Son Preference, Sex Selection and Gender Inequality in

India
Deepankar Basu

February 15, 2008

Abstract
In the backdrop of the debate on missing women (Oster, 2005), this paper empirically
tests for two competing explanations of the increasing sex ratio at birth (SRB) in India:
prevalence of hepatitis B and human intervention in the form of sex selective abortion
or female infanticide. Estimating a simple model of male-preferring stopping rule using
the method of maximum likelihood with data from three rounds of the NFHS in India
(1992, 1998 and 2005), I nd that the probability of male birth varies signicantly
across birth parities. This provides evidence of human intervention. Using a dummy
on scheduled tribes as a novel proxy for the prevalence of hepatitis B in India, I re-
estimate the model by endogenizing the probability of male birth. I nd that hepatitis
B has no impact on the probability of male birth for 1992 and a negative impact
for 1998; this rules out hepatitis B as a possible explanation for the increasing SRB.
I conclude that human intervention in the form of sex-selective abortion or female
infanticide, rather than biological factors like hepatitis B, explains the increasing SRB
in India.
JEL Codes: J1, J7.
Keywords: sex ratio at birth, hepatitis B, sex-selective abortion.
1 Introduction
In the early 1990s, Amartya Sen directed attention of the world to what he called the problem
of missing women (Sen, 1990; 1992). This problem refers to abnormally high population

Department of Economics, Ohio State University, 317 Arps Hall, Columbus, OH 43210, email:
basu.15@osu.edu. I would like to thank Stephen Cosslett, Robert de Jong, Trevon Logan, Alita Nandi
and participants in the Micro Lunch at the Department of Economics, Ohio State University, the 2007
seminar series in the Department of Economics at the Cleveland State University and the seminar in the
Department of Economics, University of Massachusetts, Amherst for helpful comments.
1
sex ratios (dened as the number of males per 100 females) in certain parts of the world,
especially East, South and South-East Asia and North Africa. These skewed population sex
ratios point to a large decit of women in the adult population. By Sens estimates, close to
100 million women were missing. The import of this nding is astounding. If population
sex ratios in East, South and South-East Asia and North Africa were like what is seen in
Sub-Saharan Africa or Western Europe, then the population would have about 100 million
more women alive today! Sen squarely put the blame for this excess mortality of women on
various kinds of discrimination, both overt and covert, that women face in these societies and
called for social, political and cultural measures to ght against these deep-rooted gender
biases. Though later attempts to rene Sens calculations arrived at lower estimates (see,
for instance, Coale, 1991), the enormity of the problem is hardly diminished because the
consensus gure lies anywhere between 60 and 100 million missing women.
A recent paper (Oster, 2005) has tried to challenge the thrust of Sens argument (see,
for instance, Dubner and Levitt, 2005) by suggesting that part of the missing women
problem can be explained by biological factors. Drawing on the epidemiological literature on
hepatitis B, Oster (2005) has argued that the prevalence of hepatitis B in a population can
partly explain the skewed population sex ratios. This, according to her, is because hepatitis
B infection among women seems to increase the sex ratio at birth (SRB), i.e., carriers of
the hepatitis B virus give birth to more male ospring than non-carriers. In her calibration
exercise the prevalence of hepatitis B can account for, on the higher end, as much as 70
percent of the missing women in China; on the lower end, she claims to be able to explain
about 20 percent of the problem for India. Several countries, in Osters (2005) analysis, lie
between these two extremes.
Osters (2005) claim, it is obvious, rests crucially on the (causal) link between hepatitis
B infection rates in a population and the SRB. Several facts suggest caution in using or
interpreting this link if at all it exists. First and foremost, there is no currently available
biological (or other scientic) explanation for this supposed link, which even she admits (page
1180-81, Oster, 2005). So interpreting the statistical link that she nds, in both individual
level and cross-country data, as a causal link is at best atheoretical and hence hazardous.
Second, a recent study (Lin and Luoh, 2006) attempting to directly examine the link
between hepatitis B infection of mothers and the sex ratio of their ospring came to a
decisive negative conclusion. Using a large micro-level dataset from the National Hepatitis
B Immunization Program in Taiwan the authors nd that the marginal probability of a
positively infected (i.e., an HBsAg+) mother having a male birth is only 0.0025. Hence the
study concludes that the eect of HBV mothers on the sex ratio among their ospring is
minimal, and hence can account for only a small portion of the case of missing women
(page 2, Lin and Luoh, 2006; emphasis added). Responding to this direct negative evidence,
Blumberg and Oster (2007) shifts the focus from mothers to fathers. Oster (2005), it must
be recalled, had explicitly stated that the link from the hepatitis B virus to the ospring sex
ratio runs through the mother. There is evidence that women who are carriers of hepatitis B
give birth to a higher ratio of boys to girls than non-carriers. Since many of the countries with
missing women also have a relatively high prevalence of hepatitis B carriers, the naturally
2
occurring higher sex ratio at birth could produce a higher population sex ratio even in the
absence of excess female mortality (p. 1164, Oster, 2005; emphasis added). Faced with
direct evidence to the contrary, it seems Blumberg and Oster (2007) have shifted the burden
of the causal explanation on paternal rather than maternal infection. They claim to present
evidence for this link by taking a new look at the old data.
There are two obvious problems with this new argument - the argument about the pa-
ternal infection route for the eect - presented in Blumberg and Oster (2007). First, this
claim remains as atheoretical as the original claim; there is no biological explanation for the
supposed eect of hepatitis B on the sex ratio at birth, and that holds as much for maternal
as for paternal infection. Second, it is well known that the hepatitis B virus can get trans-
mitted through sexual intercourse (page 1177, Oster, 2005). Hence, it would matter little
whether the mother or the father was the original infectant; in the course of time, both would
be carries of the virus and it is not clear why the distinction is at all important between
paternal and maternal infection.
Third, Sub-Saharan Africa is a region that is well-known for high hepatitis B infection
rates, compared to world averages, but which has no female decit in striking contrast to
Asia. Fourth, several studies have documented considerable variation in the SRB when
measured at dierent birth parities (Lin and Luoh, 2007), or when measured for dierent
sex-composition of existing children (Ebenstein, 2007), or when measured for dierent ge-
ographical locations (Hesketh and Wei Xing, 2006), or when measured across time (Das
Gupta, 2005).
Though variation of the SRB across birth parities or geographical location or time does
not per se rule out the role of hepatitis B in the determination of the SRB, it certainly
points to other important factors at work, especially sex selective abortion. Responding
to Das Guptas (2005) critique, Oster (2006) correctly distinguishes, in my opinion, the
marginal from the average. While variation in the SRB indicates marginal eects of human
intervention, Osters (2005) argument was about the average eect of hepatitis B on the
SRB. Hence, evidence of variation in the SRB does not constitute a valid critique of Osters
(2005) position.
The evidence on variation in the SRB is of course well known and convincing in several
cases. For instance, drawing on a strand of the demography literature, Ebenstein (2007)
provides a forceful critique of the hepatitis B explanation for missing women in China by
marshalling evidence of variation in the SRB. Using Census data, he computes the SRB
for dierent birth parities and dierent combinations of already-existing children
1
. He nds
that whereas the SRB for the rst birth is close to what is observed in Western Europe
or Sub-Saharan Africa, the SRB increases steeply for later births. Looking at the male
fraction of next birth at dierent birth parities throws up even more striking results. At
the second birth parity, he nds that the male fraction of next birth for families with one
daughter is higher than for families with one son; at the third birth parity, he nds that
1
Demographers interested in studying the eects of son preference have typically looked at variations in
the SRB over birth parities, over the sex composition of existing children, over geographical location, etc.
For a recent study on similar issues see Hesketh and Wei Xing, 2006 and the references therein.
3
the male fraction of next birth is highest for families with two daughters, which is lower
for families with a son and a daughter and even lower for families with two sons. Since the
SRB for the rst birth is statistically indistinguishable from the normal gure for human
populations, this is conclusive evidence against the claim that hepatitis B aects the SRB,
at least in China; and, since there is considerable variation across parities and conditional on
the sex composition of existing children, this shows that human intervention is important.
Ebenstein (2007), therefore, concludes that it is human intervention in the form of sex-
selective abortion, rather than biological factors like the prevalence of hepatitis B, that seem
to be more important in China. Similar nding for India has been reported by Jha, et al.
(2006).
In the backdrop of this debate on missing women, this paper tests for the validity of
the two alternative explanations for the increasing SRB in India over the past two decades:
biology (in the form of hepatitis B) and human intervention (in the form of sex-selective
abortion and/or female infanticide). My testing procedure rests on distinguishing the eects
of biology from human intervention on the SRB. I rely on the simple observation that sex-
selective abortion will cause the probability of male birth to vary across birth parity (or
birth order), whereas hepatitis B, by itself, cannot do so. Since sex-selective abortion is a
reection of son preference and other deep gender biases in society, it can be expected to
vary the probability of male birth across birth parity, as couples determine the sex of foetuses
and induce abortion of female foetuses based on the sex-composition of existing children.
This kind of intervention would be socially and culturally determined and would vary across
birth order in a way that biological factors like hepatitis B cannot induce.
For instance, if one were to plot the SRB (or the probability of male birth
2
) against the
birth parity (with the probability of male birth on the y-axis and the birth parity on the
x-axis), then a at line at a y-value higher than 0.514
3
would indicate that hepatitis B is
important and moreover that it is the only factor that increases SRB. If the plot were to be
a curve which is not at, that would suggest that sex-selective abortion is also important.
But why would the probability of male birth (or the SRB) vary across birth parities?
If there is son preference in the population such that couples target a particular number
of sons and also practice sex-selective abortion, then the probability of male birth would
generally tend to increase with the rst few births
4
. This is because the fraction of couples
who decide to have a second child would have an over-representation of those who have had
a girl as the rst child. Son preference would, therefore, imply that the probability of any
of these couples inducing a sex-selective abortion would be higher than the corresponding
probability for the whole population. The variation of the probability of male birth across
birth parities might display a dierent pattern when couples practice female infanticide which
seems to have been the case in India prior to the late-1980s. But in both cases - sex selective
abortion and female infanticide - it will be the case that the probability will vary signicantly
2
SRB is a one-one mapping of the fraction of male births; hence both provide the same information and
either can be used for analysis.
3
Probability of a male birth of 0.514 is widely accepted accepted as the average for human populations
(for instance, see Hesketh and Wei Xing, 2006)
4
For evidence on the presence of son preference see (Arnold, et al., 1998 and Larsen, et al., 1998).
4
across birth parities.
Whereas Jha, et al. (2006) and Ebenstein (2007) among others, have worked with SRBs
for India and China respectively, computed directly from Census and other survey data, I
will instead work with the probability of male birth. Since the probability of a male birth
cannot be observed, I will instead estimate it (along with other parameters of my model)
from the data. This approach has two advantages over those that work directly with ratios
computed from the data (for instance the sex ratios at birth) to infer sex selection, like Jha,
et al. (2006) and Ebenstein (2007). First, my estimates emerge from a behavioral model
of the households practicing sex selection (or neonatal infanticide), as opposed to a purely
empirical approach of looking at the computed ratios.
Second, my model allows one to endogenize the probability of male birth and therefore
study the eect of covariates on sex selection jointly with related decision-making, like son
targeting and household size decisions. There are at least two reasons why endogenizing the
probability of male birth might be important. First, there is considerable evidence pointing
to the existence of strong son preference in India (see, for instance, Basu and de Jong, 2007).
Since son targeting behaviour and family size decisions might have information about sex
selection, joint estimation of the eects of covariates is more ecient. Second, this is also
important from a policy perspective because designing appropriate policies to tackle the
problem of missing women would require knowledge of its eective determinants. Eect
of covariates on sex selective behaviour will be naturally provided by my approach whereas
an approach based on empirically observed ratios will have to work out the eect of covariates
in a separate exercise which might not account for son targeting behaviour.
For the purposes of critically examining the relationship of hepatitis B to the SRB, I
re-work the empirical model developed in Basu and de Jong (2007) in two steps. In the
rst step, I let the probability of male birth remain exogenous but allow it to vary across
birth parities. I estimate these probabilities from the data using the method of maximum
likelihood
5
and test the null hypothesis of their equality. I use detailed birth history data
provided by the three waves of National Family and Health Survey (NFHS) in India and nd
strong evidence of variation in the probability of male birth across birth parities for 1992,
1998 and 2005. I conclude that human intervention in the form of sex-selective abortion (or
female infanticide) has been prevalent in India over the last two decades.
Of course, as noted already, variation of the estimated probability of male birth across
birth parities is not, by itself, sucient to rule out the eect of biological factors like hepatitis
B. It is in principle possible for both eects to be present. Thus, a plot of the estimated
probabilities of male birth which is both shifted upwards of 0.514 and is also not at might
result from the combination of sex-selective abortion and hepatitis B. One could adopt any of
the following two empirical strategies to infer the eect of biology on the SRB. First, one could
test whether the probability of male birth is statistically indistinguishable from the normal
gure for human populations at any birth parity, as Ebenstein (2007) does. Alternatively,
5
Note that these estimated probabilities and the fraction of male births both converge in probability
to the true probabilities of male birth and so in large samples they provide good approximations for the
population SRB.
5
one could control for the eect of hepatitis B directly in the estimation procedure as Lin and
Luoh (2006) do.
Since all the estimated probabilities of male birth, i.e., at all birth parities and for all
the three time periods, are statistically signicantly greater than 0.514, I rule out the rst
strategy. Thus, I adopt the second strategy which means that I need to include a variable
measuring the incidence of hepatitis B in the estimation procedure. Since NFHS does not
provide data about the prevalence of hepatitis B I had to turn to a dierent source to
nd an observable variable to account for the prevalence of hepatitis B in the population:
epidemiology of hepatitis B in India.
The epidemiological literature on hepatitis B in India suggested a proxy variable for
prevalence of hepatitis B. Several studies, spanning the last two decades, provide evidence
that the prevalence of hepatitis B is signicantly higher in the tribal (ST henceforth) popula-
tion in India compared the general population. A summary of nine studies of 21 tribes spread
across 8 Indian states relating to this issue can be found in Murhekar and Zodpey (2005)
which concludes by saying: Our study indicates that the prevalence of HBsAg among the
scheduled tribes of India is much higher than in the general population (page 270, Murhekar
and Zodpey, 2005). This solved the problem of data unavailability on hepatitis B because
the NFHS data-set identies the ethnicity of the respondent, i.e., whether she belongs to a
scheduled tribe or not. Thus, a dummy variable on membership to a scheduled tribe can be
legitimately used as a proxy for the prevalence of hepatitis B. Note that this procedure to
control for any eect of hepatitis B infection on the SRB does not depend on the route of
infection; thus my procedure is valid irrespective of whether maternal or paternal infection
aects the SRB. Hence I can potentially address the arguments not only in Oster (2005) but
also in Blumberg and Oster (2007).
This brings me to the second stage of my empirical analysis. Now I endogenize the
probability of male birth while constraining it to be equal across birth parities
6
. I allow
the probability of male birth to depend on the proxy for hepatitis B and also control for
other covariates like income, education, geographical locations, etc. I re-estimate the full
model, once again using MLE, using NFHS data for India for 1992 and 1998. Estimation
of the full model with endogenous (but equal probabilities of birth across parities) throws
up interesting results. The estimate on the proxy for hepatitis B come out with a negative
sign; for 1992, it is not signicant but for 1998, it is signicant at the 5 percent level. This
shows that either there is no statistically signicant eect or a negative eect of hepatitis B
infection on the probability of male birth. Juxtaposed with the nding that the probability
of male birth varies signicantly across birth parities, this suggests that social rather than
biological factors are responsible for the increasing SRB in India.
The rest of the paper is organized as follows: in the next section, I present the basic em-
6
One could, in principle, endogenize the probability of male birth and allow it to vary across birth
parities. This would, of course, increase the computational burden substantially. One way to avoid the
additional computational burden would be to adopt a two step estimation procedure. The rst step would
be a probit regression for each birth parity and the second step would use estimated probabilities in the
maximum likelihood estimation and correct standard errors for the two step procedure. Work on this project
is currently underway.
6
pirical model that I use in this paper. In the following section I present results of maximum
likelihood estimation of my empirical model with variable and estimated probability of male
birth across birth parities using NFHS data for 1992, 1998 and 2005 for India; this section
presents evidence of sex-selective abortion in India in the last two decades. In section 4, I
present results of estimating the model with endogenous probability of male birth. One of
the determinants of the probability of male birth is membership in a scheduled tribe (the
proxy for prevalence of hepatitis B). This section presents evidence to show that prevalence
of hepatitis B cannot explain any part of the missing women problem in India and com-
plements similar ndings for China by Ebenstein (2007). The nal section concludes the
paper.
2 The Empirical Model
In this section, I develop my empirical model by re-working the corresponding model in Basu
and de Jong (2007). The empirical analysis in Basu and de Jong (2007) jointly analyzed the
eect of covariates on son targeting behaviour, fertility behaviour and the interaction be-
tween the two for India, a country which displays the presence of son preference manifested
in dierential stopping behaviour (DSB). Note that DSB or male-preferring stopping rules
on fertility behaviour have been studied extensively in the sociology and anthropology liter-
ature; this literature has pointed out that DSB is a manifestation of strong son preference
(i.e., a strong preference for male rather than female ospring) in society. DSB leads to a
disadvantageous position of women in society even in the absence of direct discrimination
within the household, as argued in Basu and de Jong (2007).
The disadvantageous position of women arise because of two consequences of DSB. First,
girls are born, on average, into larger families and are thus required to share resources with
a larger sibling cohort. Second, girls are born, again on average, as relatively elder children
within families and are thus required to carry a larger share of the burden of housework.
Both these eects ensure a disadvantageous position for women at the aggregate level even
in the absence of household-level discrimination. Direct discrimination in the allocation of
resources, if present, add to the disadvantage already created by male-preferring stopping
rules on fertility behaviour.
The empirical model in Basu and de Jong (2007) derived the joint probability of observing
a particular birth sequence of living children and desired maximum number of children for all
families with completed birth histories; parametrizing this likelihood function appropriately,
the eect of covariates on son targeting and fertility behaviour was estimated by using the
method of maximum likelihood. While deriving the likelihood, the probability of male birth
was assigned a constant value of 0.514 across birth parities. I relax that assumption in
this paper in three ways. First, instead of assigning a particular value to this probability, I
estimate it jointly with other parameters of the model. Second, I let the probability of male
birth vary across birth parities. Third, I endogenize the probability of male birth and allow
it to be jointly determined by covariates in the model.
The following derivation of the likelihood function relies heavily on the corresponding
7
section in Basu and de Jong, 2007. Let me start by introducing some notation. Let S
i
denote the completed birth sequence for the i
th
family; for instance, S
i
could be BBG (where
B stands for boy and G for girl). Let N
i
denote the maximum number of children that family
i would like to have; let k
i
denote the target number of boys for family i. To estimate the
probabilities of male births at dierent birth parities jointly with the eect of covariates on
targeting and fertility behaviour, I will calculate the joint likelihood of observing a given
birth sequence (S
i
) and desired maximum number of children (N
i
), parametrize the joint
probability appropriately and then maximize it over the parameter space; in other words, I
will compute a parametric expression for P(S
i
, N
i
) where P(.) denotes probability.
To proceed let me introduce T
i
, a dichotomous unobservable variable which indicates
whether family i targets sons or not. T
i
= 0 means that the family does not target; and
T
i
= 1 implies that family i is a son targeter. Finally, I let T
i
to be determined by a vector
of covariates, X
i
, in the following manner:
T
i
=

0 if X

i
+
i
0
1 if X

i
+
i
> 0
(1)
where X
i
is a (k 1) vector of co-variates which determine whether a particular family
targets sons or not, is a (k 1) vector of parameters to be estimated and
i
N(0, 1)
captures the unobservable, omitted, stochastic factors that aect son targeting behaviour.
To get the likelihood for the observed birth sequence and maximum number of children
for the i
th
family, note that
P(S
i
, N
i
) = P(S
i
, N
i
|T
i
= 0)P(T
i
= 0) + P(S
i
, N
i
|T
i
= 1)P(T
i
= 1)
= P(S
i
|N
i
, T
i
= 0)P(N
i
|T
i
= 0)P(T
i
= 0)
+ P(S
i
|N
i
, T
i
= 1)P(N
i
|T
i
= 1)P(T
i
= 1)
= P(S
i
|N
i
, T
i
= 0)P(N
i
|T
i
= 0)(X

i
)
+ P(S
i
|N
i
, T
i
= 1)P(N
i
|T
i
= 1)[1 (X

i
)] (2)
where (.) denotes the standard normal cdf. Note that, because of DSB, when a family does
not target sons, the eective stopping rule for childbirth becomes the maximum number of
children that the family desires to have, N
i
; hence the probability of observing S
i
given N
i
when the family does not target sons is
P(S
i
|N
i
, T
i
= 0) = I(n(S
i
) = N
i
)
m
i

j=1
(q
1B
j
j
) (1 q
j
)
B
j
, (3)
where I(.) denotes the indicator function and n(S
i
) denotes the number of children in birth
sequence S
i
, B
j
= 0 if the j
th
child is a boy and B
j
= 1 if the child is a girl, m
i
is the number
of children alive in family i, and q
j
is the probability of a male birth at birth parity j (which,
at this stage, I ssume to be constant for all families in the population). The indicator function
in the product is meant to rule out the possibility that a family which stops childbearing
8
before hitting the desired maximum number of children could be a non-targeter. Any family
which stops childbearing before hitting the ceiling, N
i
, has to be a son targeter in my model.
In the special case when the probability of male birth does not change across birth parities
(denoted, in this case, by q) and b
i
is the number of boys in family i, g
i
is the number of
girls in the same family, I have:
P(S
i
|N
i
, T
i
= 0) = I(n(S
i
) = N
i
)(q
b
i
) (1 q)
g
i
(4)
This gives us the rst term in (2). The second term in (2) comes from son targeters; so
I need to compute P(S
i
|N
i
, T
i
= 1). Note that when a family targets sons, its target, k
i
,
can range anywhere from 1 to N
i
1; targeting k
i
= N
i
sons with a ceiling for the desired
maximum number of children at N
i
is equivalent to not targeting. Since I cannot observe
k
i
(the target number of sons for a family), I condition on k
i
and then integrate it out as
follows:
P(S
i
|N
i
, T
i
= 1) =
N
i
1

k
i
=1
P(S
i
|N
i
, k
i
, T
i
= 1)P(k
i
|N
i
, T
i
= 1) (5)
where P(S
i
|N
i
, k
i
, T
i
= 1) is the probability of observing a birth sequence S
i
given N
i
, k
i
and
T
i
= 1 (son targeting)
7
; P(k
i
|N
i
, T
i
= 1) is the probability of targeting k
i
sons given that the
desired maximum number of children is N
i
. Since these targeting (conditional) probabilities
cannot be observed, in my model, I treat them as parameters and estimate them jointly with
other parameters. Using (5), therefore, (2) becomes:
P(S
i
, N
i
) = P(S
i
|N
i
, T
i
= 0)P(N
i
|T
i
= 0)(X

i
)
+ [1 (X

i
)]
N
i
1

k
i
=1
P(S
i
|N
i
, k
i
, T
i
= 1)P(k
i
|N
i
, T
i
= 1)P(N
i
|T
i
= 1).
Next, I assume that N
i
conditional on T
i
is distributed as a Poisson random variable with
conditional mean given by
i
. I try to capture two crucial facts with a Poisson distribution:
one, that N
i
conditional on T
i
is a count variable; and two that there is interaction between
the decision of son targeting and the desired maximum number of children (the interaction
term appears in the expression for the conditional mean,
i
).
P(N
i
|T
i
) =
exp(
i
)[
i
]
N
i
N
i
!
,
where

i
= exp(Z

i
+ T
i
). (6)
Note that in the above expression captures the eect of son targeting on the total fertility
rate and Z
i
is a set of covariates which aects desired family size. Moreover, since T
i
is a
7
The method used for computing these conditional probabilities can be found in Basu and de Jong, 2007.
9
dichotomous variable, I have
P(N
i
|T
i
= 0) =
exp(exp(Z

i
))[exp(Z

i
)]
N
i
N
i
!
(7)
and
P(N
i
|T
i
= 1) =
exp(exp(Z

i
+ ))[exp(Z

i
+ )]
N
i
N
i
!
(8)
Using (7), (8) and (5), I can write (2) as:
P(S
i
, N
i
) = P(S
i
|N
i
, T
i
= 0)
exp(exp(Z

i
))[exp(Z

i
)]
N
i
N
i
!
(X

i
)
+
N
i
1

k
i
=1
P(S
i
|N
i
, k
i
, T
i
= 1)P(k
i
|N
i
, T
i
= 1)

exp(exp(Z

i
+ ))[exp(Z

i
+ )]
N
i
N
i
!
[1 (X

i
)] (9)
The log-likelihood for the observed sample, then, becomes:
l = ln(L) =
n

i=1
ln(P(S
i
, N
i
)) (10)
where n is the number of families in the sample and P(S
i
, N
i
) is substituted from (9). Maxi-
mizing this function over the parameter space gives me estimates of the relevant parameters
(which includes the probabilities of male birth at dierent birth parities).
3 Evidence for Sex-Selective Abortion
In this section I present results of estimating the empirical model developed in the earlier
section with three waves of NFHS data from India: 1992, 1998 and 2005. There are several
things to note about how the NFHS data is used for estimating the model. First, I restrict
myself to children who were alive at the time of the interviews; this is motivated by the
observation that son targeting behaviour responds to the sex composition of existing children
who are alive. Second, from all the respondents in the data, I pick up families which can
be reasonably assumed to have completed their fertility history. This is necessary because
the likelihood function derived in the earlier section is valid only for completed sequences of
births and for children who are currently alive. Third, I do not take account of the fact that
some of the births might be accidental
8
. Fourth, for families which have an actual number
8
For instance, a family which has stopped after a boy and a girl (in that order) might have been targeting
one boy and have had the second child accidentally. If this family reports a desired maximum of 2, in my
model, this family would come out as a non-targeter whereas in fact it was targeting. Given the fact that
contraception use has been rapidly increasing in India though, the proportion of accidental births can be
expected to fall over time and thus not aect the results signicantly.
10
of children greater than what they report as their desired maximum, I recode their desired
maximum as the actual number of children alive.
One possible shortcoming of my approach should be noted immediately. In the NFHS
questionnaire, there are questions both about the desired number of sons and the desired
total number of children. While it is true that answers to both these questions are possibly
unreliable or untrue, I need to assume that at least one of these is an approximation to reality.
In writing the likelihood and also in the estimation, I assume that the desired total number
of children is observed (or truly revealed) whereas the desired number of sons is either not
observed or not truly revealed. The alternative would have been either to assume that both
are truly revealed or to assume that the information about the son target was true while the
information about the total was not. The assumption that one is true is, of course, better
than assuming that both are: in the former case requirements of reliability from the data
are less demanding. Again, information on (desired) total number of children might be more
reliable than information on the desired son target because couples might be less willing to
reveal their son preference. The approach in this paper is one possible approach and future
research should certainly explore alternative ways of deriving the empirical model.
In presenting the results I focus on two aspects. First, I present estimates of the proba-
bility of male birth at dierent birth parities
9
; these estimates emerge from estimating the
unrestricted model, i.e., allowing the probabilities to vary across birth parities. Second, I test
the null hypothesis that all these estimated probabilities are equal. To do so, I re-estimate
the model under the null (the restricted model) i.e., with the restriction that the probability
of male birth is same at all birth parities; then I use the likelihood ratio (LR) statistic to
test the null of equality of birth probabilities. A large (positive) value of the LR statistic is
interpreted as evidence against the null hypothesis.
Before proceeding, let us pause to think as to why the probability of male birth might
vary across birth parities? To answer this question let us think of a hypothetical situation
where couples have strong preference for male ospring (as is the case for India). Moreover,
this son preference has two distinct manifestations: one, couples target a particular number
of sons by practicing male-preferring stopping rules (as discussed, for instance, in Basu and
de jong, 2007); and two, couples practice sex-selective abortion, again, in their attempt to
attain their target number of boys. These two together will tend to increase the probability
of male birth with the rst few birth. To see this, let us start with a group of couples who
are at the start of their fertility history. After the rst child is born, a certain fraction
will stop childbearing. Because of son targeting, a large proportion of the couples who stop
childbearing will be those who have had a son as the rst child. Thus, a large proportion
of the couples who go on to have a second child will be those who have a daughter as the
already-existing child. Assuming that the probability of choosing sex-selective abortion is
higher for a couple with a daughter than one without a child, we see that the fraction of
couples who will choose sex-selective abortion will be higher for those deciding to have a
second child than for the original group that we started with. Hence, the probability of
9
For the estimation I consider the rst through sixth birth parities. R code for estimation is available
from the author upon request.
11
male birth will be higher for the second than for the rst birth. Of course, there is a cost
associated with sex selective abortion, both material and psychological; and this cost acts as
a countervailing force on couples desires to attain their target for boys. Thus, the probability
of male birth might be expected to dip back after rising for a few births if sex selection is
widely prevalent.
This argument would lose some of its force if there were biological factors which led
to variation in the probabilities of male births across birth parities. Demographers and
population biologists have analyzed various biological, environmental and social factors as
possible determinants of the sex ratio at birth (SRB) in human populations. Though birth
order (or parity) has gured in many such studies as a possible determinant of the SRB, the
current consensus seems to be that birth order does not aect the SRB (Ulizi and Zonta,
1995; Hesketh and Xing, 2006). More directly, I ran probit regressions with the sex of the
child as the dependent variable and used dummies for birth parities as regressors, apart from
the conventional controls. I did not nd signicant coecients on any of the birth parity
dummies
10
. Both these pieces of evidence allows me to rule out biology, pure and simple, as
a possible determinant of the variation in the probability of male birth across birth parities.
Estimation results about the probabilities of male birth are presented in Tables (1),
(2) and (3) for years 1992, 1998 and 2005 respectively; Figure (1) visually summarizes the
information in these tables about the probabilities of male birth across birth parities. The
rst thing to notice about the estimated probabilities for all the three time periods is that
they are all signicantly higher than the normal gure of 0.514
11
; this suggests that either
sex-selective abortion or hepatitis B or both are in operation in India right from the rst
birth parity. This nding is dierent from Ebensteins (2007) result that the SRB for the
rst birth is close to the average and is more in line with the data in Jha, et al (2005) where
the fraction of male births at the rst birth is 0.53.
The second thing to notice is that the LR statistic, which is distributed as a
2
random
variable with 5 degrees of freedom, is signicant at the 10 percent level in 1998 but is signif-
icant at the 1 percent level for both 1992 and 2005. Thus, evidence of human intervention
through sex-selective abortion or neonatal infanticide is strong for both 1992 and 2005, and
a little weaker for 1998. Since ultra-sound technology started becoming available in India
only from the late-1980s onwards, it seems that prior to the availability of this technology
neonatal infanticide (or under-reporting of girls) was prevalent as shown by the large LR
statistic for 1992. This is an issue that needs to be explored further.
Probabilities for a male birth for dierent birth parities have been plotted for all the
three time periods in Figure (1). The fact that these probabilities vary considerably across
birth parities strikes one immediately. It also seems to be the case that the form and pattern
of human intervention has changed over the last two decades. Whereas there was heavy
intervention, either in the form of infanticide or non-reporting of girls, in the early birth
parities for both 1992 and 1998, intervention seems to have shifted more towards the later
10
I have not reported the results of the probit regression in the text of the paper; these can, however, be
obtained upon request.
11
The t-statistics in the tables arise from the test of the probabilities being equal to 0.514.
12
births in recent years. This emerges from the fact that the estimated probabilities decrease
over the birth parities for 1992 and 1998, whereas it has a more complicated behaviour for
2005: it increases from the rst through the third birth and then decreases for the next two
birth before rising sharply again.
It is intuitively clear that sex-selective abortion and neonatal infanticide entail very dif-
ferent psychological costs for the couple, especially for the mother. Whereas abortion in the
early stages of pregnancy might be stressful, neonatal infanticide is considerably more so.
The psychological cost of neonatal infanticide, moreover, might increase sharply with the
presence of children in the family. Since most of the respondents interviewed in the 1992
wave of the NFHS did not have access to ultrasound technology for sex determination of
the foetus
12
and yet the probability of male birth is considerably higher in the early birth
parities, it suggests that either neonatal infanticide or under-reporting of girls was common
practice in the early and mid 1980s. The downward trend in the birth probabilities might
then be explained by the fact that the existence of children reduces the probability of parents
to accept neonatal infanticide. On the other hand, most of the respondents interviewed in
the 2005 wave of the NFHS might be expected to have had easy access to ultrasound technol-
ogy. Hence, the practice of neonatal infanticide might be expected to have given way to sex
selective abortion. This might be one way to explain the dierent shapes of the probability
plots for 1992 and 2005, with 1998 being a period of transition.
4 Does Biology Have a Role to Play?
The fact that the estimated probabilities of male birth vary signicantly across birth parities,
as seen in the previous section, provides strong evidence of human intervention in the form
of sex-selective abortion And possibly also neonatal infanticide. But this, by itself, does
not rule out a role for biology. As seen in Figure (1), all points on the probability plot is
signicantly higher than the average gure of 0.514. Hence it is conceivable that both biology
and human intervention has a role to play.
Since NFHS data does not provide any information about hepatitis B, I need to look
elsewhere for information on hepatitis B. Since I do not have direct data on hepatitis B, the
estimation strategy that I will adopt is to use a proxy for hepatitis B. A suitable proxy is
suggested by the epidemiological literature on hepatitis B in India. Murhekar and Zopdey
(2005) summarize a substantial body of epidemiological research into the incidence of hep-
atitis B among tribal populations in India. Bringing together nine studies about 21 tribes
across 8 states in India, they conclude that the prevalence of hepatitis B is signicantly
higher in the tribal population as compared to the general population. I use this observation
to construct a proxy for the prevalence of hepatitis B by constructing a dummy variable on
whether the respondent is a tribal or not; I use this dummy to capture the possible eect
of hepatitis B on the probability of male birth. Note that this empirical strategy is valid
12
Ultrasound technologies for sex determination became prevalent in India only from the mid-1980s on-
wards.
13
irrespective of whether hepatitis B infection aects the sex ratio through maternal or pater-
nal infection; thus, my method can test the claim of the eect of hepatitis B on the SRB
presented in Oster (2005) which had argued for the maternal route and also Blumberg and
Oster (2007) which had argued for the paternal route of the eect.
To incorporate the eect of hepatitis B on the probability of a male birth directly I add
an equation to the empirical model. This equation determines the probability of male birth
for family i as a function of covariates. For our purposes the most important covariate in this
equation is the dummy variable dened on whether the respondent belongs to a scheduled
tribe (ST) or not; we call this variable hepb and note that it captures the possible eect of
hepatitis on SRB.
hepb
i
=

1 if family belongs to a ST
0 if family does not belong to a ST
(11)
As additional controls, I use standard covariates: age of the mother, years of formal education
of the mother, a dummy variable on whether the mother works outside the household for
a wage, a dummy variable on whether the household resides in a rural area, two income
dummies (poor and middle
13
), a dummy on whether the family belongs to the largest religious
community (Hindu), a dummy on whether the household is a joint family. Then, the equation
determining probability of male birth for family i is:
p
i
= (w

i
) (12)
where w
i
are the covariates determining the probability of male birth (including hepb and a
constant) and is the vector of corresponding parameters.
Results of the estimation for 1992 and 1998 are presented in Tables (4) and (5)
14
. The
most striking aspect of the results in Tables(4) and (5) is that the coecient on hepb is
not statistically signicantly dierent from zero in 1992 and is signicantly negative in 1998.
This demonstrates either that the prevalence of hepatitis B does not have any signicant
impact on the SRB or that it has a negative impact
15
in India. In either case the positive
association between prevalence of hepatitis B and the SRB, whereby hepatitis B infection
rates is supposed to increase the SRB, is not observed in the data for India. In Figure (2)
and (3), I present the distribution of the estimated probability of male birth for families in
the sample. The median of the distribution is shifted far away to the right of the normal
value of 0.514 for both 1992 and 1998 giving indication of huge human intervention.
13
Since there are no income variables in DHS, we had to construct income/class dummies in the following
manner: a respondent is designated rich if she owns a car/truck and has electricity connection in her house;
a respondent is designated middle if she has electricity in her house but does not own a car/truck; and a
respondent is designated poor if she neither owns a car/truck nor has an electricity connection to her house.
14
I have not been able to do a similar estimation for 2005 because the preliminary data released by DHS
does not include the variable that gives information about membership in a scheduled tribe.
15
A signicantly negative value of the parameter on the proxy for hepatitis B suggest that, contrary to
Osters (2005) contention, the prevalence of hepatitis B in India increases the probability of a female birth.
14
The other determinants of the probability of male birth (or equivalently sex selection) are
as follows: age, years of formal education, working outside the household and contraception
use by the mother reduces the probability of sex selective abortion; geographical location
(i.e., rural households), living in a joint family and religion of the mother (i.e., whether she
belongs to the largest religious community, Hindu, or not) increases the probability of sex
selection. Whereas for 1992, belonging to the Hindu religious community does not have any
impact on sex selection, this shows up as increasing the probability of sex selective abortion
for 1998.
If we turn to the determinants of son targeting behaviour we nd the following: age
has a negative impact while formal years of education has no impact on the probability
of son targeting behaviour; use of contraception, rural location of the household, working
outside the household by the mother and income seems to increase the probability of son
targeting behaviour (whereby couples target a particular number of sons). Compared to
rich households, poor households have a higher probability of son targeting; middle-class
households, on the other hand, do not display any signicant dierence in son targeting
behaviour when compared to rich households.
The eect of belonging to the largest religious community changes sign, while retaining
statistical signicance, between 1992 and 1998: whereas hindu households were more proba-
ble, compared to non-hindu households, to practice son targeting in the early and mid-1980s,
in the early and mid-1990s the data shows them to be less probable to do the same. Taken
together with the nding that hindu households have become more probable to practice sex
selective abortion between mid-1980s to the mid-1990s, this suggests that the manifestation
of son preference has shifted over time for hindu households. Whereas son preference got
manifested in targeting behaviour and male preferring stopping rules on fertility in the earlier
period, the manifestation of son preference seems to have gradually shifted to more direct
forms of intervention like sex selective abortion.
When we turn to the determinants of average family size, we observe the following:
formal years of education of the mother has a strong negative impact on average family size;
age, rural location and family income seem to increase the average family size signicantly.
Membership in the largest religious community shows a reversal of sign over the two periods.
Whereas hindu households had higher average family size in the earlier period (1992), they
have a lower average family size in the latter period (1998). This is consistent with the
fact observed earlier that hindu households have gradually shifted from son targeting to
sex selection: son targeting increases the average family size (as shown by the sign of the
coecient on the interaction term in Table 4 and 5) and so a shift away from son targeting
could be expected to reduce average family size as observed.
5 Conclusion
The problem of missing women in parts of Asia and North Africa, whereby social inequalities
outweigh womens natural survival advantage and give rise to a large decit of women in the
population, has been highlighted by Sen (1990, 1992), among others. Oster (2005) challenged
15
the thrust of Sens argument by proposing a biological mechanism to explain at least part of
the missing women problem. She suggested that the prevalence of hepatitis B in a population
skews the SRB in favour of male ospring and could thus explain part of the female decit
in the adult population.
There is of course a dierent mechanism which can also aect the SRB: human interven-
tion in the form of sex-selective abortion. A cultural milieu marked by strong son prefer-
ence has combined with the gradual spread of ultra-sound technology in India over the last
two decades to markedly increase the prevalence and acceptance of sex-selective abortion
16
.
Hence, there are two competing hypotheses about the increasing SRB in India over the last
two decades: biology and sex-selective abortion.
In this paper I empirically test for these alternative hypotheses in two steps. In the rst
step I use a simple model of fertility behaviour (adopted from Basu and de Jong, 2007) to
demonstrate that the probability of male birth varies signicantly across birth parities; this
is evidence of human intervention in the form of sex-selective abortion since prevalence of
hepatitis B by itself cannot account for such variation across birth parities. In the second
step, I use a novel proxy for the prevalence of hepatitis B - membership in a scheduled tribe
- to directly estimate the eect of hepatitis B on the probability of male birth. I nd no
statistically signicant eect for 1992 and a negative impact for 1998. Based on these two
steps I conclude that biological factors, like the prevalence of hepatitis B, have no eect on
the SRB in India; rather, it is direct human intervention in the form of sex-selective abortion
which is more important.
References
Arnold F, Choe M. K. and T. K. Roy. (1998): Son Preference, the Family-building Process
and Child Mortality in India. Population Studies, 52, 301-315.
Basu, D. and R. de Jong. (2007): Son Preference and Gender Inequality. Working paper,
The Ohio State University. Available from http://web.econ.ohio-state.edu/basu/
Blumberg, B. and E. Oster. (2007): Hepatitis B and Sex Ratios at Birth: Fathers or Moth-
ers? Working paper, University of Chicago.
Coale, A. J. (1991): Excess Female Mortality and the Balance of the Sexes in the Popula-
tion: An Estimate of the number of missing females, Population and Development Review,
17, 517-23.
Das Gupta, M. (2005): Explaining Asias Missing Women: A New Look at the Data,
Population and Development Review, 31(3), September, 529-35.
16
For recent reporting on this issue, see the detailed coverage provided in BBC at the following website:
http://news.bbc.co.uk/2/hi/south asia/4592890.stm
16
Dubner, S. J. and S. D. Levitt. (2005): The Search for 100 Million Missing Women: An
Economics Detective Story, Available at http://www.slate.com/id/2119402/
Ebenstein, A. Y. (2007): Fertility Choices and Sex Selection in Asia: Analysis and Policy.
Available from http://www.demog.berkeley.edu/ebenstei/
Hesketh, T. and Z. Wei Xing. (2006): Abnormal sex ratios in human populations: Causes
and consequences, Proceedings of the National Academy of Sciences of the United States of
America, 103, 13271-13275.
Jha, P., Kumar R., Vasa, P., Dhingra, N., Thiruchelvam, D., and R. Moineddin. (2006):
Low female-to-male sex ratio of children born in India: National survey of 1.1 million house-
holds. Lancet, 267, 211-18.
Larsen U., Chung W. and M. Das Gupta. (1998): Fertility and Son Preference in Korea.
Population Studies, 52, 317-325.
Lin, M-J. and M-C. Luoh. (2007): Can Hepatitis B Mothers Account for the Number of
Missing Women? Evidence from Three Million Newborn in Taiwan. Working paper, Na-
tional Taiwan University.
Murhekar, M. V. and S. P. Zodpey. (2005): Hepatitis B virus infection among Indian tribes:
Need for Vaccination Program, Indian Journal of Gastroenterology, 24, Nov-Dec, 269-270.
Oster, E. (2005): Hepatitis B and the Case of the Missing Women, Journal of Political
Economy, 113 (6), 1163-1216.
(2006): Explaining Asias Missing Women: A New Look at the Data - Comment, Pop-
ulation and Development Review, 32(2), June, 323-27.
Sen A. (1990): More than 100 million Women are Missing. New York Review of Books,
December 20.
(1992): Missing Women: Social inequality outweighs womens survival advantage in Asia
and North Africa, British Medical Journal, 304, 7 March, 587-88.
17
Table 1: Probabilities of Male Birth: India, 1992
Estimate Std Err t-stat

Unrestricted Model
Birth 1 0.5549819 0.002237104 18.32
Birth 2 0.5472148 0.002268634 14.64
Birth 3 0.5415317 0.002665543 10.33
Birth 4 0.5433656 0.003558688 8.25
Birth 5 0.5374565 0.005238508 4.48
Birth 6 0.5380585 0.008996070 2.67
Restricted Model
0.5471772 0.001310896 25.52
Likelihood Ratio Test
LR-Stat = 60.59017

t-stat tests the equality of the probability to 0.5

Table 2: Probabilities of Male Birth: India, 1998
Estimate Std Err t-stat

Unrestricted Model
Birth 1 0.5465632 0.002111326 15.42
Birth 2 0.5475338 0.002141297 15.66
Birth 3 0.5460234 0.002595897 12.34
Birth 4 0.5404132 0.003559384 7.42
Birth 5 0.5428249 0.005362014 5.38
Birth 6 0.5384422 0.009297505 2.63
Restricted Model
0.5460816 0.001261171 24.62
Likelihood Ratio Test
LR-Stat = 9.751398

t-stat tests the equality of the probability to 0.5

18
Table 3: Probabilities of Male Birth: India, 2005
Estimate Std Err t-stat

Unrestricted Model
Birth 1 0.5370551 0.001904300 12.11
Birth 2 0.5384106 0.001941149 12.58
Birth 3 0.5435486 0.002493767 11.85
Birth 4 0.5389048 0.003561080 6.99
Birth 5 0.5316910 0.005526420 3.20
Birth 6 0.5520212 0.009668408 3.93
Restricted Model
0.5389151 0.001176208 20.83
Likelihood Ratio Test
LR-Stat = 38.41299

t-stat tests the equality of the probability to 0.5

19
1 2 3 4 5 6
0
.
5
1
0
.
5
2
0
.
5
3
0
.
5
4
0
.
5
5
0
.
5
6
Probability of Male Birth at Various Parities
Birth Parity
P
r
o
b
a
b
i
l
i
t
y

o
f

M
a
l
e

B
i
r
t
h
1992: LRStat=60.59
1998: LRStat=9.75
2005: LRStat=38.41
NORMAL RANGE
Figure 1: Probabilities of Male Birth
20
Table 4: ML Estimation Results: India, 1992
Estimate Std Err t-stat
Interaction
0.041328547 0.0078294176 5.2786233
Son Targeting
const -1.180412004 0.0943112033 -12.5161377
age -0.004970965 0.0014525798 -3.4221629
edu 0.000919164 0.0015499562 0.5930258
work 0.047472333 0.0220635002 2.1516229
contra 0.562338655 0.0233832050 24.0488271
rural 0.080335800 0.0255730249 3.1414273
jfam 0.017228990 0.0313973950 0.5487395
hindu 0.082218390 0.0391199421 2.1017002
poor 0.235484310 0.0796044907 2.9581787
middle 0.064844258 0.0768239081 0.8440635
Family Size
age 0.019968710 0.0002991885 66.7428959
edu -0.006931766 0.0004597032 -15.0787863
work -0.006682362 0.0054539650 -1.2252300
rural 0.065204498 0.0061701088 10.5678036
poor 0.546487374 0.0125697422 43.4764186
middle 0.483538399 0.0118513316 40.8003434
hindu 0.159631376 0.0091767188 17.3952564
Probability of Male Birth
const 0.296341027 0.0308139444 9.6171079
age -0.005152185 0.0004714838 -10.9275974
edu -0.001648216 0.0004921433 -3.3490576
work -0.017954897 0.0069637638 -2.5783323
contra -0.065163441 0.0076913617 -8.4722892
rural 0.017694529 0.0078220953 2.2621213
jfam 0.021459397 0.0106254590 2.0196207
hindu -0.022428115 0.0118625618 -1.8906637
poor 0.030035116 0.0247818994 1.2119780
middle 0.017585321 0.0239076978 0.7355506
hep B -0.002220185 0.0082496489 -0.2691248
Loglikelihood = -185131.6
21
Table 5: ML Estimation Results: India, 1998
Estimate Std Err t-stat
Interaction
0.051142619 0.0079015108 6.4725114
Son Targeting
const -1.088220639 0.0937427136 -11.6085891
age -0.004801223 0.0014083488 -3.4091148
edu 0.001447491 0.0023039247 0.6282717
work 0.123451287 0.0219811232 5.6162411
contra 0.601743066 0.0239045386 25.1727539
rural 0.061501102 0.0256545869 2.3972751
jfam 0.059366924 0.0294368290 2.0167568
hindu -0.172891209 0.0260537906 -6.6359330
poor 0.190235723 0.0750769517 2.5338765
middle 0.050295134 0.0712018759 0.7063737
Family Size
age 0.020509100 0.0002893944 70.8690199
edu -0.017713758 0.0006020142 -29.4241516
work 0.002384243 0.0052262292 0.4562072
rural 0.049110498 0.0060996461 8.0513684
poor 0.600831110 0.0128363083 46.8071580
middle 0.527397866 0.0118702427 44.4302514
hindu -0.058234297 0.0062398677 -9.3326172
Probability of Male Birth
const 0.259148612 0.0291972280 8.8757950
age -0.004264060 0.0004515681 -9.4427836
edu -0.002788472 0.0007455311 -3.7402495
work -0.034702822 0.0067136843 -5.1689683
contra -0.089631531 0.0077679448 -11.5386415
rural 0.021856430 0.0077549287 2.8183922
jfam 0.047881847 0.0099592898 4.8077572
hindu 0.023770794 0.0080154070 2.9656378
poor 0.020755695 0.0229687105 0.9036509
middle 0.007980385 0.0217812278 0.3663882
hep B -0.023605230 0.0105571004 -2.2359577
Loglikelihood = -201033.0
22
0.45 0.50 0.55 0.60
0
5
1
0
1
5
2
0
Distribution of Prob of Male Birth: MLE, 1992
probability of male birth
d
e
n
s
i
t
y
normal=0.514
median=0.5502
Figure 2: Distribution of the Probability of Male Birth, 1992
23
0.40 0.45 0.50 0.55 0.60
0
5
1
0
1
5
2
0
Distribution of Prob of Male Birth: MLE, 1998
probability of male birth
d
e
n
s
i
t
y
normal=0.514
median=0.5487
Figure 3: Distribution of the Probability of Male Birth, 1998
24