1

Property Rights and Gender Bias:

Evidence from Land Reform in West Bengal *

12 March 2016

Sonia Bhalotra University of Essex

Dilip Mookherjee Boston University

Abhishek Chakravarty University of Essex

Francisco J. Pino University of Chile

Abstract:

Formalization of land rights can increase productivity and welfare, but may also worsen gender inequality when men have stronger property rights than women. We document male-biased fertility stopping, sex selection, and infant survival improvements flowing from land reforms in the Indian state of West Bengal. The reforms increased infant survival in both Hindu and Non-Hindu families. However, boys’ survival rates increased far more relative to girls in Hindu families, which exhibit greater son preference than Non-Hindu families. The effects persist despite conditioning on post-reform increases in agricultural yields. Regional income growth therefore does not mitigate increased gender inequality following reforms. JEL Classification: I14, I24, J71, O15 Keywords: Land reform, property rights, infant mortality, gender bias

*We are grateful to S. Anukriti, Maitreesh Ghatak, Giacomo de Giorgi, Tarun Jain, Stephan Litschig, Pushkar Maitra,

Giovanni Mastrobuoni, Patrick Nolen, Imran Rasul, Debraj Ray, Sanchari Roy, Alessandro Tarozzi, and participants

at several conferences, seminars, and workshops for their comments. Funding from the MacArthur Foundation

Inequality Network, NSF Grant No. 0418434, and the Centre for Social Conflict and Cohesion Studies

(CONICYT/FONDAP/15130009) is greatly acknowledged. All errors remain our own.

2

1. Introduction

Secure property rights are considered a cornerstone of economic development. Developing

countries with large populations dependent on agriculture have devoted much attention to land

rights. During 1955-2000 a billion people and nearly as many hectares were affected by land

reforms (Lipton, 2009). Extensive research demonstrates the importance of land security in

increasing agricultural productivity, facilitating access to credit, and reducing poverty (E.g. see

Banerjee et al., 2002; Besley & Burgess, 2000; Besley, 1995; Besley & Ghatak, 2010; Besley et al.,

2012; Goldstein & Udry, 2008; Hornbeck, 2010). However, there is little research on the impact

of improved land rights on the gender difference in child survival in developing countries, which

is significantly biased in favour of males (Klasen & Wink, 2002; Sen, 2003). Greater land security

may increase incomes, reducing both male and female child mortality. It may however also cause

the returns to parental investments in children to diverge in favour of sons when rights to land

ownership, cultivation, and inheritance are institutionally male-biased. Increased gender

inequality in child survival is therefore a potential consequence of land reform, which may help

explain why income growth in poor countries may worsen gender inequality.

There is growing evidence that the gender difference in human capital responds to changes in

women’s earnings opportunities and inheritance rights in developing countries (Deininger et al.,

2013; Jensen, 2010; Roy, 2015), and that economic development does not reduce gender

discrimination if male-biased cultural norms persist (Duflo, 2012; Jayachandran, 2014). In this

paper we examine how sharecropper registration during the land reform programme Operation

Barga affected infant mortality by gender, the sex ratio at birth, and gender-biased fertility

stopping. Operation Barga was initiated by the state government of West Bengal in the 1970s to

provide secure and heritable rights to sharecropping tenants by registering their tenancy and

guaranteeing them a minimum share of sharecropped rents. We show that the male advantage in

infant survival and the sex ratio at birth increased after this land reform. We further show that

these results obtain even after conditioning on post-reform increases in regional agricultural

yields, thereby indicating that the results reflect gender bias in property bequest motives rather

than wealth effects. 1 The increased gender gap in survival is concentrated in the Hindu

community, which historically shows greater daughter-aversion in health investments than Non-

Hindu communities (E.g. see Bhalotra, et al., 2010; Bhalotra & Zamora, 2009). Together these

1 While we condition on the effects of increased post-reform regional productivity, and thereby regional income, the institutional mechanisms we identify include increases in household returns to the labour of male children relative to female children, which may evolve independently of regional aggregate income growth.

3

findings indicate that institutional and cultural norms may cause the formalisation of property

rights to widen gender inequality while increasing incomes.

To the best of our knowledge, Almond et al. (2013) is the only other study regarding the impact

of property rights formalisation on gender inequality in the presence of son preference,

specifically examining the impact of land reforms in China on male bias in the sex ratio at birth.

Their paper also finds an increased proportion of male births in response to improved land

rights. Our findings are more diverse, as they identify household responses to land reform in

post-natal discrimination in addition to sex selective behaviour. Our results are also the first to

show that the increased parental bias against girls arises from institutional channels independent

of wealth effects as property rights over land improve.

Our evidence comes from two separate but complementary empirical exercises. We first combine

the 1992-93 and 1998-99 waves of National Family Health Survey (NFHS) data with district

sharecropper registration rates from Banerjee et al. (2002), and use a quadruple difference-in-

differences strategy to ascertain the impact of district-level land reform on infant mortality risk

by child gender, and the gender of the firstborn child in the household. We find that Hindu boys

experience a significant decline in infant mortality risk of 4.0-5.9 percentage points once tenancy

registration rates in their district of birth exceed 50%. Hindu girls with firstborn older brothers

face no parental discrimination, and also experience an infant mortality decline of 5.1-8.6

percentage points following high district registration. Hindu girls with firstborn older sisters

however see no improvement in infant survival after high district registration, with estimates

even suggesting a post-reform increase in mortality of 1.1-2.3 percentage points. Therefore,

Hindu girls without firstborn older brothers appear to face parental discrimination after the

reform, and the child sex ratio amongst Hindu families becomes more male-biased at age 1. In

contrast, amongst Non-Hindu families, we find all children experienced similar declines in

mortality risk of 7.2-7.8 percentage points as a result of the reform regardless of their gender, or

that of the firstborn child. Firstborn children’s gender in India is plausibly exogenous2, as sex

selection by son-preferring households is typically manifested only at second or later births if the

first child is female (Bhalotra & Cochrane, 2010). This allows us to identify programme impacts

without confounding bias from differential mortality trends by child gender across districts that

may correlate with registration rates. The fact that the fall in relative survival chances of girls

were confined to specific combinations of religion, birth order and gender of the first child

2 We explicitly test whether this assumption holds in our dataset, and show in our results that land reform has no statistically significant impact on infant survival, sex ratio at birth, and fertility-stopping at birth order 1.

4

render unlikely the possibility that they were driven by other health policy reforms or unobserved

shocks correlated with the land reforms Our results are also robust to controls for a district-

specific linear time trend and district-level public programmes such as the number of health

institutions and kilometres of surfaced road per capita which would similarly affect Hindu and

Non-Hindu households.

We additionally find that Hindu families increase sex selection in response to land reforms, with

the probability of births being male at birth order 2 or higher increasing by 4.8 percentage points

when the district registration rate exceeds 50%. This probability rises to 5.5 percentage points if

the firstborn child in the family is a girl, which ostensibly increases the parental desire for

following births to be male as land security improves. Hence, the sex ratio at birth in Hindu

families is more male-biased following land reforms, and this male bias is further accentuated by

increased infant mortality of girls without firstborn older brothers as described above.

Consistent with the previous literature, and with our infant mortality results, we find no evidence

of such sex selective behaviour in Non-Hindu families in response to land reforms. We do

however see that Hindu as well as Non-Hindu families practice increased son-biased fertility

stopping in response to land reform. When their first child is male, Hindu families are 10.4-10.8

percentage points likelier to cease childbearing after a second child once district registration

exceeds 25%. However, land reform does not induce fertility stopping amongst these families

when their firstborn child is female. Therefore, the similar decline in post-reform mortality for

Hindu girls with firstborn older brothers and Hindu boys is partially explained by the fact that

these girls have fewer siblings and face less competition for resources. In Non-Hindu families,

district registration above 50% reduces the probability of a third birth by 9.6 percentage points,

and this probability declines more sharply by a further 18.2 percentage points when the first

child in the household is male and registration exceeds 25%. This behaviour is again consistent

with the previous literature, which shows that in India, Muslim households (the overwhelming

majority of Non-Hindu households in our sample) exhibit son preference by altering fertility but

not via postnatal discrimination (Bhalotra & Cochrane, 2010; Bhalotra & Zamora, 2009).

We subsequently seek to corroborate the preceding results and attempt to understand causal

mechanisms better with more disaggregated data concerning household economic status, land

productivity, and land reform implementation at the village level (Bardhan & Mookherjee, 2010;

2011). This is a panel data set from surveys of 2,400 households across 89 villages in West

Bengal, with unique information on household immigrant status, land holdings, and the village-

5

level share of land that was registered under Operation Barga. Using household fixed-effect

regressions of birth-cum-survival, which examine survival probabilities net of fertility changes,

we show that a 1% increase in the share of cultivable village land registered in the previous three

years significantly increased the likelihood of a surviving boy3 in Hindu landless households by

2.5 percentage points in any given year. There was no corresponding increase in likelihood of a

surviving girl in Hindu landless households, and in Hindu landed households there was a

significant decline in the probability of a surviving girl by 1.1 percentage points, indicating that

the increase in male surviving offspring in the household is due to discrimination. These results

are robust to the inclusion of controls for several agricultural support programmes implemented

by the government, and also survive a more stringent specification with district-year fixed effects

that control for time-varying unobservables at the district level. There are no similar effects of

land registration on the probability of surviving boys or girls in Non-Hindu households,

consistent with our other results. The effects are concentrated amongst immigrant landless

households, who were most likely to benefit from sharecropper registration, and native landed

households who were the most likely to lose land privileges after reforms.

All our findings are robust to the inclusion of land rice productivity measures as controls, ruling

out rising income from tenant registration and other programmes such as HYV minikit

distribution as a mechanism for increased gender bias. Rather, biases against women arising from

cultural institutions are the likely mechanisms behind our findings. How such institutions shape

parental preferences in different communities therefore fundamentally determine who gains

from improved land rights. This is in contrast to Almond et al. (2013), which also finds increased

sex selection in favour of sons in China in response to land reform, but does not separate rising

post-reform income and institutional factors as causal mechanisms.

This paper contributes both to the literature on land reforms and on gender bias in developing

countries by establishing a clear connection between them. West Bengal is one of the only Indian

states to have achieved notable success in land reform, and the impacts of Operation Barga on

economic outcomes since its initiation have been closely studied (E.g. see Banerjee et al., 2002;

Bardhan & Mookherjee, 2011; Bardhan, Mookherjee, & Kumar, 2012; Bardhan et al., 2013).

However, we are not aware of any study examining human development impacts of Operation

Barga, with the exception of Deininger et al. (2011) who find that the programme increased

years of schooling of children from poor and lower caste households. Furthermore, India has

3 A surviving child is defined as a child observed in the household at the time of the survey.

6

high prevailing son preference. There is substantial research documenting parental discrimination

against girl children in postnatal health investments in India, which in combination with sex-

selective abortion has significantly skewed the sex ratio (E.g. see Anderson & Ray, 2010; Babu et

al., 1993; Barcellos et al., 2013; Behrman, 1988; Bhalotra, 2010; Bhalotra & Cochrane, 2010;

Borooah, 2004; Chakravarty, 2010; Jayachandran & Kuziemko, 2011; Murthi et al., 1995; Oster,

2009; Rose, 2000). However, there is a paucity of evidence concerning the role of property

inheritance norms in generating gender imbalance, despite widespread discussion of this issue

(Bardhan, 1974; Rahman & Rao, 2004; Jain, 2014; Carranza, 2014).

The rest of the paper is organised as follows. Section 2 provides a background discussion of

land reform in India, Operation Barga in West Bengal, and prevailing son preference norms.

Section 3 presents the proposed mechanisms of impact that motivate the empirical analysis. In

Section 4 we discuss the data that we use in our estimations. Section 5 outlines our empirical

methodology, and in Section 6 we present our results. Section 7 concludes.

2. Background

2.1 Historical Context

Upon national independence in 1947, the Indian central government began three main types

of land reforms to address large historical inequalities in land distribution. These were abolition

of intermediaries, new tenancy laws to protect against eviction and extraction of excessive rental

crop shares by landlords, and land ceilings to limit the amount of land held by any one

household with the aim of vesting and redistributing surplus land to small farmers.

Implementation of the reforms was left to individual state governments, and barring

intermediary abolition in nearly all states landlords were able to subvert the remaining reform

measures by way of pre-emptive tenant evictions and parcelling land to relatives to avoid state

confiscation of above-ceiling holdings (Appu, 1996). Variation in state-level reform

implementation and legislation over time has been used in previous studies to empirically

estimate land reform impacts on poverty, equity, and human capital (Besley and Burgess, 2000;

Ghatak & Roy, 2007; Ghosh, 2008).

Reforms in the state of West Bengal were spurred by the results of the 1977 state assembly

election. The Left Front coalition won an absolute majority, which it retained until 2011. This

new government promptly created a three-tier system of local governments called panchayats,

7

which for the first time would be democratically elected. These tiers in descending order of size

of jurisdiction were district, block, and finally the gram panchayat that operated at the village level

with a jurisdiction of 10-15 hamlets (mouzas). Many national development programmes as well as

aspects of new state welfare initiatives such as Operation Barga were then decentralised to these

gram panchayats, who were largely responsible for selecting local eligible beneficiaries and lobbying

the upper tiers of the new system for funds (Bardhan & Mookherjee, 2011).

2.2 Operation Barga and the Green Revolution

West Bengal, along with Kerala, is an exceptional state in terms of the effort with which the

state government pursued land reforms. The Left Front implemented Operation Barga

rigorously to consolidate its vote base among small farmers, leading to higher sharecropper

registration in areas where it faced greater competition in newly instituted local elections

(Bardhan & Mookherjee, 2010). Registration gave sharecroppers permanent, hereditary tenancy

rights, and limited the share of the crop payable as rent to landlords to 25 percent.4 By 1981 over

1 million sharecropper tenants were registered, and almost 1.5 million by 1990 (Lieten, 1992).

Estimates of the fraction of sharecroppers registered in the state range from 45% (Bardhan &

Mookherjee, 2011), to 65% (Banerjee et. al., 2002), to as high as 80% (Lieten, 1992).

Besides Operation Barga, the state also aimed to vest land held by households above the

stipulated ceiling of 12.5 acres and redistribute it to the landless and small landowners in small

plots (or pattas). Most vesting of land had already taken place by 1978, so the Left Front

government’s main role was in redistributing this land. Appu (1996) estimates that 6.72 percent

of state operated area was distributed by 1992; several times the national average of 1.34 percent.

However, this land was redistributed in small plots (less than half an acre on average in the

sample of farms in Bardhan & Mookherjee (2011)), and was of low quality for cultivation as

landlords would only part with their lowest quality above-ceiling holdings. Hence unlike tenant

registration, land redistribution had virtually no impact on agricultural productivity.

Importantly, there were other government rural initiatives launched in the state at the same time

that were aimed at boosting agricultural productivity and reducing poverty. Alongside Operation

Barga, the state government also distributed minikits containing high yield variety (HYV) seeds,

4 This share rose to 50 percent if landlords provided all non-labour inputs.

8

fertilisers, and insecticides to farmers throughout the state via gram panchayats.5 Land reform in

combination with minikit distribution led to a substantial increase in agricultural yields in West

Bengal over the 1980s, transforming the state into one of the best agricultural performers in the

country and leading this period to be called West Bengal’s Green Revolution. This period is also

associated with significant declines in poverty and growth in rural employment. Banerjee et al.

(2002) attributed the increase in yields to land reform, citing decreased Marshall-Mill

sharecropping distortions from increased tenancy security. Bardhan & Mookherjee (2011)

however shows that while decreased inefficiencies played a role in increasing yields, it was largely

minikit distribution that was responsible for the agricultural growth in this period. 6 Other

programmes administered in the 1980s with gram panchayats targeting local beneficiaries include

the Integrated Rural Development Programme that provided subsidised credit, and employment

initiatives such as the Food for Work programme, the National Rural Employment Programme,

and the National Rural Employment Guarantee Programme.

2.3 Community Differences in Son Preference

The majority Hindu community in India traditionally exhibits greater son preference than

other religious communities, as evidenced by conditional sex ratios in the population and

empirical evidence on child mortality and education that reflect childhood parental investments

(Bhalotra & Zamora, 2009; Bhalotra & Cochrane, 2010; Bhalotra et al., 2010). The literature in

this regard has focused on Hindu-Muslim differences, as other religious communities make up a

very small part of the population.7

While no definitive explanation has been agreed upon for the differing degrees of son

preference between the Indian Hindu and Muslim communities, existing arguments such as the

Dyson-Moore hypothesis base them in marital institutions and inheritance practices. In North

India including West Bengal, Hindu marriage is exogamous for women, who leave their natal

family village to marry into families in villages much further away to avoid marrying a possible

relative. The distance from natal family after marriage reduces Hindu women’s bargaining power

and also their claim to natal family land, which is seen as bringing no reciprocal benefit and lost

5 The crops for which seeds were distributed were rice, potatoes, oilseeds, and some other vegetables according to Bardhan & Mookherjee (2011). 6 A companion paper Bardhan, Mookherjee, & Kumar (2012) also shows that tenancy reform crowded in large private investments in irrigation, the growth-inducing effects of which were far greater than those of reduced Marshall-Mill distortions. 7 We do the same in this section, as Hindu and Muslim children constitute about 98% of our estimation sample.

9

to the family when daughters inherit. Sons on the other hand care for parents and natal family

members in their old age by remaining with the natal family and working the family land,

eventually inheriting it upon the death of the family patriarchs. Cultural taboos against Hindu

women sharing public spaces with men and working agricultural land also often prevent them

from claiming and cultivating land (Agarwal, 2003). The bridal dowry practice also often entails

loss or mortgage of family land at the time of a daughter’s marriage. With regard to Operation

Barga specifically, Gupta (2002) finds from interviews of 870 households in two West Bengal

districts that 99% of households reported dowry being a serious concern, and that mortgaging

barga land to meet dowry payments was a common practice. She also finds that dowry was largely

a Hindu practice, but that the custom has penetrated younger generations of Muslims.

Under the Mitakshara Hindu doctrine followed in North India, women in fact have no claim to

joint family property, whereas men are entitled at birth to a share of such family property held by

their fathers, paternal grandfathers, and paternal great-grandfathers.8 In South India close-kin

marriages are more prevalent for Hindu women, allowing them to inherit a greater share of

ancestral land despite prevailing Mitakshara doctrine as they reside close enough to participate in

cultivation on natal family land after marriage. These marital institutions have been used to

explain more favourable female-male sex ratios in South India compared to North India

(Chakraborty & Kim, 2010). In West Bengal the Dayabhaga Hindu system of inheritance is

followed where the concept of joint family property is absent, and all of a Hindu male’s property

is subject to equal claims by his widow, sons, and daughters upon his intestate death (Lingat,

1973). While this appears more gender-equal than the Mitakshara system in theory, in practice

Hindu women nearly always relinquish their inheritance claims to their brothers and sons so as to

avoid social exclusion, intimidation, and losing the family safety net in times of financial crisis

(Agarwal, 2003). Hindu upper caste women also do not physically work agricultural land due to

prevailing social norms. Lower caste women have higher work-force participation rates in

agriculture as wage labourers, but still female employment rates in agriculture in the state have

been persistently low.9 Hindu women therefore are very much financially dependent on their

male kin, leading them to give up their rights to family land to avoid losing that support.

8 Some Indian states have since made reforms to the Hindu Succession Act of 1956 to give women equal inheritance rights to joint family property, but these reforms still explicitly exclude agricultural land from their purview. 9 In the 1991 Census of India only 11.1 percent of women in West Bengal reported having any form of employment, and only 54.1% of the employed women were cultivators or agricultural labourers. National Sample Survey data also reflects decreasing female rural employment and increased casualisation of female agricultural labour since the late 1980s.

10

Muslim communities follow inheritance practices based in the Shariat, which guarantees women

at least half as much inheritance as their closest male counterpart inheritors. Consanguineous

marriage is also practiced to keep all ancestral property within the family, allowing Muslim

women to remain close to their natal families after marriage and inherit more family property in

practice similar to Hindu women in South India.10 Marital dowry is also less prevalent among

Muslims, and abortion, sex selective or otherwise, is strictly forbidden under the Shariat. The

effect of these institutions arguably reduces parental neglect of Muslim female children

compared to Hindu female children in many parts of the country including West Bengal, despite

the fact that the Muslim minority population experience nationally higher levels of poverty than

the Hindu majority and Muslim female labour force participation in West Bengal is even lower

than that of Hindu women (Nasir & Kalla, 2006; Chakraborty & Chakraborty, 2010).

3. Causal Impacts and Identification

3.1 Mechanisms of Interest

The preceding discussion of marriage and inheritance norms suggest Operation Barga could

intensify male-biased fertility and child investment patterns, especially for Hindu families whose

land endowments increased as a result of the reforms. Beneficiaries of the programme and

agricultural tenants in general benefited from increased land security and a greater share of

agricultural output due to the programme, while landlords faced reduced land security and rents.

As outlined previously, the programme also created indirect general equilibrium effects in the

economy that affected infant mortality rates. The programme increased land productivity across

all farm sizes, albeit without increasing farm employment or wages. It also activated the land

market, with large landowners increasing sales of their holdings to small landowners11. Both the

direct effect of changed property rights and the indirect wealth effects of increased productivity

arising from the programme induce their own substitution and income effects that alter

household desired fertility and health investments in their children. We condition on productivity

increases in our empirical analysis, and focus on the net impact of the combined substitution and

income effects on infant mortality arising from changing property rights. While the income effect

will raise parental values of both boys and girls, the substitution effect is likely to be biased in

favour of sons, and particularly so amongst Hindu households who will not perceive girls to be

10 Bittles (2002) reported that 23% of Muslims in India practiced consanguineous marriages in 1992–1993. 11 The selling and exchanging of land for marriage expenses and dowry of Hindu daughters formed a significant part of the land market transactions brought to life by Operation Barga, presumably increasing son preference (Gupta, 1993; Kodoth, 2005).

11

“close” substitutes for boys. Also in the presence of primogeniture, and from empirical evidence

on the patterns of how son preference manifests in Indian sex ratios, we anticipate that the

substitution effect is most evident amongst Hindu families who do not have a firstborn son.

It will be helpful to clarify the nature of wealth and substitution effects in parental preferences

and how these may vary with household and village characteristics. We can represent the value

placed on child i by parents in household j in a given village as follows,

vi = (K - δ1 gi) exp[(α – δ2 gi)Wj + π(1 – fj )(1- gi) θ(β){lj + βt(lj)}] (1)

where gi is a dummy variable taking value 1 if child i is a girl. δ1 > 0 is an intrinsic gender bias

against girl children. Wj is household wealth, α > 0 is a common wealth effect on child values,

and δ2 >0 is a gender bias in the wealth effect. Household wealth Wj is constituted as,

Wj = θ(β){lj + βt(lj)} + F (2)

where θ is village land productivity, which is an increasing function in the fraction β of cultivable

land that is tenanted and registered under Operation Barga. lj is land owned by the household,

and t(lj) is net land leased by the household. F is household non-land wealth. Households owning

more land lease in less, so t is a decreasing function of l with t(0) > 0, and t(l*) = 0 for some l* >

0. Therefore those with land holdings less than l* lease in land from those with land holdings

greater than l*. 12 The parameter π represents the ‘substitution’ effect (which is larger for

Hindus), that is the added value of having a boy to inherit the household tenancy rights valued at

θ(β){lj + βt(lj)}. The substitution effect manifests only if the household has no firstborn son, i.e.

the dummy variable for the first child in household j being male, represented by fj, equals zero.

The general equilibrium effect of Operation Barga in the village comes through increased village

land productivity β, which applies uniformly for all landowners. A partial equilibrium property

rights effect is represented by the dependence of lj + βt(lj) on β, which is increasing in lj for those

owning less than l* and decreasing in lj for those owning more. We assume that landless

households form the bulk of sharecropping tenants that benefit directly from registration. 13

12 As a further restriction, land cultivated by the household is l + t(l) which must be non-negative and increasing in l, so the t function must have a slope less than one in absolute value. 13 Bardhan et al. (2014) reports that 80% of surveyed households were landless or marginal landowners.

12

For these landless households, Operation Barga unambiguously raises household wealth and

value placed on children through a direct improvement in property rights on rented land and

increased village land productivity β. The curvature of the logistic function in (1) also means that

for a unit increase in wealth, there is a larger increase in child value the smaller the size of the

household’s landholding. However, due to the gender bias parameters δ1 and δ2 the value placed

on boys will increase more than for girls, especially in Hindu families.

For landowners with holdings greater than l*, the sign of the wealth effect is ambiguous as the

general equilibrium effects and property rights effects operate in different directions: we cannot

sign the slope of θ(β){lj + βt(lj)} with respect to β, unless the θ function is assumed to be rising

fast enough. Likewise the substitution effect cannot be signed. In any case, the curvature of the

logistic function implies that a unit increment in wealth will lead to smaller increases in child

value for this category of households than for landless households. 14 Additionally, larger

landowning households are wealthier than the landless, and therefore place a higher value on

children to begin with.

Increasing values placed on children will translate into reductions in infant mortality rates.

Effects on fertility are more complicated, owing to possible quantity-quality trade-offs of the

sort emphasized by Becker and Lewis (1973). Son preference and male-biased inheritance

patterns also may lead to a fertility decline that manifests through son-biased fertility stopping,

which would increase infant survival amongst children who grow up in smaller families.

3.2 Other Programmes and Identification

Bardhan, Mookherjee, and Kumar (2012) find that sharecropper registration crowded in

significant private irrigation investments that triggered large productivity spill-overs across both

tenant and non-tenant farms. Hence Operation Barga potentially had large indirect impacts on

small cultivators and wage labourers who were not involved in sharecropping as landlords or

tenants. The 1980s period of accelerated rice and foodgrains production in West Bengal also saw

disproportionate increases in consumption expenditure by the poor and reduced consumption

inequality (Chattopadhyay, 2005). Reduced groundwater costs from irrigation investments

undoubtedly played a role in this period of high agricultural growth and labour-absorptive

14 We include marginal landowners with lj less than l* in the same category as large landowners in the empirical analysis. This allows for clean estimation of the effects of improved property rights on infant mortality among the landless without confounding wealth effects from landholdings, and serves only to bias estimated coefficients towards zero among landed households.

13

expansion. As men were the primary wage labourers and small cultivators in the state, their

increased incomes from registration-induced irrigation investment could have increased son-bias

in child health investments among these households if the substitution effect dominated the

income effect of higher labour returns on these investments.

HYV minikit distribution was however the likely primary driver of yield increases and

agricultural growth in West Bengal in the 1980s as noted earlier. The simultaneity of this

programme with sharecropper registration requires us to account for the effects of increased

land productivity on son preference to empirically identify the impact of registration. We

therefore use an estimation approach that conditions on the effects of regional land productivity,

while estimating the impact of sharecropper registration on infant mortality risk.15 By doing this

we control for the combined income and substitution effects of any yield increases in the

aggregate economy via HYV minikit distribution and other programmes on son-bias in parental

health investments that determine infant mortality. At the same time, conditioning on

productivity effectively conditions on income and substitution effects from increased regional

agricultural output arising from sharecropper registration as well, whether from crowded-in

private irrigation investments or increased returns from improved input use on sharecropped

land. This allows us to isolate the net effect of sharecropper registration on infant mortality

through causal channels which are based in institutional preference structures and property rights

improvements, independent of the effects of increased regional yields.

From the discussion thus far, we can make a number of predictions about what the empirical

analysis should uncover. Averaging across households in the village, if there are enough landless

and marginal landowning households relative to the landed, the average effects will be

qualitatively similar to those for the former category. Specifically, village average infant mortality

amongst boys will decline by more than amongst girls. These results will be robust to controls

for productivity, and will be more pronounced for Hindu households. Disaggregating across

household categories, these effects are expected particularly for the landless, and for Hindu

families without a firstborn son. We can further predict that the sex ratio at birth will become

more male-biased among Hindu families without firstborn sons, and that fertility stopping will

become more son-biased in response to land reform.

15 Rice is the major crop in West Bengal, accounting for more than 70% of gross cropped area consistently during 1971-1991 according to state government economic reviews.

14

4. Data and Descriptive Statistics

4.2 NFHS Survey Data

The first dataset we use is the National Family and Health Survey (NFHS). The survey uses

interviews with women aged 15-49 to collect information on their fertility history, their children’s

health and mortality outcomes, household wealth, and their own health and education.

Households are chosen for interview using stratified random sampling, with the probability of

selection weighted by the area population as per the last Indian national census. These data

provide us the advantage of being able to exploit information on children’s district and year of

birth, gender, and birth order to analyse land reform impacts across districts and time by child

gender, and by whether the firstborn child in the household is male or female. We use the 1992-

93 and 1998-99 waves of NFHS data for West Bengal, and match surveyed women’s district of

residence to available data on annual district rates of sharecropper registration from 14 districts.

The registration data is from Banerjee et al. (2002), kindly granted to us by the authors.

Panels A and B of Table 1 outline descriptive characteristics of children in the NFHS sample

born during 1967-93 and their mothers. Panel A shows that the 6,468 mothers in this sample

have 3.42 years of education on average, and have an average of 3.39 births. 75% of mothers are

Hindu, and 20.5% of them belong to a scheduled caste or tribe. 68% of mothers reside in rural

areas, and the average age at which they give birth is 19.03 years. Panel B shows that among the

20,148 children born during these years, the neonatal and infant mortality rates are 6.4% and

9.4% respectively. The probability that a child is male is 51.1%, and the probability the child has

a younger sibling is 71.8%.

We further use district-level data on yields and area under cultivation of rice in West Bengal from

the ICRISAT Village Dynamics in South Asia (VDSA) database to construct measures of annual

district rice productivity in thousands of tonnes of output per one thousand hectares.16 We also

collect information from the annual Economic Survey reports of the West Bengal government

to control for the effects of other programmes and infrastructure in our 14 districts. We

specifically use information on the annual number of medical institutions per capita, kilometres

of surfaced roads per capita, and hectares of patta land distributed per capita in each district.

Panel C of Table 1 shows descriptive statistics of annual district rice productivity for our 14

16 We construct and control for similar productivity measures of other cereals, but we do not report these coefficients as these crops form a very small part of total crop yields in West Bengal as described in Section 3.

15

sample districts, as well as annual measures of these other district healthcare coverage and

infrastructure measures for the years 1977-1990.17 On average, 1.47 thousand tonnes of rice were

produced per thousand hectares in each district over this period. Also on average each individual

had access to 0.21 kilometres of surfaced road and 0.06 medical institutions in each district over

the same period. We condition on the logs of district rice productivity and of the district

healthcare and infrastructure measures explicitly in our regressions to ensure that our results are

in fact capturing the impact of land reforms rather than alternative policy measures.

Figure 1 shows the evolution of the mean registration rate across our 14 districts over time.

Sharecropper registration appears to have been occurring most quickly up to 1985, after which

the pace of registration slows considerably and remains slow into the 1990s. We exploit this fact

in our robustness checks, when we verify that sharecropper registration was not targeted to

districts with higher pre-reform trends in the female-male difference in mortality and sex ratio at

birth, or gender-biased fertility stopping. In regressions using this registration data, we restrict

the sample to cohorts born after 1978 as there is no positive registration recorded in the data

before this year, despite registration having taken place under the previous government. The final

cohort in our sample is those children born in 1991, as we do not have information on the other

district-level programmes besides land reform outlined above after this year.

4.3 Village Survey Data

Our second dataset is compiled from primary survey data collected from household heads in

West Bengal, regarding their family history, land ownership, immigration status, and other

household characteristics since 1967.18 The survey questionnaire elicited information from the

head about all members residing in the household in 2004, and the year they were born or joined

the household. The survey reports the births of children in the household who have survived till

2004. Hence our estimates using this data will capture the combination of both household

fertility responses and as well as changes in child survival probability, or birth-cum-survival, in

response to land reform. We do not disaggregate land reform impacts on birth-cum-survival by

the gender of the firstborn child as we do with the NFHS data, as non-surviving children are not

reported in survey, making inferring birth order difficult.

17 These are the years for which these measures enter the regressions as controls; see Section 5. 18 Further details on this survey data are in Bardhan & Mookherjee (2010), Bardhan & Mookherjee (2011), and Bardhan et al. (2014).

16

The survey data was used to construct a household panel for a sample of 2,400 households from

89 villages. For approximately two-thirds of this sample, a consistent history of household

landholdings and demographics could be constructed (comprising the ‘restricted sample’). For

the rest a consistent history could be constructed under specific assumptions of nature of recall

errors. Accordingly, while we report only results from the restricted sample, we verify

independently that the results do not differ qualitatively in the full sample.19

We further combine the household survey data with data on land reforms implemented in each

of the 89 surveyed villages during 1971-2003 for both the land distribution and tenant

registration arms of the land reform programme, collected directly from Block Land Records

Offices. This data on land reforms is of significantly higher quality than the district-level

registration rates in Banerjee et al. (2002) that we match with the NFHS data, as it is compiled

firsthand from official land records rather than from indirect sources. Further, it directly

measures cultivable village land affected by land reform every year, rather than rate of

registration among individual sharecroppers which may not accurately capture the extent of

sharecropping tenancy and sharecropped land in a given geographical region. As such, this data

is a unique resource on land reforms in West Bengal. As a further advantage, this data allows us

to measure the progress of both the distribution and tenant registration arms of the land reform

programme in the actual village where surveyed households reside, which is a much more direct

and local measure of household exposure to land reform than registration rates reported at the

more aggregate geographic and administrative level of the district.

We measure the extent of reform activity as percent village cultivable land distributed to

households in small plots (or pattas), and percent cultivable village land registered as under

sharecropping (under barga) in the three years prior to a reported surviving birth in a household

residing in that village. The survey data show that approximately 15% of surveyed households

had received patta land by 1998. However, the distributed plots were small and of poor quality,

and were also not eligible to be used as collateral for subsidised credit. Land distribution was

therefore largely ineffective in increasing rice productivity. In contrast, plots registered under

barga were of a much larger size (1.5 acres on average), and could be used as collateral for loans

from state financial institutions, therefore yielding much greater positive impacts on rice

productivity than patta land distribution (Bardhan & Mookherjee, 2011). The survey data show

that about 48% of sharecropper tenants, or 6% of surveyed households, had been registered by

19 The results from the full sample are in Table A.8 in the Appendix.

17

the late 1990s. Figure 2 shows the average share of village cultivable land registered over the

years 1867-1998 for the 89 surveyed villages. As in Figure 1, we see that registration peaks in

1985. The decline in average share of cultivable land thereafter is due to the slowdown in

registration, combined with an increase in village cultivable land on average over the 1980s.

Empirical Strategy

5.1 DHS Data: Quadruple Difference-in-Differences

Using the NFHS surveys and the accompanying data on sharecropper registration, we exploit

the variation in registration intensity across districts along with variation in child gender and the

gender of the firstborn child in the household in a quadruple difference-in-differences strategy

to identify programme effects on infant mortality. We implement the following specification,

yijt = α + β1 SCROP50jt-1 * FIRSTSONi * FEMALEi

+ β2 SCROP25 jt-1* FIRSTSONi * FEMALEi + δ1 SCROP50 jt-1 * FEMALEi + δ2 SCROP25 jt-1* FEMALEi + η1 SCROP50 jt-1 * FIRSTSONi + η2 SCROP25 jt-1* FIRSTSONi + φ1 FIRSTSONi + φ2 FEMALEi + φ3 FIRSTSONi * FEMALEi + γ1 SCROP50jt-1 + γ2 SCROP25jt-1 + λ Xit + dt + θj + εit (3)

where yijt is a dummy variable taking value 1 if child i born in district j in year t dies aged 0-12

months and 0 otherwise when we are considering infant mortality. We define dummy variables

SCROP25jt-1 and SCROP50jt-1 which take value 1 if sharecropper registration rate in the district j

where child i resides reaches at least 25% or 50% respectively in the year preceding his birth year,

and 0 otherwise. These are our measures of low and high intensity sharecropper registration.20 It

is worth noting that these measures are such that SCROP25jt-1 is always equal to 1 when

SCROP50jt-1 is equal to 1, so that high intensity sharecropper registration captures a cumulative

effect of programme intensity on infant mortality. The omitted category of children constitutes

those born in districts where registration is less than 25% in the year preceding birth. We further

define a gender indicator FEMALEi taking the value 1 if child i is a girl, and the indicator

20 We chose these levels of registration as cut-offs for our treatment indicators based on estimates from a flexible specification that tested for significant effects of cumulative sharecropper registration rates in 10 percent increments. These results are available from the authors upon request.

18

FIRSTSONi taking value 1 if the firstborn child born to the mother of child i is male. The

regression is limited to children of birth order 2 or higher.

Table 2 shows the evolution of sharecropper registration rates for the 14 districts in our sample.

The light grey cells indicate the year that 25% registration of all sharecroppers is achieved, and

dark grey cells indicate the year when 50% registration is achieved. There is a lot of variation

across districts in the years in which these thresholds are achieved in programme implementation

that we exploit in specification (3). This also alleviates concerns that these years coincide closely

across districts, which would mean we have little dynamic variation in implementation and our

estimates are largely driven by cross-sectional variation across districts. We also report results

from estimating (3) including children born in border districts in the neighbouring state Bihar as

a control group in the regression sample, as these children are never exposed to land reforms.21

The β parameters on the firstborn son-female child-registration rate interaction terms will

therefore capture the differential impact of sharecropper registration intensity on infant

mortality of female children with firstborn older brothers. These impacts are over and above

registration impacts on infant mortality of girls with firstborn older sisters captured in δ

parameter estimates, and on the infant mortality of boys with firstborn older brothers captured

in η parameter estimates. The programme effects for boys with older sisters are captured in the γ

parameter estimates, and the untreated counterpart estimates are captured in the φ estimates. The

four dimensions across which we are taking differences to achieve identification are therefore

districts, year of birth, child gender, and the gender of the firstborn child in the household. The

impacts are identified independently of child birth year and district fixed effects captured in

dummies dt and θj. We also present estimates from mother fixed effect specifications, and after

including a district-specific linear trend in child birth year to control for district trends in

unobservables that may simultaneously determine sharecropper registration rates and infant

mortality risk. The covariate vector Xit includes indicators for child birth order, indicators for

household religion and caste, whether the household is rural, mother’s educational attainment,

and finally linear and quadratic terms of the mother’s age at which the child is born. εit is an

idiosyncratic error term.

21 These specifications do not include district health and infrastructure programme regressors, as these are not available for Bihar.

19

To explicitly control for the effects of minikit distribution and similar initiatives on infant

mortality via increased agricultural acreage and yields, we estimate specifications including the log

of district productivity of rice (LN RICE PRjt-1) in the year prior to the child’s birth as a

regressor, along with the corresponding interaction terms with the firstborn son and female child

indicators. To further control for any confounding effects of public health improvements,

infrastructural development, and the other arm of the land reform, we also include controls for

the logs of medical institutions per capita, kilometres of surfaced road per capita, and hectares

of patta land distributed per capita in the district in the year preceding the child’s birth along with

the necessary interaction terms.

Differential fertility-stopping by child gender may have increased due to sharecropper

registration, as discussed in Section 3. Again exploiting the plausible hypothesis that the gender

of the firstborn child is random, we re-estimate (3) with the new outcome variable indicator

taking value 1 if child i has a younger sibling and 0 otherwise. However, in this specification we

are only concerned with how having a firstborn son alters fertility-stopping as sharecropper

registration increases, and not how this varies with the gender of child i. We therefore do not

include the interaction terms of the firstborn son indicator with the female child indicator, the

two-way interaction terms of the female child indicator with the registration intensity indicators,

or the three-way interaction terms of the female child, firstborn son, and registration intensity

indicators. We also estimate these specifications sequentially for separate samples of children by

ascending birth order, so as to isolate the level of total fertility at which households alter

childbearing in response to land reform.22

If households also alter their sex selective behaviour in response to land reforms, it would

endogenously change the sample composition of births we consider in the infant mortality

regressions, and potentially worsen the child sex ratio at birth; an even earlier stage than infancy.

To examine whether this is the case, we use the same specification as for the fertility-stopping

analysis, and estimate it for the outcome variable taking value 1 if child i is male and 0 otherwise.

These estimations will reveal if sex selection and postnatal discrimination act in a mutually

reinforcing fashion in response to land reforms.

22 The results show no impact of land reform on fertility-stopping after the first birth, and are in Appendix Table A.1. The first statistically significant effects are on the probability of births following the second birth, so we do not report results for births at birth order three or higher, as these samples are endogenously selected.

20

We estimate (3) and its variations for fertility-stopping and the probability of male births using

OLS on the sample of children of birth order two or higher born during 1978-91.23 We carry out

separate estimations for Hindu and Non-Hindu children to account for the different preference

structures over child gender between communities. As there are only 14 districts, we report

results with wild cluster-bootstrapped standard errors as recommended in Cameron et al. (2008),

using the procedure outlined in Busso et al. (2013).

5.2 Test for Targeting of Sharecropper Registration

Our results are subject to reverse causality bias if sharecroppers were more rapidly

registered in districts where families, particularly Hindu families, historically discriminated more

against daughters when they did not have firstborn sons, committed more sex selection, or

engaged in greater son-biased fertility stopping than in other districts prior to Operation Barga.

To test for whether such targeting of registration took place, we assign districts as “treated” or

“untreated” if they had achieved above or below-median levels of registration in 1985

respectively. This exploits the fact that registration occurred at its fastest pace during 1980-1985,

as we see in Figure 1. We then regress our outcomes amongst children of birth order 2 or higher

born before the programme during 1958-77 on a linear time trend in years, interacted with the

above-median registration treatment indicator in 1985. This reveals if pre-programme trends in

our outcomes of interest in a district correlate with that district becoming a “treated” district in

the future. We further disaggregate this correlation between pre-reform trends in outcomes and

registration intensity in 1985 by child gender and gender of the firstborn child. The specification

we estimate is the following,

yijt = α + β1 TRT * TREND * FIRSTSONi * FEMALEi

+ β2 TRT * TREND * FEMALEi + β3 TRT * TREND * FIRSTSONi

+ λ Xit + dt + θ + εit (4)

where TRT is the indicator for above-median district registration in 1985, and TREND is the

linear time trend for years 1958-77 during which the children in this pre-reform sample are born.

The covariates included in Xit are the same as in (3), except for the controls for other district

programmes and infrastructure which are not available for the pre-reform years. The covariates

23 We also check for consistency of estimates by including firstborn children in the sample and coding the firstborn son indicator as zero for these firstborns, and also by restricting the sample to the first two children only. The results do not change qualitatively, and are available from the authors upon request.

21

also include all the remaining three-way and two-way interactions between TRT, TREND, and

the female child and firstborn son indicators.

5.3 Village Data: Fixed Effect Regressions

The main dependent variable we examine in the household head survey data is the event

that a boy or girl who survived until 2003 was born into a household in a given year. Differences

in effects between boys and girls are likely to reflect gender differences in mortality, as chances

of sex-selective abortion during this period in rural West Bengal were low. Since reported births

are of children who survived till 2003, this outcome measures both birth and survival probability.

The specification we use is,

yijt = α + β1 BARGA_LANDjt-1 + β2 PATTA_LANDjt -1 + λ Xijt + dt + θi + εijt (5)

where the outcome yit takes value 1 if a surviving boy (or girl) is born in household i in village j in

year t and 0 otherwise. This outcome variable captures birth-cum-survival, as it takes value 1

when the surviving child born in year t is a boy (girl), and value 0 if there is no birth in year t, or

if there is a surviving birth in year t that is a girl (boy). Hence estimating (5) separately for

surviving male and female births will capture increases in gender-specific survival rates in

positive coefficients, and declines in fertility combined with reduced survival rates in negative

coefficients. BARGA_LANDjt-1 and PATTA_LANDjt-1 measure the share of cultivable land in

village j that was registered and distributed respectively in the three years preceding year t. The

coefficients β1 and β2 capture respective impacts of the two arms of the programme. While land

distribution benefited many more farmers than tenancy registration, distributed plots were

generally smaller and of lower quality than those covered by tenant registration, and were also

not eligible to be used as collateral for subsidised credit (Bardhan & Mookherjee, 2011). We

therefore do not report the estimates of β2.24 We also do not disaggregate land reform impacts

by the gender of the firstborn child as we do in specification (3), as we do not observe the birth

order of children present in the household in the village data. The terms dt and θi are year and

household fixed effects respectively, and εijt is an idiosyncratic error term.

The regressors Xijt include lagged land owned by the household, an above-ceiling indicator

(whether it owned more land than permitted by the land ceiling), and immigrant status indicators

24 These estimates are available from the authors upon request.

22

(whether the household immigrated after 1967, and year of immigration) which are available for

the full time period of 1971-2003.25 In alternate specifications we include controls constructed

from the farm-level dataset of Bardhan and Mookherjee (2011), which include logs of annual

village rainfall, village land productivity, price of rice, local government expenditures on roads

and irrigation, and kilometres of surfaced road and area irrigated by canals in the district. These

regressors control for potential confounding effects of local public spending on other

programmes and economic shocks. This information is only available for years 1982-95,

restricting the time coverage of the sample, but including these regressors does not qualitatively

change the results. Further, we also estimate (5) inclusive of district-year fixed effects, which

controls for all time-varying unobservables at the district level. The majority of public

infrastructural and health spending decisions affecting individual villages are made at by the

district level of government, making this a relevant robustness check.26

As with specification (3), we present separate estimates for Hindu and Non-Hindu households.

We further disaggregate results by whether households own land, and by household immigrant

status. This is because landless immigrant households are the most likely beneficiaries of tenancy

registration, while native landed households are the most likely to lose privileges on land. We

define landowning classes using pre-reform reported household landholdings in 1977, to avoid

endogenous sample selection on landholdings that may change due to the reform. We also

estimate (5) on years 1978 and after to match the cohorts used to estimate (3).

6. Results

6.1 Firstborn Sons, District Registration Rates, and Mortality

Table 3 reports the estimates from (3) without rice productivity controls and district linear

trends for infant mortality.27 In the pooled sample of children, we find in column (1) that once

districts achieve 50% registration, boys born the following year experience a 5.0 percentage point

decrease in infant mortality risk. The effect is highly significant at the 1% level. There is no

differential effect on mortality risk for boys by gender of the firstborn child. Girls with firstborn

25 In some specifications we include the share of cultivable village land transacted in the three years preceding year t to examine if income effects from these post-reform transactions explain any impacts we find. The results do not change significantly, and are available upon request. 26 Results from specifications including district-year fixed effects are in Appendix Tables A.7. The results including the full set of controls are available from the authors upon request. 27 The results including children from untreated border districts in neighbouring state Bihar are in Appendix Table A.2, and show identical patterns to those reported here. The estimates for rice productivity controls along with those in Table 3 are reported in Appendix Table A.4.

23

older sisters however experience no similar reduction in infant mortality risk, as the 50%

registration-female interaction term shows a 5.1 percentage point higher estimated infant

mortality risk than boys, which is significant at the 1% level. Girls with a firstborn older brother

in contrast have a similar or even marginally larger reduction in infant mortality risk than boys as

indicated by the 50% registration-female-firstborn son interaction term, with the estimate

suggesting a 6.0 percentage point decline in their infant mortality after high registration,

significant at the 10% level.28 In column (2) we introduce the log of district rice productivity and

its corresponding interactions with the female child and firstborn son indicators as controls. The

results are largely unchanged, except for increasing the infant mortality decline for girls with

firstborn older brothers to 7.5 percentage points. In column (3) we additionally control for

district-specific linear time trends. Including the trends increases the estimated mortality decline

for male children to 6.3 percentage points once district registration exceeds 50%, and the

coefficient remains highly significant at the 1% level. We continue to find that girls without

firstborn older brothers have unchanged, or even marginally greater infant mortality risk of up to

1.7 percentage points at high registration, while girls with firstborn older brothers experience an

estimated reduction in infant mortality risk of 9.0 percentage points.

We next examine community-specific results by re-estimating the specifications in columns (1)-

(3) for Hindu and Non-Hindu children separately. The results are reported in columns (4)-(6)

and (7)-(9) respectively. Columns (4)-(6) show an identical pattern of results for Hindu

households to those for the full sample in columns (1)-(3). In column (4), boys show an

estimated infant mortality decline of 4.0 percentage points once district registration exceeds

50%, marginally significant at the 10% level. Girls without older firstborn brothers show no

mortality decline in contrast, with the estimate even suggesting an increase in infant mortality of

up to 2.3 percentage points after high registration. Girls with firstborn older brothers however

exhibit a decline in infant mortality of 5.1 percentage points. The introduction of controls for

log of district rice productivity and its corresponding interaction terms with the firstborn son

and female child indicators in column (5) increases the estimated infant mortality decline for

boys to 5.2 percentage points in response to high registration. Girls without firstborn older

brothers show no decline in infant mortality risk, or even a marginal increase of up to 1.7

percentage points. Girls with firstborn older brothers experience a reduction in mortality risk of

8.0 percentage points, larger than in column (4) before productivity controls were introduced.

28 The estimated effects for girls with firstborn older brothers are arrived at by adding the statistically significant coefficients from the two-way and three-way interactions terms of the firstborn son, female child, and registration intensity indicators.

24

Including the district-linear time trend in column (6) strengthens the results further, with the

infant mortality reduction for boys rising to 5.9 percentage points, more precisely estimated at

5% level significance. Girls without firstborn older brothers continue to show no such decline,

with even a suggested increase in infant mortality of up to 1.1 percentage points. Girls with

firstborn brothers show a mortality decline of 8.6 percentage points, which remains significant at

the 5% level. Therefore, the estimates suggest that as sharecropper registration increases, Hindu

families invest in the health of their sons, but only do the same for their daughters if they already

have a firstborn son.

For Non-Hindu children, we find in column (7) that district sharecropper registration exceeding

50% reduces infant mortality among all children, regardless of their gender and the gender of

the firstborn child, by a highly significant 7.2 percentage points. This is evident from the fact that

nearly all of the two-way and three-way interaction terms of the high registration indicator with

the female child and firstborn indicators show no statistically significant differential impacts from

the baseline estimated effect of registration on male children. The only such differential impact

is found for male children with firstborn older brothers, who appear to benefit less than male

children with older firstborn sisters, but even this differential effect vanishes in columns (8) and

(9) once rice productivity controls and the district-linear time trend are included in the

specification. The inclusion of the rice productivity controls in column (8) also reduces the

estimated reduction in infant mortality from high registration to a statistically insignificant 4.6

percentage points, indicating that productivity is an important competing mechanism in infant

mortality reduction that correlates with sharecropper registration as we would expect. However

the further introduction of the district-linear time trend in column (9) again increases this

estimated beneficial impact of high registration to 7.8 percentage points, which is significant at

the 5% level. The district-linear time trend therefore performs an important role in capturing

otherwise omitted district-specific differential trends in infant mortality that bias our coefficient

of interest towards zero. Columns (7)-(9) therefore show that Non-Hindu families, unlike Hindu

families, do not discriminate by child gender when distributing the health investment gains from

district sharecropper registration amongst their children.

25

6.2 Gender-Differentiated Fertility Stopping

The estimated impacts of sharecropper registration on the probability a child of birth order 2

has a younger sibling are in Table 4.29 In columns (1)-(3) we find a large, stable, and statistically

significant reduction of 11.4-11.8 percentage points in the probability of a third birth once

district sharecropper registration exceeds 25%, but only when the first child in the family is male.

This indicates that land reforms did induce households to reduce fertility, but the fertility

stopping response was significantly son-biased. There is a suggestion of a smaller fertility decline

in response to land reform among families with firstborn daughters as well in column (2), upon

introducing rice productivity controls to the specification, but this effect does not survive the

further introduction of district-linear time trend controls in column (3). Nevertheless, the

coefficient does increase substantially upon the introduction of the rice productivity controls,

indicating the importance of separating the fertility impacts of increasing rice yields due to

registration and other programmes from those of registration. These results also show that

reduced sibling competition for resources is a potentially relevant causal channel for the

estimated infant mortality reduction we find for boys as well as girls with older firstborn brothers

in response to land reforms.

Columns (4)-(6) of Table 4 show the results for the Hindu sample. The same pattern of effects

is found as in the pooled sample in columns (1)-(3), with the probability of a third birth

declining by a statistically significant 10.4-10.8 percentage points once district registration

exceeds 25% , but only if the first child is male. There are no perceptible effects on fertility

stopping after the second birth if the first child is a daughter, again indicating a substantially son-

based fertility stopping response by Hindu households in response to land reforms. Columns (7)-

(9) report the corresponding results for Non-Hindu households. Notably, the fertility decline

among these households in response to land reforms is much larger than in Hindu households.

In column (7), there is a marginally significant decline in the probability of a third birth of 7.9

percentage points irrespective of the gender of the first child once district registration exceeds

50%. This effect increases to 8.7 percentage points in column (8) once rice productivity controls

are included, and is also more precisely estimated, achieving significance at the 5% level.

Introducing the district-linear time trend in column (9) increases the estimated effect to 9.6

percentage points, still marginally significant at the 10% level. Over and above this decline in

29 The results including children from untreated border districts in neighbouring state Bihar are in Appendix Table A.3, and show identical patterns to those reported here. The estimates for rice productivity controls along with those in Table 4 are in Appendix Table A.5.

26

fertility in response to land reforms, there is an additional large and statistically significant decline

in third births once district registration exceeds 25% in all columns (7)-(9) when the first child is

male. This additional son-biased fertility stopping response to land reforms is nearly twice as

large in Non-Hindu families as in Hindu families, at 17.9-18.2 percentage points. Hence Non-

Hindu families also appear to exhibit son preference, but do so by adjusting their fertility rather

than by increasing post-natal discrimination against girl children.

6.3 Sex Ratio at Birth

In Table 5 we report the estimated impact of sharecropper registration on the probability a

birth is male.30 For the pooled sample of all children, we find in column (1) that registration has

no effect on the sex ratio at birth for firstborns. In column (2), we find that the probability of a

birth at birth order 2 or higher being male increases by 3.9 percentage points after district

registration exceeds 50%. The estimate is significant at the 10% level. This estimate increases to

4.5 percentage points in column (3) if the firstborn child is female. As in Table 1, we re-estimate

the specifications in columns (1)-(3) on the samples of Hindu and Non-Hindu children

separately, and report the results in columns (4)-(6) and (7)-(9) respectively. We find in columns

(4) and (7) that sharecropper registration has no significant impact on the sex ratio at birth

among Hindu and Non-Hindu firstborns respectively, as in the pooled sample in column (1). We

find, as with infant mortality, that the increased probability of births being male at birth order 2

or higher is driven entirely by Hindu households. The estimates in column (5) indicate that this

probability increases by a highly significant 4.8 percentage points in Hindu families after districts

achieve 50% registration. This probability increases further to 5.5 percentage points when the

firstborn child is a female, and the coefficient is significant at the 5% level. Columns (8) and (9)

show no such effects on the sex ratio at birth in Non-Hindu families, consistent with the

previous literature. Hence, land reform appears to increase both sex selection and postnatal

discrimination in favour of male children in Hindu families, while also exacerbating son-biased

fertility stopping in both Hindu and Non-Hindu families.

30 The sex ratio results including estimates for log-rice productivity are in Appendix Table A.6. For brevity, we only report estimates from the full specification inclusive of rice productivity controls and district-linear time trends in Table 5 for children of birth order 1, and children of birth order 2 or higher separately.

27

6.4 Was Sharecropper Registration Targeted?

The results from specification (4) estimated for infant mortality are in Table 6. Across

columns (1)-(3), we find no statistically significant correlations in the pooled sample of children

between the pre-reform trend in infant mortality and the intensity of registration in the district

in 1985, by either the gender of the child or the firstborn sibling. The coefficients are all also

nearly identical to zero. There are similarly no significant correlations in the corresponding

regressions for Hindu and Non-Hindu households in columns (4)-(6) and (7)-(9) respectively,

lending credibility to our results. The results from the same test for the probability of male births

and having a younger sibling are in Table 7. In each of the pooled, Hindu, and Non-Hindu

samples of children, we see no correlations between the pre-reform trend in outcomes during

1958-77 and registration intensity in 1985, with the estimated coefficients close to zero. There is

hence little sign of sharecropper registration being targeted to districts based on pre-reform

trends in outcomes.

6.5 Village-Level Tenancy Registration and Surviving Births

Table 8 shows the estimated impacts from (5) of Operation Barga on the likelihood of

surviving children being born into households in different landowning classes from the village

survey data.31 Columns (1)-(6) show the programme effects on the likelihood of a surviving girl,

and columns (7)-(12) show the effects on the likelihood of a surviving boy. Column (1) shows a

significant negative effect of 0.8 percentage points for each percent increase of village land

registered on the likelihood of surviving girls being present in the household. This effect

captures a combination of reduced fertility in response to registration, and also potentially

increased gender discrimination that reduces female child survival. However, as the NFHS data

shows that household fertility declines significantly in response to tenancy registration, and

female infant mortality does not increase except possibly for the subsample of girls with

firstborn older sisters, the fertility decline is the likeliest dominant channel of impact. In column

(2) we include average farm-level rice productivity in the village in the year prior to the child’s

birth as a regressor to control for the effects of increased income and other programmes such as

minikit distribution. This data is only available for years 1982-95, which reduces the sample

substantially, but the impact of registration on the probability of a surviving girl remains negative

31 The results including controls for district-year fixed effects, the results from the full sample, and the results inclusive of all village and district controls for years 1982-1995 are all qualitatively identical, and are in Appendix Tables A.7, A.8, and A.9 respectively.

28

at 1.2 percentage points and is still highly significant. Columns (3) and (4) show no programme

effects in landless families on the probability of a surviving girl. In contrast, the reduced

probability of surviving girls as a result of registration occurs entirely in landed families, with the

coefficient showing a highly significant decline in this probability of 1.1 percentage points in

column (5), and an increased magnitude of 1.6 percentage points conditional on farm yields in

column (6). Notably, increased farm productivity also reduces the probability of a surviving girl

in landed households.

Turning to the likelihood of surviving male children being present in the household, column (7)

shows that tenancy registration significantly increases this probability by 1.1 percentage points.

As the fertility decline we find in the DHS data in response to registration would exert a negative

effect on this coefficient, this positive programme effect is unambiguously due to increased

survival of male children relative to female children consistent with our proposed hypotheses.

The effect continues to be positive and statistically significant at 0.6 percentage points after

conditioning on farm yields in column (8), despite the reduction in sample size. This increase in

male child survival arising from registration is much larger for landless than landed households,

at 2.6 percentage points in column (9) and 5.4 percentage points in column (10) conditional on

farm yields, compared to 0.7 percentage points in column (11), which is again consistent with

our predictions. Increased farm yields also increase the probability of surviving male births in

landless households as we would expect, as the property rights and wealth effects of registration

on fertility are both positive among the landless.

Tables 9 and 10 explore the heterogeneity of the effect on male and female surviving births

respectively by dividing the sample between Hindu and non-Hindu households (classified using

the name of the household head).32 The results show that the findings in Table 8 are driven

entirely by Hindu households, as we find in the NFHS data. In Table 9 we observe a positive

and significant effect of percent village land registered on male surviving births of 2.5 percentage

points and 0.7 percentage points in landless and landed Hindu households in columns (2) and (3)

respectively. The landless show a stronger response consistent with our predictions, and the

coefficient magnitudes are nearly identical to those for the pooled sample in Table 8, indicating

that the positive discrimination in favour of male children is concentrated in Hindu households.

There are no significant programme effects found for non-Hindu households. The

corresponding results for girls in Table 10 show a negative and significant effect of land

32 For brevity we do not report the results from specifications including average farm rice productivity in further results from the village data. These specifications show qualitatively identical results, and are available upon request.

29

registration on the likelihood of surviving female births in Hindu households of 1.1 percentage

points, coming from landed households in column (3). Hence landed Hindu households increase

health investments in their sons in response to tenancy reform to improve their survival chances

as we see in Table 9, but they do not do the same for their daughters and the fertility decline

effect dominates in Table 10. This is consistent with the evidence from the NFHS data, which

also shows a widening difference in the survival advantage of sons due to tenant registration. It is

worth noting that in non-Hindu households, the coefficient estimates indicate large (but

imprecisely) estimated increases in survival probability for both sons and daughters in Tables 9

and 10, which is again consistent with what we find in the NFHS data.

Tables 11 and 12 examine Hindu male and female child survival probabilities respectively by

household immigrant status. Table 11 shows that the positive effect of registration on Hindu

male surviving births is concentrated in native landed households and immigrant landless

households in columns (3) and (5). Additionally, the increase in survival probability for male

children is greater among the landless than the landed households, consistent with our

predictions. For surviving female births in Table 12, there is a significant effect of the fertility

decline in response to registration among native landed households in column (3), but this is not

offset by a large increase in survival probability, as it is for boys in Table 11. Hence we again

capture greater relative survival gains to sons due to tenancy registration as per our hypotheses.

7. Conclusions

Our findings show a strong exacerbating effect of increased property rights security on

gender discrimination in Hindu families, with parents showing greater son-bias in sex selective

behaviour, fertility-stopping, and postnatal health investments. In Non-Hindu families on the

other hand, there is a decline in infant mortality for both boys and girls in response to tenancy

registration, and son preference only manifests in fertility stopping behaviour. Institutional

differences in son preference therefore play a significant role in determining the gender

distribution of benefits from improving property rights. In contrast, there is evidence that land

reform alters existing gender-unequal institutions in favour of women in alternative settings.

Tenure regularisation has been argued to significantly improve women’s tenurial and inheritance

claims to land in Rwanda (Ali et al., 2014), and joint spousal titling seemingly increased women’s

intra-household bargaining power in Peru (Wiig, 2013). The rigidity of gender institutions

30

governing land ownership is hence also a likely critical factor in the effectiveness of property

rights formalisation in reducing gender discrimination.

The distribution of land rights within the household underpins family welfare in developing

countries such as India where the majority of the population is engaged in agricultural

occupations. As such, while our results illustrate the importance of ensuring women are directly

benefited by property rights formalisation, the potential for resulting intra-household conflict

also requires acknowledgement. Indeed, though land reforms in India have seemingly reduced

overall suicide rates, land reforms that benefited women relatively more than men have been

argued to increase both female and male suicides due to increased familial tensions (Anderson &

Genicot, 2015). Welfare considerations have also become more urgent as rapid industrialisation

has led to increased government acquisition of agricultural land in India and other fast-growing

developing countries, which can reverse welfare gains from tenancy and land ownership reforms

without adequate compensation to farmers who lose their land as a result (Ghatak &

Mookherjee, 2014). Property rights formalisation therefore requires consideration of

distributional considerations both within households and between entire sections of the

population to prevent unintended welfare losses in the economic development process.

References

Agarwal, B., 2003. Gender and Land Rights Revisited: Exploring New Prospects via the State,

Family, and Market. Journal of Agrarian Change, 3(1,2), pp. 184-224.

Ali, D. A., Deininger, K. & Goldstein, M., 2014. Environmental and Gender Impacts of Land

Tenure Regularization in Africa: Pilot evidence from Rwanda. Journal of Development Economics,

Volume 110, pp. 262-275.

Almond, D., Li, H. & Zhang, S., 2013. Land Reform and Sex Selection in China. NBER Working

Paper No. 19153.

Anderson, S. & Genicot, G., 2015. Suicide and Property Rights in India. Journal of Development

Economics.

Anderson, S. & Ray, D., 2010. Missing Women: Age and Disease. Review of Economic Studies, 77(4),

pp. 1262-1300.

Appu, P. S., 1996. Land reforms in India: A Survey of Policy, Legislation, and Iimplementation. New

Delhi: Vikas Publishing House.

31

Babu, S. C., Thirumaran, S. & Mohanam, T., 1993. Agricultural Productivity, Seasonality, and

Gender Bias in Rural Nutrition: Empirical Evidence from South India. Social Science and Medicine,

pp. 1313-1319.

Banerjee, A., Gertler, M. & Ghatak, M., 2002. Empowerment and Efficiency: Tenancy Reform in

West Bengal. Journal of Political Economy.

Barcellos, S. H., Carvalho, L. & Lleras-Muney, A., 2013. Child Gender and Parental Investments

in India: Are Boys and Girls Treated Differently?. American Economic Journal: Applied Economics.

Bardhan, P., 1974. On life and death questions. Economic and Political Weekly.

Bardhan, P., Luca, M., Mookherjee, D. & Pino, F., 2014. Evolution of Land Distribution in West

Bengal 1967-2004: Role of Land Reform and Demographic Changes. Journal of Development

Economics.

Bardhan, P. & Mookherjee, D., 2010. Determinants of Redistributive Politics: An Empirical

Analysis of Land Reforms in West Bengal, India. The American Economic Review, pp. 1572-1600.

Bardhan, P. & Mookherjee, D., 2011. Subsidized Farm Input Programs and Agricultural

Performance: A Farm-Level Analysis of West Bengal's Green Revolution, 1982-1995. American

Economic Journal: Applied Economics, pp. 186-214.

Bardhan, P., Mookherjee, D. & Kumar, N., 2012. State-Led or Market-Led Green Revolution?

Role of Private Irrigation Investment vis-a-vis Local Government Programs in West Bengal's

Farm Productivity Growth. Journal of Development Economics, pp. 222-235.

Behrman, J. R., 1988. Intrahousehold Allocation of Nutrients in Rural India: Are Boys

Favoured? Do Households Exhibit Inequality Aversion?. Oxford Economic Papers, pp. 32-54.

Besley, T., 1995. Property Rights and Investment Incentives: Theory and Evidence from Ghana.

Journal of Political Economy.

Besley, T. & Burgess, R., 2000. Land Reform, Poverty Reduction, and Growth: Evidence from

India. The Quarterly Journal of Economics, pp. 389-430.

Besley, T. & Ghatak, M., 2010. Property Rights and Economic Development. In: D. Rodrik &

M. Rosenzweig, eds. Handbook of Development Economics. s.l.:s.n., pp. 4525-4595.

Besley, T., Leight, J., Pande, R. & Rao, V., 2016. Long-Run Impacts of Land Regulation:

Evidence from Tenancy Regulation in India. Journal of Development Economics, Volume 118, pp. 72-

87.

Bhalotra, S., 2010. Fatal Fluctuations? Cyclicality in Infant Mortality in India. Journal of

Development Economics, pp. 7-19.

Bhalotra, S. & Cochrane, T., 2010. Where Have All the Young Girls Gone? Identification of Sex

Selection in India. IZA Discussion Paper No. 5381.

32

Bhalotra, S., Valente, C. & van Soest, A., 2010. The Puzzle of Muslim Advantage in Child

Survival in India. Journal of Health Economics, 29(2), pp. 191-204.

Bhalotra, S. & Zamora, B., 2009. Social Divisions in Education in India. In: A. Sharif & R.

Basant, eds. Handbook of Muslims in India. Delhi: Oxford University Press.

Bittles, A., 2002. Endogamy, Cosanguinuity, and Community Genetics. Journal of Genetics, 81(3).

Borooah, V. K., 2004. Gender Bias Among Children in India in Their Diet and Immunisation

Against Disease. Social Science and Medicine, pp. 1719-1731.

Busso, M., Gregory, J. & Kline, P., 2013. Assessing the Incidence and Efficiency of a Prominent

Place Based Policy. American Economic Review.

Cameron, A., Gelbach, J. & Miller, D., 2008. Bootstrap-Based Improvements for Inference with

Clustered Errors. Review of Economics and Statistics.

Carranza, E., 2014. Soil Endowments, Female Labor Force Participation, and the Demographic

Deficit of Women in India. American Economic Journal: Applied Economics.

Chakraborty, I. & Chakraborty, A., 2010. Female Work Participation and Gender Differential in

Earning in West Bengal, India. Journal of Quantitative Economics, 8(2).

Chakraborty, T. & Kim, S., 2010. Kinship Institutions and Sex Ratios in India. Demography, 47(4),

pp. 989-1012.

Chakravarty, A., 2010. Supply Shocks and Gender Bias in Child Health Investments: Evidence

from the ICDS programme in India. The B.E. Journal of Economic Analysis and Policy.

Chattopadhyay, A. K., 2005. Distributive Impact of Agricultural Growth in West Bengal.

Economic and Political Weekly, 40(53).

Dasgupta, B., 1984. Sharecropping in West Bengal: From Independence to Operation Barga.

Economic and Political Weekly, 19(26).

Deininger, K., Goyal, A. & Nagarajan, H., 2013. Women's Inheritance Rights and

Intergenerational transmission of resources in India. Journal of Human Resources, 48(1), pp. 114-

141.

Deininger, K., Jin, S. & Yadav, V., 2011. Long Term Effects of Land Reform on Human Capital

Accumulation: Evidence from West Bengal. UNU-WIDER Working Paper No. 2011/82.

Duflo, E., 2012. Women Empowerment and Economic Development. Journal of Economic

Literature, 50(4), pp. 1051-1079.

Ghatak, M. & Mookherjee, D., 2014. Land Acquisition for Industrialization and Compensation

of Displaced Farmers. Journal of Development Economics.

Ghatak, M. & Roy, S., 2007. Land Reform and Agricultural Productivity in India: A Review of

the Evidence. Oxford Review of Economic Policy.

33

Ghosh, A., 2008. The Effect of Land Reforms on Long-Term Health and Well-Being in India.

RAND Dissertations (mimeo).

Goldstein, M. & Udry, C., 2008. The Profits of Power: Land Rights and Agricultural Investment

in Ghana. Journal of Political Economy.

Gupta, J., 1993. Land, Dowry, Labour: Women in the Changing Economy of Midnapur. Social

Scientist, pp. 74-90.

Gupta, J., 2002. Women Second in the Land Agenda. Economic and Political Weekly.

Hornbeck, R., 2010. Barbed Wire: Property Rights and Agricultural Development. The Quarterly

Journal of Economics, 125(2).

Jain, T., 2014. Where There is a Will: Fertility Behaviour and Sex Bias in Large Families. The

Journal of Human Resources, 49(2).

Jayachandran, S., 2014. Fertility Decline and Missing Women. NBER Working Paper no. 20272.

Jayachandran, S., 2014. The Roots of Gender Inequality. NBER working paper no. 20380.

Jayachandran, S. & Kuziemko, I., 2011. Why Do Mothers Breastfeed Girls Less Than Boys?

Evidence and Implications for Child Health in India. The Quarterly Journal of Economics, pp. 1485-

1538.

Jensen, R., 2010. Economic Opportunities and Gender Differences in Human Capital:

Experimental Evidence for India. NBER Working Paper No. w16021.

Klasen, S. & Wink, C., 2002. A Turning Point in Gender Bias in Mortality? An Update on the

Number of Missing Women. Population and Development Review, 28(2).

Kodoth, P., 2005. Fostering Insecure Livelihoods: Dowry and Female Seclusion in Left

Developmental Contexts in West Bengal and Kerala. Economic and Political Weekly, 40(25), pp.

2543-2554.

Lieten, G. K., 1992. Continuity and Change in Rural West Bengal. New Delhi: Sage Publications.

Lingat, R., 1973. The Classical Law of India. New York: Oxford University Press.

Lipton, M., 2009. Land Reform in Developing Countries: Property Rights and Property Wrongs.

s.l.:Routledge.

Murthi, M., Guio, A.-C. & Dreze, J., 1995. Mortality, Fertility, and Gender Bias in India: A

District-Level Analysis. Population and Development Review, pp. 745-782.

Nasir, R. & Kalla, A., 2006. Kinship System, Fertility, and Son Preference Among the Muslims:

A Review. Anthropologist, 8(4), pp. 275-281.

Oster, E., 2009. Does Increased Access Increase Inequality? Gender and Child Health

Investments in India. Journal of Development Economics, pp. 62-76.

34

Rahman, L. & Rao, V., 2004. The Determinants of Gender Equity in India: Examining Dyson

and Moore's Thesis with New Data. Population and Development Review, 30(2), pp. 239-268.

Rose, E., 2000. Gender Bias, Credit Constraints, and Time Allocation in Rural India. Economic

Journal, pp. 738-758.

Roy, S., 2015. Empowering Women? Inheritance rights, Female Education and Dowry Payments

in India. Journal of Development Economics.

Sen, A. K., 2003. Missing Women -- Revisited: Reduction in Female Mortality has Been

Counterbalanced by Sex-Selective Abortions. BMJ: British Medical Journal.

Wiig, H., 2013. Joint Titling in Rural Peru: Impact on Women’s Participation in Household

Decision-Making. World Development.

35

Figure 1: Average Sharecropper Registration Rate by Year

Figure 2: Average Cultivable Village Land Registered by Year

Notes: The figure shows the average rate of completed sharecropper registration across

the 14 West Bengal districts in the Banerjee at al. (2002) data during 1975-1991.

Notes: The figure shows the average percent of village cultivable land registered across

the 89 villages from the household panel data survey during the years 1967-1998. The

percent of cultivable land registered declines after 1985 as registration slowed during this

period, while the amount of cultivable land increased on average.

Year

Per

cen

t C

ult

ivab

le L

and

Reg

iste

red

36

Table 1: NFHS and District Summary Statistics

Freq. Mean S.D. Min. Max.

(1) (2) (3) (4) (5)

Panel A Mother Characteristics: 1967-1993

Years of Education 6,443 3.416 4.297 0 18

Age at Birth 6,468 19.034 3.532 5 40

Total Births 6,468 3.386 2.010 1 11

Hindu 6,468 0.750 - 0 1

Scheduled Caste/Tribe 6,468 0.205 - 0 1

Rural 6,468 0.680 - 0 1

Panel B Child Outcomes: 1967-1993

Infant Death 20,148 0.094 - 0 1

Neonatal Death 20,148 0.064 - 0 1

Male Child 20,148 0.511 - 0 1

Has Younger Sibling 20,148 0.718 - 0 1

Panel C District Productivity and Programmes: 1977-1990

Rice Productivity 196 1.473 0.434 0.720 2.595

Patta Area Per Capita 196 6.518 4.937 0.321 17.986

Surfaced Roads Per Capita 196 0.208 0.067 0.115 0.392

Medical Institutions Per Capita 196 0.056 0.016 0.033 0.115

Notes: Panel A shows mother characteristics, and Panel B shows child outcomes for cohorts

born during 1967-1993. Panel C shows productivity and programme statistics in the 14

districts with sharecropper registration data for years 1977-1990, which are the years for

which they enter as controls in the regressions. Neonatal death takes value 1 if the child dies

aged 0-1 months, and infant death takes value 1 if the child dies aged 0-12 months.

Table 2 – District Sharecropper Registration Rates by Year

1980 1982 1984 1986 1988 1990 1992

Kochibihar 19.96 39.42 42.34 50.54 54.65 57.46 58.31

Jalpaiguri 18.44 30.88 32.30 34.95 40.25 40.25 40.25

Darjeeling 14.49 23.40 24.52 28.22 28.84 28.84 28.84

West Dinajpur 34.88 66.41 70.77 73.86 75.63 76.74 76.77

Maldah 44.15 60.93 69.75 74.40 76.57 78.32 79.24

Murshidabad 15.82 43.01 44.69 53.09 56.80 58.89 60.79

Nadia 22.99 37.20 42.79 49.52 52.74 53.35 55.71

24-Parganas 15.22 38.54 42.84 47.67 49.51 49.84 54.05

Howrah 31.05 44.56 48.67 54.63 56.12 57.32 58.22

Hooghly 25.50 40.94 46.26 55.11 58.60 61.62 63.45

Midnapur 16.79 44.89 55.70 61.09 62.84 63.64 63.82

Bankura 29.69 68.67 74.48 84.46 89.66 92.56 93.83

Burdwan 11.00 35.64 39.45 45.74 49.60 51.42 53.31

Birbhum 25.01 59.26 72.49 91.37 95.67 98.30 100.00

Mean 23.21 45.27 50.50 57.48 60.53 62.04 63.33

Notes: The table shows district sharecropper registration rates by year as reported in Banerjee et. al. (2002).

37

Table 3 – Infant Mortality, Firstborn Sons, and Registration

Infant Death

All Children Hindu Children Non-Hindu Children

(1) (2) (3) (4) (5) (6) (7) (8) (9)

SCROP50 t-1 * FIRSTSON * FEMALE -0.060* -0.075* -0.073* -0.074** -0.097** -0.097** -0.033 -0.029 -0.028

(0.029) (0.039) (0.039) (0.032) (0.046) (0.048) (0.048) (0.054) (0.055)

SCROP25 t-1 * FIRSTSON * FEMALE 0.033 0.030 0.031 0.021 0.017 0.019 0.062 0.063 0.046 (0.033) (0.038) (0.038) (0.05) (0.051) (0.051) (0.041) (0.045) (0.044)

SCROP50 t-1 * FEMALE 0.051*** 0.046** 0.046** 0.063*** 0.069** 0.070** 0.028 -0.001 -0.001 (0.02) (0.022) (0.023) (0.023) (0.03) (0.029) (0.039) (0.046) (0.046)

SCROP25 t-1 * FEMALE -0.025 -0.029 -0.029 -0.024 -0.025 -0.027 -0.039 -0.046 -0.036 (0.03) (0.032) (0.032) (0.033) (0.034) (0.033) (0.045) (0.046) (0.048)

SCROP50 t-1 * FIRSTSON 0.028 0.040 0.039 0.015 0.043 0.043 0.046* 0.029 0.029 (0.023) (0.029) (0.028) (0.028) (0.031) (0.032) (0.026) (0.031) (0.031)

SCROP25 t-1 * FIRSTSON -0.037 -0.035 -0.034 -0.041 -0.034 -0.035 -0.027 -0.031 -0.016 (0.023) (0.024) (0.023) (0.036) (0.035) (0.036) (0.029) (0.034) (0.03)

SCROP50 t-1 -0.050*** -0.050*** -0.063*** -0.040* -0.052* -0.059** -0.072** -0.046 -0.078**

(0.02) (0.021) (0.023) (0.021) (0.025) (0.026) (0.035) (0.029) (0.035)

SCROP25 t-1 0.019 0.012 0.001 0.006 0.000 -0.003 0.051 0.035 -0.005

(0.023) (0.021) (0.020) (0.03) (0.029) (0.031) (0.030) (0.036) (0.039)

FIRSTSON * FEMALE -0.019 -0.025 -0.022 -0.062 -0.104 -0.022 0.045 0.087 0.079 (0.13) (0.143) (0.140) (0.163) (0.178) (0.140) (0.222) (0.236) (0.232)

FEMALE 0.192 0.163 0.164 0.181 0.170 0.164 0.209 0.178 0.179 (0.152) (0.132) (0.126) (0.147) (0.134) (0.126) (0.247) (0.259) (0.256)

FIRSTSON 0.024 0.038 0.032 0.011 0.070 0.032 0.016 -0.026 -0.024

(0.114) (0.132) (0.134) (0.17) (0.198) (0.134) (0.127) (0.148) (0.142)

District FE x x x x x x x x x District Rice Productivity x x x x x x District-Birth Year Trend x x x Observations 8,367 8,367 8,367 5,448 5,448 5,448 2,919 2,919 2,919 Pre-Reform Outcome Mean 0.098 0.098 0.098 0.107 0.107 0.107 0.074 0.074 0.074 Cohorts 1978-91 1978-91 1978-91 1978-91 1978-91 1978-91 1978-91 1978-91 1978-91 Districts 14 14 14 14 14 14 14 14 14 Notes: Wild cluster bootstrapped standard errors in parentheses. All specifications also include the female child, birth year fixed effects, birth order fixed effects, year of interview fixed effects, indicators for household religion and caste, whether the household is rural, mother’s educational attainment, and linear and quadratic terms of the mother’s age at which the child is born. Specifications for children of birth order 2 or higher also include the firstborn son indicator and its corresponding interaction terms. The district covariates include logs of patta land area distributed, number of medical institutions, and kilometres of surfaced road per capita and their corresponding interactions with the female child and the firstborn son indicators. *** p<0.01, ** p<0.05, * p<0.1

38

Table 4 – Younger Siblings and Sharecropper Registration

Child Has a Younger Sibling All Children Hindu Children Non-Hindu Children (1) (2) (3) (4) (5) (6) (7) (8) (9)

SCROP50 t-1 * FIRSTSON -0.007 0.009 0.012 -0.011 0.003 0.008 0.011 0.020 0.019 (0.045) (0.044) (0.044) (0.050) (0.052) (0.051) (0.053) (0.049) (0.050)

SCROP25 t-1 * FIRSTSON -0.118*** -0.114** -0.116** -0.106** -0.104** -0.108** -0.179*** -0.179*** -0.182** (0.053) (0.053) (0.055) (0.051) (0.049) (0.052) (0.078) (0.082) (0.086)

SCROP50 t-1 -0.037 -0.048* -0.048 -0.024 -0.035 -0.038 -0.079* -0.087** -0.096*

(0.029) (0.028) (0.032) (0.033) (0.034) (0.039) (0.04) (0.037) (0.051)

SCROP25 t-1 0.022 -0.001 -0.020 0.016 -0.008 -0.058 0.010 0.000 0.036

(0.041) (0.042) (0.056) (0.058) (0.062) (0.079) (0.063) (0.06) (0.083)

FIRSTSON -0.315 -0.223 -0.201 -0.436 -0.343 -0.347 0.108 0.150 0.205 (0.209) (0.192) (0.193) (0.258) (0.251) (0.251) (0.133) (0.119) (0.138)

District FE x x x x x x x x x District Rice Productivity x x x x x x District-Birth Year Trend x x x Observations 2,686 2,686 2,686 1,919 1,919 1,919 767 767 767 Pre-Reform Outcome Mean 0.839 0.839 0.839 0.808 0.808 0.808 0.952 0.952 0.952 Cohorts 1978-91 1978-91 1978-91 1978-91 1978-91 1978-91 1978-91 1978-91 1978-91 Districts 14 14 14 14 14 14 14 14 14 Notes: Wild cluster bootstrapped standard errors in parentheses. All specifications also birth year fixed effects, birth order fixed effects, year of interview fixed effects, indicators for household religion and caste, whether the household is rural, mother’s educational attainment, and linear and quadratic terms of the mother’s age at which the child is born. The district covariates include logs of patta land area distributed, number of medical institutions, and kilometres of surfaced road per capita and their corresponding interactions with the female child and the firstborn son indicators. *** p<0.01, ** p<0.05, * p<0.1

39

Table 5 – Male Births and Sharecropper Registration

Child is Male

All Children Hindu Children Non-Hindu Children

B. Ord. 1 B. Ord. >1 B. Ord. >1 B. Ord. 1 B. Ord. >1 B. Ord. >1 B. Ord. 1 B. Ord. >1 B. Ord. >1

(1) (2) (3) (4) (5) (6) (7) (8) (9)

SCROP50 t-1 * FIRSTSON - - -0.011 - - -0.014 - - 0.004 (0.020) (0.019) (0.047)

SCROP25 t-1 * FIRSTSON - - -0.005 - - 0.005 - - -0.033 (0.027) (0.031) (0.062)

SCROP50 t-1 -0.020 0.039* 0.045* -0.065 0.048** 0.055** 0.091 0.042 0.039 (0.037) (0.022) (0.025) (0.050) (0.024) (0.029) (0.064) (0.035) (0.041)

SCROP25 t-1 -0.010 0.048 0.051 0.063 0.031 0.030 -0.149 0.096 0.111

(0.034) (0.047) (0.058) (0.053) (0.050) (0.054) (0.098) (0.077) (0.095)

FIRSTSON - -0.007

0.094

- -0.008

0.186 - -0.007

-0.005

(0.009)

(0.096)

(0.010)

(0.113) (0.017)

(0.262)

District FE x x x x x x x x x District Covariates x x x x x x x x x District-Birth Year Trend x x x x x x x x x Observations 3,248 8,367 8,367 2,323 5,448 5,448 925 2,919 2,919 Pre-Reform Outcome Mean 0.449 0.494 0.494 0.433 0.493 0.493 0.488 0.498 0.498 Cohorts 1978-91 1978-91 1978-91 1978-91 1978-91 1978-91 1978-91 1978-91 1978-91 Districts 14 14 14 14 14 14 14 14 14 Notes: Wild cluster bootstrapped standard errors in parentheses. All specifications also include birth year fixed effects, birth order fixed effects, year of interview fixed effects, indicators for household religion and caste, whether the household is rural, mother’s educational attainment, and linear and quadratic terms of the mother’s age at which the child is born. The district covariates include logs of patta land area distributed, number of medical institutions, and kilometres of surfaced road per capita and their corresponding interactions with the female child and the firstborn son indicators. *** p<0.01, ** p<0.05, * p<0.1

40

Table 6 – Test of Targeted Registration: Infant Mortality

Infant Death

All Children Hindu Children Non-Hindu Children

(1) (2) (3) (4) (5) (6) (7) (8) (9)

TRT * TREND -0.002 -0.005 -0.001 -0.000 -0.004 -0.002 -0.009 -0.011 -0.004

(0.005) (0.007) (0.008) (0.004) (0.008) (0.009) (0.009) (0.012) (0.012)

TRT * TREND * FEMALE - 0.007 0.006 - 0.008 0.008 - 0.006 0.001 (0.008) (0.008) (0.009) (0.011) (0.009) (0.009)

TRT * TREND * FIRSTSON * FEMALE - - 0.003 - - 0.001 - - 0.007 (0.006) (0.006) (0.007)

District FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Observations 3,389 3,389 3,389 2,428 2,428 2,428 961 961 961 Cohorts 1958-77 1958-77 1958-77 1958-77 1958-77 1958-77 1958-77 1958-77 1958-77 Districts 14 14 14 14 14 14 14 14 14 Notes: Wild cluster bootstrapped standard errors in parentheses. Samples include children of birth order 2 or higher. All specifications also include the female child and firstborn son indicators and their two-way interactions with the trend, birth year fixed effects, indicators for household religion and caste, whether the household is rural, mother’s educational attainment, and linear and quadratic terms of the mother’s age at which the child is born. *** p<0.01, ** p<0.05, * p<0.1

Table 7 – Test of Targeted Registration: Male Births and Fertility

All Children Hindu Children Non-Hindu Children

Male Child Younger Sibling Male Child Younger Sibling Male Child Younger Sibling

(1) (2) (3) (4) (5) (6)

TRT * TREND 0.001 0.001 0.001 0.001 0.002 0.002

(0.004) (0.004) (0.005) (0.005) (0.008) (0.004)

TRT * TREND * FIRSTSON - -0.009 - -0.012 - 0.003 (0.007) (0.008) (0.008)

District FE Yes Yes Yes Yes Yes Yes Observations 3,389 1,369 2,428 1,015 961 961 Cohorts 1958-77 1958-77 1958-77 1958-77 1958-77 1958-77 Districts 14 14 14 14 14 14 Notes: Wild cluster bootstrapped standard errors in parentheses. Samples for the sex ratio regressions include children of birth order 2 or higher, and of birth order 2 for the fertility regressions. The specifications for the probability of having a younger sibling also include the firstborn son indicator and its interaction with the trend, birth year fixed effects, indicators for household religion and caste, whether the household is rural, mother’s educational attainment, and linear and quadratic terms of the mother’s age at which the child is born. *** p<0.01, ** p<0.05, * p<0.1

41

Table 8: Land Reform and Surviving Births

Dep. Variable: 1=surviving girl,

0=no birth or surviving boy 1=surviving boy,

0=no birth or surviving girl

Land category: All Landless Some land All Landless Some land

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

Agricultural land -0.001 -0.001 - - -0.001* -0.001 -0.000 0.003 - - -0.000 0.003

(0.001) (0.005) (0.001) (0.005) (0.001) (0.004) (0.001) (0.004) % Land registered -0.008*** -0.012*** 0.000 0.001 -0.011*** -0.016*** 0.011*** 0.006*** 0.026*** 0.054*** 0.007*** -0.005*

(0.001) (0.002) (0.003) (0.003) (0.001) (0.002) (0.002) (0.002) (0.009) (0.002) (0.001) (0.003) Farm Productivity - -0.023** - -0.002 - -0.044*** - 0.024* - 0.029* - 0.023

(0.011) (0.016) (0.016) (0.012) (0.016) (0.015) Household FE x x x x x x x x x x x x Observations 44,590 7,869 19,548 3,627 25,042 4,242 44,590 7,869 19,548 3,627 25,042 4,242 Households 2,286 1,928 1,144 892 1,142 1,036 2,286 1,928 1,144 892 1,142 1,036 Years 1978-99 1982-95 1978-99 1982-95 1978-99 1982-95 1978-99 1982-95 1978-99 1982-95 1978-99 1982-95 Notes: Robust standard errors in parentheses, adjusted for clustering on villages. The variable % land registered is computed as the sum over the previous three years of the share of land affected by each program over the total cultivable land in each village, using official land records. All regressions also include immigrant status indicators, and year and household fixed effects, and % land distributed calculated the same way as % land registered. The land category of the household is defined based on year 1977 landholdings, and regressions include data for years 1978–1999. *** p<0.01, ** p<0.05, * p<0.1

42

Table 9: Surviving Male Births by Religion

Dep. Variable: 1=surviving boy,

0=no birth or surviving girl

Religion: Hindu Non-Hindu

Land category: All Landless Some land All Landless Some land

(1) (2) (3) (4) (5) (6)

Agricultural land 0.001 - 0.001 -0.006 - -0.005 (0.001) (0.001) (0.004) (0.004)

% Land registered 0.010*** 0.025*** 0.007*** 0.223 0.317 0.152 (0.002) (0.007) (0.001) (0.145) (0.385) (0.212)

Household FE x x x x x x Observations 35,018 15,549 19,469 9,572 3,999 5,573 Households 1,822 934 888 464 210 254 Years 1978-99 1978-99 1978-99 1978-99 1978-99 1978-99

Notes: Robust standard errors in parentheses, adjusted for clustering on villages. The variable % land registered is computed as the sum over the previous three years of the share of land affected by each program over the total cultivable land in each village, using official land records. All regressions also include a land ceiling indicator, immigrant status indicators, and year and household fixed effects, and % land distributed calculated the same way as % land registered. The land category of the household is defined based on year 1977 landholdings, and regressions include data for years 1978–1999. Religion of the household is the religion of the household head. *** p<0.01, ** p<0.05, * p<0.1

Table 10: Surviving Female Births by Religion

Dep. Variable: 1=surviving girl,

0=no birth or surviving boy

Religion: Hindu Non-Hindu

Land category: All Landless Some land All Landless Some land

(1) (2) (3) (4) (5) (6)

Agricultural land -0.000 - -0.000 -0.005*** - -0.005*** (0.001) (0.001) (0.002) (0.002)

% Land registered -0.009*** -0.001 -0.011*** 0.179 0.116 0.222 (0.001) (0.003) (0.001) (0.187) (0.266) (0.183)

Household FE x x x x x x Observations 35,018 15,549 19,469 9,572 3,999 5,573 Households 1,822 934 888 464 210 254 Years 1978-99 1978-99 1978-99 1978-99 1978-99 1978-99

Notes: Robust standard errors in parentheses, adjusted for clustering on villages. The variable % land registered is computed as the sum over the previous three years of the share of land affected by each program over the total cultivable land in each village, using official land records. All regressions also include a land ceiling indicator, immigrant status indicators, and year and household fixed effects, and % land distributed calculated the same way as % land registered. The land category of the household is defined based on year 1977 landholdings, and regressions include data for years 1978–1999. Religion of the household is the religion of the household head. *** p<0.01, ** p<0.05, * p<0.1

43

Table 11: Surviving Hindu Male Births by Immigrant Status

Dep. Variable: 1=surviving boy,

0=no birth or surviving girl

Native/Immigrant: Native Immigrant

Land category: All Landless Some land All Landless Some land

(1) (2) (3) (4) (5) (6)

Agricultural land 0.001 - 0.001 -0.012 - -0.005 (0.001) (0.001) (0.010) (0.009)

% Land registered 0.008*** 0.011 0.006*** 0.039*** 0.038*** 0.564 (0.002) (0.013) (0.001) (0.005) (0.003) (0.481)

Household FE x x x x x x Observations 29,099 10,471 18,628 5,919 5,078 841 Households 1,340 491 849 482 443 39 Years 1978-99 1978-99 1978-99 1978-99 1978-99 1978-99

Notes: Robust standard errors in parentheses, adjusted for clustering on villages. The variable % land registered is computed as the sum over the previous three years of the share of land affected by each program over the total cultivable land in each village, using official land records. All regressions also include a land ceiling indicator, year of immigration controls, and year and household fixed effects, and % land distributed calculated the same way as % land registered. The land category of the household is defined based on year 1977 landholdings, and regressions include data for years 1978–1999. Religion of the household is the religion of the household head. *** p<0.01, ** p<0.05, * p<0.1

Table 12: Surviving Hindu Female Births by Immigrant Status

Dep. Variable: 1=surviving girl,

0=no birth or surviving boy

Native/Immigrant: Native Immigrant

Land category: All Landless Some land All Landless Some land

(1) (2) (3) (4) (5) (6)

Agricultural land 0.000 - -0.000 -0.014 - -0.004 (0.001) (0.001) (0.009) (0.008)

% Land registered -0.010*** -0.003 -0.011*** 0.001 -0.000 -0.103 (0.001) (0.006) (0.001) (0.003) (0.003) (0.465)

Household FE x x x x x x Observations 29,099 10,471 18,628 5,919 5,078 841 Households 1,340 491 849 482 443 39 Years 1978-99 1978-99 1978-99 1978-99 1978-99 1978-99

Notes: Robust standard errors in parentheses, adjusted for clustering on villages. The variable % land registered is computed as the sum over the previous three years of the share of land affected by each program over the total cultivable land in each village, using official land records. All regressions also include a land ceiling indicator, year of immigration controls, and year and household fixed effects, and % land distributed calculated the same way as % land registered. The land category of the household is defined based on year 1977 landholdings, and regressions include data for years 1978–1999. Religion of the household is the religion of the household head. *** p<0.01, ** p<0.05, * p<0.1

44

Appendix A: Additional Tables

Table A.1: Infant Death, Younger Siblings, and Sharecropper Registration: Birth Order 1

All Children Hindu Children Non-Hindu Children

Infant Death Younger Sibling Infant Death Younger Sibling Infant Death Younger Sibling

(1) (2) (3) (4) (5) (6)

SCROP50 t-1 * FEMALE 0.014 0.007 0.011 0.026 0.025 -0.025 (0.025) (0.021) (0.026) (0.032) (0.042) (0.044)

SCROP25 t-1 * FEMALE -0.041 -0.005 -0.029 -0.003 -0.052 -0.022 (0.031) (0.017) (0.037) (0.026) (0.062) (0.039)

SCROP50 t-1 0.000 -0.015 0.017 -0.039 -0.033 0.038 (0.025) (0.030) (0.028) (0.034) (0.043) (0.04)

SCROP25 t-1 0.016 0.008 0.053 0.025 -0.057 -0.032 (0.048) (0.037) (0.053) (0.052) (0.055) (0.027)

FEMALE

-0.044 -0.245 -0.117 -0.286 0.169 -0.093

(0.12) (0.181) (0.17) (0.268) (0.288) (0.156)

District FE x x x x x x District Covariates x x x x x x District-Birth Year Trend x x x x x x Observations 3,248 3,248 2,323 2,323 925 925 Cohorts 1978-91 1978-91 1978-91 1978-91 1978-91 1978-91 Districts 14 14 14 14 14 14

Notes: Wild cluster bootstrapped standard errors in parentheses. Specifications also include birth year fixed effects, year of interview fixed effects, indicators for household religion and caste, whether the household is rural, mother’s educational attainment, and linear and quadratic terms of the mother’s age at which the child is born. The district covariates include logs of for rice and cereal productivity, patta land area distributed, number of medical institutions, and kilometres of surfaced road per capita and their corresponding interactions with the female child and the firstborn son indicators. *** p<0.01, ** p<0.05, * p<0.1

45

Table A.2: Infant Mortality, Firstborn Sons, and Registration: Bihar Control Districts

Infant Death

All Children Hindu Children Non-Hindu Children

(1) (2) (3) (4) (5) (6) (7) (8) (9)

SCROP50 t-1 * FIRSTSON * FEMALE -0.058* -0.067* -0.067* -0.077** -0.090** -0.089** -0.023 -0.027 -0.030

(0.029) (0.033) (0.033) (0.033) (0.040) (0.038) (0.045) (0.048) (0.048)

SCROP25 t-1 * FIRSTSON * FEMALE 0.020 0.007 0.008 0.010 -0.009 -0.007 0.044 0.048 0.040 (0.030) (0.035) (0.036) (0.039) (0.046) (0.046) (0.042) (0.05) (0.05)

SCROP50 t-1 * FEMALE 0.051** 0.051** 0.051** 0.066** 0.071** 0.071** 0.024 0.017 0.017 (0.024) (0.025) (0.023) (0.029) (0.031) (0.031) (0.032) (0.035) (0.036)

SCROP25 t-1 * FEMALE -0.006 -0.009 -0.010 -0.002 0.000 -0.001 -0.030 -0.044 -0.041 (0.023) (0.028) (0.029) (0.026) (0.029) (0.029) (0.033) (0.046) (0.047)

SCROP50 t-1 * FIRSTSON 0.027 0.031 0.031 0.020 0.030 0.030 0.041 0.035 0.035 (0.019) (0.023) (0.022) (0.024) (0.028) (0.028) (0.025) (0.03) (0.03)

SCROP25 t-1 * FIRSTSON -0.024 -0.019 -0.017 -0.029 -0.012 -0.013 -0.005 -0.020 -0.010 (0.024) (0.025) (0.026) (0.034) (0.033) (0.033) (0.028) (0.034) (0.032)

SCROP50 t-1 -0.043** -0.042** -0.061*** -0.035 -0.040* -0.054** -0.068** -0.055 -0.082**

(0.021) (0.021) (0.023) (0.022) (0.022) (0.025) (0.036) (0.034) (0.038)

SCROP25 t-1 0.015 0.016 -0.009 -0.001 -0.004 -0.016 0.051 0.051 -0.003

(0.02) (0.021) (0.022) (0.027) (0.025) (0.03) (0.033) (0.034) (0.037)

FIRSTSON * FEMALE -0.003 -0.004 -0.003 0.019 0.016 0.016 -0.052 -0.049 -0.044 (0.029) (0.028) (0.029) (0.033) (0.033) (0.033) (0.034) (0.034) (0.035)

FEMALE -0.023 -0.023 -0.023 -0.048* -0.047* -0.043 0.042 0.037 0.030 (0.025) (0.023) (0.024) (0.026) (0.027) (0.026) (0.038) (0.035) (0.038)

FIRSTSON 0.014 0.014 0.012 0.010 0.013 0.012 0.006 0.002 -0.006

(0.023) (0.024) (0.023) (0.03) (0.031) (0.031) (0.026) (0.027) (0.029)

District FE x x x x x x x x x District Productivity Controls x x x x x x District-Birth Year Trend x x x Observations 9,355 9,355 9,355 6,236 6,236 6,236 3,119 3,119 3,119 Cohorts 1978-91 1978-91 1978-91 1978-91 1978-91 1978-91 1978-91 1978-91 1978-91 Districts 19 19 19 19 19 19 19 19 19 Notes: Wild cluster bootstrapped standard errors in parentheses. All specifications also include the female child, birth year fixed effects, birth order fixed effects, year of interview fixed effects, indicators for household religion and caste, whether the household is rural, mother’s educational attainment, and linear and quadratic terms of the mother’s age at which the child is born. Specifications for children of birth order 2 or higher also include the firstborn son indicator and its corresponding interaction terms. *** p<0.01, ** p<0.05, * p<0.1

46

Table A.3: Younger Siblings and Registration: Bihar Control Districts

Child Has a Younger Sibling

All Children Hindu Children Non-Hindu Children

(1) (2) (3) (4) (5) (6) (7) (8) (9)

SCROP50 t-1 * FIRSTSON 0.001 -0.001 -0.001 0.001 0.000 0.004 0.004 -0.010 -0.015

(0.044) (0.044) (0.043) (0.049) (0.052) (0.051) (0.053) (0.046) (0.046)

SCROP25 t-1 * FIRSTSON -0.094*** -0.096** -0.099** -0.085** -0.083* -0.091** -0.158** -0.183** -0.176** (0.036) (0.044) (0.045) (0.037) (0.043) (0.044) (0.065) (0.077) (0.078)

SCROP50 t-1 -0.036 -0.034 -0.040 -0.018 -0.017 -0.033 -0.085* -0.078* -0.070

(0.027) (0.026) (0.031) (0.032) (0.033) (0.039) (0.049) (0.044) (0.043)

SCROP25 t-1 0.027 0.015 -0.019 0.032 0.019 -0.048 0.011 0.003 0.044

(0.032) (0.038) (0.053) (0.045) (0.052) (0.076) (0.042) (0.037) (0.065)

FIRSTSON -0.008 -0.008 -0.003 -0.029 -0.029 -0.023 0.081** 0.076* 0.065 (0.022) (0.021) (0.02) (0.026) (0.025) (0.022) (0.039) (0.042) (0.047)

District FE x x x x x x x x x District Productivity Controls

x x

x x

x x

District-Birth Year Trend x x x Observations 2,935 2,935 2,935 2,126 2,126 2,126 809 809 809 Cohorts 1978-91 1978-91 1978-91 1978-91 1978-91 1978-91 1978-91 1978-91 1978-91 Districts 14 14 14 14 14 14 14 14 14 Notes: Wild cluster bootstrapped standard errors in parentheses. The sample in each specification is children of birth order 2. All specifications also include the female child, birth year fixed effects, year of interview fixed effects, indicators for household religion and caste, whether the household is rural, mother’s educational attainment, and linear and quadratic terms of the mother’s age at which the child is born. *** p<0.01, ** p<0.05, * p<0.1

47

Table A.4: Infant Mortality, Firstborn Sons, and Registration – Rice Productivity Estimates

Infant Death

All Children Hindu Children Non-Hindu Children

B. Or. 1 B. Or. >1 B. Or. 1 B. Or. >1 B. Or. 1 B. Or. >1

(1) (2) (3) (4) (5) (6)

SCROP50 t-1 * FIRSTSON * FEMALE - -0.073* - -0.097** - -0.028

(0.039) (0.048) (0.055)

SCROP25 t-1 * FIRSTSON * FEMALE - 0.031 - 0.019 - 0.046 (0.038) (0.051) (0.044)

SCROP50 t-1 * FEMALE 0.014 0.046** 0.011 0.070** 0.025 -0.001 (0.025) (0.023) (0.026) (0.029) (0.042) (0.046)

SCROP25 t-1 * FEMALE -0.041 -0.029 -0.029 -0.027 -0.052 -0.036 (0.031) (0.032) (0.037) (0.033) (0.062) (0.048)

SCROP50 t-1 * FIRSTSON - 0.039 - 0.043 - 0.029 (0.028) (0.032) (0.031)

SCROP25 t-1 * FIRSTSON - -0.034 - -0.035 - -0.016 (0.023) (0.036) (0.03)

SCROP50 t-1 0.000 -0.063*** 0.017 -0.059** -0.033 -0.078**

(0.025) (0.023) (0.028) (0.026) (0.043) (0.035)

SCROP25 t-1 0.016 0.001 0.053 -0.003 -0.057 -0.005

(0.048) (0.020) (0.053) (0.031) (0.055) (0.039)

LN RICEPR t-1 * FIRSTSON * FEMALE - 0.043 - 0.087 - 0.087

(0.049) (0.068) (0.068)

LN RICEPR t-1 * FEMALE -0.058 0.035 -0.020 -0.009 -0.170* -0.009

(0.042) (0.044) (0.049) (0.053) (0.093) (0.053)

LN RICEPR t-1 * FIRSTSON - -0.041 - -0.109 - -0.109

(0.048) (0.065) (0.065)

LN RICEPR t-1 0.085* -0.102** 0.062 0.006 0.206** -0.308***

(0.046) (0.046) (0.053) (0.048) (0.09) (0.121)

FIRSTSON * FEMALE - -0.025

- -0.096

- 0.085 (0.141)

(0.177)

(0.219)

FEMALE -0.044 0.163

-0.117 0.171

0.169 0.174 (0.12) (0.131)

(0.170) (0.131)

(0.288) (0.247)

FIRSTSON - 0.034

- 0.061

- -0.017

(0.137)

(0.197)

(0.150)

District FE Yes Yes Yes Yes Yes Yes District Covariates x x x x x x District-Birth Year Trend x x x x x x Observations 3,248 8,367 2,323 5,448 925 2,919 Cohorts 1978-91 1978-91 1978-91 1978-91 1978-91 1978-91 Districts 14 14 14 14 14 14

Notes: Wild cluster bootstrapped standard errors in parentheses. All specifications also include birth year fixed effects, birth order fixed effects, year of interview fixed effects, indicators for household religion and caste, whether the household is rural, mother’s educational attainment, and linear and quadratic terms of the mother’s age at which the child is born . The district covariates include logs of for rice and cereal productivity, patta land area distributed, number of medical institutions, and kilometres of surfaced road per capita and their corresponding interactions with the female child and the firstborn son indicators. *** p<0.01, ** p<0.05, * p<0.1

48

Table A.5: Younger Siblings and Sharecropper Registration – Rice Productivity Estimates

Child has a Younger Sibling

All Children Hindu Children Non-Hindu Children

B. Ord. 1 B. Ord. 2 B. Ord. 1 B. Ord. 2 B. Ord. 1 B. Ord. 2

(1) (2) (4) (5) (7) (8)

SCROP50 t-1 * FIRSTSON - 0.012 - 0.008 - 0.019 (0.044) (0.051) (0.05)

SCROP25 t-1 * FIRSTSON - -0.116** - -0.108** - -0.182** (0.055) (0.052) (0.086)

SCROP50 t-1 * FEMALE 0.007 - 0.026 - -0.025 - (0.021) (0.032) (0.044)

SCROP25 t-1 * FEMALE -0.005 - -0.003 - -0.022 - (0.017) (0.026) (0.039)

SCROP50 t-1 -0.015 -0.048 -0.039 -0.038 0.038 -0.096*

(0.030) (0.032) (0.034) (0.039) (0.04) (0.051)

SCROP25 t-1 0.008 -0.020 0.025 -0.058 -0.032 0.036

(0.037) (0.056) (0.052) (0.079) (0.027) (0.083)

LN RICEPR t-1 * FIRSTSON - -0.086 - -0.083 - -0.050

(0.072) (0.085) (0.052)

LN RICEPR t-1 - -0.061 - -0.047 - -0.093

(0.075) (0.09) (0.086)

LN RICEPR t-1 * FEMALE 0.026 - 0.036 - 0.001 -

(0.038) (0.045) (0.067)

LN RICEPR t-1 -0.030 - -0.051 - 0.090 -

(0.066) (0.081) (0.103)

FEMALE -0.245 - -0.286 - -0.093 - (0.181) (0.268) (0.156)

FIRSTSON - -0.201 - -0.347 - 0.205

(0.193) (0.251) (0.138)

District FE Yes Yes Yes Yes Yes Yes District Covariates x x x x x x District-Birth Year Trend x x x x x x Observations 3,248 2,686 2,323 1,919 925 767 Cohorts 1978-91 1978-91 1978-91 1978-91 1978-91 1978-91 Districts 14 14 14 14 14 14

Notes: Wild cluster bootstrapped standard errors in parentheses. Specifications also include the female child and firstborn son indicators for children of birth order 1 and 2 respectively, birth year fixed effects, year of interview fixed effects, indicators for household religion and caste, whether the household is rural, mother’s educational attainment, and linear and quadratic terms of the mother’s age at which the child is born. The district covariates include logs of patta land area distributed, number of medical institutions, and kilometres of surfaced road per capita and their corresponding interactions with the female child and the firstborn son indicators. *** p<0.01, ** p<0.05, * p<0.1

49

Table A.6: Male Births and Registration – Rice Productivity Estimates

Child is Male All Children Hindu Children Non-Hindu Children B. Or. 1 B. Or. >1 B. Or. 1 B. Or. >1 B. Or. 1 B. Or. >1 (1) (2) (3) (4) (5) (6)

SCROP50 t-1 * FIRSTSON - -0.011 - -0.014 - 0.004 (0.020) (0.019) (0.047)

SCROP25 t-1 * FIRSTSON - -0.005 - 0.005 - -0.033 (0.027) (0.031) (0.062)

SCROP50 t-1 -0.020 0.045* -0.065 0.055** 0.091 0.039

(0.037) (0.025) (0.050) (0.029) (0.064) (0.041)

SCROP25 t-1 -0.010 0.051 0.063 0.030 -0.149 0.111

(0.034) (0.058) (0.053) (0.054) (0.098) (0.095)

LN RICEPR t-1 * FIRSTSON - -0.014 - -0.034 - -0.010 (0.029) (0.039) (0.084)

LN RICEPR t-1 -0.064 -0.009 -0.010 0.023 -0.241* -0.062 (0.098) (0.051) (0.131) (0.050) (0.136) (0.098)

FIRSTSON - 0.094

- 0.186 - -0.005

(0.096)

(0.113) (0.262)

District FE x x x x x x District Covariates x x x x x x District-Birth Year Trend x x x x x x Observations 3,248 8,367 2,323 5,448 925 2,919 Cohorts 1978-91 1978-91 1978-91 1978-91 1978-91 1978-91 Districts 14 14 14 14 14 14

Notes: Wild cluster bootstrapped standard errors in parentheses. All specifications also include birth year fixed effects, birth order fixed effects, year of interview fixed effects, indicators for household religion and caste, whether the household is rural, mother’s educational attainment, and linear and quadratic terms of the mother’s age at which the child is born. Specifications for children of birth order 2 or higher also include the firstborn son indicator. The district covariates include logs of for rice and cereal productivity, patta land area distributed, number of medical institutions, and kilometres of surfaced road per capita and their corresponding interactions with the female child and the firstborn son indicators. *** p<0.01, ** p<0.05, * p<0.1

50

Table A.7: Land Reform and Surviving Births – District-Year Fixed Effects

Dep. Variable: 1=surviving girl,

0=no birth or surviving boy 1=surviving boy,

0=no birth or surviving girl

Land category: All Landless Some land All Landless Some land

(1) (2) (3) (4) (5) (6)

Agricultural land -0.001 - -0.001 -0.000 - -0.000 (0.001) (0.001) (0.001) (0.001)

% Land registered -0.007*** 0.000 -0.009*** 0.009*** 0.020*** 0.008*** (0.001) (0.003) (0.001) (0.003) (0.007) (0.001)

Household FE x x x x x x District-Year FE x x x x x x Observations 42,304 18,404 23,900 42,304 18,404 23,900 Households 2,268 1,126 1,142 2,268 1,126 1,142 Years 1978-99 1978-99 1978-99 1978-99 1978-99 1978-99

Notes: Robust standard errors in parentheses, adjusted for clustering on villages. The variable % land registered is computed as the sum over the previous three years of the share of land affected by each program over the total cultivable land in each village, using official land records. All regressions also include a land ceiling indicator, immigrant status indicators, and year and household fixed effects, and % land distributed calculated the same way as % land registered. The land category of the household is defined based on year 1977 landholdings, and regressions include data for years 1978–1999. *** p<0.01, ** p<0.05, * p<0.1.

Table A.8: Land Reform and Surviving Births – Full Sample

Dep. Variable: 1=surviving girl,

0=no birth or surviving boy 1=surviving boy,

0=no birth or surviving girl

Land category: All Landless Some land All Landless Some land

(1) (2) (3) (4) (5) (6)

Agricultural land -0.001 - -0.001 -0.000 - 0.000 (0.000) (0.000) (0.001) (0.001)

% Land registered -0.009*** -0.003 -0.010*** 0.009*** 0.042*** 0.005*** (0.001) (0.006) (0.001) (0.002) (0.011) (0.001)

Household FE x x x x x x Observations 56,461 23,154 33,307 56,461 23,154 33,307 Households 2,286 1,230 1,403 2,286 1,230 1,403 Years 1978-99 1978-99 1978-99 1978-99 1978-99 1978-99

Notes: Robust standard errors in parentheses, adjusted for clustering on villages. The variable % land registered is computed as the sum over the previous three years of the share of land affected by each program over the total cultivable land in each village, using official land records. All regressions also include a land ceiling indicator, immigrant status indicators, and year and household fixed effects, and % land distributed calculated the same way as % land registered. The land category of the household is defined based on year 1977 landholdings, and regressions include data for years 1978–1999. *** p<0.01, ** p<0.05, * p<0.1.