
Indian vital statistics. Although the SRS data is a much larger sample than the NFHS, it still
suffers from sampling error and misreporting (see Saikia et al, 2011 and Mahapatra,
2010). Publications for infant mortality are not available for the exact census years and
data are missing for some states. We therefore use 1985 data to adjust census counts of
1981 and 1990 data to adjust census counts of 1991. Infant mortality data is missing for
the following states and years: Arunachal Pradesh (1985, 1990), Delhi (1985, 1990), Goa
(1985, 1990), Himachal Pradesh (1990), Jammu & Kashmir (1990), Manipur (1985, 1990),
Meghalaya (1985, 1990), Mizoram (1985, 1990), Nagaland (1985, 1990), Sikkim (1985,
1990), and Tripura (1985, 1990). For these cases, we use data from the closest year
available.
Son Preference Measures
Mothers were asked in NFHS survey about desired number of children (n) , desired
number of boys (rB), desired number of girls (rG), and desired number of children for
which sex would not matter (rO = n‐ rB ‐ rG ). We take numeric responses to these questions
(about 95 percent of the cases) and define two alternative measures of son preference as
follows:
The first measure is computed as
0.5 ∗
0.5 ∗
The second measure is computed as
∑

∑

∑

∑

where
!
!

2

1
! 2

The intuition behind the second measure is the following. Consider a woman who reports
wanting n children. If she has neutral preference for boys and girls (rB=0 and rG=0), her
ideal MFR is given by (n/2)/(n/2)=1. Now suppose she wants rB boys (rB>0). To find the
corresponding ideal MFR, we first find all possible sequences that would violate her
where k is the actual number of boys. We then calculate the
preference
associated probabilities for these sequences, Pr(B=k), for all relevant k, where Pr(B=k), is
the pdf for a binomial distribution
, , given by the equation above. Next we
hypothetically re‐assign sex to the undesired sequences to satisfy her preference
(assuming she can selectively abort girls) and compute the resulting ideal MFR.
Specifically, we adjust the neutral ideal MFR by adding to the numerator and subtracting
from the denominator the total number of desired boys from all the undesirable
1
).
sequences (∑
For example, suppose a family wants 3 children, 1 boy, and the other two do not matter
(n=3, rB=1 and rG=0). There are eight possible sequences of boys and girls born into this
family, each with probability 1/8. There is only one sequence (three girls) that is
undesired. The family could replace this with one boy and two girls (e.g. by selectively
aborting a third girl until it conceives a boy). Across the 24 births in the eight sequences
this would yield 13 boys and 11 girls, or an MFR of 1.18. And this can be verified by the
above formula:

2

3
2
3
2

∑

1

1

1

∑

1

1

1

1.18

3
2
3
2

1

0

0

1

0

0

3
2
3
2

1
2
1
2

Similar calculations can be done for the case where she wants rG girls (rG>0). The general
formula above includes both cases (assuming families would selectively abort boys as well
as girls).

One problem with these measures (ideal MFR1 and Ideal MFR2) is that while they work
well for most cases, they might become ill‐defined in some extreme cases. For example,
for women who report that they want all boys (
,
0), these measures
become ∞ and thus the data is missing.

3

We therefore consider an additional measure of son preference: an indicator for wanting
more sons than girls:
1
Finally, we construct a measure for a child of wanted gender. Specifically, we define an
0.5 ∗
indicator for a female birth being wanted to be 1 if the desired number of girls (
) is greater than the actual number of older sisters alive at the girl’s conception and 0
otherwise. The indicator is defined in a similar way for a male birth.

4

Appendix B – Robustness Tests
We consider alternative definitions for nutritional outcomes: we use the continuous z‐
scores and also examine an alternative measure applied sometimes in the medical
literature defined as the ratio of a measured value in the individual to the median value
of the reference population for the same age and sex, expressed as a percentage: percent
of the median. We also tested the sensitivity of our results to the use of specific growth
standards. As noted above, in 2006 the World Health Organization published new growth
standards for assessing children’s nutritional status. We re‐defined the anthropometric
indicators using the new WHO tables and re‐estimated the main specification. Estimates
using these alternative outcomes reported in Appendix Table A3 confirm our main
findings – girls’ nutritional outcomes improved in places with increasing MFR. In this case,
the coefficient for the interaction term between MFR and female is also significant for the
height outcomes, although, as before, the main MFR effect points to a worse outcome
for boys (but not significant). Taken together, the results suggest an increase in the
practice of prenatal sex selection leads to an improvement in girls’ nutritional outcomes
(especially weight and weight for height) among surviving girls.
We also performed additional tests to assess the validity of our identifying
assumptions and check for the robustness of the main results. The results of these tests
are reported in Appendix Table A4. In order to facilitate comparison, we reproduce the
estimates obtained in our main specification in the first row of the table. Overall, the
additional estimates are qualitatively similar across various models and specifications.
First, we assess the validity of our main identifying assumption that there are no
changes in unobserved factors associated with MFR that could affect girls’ outcomes.
Note that this is partially addressed by the triple‐difference strategy which controls for
state‐level time‐varying factors that affect boys and girls similarly by including a main MFR
effect. As we note before, the main effect of MFR is never significant and is inconsistent
across outcomes suggesting that conditional on other covariates included in our model,
changes in MFR are not associated with changes in other factors that affect boys’
outcomes. We also consider a more flexible specification with the main MFR effect being
replaced by a set of full interactions between state‐ and year‐of‐birth fixed effects to allow
for unobserved state‐specific time‐varying factors. Estimates for the interaction term
between female and MFR are virtually unchanged for the three outcomes (see Table A4
row 2) suggesting that our results are unlikely to be driven by state‐time‐varying
unobserved factors associated with MFR that affect child outcomes.

5

Still, there may be state‐time‐varying factors that differentially affect boys and girls.
This could take place if increases in MFR are related to more rapid economic development
and modernization, which in turn may affect girls’ and boys’ health outcomes
differentially (for example, by providing better health care access or by means of a
reduction in discrimination against girls). A possible way to account for unobserved time‐
varying factors that differentially affect both genders would be to control for state‐
gender‐specific trends. However, this approach is less suitable to our set‐up as we do not
have enough observations from cohorts born before prenatal sex selection to extrapolate
a long‐term trend. Moreover, MFR evolves following a close to linear trend after the
diffusion of prenatal sex selection. Therefore, the inclusion of gender‐state‐specific linear
trends absorbs most of the variation in MFR and magnifies any possible measurement
error in this variable (Griliches and Hausman, 1986).
We therefore apply an alternative approach where we estimate expanded versions of
equation (1) in which we control for household and time‐varying state characteristics and
their interactions with gender. Specifically, we estimate three models with alternative
sets of additional controls. In the first set, we add interactions between household
covariates and a female dummy to the basic model. In the second and third specifications,
we include, in addition to gender‐specific controls for household characteristics, two
alternative sets of state‐level time‐varying covariates interacted with gender. The first set
of state variables is compiled from the NFHS data and includes state means of the wealth
index and the proportion of households in urban areas, proportion with electricity access,
and proportion with access to TV. The alternative set contains state information on net
domestic product and per‐capita net domestic product (compiled from reports of the
India’s central bank ‐‐ the Reserve Bank of India).
The estimates for the three expanded models, reported in rows 3 through 5 of Table
A4, are largely similar to those obtained in our main specification suggesting that our
results are unlikely to be driven by unobserved changes at the state level correlated with
MFR that had a differential impact on boys and girls.
We further assess the likelihood of the main identifying assumption by examining the
sensitivity of our results to the inclusion of additional covariates associated with
improvement in health outcomes. We first add to our main specification controls for birth
weight and its interaction with female. While birth weight could itself be affected by
prenatal sex selection, we are interested in the sensitivity of our results to the inclusion
of this covariate and its interaction with female as it should absorb any changes at the
state level that differentially affect boys and girls nutritional status at birth.
Unfortunately, only 30 percent of the sample has data on birth weight. However, almost
all children (99 percent) have valid information regarding their size at birth as reported

6

by mothers (very large, larger than average, average, smaller than average or very small).
We find that mother’s report on size at birth is highly correlated with birth weight for the
subsample of children who have valid data on birth weight. We therefore impute birth
weight for those with missing data using size at birth and all additional covariates.1 We
then re‐estimate our main model after including controls for birth weight, an indicator for
imputed birth weight, and their interactions with female. In an alternative specification,
we include indicators for size at birth and their interactions with female instead of birth
weight.
The results of these two specifications are reported in rows 6 and 7 of Table A4 and
confirm that our estimates of MFR and their interaction with female are largely
unaffected by the inclusion of these additional covariates.2 This result suggests that for
our estimates to be biased, there should be unobserved time‐varying factors correlated
with MFR that differentially affect boys’ and girls’ nutritional outcomes after birth but not
at birth.
As an additional check, we also estimated a specification that controls for delivery at
a health facility and its interaction with female in an attempt to proxy for improvements
in the availability of health care that could differentially affect boys and girls outcomes.
Again in this case, we find that our results (reported in row 8 of the table) are not sensitive
to the addition of these additional covariates.
In rows 9‐11 of the table we assessed the robustness of our results to the specific
definition of the main explanatory variable. Note that our basic specification includes MFR
in a linear fashion although it may be the case that the relationships between MFR and
the outcomes of interest are nonlinear. We therefore re‐estimated equation (1) after
replacing MFR with its natural logarithm. The estimated effects reported in row 9 are of
comparable magnitude to our main results.3 We also experimented with different ways
1

Specifically, we impute birth weight by estimating separate regressions for each gender and
survey round of birth weight on size at birth, year of birth, state fixed effects, and all household
covariates.
2
Note that if prenatal sex determination affects birth weight the estimates from these two
specifications that control for birth weight and its interaction with female would be attenuated.
Interestingly, we find that our estimates are not reduced after controlling for birth weight. Indeed,
in an additional analysis we estimated two models where we regressed birth weight or an indicator
for small size at birth on MFR and its interaction with female and all additional covariates specified
in equation (1) and found that the estimates of MFR and its interaction with female were not
significant.
3
For example, the coefficient on female*ln(MFR) in the underweight regression is ‐0.609
(s.e.=0.147) which implies that a 0.2 ‐point increase in MFR from 1.05 to 1.25 (or 17%) is associated
with a decrease of 10 percentage points (17*0.609) in the proportion of underweight girls, which
is similar to the results reported in Table 5.

7

of smoothing MFR. For example, we considered both a shorter and a longer window for
computing the moving average. The results, which are reported in rows 10 and 11, are
similar to those obtained from our main specification.
Finally, we also looked at the sensitivity of our results to the weighting scheme of the
data by re‐estimating our model using unweighted data (row 12) and using state weights
(row 13). All the estimates are similar to the main results.
We also estimate the impacts of prenatal sex selection using a discrete version of the
triple‐differences approach, where we stratify states by region and compare between
changes in outcomes of girls relative to boys in northern and western states (regions with
the sharpest increase in sex ratios at birth) versus all the rest. Results were very much in
line with the findings presented in this paper. Namely, we observe an improvement in
girls’ nutritional outcomes (relative to boys) in north‐western states. For details, see Hu
and Schlosser (2012).

Appendix C – Exploring Differential Recall of Death by Gender and Its Association with
Sex‐Selection
There is no direct way to measure underreporting of deaths in the NHFS. However, we
can partially address this issue by comparing deaths rates of the same cohort over the
survey rounds, both for the full sample and by gender. Specifically, we compute the death
rate for children during their first year of life (among children who were older than 1 year
of age) and compare differences in death rates of the cohorts that are observed in two
survey rounds.4
We do this by estimating the following model:
_0_1

∗

Where Died_0_1 is an indicator for whether the child i born in state s in year t and
observed in survey round j died between ages 0 and 11 months. and are vectors of
state and year of birth fixed effects respectively. Post is an indicator for the later survey
round (round 2 for cohorts 1989‐1991 and round 3 for cohorts 1996‐1998) and female is
an indicator for a female child. If there are differential changes in the reporting of female
deaths over time, we should find that is statistically significant.

4

Given the age restrictions, we can compare death rates of the 1989‐1991 cohorts between rounds
1 and 2 of the survey and death rates of the 1996‐1998 cohorts between rounds 2 and 3 of the
survey.

8

We estimate the model comparing death rates in rounds 1 and 2 and in rounds 2 and 3
and also pooling the three survey rounds. The results are reported in Appendix Table A.8.
As seen in columns 1 through 3, the coefficient of the interaction term between Female
and Post is very small and never significant suggesting that there are no changes in
differential recall of deaths by gender.
We also examine whether underreporting of deaths is related to sex selection. To do so,
we extended the model to include also a proxy for sex‐selection (MFR) interacted with a
post dummy (indicating the later survey round) and female dummy and the 3‐way
interactions. To proxy for sex selection, we used the state‐level MFR at age 0 from the
2001 Census because we think it is useful to have an MFR measure from a different
source than the NFHS and that the year 2001 reflects more clearly where prenatal sex
selection takes place.
As can be seen in the results reported in column 4 of the same table (Appendix Table A.8),
there is a positive coefficient on MFR*female, suggesting that states that have prenatal
sex selection have also relatively higher female mortality rates. However, there is no
evidence of any association between recall or differential recall by gender and prenatal
sex selection (i.e. the coefficients of the interactions MFR*post and MFR*female*post are
not statistically significant from zero, either when looking at each separately or when
looking at them jointly).
To summarize, we find no evidence suggesting that there were differential changes in
underreporting (or recalls) of female deaths in the NFHS or that underreporting was
related to sex selection.

9

A

Appendix

Figure A.1: F-Values for Structural Break Test in Differential Probability of a Male Birth by Year of

0

0

10

10

20

20

30

30

Birth

1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006
Year
High income households in son preference states
Other households

(a) Parity 2

1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006
Year
High income households in son preference states
Other households

(b) Parity 3

Notes: The figure plots F-values for structural break tests in the differential probability of a male birth by year of birth. The full
line plots F-values for households in the 4th or 5th wealth quintile who live in states with son preferences above median (according
to first round of NFHS). The dotted line plots F-values for the sample based on all other households. See data appendix for the
definition of son preference measure. See Figure 2 for details on the probability of male series.

Table A1. State Characteristics
Son preference
Wants more
1990 NDP
Wealth HH with
Ideal Ideal
sons than
PC (Rs.) Urban Index electricity MFR I MFR II
daughters
(1)

North
Delhi
Haryana
Himachal Pradesh
Jammu & Kashmir
Punjab
Rajasthan
West
Gujarat
Maharashtra
Northeast
Arunachal Pradesh
Assam
Manipur
Meghalaya
Mizoram
Nagaland
Tripura
Sikkim
Central
Madhya Pradesh
Uttar Pradesh
East
Bihar
Orissa
West Bengal
South
Andhra Pradesh
Goa
Karnataka
Kerala
Tamil Nadu

TV
exposure

Mother's
Illiterate

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

11,057
7,508
4,910
3,816
8,318
4,191

0.92
0.26
0.10
0.18
0.28
0.20

4.79
3.90
3.62
3.74
4.26
2.79

0.96
0.86
0.92
0.88
0.94
0.54

1.31
1.45
1.33
1.50
1.48
1.62

1.28
1.43
1.31
1.49
1.47
1.62

0.302
0.451
0.367
0.492
0.480
0.577

0.83
0.49
0.47
0.50
0.57
0.18

0.37
0.64
0.50
0.57
0.53
0.82

5,891
7,439

0.35
0.42

3.60
3.54

0.78
0.76

1.46
1.31

1.36
1.28

0.424
0.359

0.39
0.47

0.55
0.50

5,398
4,281
3,976
4,375
4,474
4,990
3,370
5,302

0.15
0.12
0.32
0.19
0.49
0.21
0.20
0.14

3.17
2.44
3.55
3.10
3.82
3.64
2.96
3.73

0.62
0.20
0.64
0.43
0.76
0.78
0.47
0.80

1.50
1.41
1.37
1.01
1.18
1.14
1.31
1.16

1.46
1.41
1.37
1.00
1.18
1.14
1.31
1.15

0.430
0.436
0.435
0.143
0.331
0.278
0.327
0.218

0.29
0.18
0.38
0.24
0.25
0.23
0.34
0.56

0.70
0.59
0.48
0.51
0.08
0.43
0.42
0.49

4,049
3,590

0.22
0.20

2.85
2.62

0.65
0.34

1.55
1.68

1.52
1.66

0.516
0.565

0.27
0.19

0.74
0.76

2,660
3,077
4,673

0.15
0.15
0.27

2.32
2.42
2.67

0.17
0.29
0.30

1.65
1.45
1.28

1.62
1.43
1.28

0.556
0.451
0.319

0.13
0.16
0.33

0.78
0.67
0.51

4,531
8,797
4,598
4,200
4,983

0.26
0.50
0.33
0.28
0.35

3.20
4.32
3.27
3.89
3.42

0.65
0.92
0.66
0.61
0.66

1.33
1.24
1.25
1.15
1.07

1.29
1.21
1.24
1.14
1.07

0.329
0.278
0.271
0.183
0.115

0.39
0.71
0.40
0.42
0.50

0.69
0.34
0.61
0.16
0.50

Notes: The table reports selected economic and demographic characteristics by state and measures of son preference. Data on net
domestic product per capita (NDP PC) in column 1 are reported in Rupees and are based on reports of the Reserve Bank of India. Data
for columns 2‐9 are based on state averages from the first round of the NFHS. Tabulations for Sikkim are based on the second round of
the NFHS as Sikkim was not sampled in the first round. See data appendix for definition of son preference measures.

Table A2. Are Missing Anthropometric Values Correlated with Gender or MFR?
Basic specification
Full controls
Sample
Outcome
Females
Female
Female
size
mean
‐Males
x MFR
x MFR
MFR
MFR
(1)
(2)
(3)
(4)
(5)
(6)
(7)
Missing weight

91,374

0.160

‐0.002
(0.002)

0.620
(0.368)

‐0.048
(0.091)

0.605
(0.352)

‐0.042
(0.093)

Missing weight for height

91,374

0.246

‐0.001
(0.003)

0.828
(0.901)

‐0.125
(0.082)

0.877
(0.916)

‐0.124
(0.091)

Missing height

91,374

0.248

0.000
(0.003)

0.838
(0.899)

‐0.118
(0.084)

0.888
(0.913)

‐0.118
(0.093)

Notes: The dependent variables are indicators for missing values in weight (row 1), weight for height (row 2), or height (row 3).
Means of the dependent variables are reported in Column 2. Gender differences in outcomes are reported in column 3. Columns 4
and 5 report regression estimates for MFR and MFR interacted with a female dummy from a linear probability model that controls
for state and year of birth fixed effects and their interactions with gender. Columns 6 and 7 report regression estimates from a
model that controls also for twin status, mother's age, mother's education, mother's age at first birth, indicators for mother's
religion, father's education, mother's mass media exposure, wealth, and rural/urban status. The sample includes children aged 0 to
35 months born in the last two births of ever married women sampled in rounds 1‐3 of the NFHS surveys. Observations are
weighted using national‐level weights. Standard errors clustered at the state level are reported in parenthesis.

Table A3. Effects on Nutritional Status of Children
Using Alternative Anthropometric Indices

Basic specification
Sample
size
(1)

Outcome
mean
(2)

Weight for age
z‐score

76314

Weight for height
z‐score

Full controls

MFR
(3)

Female
x MFR
(4)

MFR
(5)

Female
x MFR
(6)

‐1.873

0.727
(0.864)

1.038
(0.453)

0.794
(0.847)

1.060
(0.358)

69784

‐0.980

1.506
(1.225)

0.926
(0.394)

1.507
(1.235)

0.963
(0.408)

Height for age
z‐score

69571

‐1.729

‐1.126
(0.867)

0.430
(0.416)

‐0.911
(0.739)

0.683
(0.312)

Weight for height
percent of reference median

76314

79.944

9.042
(8.352)

11.665
(5.387)

8.819
(8.148)

11.498
(3.790)

Weight for height
percent of reference median

69784

91.212

13.706
(11.570)

9.893
(4.130)

14.269
(11.497)

9.711
(4.216)

Height for age
percent of reference median

69571

93.390

‐4.195
(3.333)

1.451
(1.537)

‐3.095
(2.782)

2.312
(1.181)

Underweight
new‐zscore

77,195

0.440

‐0.054
(0.273)

‐0.519
(0.154)

‐0.019
(0.258)

‐0.554
(0.139)

Wasted
new‐zscore

69,883

0.221

‐0.040
(0.275)

‐0.246
(0.091)

0.060
(0.250)

‐0.211
(0.093)

Stunted
new‐zscore

70,221

0.499

0.378
(0.201)

‐0.162
(0.132)

0.260
(0.203)

‐0.216
(0.119)

Outcome

Notes: Means of the dependent variables are reported in Column 2. Columns 3 and 4 report regression estimates for
MFR and MFR interacted with a female dummy from a linear probability model that controls for state and year of birth
fixed effects and their interactions with gender. Columns 5 and 6 report regression estimates from a model that
controls also for twin status, mother's age, mother's education, mother's age at first birth, indicators for mother's
religion, father's education, mother's mass media exposure, wealth, and rural/urban status. The sample includes
children aged 0 to 35 months born in the last two births of ever married women sampled in rounds 1‐3 of the NFHS
surveys. Sample sizes are smaller for wasted and stunted because height measures were not taken in round 1 for the
following 5 states: Andhra Pradesh, Himachal Pradesh, Madhya Pradesh, Tamil Nadu and West Bengal. Observations
are weighted using national‐level weights. Standard errors clustered at the state level are reported in parenthesis.

Table A4. Robustness Checks
Underweight
Wasted
Female
Female
x MFR
x MFR
MFR
MFR
(1)
(2)
(3)
(4)
1 Main Results

Stunted
Female
x MFR
MFR
(5)
(6)

‐0.095
(0.318)

‐0.536
(0.132)

‐0.029
(0.275)

‐0.250
(0.104)

0.386
(0.236)

‐0.293
(0.164)

---

‐0.560
(0.134)

---

‐0.267
(0.101)

---

‐0.304
(0.165)

3 Household covariates interacted with
gender

‐0.107
(0.319)

‐0.514
(0.126)

‐0.036
(0.278)

‐0.236
(0.105)

0.379
(0.239)

‐0.276
(0.164)

4 State varying controls 1 + HH
covariates interacted with gender

‐0.091
(0.254)

‐0.433
(0.115)

0.140
(0.181)

‐0.135
(0.114)

0.415
(0.175)

‐0.217
(0.112)

5 State varying controls 2 + HH
covariates interacted with gender

‐0.303
(0.289)

‐0.384
(0.092)

‐0.009
(0.240)

‐0.227
(0.110)

0.168
(0.170)

‐0.206
(0.127)

6 Birth weight
interacted with gender

‐0.055
(0.327)

‐0.549
(0.149)

‐0.028
(0.270)

‐0.250
(0.124)

0.415
(0.269)

‐0.303
(0.175)

7 Size at birth
interacted with gender

‐0.085
(0.325)

‐0.525
(0.129)

‐0.031
(0.282)

‐0.235
(0.100)

0.383
(0.237)

‐0.276
(0.150)

8 Delivery at health
facility interacted with gender

‐0.092
(0.318)

‐0.518
(0.132)

‐0.029
(0.274)

‐0.243
(0.105)

0.395
(0.235)

‐0.288
(0.166)

9 Log MFR

‐0.099
(0.344)

‐0.609
(0.147)

‐0.029
(0.302)

‐0.295
(0.105)

0.463
(0.254)

‐0.330
(0.188)

10 5‐year span for MFR

‐0.103
(0.269)

‐0.426
(0.131)

‐0.015
(0.199)

‐0.184
(0.077)

0.175
(0.205)

‐0.222
(0.161)

11 9‐year span for MFR

0.130
(0.386)

‐0.591
(0.159)

0.080
(0.295)

‐0.316
(0.138)

0.540
(0.308)

‐0.236
(0.193)

12 Unweighted regression

0.011
(0.235)

‐0.474
(0.112)

0.151
(0.203)

‐0.290
(0.103)

0.186
(0.137)

‐0.245
(0.101)

13 State weights

0.011
(0.240)

‐0.471
(0.111)

0.136
(0.204)

‐0.286
(0.095)

0.201
(0.142)

‐0.267
(0.096)

2 State x Year of birth FE

Notes: The table reports estimates from various robustness checks. See Appendix B of the paper for a detailed explanation of each test.

Table A5. Difference in Difference Models for Nutritional Outcomes

Sample size
Boys
Girls

Outcome mean
Boys
Girls

MFR from NFHS
Boys
Girls

MFR from Census and
inter‐census
interpolation
Boys
Girls

MFR from Census adjusted
for infant mortality and
inter‐census interpolation
Boys
Girls

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

Underweight

39475

36839

0.477

0.493

‐0.107
(0.319)

‐0.621
(0.301)

‐0.035
(0.250)

‐0.448
(0.224)

‐0.033
(0.254)

‐0.442
(0.223)

Wasted

36178

33606

0.186

0.171

‐0.036
(0.278)

‐0.272
(0.257)

‐0.125
(0.223)

‐0.418
(0.212)

‐0.120
(0.221)

‐0.404
(0.214)

Stunted

36081

33490

0.426

0.440

0.379
(0.239)

0.103
(0.231)

0.380
(0.291)

0.109
(0.222)

0.345
(0.292)

0.089
(0.221)

Weight for age
z‐score

39475

36839

‐1.864

‐1.882

0.836
(0.846)

1.819
(0.825)

0.487
(0.705)

1.131
(0.701)

0.472
(0.712)

1.080
(0.701)

Weight for height
z‐score

36178

33606

‐0.991

‐0.969

1.530
(1.240)

2.456
(1.310)

1.341
(1.096)

2.137
(1.184)

1.243
(1.088)

2.013
(1.182)

Height for age
z‐score

36081

33490

‐1.722

‐1.735

‐0.873
(0.736)

‐0.268
(0.695)

‐0.851
(0.865)

‐0.555
(0.637)

‐0.778
(0.870)

‐0.501
(0.637)

Weight for age
percent of reference median

39475

36839

80.269

79.596

9.304
(8.153)

19.904
(8.418)

5.269
(7.197)

11.613
(7.909)

5.077
(7.257)

10.981
(7.933)

Weight for height
percent of reference median

36178

33606

91.464

90.942

14.490
(11.529)

23.853
(12.635)

12.679
(10.171)

20.311
(11.469)

11.717
(10.096)

19.057
(11.462)

Height for age
percent of reference median

36081

33490

93.476

93.297

‐2.944
(2.768)

‐0.945
(2.684)

‐2.985
(3.212)

‐2.086
(2.461)

‐2.705
(3.231)

‐1.879
(2.461)

Notes: Sample sizes are reported in columns 1 and 2. Means of the dependent variables are reported in Column 2 and 3. Columns 5‐10 report differences‐in‐difference
estimates from models that control for state, survey round, year of birth fixed effects. In addition, the models include the following covariates: twin status, mother's age,
mother's education, mother's age at first birth, indicators for mother's religion, father's education, mother's mass media exposure, wealth, and rural/urban status. Estimates
in columns 5 and 6 come from models that use MFR based on the NFHS data as the main explanatory variable. Estimates in columns 7 and 8 come from models that use MFR
constructed from census data and inter‐census interpolation and estimates in columns 9 and 10 come from models that use the same data but adjust census numbers for
infant mortality. See data Appendix for details about the construction of the MFR series. The sample includes children aged 0 to 35 months born in the last two births of ever
married women sampled in rounds 1‐3 of the NFHS surveys. Observations are weighted using national‐level weights. Standard errors clustered at the state level are reported
in parenthesis.

Table A6. Family Size by Probability of Ultrasound Use
A. Full Sample
Prob. of ultrasound use

4.560
(22.483)

Year of birth

‐0.015
(0.007)

Prob. of ultrasound use x Year of birth

‐0.004
(0.011)

Female x prob. of ultrasound

22.707
(12.019)

Female x Year of birth

0.003
(0.003)

Female x prob. of ultrasound x year of birth

‐0.011
(0.006)

Number of observations

97347

B. Low prob. of ultrasound sample (25 percentile ‐ Prob.< 0.078)
Year of Birth

‐0.003
(0.008)

Female x Year of birth

0.003
(0.006)

Number of observations

24,630

Notes: The table reports changes in family size within states and over time
by the probability of ultrasound use. Panel A uses the full sample and Panel
B includes only households with a predicted probability of ultrasound use
below the 25 percentile (prob.<0.078). Estimates come from models that
control for state fixed effects interacted with gender. Observations are
weighted using national‐level weights. Bootstrapped standard errors
clustered at the state level are reported in parenthesis.

Table A7. Son Preference and MFR
MFR of cohorts born between 5‐10 years
prior to survey date
Ever married
Ever married
age 15‐20
age 15‐20
with no children
(1)
(2)
Ideal MFR I

‐0.105
(0.225)
1.329

0.070
(0.234)
1.300

Ideal MFR II

0.034
(0.197)
1.307

0.135
(0.211)
1.285

Wants more sons
than daughters

‐0.111
(0.259)
0.332

0.209
(0.232)
0.310

Number of women

26,987

12,452

Notes: The table reports regression estimates of son preference on the state
average MFR of cohorts born between 5 and 10 years prior to survey date. Rows one
and two report regression estimates where the dependent variable is defined based
on the ratio of the ideal number of sons to the ideal number of daughters reported
by mothers (ideal MFR). The third row reports regression estimates where the
dependent variable is an indicator that equals one if the mother reported a larger
ideal number of sons relative to her ideal number of daughters. See Data Appendix
for more details about the definition of son preference measures. All estimates
come from models that control for state fixed‐effects and indicators for mother's
year of birth, mother's age, mother's education, mother's age at first birth, mother's
religion, father's education, mother's mass media exposure, wealth, and rural/urban
status. The sample in column 1 includes all ever married women of ages 15‐20
sampled in rounds 1‐3 of the NFHS surveys. The sample in column 2 includes all ever
married women of ages 15‐20 with no children sampled in rounds 1‐3 of the NFHS
surveys. Means of the dependent variables are reported in italics. Observations are
weighted using national‐level weights. Standard errors clustered at the state level
are reported in parenthesis.

Post

Table A8. Assessing Differential Recall of Female Deaths by Survey Round and Associations with MFR
Pooling all rounds +
Round 1 vs. Round 2 Round 2 vs. Round 3
Pooling all rounds
Interaction with MFR
(1)
(2)
(3)
(4)
0.005
0.019
0.011
‐0.002
(0.006)
(0.005)
(0.005)
(0.064)

Female

‐0.008
(0.005)

‐0.003
(0.004)

‐0.006
(0.003)

‐0.100
(0.051)

Female x Post

0.000
(0.007)

‐0.003
(0.008)

‐0.001
(0.006)

‐0.015
(0.073)

MFR x Female

‐‐

‐‐

‐‐

0.085
(0.045)

MFR x Post

‐‐

‐‐

‐‐

0.011
(0.057)

MFR x Female x Post

‐‐

‐‐

‐‐

0.013
(0.063)

0.076

0.070

0.073

0.073

‐‐

‐‐

0.12

56,285

124,419

124,419

Outcome mean

F‐Stat
‐‐
(MFR x Post = MFR x Female x Post = 0)
Number of observations

68,134

Notes: The table reports estimates from a regression that examines whether the death rate of the same cohort changed over time
differentially by gender. The sample in column (1) includes children born between 1989‐1991 who appear in rounds 1 or 2 of the
survey. The sample in column (2) includes children born between 1996‐1998 who appear in rounds 2 or 3 of the survey. The sample
in columns (3) and (4) includes the union of columns (1) and (2). In addition to the listed variables, all models control for state and
year of birth fixed effects. Observations are weighted using national‐level weights. Standard errors clustered at the state level are
reported in parenthesis.

Working Paper Series
A series of research studies on regional economic issues relating to the Seventh Federal
Reserve District, and on financial and economic topics.
Risk Taking and the Quality of Informal Insurance: Gambling and Remittances in Thailand
Douglas L. Miller and Anna L. Paulson

WP-07-01

Fast Micro and Slow Macro: Can Aggregation Explain the Persistence of Inflation?
Filippo Altissimo, Benoît Mojon, and Paolo Zaffaroni

WP-07-02

Assessing a Decade of Interstate Bank Branching
Christian Johnson and Tara Rice

WP-07-03

Debit Card and Cash Usage: A Cross-Country Analysis
Gene Amromin and Sujit Chakravorti

WP-07-04

The Age of Reason: Financial Decisions Over the Lifecycle
Sumit Agarwal, John C. Driscoll, Xavier Gabaix, and David Laibson

WP-07-05

Information Acquisition in Financial Markets: a Correction
Gadi Barlevy and Pietro Veronesi

WP-07-06

Monetary Policy, Output Composition and the Great Moderation
Benoît Mojon

WP-07-07

Estate Taxation, Entrepreneurship, and Wealth
Marco Cagetti and Mariacristina De Nardi

WP-07-08

Conflict of Interest and Certification in the U.S. IPO Market
Luca Benzoni and Carola Schenone

WP-07-09

The Reaction of Consumer Spending and Debt to Tax Rebates –
Evidence from Consumer Credit Data
Sumit Agarwal, Chunlin Liu, and Nicholas S. Souleles

WP-07-10

Portfolio Choice over the Life-Cycle when the Stock and Labor Markets are Cointegrated
Luca Benzoni, Pierre Collin-Dufresne, and Robert S. Goldstein

WP-07-11

Nonparametric Analysis of Intergenerational Income Mobility
with Application to the United States
Debopam Bhattacharya and Bhashkar Mazumder

WP-07-12

How the Credit Channel Works: Differentiating the Bank Lending Channel
and the Balance Sheet Channel
Lamont K. Black and Richard J. Rosen

WP-07-13

Labor Market Transitions and Self-Employment
Ellen R. Rissman

WP-07-14

First-Time Home Buyers and Residential Investment Volatility
Jonas D.M. Fisher and Martin Gervais

WP-07-15

1

Working Paper Series (continued)
Establishments Dynamics and Matching Frictions in Classical Competitive Equilibrium
Marcelo Veracierto

WP-07-16

Technology’s Edge: The Educational Benefits of Computer-Aided Instruction
Lisa Barrow, Lisa Markman, and Cecilia Elena Rouse

WP-07-17

The Widow’s Offering: Inheritance, Family Structure, and the Charitable Gifts of Women
Leslie McGranahan

WP-07-18

Incomplete Information and the Timing to Adjust Labor: Evidence from the
Lead-Lag Relationship between Temporary Help Employment and Permanent Employment
Sainan Jin, Yukako Ono, and Qinghua Zhang

WP-07-19

A Conversation with 590 Nascent Entrepreneurs
Jeffrey R. Campbell and Mariacristina De Nardi

WP-07-20

Cyclical Dumping and US Antidumping Protection: 1980-2001
Meredith A. Crowley

WP-07-21

Health Capital and the Prenatal Environment:
The Effect of Maternal Fasting During Pregnancy
Douglas Almond and Bhashkar Mazumder

WP-07-22

The Spending and Debt Response to Minimum Wage Hikes
Daniel Aaronson, Sumit Agarwal, and Eric French

WP-07-23

The Impact of Mexican Immigrants on U.S. Wage Structure
Maude Toussaint-Comeau

WP-07-24

A Leverage-based Model of Speculative Bubbles
Gadi Barlevy

WP-08-01

Displacement, Asymmetric Information and Heterogeneous Human Capital
Luojia Hu and Christopher Taber

WP-08-02

BankCaR (Bank Capital-at-Risk): A credit risk model for US commercial bank charge-offs
Jon Frye and Eduard Pelz

WP-08-03

Bank Lending, Financing Constraints and SME Investment
Santiago Carbó-Valverde, Francisco Rodríguez-Fernández, and Gregory F. Udell

WP-08-04

Global Inflation
Matteo Ciccarelli and Benoît Mojon

WP-08-05

Scale and the Origins of Structural Change
Francisco J. Buera and Joseph P. Kaboski

WP-08-06

Inventories, Lumpy Trade, and Large Devaluations
George Alessandria, Joseph P. Kaboski, and Virgiliu Midrigan

WP-08-07

2

Working Paper Series (continued)
School Vouchers and Student Achievement: Recent Evidence, Remaining Questions
Cecilia Elena Rouse and Lisa Barrow
Does It Pay to Read Your Junk Mail? Evidence of the Effect of Advertising on
Home Equity Credit Choices
Sumit Agarwal and Brent W. Ambrose

WP-08-08

WP-08-09

The Choice between Arm’s-Length and Relationship Debt: Evidence from eLoans
Sumit Agarwal and Robert Hauswald

WP-08-10

Consumer Choice and Merchant Acceptance of Payment Media
Wilko Bolt and Sujit Chakravorti

WP-08-11

Investment Shocks and Business Cycles
Alejandro Justiniano, Giorgio E. Primiceri, and Andrea Tambalotti

WP-08-12

New Vehicle Characteristics and the Cost of the
Corporate Average Fuel Economy Standard
Thomas Klier and Joshua Linn

WP-08-13

Realized Volatility
Torben G. Andersen and Luca Benzoni

WP-08-14

Revenue Bubbles and Structural Deficits: What’s a state to do?
Richard Mattoon and Leslie McGranahan

WP-08-15

The role of lenders in the home price boom
Richard J. Rosen

WP-08-16

Bank Crises and Investor Confidence
Una Okonkwo Osili and Anna Paulson

WP-08-17

Life Expectancy and Old Age Savings
Mariacristina De Nardi, Eric French, and John Bailey Jones

WP-08-18

Remittance Behavior among New U.S. Immigrants
Katherine Meckel

WP-08-19

Birth Cohort and the Black-White Achievement Gap:
The Roles of Access and Health Soon After Birth
Kenneth Y. Chay, Jonathan Guryan, and Bhashkar Mazumder

WP-08-20

Public Investment and Budget Rules for State vs. Local Governments
Marco Bassetto

WP-08-21

Why Has Home Ownership Fallen Among the Young?
Jonas D.M. Fisher and Martin Gervais

WP-09-01

Why do the Elderly Save? The Role of Medical Expenses
Mariacristina De Nardi, Eric French, and John Bailey Jones

WP-09-02

3

Working Paper Series (continued)
Using Stock Returns to Identify Government Spending Shocks
Jonas D.M. Fisher and Ryan Peters

WP-09-03

Stochastic Volatility
Torben G. Andersen and Luca Benzoni

WP-09-04

The Effect of Disability Insurance Receipt on Labor Supply
Eric French and Jae Song

WP-09-05

CEO Overconfidence and Dividend Policy
Sanjay Deshmukh, Anand M. Goel, and Keith M. Howe

WP-09-06

Do Financial Counseling Mandates Improve Mortgage Choice and Performance?
Evidence from a Legislative Experiment
Sumit Agarwal,Gene Amromin, Itzhak Ben-David, Souphala Chomsisengphet,
and Douglas D. Evanoff

WP-09-07

Perverse Incentives at the Banks? Evidence from a Natural Experiment
Sumit Agarwal and Faye H. Wang

WP-09-08

Pay for Percentile
Gadi Barlevy and Derek Neal

WP-09-09

The Life and Times of Nicolas Dutot
François R. Velde

WP-09-10

Regulating Two-Sided Markets: An Empirical Investigation
Santiago Carbó Valverde, Sujit Chakravorti, and Francisco Rodriguez Fernandez

WP-09-11

The Case of the Undying Debt
François R. Velde

WP-09-12

Paying for Performance: The Education Impacts of a Community College Scholarship
Program for Low-income Adults
Lisa Barrow, Lashawn Richburg-Hayes, Cecilia Elena Rouse, and Thomas Brock
Establishments Dynamics, Vacancies and Unemployment: A Neoclassical Synthesis
Marcelo Veracierto

WP-09-13

WP-09-14

The Price of Gasoline and the Demand for Fuel Economy:
Evidence from Monthly New Vehicles Sales Data
Thomas Klier and Joshua Linn

WP-09-15

Estimation of a Transformation Model with Truncation,
Interval Observation and Time-Varying Covariates
Bo E. Honoré and Luojia Hu

WP-09-16

Self-Enforcing Trade Agreements: Evidence from Antidumping Policy
Chad P. Bown and Meredith A. Crowley

WP-09-17

Too much right can make a wrong: Setting the stage for the financial crisis
Richard J. Rosen

WP-09-18

4

Working Paper Series (continued)
Can Structural Small Open Economy Models Account
for the Influence of Foreign Disturbances?
Alejandro Justiniano and Bruce Preston

WP-09-19

Liquidity Constraints of the Middle Class
Jeffrey R. Campbell and Zvi Hercowitz

WP-09-20

Monetary Policy and Uncertainty in an Empirical Small Open Economy Model
Alejandro Justiniano and Bruce Preston

WP-09-21

Firm boundaries and buyer-supplier match in market transaction:
IT system procurement of U.S. credit unions
Yukako Ono and Junichi Suzuki
Health and the Savings of Insured Versus Uninsured, Working-Age Households in the U.S.
Maude Toussaint-Comeau and Jonathan Hartley

WP-09-22

WP-09-23

The Economics of “Radiator Springs:” Industry Dynamics, Sunk Costs, and
Spatial Demand Shifts
Jeffrey R. Campbell and Thomas N. Hubbard

WP-09-24

On the Relationship between Mobility, Population Growth, and
Capital Spending in the United States
Marco Bassetto and Leslie McGranahan

WP-09-25

The Impact of Rosenwald Schools on Black Achievement
Daniel Aaronson and Bhashkar Mazumder

WP-09-26

Comment on “Letting Different Views about Business Cycles Compete”
Jonas D.M. Fisher

WP-10-01

Macroeconomic Implications of Agglomeration
Morris A. Davis, Jonas D.M. Fisher and Toni M. Whited

WP-10-02

Accounting for non-annuitization
Svetlana Pashchenko

WP-10-03

Robustness and Macroeconomic Policy
Gadi Barlevy

WP-10-04

Benefits of Relationship Banking: Evidence from Consumer Credit Markets
Sumit Agarwal, Souphala Chomsisengphet, Chunlin Liu, and Nicholas S. Souleles

WP-10-05

The Effect of Sales Tax Holidays on Household Consumption Patterns
Nathan Marwell and Leslie McGranahan

WP-10-06

Gathering Insights on the Forest from the Trees: A New Metric for Financial Conditions
Scott Brave and R. Andrew Butters

WP-10-07

Identification of Models of the Labor Market
Eric French and Christopher Taber

WP-10-08

5

Working Paper Series (continued)
Public Pensions and Labor Supply Over the Life Cycle
Eric French and John Jones

WP-10-09

Explaining Asset Pricing Puzzles Associated with the 1987 Market Crash
Luca Benzoni, Pierre Collin-Dufresne, and Robert S. Goldstein

WP-10-10

Prenatal Sex Selection and Girls’ Well‐Being: Evidence from India
Luojia Hu and Analía Schlosser

WP-10-11

6