Jayachandran and Pande 2017

American Economic Review 2017, 107(9): 2600–2629
https://doi.org/10.1257/aer.20151282
Why Are Indian Children So Short?

The Role of Birth Order and Son Preference†
By Seema Jayachandran and Rohini Pande*
Child stunting in India exceeds that in poorer regions like sub-

Saharan Africa. Data on over 168,000 children show that, relative
to Africa, India’s height disadvantage increases sharply with
birth order. We posit that India’s steep birth order gradient is due
to favoritism toward eldest sons, which affects parents’ fertility
decisions and resource allocation across children. We show that,
within India, the gradient is steeper for high-son-preference regions
and religions. The gradient also varies with sibling gender as
predicted. A back-of-the-envelope calculation suggests that India’s
steeper birth order gradient can explain over one-half of the India-
Africa gap in average child height. (JEL I12, J13, O15, Z12, Z13)
One in four children under age five, worldwide, is so short as to be classified as

stunted (UNICEF 2014). Child stunting—a key marker of child malnutrition—casts
a long shadow over an individual’s life. On average, people who are shorter as chil-
dren are less healthy, have lower cognitive ability, and earn less as adults.1
India, where over 30 percent of the world’s stunted children live, stands out in
particular (UNICEF 2013). Its child stunting rate is over 40 percent, an outlier even
among poor countries (IIPS 2010).2 Figure 1 graphs average child height-for-age for
sub-Saharan African (hereafter, African) countries and Indian states against income.
* Jayachandran: Department of Economics, Northwestern University, 2211 Campus Drive, Evanston, IL 60208
(email: [email protected]); Pande: Harvard Kennedy School, Harvard University, Mailbox 46, 79 JFK
Street, Cambridge, MA 02138 (email: [email protected]). We thank Patrick Agte, Alejandro Favela, Lydia
Kim, Suanna Oh, and Alexander Persaud for excellent research assistance. We are grateful to Abhijit Banerjee, Jere
Behrman, Diane Coffey, Angus Deaton, Rebecca Dizon-Ross, Jean Drèze, Erica Field, Dominic Leggett, Nachiket
Mor, Debraj Ray, Dean Spears, Tomasz Strzalecki, Alessandro Tarozzi, several seminar participants, and three
anonymous referees for helpful comments. We thank the International Growth Centre, National Science Foundation
(Jayachandran) and Harvard’s Women and Public Policy Program (Pande) for funding. The authors declare that
they have no relevant or material financial interests that relate to the research described in this paper.
†
Go to https://doi.org/10.1257/aer.20151282 to visit the article page for additional materials and author
disclosure statement(s).
1
Stunting is defined as having a height-for-age that is two standard deviations or more below the worldwide
reference population median for one’s gender and age in months. Child height is predictive of adult height (Healy
et al. 1956) and taller adults have greater cognitive skills (Glewwe and Miguel 2008; Guven and Lee 2015), fewer
functional impairments (Barker and Osmond 1986; Barker et al. 1993; Gould 1989), and higher earnings (Strauss
and Thomas 1998; Case and Paxson 2008; Hoddinott et al. 2013).
2
Unlike in western countries where economic growth was accompanied by rapid height increases (Floud et al.
2011), recent economic growth has only modestly increased average height in developing countries (Deaton 2007).
Between 1992 and 2005, India’s annual economic growth exceeded 6 percent, yet stunting declined by just 0.6 per-
centage points (1.3 percent) per year (Tarozzi 2012).
2600
VOL. 107 NO. 9 Jayachandran and Pande: Why Are Indian Children So Short? 2601
−0.5
SEN
Mean height-for-age z-score

GHA
TN
STP KE
−1
COG GO
SLE SWZ
KEN NAM
MN CMR
GIN JM NA NGA
ZWE HP
MLI SK
LBR UGA AP PJ
LSO TR
TCD AR
TZA ZMB MZ
COD RJWB
−1.5
KA
OR DL
TZA MG MH
AS
MDG
NER RWA LSO
UC HR
ETH MP
JH
CH GJ
BH UP African countries Africa fitted values
−2
MWI
Indian states India fitted values
6 7 8 9
ln(GDP per capita) in birth year
Figure 1. Child Height versus National GDP
Notes: The light and dark circles represent sub-Saharan African countries and Indian states, respectively. The aver-
ages are calculated over all children less than 60 months old. The lines represent the best linear fit for each sample.
National GDP data are based on the Penn World Table 9.0 (Feenstra, Inklaar, and Timmer 2015).
Both regions exhibit a positive correlation between income and child height, but
the curve for India is lower; at a given level of income, Indian children are shorter.
Given that India performs better than African countries on most health and develop-
ment indicators, this contrast is striking and is the focus of this paper.3
Using survey data on over 168,000 children from India and 25 African countries,
we demonstrate a steeper height drop-off for later-born children in India. India’s
relative height disadvantage materializes for second-born children and increases for
third and higher order births, at which point mean height-for-age for Indian children
is lower than that of African children by 0.3 standard deviations of the worldwide
distribution. The steeper drop-off with birth order in India than Africa also holds
for weight-for-age, child hemoglobin and, importantly, for an array of prenatal and
postnatal health inputs.
We use several approaches to ensure that differential household selection into
high fertility is not generating the observed steeper birth order gradient in India.
First, the same patterns hold when we control for maternal and neighborhood char-
acteristics that are correlated with total fertility and child outcomes. Second, the
results are robust to estimating the patterns using only between-sibling variation in
child height (i.e., holding family size and other family characteristics fixed). Finally,
we consider two different samples where mothers have likely completed fertility,
and in both cases the birth order patterns are robust to flexibly controlling for total
fertility in parallel to birth order.
3
India outperforms African countries on maternal mortality, life expectancy, food security, poverty incidence,
and educational attainment (Gwatkin et al. 2007). Yet, India has the fifth highest stunting rate among 81low-income
and low-middle-income countries with comparable child height data (UNICEF 2013), despite being in the middle
of the group (rank 43) for GDP per capita.
2602 THE AMERICAN ECONOMIC REVIEW september 2017
Turning to the underlying mechanism at work, we propose that eldest son prefer-
ence in India—encompassing both a desire to have at least one son and for the son to
be healthy—influences parents’ fertility decisions and how they allocate resources
across children, leading to the steep birth order gradient in height. Eldest son pref-
erence can be traced to the patrilocal and patrilineal Hindu kinship system: aging
parents typically live with, and bequeath property to, their eldest son (Dyson and
Moore 1983; Gupta 1987). Further, Hindu religious texts emphasize post-death rit-
uals which can only be conducted by a male heir (Arnold, Choe, and Roy 1998).
(Henceforth, “son preference” is shorthand for eldest son preference.)4
The data support several specific predictions of this hypothesis. First, within
India, the birth order gradient is shallower among children living in Indian states
that practice matrilineality (Kerala and the eight Northeast states) and those with a
less male-skewed sex ratio.5 We also observe a shallower birth order gradient within
India among Muslim children; relative to Hinduism, Islam places less emphasis on
having a son. Finally, consistent with the Hindu-Muslim difference in eldest son
preference, we show that the birth order gradient in India exceeds that in the neigh-
boring countries of Bangladesh and Pakistan.6
Second, the son-preference mechanism predicts that the drop-off in height across
successive children will depend on their older siblings’ gender composition. Among
boys, we posit that eldest sons receive more health and nutrition inputs than their
brothers. Consistent with the fact that sons born at high birth order are less likely
to be the family’s first son, we observe a birth order gradient among boys. We also
exploit variation in when the family’s first son is born, and show that a son born at
birth order 2 is taller in India than Africa if and only if his older sibling is a girl: i.e.,
he is the family’s eldest son.
Among girls, older siblings’ gender composition affects resources in two ways,
both of which disfavor later-born daughters. First, there is the “sibling rivalry effect”:
a girl born at later birth order has (by definition) more older siblings. It follows that
she is also more likely to have an elder brother, and likely to fare poorly when com-
peting with him for family resources (Garg and Morduch 1998; Pande 2003).
The second mechanism is fertility-stopping behavior, which instead confers a
disadvantage to a later-born daughter who lacks brothers. When a daughter is born
into a family with only girls, her parents are likely to continue having children in
their quest for a son, exceeding their originally desired family size. Thus, the birth
of a late-parity girl is akin to a negative income per capita shock and fewer resources
are expended on her. Both mechanisms generate a birth order gradient among girls,
and the net effect of having an older brother on the resources allocated to a girl is
ambiguous. Empirically, we find that the net effect is positive, pointing to the impor-
tance of the fertility-stopping mechanism.
4
India also exhibits a general preference toward all sons. In Section III, we discuss the role of eldest son pref-
erence versus general son preference in explaining the patterns we find.
5
In Kerala and the Northeast states, some but not all population groups follow matrilineal practices.
6
Using country comparison groups other than Africa also helps address the concern that it may be Africa that
is special, not India. In addition to the within-South-Asia comparison, we also compare Indian children to those in
countries with similar GDP and those in countries with genetically similar population groups. In all cases, India
exhibits the steepest birth order gradient.
These two mechanisms affecting girls provide a third testable prediction among
daughters with no elder brother. While a lack of sibling rivalry will increase both
prenatal and postnatal investments in these daughters relative to those with an elder
brother, the fertility-stopping effect reduces the postnatal investment they receive
because, after their birth, parents realize they need to try again for a son. We show
that, relative to Africa and to prenatal inputs, girls in India receive fewer postnatal
resources if their family does not yet have a son. Theories for the steep Indian birth
order gradient other than son preference seem unlikely to explain the several pat-
terns we find in the data.
The birth order differences in health inputs and height directly demonstrate stark
inequality across Indian children. This inequality is also informative about the aver-
age Indian deficit in child height. At a most basic level, the decomposition by birth
order and gender is informative about the genetic potential of Indian children: unless
genotypes vary systematically by birth order, all Indian children have the genetic
potential to be at least as tall as the observed average height of first-borns. The
absence of a height deficit for Indian first-borns, thus, suggests that genetics cannot
explain most of India’s height deficit compared to Africa.
More directly, we argue that parents’ unequal allocation of resources across
children affects average height in India. Mirroring the India-Africa patterns, our
within-India analyses show that there is higher average child height and a shal-
lower birth order gradient for Indian regions with less son preference. For these
subgroups, we observe a more equal allocation of resources, but not more inputs to
children on average. Consistent with height production functions in the literature,
these patterns suggest that height exhibits diminishing returns to inputs (Steckel
1995).7 With diminishing returns to inputs, unequal investment across children will
depress average child height.
To quantify the link between birth order gradient differences across India and
Africa and the average child height deficit in India, we conduct two back-of-the-
envelope calculations. The first asks how much of the gap is explained by the birth
order gradient, and the second how much is explained specifically by eldest son
preference generating a birth order gradient. As a starting point, we demonstrate
a negative correlation between the birth order gradient and level of height across
African countries. For the first exercise, we multiply this estimated correlation by
the India-Africa gap in the birth order gradient. When we compare the resulting
estimate to the observed India-Africa height gap (adjusted for GDP per capita), we
find that the birth order gradient explains over one-half of the Indian height puzzle.
Next, we use the low eldest son preference states of Kerala and the Northeast as a
proxy for India without eldest son preference and conduct a similar calculation. We
find that the birth order gradient rooted in eldest son preference explains one-third
of the India-Africa height gap. While not rigorously establishing the link between
the gradient and level, these two accounting exercises are suggestive that eldest
son preference and, relatedly, the birth order gradient have quantitatively important
implications for average height in India.
7
We also use Indian Human Development Survey data to directly demonstrate that child height exhibits dimin-
ishing returns to household income and expenditures in India.
Our paper complements, and adds to, the literature on the environmental deter-
minants of child height. Spears (2013), for instance, focuses on open defecation as
a cause of the Indian height disadvantage.8 We also contribute to the literature on
how cultural gender preferences and gender gaps in perceived returns to investment
cause unequal resource allocation across siblings (Rosenzweig and Schultz 1982;
Behrman 1988; Garg and Morduch 1998; Oster 2009). Our contribution is to show
how gender preferences, by accentuating birth order gradients, can explain a signif-
icant fraction of child stunting in India. To the best of our knowledge, ours is the
first paper to examine how cultural norms of son preference influence birth order
effects and, more generally, is one of the few papers in economics to study why birth
order effects arise.9 Finally, we contribute to the literature on the unintended conse-
quences of son preference by demonstrating how dynamic fertility decisions cause
inequality in health outcomes between genders, among brothers, and even among
sisters (Sen 1990; Clark 2000; Jensen 2003; Jayachandran and Kuziemko 2011).
The remainder of the paper is organized as follows. Section I describes the data
and presents descriptive statistics for the sample. Section II presents evidence on
the birth order gradient in the Indian height disadvantage, and Section III presents
evidence on eldest son preference as the root cause and also tests alternative expla-
nations for the within-family patterns. Section IV concludes.
I. Background and Data Description
The established link between child stunting and adverse long-term outcomes, as
well as the relative ease of measuring child height (versus, say, keeping a compre-
hensive food diary for a child) has led to the widespread use of height as a marker
of child malnutrition. However, and especially for cross-country comparisons, it is
important to account for the other key factor determining height: genetic potential.
A common norm, and the one we follow, is to create the child’s height-for-age
(HFA) z-score based on the World Health Organization (WHO) universally applica-
ble growth standard for children aged zero to five years.10 A z-score of 0 represents
the median of the gender- and age-specific reference population, and a z-score of −2
indicates that the child is two standard deviations below that reference-population
median, which is the cutoff for being considered stunted. Our primary outcome of
interest is the HFA z-score because it is the child health measure that has been most
often linked to later-life outcomes and is viewed as the best cumulative measure of
8
Much of the recent evidence weighs against genetic explanations in explaining India’s height deficit (Coffey et
al. 2013). For instance, work examining the height of Indian children whose parents migrate to rich countries finds
significant narrowing in the gap between Indian-born children and worldwide norms (Tarozzi 2008; Proos 2009).
9
Price (2008) studies parental care-giving time as a reason for birth order effects. Birth order gradients have
been documented for outcomes as varied as IQ, schooling, height, and personality (Behrman and Taubman 1986;
Sulloway 1996; Black, Devereux, and Salvanes 2005, 2011; Belmont, Stein, and Susser 1975; Horton 1988).
10
The WHO standard describes how children should grow if they receive proper nutrition and health care. It
is premised on the fact that the height distribution among children under age five who receive adequate nutrition
and health care has been shown to be similar in most ethnic groups (de Onis et al. 2006; WHO Multicentre Growth
Reference Study Group 2006a). The WHO constructs the height distribution using a sample of children from six
affluent populations across five continents (Brazil, Ghana, India, Norway, Oman, and the United States) with no
known health or environmental constraints to growth and who received recommended nutrition and health inputs
(WHO Multicentre Growth Reference Study Group 2006b).
child malnutrition. As a robustness check we also consider weight for age (WFA)
z-scores (which are similarly defined) as an outcome variable.
The 2005–2006 National Family Health Survey (NFHS-3) is our data source for
Indian children; it employs the same sampling methodology and survey instrument
as the internationally used Demographic and Health Surveys (DHS). Following the
previous literature on the puzzle of Indian malnutrition, we use sub-Saharan African
children as the comparison group for Indian children (Ramalingaswami, Jonsson,
and Rohde 1996). Sub-Saharan Africa’s level of development is similar to (but, on
average, lower than) India. The comparison group comprises the 25 sub-Saharan
African countries (from now on, African countries) where DHS Surveys collected
child anthropometric data and occurred between 2004 and 2010 (to ensure a com-
parable time period to NFHS-3). Two African countries (Tanzania and Lesotho)
were surveyed twice in this time period, and we include both survey rounds. The
DHS sample in our main analysis refers to the 27 Demographic and Health Surveys
for African countries plus India’s NFHS-3 survey. Our robustness checks use DHS
surveys from other regions.
The DHS interviews women aged 15 to 49 years, and measures height for their
children aged 5 and under. Our sample comprises the 168,108 children with anthro-
pometric data.11 Table 1 provides summary statistics, and the online Data Appendix
provides other survey details. The average Indian child is slightly older than the aver-
age African child (30.2 months versus 28.3 months). A comparison of HFA z-scores
shows that Indian children are shorter than African children (−1.51 and −1.35,
respectively). We define child birth order based on all children, currently alive or
deceased, ever born to a mother. As African women have more children (3.9) than
their Indian counterparts (2.7), the mean birth order in Africa (3.7) exceeds that in
India (2.6). Lower total fertility in India implies that despite similar mothers’ age at
first birth, average maternal age at birth for children in our sample is lower in India
(25 years) than Africa (27 years). The average spacing between births is reasonably
high and similar in India (36 months) and Africa (39 months). We also define a sub-
sample with likely completed fertility comprising the 49,880 mothers who state that
they do not want more children or are sterilized or infertile.
In addition, we use data on prenatal and postnatal health-related behaviors.
Prenatal behavior includes the number of prenatal care visits, whether the pregnant
woman received tetanus shots and iron supplementation, and delivery at a facility;
India typically outperforms Africa on these measures (for example, 69 percent of
the time, pregnant women in India took iron supplements, compared to 62 percent
in Africa). Data on health inputs for young children include whether they had a
medical checkup within the first two months of life, whether they received iron
supplementation, and the total number of vaccinations. India has higher vaccination
rates, while postnatal checkups and child iron supplementation are more common
in Africa. Table 1 also summarizes our control variables. Reflecting differences in
the number of countries covered and total sample size, Africa has close to 3 times
as many primary sampling units as India (10,366 and 3,822, respectively), while
Following WHO guidelines on handling outlier values, we exclude observations with a HFA z-score > 6 or
11
< −6 (>5 or <

−6 for WFA) as these are likely to be erroneous.
Table 1—Summary Statistics
India Africa India Africa

subsample subsample subsample subsample
Mother’s age at birth (years) 24.75 26.96 Child’s age (months) 30.20 28.27
[5.23] [6.86] [16.90] [17.06]
Mother’s total children born 2.74 3.88 Child is a girl 0.48 0.50
[1.82] [2.54] [0.50] [0.50]
Mother’s desired fertility 2.47 4.62 Child’s birth order 2.62 3.74
[0.96] [1.47] [1.80] [2.48]
Mother wants more children 0.34 0.67 Child’s HFA z-score −1.51 −1.35
[0.47] [0.46] [1.81] [1.94]
Mother completed her 0.67 0.33 Child is stunted 0.40 0.38
fertility [0.47] [0.47] [0.49] [0.48]
Mother is literate 0.58 0.50 Child’s WFA z-score −1.53 −0.88
[0.49] [0.50] [1.33] [1.42]
Mother’s height (meters) 1.52 1.58 Child’s hemoglobin level 10.28 10.15
[0.06] [0.07] (g/dl) [1.57] [1.68]
Mother took iron 0.69 0.62 Child is deceased 0.05 0.07
supplements [0.46] [0.48] [0.22] [0.26]
Mother’s total tetanus shots 1.87 1.41 Child taking iron pills 0.06 0.11
[0.94] [1.20] [0.23] [0.32]
Total prenatal visits 4.04 3.85 Child’s total vaccinations 6.61 6.24
[3.48] [3.07] [2.80] [3.12]
Delivery at health facility 0.45 0.47 Birth spacing (months) 36.16 38.69
[0.50] [0.50] [20.32] [20.63]
Postnatal check within 0.09 0.30 Diarrhea in last two weeks 0.09 0.16
two months [0.29] [0.46] [0.29] [0.36]
Average pooled inputs 0.33 0.38 Open defecation 0.46 0.32
[0.28] [0.30] [0.50] [0.47]
Percent nonresident among 0.02 0.10 Land scarcity 5.03 2.56
children [0.04] [0.08] – [1.17]
Number of adult females 1.85 1.60 Number of PSUs 3,822 10,366
in household [1.09] [1.06]
log GDP per capita 7.78 7.36 Main sample of children 42,069 126,039
(in child’s birth year) [0.10] [0.65]
Notes: The means of the specified variables are calculated separately for the India and Africa subsamples. Standard
deviations appear in brackets. The following variables are summarized at the mother level: total children born, moth-
er’s desired fertility, wants more children, mother completed her fertility, mother is literate, and mother’s height.
Total prenatal visits, mother took iron supplements, total tetanus shots, postnatal check within two months are also,
in effect, summarized at the mother level because they are only available for the most recent birth. Variables summa-
rized at the child level include: mother’s age at birth, birth spacing (the birth interval between a child and his or her
older sibling), delivery at health facility, average pooled inputs, all child variables (first ten variables in the second
column), diarrhea in last two weeks, open defecation, percent nonresident among children, number of adult females
in the household, and log GDP per capita in child’s birth year. Land scarcity is summarized at the country level.
maternal literacy is higher in India. The online Data Appendix provides further
details on variables entering the analysis.
Our within-India analysis uses two datasets. First, we pool all three waves of NFHS
(conducted in 1992–1993, 1998–1999, and 2005–2006), which gives us a sample
of over 90,000 Indian children. Second, we use the Indian Human Development
Survey (IHDS), a two-wave survey conducted in 2005 and 2012. The IHDS panel
structure and seven-year gap between waves allow us to focus on families that had
no children between the two waves and therefore (almost surely) completed fertility.
While the sample size among families with completed fertility is relatively small
(3,615 children under age 5 with height data in wave 1), the IHDS analysis allows
us to show that our results are robust to controlling for family size. Online Appendix
Table 1 provides summary statistics for these two datasets.
II. Birth Order and Child Outcomes
We start by documenting the key fact that underlies our analysis: the steeper birth
order gradient in child height in India relative to Africa. We then discuss endogene-
ity concerns, and provide relevant robustness checks. We conclude this section by
documenting similar gradients in parental inputs that likely influence height and in
other child health outcomes.
A. Child Height
Basic Finding.—Figure 2 plots the average child height-for-age (HFA) z-scores

for India and Africa, separately by birth order. An Indian deficit emerges at birth
order 2 and widens for birth order 3 and higher.
Table 2 examines this pattern via regression analysis. In column 1 we show
the average India-Africa height gap, pooling all children. Indian children are, on
average, 0.08 standard deviations shorter than African children, a difference that is
significant at the 1 percent level. As shown in column 1, online Appendix Table 2,
this average deficit remains significant and increases to −0.16 standard deviations if
one controls for PPP-adjusted GDP per capita in the child’s birth year; India is richer
than the African comparison group, on average.
We next disaggregate the height disadvantage by birth order. The outcome vari-
able remains HFA for child iborn to mother min country c . We estimate
HFA i  mc = α1 I c    + α2 I c    × 2nd Child i  mc  + α3 I c    × 3rd+ Child i  mc 
(1)
+ β1 2nd Child i  mc  + β2 3rd+ Child i  mc  + γ X i  mc  + ϵimc ,
where I c  is an indicator for Indian children; α

1is the India gap for first-born chil-
dren (omitted birth order category), and α2 and α3capture how the gap differs for
second-born children and third-and-higher birth order children; Xi  mcis a vector of
controls that always includes child age dummy variables (in months) to account for
nonlinear patterns of z-scores and age. We also expand the set of controls to check
the robustness of our results, as described below. Throughout, standard errors are
clustered at the mother level.
Column 2 of Table 2 shows that the Indian height disadvantage opens up at birth
order 2: the interaction of India and being second-born is − 0.14and highly sig-
nificant. The Indian disadvantage remains significant and increases, with third and
higher births having a height z-score gap of − 0.28compared to African children
(sum of main effect and interaction term). 12
12
In Section IIIF, we conduct an accounting exercise that quantifies how much this gradient contributes to the
average height deficit in India compared to Africa.
−1.2

−1.30 −1.29 −1.30
−1.4
−1.38
−1.42
−1.6
Africa
India
−1.8 −1.75
Birth order 1 Birth order 2 Birth order 3+
Figure 2. Child Height in India and Africa, by Child’s Birth Order
Notes: The figure depicts the mean child height-for-age z-scores for sub-Saharan Africa and India, by the birth order
of the child. The mean is calculated over all children less than 60 months old.
Table 2—India’s Differential Birth Order Gradient in Child Height and Related Outcomes
HFA WFA Hb
z-score Stunted z-score level Deceased
(1) (2) (3) (4) (5) (6) (7) (8) (9)
India −0.082 0.092
[0.011] [0.018]
India ×
2nd child −0.144 −0.161 −0.110 −0.243 0.051 −0.146 −0.094 0.003
[0.025] [0.027] [0.063] [0.048] [0.007] [0.020] [0.030] [0.004]
India ×
3rd+ child −0.377 −0.227 −0.193 −0.436 0.064 −0.198 −0.159 0.002
[0.024] [0.032] [0.092] [0.085] [0.009] [0.024] [0.036] [0.004]
2nd child 0.023 −0.011 −0.097 −0.167 0.009 0.009 −0.011 −0.014
[0.015] [0.017] [0.053] [0.027] [0.004] [0.012] [0.022] [0.002]
3rd+ child −0.066 −0.118 −0.169 −0.334 0.036 −0.063 −0.037 −0.011
[0.013] [0.019] [0.074] [0.044] [0.005] [0.014] [0.025] [0.003]
Africa mean of outcome −1.351 −1.351 −1.351 −1.351 −1.351 0.375 −0.877 10.150 0.071
Child’s age dummies × India No No Yes Yes Yes Yes Yes Yes Yes
Mother’s literacy × India No No Yes Yes No Yes Yes Yes Yes
Mother’s age at birth × India No No Yes Yes No Yes Yes Yes Yes
PSU fixed effects No No Yes Yes No Yes Yes Yes Yes
Mother fixed effects No No No No Yes No No No No
Completed fertility sample No No No Yes No No No No No
Observations 168,108 168,108 167,737 66,566 83,228 167,737 167,737 88,838 199,514
Notes: Standard errors are clustered by mother and appear in brackets. HFA z-score is the child’s height-for-age
z-score, Stunted is defined as having an HFA z-score ≤ −2, WFA z-score is the child’s weight-for-age z-score, and
Hb level is the child’s hemoglobin level; 2nd child is an indicator for children whose birth order is 2; 3rd+ child is
an indicator for children whose birth order is 3 or higher. Child age dummies are included in all columns. In col-
umns 3–4 and 6–9, the main effect India is absorbed by PSU fixed effects. In column 5, the main effect India is
absorbed by mother fixed effects. Columns 3–4 and 6–9 include PSU fixed effects, a linear and a quadratic variable
for mother’s age at birth, mother’s literacy, and mother’s literacy, maternal age, and child age dummies interacted
with India. In columns 3 and 5–9, the sample is restricted to PSUs with at least two children aged 1–59 months. In
column 4, the sample is restricted to children whose mothers report that they do not want to have more children,
are sterilized, or are infecund. Column 4 includes total fertility dummies, top-coded at 6 children, and total fertility
dummies interacted with India. In column 8, Hb level is defined for children 6 months or older and is not available
for six surveys. In column 9, the sample consists of ever-born children aged 13–59 months. See the online Data
Appendix for further details.
Endogeneity Concerns.—The ideal data for examining differences in the birth

order gradient across India and Africa would use households that had completed fer-
tility and would have height data for all children. This would allow us to control for
total family size in parallel to birth order and ensure that the estimates are not con-
founding the effects of birth order and family size (Black, Devereux, and Salvanes
2005). In this case, birth order would be orthogonal to family characteristics (so
adding mother fixed effects might improve precision but would not change the birth
order coefficients).
However, the nature of DHS sampling implies that a large fraction of house-
holds in our sample have not completed childbearing. The best proxy for intended
family size is a survey question on the mother’s desired fertility, but it is not asked
prior to childbearing, rendering it potentially endogenous to a woman’s fertility out-
comes. Moreover, actual and desired fertility often differ in countries where access
to contraception is limited. Hence, our regressions cannot control for total family
size in general, raising an omitted variable bias concern; the birth order variable
in between-household comparisons could be proxying for high-fertility families.
Higher birth order children are more likely to come from larger families, and family
size could be correlated with child height; family size could affect child height via
its effect on the available resources per child, plus larger families tend to be poorer.
One response is to include family fixed effects; eventual total family size is held
fixed when making comparisons across siblings, despite our not directly observing
it. The one caveat is that DHS surveys only provide height data for children age five
and younger, raising the possibility of endogenous selection into the relevant sam-
ple; for example, the sample in regressions with family fixed effects will typically
have shorter birth spacing than the full sample and birth spacing could differ across
India and Africa.
Given this, we address endogeneity concerns in multiple ways. First, and as we
describe immediately below, we conduct a set of robustness checks using the DHS
data. Second, in Section III we conduct a parallel within-India analysis for children
of women who have completed fertility using the IHDS data. Finally, we consider
comparison groups of countries other than Africa; if the nature of differential selec-
tion is specific to the India-Africa comparison then we should not observe a steeper
birth order gradient when we use alternative comparison groups.
Our first approach to address endogeneity is to include a rich set of covariates in
parallel to birth order: primary sampling unit (PSU) fixed effects, maternal literacy,
maternal age, and child age. In rural areas a PSU is a village, and in urban areas
it is a neighborhood. PSU fixed effects control for many dimensions of economic
and health status, as well as unobserved environmental conditions. For instance,
fertility outcomes are highly correlated within PSUs. We also control for maternal
literacy, which again is highly correlated with observed fertility.13 Our other two
controls relate to maternal and child age. Within families, birth order is correlated
with maternal age. The public health literature identifies a nonlinear relationship
13
For the completed fertility sample, in India (Africa) the average standard deviation of fertility within a PSU is
1.27 (2.01) while the standard deviation across all of India (Africa) is 1.88 (2.76). Pooling regions, literate women
with completed fertility have 1.7 fewer children than illiterate women. In India (Africa), literate women with com-
pleted fertility have 1.6 (1.9) fewer children than illiterate women.
between maternal age and child health (specifically, early and late pregnancies are
associated with worse outcomes: see Fall et al. 2015). To ensure that birth order
effects do not proxy for maternal age, we include a quadratic in maternal age in the
controls.14 Finally, child age is correlated with birth order within families; among
siblings, the higher birth order child will, by definition, be younger. We therefore
use child age dummies as covariates. Importantly, for each of these covariates, we
include the interaction with the India dummy. Column 3 of Table 2 shows that the
addition of these control variables reduces the magnitude but not significance of
the Ic    × 3rd+ Childcoefficient, and does not appreciably change the I c    × 2nd Child
coefficient. This specification will be our main specification in later analyses.
Next, we classify women who stated they do not want any more children or who
have been sterilized as likely to have completed fertility. We then reproduce col-
umn 3 for children belonging to this subsample of households where the mother
has likely completed fertility (our sample size is roughly 40 percent of the original
sample). We use the observed number of children for these households as their total
family size, and in our regression include controls for family size dummies inter-
acted with India. Column 4 shows that our results on the birth order gradient hold,
although they are less precisely estimated. Finally, similar in spirit to controlling for
actual family size is to control for desired family size. Keeping in mind the caveats
mentioned earlier, column 1 of online Appendix Table 2 shows that our results are
robust to controlling for desired fertility.
In column 5 of Table 2 we report regressions that include mother fixed effects. By
only using within-family comparisons for identification, we fully control for family
size differences. Birth order and child’s age are strongly correlated within a family,
so we continue to control for I c    × ChildAge. Requiring there to be at least 2 children
from a family in the sample reduces the sample size to 83,228 children. The effec-
tive sample size is even smaller: the birth order coefficients are identified off of the
42,524 children (13,550 for India and 28,974 for Africa) with one or more siblings
in the sample with a different birth order than them (i.e., not simply multiple births)
and where at least one sibling is birth order 1 or 2 (so that not all siblings fall in our
3rd+ Child category). The Indian birth order gradient remains statistically signifi-
cant, and the results are similar though somewhat larger in magnitude to those in
columns 2 and 3. Consistent with findings in many settings that low-parity children
have better outcomes, we observe a negative birth order gradient in Africa (the coef-
ficients on 2 nd Child and 3rd+ Childare negative and significant). The key finding
is that the birth order gradient in child height is twice as large in India as in Africa.
The mother fixed effects specification is an important robustness check that de
facto includes fixed effects for eventual total family size, which does not vary within
family. The drawback is that the birth order gradient is identified off less than one-
half of the sample. A specific concern is that siblings with shorter than average
birth spacing identify the mother fixed effects estimates and, therefore, selection
based on birth spacing might differ between India and Africa. Reassuringly, aver-
age birth spacing in this subsample is reasonably high and similar across India and
Africa (26 months versus 29 months). Moreover, because the mother fixed effects
14
We de facto control for mother’s current age, as it is the sum of child age and mother’s age at birth.
s pecification includes child age (in months) dummies, we are de facto controlling
for birth spacing between the siblings in the sample. As a further robustness check,
we also control directly for a child’s birth spacing from his or her older sibling
interacted with India. In the specification with household controls (online Appendix
Table 3, column 2), the birth order gradient is very similar to our main results. With
mother fixed effects (column 3), India’s birth order gradient remains statistically
significantly steeper for children of birth order 3 and higher; the coefficient for the
birth order 2 interaction becomes smaller but remains marginally significant.15
Next, we use alternative geographic comparison groups to check whether what
we interpret as an abnormally steep birth order gradient in India is actually an
abnormally shallow gradient in Africa. In online Appendix Table 4, columns 1–3 we
define the comparison group for India economically: the comparison group com-
prises 25 country surveys (between 2004 and 2010) for which country GDP per
capita in the survey year was within 50 percent (either higher or lower) of India’s
2005–2006 GDP per capita. India exhibits a significantly stronger birth order gradi-
ent than this alternative comparison group.
In columns 4–6 we define the comparison group in terms of genetic similar-
ity. Recent genome studies that use modern-day genetic distance between ethnic
groups to reconstruct prehistoric migration patterns find evidence of Indo-European
migration and genetic similarity between India, Europe, Central Asia, and West Asia
(Cavalli-Sforza, Menozzi, and Piazza 1994). We use 16 European and Central and
West Asian countries with DHS surveys as the comparison group, and again find a
significantly stronger birth order gradient in India than in the comparison group.16
Finally, we compare India to its two South Asian neighbors. The puzzle of stunt-
ing is often framed as the “South Asian enigma,” but our hypothesis that son prefer-
ence is the root cause predicts that the birth order gradient should be steeper in India
than Bangladesh and Pakistan (which are majority Muslim countries; Islam has less
eldest son preference than Hinduism). Columns 7–9 show that the birth order gradi-
ent is indeed steeper in India. (In Section III, we will show that a steeper birth order
gradient among Hindus than Muslims also holds within India.)17
15
Online Appendix Table 3 reports further robustness checks. In columns 4 and 5 we find similar estimates
when we restrict the sample to children who are birth order 4or lower. Our estimates are also robust to considering a
more endogenous definition of birth order, namely birth order among currently living children (columns 6 and 7). In
Jayachandran and Pande (2015) we show that the results are robust to excluding African countries with fertility that
is above the median of our full African sample. Further, to check that polygamy (which is more common in Africa
than India) is not biasing estimates, we show that the results are similar when the sample is restricted to mothers
who have only had children with one partner.
16
One difference is the absence of a first-born advantage in India, which is unsurprising given that this compar-
ison group is significantly richer than the African comparison group.
17
Jayachandran and Pande (2015) report a placebo test for whether India’s birth order gradient is an outlier
among countries. We compare each country in our sample to the rest of the sample and find that India is the only
country with a significantly steeper birth order gradient than the rest of the sample. This is the case even when, to
account for India’s larger sample size, we aggregate African countries to regions. A second placebo test uses the
25 African countries and 29 Indian states in the sample, randomly selects 29 countries or states to comprise a pla-
cebo India, and estimates the differential Indian birth order gradient, repeating the exercise 500 times. The actual
India × 2nd Childand I ndia × 3rd+ Childcoefficients are in the bottom 1 percent of the distribution of estimates,
i.e., have a p-value < 0.01.
B. Other Health Outcomes
We focus on the continuous HFA z-score but of specific policy relevance is the
stunting rate (a measure of overall child malnutrition): the incidence of children
with a HFA z-score ≤ −2. Column 6 of Table 2 shows that the steep Indian birth
order gradient holds for stunting: relative to their African counterparts, the disad-
vantage for Indian second-borns is 5 percentage points, and for third-borns, 6 per-
centage points (statistically significant at the 1 percent level). Thus, the high birth
order penalty for stunting is two to three times as large in India as in Africa. A
similar pattern holds when height in centimeters is the outcome (online Appendix
Table 3, columns 8 and 9).
If the birth order gradient in height reflects unequal resource allocation across
children then we would also predict a birth order gradient in other health outcomes.
To examine this possible mechanism, we start by considering other health outcomes
that are likely also affected by parents’ resource allocation. Table 2, columns 7 and
8 show a differentially steep birth order gradient in weight-for-age and hemoglobin
in India. Column 9 examines infant mortality. This is a negative health outcome, so
we would predict positive India-birth order interaction terms. The point estimates
are indeed positive, though statistically insignificant. Examining infant mortality
also serves a different purpose: it addresses the concern that mortality selection
might underlie India’s strong birth order gradient. For mortality selection to explain
the height patterns, we would need India to have a negative differential birth order
gradient in infant mortality; weak later-born children survive at a high rate, gener-
ating a negative birth order gradient in survivors’ height. However, we observe the
opposite pattern.
C. Health Inputs
Next, in Table 3 we examine birth order gradients in prenatal and postnatal child
investments. The prenatal input information is based on retrospective information
about inputs in utero and at childbirth; the prenatal outcomes (and some of the post-
natal outcomes) are only available for the youngest child in the family. We estimate
equations of the form of equation (1), and the set of covariates includes PSU fixed
effects and controls for maternal literacy, mother’s and child’s age, and their inter-
actions with the India dummy.
In columns 1 to 4 we examine whether the steeper gradient holds for each pre-
natal input. On average, Indian women are more likely to obtain prenatal care, take
iron supplements, and receive tetanus shots during pregnancy but are less likely to
deliver at a health facility. However, for all outcomes other than tetanus shots, we
observe a significantly sharper decline with birth order in India than in Africa. The
gradient magnitudes are large enough that for two of the three inputs where the
India average exceeds the Africa average (prenatal visits and iron supplementation),
later-born Indian children receive fewer inputs than their African counterparts.18
18
As the main effect for India is absorbed by PSU fixed effects, the tables do not report the gap among first-
borns. The comparison of absolute levels is based on a specification without PSU fixed effects.
Table 3—Child Health Inputs
Prenatal inputs Postnatal inputs

Mother’s Postnatal
Total Mother total Delivery check Child Child’s Average
prenatal took iron tetanus at health within two taking total pooled
visits supplements shots facility months iron pills vaccinations inputs
(1) (2) (3) (4) (5) (6) (7) (8)
India ×
2nd child −0.525 −0.031 −0.019 −0.040 −0.009 −0.008 −0.203 −0.011
[0.052] [0.008] [0.018] [0.006] [0.013] [0.005] [0.039] [0.003]
India ×
3rd+ child −1.012 −0.071 −0.036 −0.092 0.014 −0.010 −0.462 −0.033
[0.060] [0.009] [0.021] [0.008] [0.014] [0.006] [0.051] [0.004]
2nd child −0.181 −0.014 −0.112 −0.088 0.005 −0.004 −0.098 −0.044
[0.029] [0.005] [0.013] [0.004] [0.010] [0.004] [0.025] [0.002]
3rd+ child −0.431 −0.031 −0.206 −0.133 −0.022 −0.013 −0.207 −0.071
[0.033] [0.005] [0.014] [0.004] [0.011] [0.005] [0.030] [0.003]
Africa mean of outcome 3.846 0.622 1.415 0.472 0.302 0.113 6.245 0.380
India mean of outcome 4.041 0.689 1.872 0.450 0.090 0.055 6.607 0.334
Age and other controls Yes Yes Yes Yes Yes Yes Yes Yes
Observations 115,343 117,686 117,199 167,377 35,888 91,936 122,898 167,724
Notes: Standard errors are clustered by mother and appear in brackets. Control variables included are child age
dummies, mother’s literacy, maternal age, PSU fixed effects, and child age dummies, mother’s literacy, and mater-
nal age interacted with India. The main effect India is absorbed by PSU fixed effects. Total prenatal visits, mother
took iron supplements, mother’s total tetanus shots, and postnatal check within 2 months are only available for
the youngest living child in the family; postnatal check within 2 months is collected in only 13 African surveys.
Delivery at health facility, child taking iron pills, and total vaccinations are available for all births in the past
five years; child taking iron pills is collected in only ten African surveys; total vaccinations uses children ages
13–59 months, as the recommended age for some is up to 1 year. In column 8, the average across four prenatal and
three postnatal inputs is used to create the outcome. The dummies are (i) total prenatal visits > 3; (ii) mother took
iron supplements; (iii) mother’s total tetanus shots > 2; (iv) child was delivered at a health facility; (v) child is tak-
ing iron pills; (vi) total vaccinations > 8; (vii) child had postnatal check within two months of birth. See the online
Data Appendix for further details.
Columns 5 to 7 analyze three postnatal investments. The prevalence of postnatal

checkups is much lower in India than Africa (reflecting an Indian social norm of
maternal home confinement for 40 days after birth) and child iron pill consumption
is also lower. However, Indian children are more likely to get vaccinated. There is
no differential birth order gradient across India and Africa for postnatal checkups
and iron pill consumption. In contrast, vaccinations show a strong negative and sta-
tistically significant India birth order gradient.19 In column 8 we summarize our
findings and show that the steeper birth order gradient holds for a composite input
measure: the average pooled inputs received by a child. This measure is the average
value of seven indicator variables. For the three input variables that are originally
multivalued (total prenatal visits, total tetanus shots, and total vaccines), we con-
struct dummy variables that equal 1 if the original measure exceeds the sample
median. In sum, to the extent that child health inputs affect child height, this birth
order gradient in inputs is consistent with a behavioral basis for the height birth
order gradient.
19
We do not examine breastfeeding as an outcome because the choice of how long to breastfeed is determined
both by its health benefits and subsequent fertility (Jayachandran and Kuziemko 2011).
III. Culture and Height Deficits
The Indian birth order gradient in child height is steeper than that in Africa and
several alternative comparison groups including India’s neighboring countries of
Bangladesh and Pakistan. An important difference between India and comparator
countries lies in the religious make-up of the population: roughly four-fifths of
India’s population is Hindu.
In this section, we provide two types of evidence that eldest son preference—
which follows from the tenets of Hinduism—is an important mechanism underlying
the steep child height gradient in India. First, we use regional and religious variation
within India to show that the birth gradient in height is shallower when son prefer-
ence is lower. Second, we show that the Indian birth order gradient varies with the
gender composition of siblings in a manner consistent with eldest son preference.
A. Within-India Evidence
We identify multiple subgroups within India that are marked by lower than aver-
age son preference and examine whether, relative to the rest of India, these sub-
groups have a less negative birth order gradient.
To do so, we use two different datasets. First, we use the pooled sample of all
three NFHS waves, conducted in 1992–1993, 1998–1999, and 2005–2006. By pool-
ing the three waves, we gain statistical power. Second, we use a c ompleted-fertility
sample from the Indian Human Development Survey (IHDS), a panel with two
waves collected seven years apart. To construct our IHDS sample, we use the second
wave to identify mothers who had completed fertility by the first wave: nonpregnant
women who did not give birth after the first wave.20 Among children born to these
mothers, we examine height-for-age of children who were under age five in wave 1.
To examine whether the birth order gradient is muted in regions and among social
groups that exhibit lower son preference, we estimate a model analogous to equa-
tion (1) with one difference: the indicator for India is replaced by an indicator for
the low-son-preference subgroup. In regressions using the IHDS sample, we also
control for family size dummies in parallel to birth order, i.e., include fixed effects
for family size interacted with the son-preference proxy.
We begin by comparing matrilineal Indian states—Kerala and the eight
Northeastern states—with the rest of India. Matrilineality—which is associated with
kinship practices that favor boys less and do not prioritize eldest sons—is more
common in these states (Oommen 1999; Chakrabarti and Chaudhuri 2007; Gneezy,
Leonard, and List 2009). Column 1 of Table 4 shows that the birth order gradient
in height is significantly more muted in matrilineal states. A comparison of subsa-
mple means provides suggestive evidence that differences in the gradient influence
20
The approach of using a long gap of no childbearing to identify completed fertility complements is arguably
superior to using mothers’ stated desire to not have more children to measure completed fertility in DHS surveys,
which we employed in Table 2, column 4. Mothers who completed their fertility have, on average, 3.11 children
in IHDS as compared to 3.22 among mothers in the NFHS-3 sample who we identified as likely having completed
fertility. With our IHDS analysis, we can enlarge the sample slightly by also including women who gave birth after
wave 1, but just not in the five years preceding wave 2; we find similar and somewhat more precisely estimated
effects.
Table 4—Cultural Norms and Child Height: Within-India Evidence
Low son preference proxy Kerala and Northeast Below-median child sex ratio Muslims
HFA WFA HFA HFA WFA HFA HFA WFA HFA
z-score z-score z-score z-score z-score z-score z-score z-score z-score
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Low son pref proxy 0.078 0.008 1.040 0.078 0.039 0.374 −0.027 0.034 0.212
× 2nd child [0.039] [0.029] [0.515] [0.030] [0.023] [0.236] [0.047] [0.035] [0.360]
Low son pref proxy 0.108 0.069 1.793 0.081 0.044 1.065 0.184 0.156 −0.279
× 3rd+ child [0.045] [0.033] [1.043] [0.036] [0.027] [0.372] [0.055] [0.041] [0.568]
2nd child −0.185 −0.154 −0.578 −0.207 −0.173 −0.650 −0.159 −0.153 −0.573
[0.017] [0.013] [0.116] [0.020] [0.015] [0.140] [0.017] [0.013] [0.123]
3rd+ child −0.422 −0.350 −0.472 −0.437 −0.363 −0.738 −0.412 -0.354 −0.413
[0.020] [0.015] [0.183] [0.024] [0.019] [0.218] [0.021] [0.016] [0.193]
Low son pref group mean −1.388 −1.198 −1.407 −1.561 −1.491 −1.485 −1.732 −1.602 −1.227
of outcome
High son pref group mean −1.710 −1.648 −1.557 −1.721 −1.622 −1.584 −1.691 −1.628 −1.575
of outcome
Sample NFHS 1-3 NFHS 1-3 IHDS 1 NFHS 1-3 NFHS 1-3 IHDS 1 NFHS 1-3 NFHS 1-3 IHDS 1
Age and other controls Yes Yes Yes Yes Yes Yes Yes Yes Yes
Observations 95,125 95,125 3,615 95,125 95,125 3,615 82,084 82,084 3,405
Notes: Standard errors are clustered by mother and appear in brackets. In columns 1–2, 4–5, and 7–8, the sample
uses NFHS 1-3. NFHS-1 only has data for children aged four years and younger, and NFHS-2 only has data for
children aged 3 years and younger. In columns 3, 6, and 9, the sample uses IHDS 1 and consists of children aged
1–59 months in IHDS 1 whose mothers (likely) had completed their fertility, i.e., their mothers did not give birth
between IHDS 1 and IHDS 2. All columns include child age dummies, maternal age, mother’s literacy, and child
age dummies, maternal age, and mother’s literacy interacted with Low son pref proxy. Columns 1–2, 4–5, and
7–8 include survey and PSU fixed effects, and survey and PSU fixed effects interacted with Low son pref proxy.
Columns 3, 6, and 9 include total fertility dummies, top-coded at 6 children, and total fertility dummies interacted
with Low son pref proxy. In columns 1–2 and 4–5, the main effect Low son pref proxy is absorbed by PSU fixed
effects. In columns 3 and 6–9, the main effect low son pref proxy is included but not shown. Kerala and Northeast
include Arunachal Pradesh, Assam, Kerala, Manipur, Meghalaya, Mizoram, Nagaland, Sikkim, and Tripura. Child
sex ratio is defined as the number of boys aged 0–6 years over the number of girls aged 0–6 years in the respondent’s
state-by-rural cell. In columns 7–9, the sample is restricted to Muslims and Hindus. See the online Data Appendix
for further details.
average child height: average child height in matrilineal states exceeds that in the
rest of India. Column 2 shows that the same pattern holds for weight-for-age. In
column 3 we estimate this relationship using the IHDS sample. We observe the same
pattern—a significantly shallower birth order gradient for child height in matrilineal
states.21
Next we examine heterogeneity by the child sex ratio, calculated for each state-
by-urban cell (which is the finest administrative level at which we can match census
sex ratio data to NFHS). The sex ratio, as defined, is increasing in the proportion
male. In columns 4 to 6 we find that, as predicted, low-sex-ratio regions have a
shallower birth order gradient. We continue to see a negative correlation between
the steepness of birth order gradient and average child height: the subsample means
show that average child height is higher in low-sex-ratio regions.
Finally, we examine differences by religion: compared to Hinduism, Islam places
less emphasis on needing a son for religious ceremonies, and Islamic inheritance
rules disfavor women less. Son preference, in turn, is weaker among Muslims; for
21
Online Appendix Table 5 reports similar specifications examining weight-for-age using IHDS data and
prenatal and postnatal inputs using NFHS data.
example, the sex ratio is less skewed among Muslims than Hindus (Borooah and Iyer
2005) and the gender gap in child mortality is smaller (Bhalotra, Valente, and Soest
2010). Using our NFHS data we find that, relative to Hindus, Muslim Indians have
a much more muted birth order gradient in HFA and WFA for birth order three and
higher (columns 7 and 8). However, we do not observe a differential Hindu-Muslim
height gradient using IHDS data (column 9). It could be that our covariates do not
fully control for unobserved socioeconomic characteristics by religion; this may
also be the reason why subsample averages show that Muslim children have rel-
atively lower HFA and WFA z-scores. Consistent with Muslim families being, on
average, poorer, online Appendix Table 5 shows that child inputs are lower among
Muslim families. Importantly, these resources are more equally distributed across
birth order among Muslims.
B. Favoritism toward Eldest Sons and Birth Order Gradients
We now use the DHS sample to test a series of predictions that follow if the child
height gradient stems from parents’ eldest son preference.
Prediction 1: Relative to African counterparts, both boys and girls in India will
exhibit a steeper birth order gradient.
Among boys this is straightforward: the eldest son, by definition, has the low-
est birth order among sons in the family and will be favored over his siblings.
Importantly, this gradient should be absent if, instead, parents exhibit general son
preference, i.e., they favor sons over daughters but not the eldest son particularly.
Among girls, eldest son preference disfavors later-born girls in two ways. First,
by virtue of having more older siblings, a later-born girl is more likely to have an
elder brother and be in competition with him for resources.
Second, parents’ desire for a son affects their fertility decisions. Consider a fam-
ily with a desired fertility of two children and which wants at least one son. Ex ante
the preferences are compatible because the likelihood of any child being male is
(very close to) 50 percent. If the first-born is a daughter, then parents realize that
they may need to exceed desired fertility to ensure a son. They will decide expendi-
tures on this daughter given their available resources and an expected family size of
three. If their second child is also a girl, then parents certainly need to exceed their
desired fertility of two in order to have a son. Consequently, and assuming fixed
family resources, the second daughter will receive fewer early-life resources than
her older sister because the expected family size has increased from three to four.22
It is theoretically ambiguous whether the male or female gradient in India should
be steeper. Eldest son preference directly drives unequal allocation of resources
across brothers and, thus, might have a particularly strong effect on allocation and
outcomes across boys. In contrast, the birth order gradient for girls is not generated
22
In our data, the majority of Indian mothers report an ideal family size of two children. Families strongly want
at least one son and thereafter prefer having a roughly balanced gender composition (Jayachandran 2017). Myopia
among parents such that they only update their fertility plans when it is certain that they need to exceed their desired
fertility will amplify the extent to which the birth of a second or later daughter is a positive shock to expected family
size and thus to future expenses.
by parents having discriminatory preferences across daughters per se. It is also pos-
sible that inequality among girls is larger if having more children than originally
planned has important consequences for resources allocated to each child. We turn
to the data and estimate the gender-specific gradients:
(2) Y icm = α1 I c    + δ1 I c    × Girl + δ2 I c    × Girl × 2nd Childimc
 
+ δ3 I c    × Girl × 3rd+ Childimc

    + β1 2nd Childimc
    + β2 3rd+ Childimc
 
+ β3 Girl × 2nd Childimc

    + β4 Girl × 3rd+ Childimc
    + β5 Girlimc
 
+ α2 I c    × 2nd Childi  mc  + α3 I c    × 3rd+ Childi  mc  + γ Xi  mc  + ϵimc.
This is an expanded form of equation (1), where the key additional regressors are
the triple interaction between India, birth order, and being a girl. We are interested
in δ 2 and δ 3 , which test whether India’s steep birth order gradient is stronger among
girls or boys.
Column 1 of Table 5 shows a similarly steep birth order gradient for Indian boys
and girls; the triple interactions of India, higher birth order, and the girl dummy,
while negative, are statistically insignificant. However, unlike with boys, the first-
born height advantage is absent for Indian girls (relative to their African counter-
parts). Specifically, the main effect for India implies that, on average, first-born
Indian sons are 0.15z-score points taller than their African counterparts.
The coefficients remain fairly similar when we include additional covariates (col-
umn 2) and mother fixed effects (column 3). Finally, in column 4 we show that the
same pattern holds when we consider weight-for-age as the outcome variable.
While the birth order gradient does not differ by gender, there are two reasons
to expect a level difference by gender in India. First, if eldest sons receive more
resources than all other children, then sons on average will fare better than daugh-
ters. Second, the gender composition of children influences fertility behavior: in
India, relative to Africa, the birth of a girl in a family with only daughters increases
mothers’ desire for additional children (Jayachandran and Pande 2015). Thus, rel-
ative to sons, daughters in India are more likely to belong to larger than planned
families that lack adequate resources for their children (Clark 2000; Jensen 2003).
These two effects, together, yield a second prediction.
Prediction 2: The India-Africa height gap will be more pronounced among

girls.
Table 5, column 5 summarizes the average gender bias in the Indian height defi-
cit. The India dummy is small and insignificant and the coefficient on I ndia × Girl
is −0.14. Thus, overall, only Indian girls show a child height disadvantage relative
to Africa and this gender deficit remains significant when we include additional
covariates (column 6), and also when we estimate a regression with mother fixed
effects (column 7). Column 8 shows that Indian girls are relatively disadvantaged in
terms of weight-for-age as well.
Table 5—Child Gender and the Birth Order Gradient in Height
HFA WFA HFA WFA

z-score z-score z-score z-score
(1) (2) (3) (4) (5) (6) (7) (8)
India 0.148 −0.011
[0.026] [0.014]
India ×
Girl −0.111 −0.143 −0.147 −0.098 −0.116
[0.036] [0.020] [0.019] [0.032] [0.014]
India ×
2nd child −0.107 −0.152 −0.228 −0.122
[0.036] [0.040] [0.069] [0.030]
India ×
3rd+ child −0.352 −0.221 −0.414 −0.175
[0.033] [0.047] [0.097] [0.035]
India ×
2nd child ×
Girl −0.076 −0.045 −0.024 −0.047
[0.053] [0.057] [0.101] [0.043]
India ×
3rd+ child × Girl −0.051 −0.048 −0.030 −0.064
[0.047] [0.067] [0.092] [0.049]
Africa mean of outcome −1.575 −1.575 −1.575 −1.575 −1.351 −1.351 −1.351 −1.351
Age and other controls No Yes No Yes No Yes No Yes
Mother fixed effects No No Yes No No No Yes No
Observations 168,108 165,596 83,228 165,596 168,108 167,737 83,228 167,737
Notes: Standard errors are clustered by mother and appear in brackets. Child age dummies are included in all
regressions. Columns 2, 4, 6, and 8 additionally include mother’s literacy, maternal age, and PSU fixed effects. In
columns 2 and 4, child age dummies, maternal age, and mother’s literacy are interacted with Girl, India, and India
×Girl and PSU fixed effects are interacted with Girl. In columns 3 and 7, the main effect India is absorbed by
mother fixed effects. In columns 2 and 4, the main effects India and India ×
Girl are absorbed by PSU fixed effects
and their interactions with Girl. In columns 6 and 8, the main effect of India is absorbed by PSU fixed effects.
The main effect of Girl is included in all regressions but not shown. In columns 1–3, coefficients for 2nd child and
3rd+ child, 2nd child ×Girl, and 3rd+ child ×Girl are included in the regression but not shown. See the online
Data Appendix for further details.
Eldest versus General Son Preference.—An overall gender gap in child height
would also be observed if parents favor all sons and not just their eldest sons. This
raises the question of whether India’s birth order gradient and height gap are driven
by eldest son preference or general son preference. As we elaborate on below, while
both types of son preference are present—Indian parents favor all sons over daugh-
ters and also favor the eldest son over other sons—eldest son preference appears to
be what causes the birth order gradient.
As evidence that India exhibits general son preference, when we compare boys
and girls with an older brother, the boys enjoy a relative height advantage over the
girls in India. In other words, even non-eldest sons are favored over girls. However,
general son preference cannot by itself explain two patterns we see in the data.
First, general son preference does not predict a birth order gradient among boys
(Prediction 1). Second, the within-India evidence using matrilineal states indicates
that the birth order gradient is linked to eldest son preference. We showed earlier
that matrilineal states have a shallower gradient and a higher level of height than
the rest of India. If what differentiates these states from the rest of India is weaker
eldest son preference, then both girls and non-eldest sons should do better in these
states. If, instead, matrilineal states differ in their general son preference, then we
should see a smaller girl disadvantage in these states but no obvious gains among
non-eldest boys in these states. Figure 3 plots average child height in all-India and
Panel A. Girls Panel B. Non-eldest sons Panel C. Eldest sons

−0.5 −0.5 −0.5
−1 −1 −1
KE & NE
KE & NE
IND
KE & NE
−1.5 IND −1.5 −1.5
IND
−2 −2 −2
6 7 8 9 6 7 8 9 6 7 8 9
ln(GDP per capita in birth year)
India Kerala and Northeast Africa Africa fitted values
Figure 3. Height of Indian Children Relative to Africa
Notes: The difference between the gap between the Africa fitted line and Kerala and Northeast and the gap between
the Africa fitted line and India is 0.231 for girls, 0.235 for non-eldest sons, and 0.098 for eldest sons. GDP data are
based on the Penn World Table 9.0 (Feenstra, Inklaar, and Timmer 2015).
the matrilineal states in comparison to the Africa sample and shows that both girls
and non-eldest sons fare much better in matrilineal states than the rest of India,
while eldest sons enjoy a much smaller gain.23 In other words, what distinguishes
the matrilineal states, where we see a shallow gradient, from the rest of India is how
not just girls but also non-eldest sons are treated.
C. Implications of Son-Biased Fertility-Stopping Behavior
We now further drill down on the implications of son-biased fertility-stopping

behavior for resource allocation across siblings. The birth of a daughter with no
older brothers causes her parents to exceed their intended fertility to try again for a
son, reducing resources spent on her. Our first test seeks to provide evidence on the
“try again” mechanism, separate from the sibling rivalry mechanism that posits that
having a brother worsens outcomes for a girl because she has to compete with him
for resources (Garg and Morduch 1998).
At the prenatal stage, both mechanisms benefit daughters without an older
brother: there is no sibling rivalry and parents will invest in her while she is in utero
23
Less eldest son preference might also imply that eldest sons do worse in matrilineal states than the rest of
India given a household budget constraint. The fact that they do slightly better likely derives from higher overall
spending on children in matrilineal societies (Lowes 2017). Also, while girls’ and non-eldest boys’ disadvantage
relative to Africa is smaller in matrilineal states than in the rest of India, it still exists (online Appendix Table 6). We
return to this fact while doing the accounting exercises in Section IIIF.
anticipating that, with 50 percent chance, they are investing in their eldest son.24
Post-birth, the negative effects of son-biased fertility-stopping behavior material-
ize as parents reoptimize fertility and expenditure decisions. Thus, at the postnatal
stage, not having an older brother disadvantages girls through the fertility continu-
ation mechanism.
Prediction 3: Relative to African counterparts, later parity girls with no older

brothers in India face larger disadvantages in postnatal than prenatal investments.
Table 6 reports results from regressions estimated at the input-level for the sample
of girls. We consider the set of inputs reported in Table 3 and distinguish between
prenatal and postnatal inputs. In columns 1 and 2 we see a positive and significant
coefficient on I  c  × PrenatalInputs × NoElderBro. This tells us that parents allocate
more prenatal inputs during a pregnancy when they do not have any sons. Strikingly,
this pattern is exactly reversed for postnatal inputs, and we observe a negative and
significant coefficient on I c    × NoElderBro.
Given this evidence, we examine how height deficits vary with the gender of elder
siblings. If the fertility-stopping mechanism dominates the sibling rivalry mecha-
nism, then Indian daughters with only sisters as elder siblings should do relatively
worse than their African counterparts, and vice versa.
Among boys, eldest sons in India should do well, but those born at late parity
may suffer as their parents expended resources on a more-than-planned number of
daughters. A family with desired fertility of two children and an eldest son born at
birth order 1 or 2 need not exceed desired fertility. By contrast, while an eldest son
born at birth order 3 might fare better than his sisters and better than any subse-
quent sons, across families, he might be disadvantaged relative to eldest sons born
at earlier birth order because his family expended resources on his older sisters and
exceeded its desired fertility. Prediction 4 summarizes.
Prediction 4: Relative to African counterparts, outcomes for Indian children

will vary with sibling composition and birth order as follows:
(i) If fertility stopping effects dominate sibling rivalry effects, then later parity
girls with no older brothers will show larger height deficits.
(ii) High birth order eldest sons fare worse than eldest sons born at lower birth
order.
To test (i), we estimate a model based on equation (2), but adding in interactions
with NoElderBro. The coefficient on I c    × NoElderBrocaptures the differential out-
come for a family’s eldest son in India, and the coefficient on I c    × Girl × NoElderBro
captures the differential outcome for a girl in India who either only has sisters as older
siblings or is the first-born. In Table 6, column 3 we observe a positive coefficient
24
Specifically, relative to postnatal investments, prenatal investments for a daughter will be better in a family
that has no son, as it will be based on expected, not realized, gender. Later in this section, we discuss robustness to
prenatal sex determination.
Table 6—Heterogeneity by the Gender of Older Siblings
Pooled Pooled HFA HFA WFA WFA

inputs inputs z-score z-score z-score z-score
(1) (2) (3) (4) (5) (6)
India ×
Prenatal input 0.047
[0.004]
India ×
Prenatal input ×
No elder brother 0.057 0.034
[0.006] [0.007]
India ×
Girl −0.021 0.008
[0.066] [0.050]
India ×
Girl ×No elder brother −0.091 −0.103 −0.088 −0.110
[0.055] [0.058] [0.041] [0.042]
India ×
No elder brother −0.036 −0.025 0.060 0.024 0.059 0.081
[0.007] [0.006] [0.038] [0.040] [0.029] [0.029]
India −0.022 0.088 −0.488
[0.007] [0.047] [0.036]
India ×
2nd child −0.028 −0.034 −0.080 −0.142 −0.098 −0.084
[0.007] [0.005] [0.041] [0.044] [0.031] [0.033]
India ×
3rd+ child −0.120 −0.060 −0.311 −0.208 −0.239 −0.119
[0.007] [0.007] [0.046] [0.056] [0.035] [0.042]
India ×
Girl × 2nd child −0.118 −0.093 −0.086 −0.099
[0.059] [0.064] [0.045] [0.047]
India ×
Girl × 3rd+ child −0.122 −0.122 −0.139 −0.143
[0.065] [0.080] [0.049] [0.058]
Africa mean of outcome 0.385 0.385 −1.351 −1.351 −0.877 −0.877

p-value: India + India ×
No elder brother = 0 0.000 0.000
p-value: India ×
No elder brother + 0.448 0.070 0.343 0.355
India ×Girl ×No elder brother = 0
2nd child + 0.037 0.000
India ×No elder brother = 0
3rd+ child + 0.000 0.000
India ×No elder brother = 0
Sample Girls Girls Children Children Children Children
Age and other controls No Yes No Yes No Yes
Observations 379,055 377,922 168,108 165,596 168,108 165,596
Notes: Standard errors are clustered by mother and appear in brackets. Control variables include child age dum-
mies. Even columns control for maternal age, mother’s literacy, PSU fixed effects, and child age dummies, mater-
nal age, and mother’s literacy interacted with India. Column 2 additionally includes child age dummies, maternal
age, and mother’s literacy interacted with Prenatal input and India ×Prenatal input and PSU fixed effects interacted
with Prenatal input. Columns 4 and 6 includes child age dummies, maternal age, and mother’s literacy interacted
with Girl and India ×
Girl, and PSU fixed effects interacted with Girl. In column 2, the main effects India and India
×Prenatal input are absorbed by PSU fixed effects and their interactions with Prenatal input. In columns 4 and 6,
the main effects India and India ×Girl are absorbed by PSU fixed effects and their interactions with Girl. In col-
umns 1–2, all other main effects (2nd child, 3rd+ child, No elder brother, Prenatal input, Prenatal input × No elder
brother) are included but not shown. In columns 3–6, all other main effects (2nd child, 3rd+ child, Girl, 2nd child
×Girl, 3rd+ child ×Girl, No elder brother, Girl ×No elder brother) are included but not shown. The sample
in columns 1–2 is girls aged 1–59 months, and the sample in columns 3–6 is the main sample of children aged
1–59 months. See the online Data Appendix for further details.
on I c    × NoElderBro ( p-value of 0.12). Adding in the main effect, relative to his
African counterpart, an Indian eldest son enjoys a significant 0.15z-score height
advantage. The coefficient on I c    × Girl × NoElderBroshows that the opposite is
true for girls: having no older brother is worse than having an older brother. The
net effect for girls of lacking an elder brother is negative in India relative to Africa
(I c    × NoElderBro + I c    × Girl × NoElderBro), but insignificant. In column 4, this
finding of lower height for girls in India who only have sisters as elder siblings is
marginally significant ( p-value of 0.07). Thus, the son-biased fertility mechanism
appears to slightly dominate, such that not having an older brother on net disadvan-
tages girls.
Column 3 also allows us to examine whether eldest sons born at later parity
are advantaged as long as they are born within their family’s desired family size.
I c  + I c    × 2nd Child + I c    × NoElderBro , which gives the relative Indian advan-
tage for an eldest son at birth order 2, is positive and significant ( p-value of 0.04).
Meanwhile, an eldest son born at birth order 3 does worse in India than Africa,
which is consistent with Prediction 4(ii), assuming that families want two chil-
dren (the modal preference in India). Columns 5 and 6 show similar patterns using
weight-for-age as the outcome.
In unreported results, we observe a birth order gradient between a family’s sec-
ond and third sons suggesting that our model cannot explain all birth order pat-
terns across siblings. Nonetheless, taken together the observed patterns in the data
point to eldest son preference being an important determinant of resource allocation
across siblings and fertility stopping behaviors—and consequently child height—in
India. In Section IIIF, we provide an accounting exercise that quantifies the fraction
of the India-Africa height gap that can be explained by the eldest son preference
mechanism.
D. Sex Selection as a Potential Confounder
The incidence of sex-selective abortions is higher in India than Africa, rais-

ing concerns about selection into high fertility in India. Wealthier households in
India are more likely to use sex-selection techniques, as measured both by use of
ultrasound and incidence of skewed sex ratios. If poor families who do not have a
son within their desired family size try again for a son, while wealthy families use
sex-selection, then, first, this could cause poor families to be overrepresented at high
birth order in India (relative to Africa). Second, if wealthy households who have
girls at low birth order are particularly likely to engage in sex-selection, then girls
at birth order 2 or 3 may belong to relatively poor households in India, so treating
gender as an exogenous variable is problematic. This could create selection bias
when we include interactions with the child’s or siblings’ gender.
On the first concern, Table 2 already showed that our results are not driven by
differential household selection into high fertility: the India-Africa gap in the birth
order gradient is robust to allowing for differential effects of socioeconomic vari-
ables in India, and importantly to the inclusion of family size fixed effects interacted
with India.
To address the concern that endogenous child gender in India may bias our esti-
mates, we replicate our main results with the first NFHS survey. This survey was
conducted in 1992–1993, which was prior to significant sex-selection by Indian
families (Jha et al. 2011). To create the comparison sample of African countries, we
consider the 18 African countries with DHS surveys between 1991 and 1997. First,
online Appendix Table 7 shows that the differential birth order gradient in height
and weight between India and Africa also holds for this sample, and the magnitudes
are comparable to our main results. Next, online Appendix Table 8 replicates the
results that use the child’s or his or her siblings’ gender. Columns 1 and 4 show an
overall height and weight deficit for girls. We also continue to find a significantly
steeper birth order gradient in India than Africa for both genders (columns 2 and
5). Columns 3 and 6 find some evidence that eldest sons are the most advantaged
in India and having an older brother is a net positive for girls; the point estimate is
similar to the result in NFHS-3 but is statistically insignificant.25
A different worry is whether the fertility-continuation behavior that we posit is a
contributor to the birth order gradient was made obsolete by sex-selective abortion
by the time of NFHS-3. To allay this concern, we show that the “try again” approach
was still prevalent at the time of NFHS-3. Even when sex-selective abortion is avail-
able, it is financially, physically, and psychologically costly, and many families con-
tinue to use son-biased fertility-stopping rules. Online Appendix Table 9 shows this
by examining, first, whether families have gone beyond their desired fertility and,
second, whether they want more children even if they are already at or beyond their
desired fertility. In India, families are more likely to go beyond their desired fertility
or want to go beyond their desired fertility if they do not yet have a son. The prev-
alence of fertility continuation is, as expected, somewhat diminished in NFHS-3
compared to NFHS-1, but importantly, it is still very pronounced in NFHS-3. For
example, not having a son yet is associated with a 19-percentage-point increase in
the desire to go beyond one’s desired family size to have another child in India.
E. Alternative Explanations
We conclude this section by examining a set of alternative explanations for India’s

steep birth order gradient in height. Online Appendix Table 10 reports the results.
Health Conditions
Maternal Health: Indian mothers are, on average, six centimeters shorter than
African mothers. To examine whether maternal health endowment has differential
effects on child height by birth order, column 1 presents our basic birth order regres-
sion adding in interactions between mother’s height and birth order.26 The test is
whether including mother’s height “knock outs” the stronger birth order gradient in
India, and it does not: the coefficients on M other’sHeight × BirthOrder dummies
are small and insignificant, and the steep Indian birth order gradient remains.27
Disease Environment: Spears (2013) highlights India’s high rate of open defeca-
tion as a contributor to child stunting. Even absent changes in a household’s sanitation
infrastructure, later-born children may have a worse disease environment because
25
The very limited data on postnatal health inputs in NHFS-1 makes it infeasible to estimate regressions that
use pooled inputs.
26
This possibility is related to Deaton and Drèze’s (2009) gradual catch-up hypothesis, which posits that if a
mother’s poor nutrition and health as a child, in turn, affect her children’s size, then India’s height gap could take
generations to close.
27
In Jayachandran and Pande (2015), we examined India-Africa differences in how, holding constant number of
children born, maternal nutritional inputs and outcomes vary with a woman’s pregnancy status: relative to African
mothers, Indian women show a greater drop-off in food consumption and hemoglobin levels across successive preg-
nancies. These India-Africa differences are absent among nonpregnant mothers. This suggest that differences in
older siblings expose them to pathogens or because they receive lower-quality care.
Column 2 shows that there is no appreciable birth order gradient for diarrhea in
India. Column 3 directly shows that controlling for the rate of open defecation does
not diminish the magnitude of the India-Africa birth order gradient in child height.
Other Cultural Norms
Communal Child-Rearing: The presence of older siblings will typically reduce

the time parents can devote to later-born infants. This constraint may be less strict
in Africa, which has a strong norm of relatives and neighbors helping raise chil-
dren (Goody 1982), allowing greater investments in later-born children. To test this
hypothesis, we consider two PSU-level “communal child-rearing” proxies: the pro-
portion of women’s children aged ten years and younger who are nonresident in
their household and the number of adult females in the household. While both prox-
ies are higher in Africa, the India-Africa birth order gradient is robust to inclusion of
either proxy (columns 4 and 5).
Land Scarcity: In Africa, where land is more abundant, parents might value a
larger number of children as farm help, and this could imply that early- and later-
born children are more equally valued. This, in turn, could have engendered an
African norm of valuing higher birth order children more. In column 6 we include
the 1961 ratio of population to land area as a proxy for historical land scarcity. By
this metric, while land is indeed more scarce in India than Africa, it cannot explain
why height drops off so steeply with birth order in India.
In sum, we find limited evidence that these alternative explanations can cause
a large differential birth order gradient in height in India compared to Africa.
Moreover, they do not predict several other patterns observed: how height varies
with older siblings’ gender, how health inputs vary with birth order and gender, and
how having an older brother differentially impacts girls’ prenatal versus postnatal
inputs. In this sense, eldest son preference is likely unique in offering a parsimoni-
ous explanation for not just the birth order gradient but also a suite of other facts.
F. Impact on Average Height
The inequality across children in health inputs and outcomes that we document
is important per se if we value equity. But does it also affect the average height gap
across India and Africa, which is an important motivation for our paper?
Our within-India comparisons provide suggestive evidence. There is higher aver-
age height and a shallower birth order gradient for Indian subgroups with less son
preference such as Kerala and the Northeast. A second piece of supportive evidence
comes from the literature documenting diminishing returns to inputs for adult height
(Steckel 1995). While our data do not allow for a rigorous estimation of a similar
height production function for child height, using the IHDS data we find that child
preferences, not a steeper decline in Indian household resources over the life cycle relative to African counterparts,
underlies observed birth order patterns.
height exhibits diminishing returns to household income and expenditures in India

(online Appendix Table 11). We use the IHDS because, unlike NFHS, it provides
measures of family resources that have a cardinal interpretation (income, expendi-
tures). The ideal data, which we lack, are child-level expenditures.
The facts above suggest a link, but leave open the question of how much of the level
deficit is explained by the gradient. To this end, we conduct two b ack-of-the-envelope
calculations to estimate: (i) how much of the India-Africa height gap is explained by
the birth order gradient; and (ii) how much is explained specifically by eldest son
preference generating a birth order gradient.
The first step is to estimate the correlation between the birth order gradient and
level of height across countries in our African subsample. Up until now, our regres-
sions have quantified the birth order gradient via the coefficients on birth order 2 and
on birth order 3 and higher (relative to birth order 1). For our accounting exercise we
need to collapse this information into a single country-specific summary measure,
and we do this in two ways. Our first gradient proxy is defined as the average height
gap between first- and second-borns and between first- and third-borns and higher,
weighted by the observed distribution of birth order in that country. To obtain our
second proxy we estimate, separately for each country, a regression of height on a
linear birth order variable, top-coded at birth order 3, and then use the regression
coefficient as the gradient proxy. The first approach has the advantage that it does
not impose linearity (i.e., that the drop-off in height from birth order 1 to 2 is the
same as the drop-off from birth order 2 to 3+), while the second approach uses a
measure derived from regression analysis, paralleling the analyses presented earlier
in the paper.
Online Appendix Table 2 shows the correlation between the HFA z-score and
each gradient proxy. We use only the African sample so that this calculation is not
“assuming the answer” by comparing India (with its steeper gradient and lower
average height) to Africa. The regression controls for child age dummies and real
GDP per capita in the child’s birth year. Column 2 uses the first gradient proxy, and
we observe a strong positive correlation between the level of height and the gradient
proxy.28 For (i)—quantifying the role of the birth order gradient—we multiply the
coefficient of 0.400 from column 2 by the India-Africa gap in the birth order gradi-
ent. Assuming that Indian children showed the same relationship between average
child height and the birth order gradient as African children, this product provides
an estimate of how much India’s steeper-than-usual birth order gradient depresses
its average height compared to Africa. We can then compare this explained amount
(−0.106 z-score points) to the overall India-Africa height gap (adjusted for GDP per
capita) in column 1 of −0.162 z-score points. This exercise suggests that the birth
order gradient accounts for 65 percent of the Indian height puzzle. When we repeat
the exercise using the second gradient proxy, we again find a significant correlation
between the HFA z-score and the gradient proxy (column 3), and the birth order
gradient accounts for 84 percent of the Indian height deficit.
28
Online Appendix Figure 1 shows the scatter plot of average height and the gradient proxy. Panel A and
panel B correspond to the gradient proxies in columns 2 and 3, respectively. India is roughly on the regression line,
but slightly below, consistent with the birth order gradient explaining much but not all of India’s height deficit in
our accounting exercise.
For (ii), we use the matrilineal states of Kerala and the Northeast as a proxy for
India without eldest son preference and conduct a similar calculation. Using the
first gradient proxy, Kerala and the Northeast have an average gradient that is 0.129
z-score points smaller in magnitude than the rest of India. This gradient difference
multiplied by the gradient-level correlation accounts for a −0.052 z-score level defi-
cit. Thus, the birth order gradient rooted in eldest son preference explains 32 per-
cent of the India-Africa height gap (or 43 percent using the second gradient proxy).
The fact that one-half (32 percent/65 percent with the first gradient proxy and
43 percent/84 percent with the second) rather than all of the birth order gradient
effect is explained by eldest son preference in this exercise could be due to Kerala
and the Northeast having some eldest son preference compared to Africa even
though they have low eldest son preference compared to the rest of India,29 or to
other mechanisms such as those discussed in the previous subsection also contribut-
ing to India’s steep gradient.
These accounting exercises by no means establish a causal link between the gra-
dient and level, but they are suggestive that eldest son preference and the birth order
gradient have quantitatively important implications for average height in India.
IV. Conclusion
This paper compares child height-for-age in India and Africa in order to shed
light on India’s puzzlingly high rate of stunting. Several facts point to intrafamily
allocation decisions as a key factor. First, India’s height disadvantage emerges with
second-born children and increases with birth order. Second, investments in succes-
sive pregnancies and higher birth order children decline faster in India than Africa.
We examine a specific mechanism that could drive India’s steep birth order gra-
dient in child height: eldest son preference. We compare subgroups within India
and show that subgroups with lower son preference exhibit a shallower birth order
gradient. We then derive a set of predictions linking the extent of unequal resource
allocation within a family to the gender composition of siblings and find that these
predictions are supported in the data. We take this as evidence that son-biased
fertility stopping rules are an important factor linking eldest son preference and
the observed birth order gradient in child height. Finally, an accounting exercise
suggests that roughly two-thirds of India’s child height deficit (relative to Africa)
can be explained by India’s steeper birth order gradient. One-half of this can be
attributed to eldest son preference in India.
One might expect that unequal allocation in the household will matter less as
India develops. With greater financial resources, families could provide all children
with enough food and health care to achieve their height potential. The time trend
in India can shed some light on this: between the earlier and recent NFHS waves
of data, the stunting rate in India indeed fell, but it remains very high. Malnutrition
could eventually be an obsolete problem, but India still appears decades away from
this achievement. Moreover, the inequality across birth order has persisted over
time. Parents’ son preference and unequal investment in children do not seem to
29
Online Appendix Table 6 shows that in India’s matrilineal states, girls have a height deficit relative to Africa
and there is a steeper birth order gradient than in Africa.
be diminishing. Even cross-sectionally, richer families in India show a steeper birth

order gradient than poorer ones.30
Thus, even once the problem of chronic malnutrition is solved, other import-
ant human capital investments might remain unequally allocated within families.
This matters, first, because if the investments have diminishing returns, unequal
investment will depress India’s total human capital and economic growth. Second,
intrafamily inequality might widen societal inequality which, in turn, could limit
the economic opportunities available to many people, exacerbate societal dis-
cord, or have other attendant ills.31 Third, and perhaps most important, most
societies value equality per se (Sen 1992). For all these reasons, policies to coun-
teract the intrafamily allocation decisions that parents are making, such as poverty-
alleviation programs targeted toward specific children, could be very valuable. The
need for such policies might be especially strong in India, given the level and per-
sistence of its intrafamily inequality.
REFERENCES
Arnold, Fred, Minja Kim Choe, and T. K. Roy. 1998. “Son Preference, the Family-Building Process
and Child Mortality in India.” Population Studies 52 (3): 301–15.
Barker, David J. P., K. M. Godfrey, P. D. Gluckman, J. E. Harding, J. A. Owens, and J. S. Robinson.
1993. “Fetal Nutrition and Cardiovascular Disease in Adult Life.” The Lancet 341 (8850): 938–41.
Barker, David J. P., and Clive Osmond. 1986. “Infant Mortality, Childhood Nutrition, and Ischaemic
Heart Disease in England and Wales.” The Lancet 327 (8489): 1077–81.
Behrman, Jere R. 1988. “Nutrition, Health, Birth Order and Seasonality: Intrahousehold Allocation
among Children in Rural India.” Journal of Development Economics 28 (1): 43–62.
Behrman, Jere R., and Paul Taubman. 1986. “Birth Order, Schooling, and Earnings.” Journal of Labor
Economics 4 (3): S121–45.
Belmont, Lillian, Zena A. Stein, and Mervyn W. Susser. 1975. “Comparison of Associations of Birth
Order with Intelligence Test Score and Height.” Nature 255 (5503): 54–56.
Bhalotra, Sonia, Christine Valente, and Arthur Van Soest. 2010. “The Puzzle of Muslim Advantage in
Child Survival in India.” Journal of Health Economics 29 (2): 191–204.
Black, Sandra E., Paul J. Devereux, and Kjell G. Salvanes. 2005. “The More the Merrier? The Effect
of Family Size and Birth Order on Children’s Education.” Quarterly Journal of Economics 120 (2):
669–700.
Black, Sandra E., Paul J. Devereux, and Kjell G. Salvanes. 2011. “Older and Wiser? Birth Order and
IQ of Young Men.” CESifo Economic Studies 57 (1): 103–20.
Borooah, Vani K., and Sriya Iyer. 2005. “Religion, Literacy and the Female-to-Male Ratio.” Economic
& Political Weekly 40 (5): 419–27.
Case, Anne, and Christina Paxson. 2008. “Stature and Status: Height, Ability, and Labor Market Out-
comes.” Journal of Political Economy 116 (3): 499–532.
Cavalli-Sforza, L. Luca, Paolo Menozzi, and Alberto Piazza. 1994. The History and Geography of
Human Genes. Princeton, NJ: Princeton University Press.
Chakrabarti, Anindita, and Kausik Chaudhuri. 2007. “Antenatal and Maternal Health Care Utiliza-
tion: Evidence from Northeastern States of India.” Applied Economics 39 (6): 683–95.
30
The analysis behind this finding is reported in Jayachandran and Pande (2015), where we examine
c ross-sectional heterogeneity by wealth within India to infer the likely effect of economic development (given the
caveat that the wealthy may also differ on unobserved characteristics that predict son preference such as caste). We
also find that at each birth order, children in wealthier households (defined as those in the top two quintiles of the
wealth index) are taller than those in less wealthy families, and that even among the wealthy, stunting is prevalent
(the rate is 21 percent).
31
The effects of intrafamily inequality may be distinct from those of general inequality. For instance, it
might translate into differences in bargaining power enjoyed by siblings once they form independent households
(Chiappori and Meghir 2015).
Chiappori, Pierre-André, and Costas Meghir. 2015. “Intrahousehold Inequality.” In Handbook of

Income Distribution, Vol. 2A, edited by Anthony B. Atkinson and François Bourguignon, 1369–
1418. Amsterdam: Elsevier.
Clark, Shelley. 2000. “Son Preference and Sex Composition of Children: Evidence from India.”
Demography 37 (1): 95–108.
Coffey, Diane, Angus Deaton, Jean Drèze, Dean Spears, and Alessandro Tarozzi. 2013. “Stunting
among Children.” Economic & Political Weekly 48 (34).
Deaton, Angus. 2007. “Height, Health and Development.” Proceedings of the National Academy of Sci-
ences 104 (33): 13232–37.
Deaton, Angus, and Jean Drèze. 2009. “Food and Nutrition in India: Facts and Interpretations.” Eco-
nomic & Political Weekly 46 (7): 42–65.
de Onis, Mercedes, Adelheid W. Onyango, Elaine Borghi, Cutberto Garza, and Hong Yang. 2006.
“Comparison of the World Health Organization (WHO) Child Growth Standards and the National
Center for Health Statistics/WHO International Growth Reference: Implications for Child Health
Programmes.” Public Health Nutrition 9 (7): 942–47.
Dyson, Tim, and Mick Moore. 1983. “On Kinship Structure, Female Autonomy, and Demographic
Behavior in India.” Population and Development Review 9 (1): 35–60.
Fall, Caroline, Harshpal Singh Sachdev, Clive Osmond, Maria Clara Restrepo-Mendez, Cesar Victora,
Reynaldo Martorell, Aryeh D. Stein et al. 2015. “Association between Maternal Age at Childbirth and
Child and Adult Outcomes in the Offspring: A Prospective Study in Five Low-Income and Middle-
Income Countries (COHORTS Collaboration). The Lancet 3 (7): e366–77.
Feenstra, Robert C., Robert Inklaar, and Marcel P. Timmer. 2015. “The Next Generation of the Penn
World Table.” American Economic Review 105 (10): 3150–82.
Floud, Roderick, Robert W. Fogel, Bernard Harris, and Sok Chul Hong. 2011. The Changing Body:
Health, Nutrition, and Human Development in the Western World Since 1700. Cambridge, UK:
Cambridge University Press.
Garg, Ashish, and Jonathan Morduch. 1998. “Sibling Rivalry and the Gender Gap: Evidence from
Child Health Outcomes in Ghana.” Journal of Population Economics 11 (4): 471–93.
Glewwe, Paul, and Edward A. Miguel. 2008. “The Impact of Child Health and Nutrition on Education
in Less Developed Countries.” In Handbook of Development Economics, Vol. 4, edited by T. Paul
Schultz and John Strauss, 3561–3606. Amsterdam: Elsevier.
Gneezy, Uri, Kenneth L. Leonard, and John A. List. 2009. “Gender Differences in Competition: Evi-
dence From a Matrilineal and a Patriarchal Society.” Econometrica 77 (5): 1637–64.
Goody, Esther N. 1982. Parenthood and Social Reproduction: Fostering and Occupational Roles in
West Africa. Cambridge, UK: Cambridge University Press.
Gould, Jeffrey B. 1989. “The Low-Birth-Weight Infant.” In Human Growth: A Comprehensive Trea-
tise, Vol. 1: Developmental Biology and Prenatal Growth, edited by Frank Falkner and J. M. Tan-
ner, 391–413. New York: Plenum.
Gupta, Monica Das. 1987. “Selective Discrimination against Female Children in Rural Punjab, India.”
Population and Development Review 13 (1): 77–100.
Guven, Cahit, and Wang-Sheng Lee. 2015. “Height, Aging and Cognitive Abilities across Europe.”
Economics & Human Biology 16: 16–29.
Gwatkin, Davidson R., Shea Rutstein, Kiersten Johnson, Eldaw Suliman, Adam Wagstaff, and Agbessi
Amouzou. 2007. “Socio-Economic Differences in Health, Nutrition, and Population within Devel-
oping Countries: An Overview.” Health, Nutrition and Population Network Informal Paper Series.
Washington, DC: World Bank.
Healy, M. J. R., R. D. Lockhart, J. D. MacKenzie, J. M. Tanner, and R. H. Whitehouse. 1956. “Aber-
deen Growth Study: I. The Prediction of Adult Body Measurements from Measurements Taken
Each Year from Birth to 5 Years.” Archives of Disease in Childhood 31 (159): 372–81.
Hoddinott, John, Jere R. Behrman, John A. Maluccio, Paul Melgar, Agnes R. Quisumbing, Manuel
Ramirez-Zea, Aryeh D. Stein, Kathryn M. Yount, and Reynaldo Martorell. 2013. “Adult Conse-
quences of Growth Failure in Early Childhood.” American Journal of Clinical Nutrition 98: 1170–78.
Horton, Susan. 1988. “Birth Order and Child Nutritional Status: Evidence from the Philippines.” Eco-
nomic Development and Cultural Change 36 (2): 351–54.
International Institute for Population Sciences (IIPS). 2010. District Level Household and Facility
Survey (DLHS-3), 2007–08. Mumbai: International Institute for Population Sciences.
Jayachandran, Seema. 2017. “Fertility Decline and Missing Women.” American Economic Journal:
Applied Economics 9 (1): 118–39.
Jayachandran, Seema, and Ilyana Kuziemko. 2011. “Why Do Mothers Breastfeed Girls Less than
Boys: Evidence and Implications for Child Health in India.” Quarterly Journal of Economics 126
(3): 1485–1538.
Jayachandran, Seema, and Rohini Pande. 2015. “Why Are Indian Children So Short?” National
Bureau of Economic Research Working Paper 21036.
Jayachandran, Seema, and Rohini Pande. 2017. “Why Are Indian Children So Short? The Role of
Birth Order and Son Preference: Dataset.” American Economic Review. https://doi.org/10.1257/
aer.20151282.
Jensen, Robert. 2003. “Equal Treatment, Unequal Outcomes? Generating Sex Inequality through Fer-
tility Behavior.” http://watsoninstitute.org/pub/Jensen_FertPrefs.pdf (accessed June 22, 2017).
Jha, Prabhat, Maya A. Kesler, Rajesh Kumar, Faujdar Ram, Usha Ram, Lukasz Aleksandrowicz,
Diego G. Bassani, Shailaja Chandra, and Jayant K. Banthia. 2011. “ Trends in Selective Abortions
of Girls in India: Analysis of Nationally Representative Birth Histories from 1990 to 2005 and Cen-
sus Data from 1991 to 2011.” The Lancet 377 (9781): 1921–28.
Lowes, Sara. 2017. “ Matrilineal Kinship and Spousal Cooperation: Evidence from the Matrilin-
eal Belt.” https://scholar.harvard.edu/files/slowes/files/lowes_matrilineal.pdf (accessed June 22,
2017).
Oommen, M. A., ed. 1999. Rethinking Development: Kerala’s Development Experience. Vol. 2, New
Delhi: Concept Publishing Company.
Oster, Emily. 2009. “Does Increased Access Increase Equality? Gender and Child Health Investments
in India.” Journal of Development Economics 89 (1): 62–76.
Pande, Rohini. 2003. “Selective Gender Differences in Childhood Nutrition and Immunization in
Rural India: The Role of Siblings.” Demography 40 (3): 395–418.
Price, Joseph. 2008. “Parent-Child Quality Time: Does Birth Order Matter?” Journal of Human
Resources 43 (1): 240–65.
Proos, Lemm A. 2009. “Growth and Development of Indian Children Adopted in Sweden.” Indian
Journal of Medical Research 130: 646–50.
Ramalingaswami, Vulimiri, Urban Jonsson, and Jon Rohde. 1996. Commentary: The Asian Enigma.
New York: UNICEF, The Progress of Nations.
Rosenzweig, Mark R., and T. Paul Schultz. 1982. “Market Opportunities, Genetic Endowments, and
Intrafamily Resource Distribution: Child Survival in Rural India.” American Economic Review 72
(4): 803–15.
Sen, Amartya. 1990. “More than 100 Million Women Are Missing.” New York Review of Books,
December 20.
Sen, Amartya. 1992. Inequality Reexamined. Cambridge, MA: Harvard University Press.
Spears, Dean. 2013. “The Nutritional Value of Toilets: How Much International Variation in Child
Height Can Sanitation Explain?” http://riceinstitute.org/research/the-nutritional-value-of-toilets-
how-much-international-variation-in-child-height-can-sanitation-explain/ (accessed June 22,
2017).
Steckel, Richard H. 1995. “Stature and the Standard of Living.” Journal of Economic Literature 33
(4): 1903–40.
Strauss, John, and Duncan Thomas. 1998. “Health, Nutrition, and Economic Development.” Journal
of Economic Literature 36 (2): 766–817.
Sulloway, Frank J. 1996. Born To Rebel: Birth Order, Family Dynamics, and Creative Lives. New
York: Pantheon.
Tarozzi, Alessandro. 2008. “Growth Reference Charts and the Nutritional Status of Indian Children.”
Economics and Human Biology 6 (3): 455–68.
Tarozzi, Alessandro. 2012. “Some Facts about Boy versus Girl Health Indicators in India: 1992–2005.”
CESifo Economic Studies 58 (2): 296–321.
UNICEF. 2013. Improving Child Nutrition: The Achievable Imperative for Global Progress. New York:
UNICEF.
UNICEF. 2014. Levels and Trends in Child Malnutrition: UNICEF/WHO/The World Bank Joint Child
Malnutrition Estimates. New York: UNICEF.
World Health Organization (WHO) Multicentre Growth Reference Study Group. 2006a. “Assess-
ment of Differences in Linear Growth among Populations in the WHO Multicentre Growth Refer-
ence Study.” Acta Paediatrica 450 (S): 56–65.
World Health Organization (WHO) Multicentre Growth Reference Study Group. 2006b. WHO Child
Growth Standards: Methods and Development. Geneva: World Health Organization.
This article has been cited by:
1. Alberto Alesina, Paola Giuliano, Nathan Nunn. 2018. Traditional agricultural practices and the sex
ratio today. PLOS ONE 13:1, e0190510. [Crossref]
2. Shatanjaya Dasgupta, Erin K. Fletcher. 2017. Paying for Violence? Spousal Abuse and Son Preference
in India. The Journal of Development Studies 1. [Crossref]

Jayachandran and Pande 2017

Uploaded by

Copyright:

Available Formats

Jayachandran and Pande 2017

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

Jayachandran and Pande 2017

Uploaded by

Copyright:

Available Formats

American Economic Review 2017, 107(9): 2600–2629

Why Are Indian Children So Short?

By Seema Jayachandran and Rohini Pande*

Child stunting in India exceeds that in poorer regions like sub-

One in four children under age five, worldwide, is so short as to be classified as

Mean height-for-age z-score

Figure 1. Child Height versus National GDP

I. Background and Data Description

< −6​ (​>​5 or <

Table 1—Summary Statistics

India Africa India Africa

II. Birth Order and Child Outcomes

Basic Finding.—Figure 2 plots the average child height-for-age (HFA) z-scores

where ​​I c​ ​​​is an indicator for Indian children; ​α

Mean height-for-age z-score

Figure 2. Child Height in India and Africa, by Child’s Birth Order

Endogeneity Concerns.—The ideal data for examining differences in the birth

B. Other Health Outcomes

Table 3—Child Health Inputs

Prenatal inputs Postnatal inputs

Observations 115,343 117,686 117,199 167,377 35,888 91,936 122,898 167,724

Columns 5 to 7 analyze three postnatal investments. The prevalence of postnatal

III. Culture and Height Deficits

Table 4—Cultural Norms and Child Height: Within-India Evidence

B. Favoritism toward Eldest Sons and Birth Order Gradients

+ ​δ​3​​ ​I c​ ​​ × Girl × 3rd+ Chil​dimc

+ ​β​3​​ Girl × 2nd Chil​dimc

Prediction 2: The India-Africa height gap will be more pronounced among

Table 5—Child Gender and the Birth Order Gradient in Height

HFA WFA HFA WFA

Observations 168,108 165,596 83,228 165,596 168,108 167,737 83,228 167,737

Panel A. Girls Panel B. Non-eldest sons Panel C. Eldest sons

India Kerala and Northeast Africa Africa fitted values

Figure 3. Height of Indian Children Relative to Africa

C. Implications of Son-Biased Fertility-Stopping Behavior

We now further drill down on the implications of son-biased fertility-stopping

Prediction 3: Relative to African counterparts, later parity girls with no older

Prediction 4: Relative to African counterparts, outcomes for Indian children

Table 6—Heterogeneity by the Gender of Older Siblings

Pooled Pooled HFA HFA WFA WFA

Africa mean of outcome 0.385 0.385 −1.351 −1.351 −0.877 −0.877

Observations 379,055 377,922 168,108 165,596 168,108 165,596

D. Sex Selection as a Potential Confounder

The incidence of sex-selective abortions is higher in India than Africa, rais-

We conclude this section by examining a set of alternative explanations for India’s

Other Cultural Norms

Communal Child-Rearing: The presence of older siblings will typically reduce

F. Impact on Average Height

height exhibits diminishing returns to household income and expenditures in India

be diminishing. Even cross-sectionally, richer families in India show a steeper birth

Chiappori, Pierre-André, and Costas Meghir. 2015. “Intrahousehold Inequality.” In Handbook of

You might also like

< −6 (>5 or <

where I c  is an indicator for Indian children; α

+ δ3 I c    × Girl × 3rd+ Childimc

+ β3 Girl × 2nd Childimc