Professional Documents
Culture Documents
DOI 10.1007/s11113-006-9017-2
Received: 21 November 2003 / Accepted: 12 May 2005 / Published online: 14 March 2007
Springer Science+Business Media B.V. 2007
R. P. Pande (&)
International Center for Research on Women, 1717 Massachusetts Ave., N.W., Suite 302,
Washington, DC 20036, USA
e-mail: rpande@icrw.org
N. M. Astone
Department of Population and Family Health Sciences, The Johns Hopkins School of
Hygiene and Public Health, Baltimore, MD, USA
123
2 R. P. Pande, N. M. Astone
family size. Our results suggest that policy makers seeking to influence son
preference need to identify and target different policy levers to women in
different fertility and social contexts, rather than try an approach of one size
that fits all.
Introduction
‘‘May he (Pragâpati) elsewhere afford the birth of a female, but here he shall
bestow a man!’’ This hymn (Atharva Veda, Book VII, verse 11, translation by
Bloomfield, 1897) is an ancient Indian charm for women to give birth to sons.
Dating to circa 800 B.C.E., it reflects a sentiment that persists in India. Much
research has been done on demographic manifestations of this son preference,
particularly girls’ excess mortality and, recently, sex-selective abortion. There
is less research on son preference itself. In this paper we address this gap by
empirically examining son preference directly as the outcome of interest.
Although scholars are aware that gender inequality arises from systemic
factors, they are often unable to quantitatively disentangle the social struc-
tures and norms that are the ultimate cause of gender inequality, and that
contribute to the perseverance of son preference, from the individual deter-
minants of son preference. In our analysis, we test the independent effect of
both structural and social factors that set the norms for gender preference, and
characteristics of individual women, their households and their immediate
communities that influence how they respond to these norms.
Background
Son preference and discrimination against girl children are widespread in the
Middle East and North Africa (Yount, 2001) and in South and East Asia
(Chan & Yeoh, 2002). In Asia, this has been documented in China (Banister,
2003), Korea (Park & Cho, 1995), Vietnam (Belanger, 2002), Nepal (Leone,
Matthews, & Zuanna, 2003), and Bangladesh (Bairagi, 2001), as well as India
(Mishra, Roy, & Retherford, 2004), the setting for our study.
In the past, researchers interested in son preference have had as their
central concern its adverse consequences, particularly excess female infant
and child mortality or the poor health of girl children relative to boys (Basu,
1989; Das Gupta, 1987; Das Gupta & Shuzhuo, 1999; Makinson, 1994; Miller,
1981; Pande, 2003; Pande & Yazbeck, 2003, among others). Others have
examined the role of son preference in slowing the transition to low fertility as
couples bear children until they have sufficient boys (Arnold, Minja, & Roy
1998; Clark 2000; Das Gupta & Bhat 1997; Leone et al., 2003; Yount, Lang-
sten, & Hill, 2000, among others). The advent of technology permitting pre-
123
Explaining son preference in rural India 3
natal sex selection has shifted the focus of scholars and policymakers to
sex-selective abortion and consequent distorted sex ratios as manifestations of
son preference (Arnold, Kishor, & Roy, 2002; Murphy, 2003; Oomman &
Ganatra, 2002).
In the Indian setting, distorted sex ratios have a long history. Over the
100-year period from 1901 to 2001 the population sex ratios from Indian
censuses (expressed in India as the number of women per 1,000 men) have
shown more or less a continuous decline, from 972 females per 1,000 males in
1901 to 933 females per 1,000 males in 2001 (Banthia, 2001, p. 3).
In this period, the only noticeable aberration was in 1981, when the situa-
tion seemed to reverse with sex ratios becoming somewhat less masculine,
from 931 women per 1,000 men in 1971 to 934 in 1981. Any relief at this
change was cut short, however, by the 1991 census, which showed the lowest
sex ratio of the century, at only 929 women per 1,000 men. Further, between
1981 and 1991, the sex ratio for children between 0 and 6 years of age
decreased at a much faster pace than the overall sex ratio (from 962 girls per
1,000 boys in 1981 to 945 in 1991). The 1991 juvenile sex ratio was lower than
had been reported by any census since at least 1961. The 2001 census shows
that this trend is continuing (Banthia, 2001, pp. 3 & 8).
This drop in the 1991 census, reflecting that the last census of the 20th
century showed the worst sex ratios in this century, resulted in active debate
among academics and policymakers in India and elsewhere about the reasons
for this decline, particularly following as it did on the heels of an apparent
improvement between 1971 and 1981 (Griffiths, Matthews, & Hinde, 2000;
Mayer, 1999). The even more serious drop in the juvenile sex ratio in the
period of the late 1980s and early 1990s adds to the interest in this period,
because it is a time when son preference, as manifested in these distorted sex
ratios, remained high or possibly increased despite other more heartening
economic and social changes such as increasing female literacy.
What determines this strong son preference that affects so many demo-
graphic outcomes, in particular at the point when India faced its worst sex
ratio of the 20th century? In identifying important factors, we take as our lead
both the literature examining the determinants of demographic outcomes
thought to be the result of son preference, as well as the broader literature on
gender inequality in India.
123
4 R. P. Pande, N. M. Astone
patterns and sometimes formal rules for the allocation of material goods,
rights, opportunities and obligations between men and women (Baltiwala,
1994; Cain, 1993; Malhotra & Schuler, 2005; Morgan & Niraula, 1995). In
India, a number of structural factors lay the foundation for gender inequality.
These include kinship and marriage norms, the organization of the agrarian
economy, and rules and rituals associated with caste and religion.
In a path-breaking article in 1983, Dyson and Moore proposed that two
broad differences in Indian kinship patterns—between ‘‘Northern’’ and
‘‘Southern’’ India—largely determine gender inequality and son preference
(Dyson & Moore, 1983). Specifically, in the northern Indian kinship system, as
Dyson and Moore and others (Arnold et al., 1998; Das Gupta, 1995; Kishor,
1995) describe it, marriage is exogamous, that is, spouses must be unrelated
through kin and often by place of birth or residence. Parents of a girl often
have to pay all marriage costs and provide a large dowry. After marriage, a
girl typically becomes a member of her husband’s family and does not have
much interaction with her natal kin. In contrast, the southern kinship pattern
is characterized by endogamous marriage, that is, marriage between certain
types of cousins or within a defined, contiguous geographical area. Women’s
sexuality and freedom of movement are less curtailed, dowry is not a major
marriage transaction, and married daughters are often likely to be on hand to
render social and financial help to their parents.
Recent studies using data from the late 1980s and early 1990s suggest some
blurring of this North–South distinction with regard to gender inequality
(Arnold et al., 1998; Dharmalingam, 1996; Rahman & Rao, 2004). Most
recently, Rahman and Rao (2004), using data from 1995, dispute the idea that
the regional patterns identified by Karve (1965) and Dyson and Moore (1983)
still hold. They call attention to the fact that in modern India, Southern brides
are as likely to pay dowry (and pay as much) as Northern brides, and that
endogamous marriage can result in more inequity between wives and hus-
bands if their prenuptial relationship was already hierarchal (e.g., uncle–niece
matches). This is in accordance with other recent research that finds Indian
girls equally at risk of facing discrimination in health care in the South as the
North (Mishra et al., 2004).
Besides region, religion plays a role in defining appropriate social and
gender norms, which in turn influence son preference. In Hinduism, the major
religion in India, sons are crucial. Among Hindus, a dead parent’s soul can
only attain heaven if that person has a son to light the funeral pyre, and
salvation can be achieved through sons who offer ancestral worship (Vlassof,
1990). Girls and women do have some importance. Giving away a daughter in
marriage (kanyadaan) is considered meritorious (Miller, 1989). Studies
suggest that son preference exists also among other religious groups such as
the Sikhs (Das Gupta, 1987) or the Muslims (Murthy, 1996).
Caste may also be associated with cultural practices that influence women’s
roles, and thus son preference, such that one may expect less son preference
among lower castes and tribals than among high castes. Compared to lower
castes, higher castes have more rigid gender stratification systems, with strictly
123
Explaining son preference in rural India 5
Village development
Although the incentives for action ultimately flow from social and economic
structures, the decisions taken to act on a preference for sons occur at the level
of the individual woman and her household. Thus, given social norms, such
decisions are influenced by characteristics of individual women and their
123
6 R. P. Pande, N. M. Astone
123
Explaining son preference in rural India 7
Data
We use data from the rural sample of the National Family Health Survey,
India (NFHS-1), 1992–1993. We focus on the NFHS-1 specifically to examine
factors explaining son preference around the time of the 1991 census. The
NFHS followed the format of the Demographic and Health Surveys (DHS),
which are large-scale household surveys conducted in Asia, Africa, and Latin
America. Data are available on individual women, their households, and their
villages. For each Indian state, a multistage systematic stratified sampling
design was adopted, where the primary sampling units were selected
systematically, with probability proportional to size. Households were then
sampled using systematic sampling with equal probability, and all eligible
women in each household were interviewed. National and state-level sampling
weights were created to reflect sampling design (IIPS, 1995).
123
8 R. P. Pande, N. M. Astone
For this paper, we use a de jure sample of 50,136 ever-married women between
13 and 49 years of age at the time of the survey who have valid data on our outcome
variable. We chose to use the rural sample because village data are available only
for rural areas. Since about 73% of India’s population was rural at the time (IIPS,
1995, p. xxxi), the results will still be applicable to the majority of the population.
We use the de jure sample (all residents) rather than the de facto sample (all those
who slept in the sampled household the previous night) because detailed house-
hold structure and village data are only available for de jure women.
We combine the data from the NFHS with data on rice and wheat pro-
duction by state between 1989 and 1993. These data were obtained from the
Economic Survey of India, 1994–1995, Directorate of Economics & Statistics,
Department of Agriculture & Cooperation of the Government of India
(Government of India, 1995, p. S-19).
Dependent variable
Respondents were asked the ideal number of children they would want.
Women who had no surviving children were asked the question: ‘‘If you could
choose exactly the number of children to have in your whole life, how many
would that be?’’ Women who had living children at the time of the survey
were asked: ‘‘If you could go back to the time you did not have any children
and could choose exactly the number of children to have in your whole life,
how many would that be?’’ Those who responded to either question with a
number were then asked the follow-up question: ‘‘How many of these children
would you like to be boys and how many would you like to be girls?’’
Responses were entered as a number of boys, girls, either, or other response.
We created an ordered categorical variable that has three categories: zero
when a respondent reports no son preference (including daughter preference,
or a response of ‘‘either’’); one when a respondent reports an ideal of one son
more than the ideal number of daughters; and two when a respondent reports
an ideal of two or more sons more than the ideal number of daughters.
A majority (88.6%) of sample women gave valid numeric responses to
these questions (IIPS, 1995). Women whose characteristics are likely to be
associated with lower son preference (for example, educated women and
women with media access) are more likely to have given a valid numeric
response (Pande, 1999). This could lead us to underestimate the strength of
son preference in the analysis. Some women who may have higher son pref-
erence (those living in a joint family or from middle or high castes, for
example) are also slightly over-represented, and thus the overall direction of
any bias is not clear a priori. Given that almost 90% of sample women gave
valid responses, any bias is expected to be small.
Explanatory variables
123
Explaining son preference in rural India 9
100
% wanting more sons than daughters
80
Rajas
Bihar
UP
Jammu
Haryana
60
Punjab
MP
Gujarat
Maharash
Arunacha
Manipur
Orissa
Assam
< 20 %
Himachal
WBengal
Mizoram
Andhra
Tripura
Karnatak
40
Nagaland
20 - 40%
Delhi
Goa
41 - 60%
Kerala
Meghalay
20
TN
61% +
0
North Central South East
123
10 R. P. Pande, N. M. Astone
We include measures of religion and caste to capture norms and beliefs that
are associated with these social systems. We categorize religion as Hindu,
Muslim, or ‘‘other.’’ Caste is divided into low castes (including untouchables),
tribes, and ‘‘other,’’ where ‘‘other’’ includes high-caste Hindus and non-Hin-
dus. The research is inconclusive as to the relative strength of son preference
across religions, and we remain agnostic about the direction of our results for
this variable. We expect women from lower castes and from tribal groups to
have lower son preference than high caste women.
To specifically capture the effect of marriage customs, we include two vil-
lage endogamy variables as indicators of local marriage norms, one measuring
the proportion of women who marry relatives (kinship endogamy) and one for
the proportion who marry within the village (territorial endogamy). For both,
we create weighted continuous variables, with an exponential weight chosen
to bound the proportion between one and zero. The closer the proportion is to
one, the more endogamous the village. Some sample villages have a small
number of women only; however, the data allow us to group villages into
(larger) districts. Thus, the final variable for each of these indicators is a
weighted sum of the village-level and district-level aggregate for that indicator
(see Pande, 1999 for details). We expect endogamy to be inversely associated
with son preference.
To measure the agrarian ecology argument for son preference, we examine
the effect on gender preference of the extent of rice and wheat production.
Using data from the Economic Survey of India 1994–1995, we created con-
tinuous variables to measure the proportion of rice, and the proportion of
wheat, to total foodgrain production averaged over 1989–1990 to 1992–1993,
in each state. Women in rice-growing areas are expected to have weaker son
preference than others.
To assess the importance of village social and economic development, we
measure village-level opportunities for women, as well as general village
development. We use two continuous variables to measure village opportu-
nities for women, namely, the proportion of literate women in a village (to
measure female education) and the proportion of women in a village
employed outside of the home (to measure female employment). These are
created in a similar manner as the endogamy variables. Measures of village
development are whether a village has access to transport such as an
all-weather road, bus station or rail station (to measure accessibility to the
outside world), whether the village has at least a middle school (on the
assumption that, when there is a school in the village girls are more likely to
go to school which, in turn, could change their mother’s valuation of girls
versus boys), and the number of other facilities in the village such as a bank,
shop, or cinema (as an indicator of wider village economic development). We
expect all indicators of village development to be inversely associated with son
preference.
We use two household wealth measures. First, we create an index that is a
composite of variables describing consumer durables and housing quality,
measured in household quintiles (Filmer & Pritchett, 2001). Second, we
123
Explaining son preference in rural India 11
include a variable for household ownership of land, and expect women from
landed families to have higher son preference than others. Further, we
examine residence in an extended household defined as one that includes
members in addition to the respondent, her husband and her unmarried minor
children, and we expect women in extended families to have higher son
preference than those in nuclear households.
Our indicators for women’s characteristics are women’s level of education
(distinguishing between primary school only; up to middle school; secondary
school and higher); women’s employment (broadly defined as any work
outside of housework); women’s cash earnings; and women’s weekly access to
radio or TV and monthly access to cinema (as broad indicators of women’s
exposure to new ideas). We expect highly educated women, those earning
cash, and those with greater media exposure to have lower son preference
than others. We include mother’s age as a control variable.
Reported gender preferences may be affected by the current sex compo-
sition of children and family size (defined as the current number of surviving
children), either as an ex-post rationalization of actual family size and com-
position, or as the result of a change—either conscious or subconscious—in
preferences based on childbearing experience. Thus we control for actual sex
composition and family size in our analyses.
The ordered logit model is used when the outcome variable is categorized on
an ordinal scale, ordered by some conceptual or subjective criteria (McCul-
lagh & Nelder, 1989) as in our case. Following the notation of McCullagh and
Nelder (1989), the probability of a response for any one category of an out-
come Y can be expressed as p1,...pk for k possible values of Y. If the categories
are ordered, as in an ordered logit, we can consider cumulative response
probabilities pj = Pr(Y £ j) rather than the category probabilities pj. These
cumulative response probabilities can be interpreted as the probability of an
outcome up to a certain category, and can be written as:
c1 ¼ p1 ; c2 ¼ p1 þ p2 ; . . . ; ck ¼ 1:
123
12 R. P. Pande, N. M. Astone
into a category higher than j, with a given set of covariates. The odds ratio for
a unit change in a particular covariate, say from x = x1 to x = x2, is given by:
cj ðx1 Þ= 1 cj ðx1 Þ
¼ expðbx ðx1 x2 ÞÞ
cj ðx2 Þ= 1 cj ðx2 Þ
where bx is the coefficient of interest. The negative sign on the coefficient means
that a higher value of the variable increases the odds of a lower value of the
outcome. For example, a negative coefficient for maternal education means that
a higher value of maternal education is associated with higher odds for a lower
value of son preference (in other words, with weaker son preference).
The data in our sample are clustered such that all observations within the
same household, community, and primary sampling unit are unlikely to be
independent of each other. Thus if one estimates standard errors on the
assumption of independence, they will be incorrect, most likely smaller than
they should be (Liang & Zeger, 1993). In previous work using these data, the
effects of clustering at the household and village levels were found to be minimal
(Pande, 1999). We correct here for clustering at the level of the primary sampling
unit, using the ‘‘cluster’’ command in Stata version 7.0.
Descriptive results
A preference for girls is rare in this population. Only 2.6% of all respondents
say they want more daughters than sons (Table 1). About half show a desire
for a balanced sex composition and say they want children of either sex, or
equal numbers of daughters and sons. Almost half of all respondents (46%)
report some son preference. Among those who report some son preference,
the majority (75.6%) want one more son than daughter.
Son preference is, of course, only one of a set of preferences people have
about their families and should be interpreted in that context. Particularly
relevant for this study are family size preferences. Table 1 shows that the
pattern of reported son preference appears to vary by the reported total
number of desired children.
The highest reported son preference is among those who want only three
children: 88% of such women want more boys than girls. Further, the strength
of reported son preference, and the percentage who report no son preference
(those in the ‘‘either’’ category), vary consistently by even and odd desired
family size. At every desired family size, those desiring an even number of
children seem to be less likely to report son preference than those desiring an
odd number of children. Further, in most cases, a majority of those desiring an
even number of children are likely to report being indifferent to, or wanting
equal numbers of, girls and boys.
123
Explaining son preference in rural India 13
Table 1 Gender preferences by family size preferences: rural Indian women, 1992–1993
123
14 R. P. Pande, N. M. Astone
Multivariate results
We present three models for the sample as a whole to address our central
question (Table 3). Model 1 presents the effects of socioeconomic charac-
teristics of sample women and their households on son preference. Model 2
adds in the effects of variables that measure village opportunities for women
in terms of employment and education, as well as village economic develop-
ment. Model 3, the final model, adds variables that measure structural or
social factors and norms, including 25 state dummies. This final model allows
us to examine the independent influence of individual, community and mac-
rosocial factors on son preference. We then run model 3 for subgroups of
sample women as defined earlier.
Most individual and household characteristics are strongly associated with son
preference when contextual factors are not controlled (model 1, Table 3).
Women who work but don’t earn, who are educated, who have regular access
to media and cinema, and who live in wealthy households are likely to have
weaker son preference than other women. As hypothesized, women from
landholding or extended households have higher son preference than those
from landless or nuclear families. Any education works to diminish son
preference. At the same time, higher levels of education are associated with
123
Table 2 Mean value for explanatory variables (weighted), for all and by family size preference: rural Indian women, 1992–1993
Who want even Who want All women Who want Who want odd
numbers of only two only three numbers of
children children children children
123
Table 2 continued
16
123
Who want even Who want All women Who want Who want odd
numbers of only two only three numbers of
children children children children
Works but doesn’t earn cash –0.229 0.000 –0.065 0.013 0.006 0.819
Earns cash –0.019 0.576 0.104 0.000 0.049 0.102
Listens to radio or TV weekly (no) –0.223 0.000 –0.173 0.000 –0.169 0.000
Goes to cinema monthly (no) –0.582 0.000 –0.490 0.000 –0.219 0.000
Household wealth (poorest)
Second quintile –0.030 0.357 0.022 0.531 0.041 0.243
Middle quintile –0.075 0.020 0.021 0.544 0.009 0.786
Fourth quintile –0.142 0.000 0.017 0.663 –0.006 0.871
Richest quintile –0.141 0.000 –0.002 0.970 –0.108 0.025
Household owns land (no) 0.179 0.000 0.049 0.050 0.020 0.407
Lives in extended household (no) 0.102 0.000 0.044 0.058 0.015 0.525
Village opportunities & development
Proportion of literate women –1.468 0.000 –0.529 0.000
Proportion of women working –0.876 0.000 –0.131 0.147
At least a middle school (no) 0.011 0.722 0.003 0.911
All-weather road, bus station or rail station (none) –0.009 0.782 –0.028 0.345
Number of general facilities (none) –0.015 0.036 –0.012 0.164
Social factors & norms
Territorial endogamy 0.039 0.768
Kinship endogamy –0.073 0.586
Proportion rice in foodgrains –0.157 0.331
17
123
Table 3 continued
18
123
Individual & household Village opportunities & Social factors and
factors development norms
significantly weaker son preference than are lower levels of education, pos-
sibly because higher levels of education provide a woman with more avenues
to increase her social and economic standing by means other than being a
mother of sons.
In model 2 we introduce characteristics of village opportunities and
development. While the effects of education, work, and media access remain
significant, household wealth is no longer important. Overall village devel-
opment in terms of facilities available in the village is inversely related with
son preference, as hypothesized. Village-level women’s employment and
education are particularly important and significantly related to lower son
preference.
In model 3, the complete model, we introduce variables to measure social
and cultural norms. Overall, this model suggests that women’s opportunities
and characteristics as measured by education and access to media are critical,
at both individual and community levels. Women’s work and household
wealth, however, are not important, and neither is village development.
Among variables measuring social norms, only caste and religion significantly
influence son preference.
Independent of all other factors, women’s education at individual and
village levels continues to be strongly associated with weaker son preference.
The effect of secondary education here is particularly noteworthy, and women
with secondary or higher education have significantly weaker son preference
than women with lower levels of education. In fact, the coefficient on this
variable is the largest of any variable in the model. This strong result supports
the view in the literature that higher levels of education change women’s view
of female worth (Malhotra et al., 2003). Similarly, regular access to media and
cinema remain significantly associated with lower son preference, suggesting
that access to ‘‘modern’’ information and ways of life can influence gender
preferences independent of other individual or structural factors. On the other
hand, women’s employment, at the individual or community level, is not
significant. This may reflect the study’s context of rural areas with low female
education and thus limited professional skills, wherein employment may be
more associated with economic necessity rather than signifying greater gender
equality as such.
Household wealth variables are also not significant. The effect of land
ownership becomes small and statistically nonsignificant, and only women in
the wealthiest quintile of households have significantly weaker son preference
than the poorest, though the coefficient is small. One interpretation of this
finding is that when land ownership is controlled, women in wealthier
households are less conservative in their gender preferences than those in the
poorest households. Another interpretation is that a woman has to be in an
extremely wealthy household before she has alternatives to sons as economic
assets.
Seventeen of the 25 state dummies are statistically significant, with 13 of
them associated with higher son preference. Though these dummies cannot be
substantively interpreted, this pattern is consistent with the existence of
123
20 R. P. Pande, N. M. Astone
factors at the state level that determine son preference, above and beyond
what can be measured by the variables included in this analysis.
In this full model, neither the endogamy variables nor the agrarian ecology
variables remain significant. In other words, the factors that have historically
been thought to mediate regional differences in son preference (and gender
equity more generally) are not found to be influential. This finding is consis-
tent with recent research suggesting that broad regional cultural and economic
patterns of endogamy or agricultural production are becoming less important
determinants of gender inequality—and thus, presumably, son prefer-
ence—than are more local cultural and social norms (Rahman & Rao, 2004).
On the other hand, caste and religion remain strong determinants of son
preference, regardless of which state the respondent may reside in, suggesting
that these social characteristics continue to influence son preference. Contrary
to our hypothesis, lower caste women have slightly higher son preference than
high castes. This may be related to the process of Sanskritization, whereby
lower castes try to emulate the higher castes by adopting customs that con-
tribute to confining women to the home (Srinivas, 1976). Tribal women, on the
other hand, have significantly lower son preference than reference women.
The religion coefficients suggest that women from Muslim households have
slightly higher son preference, and women from other, non-Hindu,
non-Muslim religions, have slightly lower son preference than do Hindu
women.
Comparing final models for all women and women with different family
size preferences
The effects of different individual and structural factors on son preference for
those who have revealed son preference (odd ideal family size) and those who
have unrevealed preferences (even ideal family size) are, on the whole, rel-
atively similar to the full sample (Table 4). Among individual characteristics,
the effects of women’s education remain strong, as does the greater influence
of post-primary education in reducing son preference. Similarly, exposure to
weekly media continues to be significant. Among the variables measuring
social norms, the weaker son preference among tribal women remains sig-
nificant across all three groups.
Women who want particularly small families, on the other hand, differ from
the full sample in some ways (Table 5; models 6 and 7 are logit models, since
the third category for the ordered logit dependent variable was not defined for
these subsamples). For instance, the effects of media and religion vary across
subgroups. The rice production coefficient, in addition, is unstable across
model specifications.
Education and the effect of tribe are consistent across models. In all the
models in Tables 4 and 5, individual women’s education (though not village
female literacy) remains a strong predictor of son preference, especially
middle or higher education. Similarly, in all the models, women from tribal
households have weaker son preference than high/other caste women.
123
Table 4 Determinants of son preference among rural Indian women, 1992–1993: by odd and even desired family sizes
Final model for all Women with odd Women with even
rural women desired family size desired family size
Works but doesn’t earn cash 0.006 0.819 –0.032 0.543 0.037 0.480
Earns cash 0.049 0.102 –0.056 0.299 0.146 0.009
Listens to radio or TV weekly (no) –0.169 0.000 –0.136 0.022 –0.237 0.000
Goes to cinema monthly (no) –0.219 0.000 –0.058 0.448 –0.279 0.004
Household wealth (poorest)
Second quintile 0.041 0.243 0.021 0.770 0.077 0.186
Middle quintile 0.009 0.786 0.014 0.836 0.034 0.556
Fourth quintile –0.006 0.871 0.023 0.732 –0.028 0.659
Richest quintile –0.108 0.025 –0.167 0.028 –0.107 0.169
Household owns land (no) 0.020 0.407 0.054 0.247 0.001 0.983
Lives in extended household (no) 0.015 0.525 0.040 0.334 0.034 0.366
Village opportunities & development
Proportion of literate women –0.529 0.000 –0.262 0.111 –0.751 0.000
Proportion of women working –0.131 0.147 0.048 0.737 –0.141 0.356
At least a middle school (no) 0.003 0.911 0.025 0.638 –0.017 0.738
All-weather road, bus station or rail station (none) –0.028 0.345 –0.070 0.162 –0.026 0.638
Number of general facilities (none) –0.012 0.164 –0.013 0.316 –0.029 0.029
Social factors & norms
Territorial endogamy 0.039 0.768 –0.213 0.318 0.079 0.705
Kinship endogamy –0.073 0.586 0.298 0.205 –0.037 0.887
Proportion rice in foodgrains –0.157 0.331 –1.279 0.000 –0.311 0.280
21
123
Table 4 continued
22
123
Final model for all Women with odd Women with even
rural women desired family size desired family size
Final model for all Women who ideally Women who ideally
rural women want two children want three children
Works but doesn’t earn cash 0.006 0.819 0.006 0.945 –0.039 0.585
Earns cash 0.049 0.102 0.269 0.002 –0.064 0.450
Listens to radio or TV weekly (no) –0.169 0.000 –0.263 0.000 –0.080 0.300
Goes to cinema monthly (no) –0.219 0.000 –0.155 0.228 –0.062 0.547
Household wealth (poorest)
Second quintile 0.041 0.243 –0.060 0.611 0.041 0.657
Middle quintile 0.009 0.786 0.035 0.763 0.033 0.723
Fourth quintile –0.006 0.871 0.006 0.961 0.062 0.494
Richest quintile –0.108 0.025 –0.023 0.863 –0.128 0.204
Household owns land (no) 0.020 0.407 0.007 0.919 0.068 0.248
Lives in extended household (no) 0.015 0.525 0.063 0.268 0.067 0.226
Village opportunities & development
Proportion of literate women –0.529 0.000 –0.333 0.147 –0.242 0.341
Proportion of women working –0.131 0.147 0.458 0.046 –0.019 0.932
At least a middle school (no) 0.003 0.911 0.024 0.770 0.064 0.440
All-weather road, bus station or rail station (none) –0.028 0.345 0.091 0.256 –0.119 0.113
Number of general facilities (none) –0.012 0.164 –0.048 0.019 0.014 0.427
Social factors & norms
Territorial endogamy 0.039 0.768 0.016 0.968 –0.318 0.186
Kinship endogamy –0.073 0.586 0.024 0.959 0.250 0.412
Proportion rice in foodgrains –0.157 0.331 –1.682 0.000 –2.411 0.000
23
123
Table 5 continued
24
123
Final model for all Women who ideally Women who ideally
rural women want two children want three children
The variation in results between the sample as a whole and the different
subsamples examined here is difficult to interpret as it is likely to be a com-
bination of differences in sample sizes, differences in the extent of revealed
son preference and ideal desired family sizes between groups, and systematic
differences in sample characteristics for the subsamples that want particularly
small families. At a minimum, however, this analysis suggests that it is
important to take account of ideal family size preferences in measuring ideal
gender preferences. In particular, women who go against the norm in wanting
small families may be different from other women in ways that warrant further
exploration when examining gender preferences.
Discussion
In this paper we have tested the hypothesis that a number of factors that have
long been associated, at the level of individual characteristics or social norms,
with demographic phenomena thought to reflect son preference, are also
important predictors of son preference when measured directly. We have
analyzed the independent, relative effects of individual and household char-
acteristics, village opportunities and development, and social norms and
structures. Finally, we have examined whether the predictors of son prefer-
ence differ by desired family size. The data we use, while not exhaustive,
provide a unique opportunity to examine these issues quantitatively.
Our study is informed by a theoretical orientation whose proponents see
gender stratification as emerging from the simultaneous influence of factors at
many levels: a woman’s own experience in her household, the characteristics
of the people with whom she lives and the household in which she resides, the
local community, and macrosocietal norms (Malhotra & Schuler, 2005). As
such, our results are consistent with this view of gender and how gender
preferences are formed: while an individual woman’s and her household’s
characteristics do influence her gender preferences, social norms are impor-
tant. In other words, it is not enough in an analysis of son preference to focus
on an individual woman’s characteristics alone or structural characteristics
alone; rather, we need to examine both simultaneously.
At the same time, our paper demonstrates an enormous influence of
women’s education, particularly education beyond primary schooling,
regardless of desired family size. Given the multivariate nature of this anal-
ysis, we can conclude that there is something in the nature of women’s
education—particularly higher education—that significantly weakens son
preference, beyond any role it may play in allowing women to be employed or
have access to media, beyond its association with higher socioeconomic status,
and net of social norms. This finding is consistent with recent research on
other aspects of gender inequality that shows a much stronger effect of
post-primary education, compared to primary education, in improving gender
equality and women’s lives (Malhotra et al., 2003).
123
26 R. P. Pande, N. M. Astone
123
Explaining son preference in rural India 27
Acknowledgments This paper is based on the first author’s doctoral thesis while at the Johns
Hopkins Bloomberg School of Public Health. Support for this research was provided by the
Hewlett Foundation, The Population Council, and the Mellon Foundation. The authors thank
Drs. Ken Hill, Anju Malhotra, and Kathryn Yount for their comments.
References
Agarwal, B. (1994). A field of one’s own: Gender and land rights in South Asia. Cambridge,
England: University Press.
Arnold, F., Kishor, S., & Roy, T. K. (2002). Sex selective abortions in India. Population and
Development Review, 28(4), 739–785.
Arnold, F., Minja, K. C., & Roy, T. K. (1998). Son preference, the family-building process and
child mortality in India. Population Studies, 52(3), 301–315.
Bairagi, R. (2001). Effects of sex preference on contraceptive use, abortion and fertility in Matlab,
Bangladesh. International Family Planning Perspectives, 27(3), 137–143.
Baltiwala, S. (1994). The meaning of women’s empowerment: New concepts from action. In
G. Sen, A. Germain, & L. C. Chen (Eds.), Population policies reconsidered: Health,
empowerment and rights (pp. 127–138). Cambridge MA: Harvard University Press.
Banister, J. (2003). Shortage of girls in China today: Causes, consequences, international com-
parisons, and solutions. Beijing Javelin Investments.
Banthia, J. K. (2001). Provisional population totals: India. Census of India 2001, Series 1, India,
Paper 1 of 2001. New Delhi, India: Office of the Registrar General.
Bardhan, P. K. (1974). On life and death questions. Economic and Political Weekly, Special
Number, 1293–1304.
Barua, A., & Kurz, K. (2001). Reproductive health-seeking by married adolescent girls in Ma-
harashtra India. Reproductive Health Matters, 9(17), 53–62.
Basu, A. M. (1989). Is discrimination in food really necessary for explaining sex differentials in
childhood mortality? Population Studies, 43(2), 193–210.
Basu, A. M., & Basu, K. (1991). Women’s economic roles and child survival: The case of India.
Health Transition Review, 1(1), 83–103.
Belanger, D. (2002). Son preference in a rural village in North Vietnam. Studies in Family
Planning, 33(4), 321–334.
Bhuiya, A., & Streatfield, K. (1991). Mothers’ education and survival of female children in a rural
area of Bangladesh. Population Studies, 45, 253–264.
Bloomfield, M. (1897). Sacred books of the East, Vol. 42. Retrieved from http://www.sacred-
texts.com/.
Bourne, K. L., & Walker, G. M. (1991). The differential effect of mothers’ education on mortality
of boys and girls in India. Population Studies, 45(2), 203–219.
Cain, M. T. (1988). The material consequences of reproductive failure in rural south Asia. In
D. Dwyer & J. Bruce (Eds.), A home divided: Women and income in the Third World
(pp. 20–38). Stanford CA: Stanford University Press.
Cain, M. T. (1993). Patriarchal structure and demographic change. In N. Federici, K. O. Mason, &
S. Sogner (Eds.), Women’s position and demographic change (pp. 43–60). Oxford, England:
Clarendon Press.
Chan, A., & Yeoh, B. (2002). Gender, family and fertility in Asia: An introduction. Asia-Pacific
Population Journal, 17(2), 5–10.
Clark, S. (2000). Son preference and sex composition of children: Evidence from India. Demo-
graphy, 37(1), 95–108.
Das Gupta, M. (1987). Selective discrimination against female children in rural Punjab, India.
Population and Development Review, 13(1), 77–100.
123
28 R. P. Pande, N. M. Astone
Das Gupta, M. (1995). Life course perspectives on women’s autonomy and health outcomes.
American Anthropologist, 97(3), 481–491.
Das Gupta, M., & Bhat, P. N. (1997). Fertility decline and increased manifestation of sex bias in
India. Population Studies, 51(3), 307–315.
Das Gupta, M., & Shuzhuo, L. (1999). Gender bias in China, South Korea and India 1920–1990:
Effects of war, famine and fertility decline. Development and Change, 30(3), 619–652.
Dharmalingam, A. (1996). The social context of family size preferences and fertility behaviour in a
south Indian village. Genus, 52(1–2), 83–103.
Dyson, T., & Moore, M. (1983). On kinship structure, female autonomy, and demographic
behavior in India. Population and Development Review, 9(1), 35–60.
Filmer, D., & Pritchett, L. H. (2001). Estimating wealth effects without expenditure data—or tears:
An application to educational enrollments in states of India. Demography, 38(1), 115–132.
Government of India (1995). Economic survey 1994–1995. New Delhi, India: Ministry of Finance,
Economic Division.
Govindaswamy, P., & Ramesh, B. M. (1996). Maternal education and gender bias in child care
practices in India. Paper presented at the Population Association of America, New Orleans
LA, 9–11 May 1996.
Griffiths, P., Matthews, Z., & Hinde, A. (2000). Understanding the sex ratio in India: A simulation
approach. Demography, 37(4), 477–488.
International Institute for Population Sciences [IIPS] (1995). National family health survey (MCH
and family planning) India 1992–93. Bombay, India: IIPS.
Jejeebhoy, S. J. (1996). Women’s autonomy and reproductive behaviour in India: Linkages and
influence of sociocultural context. Paper presented at the Seminar on Comparative Perspectives
on Fertility Transition in South Asia, Islamabad-Rawalpindi, Pakistan, 17–20 December 1996.
Karve, I. (1965). Kinship organization in India. Bombay, India: Asia Publishing House.
Kennedy, P. (1996). A guide to econometrics, 3rd edn. Cambridge MA: The MIT Press.
Kishor, S. (1993). May God give sons to all: Gender and child mortality in India. American
Sociological Review, 58(2), 247–265.
Kishor, S. (1995). Gender differentials in child mortality: A review of the evidence. In M. Das
Gupta, L. C. Chen, & T. N. Krishnan (Eds.), Women’s health in India: Risk and vulnerability
(pp. 19–54). Bombay, India: Oxford University Press.
Leone, T., Matthews, Z., & Zuanna, G. D. (2003). Impact and determinants of sex preference in
Nepal. International Family Planning Perspectives, 29(2), 69–75.
Liang, K., & Zeger, S. L. (1993). Regression analysis for correlated data. Annual Review of Public
Health, 14, 43–68.
Makinson, C. (1994). Discrimination against the female child. International Journal of Gynecology
and Obstetrics, 46(2), 119–125.
Malhotra, A., Pande, R., & Grown, C. (2003). Impact of investments in female education on gender
equality. International Center for Research on Women.
Malhotra, A., & Schuler, S. (2005). Women’s empowerment as a variable in international
development. In D. Narayan (Ed.), Measuring empowerment: Cross-disciplinary perspectives
(pp. 71–88). Washington, DC: World Bank.
Mandelbaum, D. G. (1988). Women’s seclusion and men’s honor. Sex roles in North India,
Bangladesh and Pakistan. Tucson AZ: The University of Arizona Press.
Mayer, P. (1999). India’s falling sex ratios. Population and Development Review, 25(2), 323–344.
McCullagh, P., & Nelder, J. A. (1989). Generalized linear models. Monographs on statistics and
applied probability 37. New York: Chapman and Hill.
Miller, B. D. (1981). The endangered sex: Neglect of female children in rural North India. Ithaca
NY: Cornell University Press.
Miller, B. D. (1989). Changing patterns of juvenile sex ratios in rural India, 1961 to 1971.
Economic and Political Weekly, 24(2), 1229–1236.
Mishra, V., Roy, T. K., & Retherford, R. D. (2004). Sex differentials in childhood feeding, health
care and nutritional status in India. Population and Development Review, 30(2), 269–296.
Morgan, S. P., & Niraula, B. B. (1995). Gender inequality and fertility in two Nepali villages.
Population and Development Review, 21(3), 541–561.
Muhuri, P. K., & Preston, S. H. (1991). Effects of family composition on mortality differentials by sex
among children in Matlab, Bangladesh. Population and Development Review, 17(3), 415–434.
123
Explaining son preference in rural India 29
Murphy, R. (2003). Fertility and distorted sex ratios in a rural Chinese country: Culture, state and
policy. Population and Development Review, 29(4), 595–626.
Murthi, M., Guio, A., & Dreze, J. (1995). Mortality, fertility and gender bias in India: A district-
level analysis. Population and Development Review, 21(4), 745–781.
Murthy, R. K. (1996). Fighting female infanticide by working with midwives: An Indian case
study. Gender and Development: Women and the Family, 4(2), 20–27.
Oldenburg, P. (1992). Sex ratio, son preference and violence in India: A research note. Economic
and Political Weekly, 27(49 & 50), 2657–2662.
Oomman, N., & Ganatra, B. R. (2002). Sex selection: The systematic elimination of girls.
Reproductive Health Matters, 10(19), 184–197.
Pande, R. P. (1999). Grant a girl elsewhere, here grant a boy: Gender and health outcomes among
rural Indian children. Unpublished doctoral dissertation, School of Hygiene and Public
Health, Johns Hopkins University, Baltimore, MD.
Pande, R. (2003). Selective gender differences in childhood nutrition and immunization in rural
India: The role of siblings. Demography, 40(3), 395–418.
Pande, R. P., & Yazbeck, A. (2003). What’s in a country average? Wealth, gender and regional
inequalities in immunization in India. Social Science and Medicine, 57, 2075–2088.
Park, C. B., & Cho, N. (1995). Consequences of son preference in low-fertility society: Imbalance
of the sex ratio at birth in Korea. Population and Development Review, 21(1), 59–84.
Rahman, L., & Rao, V. (2004). The determinants of gender equity in India: Examining Dyson and
Moore’s thesis with new data. Population and Development Review, 30(2), 239–268.
Srinivas, M. N. (1976). The remembered village. New Delhi, India: Oxford University Press.
Vlassoff, C. (1990). The value of sons in an Indian village: How widows see it. Population Studies,
44(1), 5–20.
Yount, K. M. (2001). Excess mortality of girls in the Middle East in the 1970s and 1980s: Patterns,
correlates, and gaps in research. Population Studies, 55(3), 291–308.
Yount, K. M., Langsten, R., & Hill, K. (2000). The effect of gender preference on contraceptive
use and fertility in rural Egypt. Studies in Family Planning, 31(4), 290–300.
123