gupta2006

Journal of Development Studies,
Vol. 42, No. 1, 110–138, January 2006
Fertility and the Household’s Economic

Status: A Natural Experiment using
Indian Micro Data
NABANITA DATTA GUPTA* & AMARESH DUBEY**
*The Danish National Institute of Social Research, Copenhagen, Denmark, **North-Eastern Hill
University, Shillong, India and NCAER, Delhi, India
Final version received January 2005
ABSTRACT We model fertility as endogenous to the family’s economic status because poor
households choose to have large families in the absence of adequate social insurance. Because of a
strong son preference in India, having two girls first can proxy an exogenous increase in fertility,
and is therefore a good instrument for fertility in determining poverty of rural households. The
1993–1994 Indian Quinquennial Survey data shows that even though poverty rates are
comparable, 74 per cent of two-girl families have a third child compared to 63 per cent of
other families. Fertility significantly positively affects poverty when treated as exogenous, but
vanishes once endogenised. These results are robust to omitting states with skewed sex ratios and
to proxying economic status by expenditures.
I. Introduction
This paper attempts to uncover the causal effect of fertility on the economic status of
rural nucleus households in India. Our hypothesis is that family size and economic
status are jointly determined, so that fertility cannot be treated as purely exogeneous
to the household’s poverty status or expenditure level. Rather, poor households in
developing countries choose to have large families because children are considered
current and future economic assets to the family in the absence of adequate public or
privately provided social insurance.
Much evidence indicates that poor families often have a high ratio of dependent
members, mostly children. Previous research in the area has shown, for example,
that that the number of children per family is significantly correlated with a family’s
poverty level, that is, households that fall below the poverty line also tend to be large
relative to the average family. For example, in Brazil, Fishlow (1972) reported that
29 per cent of all families had a size of six or more individuals, and over half of such
Correspondence Address: Nabanita Datta Gupta, The Danish National Institute of Social Research,
Herluf Trolles Gade 11, DK-1052 Copenhagen K, Denmark. Tel: þ 45 33 48 09 85; Fax: þ 45 33 48 08 33;
Email: ndg@sfi.dk
ISSN 0022-0388 Print/1743-9140 Online/06/042110-29 ª 2006 Taylor & Francis

DOI: 10.1080/00220380500356779
India: Fertility and Household Economic Status 111
families fell below the poverty line. Similarly, for Malaysia, Anand (1977) noted that
the incidence of poverty rises with family size, ranging from 24 per cent in a
household of one to 46 per cent in households with ten or more people. The World
Development Report (World Bank, 1990) observed that in Pakistan in 1984, the
poorest 10 per cent of households had an average of 7.7 members; the corresponding
national average was 6.1. In a micro-level study of poverty in India, Dubey,
Gangopadhyay and Wadhwa (1999) find a significantly positive effect of household
size on the incidence of poverty. As dependent members are most often children, the
findings on the correlation of poverty and household size indicate that the burden of
poverty falls disproportionately on children, affecting childhood nutrition and
education, perpetuating the institution of child labour and eventually passing on the
legacy of poverty to future generations.
Clearly, family size may be both a cause of poverty as well as an effect. Larger
families, especially those with larger numbers of children, are likely to have a lower
per capita income simply because of the high dependency ratio. Child labour may
mitigate this effect somewhat, but children are not paid much. Other effects of family
size on poverty may work through lowered labour supply and human capital
investments of the mother and resulting effects on family income. More significantly,
poverty may actually feed on itself by creating the incentive to have a large number
of children. Poor households in developing countries may be more likely to invest in
children as a source of support in old age, to compensate for missing markets in life
insurance and social security. Infant mortality is another factor, in that families that
expect higher mortality rates (such as the poor) may choose to have more children to
compensate for this possibility (see, for example, Ben-Porath (1976) on the
‘hoarding’ and ‘replacement’ responses to child mortality).1 Also, poorer families
may be slower to respond to a fall in child death rates, as they base their expectations
of infant mortality rates on the experiences of their parents and others in their
parent’s generation, that is, there could be imperfect information. Poor families may
also tend to have more children because they are more likely to be located in poorer
regions where the probability is greater that a single child will not earn enough in
adulthood to support his parents.
In terms of the costs of raising and bearing children, any factor that raises the
costs (both direct and indirect) of having children will lead to a decrease in fertility
cet. par. For example, a rise in women’s wages generally tends to reduce fertility
because it increases the opportunity costs of having children. On the one hand, poor
families are more likely to have higher labour force participation of their female
members because they are more dependent on their income. This will raise the
opportunity costs of having children. On the other hand, poor women who
participate in the labour market generally face lower wage rates than rich women
because of their lower amount and quality of educational and other human capital
investments. In terms of direct costs, rich families would tend to invest more in the
education of children according to the quantity–quality trade-off models, which
would then raise the direct costs of an additional child, see for example Becker and
Tomes (1976).
The above discussion suggests that fertility and economic status may be jointly
determined, and that there is a need to separate these effects in order to identify the
‘causal’ effects of family size on a household’s economic situation. Rosenzweig and
112 N. D. Gupta & A. Dubey
Evenson (1977) develop a household time-allocation model, which explicitly

accounts for joint family decisions regarding fertility and the allocation of children’s
time to schooling and work activities. They apply their model to aggregate district-
level data pertaining to the rural population of India and find that fertility among
Indian families in the 1950s was largely motivated due to the high return to
children’s labour outside the home, compared to investments in skills obtained in
schools.
In recent times, the availability of micro-data pertaining to characteristics of
households in developing countries allows a more direct examination of family
fertility decisions and their effect on a household’s living standards. While not
developing a fully structural model such as in Rosenzweig and Evanson (1977), this
paper presents reduced form analyses of the joint determination of family size and
economic status using econometric methods that account for the endogeneity of the
fertility decision in the determination of a household’s living standard. The basic
hypothesis is that fertility is intrinsically linked with poverty as children are
considered current and future economic assets to the family in developing
countries. Using the gender of the first two children as a natural experiment, we
estimate the causal effect of fertility on poverty of rural nucleus households in India
on a sample drawn from the 1993–94 Indian Quinquennial Survey (5th wave) on
consumer expenditure. Our findings indicate that fertility has a significant positive
effect on poverty when fertility is modelled as being exogeneous to the household’s
economic status. However, once endogenised, there is evidence of significant
correlation in the two outcomes, but its effect on the household’s economic status
vanishes. These results are robust to the exclusion of states with skewed sex ratios
and to modelling the family’s economic status using a linear expenditure
formulation.
The rest of the paper is organised as follows: in the next section we present
findings and statistics on poverty and fertility in India; then, we outline the natural
experiment and set up a simple theoretical model of son preference. In the following
sections, we set up the empirical model, describe the data and related issues, present
our estimation results and finally, offer some conclusions.
II. Poverty and Fertility in India

There is a huge literature on poverty in India, but it is concerned mainly with
temporal change and spatial variation. Further, the characteristics of the poor have
been studied, mostly, at aggregate levels. Some of the key findings from this
literature are that, poor households tend to be headed by illiterates and that
traditionally socially excluded groups are relatively poorer than other groups.
Several studies report that poor in general have larger family sizes.
Several researchers (Ahluwalia, 1978; Saith, 1981) have argued that rural poverty
in India can be explained largely by agricultural production and prices. Another
group of researchers (Bardhan, 1984; Gaiha, 1988; Dubey et al., 1999; Dubey,
Gangopadhyay and Wadhwa, 2001) have argued that underlying causal mechanisms
of poverty and underdevelopment can be better understood using an analytical
framework anchored in models of individual and household behaviour. For
example, Gaiha (1988) uses data from the Additional Rural Income survey carried
out by the National Council of Applied Economic Research (1968–70)2 and treats
poverty of the household as a random event that is related to village-specific,
technological and household-specific variables. In the village-specific characteristics,
he includes availability of transport, medical facilities, credit availability and
registered small-scale factories as proxy for non-agricultural employment. Among
the technological variables, Gaiha (1988) considered use of tractor, use of high
yielding variety of seeds and whether the village has electricity supply or not. The
household specific variables are demographic, the dependency burden (number of
children 414 years divided by household size) and age of the head of the household.
Among other studies, Dubey et al. (1999) try to measure the vulnerability of
female-headed households to poverty using the household level consumption
expenditure data collected by the Indian National Sample Survey Organisation in
1987–88. They use a dummy for female-headed household, years of schooling of
head of the household and household size as explanatory variables to show that
schooling of the head of the household has a significant effect on poverty in the
urban sector.
In another study, Dubey et al. (2001) use household level consumption
expenditure data collected by the Indian National Sample Survey Organisation for
two years, 1987–88 and 1993, to measure the effect of city size on poverty incidence.
They use a dummy for city size and household size and years of schooling of the
head of the household to calculate the probability of a household being poor.
All three micro studies reported here show that household size has a positive and
significant effect on the probability of a household being poor. Given that the size of
household is correlated with poverty and that the poor have larger family sizes, it is
important to examine to what extent taking into account the endogeneity of fertility
on the likelihood of a household being poor, changes the size and sign of these
reported effects.
Most population figures on India presented in the international media tend to
focus on statistics on the sheer size of the total population (about one billion
currently, and which by all projections will exceed that of China’s shortly) and the
dramatic rise in population density in the decade of the 1980s. In part, such rapid
growth is an indirect result of the decline in India’s mortality rate that has fallen
from 25 per thousand in the early 1950s to about nine per thousand in the late 1990s,
see Table 1. At the same time, analysts within India point out that there has been a
significant decline in fertility over the past five decades. This decline is attributed
more to the results of family planning efforts and successful (at times coercive) birth
control than to economic and social change (World Resources 1994–95).
National average figures, for example in Table 1, show that the fertility rate has
nearly halved since the 1950s from six children per woman to 3.2. These aggregate
figures disguise vast demographic differences among India’s various states. In fact,
the fertility transition (from high to low fertility) is nearly complete in some states,
notably Kerala and Tamil Nadu in the south, where the average fertility rate is two
children. Still, the current national average of 3.2 reported below is above the
replacement level.
Despite the significant decline in the fertility rate, families in the lower income
groups have higher fertility than average as indicated by the higher household size
among poor households. Based on the NSS Indian quinquennial sample data (43rd
and 50th rounds), we present evidence in Table 2 below that the number of children
per household declines as we go along the expenditure scale. The expenditure scale is
constructed by ranking households by their monthly per-capita total expenditures on
food and non-food items (current prices).
For each decile above, we calculate the average number of children per household.
Note that this average is lower than the average that we obtain from the rural
nucleus family households that we use in estimation of the model. The reason is that
when selecting the relevant sample households for the econometric analysis, by
construction, we leave out about 25 per cent of households with zero or one child.
The average presented in Table 2 is based on the entire sample of rural households,
Table 1. Selected health indicators for India
Measure 1951 1981 1991 Current
Crude birth rate (per 1,000 population) 40.8 33.9 29.5 26.1*
Crude death rate (per 1,000 population) 25.1 12.5 9.8 8.7*
Total fertility rate (average per women) 6.0 4.5 3.6 3.2**
Infant mortality rate (per thousand live birth) 146 110 80 70*
Child mortality rate (1–4 yrs: per 1,000 children) 57.3 41.2 26.5 22.5**
Life expectancy at birth (Years)
Male 37.2 54.1 59.7 60.4***
Female 36.2 54.7 60.9 61.8
Note: *denotes that figures are for 1999; **denotes that figures relate to 1998; ***denotes the
average figures for 1993–97.
Source: Economic Survey 2001–2002, Ministry of Finance, Government of India, New Delhi,
2002.
Table 2. Number of children per household in rural areas
1987–88 1993–94
Average no of Average no of
children per Standard children per Standard
Expenditure decile household deviation household deviation
0–10 3.37 1.74 3.20 1.71

10–20 3.16 1.79 2.92 1.68
20–30 3.03 1.80 2.79 1.81
30–40 2.86 1.74 2.59 1.69
40–50 2.79 1.82 2.54 1.75
50–60 2.69 1.94 2.31 1.70
60–70 2.49 1.79 2.18 1.71
70–80 2.38 1.85 2.05 1.73
80–90 2.19 1.84 1.89 1.67
90–100 1.90 1.77 1.69 1.70
All 2.69 1.86 2.42 1.77
Percent of HH with
42 children 0.3795 0.3217
N ¼ 82,661 (69,235) rural nucleus family households in 1987/1988 (1993/1994).

82,661 in 1987–88 and 69,235 in 1993–94. Evidence of the decline in fertility is seen in
the last row of the table, where the percentage of households with more than two
children has declined from 38 per cent to 32 per cent in this period.
At the same time, the number of people living below the poverty line in India has
decreased from about 52 per cent of the population in 1972–73 to about 30 per cent
of the population in 1987–88 (estimates appearing in the Statistical Outline of
India, 1992–93). Dubey and Gangopadhyay (1998) use the expenditure surveys of
the NSS (43rd and 50th rounds) and estimate that the overall poverty rate fell
from 47 per cent in 1987–88 to about 40 per cent in 1993–94. Both sets of estimates
define poverty in relation to the expenditure required for a daily calorie intake of
about 2,400 per person in rural areas and 2,100 in urban areas (that is, the official
poverty line). The NSS estimates however, are corrected for variations in
commodity prices across different states and union territories. While these figures
are based on simple head count ratios, Dubey and Gangopadhyay (1998) conclude
that a 6–7 percentage-point decline in the poverty rate between the 1987–88 and
1993–94 period is evidenced irrespective of the actual poverty line or the statistical
measure of poverty (head count ratio vs. poverty gap index vs. FGT measures)
being used.
Several macro factors could have contributed to this observed decline in poverty
in the 1987/88–1993/94 period, such as the impact of financial liberalisation and
economic reforms that were started in 1991 and the continuing growth in food
production in the nation’s agricultural sector with the use of high-yield varieties of
grain and the extension of rural credit facilities. At the same time, it is acknowledged
that the problem of poverty in India is closely linked to the reproductive status of
women and that a critical component of the transition to lower fertility and poverty
rates will be greater investments in women’s health and education. Yet, reliable
estimates of the magnitude of the impact of fertility on poverty are largely lacking in
the literature so far. In this paper, we attempt to uncover the causal impact of an
increase in fertility on rural poverty in India. In the next section, we outline the
‘natural experiment’ methodology.
III. Methodology
One approach that has been adopted in the labour supply literature to isolate the
exogenous effect of fertility on labour supply is the use of instrumental variables (IV)
estimation strategies based on the sibling sex-mix in families with two or more
children, see for example, Angrist and Evans (1998). In developed countries, parents
typically exhibit a preference for mixed sibling sex-composition. For example, Ben-
Porath and Welch (1976) find that in the 1970 Census, 56 per cent of families with
either two boys or two girls had a third birth, whereas only 51 per cent of families with
one boy and one girl had a third child. Angrist and Evans (1998) compare the 1980
and 1990 PUMS data and find that in 1980 (1990), among women with two or more
children, 43 per cent (41 per cent) had another birth if the first two children were of the
same sex, compared to 37 per cent (34 per cent) if the two children were of opposite
sex. Because sex-mix is virtually randomly assigned, a dummy for whether the sex of
the first child matches the sex of the second child provides a plausible instrument for
further childbearing among women with at least two children. In fact, IV estimates of
fertility on labour supply have been shown to be significant, but smaller than ordinary
least-squares estimation.3
In many developing countries, particularly in the band extending from North
Africa, through the Middle East and South Asia to East Asia, there is a strong
preference for male offspring over female offspring.4 This is because male children are
viewed as being better investments in terms of future income-producing ability in
societies where social insurance or old age income support is lacking.5 There may be
several reasons for this. First, in many societies, old-age support is exclusively the task
of male offspring by way of social practice and tradition. Even though female
offspring may be just as able to offer support, there may be a stigma associated with
receiving such support from daughters. Second, in societies where female employ-
ment is not in demand or undervalued, males may be potentially more productive
future ‘assets’. Particularly in rural areas, if women are typically not involved in
agriculture or do not have property rights to land, their income-earning ability will be
lower relative to men. Third, in societies where the practice of dowry flourishes, the
birth of a daughter is often associated with an expected future outflow of family
income.6
There are no a priori reasons for this preference to be stronger among the poor
rather than the rich. Several studies in fact suggest that the preference for sons could
be even stronger in the upper social strata. While it is true that the poor are in
general more dependent on income support from their children, poor females are
more likely to be engaged in economic activity on grounds of necessity and therefore
more likely to be viewed as contributing members of the household.7 The practice of
giving and receiving dowry is prevalent among all social classes. Sen (1985) and
Murthi, Guio and Dreze (1995) report that discrimination against girls rises with
income. Miller (1981) and Dasgupta (1987) find that the higher the education level of
the mothers, the greater in fact is discrimination against daughters. Further, several
high-caste tribes are known to practice female infanticide (Tambiah, 1973). Thus,
son preference seems to cut across income classes.
In the section below, we explore the theoretical underpinnings of such a preference
for sons.
IV. A Theoretical Model of Son Preference

We consider a simple, static model of the value to the parents of having a child, in a
society in which parents ‘invest’ in their children as a source of insurance and
support to the family in old age. These are rather different assumptions from the
standard quality/quantity model of fertility, originally developed by Becker and
Lewis (1973) and expanded later by Becker and Tomes (1976) in which children are
assumed to be consumption goods from which parents derive utility. In this model,
children are considered as pure investment goods, although a more altruistic
formulation would allow for the child’s current and future well-being to enter the
parental valuation function.
Assume that parents care about both current parental consumption as well as
future parental consumption after the cessation of work life. Assume also that there
is no social insurance or old age support available to individuals except for the
support they may get from their children.
Let Vc be the value associated with having a child. Thus, the value function is
given by sum of the discounted future net benefits stemming from the child’s
future employment (E*) and the child’s future dowry receipts net of marriage
costs (D*).8 These are the sources of parental consumption in old age. Children
also contribute to the household’s current consumption. Let the instantaneous
value of any current income earned outside the household by the child be given
by Yc. In terms of costs of having a child, there are costs incurred today of
investing in the child’s future earnings and marriage potential through the
allocation of resources to schooling (S), and other child-specific inputs (X).9,10 The
prices of these inputs are assumed to be invariant to the quantity of inputs
consumed. The parameter (is a weight that parents attach to earnings of their
child, and r is the interest rate.
Thus, the value function can be represented as:
Vc ¼ ½1=ð1 þ rÞððgE þ D Þ þ Yc Ps S Px X ð1Þ
Parents contribute both purchased inputs and as well as time inputs in the rearing of
children. For simplicity, we concentrate only on the financial inputs. A child’s
expected lifetime earnings will be a function of their innate ability (A), schooling (S),
a vector of child-specific inputs (X). The coefficient d captures the extent to which the
market undervalues work, that is, the difference between marginal productivity and
wage.
E ¼ E ðA; S; X; d Þ ð2Þ
We assume that dE*/dA 4 0, d2 E*/dA2 5 0, dE*/dS 4 0, d2E*/dS2 5 0, dE*/dX 4 0,

d2E*/dX2 5 0, dE*/dd 5 0.
The dowry component is also an increasing function of ability, schooling and
other inputs:11
D ¼ D ðA; S; XÞ: ð3Þ
Again, assume that there are positive and diminishing marginal products,
that is, dD*/dA 4 0, d2D*/dA2 5 0, dD*/dS 4 0, d2D*/dS2 5 0, dD*/dX 4 0, d2D*/
dX2 5 0.
Finally, child labour is also a function of non-schooling inputs:12
Yc ¼ Yc ðXÞ; ð4Þ
and we assume that dYc/dX 4 0, d2Yc/dX2 5 0. This relationship between non-

schooling inputs and child-labour may arise if nutrition for example, affects the
capacity to do productive work.
In terms of the model above, it is clear that a preference for a boy over a girl can
arise through several different channels, that is, Vb4Vg.
(1) First, parents may value future income earned by a son higher than future
earnings of a daughter. Such a preference could arise if traditionally old age
support has been the domain of boys and parents feel less willing to break with
tradition and accept support from girls.13 Thus:
gb > gg : ð5Þ
(2) Even if parents attach equal weight to earnings of a son compared to earnings
of a daughter, a girl may simply earn less than a boy as an adult, given the same
abilities, schooling and other inputs, if the returns to schooling and other inputs
are lower for women than men. This could arise if the market undervalues
female labour, that is, labour market discrimination against women:
Eb ðA; S; X; db Þ > Eg ðA; S; X; dg Þ; db < dg : ð6Þ
(3) Clearly, if a boy can command a bigger dowry than a girl, this will lead to
a larger relative increase in the family’s future consumption possibilities.14
That is:
Db > Dg : ð7Þ
(4) A difference in the relative value of a boy versus a girl can arise if a boy can earn
more by way of child labour than a girl, given the same inputs, that is:
Ycb ðXÞ > Ycg ðXÞ: ð8Þ
Swaminathan (1998) finds that gender and age are significant determinants of
earnings of working children in Western India. Nielsen and Dubey (2001) also
find that girls earn 5–6 per cent less than boys in weekly wages, based on wage
regressions carried out on samples of working children from the Indian NSS
data, 43rd and 50th rounds.
(5) Aside from the main effects described in (1)–(4), there could be feedback
effects arising from future earnings and future dowry gifts that affect current
resource allocation. Parents who expect lower relative earnings of a daughter
or a large financial outflow when a daughter gets married face reduced
incentives to invest in a daughter (Boserup, 1970; Rosenzweig and Schultz,
1982). Much evidence indicates that girls receive lower inputs during their
childhood than boys do, whether nutritional, medical, schooling or inputs of
other types, see for example, Subramanium (1994). One reason for this may
be that parents may want to protect a daughter against the harmful effects
of this kind of discrimination and thereby reduce schooling and other
investments that enhance future labour market productivity. A less altruistic
explanation is that parents internalise this type of gender bias in their intra-
family resource allocation decisions. A final source of gender difference in
inputs received during childhood may arise if a male child earns more than a
female child, and/or expends more effort in working for pay outside the
household. This will lead the household to compensate the boys with greater
nutrition etc., see Sen (1985).
In terms of our model, an optimal resource allocation strategy is found by

maximising the value function in terms of the inputs S and X:
Max Vc ¼ ½1=ð1 þ rÞ½gE ðA; S; XÞ þ DðA; S; XÞ þ Yc ðXÞ Ps S Px X

ð9Þ
S; X
The first order conditions are given by:
dVc =dS ¼ ½1=ð1 þ rÞ½ðdE =dS þ dD=dSÞ ¼ Ps ð10Þ
and:
dVc =dX ¼ ½1=ð1 þ rÞ½ðdE =dX þ dD=dXÞ þ dYc =dX ¼ Px : ð11Þ
Thus, schooling and other inputs should be increased up to the point where the
discounted marginal increment in future earnings plus the discounted marginal
increment in future dowry (plus, in the case of other inputs, the marginal increment
in earnings of children) is just equal to the price of schooling/other inputs. Thus, it is
clear that if parents attach more weight to future earnings of boys, gb 4 gg, or the
returns to schooling is greater for boys, dE*b/dS 4 dE*g/dS, or if additional schooling
enhances the future dowry of boys, dD*b/dS 4 dD*g/dS, then boys will receive more
schooling than girls. In the case of other inputs, the same conditions above or the
condition that the marginal increment in a boy’s child labour earnings exceed that of
a girl’s, dYcb/dX 4 dYcg/dX, will ensure that girls receive fewer childhood non-
schooling inputs than boys.
An interior solution for the optimal level of schooling and other non-schooling
inputs exists given the assumptions on the curvature properties of the future
earnings, dowry and child-labour earnings functions.
V. An Empirical Model of Poverty and Fertility

The structural model presented above illustrates the sources through which parents
develop a preference for male children, as a justification for the choice of children’s
sex mix as an instrument for further childbearing. In this section we present an
empirical model that allows us to isolate the ‘causal’ effect of fertility on poverty
by following the IV estimation strategy that has been employed in the labour
supply literature. We consider a sample of families who have already obtained two
children.15 From the discussion above it may be expected that families with two
female children will be more likely to have more children than families with one
male child and one female child. As the sex of the first two children may have
greater impacts on the family’s future economic position than on their current
income situation via the above channels, they are not correlated with current
poverty and are therefore valid instruments for future fertility among women with
two children.16
The empirical model is sketched briefly below. Let pi denote the poverty status of
the ith household. Note that pi is a zero-one dummy indicating whether or not a
household falls below the poverty line. Let xi denote whether or not the household
has more than two children. Then, the equation linking poverty to fertility is
described as:
pi ¼ a0 wi þ bxi þ Ei ; ð12Þ
where wi is a vector of other demographic controls than number of children that

affect poverty and where Ei is a normal i.i.d mean-zero disturbance term. This is a
simple probit model in which having more than two children, xi is thought to be
exogenous to the family’s poverty status.17
The alternative model proposed in this paper is that fertility and poverty are
jointly determined, as fertility is endogenous in the determination of poverty. This
leads to the following specification of a recursive simultaneous equations system
with normally distributed errors:
pi ¼ a 0 wi þ bxi þ Ei ð13Þ
xi ¼ p0 wi þ dg2i þ ui ; ð14Þ
where g2 is a zero-one dummy that indicates whether the family has had two girls
and is our instrument for expected future child-bearing among women with at least
two children, and where the error terms Ei and ui are assumed to be correlated and
normally and identically distributed. The system is estimated by full information
maximum likelihood (ML) methods applied to the joint estimation of poverty and
fertility with endogeneous fertility.
VI. Issues in Specification

Two other specification extensions should be considered to the above model. First, an
important consideration is that the sex-mix of the children may not be purely
randomly assigned in some developing countries. In India, the use of ultrasound was
first started as a ‘quality control’ measure in the population control programmes in
the 1970s and soon became a sex-screening device that was mainly used for sex-
selective abortions, see for example Arnold, Kishor and Roy (2002) for recent
estimates for India. While such prenatal screening has been declared illegal,18 in
practice the ban has proven to be ineffectual (Das Gupta 1987). Other more
traditional sex-targeting mechanisms have included the practice of female infanticide
reported within certain communities, as well as extreme neglect of female children
leading to their eventual death, see for example Sen (1984) on the relative nutritional
deficiencies of boys and girls. Subramanium (1994) presents evidence on gender bias
in other outcomes such as access to medical care. The consequences of such
phenomena is that while the normal population sex ratio is around 105 females for
100 males, in India, China and South Korea, for example, this ratio is currently below
100. Sen (1992) estimates that in the case of China, around 90 million women are
‘missing’. While it is widely believed that prenatal screening could eventually favour
females as they become relatively scarce, others such as Edlund (1999) argue that such
practices could lead to the formation of a female underclass as non-random mating
causes low-status families to prefer girls over boys, as their chances of ‘marrying up’
are higher.
Does the existence of non-random sex selection invalidate using the sex of the first
two children as a measure of an exogenous increase in fertility? This would be the
case if a correlation is induced between the gender of the first two children and
poverty, for example, if sex-targeting is only practised by the rich. However, sex-
targeting is practised uniformly across all income classes, whether by prenatal
screening methods or by the more traditional methods mentioned earlier.19 Still, the
point made by Edlund (1999) that poorer mothers would tend to bear more
daughters as sex ratios become skewed would lead to a correlation between the error
term in the poverty equation and our instrument unless additional information on
the household’s usage of sex-targeting could be incorporated. Although we lack
micro-level data on such phenomena, we use two approaches to try to correct for
this. First, we use Indian Census data to construct district-level aggregate ratios of
the number of male children to the number of female children, in various age groups
and the district level infant mortality rate. These aggregates are merged to a
household by district and used as additional controls in the fertility equation in order
to identify the effect of additional children on poverty, controlling for the existence
of non-random sex selection (Table 8, column 2). In another approach, we drop
states and UTs which have the most skewed sex ratios (in parentheses), these being
Punjab (793), Haryana (820), Delhi (865), Chandigarh (845) and Gujarat (878).20
This lowers the sample size to 12,640 households, but the coefficient to fertility in the
bivariate probit poverty equation remains about the same (Table 9).
Another consideration is the issue of hand-me-down savings for the family that
can arise from the sex-sameness of children. Particularly, these savings are seen in
the case of items such as clothing and footwear that can be easily re-used by children
of the same sex. While rural households in developing countries do not spend quite
as much on clothing and footwear as their developed country counterparts,
Rosenzweig and Wolpin (2001) estimate for a sample of 4,896 rural households in
India in 1982 (NCAER REDS data), that annual expenditures on such items on
school-aged children could amount to 8 per cent of household income. The
implication is, therefore, that having two girls could in this case, have an indirect
effect on poverty (through hand-me-down expenditure economies) in addition to the
instrumented effect on fertility. This would induce a correlation between the
instrument and the error term in the outcome equation. In order then for the
instrument to identify the causal link between fertility and poverty, some additional
restrictions are required, such as including a measure of household savings on such
items as a result of sex-sameness of children, or simply by documenting that
households with two girls in our sample do not have significantly lower expenditures
on clothing and footwear than households with mixed sex children.
Table 3 provides some evidence that hand-me-down expenditure economies are
not significant in our sample.
For example, if hand-me down economies were significant, clothing expenditures
would be significantly higher for families with one boy and one girl, compared to
families with either two boys or two girls, but the table above shows in fact that
clothing expenditures are not significantly different for families with one boy and one
girl compared to either two-boy families or two-girl families (see standard errors of
Table 3. Average monthly household expenditure (in rupees) by category and family type
Type of expenditure\Type of household 2 boys 2 girls 1 boy, 1 girl
Clothing 103.97 (5.10) 91.40 (5.67) 100.21 (3.92)

Footwear 19.01 (1.00) 17.77 (1.20) 18.98 (0.87)
Education 44.52 (1.96) 45.36 (2.22) 48.06 (1.97)
Source: NSS 1993/94 sample of rural families with two children. Mean, and standard error of
mean in parentheses.
clothing means in Table 3). The same is seen for household expenditures on
education and footwear. Thus, at least in the NSS sample these savings are not
considerable and are thereby not expected to weaken the instrument (having two
girls) in the poverty equation.
VII. Data and Descriptives

The data for this study comes from the Indian 5th Quinquennial Survey (1993–94) of
consumer expenditure conducted by the National Sample Survey Organisation. The
reference period was July 1993 to June 1994. There were approximately 115,000
households surveyed during this round (50th round). This data reports socio-
economic and demographic characteristics of the households besides item-wise
expenditure of the households. The demographic information includes age, sex, and
education along with the relationship of each member of the household. There is
some information on the immovable asset holding (land) of the sampled households.
Thus, we can identify the nucleus families with desired characteristics from the data.
From the entire sample of 115,000 households, we focused only on the 69,235
rural sector households (60 per cent of the total), mainly because children are more
likely to be considered economic assets in this group.21 From this, we derive a sub-
sample of about 14,000 households that are rural, nucleus family households in
which the wife is under 40 years of age and where there are at least two biological
children and where the second child is at least one year old.22
The notion of poverty in India is based on food energy intake method. A
normative consumption basket was derived in 1973–74 that was converted to value
terms. This works out to be Indian rupees 49.09 for the rural sector and 56.64 for the
urban sector per person per month at 1973–74 prices. The consumption basket has
remained unchanged since then. Since the consumption expenditure data collected
by the NSSO is in current rupees, the poverty line is adjusted for price changes over
time. In this paper we use the state level official poverty line (OPL) at 1993–94 prices
reported in Dubey and Gangopadhyay (1998).
The key variable in this study is the household’s poverty status, that is, if a
household falls below the state poverty line, which is determined by comparing per
capita monthly consumption expenditures (in rupees) of the household to the per
capita per month expenditure implied by the state poverty line.23 We have
information on the gender and ages of all children in the household, which allows
us to create the endogenous fertility measure, whether or not there are more than
two children, and the instrument g2, whether or not the first two children are girls.
As additional controls, we use the age of the mother, her age at first birth, the level of
schooling of both parents,24 their work status,25 work status of children, the
household’s religious affiliation (Hindu, Muslim, Christian, or other non-Hindu), its
social grouping,26 the size of the household’s loan outstanding, the total cultivatable
land owned by households (in hectares), one of the most important income
generating assets in rural India, and an indicator for female headship. In some
specifications, we use two-digit regional classifiers (state and union territories)27 or
alternatively in other specifications, the state’s (or UT’s) per capita domestic product
in 1993/1994 (current prices).
Table 4A above presents means of variables for our sample of households with
two or more children. About 31 per cent of households fall below the state poverty
line. On average, nucleus family households have around three children, although
this ranges from two (by construction) to 10. This number appears higher than for
the rural sector as a whole, (national statistics show an average of close to 2.42, see
Table 1 earlier), as our sample is based on nucleus family households with at least
two children, that is, the 25 per cent of families with zero or one child are not
included. The average acreage of cultivatable land owned by the household is 1.1
hectares. Fathers have higher levels of education than mothers, 5.6 years compared
Table 4A. Means of variables, rural households with two or more children compared to rural
households with one or more children
Rural HH with 52 children Rural HH with 51 child
Poor 0.310 (0.004) 0.308 (0.003)

No. of children 3.202 (0.011) 2.981 (0.010)
Household size 5.479 (0.014) 5.367 (0.013)
Cultivatable land 1.107 (0.019) 1.089 (0.020)
State per capita DP 6,992.208 (21.41) 7,003.478 (19.08)
Schooling of father 5.572 (0.039) 4.938 (0.035)
Schooling of mother 2.924 (0.022) 2.564 (0.020)
Father works 0.398 (0.004) 0.361 (0.004)
Mother works 0.099 (0.002) 0.085 (0.002)
Children work 0.047 (0.002) 0.040 (0.001)
Muslim 0.115 (0.003) 0.116 (0.002)
Christian 0.048 (0.002) 0.045 (0.002)
Other non-hindu 0.044 (0.002) 0.043 (0.002)
Sched.caste, tribe 0.304 (0.004) 0.308 (0.003)
Loan amount 3421.577 (202.78) 3169.103 (156.56)
Female headship 0.043 (0.002) 0.047 (0.002)
Age of mother 31.345 (0.038) 30.379 (0.038)
Age mother, 1st birth 20.465 (0.025) 20.653 (0.023)
1st child is boy 0.592 (0.004) 0.576 (0.004)
1st child is girl 0.415 (0.004) 0.424 (0.004)
2nd child is boy 0.553 (0.004) 0.487 (0.004)
2nd child is girl 0.447 (0.004) 0.408 (0.004)
1st two ch. are boys 0.342 (0.004) 0.298 (0.003)
1st two ch. are girls 0.204 (0.003) 0.189 (0.003)
More than 2 ch. 0.656 (0.004) 0.585 (0.004)
No. of rural HH with 2 or more children ¼ 14,255; all rural family HH ¼ 18,094. Mean and
standard error of mean in parentheses.
to 2.9 years. This means that both women and men are literate on average, but below
primary level (defined as schooling level 5, according to the NFEC/AEC coding
system, see note below 10). Almost 40 per cent of fathers report they are employed as
either paid labour or self-employed, while the corresponding figures for mothers and
children are 10 per cent and 5 per cent. Just over 20 per cent of the sample are non-Hindu,
while 30 per cent belong to the lowest social group.28 In terms of sex composition of
children, there is evidence of non-random sex-selection or sex-targeting (59 per cent of
households have a boy as their first child, as compared to 51 per cent in most western
countries, which would be the case if the sex of the child was purely randomly assigned,
as boys are slightly more favourably selected by nature due to their lower survival rates
in infancy). Conditional on the first child’s sex being targeted, the sex selection is
slightly less prevalent in the case of the second child (the probability of a male child here
is 0.55). Sex-targeting is again evident in the probabilities of the first two children
being boys (girls) which is 0.34 (0.20), instead of 0.25 each. Two-thirds of the sampled
rural households have more than two children (66 per cent).
Table 4A also shows the difference in means between our sample (rural households
with two or more children) and all rural family households (one or more children) as
an investigation into the representativeness of our sample. There are a few differences
in mean values across the two samples arising mainly out of definition (our sample
has more children, greater household size and higher age of the mother as per
definition), suggesting that our sample is reasonably representative of rural family
households in general. For example, poverty rates are not significantly different
across the two samples and neither are state domestic product, cultivatable land
owned, religious or social group affiliation, loan amount, female headship rate or the
sex-mix of the first child.29 The only differences that arise are families in our sample
are on average better educated, and are more likely to be working for pay.
In Table 4B we present means of variables by family type of the right hand side
variables as the difference in means between families with two boys or one boy,
Table 4B. Means of variables by family type in difference from means in families with two girls
Two boys 1 boy, 1 girl
Cultivable land 0.192 0.066 0.140 0.064

State per capita DP 7274.436 73.528 62.322 71.664
Schooling of father 0.155 0.134 70.018 0.127
Schooling of mother 70.076 0.077 70.068 0.075
Father works 70.026 0.014 0.003 0.014
Mother works 70.023 0.009 70.011 0.009
Children works 0.028 0.007 70.011 0.005
Muslim 0.010 0.009 70.009 0.008
Christian 0.010 0.007 0.008 0.006
Other non-hindu 70.001 0.006 0.007 0.006
Sched. caste, tribe 0.003 0.013 0.004 0.013
Loan amount 71350.787 506.524 256.663 714.426
Female headship 70.011 0.006 70.007 0.006
Age of mother 0.912 0.129 0.000 0.123
Age mother, 1st birth 70.288 0.085 70.226 0.082
Note: Difference in means followed by the standard error of the difference.

one girl families and the reference group, families where the first two children are
girls. Few differences emerge, with the exception of reference families having less
cultivatable land than other families, having higher state GDP, loan amount and
younger mothers than families with two boys, however, at the same time, mothers in
these families being older at time of first birth than other families. No significant
differences are seen in education of father or mother, employment status of father,
mother or children, religious affiliation or membership in the lowest social class.
Table 5 above presents clear evidence that even given the presence of sex-targeting,
74 per cent of rural nucleus families in which the first two children are girls have a
third child, compared to 63 per cent of families who have either a boy and a girl, or
two boys (the separate rates of these are 64 per cent and 62 per cent respectively) and
that this difference is significant. Thus, having two girls definitely increases the
family’s fertility more than otherwise. In the third column, we compare poverty rates
of families where the two first children are girls to other families. Poverty rates are
not significantly different between families in which the first two children are girls
and other families.
VIII. Estimation Results

Table 6 below presents estimates of the poverty probit equation in which fertility is
treated as exogeneous.
In the first specification, only schooling of both parents, religious and social
affiliation of the household, land holdings, the loan size, female headship, the age of
the mother at first birth and the fertility measure, more than two children, are
included. Most of the estimates have the expected signs. The greater the schooling
level of the mother, the lower the chances of poverty of the household while the
opposite is true for the father, indicating that education is not a good investment for
men in the rural sector, possibly increasing their risk of underemployment/mismatch
of skills or simply because poorer rural families invest in the education of males
because they lack other assets. Belonging to a scheduled caste or scheduled tribe
increases the chances of poverty, as does being Muslim (relative to Hindu).
Christians and other non-Hindus have a lower poverty risk relative to Hindus. The
more land owned by the household, the lower is its chances of being poor, and the
greater the loan size, the lower is poverty, indicating perhaps that richer households
are able to command larger loans because of the availability of collateral assets, but
the latter effect is insignificant. Female headship unexpectedly reduces poverty in this
specification. The higher the age of the mother at first birth, the greater the
Table 5. Fraction of households that have a third child, by sex of the first two children
Sex of first two children in HHs Fraction Fraction that had

with two or more children of sample a another child Poverty rate
1 boy, 1 girl or 2 boys 0.796 0.635 (0.005) 0.306 (0.004)

2 girls 0.204 0.741 (0.009) 0.328 (0.009)
Total no. of households ¼ 14,255. Standard error of mean in parentheses.

Table 6. Probit estimates of poverty equation, exogenous fertility
(1) (2) (3) (4)
Constant 70.8029*** 71.0141*** 70.5979*** 70.4067***

(0.0874) (0.0891) (0.0963) (0.1178)
Schooling father 0.0127*** 0.0135*** 0.0144*** 0.0066**
(0.0029) (0.0030) (0.0030) (0.0032)
Schooling mother 70.1151*** 70.1153*** 70.1082*** 70.1476***
(0.0055) (0.0055) (0.0056) (0.0064)
Father works – 0.2895*** 0.3072*** 0.3060***
(0.0260) (0.0261) (0.0275)
Mother works – 0.1092*** 0.1554*** 0.1270***
(0.0398) (0.0402) (0.0429)
Children work – 0.3140*** 0.3422*** 0.3902***
(0.0526) (0.0529) (0.0555)
SCST 0.3805*** 0.3310*** 0.3318*** 0.3789***
(0.0260) (0.0264) (0.0266) (0.0286)
Muslim 0.3485*** 0.3582*** 0.3289*** 0.1491***
(0.0358) (0.0361) (0.0363) (0.0397)
Christian 70.0705 70.0584 70.0291 0.2781***
(0.0568) (0.0572) (0.0577) (0.0760)
Other religion 70.2079*** 70.2064*** 70.0058 0.0694
(0.0597) (0.0572) (0.0626) (0.0847)
Land_tot 70.1193*** 70.0886*** 70.0846*** 70.0726***
(0.0079) (0.0079) (0.0079) (0.0086)
Loan amount/1000 70.0682 70.1434** 70.0961* 70.0530
(0.0492) (0.0692) (0.0562) (0.0494)
Female headship 70.1137* 0.0021 0.0061 0.0125
(0.0595) (0.0609) (0.0613) (0.0642)
Age of mother at 1st birth 0.0136*** 0.0154*** 0.0129*** 0.0103**
(0.0039) (0.0040) (0.0040) (0.0042)
More than 2 children 0.3475*** 0.3514*** 0.3568*** 0.4007***
(0.0244) (0.0247) (0.0248) (0.0261)
State dom. product/10000 – – 70.6009*** –
(0.0514)
State dom. product missing – – 70.8444*** –
(0.1833)
State dummies No No No Yes
Pseudo-R2 0.0884 0.1010 0.1093 0.1874
Log Likelihood 78043.87 77932.79 77859.31 77159.08
w2 1559.53 1781.70 1928.67 3301.62
Prob 4 w2 0.0000 0.0000 0.0000 0.0000
N ¼ 14,255 households. ***significant at 1 per cent; **significant at 5 per cent; *significant at

10 per cent.
probability the household will be poor, which is unexpected, as delaying childbirth

for investments in human capital may be expected to reduce poverty. Finally, the key
variable of interest for our study, having more than two children significantly
increases poverty of the household.
In the second specification, we add dummies for employment of the father, mother
and children of the household. The addition of these variables does not change the
estimates appreciably, suggesting that these are uncorrelated with the other variables
in the model, but all three employment indicators of the parents have positive effects
on poverty, which most likely indicates that work for pay in the rural sector is
prevalent among the poorer families. While we control for size of land holdings of
the household, we do not have information on the quality of land, and this may be
one explanation for why poorer families are forced to supplement their incomes by
working for pay, thereby inducing a positive correlation between work and poverty.
The loan size of the household now has a significant (but still negative) effect on
poverty, indicating that it is correlated with the employment variables and female
headship now has the expected positive effect on poverty. Specification 2 has a better
fit by comparison of pseudo-R2 and likelihood ratio tests. In specification 3, we add
the state’s per capita annual domestic product (in current prices, in rupees), but this
addition marginally affects the model except that once geographical differences in
domestic product are accounted for, other religious groups no longer have
significantly lower poverty rates than Hindus. The higher the state’s domestic
product, the less likely it is that a household is poor. The addition of state GDP
improves the fit. In the fourth and final column, we add a full set of state dummies to
pick up separate regional effects on poverty. Except for the fact that the coefficient
on fathers’ schooling is now halved and that Christians and people of other non-
Hindu denominations are now more likely than Hindus to be poor and that loan size
is no longer significant, all other coefficients have the same signs and significances.
This reversal in sign in the religious affiliation variable may reflect the fact that there
is a greater concentration of Christians in particular in Western India, which is
relatively richer than average. Adding state dummies definitely improves the fit as
measured by the pseudo-R2, and the coefficient on the fertility variable is still
positive and significant and remarkably robust across all the specifications.
Thus far, while we have adjusted for household composition by the use of adult
equivalent weights in the expenditure measure, we have not corrected for economies of
scale in consumption which typically arise if there are non-exclusive public goods that
the family consume in common for example, kitchen utensils used to prepare food,
electricity etc. The existence of these scale economies means that the expenditure
needed to maintain a given level of welfare rises less than proportionately with family
size. Previous work by Lanjouw and Ravallion (1995) show that the relationship
between poverty and household size can in fact vanish once economies of scale have
been accounted for. White and Masset (2003) find in Vietnam that the poverty profile
changes quite dramatically when economies of scale in household consumption
are allowed for. Thus, in the case of a country like India, where family sizes are
significantly larger than Vietnam, the relationship between poverty and family size
may be significantly altered. In order to address this issue, in Table 7, we re-estimate
the final specification from Table 6 (column 4) by adjusting our poverty measure to
reflect scale economies. That is, a household is considered poor if its expenditures per
adult equivalent adjusted for economies of scale falls below the poverty line:
Ei
4PL
ðAEi Þ1a
where a ¼ 0 (same as column 4, Table 6), 0.30 and 0.15 in columns 1, 2 and 3
respectively. First of all, it is noticeable the coefficient to the fertility variable changes
Table 7. Probit estimates, exogeneous fertility, adjusted for economies of scale
(1) (2) (3)
Constant 70.4067*** 72.818*** 71.910***

(0.1178) (0.451) (0.592)
Schooling father 0.0066** 0.002 0.009**
(0.0032) (0.005) (0.004)
Schooling mother 70.1476*** 70.138*** 70.141***
(0.0064) (0.012) (0.008)
Father works 0.3060*** 0.322*** 0.323***
(0.0275) (0.045) (0.032)
Mother works 0.1270*** 0.043 0.116**
(0.0429) (0.060) (0.047)
Children work 0.3902*** 0.003 0.187***
(0.0555) (0.081) (0.060)
SCST 0.3789*** 0.338*** 0.337***
(0.0286) (0.045) (0.033)
Muslim 0.1491*** 0.048 0.130***
(0.0397) (0.065) (0.046)
Christian 0.2781*** 0.363*** 0.299***
(0.0760) (0.113) (0.086)
Other religion 0.0694 0.064 0.153
(0.0847) (0.160) (0.103)
Land_tot 70.0726*** 70.128*** 70.095***
(0.0086) (0.025) (0.013)
Loan amount/1000 70.0530 70.123 70.105
(0.0494) (0.117) (0.079)
Female headship 0.0125 0.462*** 0.189**
(0.0642) (0.090) (0.073)
Age of mother at 1st birth 0.0103** 0.009 0.006
(0.0042) (0.007) (0.005)
More than 2 children 0.4007*** 0.123*** 0.296***
(0.0261) (0.043) (0.032)
State dummies Yes Yes Yes
Pseudo-R2 0.1874 0.154 0.170
Log Likelihood 77159.08 72418.144 74943.39
w2 3301.62 879.27 2022.76
Prob 4 w2 0.0000 0.0000 0.0000

10 per cent; (1) no adjustment, (2) a ¼ 0.30 (3) a ¼ 0.15.
when economies of scale are allowed for, dropping from 0.401 in column 1 (no
adjustment) to 0.296 (a ¼ 0.15) to 0.123 (a ¼ 0.30), however it is still significant in
each case. Not surprisingly, this indicates that children exert a smaller effect on
poverty if scale economies in consumption are assumed, and the higher the scale
parameter, the greater is the drop in value of this coefficient. Many of the other
results appear quite robust, except that female headship is significantly positive as
can be expected, and age of mother at first birth no longer is a significant
determinant. In general, the results from column 2 deviate the most, which is not
surprising given the large value of the scale parameter being assumed. As discussed
in White and Masset (2003), in developing countries where food items make up the
bulk of budgetary expenditures, a scale parameter of a ¼ 0.15 is more appropriate.
Another option could be to directly estimate the scale parameter by including
household size and household size-squared as additional regressors in a poverty
model in which expenditures are otherwise unadjusted for scale economies. When
doing so, the results obtained are similar to those in column 3 and the coefficient to
the fertility variable is estimated to be 0.267 (t value 7.56), which is approximately
the same as that found in column 3.30 Both household size and its square are
significant, respectively positive and negative, indicating the presence of economies
of scale. Of course, this method does not allow for any interactions of economies of
scale with the other determinants in the model, and therefore, we choose to retain the
method of assuming a value for the scale parameter, picking a ¼ 0.15 as the value
that is realistic in this context and using a poverty measure adjusted for economies of
scale in all the following estimations in which fertility is endogenised.
In Table 8, we present the full-information maximum likelihood (bivariate probit)
estimates of the recursive system of simultaneous equations described in the
empirical model in which fertility is endogenised, and in which the poverty measure
is corrected both for household composition and size. As an unique instrument for
fertility (that is, having more than two children) we use g2, whether or not the first
two children are girls. Estimates of both equations are presented. In terms of signs
and significance of coefficients in the poverty equation, compared to the simple
probit (Table 7, column 3), all the variables are estimated to have almost identical
effects on poverty except that the fertility coefficient is further reduced to only about
a fifth of the size of the ordinary probit estimate and is no longer significant. This
means, that once the endogeneity is taken into account, family size no longer has a
significant impact on a rural household’s poverty risk. However, we need to test for
endogeneity, and we do this formally and find that it cannot be rejected.31 Similarly,
the estimate of rho from the bivariate probit model is significant and positive,
indicating positive correlation in the residuals from the poverty and fertility
equation, that is, the correlation between Ei and ui described in the empirical model.32
Thus, all tests confirm that modelling poverty and fertility jointly is more
appropriate than treating fertility as given to poverty.
Estimates of the fertility equation are also presented in column 2, Table 8. The
results are as expected and show that fertility (the probability of having more than
two children) significantly increases with having two girls first (t-statistic of 12.53),
but also in Muslim households, belonging to SCST households (10 per cent level),
Christian households (10 per cent level), female-headed households, age of mother,
and in districts with higher infant mortality and higher male–female ratio in the 5–9
year age group. Higher infant mortality may lead families to compensate by having
more children (replacement/hoarding responses). The male–female ratio variables
are included in part to control for non-random sex-targeting, but only the ratio in
the 5–9 age group is significant. However, contrary to the hypothesis that sex-
targeting would reduce family size, the effect on fertility is positive. Fertility
significantly decreases with mother’s schooling level, if the mother or father works
for pay, and the higher the age of mother at first birth.
Earlier, in the discussion of non-random sex-selection, five north-Indian states and
territories were mentioned which have the most skewed sex ratios in the country.
Table 8. Bivariate probit estimates of poverty and fertility equations, endogeneous fertility
Poverty equation Fertility equation
Constant 77.373*** 71.640***

(0.591) (0.169)
Schooling father 0.009** 0.002
(0.004) (0.003)
Schooling mother 70.144*** 70.047***
(0.008) (0.005)
Father works 0.310*** 70.054**
(0.033) (0.026)
Mother works 0.109** 70.110***
(0.047) (0.042)
Children work 0.236*** 0.038
(0.062) (0.066)
SCST 0.336*** 0.053*
(0.033) (0.0276)
Muslim 0.150*** 0.346***
(0.046) (0.040)
Christian 0.308*** 0.096*
(0.086) (0.057)
Other religion 0.164 0.065
(0.102) (0.058)
Land_total 70.091*** 0.002
(0.013) (0.005)
Loan amount/1000 70.100 0.002
(0.079) (0.039)
Female headship 0.236*** 0.629***
(0.075) (0.074)
Age mother 1st birth 0.004 70.128***
(0.005) (0.004)
More than 2 children 0.062 _
(0.0857)
Age of mother _ 0.128***
(0.003)
Infant mortality 0–1 yr. _ 0.002***
(0.0005)
Male–female ratio 0–4 years _ 0.167
(0.206)
Male–female ratio 5–9 years _ 0.493***
(0.116)
First two children are girls _ 0.376***
(0.030)
State dummies Yes Yes
Log Likelihood 712698.3
Rho 0.164***
(0.054)
LR test of Rho ¼ 0: w2(1) ¼ 8.956, prob 4 w2 ¼ 0.0028.

10 per cent.
Omitting those households from the sample which originate from states in which
sex-targeting is most actively practised should give a cleaner estimate of the effect of
fertility on poverty. If in fact the instrument is positively correlated with the error
term in the poverty equation, for example, if sex-screening was practised exclusively
by the rich, then instrumented fertility would be biased upwards, or in other words,
omitting states with skewed sex ratios should reduce the coefficient to fertility in the
poverty equation. The results in Table 9 show that neither the poverty equation
estimates nor the fertility equation estimates move appreciably (only infant mortality
and being an SCST household are no longer significant in the fertility equation in the
reduced sample) and while the coefficient to fertility is reduced to about half its
previous value (0.036 compared to 0.062), it remains insignificantly different from
zero as before.
The focus of our study has been on identifying the causal effect of children on
whether or not a rural Indian household falls below the state poverty line. Our
reason for doing this is mainly that the poverty line, however arbitrarily defined, is a
standard and widely-used construct which makes a useful instrument for policy
purposes.
However, from a statistical standpoint, there are clearly good arguments for
adopting expenditures per person as the dependent variable, as all of the variation in
the underlying data is utilised, rather than collapsing it simply into a binary
outcome.
In Table 10, we therefore present estimates from a linear expenditure model in
which the dependent variable is the expenditures per adult equivalent adjusted for
economies of scale, a ¼ 0.15. Both the simple OLS in which fertility is treated as
exogenous and a 2SLS model in which predicted fertility replaces actual fertility (the
variables in the estimated fertility equation being the same as in Table 8, column 2)33
in the expenditure equation, with robust standard errors. The basic finding is
essentially the same, that is, in the OLS, the fertility variable significantly reduces the
household’s monthly expenditures by 55 rupees, but when endogenised, there is no
effect left on expenditures. The estimated coefficients on the other determinants are
all of the same signs and significance as in the biprobit case, except for two
noticeable differences. In the linear expenditure model, being a Muslim household
does not significantly reduce the household’s expenditures, and neither does female
headship. But, similar to the biprobit model, schooling of the mother and land
holdings reduce poverty risk, while paid employment of the mother and children and
belonging to the SCST social group are associated with higher poverty risk.
IX. Conclusions
This paper models fertility as being endogenous to a household’s poverty status. By
using a unique instrument of fertility, which is whether or not the first two children
of the household are girls, we are able to use exogenous variation in family size to
conduct a natural experiment that uncovers the causal effect of fertility on poverty of
rural nucleus households in India. Using micro data of households from the 1993–94
Indian Quinquennial Survey (5th wave) on consumer expenditure, we generate a
sample of nucleus families to empirically estimate our model. While our sample is
limited to rural nucleus households with two or more children, few significant
Table 9. Bivariate probit estimates of poverty and fertility; endogenous fertility, minus states
with skewed sex ratios
Poverty equation Fertility equation
Constant 71.721*** 71.721***

(0.178) (0.178)
Schooling father 0.009** 0.001
(0.038) (0.003)
(0.008) (0.006)
Father works 0.306*** 70.039
(0.034) (0.028)
Mother works 0.112** 70.084*
(0.049) (0.044)
Children work 0.269*** 0.004
(0.064) (0.070)
SCST 0.332*** 0.021
(0.034) (0.029)
Muslim 0.157*** 0.359***
(0.048) (0.043)
Christian 0.332*** 0.163***
(0.097) (0.062)
(0.109) (0.075)
Land_total 70.088*** 0.003
(0.014) (0.006)
Loan amount/1000 70.139 70.035
(0.092) (0.061)
Female headship 0.243*** 0.657***
(0.078) (0.081)
Age mother 1st birth 0.006 70.129***
(0.005) (0.005)
More than 2 children 0.036 _
(0.087)
Age of mother _ 0.131***
(0.003)
Infant mortality 0–1 yr. _ 0.001
(0.001)
Male–female ratio 0–4 years _ 0.143
(0.227)
Male–female ratio 5–9 years _ 0.594***
(0.124)
First two children are girls _ 0.370***
(0.032)
Log Likelihood 711,507.616
Rho 0.171***
(0.055)
LR test of Rho ¼ 0: w2(1) ¼ 9.466, prob 4 w2 ¼ 0.0021.

10 per cent.
Table 10. Linear expenditure model estimates, ols, 2sls
OLS Expenditure equation 2SLS Expenditure equation1
Constant 492.302*** 427.15***

(120.643) (71.503)
Schooling father 1.356* 1.181*
(0.775) (0.672)
(1.379) (1.349)
Father works 714.894** 712.057
(6.738) (7.914)
Mother works 720.348* 720.766***
(10.896) (7.662)
Children work 744.633*** 756.958***
(14.514) (7.957)
SCST 758.219*** 758.252***
(7.245) (5.580)
Muslim 70.687 75.711
(10.154) (15.477)
Christian 719.686 719.469
(18.355) (13.973)
(19.440) (16.795)
Land_total 15.546*** 15.153***
(1.419) (1.721)
Loan amount/1000 3.985 3.090
(9.656) (7.844)
Female headship 25.150 10.192
(15.722) (10.576)
Age mother 1st birth 70.494 0.287
(1.037) (0.846)
More than 2 children 754.827*** 4.015
(6.367) (11.591)
F-value 42.73 78.56
Prob 4 F 0.0000 0.0000
Adjusted R2 0.1186 0.115

10 per cent; dependent var ¼ expenditures per adult equivalent, adjusted for economies of
scale, a ¼ 0.15.
1
robust standard errors, predicted fertility.
differences exist across our sample and the population of rural nucleus family
households. Furthermore, while families with two girls as the first children do not
appear very different from other families in terms of their background characteristics
or poverty rates, they are much more likely to go on to have a third child than other
families. The existence of non-random sex selection does not appear to weaken the
instrument in our sample and neither do hand-me-down economies appear
substantial in our data.
Our findings indicate that fertility has a significant positive effect on poverty when
treated as exogenous. However, when modelling the simultaneity of the two
processes, there appears to be significant evidence of endogeneity and significant
correlation in the two outcomes. And, once endogenised, the effect of fertility on
poverty essentially vanishes. These results are robust to omitting states with skewed
sex ratios from the sample and to proxying economic status by expenditures instead
of poverty.
While there has been a significant drop in rural poverty in India over the last
decade, according to our results, the transition to lower fertility does not explain this
decline and the implication therefore is that other factors must have played a role in
the observed reduction in rural poverty. Macro factors such as the impact of
liberalisation and economic reforms must certainly have contributed, but results
from this study indicate that micro factors such as the schooling level of mothers and
the size of land holdings significantly raise the rural family’s economic status while
female headship and belonging to a religious or social group minority significantly
reduce it. While investments in family planning are necessary in order to give women
control over their fertility, policy makers should at the same time target
improvements in particularly women’s education, by for example raising women’s
literacy, give families access to land assets through wide-scale rural land reform as
well as concentrate on bettering the economic situation of female heads and religious
and social group minorities if the goal is to bring about an amelioration in the
economic status of rural households in India.
Acknowledgement
The authors wish to thank Partha Dasgupta, Martin Browning, Robert Michael,
Karsten Albæk, Debashish Bhattacherjee, Annapurna Shaw, Helena Skyt Nielsen
and participants of the Aarhus School of Business Economics Department Seminar
Series for useful comments. Thanks also to Peeyush Bajpai for help with initial data
processing and sample selection and to Marianne Simonsen for research assistance
with estimation.
Notes
1. For example in 1991, the two poorest social groups in India, STs and SCs, had child mortality rates of
126 and 127 per thousand compared to 26.5 for the total population (Irudaya Rajan and
Mohanchandran 2001).
2. A survey carried out to assess the impact of new agricultural technology on rural areas.
3. A related approach used by a number of studies is the use of twins as a natural experiment, see
Rosenzweig and Wolpin (1980), Bronars and Grogger (1994), Jacobsen, Pearce and Rosenbloom
(1999) etc.
4. Note that son preference is rare in sub-Saharan Africa, the Caribbean, Central Asia and most of
Southeast Asia.
5. The focus here is on economic reasons for a son preference – however, among particularly the
majority Hindu religious group, religious underpinnings exist as well, for example, having a male off-
spring light the funeral pyre is traditionally thought of as assuring immortality for the deceased.
6. Among Hindus, however, there could be a countering effect of the religious tradition known as
kanyadaan, in which giving a daughter away unconditionally in marriage gives merit to the parents.
Thus, most Hindu families do want to have a daughter among their children, once they have had a
son.
7. See Trivers and Willard (1973) for empirical evidence on this point.
8. In most parts of India, a dowry must be paid when daughters get married (Berhman et al., 1999). In
principle the dowry accrues mainly to the couple, not the parents, although certain components
(goods) may be retained by the parents. However, in the present analysis, we count dowry receipts as
incoming income to the (joint) family, nonetheless.
9. Note that not all inputs need to be child-specific, and that some inputs could be consumed in common
by all children in the household. The model could be generalised to include the latter.
10. We assume in addition that the household does not set aside savings to cover future marriage
expenditures out of current consumption.
11. We do not fully specify the dowry component in this model. Certainly, dowry must in part be
determined in a marriage market that evaluates individuals’ abilities, schooling, inputs, parental
endowments as well as other innate traits. Thus, an individual’s dowry gift/payment could also include
a match-specific component. Pryor (1977) finds in a study of the determinants of brideprice across 60
distinct populations, that the degree of female economic activity, the post-marital residence of the
couple (whether the couple will reside with the husband’s parents) and the relative decision-making
power of the spouses account for nearly half the variance in brideprices.
12. A more general formulation may allow schooling to negatively affect child labour earnings, such that
the more schooling a child is given, the less time is available for work, and hence the lower the
earnings. For example, Nielsen and Dubey (2001) show that the higher the adult/child wage ratio, the
more likely it is that children are enrolled in school.
13. In many cases extended families pool resources to form a ‘joint family’ household in which parents
cohabit with their adult male children, their spouses and children. In traditional patrilineal societies,
females may be barred by their in-laws from extending support to their parents and must instead
contribute to their husband’s household.
14. In fact, the reality is that families are net receivers of dowry in the case of boys, and net providers of
dowry in the case of girls, taking all other marriage costs into account, except in a few cases as noted
by Boserup (1970), ‘In regions where women do most of the agricultural work, it is the bridegroom
who must pay bridewealth, but where women are less actively engaged in agriculture, marriage
payments come usually from the girl’s family’ (p.48), that is, in general Dg 5 0.
15. A preference for boys will show up more clearly in the birth of a third child rather than a second child,
as most families prefer having at least two children.
16. A correlation may be present if families with daughters reduce current consumption to save for future
dowry or wedding costs. We believe this link to be rather weak, as wedding costs are low in rural areas
and neither are savings propensities high.
17. The continuous fertility measure, the number of children, has also been tried and gives qualitatively
similar results. For compactness, the results of the linear specification are not reported here, but are
available on request.
18. In 1994, the Pre-Natal Diagnostic Technologies (PNDT) Act banned the practice.
19. In fact, ultrasound fetal sex determination technology coupled with on-demand abortion (often
offered by doctors who ferry portable equipment in vans) is available in almost every village in India.
These mostly unregistered clinics charge a fee of about Rs. 500 ($10 dollars) for an ultrasound scan
plus an additional fee for an illegal abortion, see Ramachandran (2002).
20. Also known as India’s ‘Bermuda Triangle’, see Ramachandran, 2002.
21. The urbanisation level in India in 1991 was just over 25 per cent.
22. By definition, nearly all children will be biological children of the head of the household (either
mother or father) and not resident children. This is because a child is defined through its
relationship to the head of the household. This still allows for the probability that the child/children
may be the offspring of the head of the household and a previous spouse, that is, step-children. But,
as remarriage of widows is not common, the probability that children are not the biological child of
the mother is very small. Further, while selecting the initial sample of over 69,000 households, we
selected only those households where: (i) there was only one surviving mother in the household (if
there were two or more wives of the male were reported in the data); (ii) both parents were
surviving; (iii) mothers were identified either as head or spouse of the head or daughter-in-law of
the head in case grand parents of children were living with the household; (iv) the age of the
mothers was restricted to between 18 and 35 years. Further, upper income group households were
omitted and the upper age limit of children was restricted to at most 17 years. Thus, affluent
households who can afford to send their children to boarding school are also left out. These
considerations suggest that we select nucleus households with biological children only as far as
possible.
23. Poverty incidence is corrected for differences in household composition by converting household size
to adult equivalent terms, and by applying the corrected household size to the poverty line to generate
the adult equivalent poverty line, which varies across households. The consumer units used to generate
adult-equivalent poverty incidence differ by age and sex where a male aged 20–39 is the reference
person, see Appendix 1. These units are not based on observed consumption amounts within the
household, but represent instead food–energy requirements in various age–sex categories based on
body mass index. Average BMI values in different age–sex categories may indirectly reflect biases in
the inter-household allocation of resources. Therefore, for the sake of robustness, a simpler
equivalence scale giving each adult a weight of 1 and each child a weight of 0.7, irrespective of age and
sex, is tried and found to yield similar results of the specifications reported in Table 6. Finally, an
uncorrected household expenditure-based poverty measure is also tried and gives higher rural poverty
incidence, but similar conclusions with respect to fertility and other dependent variables. Thus, the
findings do not appear to be highly sensitive to the adjustment factors for adult equivalence.
24. Education is the general educational standard which is the level of education according to the NCEC/
AEC codes, with 1 indicating not literate, 2–5 literate but below primary, 5–6 primary, 7 middle, 8
secondary, 9 higher secondary, 10–13 graduate levels. This has been converted to a years of education
measure in our sample.
25. Here ‘working’ means working for pay as either self-employed in the household enterprise, own
account worker, employer, regular salaried employee, casual wage laborer, in public services or
engaged in other types of work. Unpaid family workers are not considered to be working.
26. That is, belonging to the lowest social class officially termed as the scheduled caste, a group that has
traditionally been considered untouchable, plus the group known as scheduled tribe not considered
untouchable but still a disadvantaged minority and hence denied access to education and land-
ownership.
27. There were 26 states and six union territories (UTs) in India in 1993–94. For administrative purposes,
the republic of India is classified into states and UTs.
28. This emphasises the point made earlier that the poor have a higher number of children per family. The
available statistics suggest that the lowest social groups have a higher poverty incidence and their
share in the total population is about 23 per cent as per the 1991 census.
29. The sex-mix of the second child cannot be compared across the two samples as the full sample includes
also one-child families.
30. Results available from authors.
31. We conduct a modified Hausman test in which the residual from the fertility equation is added as a
regressor in the probit poverty equation (Table 7, specification 3). The Davidson-Mackinnon robust
standard error t-test reveals a positive and significant coefficient to the residual, rejecting exogeneity.
In another specification test, a second instrument is tried, whether or not the second child is a girl, but
turns out not to add signficant explanatory power.
32. The same is also seen in a likelihood ratio test of the univariate probit versus the bivariate probit
which rejects the constrained model.
33. This first-stage fertility equation (a simple probit) produces estimates that are essentially identical
to those in Table 8, column 2. The pseudo-R2 of this probit estimation is 0.1525 and the
chi-sq(19) ¼ 2796.33, prob 4 chi-sq ¼ 0.0000, indicating good predictive power of the first-stage
equation.
References
Ahluwalia, M. S. (1978) Rural poverty and agricultural performance in India, Journal of Development
Studies, 14, pp. 298–323.
Anand, S. (1977) Aspects of poverty in Malaysia, Review of Income and Wealth, 23, pp. 1–16.
Angrist, J. D. and Evans, W. N. (1998) Children and their parents’ labor supply: evidence from exogenous
variation in family size, American Economic Review, 88(3), pp. 450–77.
Arnold, F., Kishor, S. and Roy, T. K. (2002) Sex-selective abortions in India, Population and Development
Review, 28(4), pp. 759–85.
Bardhan, P. K. (1984), Land, Labor, and Rural Poverty: Essays in Development Economics (New York:
Columbia University Press).
Becker, G. S. and Lewis, H. G. (1976) On the interaction between the quantity and quality of children.
Journal of Political Economy, 84(4(II)), pp. 143–62.
Becker, G. S. and Tomes, N. (1973) Child endowments and the quantity and quality of children, Journal of
Political Economy, 82(2(II)), S279–88.
Behrman, J. R., Foster, A. D., Rosenzweig, M. R. and Vashistha, P. (1999) Female schooling, home
teaching and economic growth, Journal of Political Economy, 107(4), pp. 682–714.
Ben-Porath, Y. (1976) Fertility responses to child mortality: micro data from Israel. Journal of Political
Economy, 84(4(II)), S163–S178.
Ben-Porath, Y. and F. Welch (1976) Do sex preferences really matter, Quarterly Journal of Economics,
90(2), pp. 285–307.
Boserup, E. (1970) Women’s Role in Economic Development (New York: St. Martin’s Press).
Bronars, S. G. and Grogger, J. (1994) The economic consequences of unwed motherhood: using twins as a
natural experiment, American Economic Review 84(5), pp. 1141–56.
Das Gupta, M. (1987) Selective discrimination against female children in rural Punjab, India. Population
and Development Review, 13, pp. 77–100.
Dubey, A. and Gangopadhyay, S. (1998) Counting the Poor: Where are the Poor in India? Sarvekshana
Analytical Report No.1, New Delhi: Department of Statistics, Government of India.
Dubey, A. S., Gangopadhyay, S. and Wadhwa, W. (1999) Female-headed households in India: incidence,
poverty and socio-economic characteristics. Planning and Policy Research Unit Paper, Indian
Statistical Institute, New Delhi.
Dubey, Amaresh, Gangopadhyay, S. and Wadhwa, W. (2001) Occupational structure and incidence of
poverty in Indian towns of different sizes, Review of Development Economics, 5(1), pp. 49–59.
Edlund, L. (1999) Son preferences, sex ratios and marriage patterns, Journal of Political Economy,
107(6(1)), pp. 1275–1304.
Fishlow, A. (1972) Brazilian size distribution of income, American Economic Review, 62, pp. 391–402.
Gaiha, R. (1988) On measuring the risk of rural poverty in India, in: T. N. Srinivasan and P. K. Bardhan
(eds), Rural Poverty in South Asia (New Delhi: Oxford University Press), pp. 219–61.
Irudaya Rajan, S. and Mohanchandran P. (2001) Infant and child mortality rates for scheduled castes and
scheduled tribes: Based on 1991 Census, Economic and Political Weekly, 36(48).
Jacobsen, J. P., Pearce III, J. W. and Rosenbloom, J. L. (1999) The effects of child-bearing on married
women’s labor supply and earnings using twin births as a natural experiment, Journal of Human
Resources, 34(3), pp. 449–74.
Lanjouw, P. F. and Ravallion, M. (1995) Poverty and household size, Economic Journal, 105, No. 433.
Miller, B. D. (1981) The Endangered Sex: Neglect of Female Children in Rural North India (Ithaca, N.Y.:
Cornell Univ. Press).
Murthi, M., Guio, A. and Dreze, J. (1995) Mortality, fertility and gender bias in India: a district level
analysis, Population and Development Review, 21, pp. 745–82.
Nielsen, H. S. and Dubey, A. (2001) Child labor: A microeconomic perspective, Indian Journal of Labour
Economics, 45, pp. 79–496.
Pryor, F. L. (1977) The Origins of the Economy (New York: Academy Press).
Ramachandran, S. (2002) Concern over India’s vanishing girls, D þ C Development and Cooperation
No.1, Jan./Feb. 2002, p. 27.
Rosenzweig, M. R. and Evenson, R. (1977) Fertility, schooling and the economic contribution of children
in rural India: An econometric analysis, Econometrica, 45(5), pp. 1065–79.
Rosenzweig, M. R. and Wolpin, K. I. (1980) Testing the quantity–quality fertility model: The use of twins
as a natural experiment, Econometrica, 48(1), pp. 227–40.
Rosenzweig, M. R. and Schultz, T. P. (1982) Market opportunities, genetic endowments, and intrafamily
resource distribution: child survival in rural India, American Economic Review, 72(4), pp. 803–15.
Rosenzweig, M. R. and K. I. Wolpin (2000) Natural ‘natural experiments’ in economics, Journal of
Economic Literature, XXXVIII(4), pp. 827–74.
Saith, A. (1981) Production, prices and poverty in rural India, Journal of Development Studies, 17.
Sen, A. (1984) Resources, Values and Development (Cambridge, M.A.: Harvard University Press).
Sen, A. (1985) Commodities and Capabilities (Amsterdam: North-Holland).
Sen, A. (1992) Missing women, British Medical Journal, 304, pp. 587–8.
Subramanium, S. (1994) Gender discrimination in intra-household allocation in India, (mimeo),
Department of Economics, Cornell University.
Swaminathan, M. (1998) Economic growth and the persistence of child labor: Evidence from an Indian
City. World Development 26, pp. 1513–28.
Tambiah, S. J. (1973) Dowry and bridewealth and the property rights of women in South Asia, in: Jack R.
Goody and Stanley J. Tambiah (eds), Bridewealth and Dowry, Cambridge Papers in Social
Anthropology, No.7. Cambridge: Cambridge University Press.
Trivers and Willard (1973) Natural selection of parental ability to vary the sex-ratio of offspring, Science,
1973, 179, 5 Jan., pp. 90–2.
White, H. and Masset, E. (2003) Constructing the poverty profile: An illustration of the importance of
allowing for household size and composition in the case of Vietnam, Development and Change, 34(1),
Jan.
World Bank (1990) World Development Report: Poverty (Oxford University Press, London).
Appendix 1. Consumer units by age and sex
Sex
Age group (in completed years) Male Female
Below 1 0.43 0.43

1–3 0.54 0.54
4–6 0.72 0.72
7–9 0.87 0.87
10–12 1.03 0.93
13–15 0.97 0.80
16–19 1.02 0.75
20–39 1.00 0.71
40–49 0.95 0.68
50–59 0.90 0.64
60–69 0.80 0.51
70 and above 0.70 0.50
Source: National Sample Survey Organisation (1997), Sarvekhshana, Vol. 21, No.2.

gupta2006

Uploaded by

Copyright:

Available Formats

You might also like

gupta2006

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

gupta2006

Uploaded by

Copyright:

Available Formats

Journal of Development Studies,

Vol. 42, No. 1, 110–138, January 2006

Fertility and the Household’s Economic

Final version received January 2005

ISSN 0022-0388 Print/1743-9140 Online/06/042110-29 ª 2006 Taylor & Francis

Evenson (1977) develop a household time-allocation model, which explicitly

II. Poverty and Fertility in India

Table 1. Selected health indicators for India

Measure 1951 1981 1991 Current

Table 2. Number of children per household in rural areas

0–10 3.37 1.74 3.20 1.71

N ¼ 82,661 (69,235) rural nucleus family households in 1987/1988 (1993/1994).

IV. A Theoretical Model of Son Preference

Vc ¼ ½1=ð1 þ rÞððgE þ D Þ þ Yc Ps S Px X ð1Þ

We assume that dE*/dA 4 0, d2 E*/dA2 5 0, dE*/dS 4 0, d2E*/dS2 5 0, dE*/dX 4 0,

and we assume that dYc/dX 4 0, d2Yc/dX2 5 0. This relationship between non-

Eb ðA; S; X; db Þ > Eg ðA; S; X; dg Þ; db < dg : ð6Þ

Db > Dg : ð7Þ

Ycb ðXÞ > Ycg ðXÞ: ð8Þ

In terms of our model, an optimal resource allocation strategy is found by

Max Vc ¼ ½1=ð1 þ rÞ½gE ðA; S; XÞ þ DðA; S; XÞ þ Yc ðXÞ Ps S Px X

The ﬁrst order conditions are given by:

dVc =dS ¼ ½1=ð1 þ rÞ½ðdE =dS þ dD=dSÞ ¼ Ps ð10Þ

dVc =dX ¼ ½1=ð1 þ rÞ½ðdE =dX þ dD=dXÞ þ dYc =dX ¼ Px : ð11Þ

V. An Empirical Model of Poverty and Fertility

where wi is a vector of other demographic controls than number of children that

VI. Issues in Speciﬁcation

Type of expenditure\Type of household 2 boys 2 girls 1 boy, 1 girl

Clothing 103.97 (5.10) 91.40 (5.67) 100.21 (3.92)

VII. Data and Descriptives

Rural HH with 52 children Rural HH with 51 child

Poor 0.310 (0.004) 0.308 (0.003)

Two boys 1 boy, 1 girl

Cultivable land 0.192 0.066 0.140 0.064

Note: Diﬀerence in means followed by the standard error of the diﬀerence.

VIII. Estimation Results

Sex of ﬁrst two children in HHs Fraction Fraction that had

1 boy, 1 girl or 2 boys 0.796 0.635 (0.005) 0.306 (0.004)

Total no. of households ¼ 14,255. Standard error of mean in parentheses.

Table 6. Probit estimates of poverty equation, exogenous fertility

(1) (2) (3) (4)

Constant 70.8029*** 71.0141*** 70.5979*** 70.4067***

N ¼ 14,255 households. ***significant at 1 per cent; **significant at 5 per cent; *significant at

probability the household will be poor, which is unexpected, as delaying childbirth

Table 7. Probit estimates, exogeneous fertility, adjusted for economies of scale

(1) (2) (3)

Constant 70.4067*** 72.818*** 71.910***

N ¼ 14,255 households. ***significant at 1 per cent; **significant at 5 per cent; *significant at

Poverty equation Fertility equation

Constant 77.373*** 71.640***

LR test of Rho ¼ 0: w2(1) ¼ 8.956, prob 4 w2 ¼ 0.0028.

Poverty equation Fertility equation

Constant 71.721*** 71.721***

LR test of Rho ¼ 0: w2(1) ¼ 9.466, prob 4 w2 ¼ 0.0021.

Table 10. Linear expenditure model estimates, ols, 2sls

OLS Expenditure equation 2SLS Expenditure equation1

Constant 492.302*** 427.15***

N ¼ 14,255 households. ***significant at 1 per cent; **significant at 5 per cent; *significant at

Appendix 1. Consumer units by age and sex

We assume that dE/dA 4 0, d2 E/dA2 5 0, dE/dS 4 0, d2E/dS2 5 0, dE*/dX 4 0,

Constant 70.8029* 71.0141* 70.5979* 70.4067*

N ¼ 14,255 households. significant at 1 per cent; significant at 5 per cent; significant at

Constant 70.4067* 72.818* 71.910***

N ¼ 14,255 households. significant at 1 per cent; significant at 5 per cent; significant at

Constant 77.373* 71.640*

Constant 71.721* 71.721*

Constant 492.302* 427.15*

N ¼ 14,255 households. significant at 1 per cent; significant at 5 per cent; significant at