Mortality, Gender and India’s Religious Demography ∗

Vani Borooah†

Quy-Toan Do‡

Sriya Iyer§

Shareen Joshi¶

4 August 2010

This paper explores the relationship between religion, infant mortality and fertility across Hindusand Muslims in India using recent data from the 2006 National Family Health Survey. The analysisshows that Muslims exhibit lower infant mortality rates relative to Hindus, and that this difference isnot adequately explained by socio-economic status, location and policy variables. We argue that thecombination of gender preference and heterogeneity in desired fertility across religious groups offers anexplanation for the observed differences in mortality rates. Our data support this view; the differencein infant mortality between Muslims and Hindus is concentrated at higher birth order and among girls,irrespective of their birth order.

JEL Classification Cide: C25, J13, J15, O53, Z12

Keywords: Religion, infant mortality, desired versus realized fertility, sex ratio, gender bias

1 Introduction

India is well-known for striking demographic patterns: excess female mortality and significant regionalvariation in fertility as well as child mortality (Visaria 1971, Dyson and Moore 1983, Sen 1992 2001,Kishor 1993, Das Gupta 2005). 1 A recent literature argues that demographic indicators also varysignificantly across India’s religious groups. Fertility is known to be higher among Muslims than it is∗For helpful discussions we are grateful to Robert Barro, Eli Berman, Maitreyi Das, Edward Glaeser, Timothy Guin-

nane, Monica Das Gupta, Partha Dasgupta, Larry Iannaccone, Siddiq Osmani, Abusaleh Shariff, Chander Velu, DavidVoas, and to participants at the American Economic Association 2007 meetings in Chicago, the ‘Religion, Economics andCulture’ 2004 Annual Meetings in Kansas City, Missouri, and the 2005 ESRC seminar on ‘Demography and Religion’ inLancaster, UK. We are also grateful to Sarina Siddhanti for excellent research assistance, We are especially grateful tothe Editor and two anonymous referees for their invaluable comments.†Faculty of Economics and Politics, University of Ulster, Northern Ireland. vk.borooah@ulst.ac.uk‡Development Economics Research Group, The World Bank. qdo@worldbank.org§Corresponding author: Faculty of Economics and St. Catharine’s College, University of Cambridge, Sidgwick Avenue,

Cambridge CB3 9DD, United Kingdom. Tel: 44 1223 335257; Fax: 44 1223 335475; Email: Sriya.Iyer@econ.cam.ac.uk¶The School of Foreign Service, Georgetown University. shareen.joshi@gmail.com1According to the 2001 Census, India has 933 females for every 1000 males, which implies that as many as 35–37

million women in India may be “missing” (Drèze and Sen 1996, Klasen 1994, Agnihotri 2000, Sen 2003, Oster 2005).Theestimate of the number of missing women is based on comparisons with Europe and North America, which have 1050females per 1000 males.

1

among Hindus (Basu 1997, Jeffery and Jeffery 1997, Iyer 2002, Dharmalingam and Morgan 2004).2

Muslim infant mortality among Muslims, at 59 per 1000, is also lower than for Hindus, which standsat 77 per 1000 (Boorah and Iyer, 2005a). Similarly, child mortality, at 83 per 1000 for Muslims, issubstantially lower than among Hindus, at 107 per 1000 (IIPS and ORC Macro International 2000).The survival advantage of Muslim children despite the higher fertility of their mothers is particularlypuzzling in light of the higher levels of poverty in this community as noted by the recent and well-publicized Sachar Committee Report (Government of India 2006).

This paper extends the literature on religion and demography in India by performing a comparativeanalysis of infant mortality across religious and caste groups using data from the National FamilyHealth Survey of 2005-06. It confirms that Muslim children show lower risks of mortality than Hinduchildren. The differential in mortality is not adequately explained by socio-economic, regional, orpolicy variables. We provide a novel explanation for these patterns; Muslims prefer larger familiesthan Hindus do, which implies that a child born to a Muslim family is more likely to be within optimalfamily size than a child born to a Hindu family. With gender preferences, this effect is even true whenthe first born is a girl. Empirically, we expect mortality differences to be starker among higher-orderbirths and among girls, irrespectively of their birth order.

Our analysis divides our population into three subgroups: non-Dalit Hindus (or upper-caste Hindus),Dalit Hindus (or lower-caste Hindus), and Muslims. We estimate survival probabilities for childrenand control for a number of important socio-economic, geographic and demographic variables suchas socio-economic status, nutrition, and female labour force participation. Our findings confirm thatMuslim infant and child mortality is considerably lower than that for the Hindus, even after accountingfor the wide range of socio-economic characteristics available to us. The effect is particularly strong forgirls. There are also differences between Dalit Hindus and upper-caste Hindus, but these differencesdiminish in significance with the inclusion of control variables. This suggests that the critical differencein survival probabilities is across, rather than within, religious groups.

Our explanation of the lower mortality of Muslims begins with the observation that they have higherfertility than Hindus, and this is not entirely explained by their observable characteristics such associo-economic status or location. We infer from this that Muslims prefer larger families than Hindusand argue that consequently, a Muslim child is more likely to be born within optimal family size andthus experience a lower risk of mortality. The higher Hindu mortality risks are likely to grow at higherorder births that come close to or exceed optimal family size. Empirical evidence provides supportfor this hypothesis. Our results also highlight the role of gender. First born Hindu girls (particularlyupper-caste Hindu girls) display higher mortality risks than their Muslim counterparts. Our conjectureis that even though both Hindus and Muslims may prefer sons over daughters, the Hindu preferencefor smaller families may induce a higher levels of daughter-aversion as early as the first birth, resultingin a higher female mortality rate for this group. Taken together, our results lead to the conclusion thatdiscussions about “missing women” and excess child mortality in India may be significantly enrichedby considering the effects of religion and differences in desired fertility.

The rest of the paper proceeds as follows: Section 2 presents the empirical findings; Section 3 explores2An often-cited figure from the National Family Health Survey conducted in 1998-99, shows that the Total Fertility

Rate (TFR) for Hindus and Muslims was 2.8 and 3.6 respectively.

2

the socio-economic determinants of infant morality. Section 4 discusses how religion, fertility andmortality might be linked. Section 5 concludes.

2 Child Mortality and Its Socio-Economic Determinants

Our description of contemporary India uses data from the 2005-06 National Family Health Survey,henceforth NFHS.3 The NFHS interviewed a total of 123,385 women in 29 states of India and isintended to be representative at both state and national levels. The religious composition of thehouseholds is consistent with the findings of the 2001 Indian Census (henceforth Census): 82 percentof households are Hindu, 13 percent are Muslim, 3 percent are Christian, 2 percent are Sikh, and theremainder are of other religious affiliations. Among Hindus, the caste composition of our populationalso mirrors the one obtained from the Census; 44% of Hindu respondents reported that they belongedto a “Scheduled Caste” or a “Scheduled Tribe”, henceforth referred to as Dalits, and 56% of the Hindupopulation was higher caste or non-Dalit; the term “Dalit” translates as “the downtrodden”, and refersspecifically to India’s scheduled castes and scheduled tribes - these are those castes and tribes recognisedby the Indian Constitution as deserving special recognition in respect of education, job reservation inemployment, and political representation.

For most of the analysis in this paper, we restrict our sample to only Hindu and Muslim women whoare married. This generates a sample of 93,796 women. 84,609 of these women report that they havehad at least one child. The remainder are childless.4 Panel A of Table 1 presents summary statistics ofkey variables for the entire female sample, as well as the limited sample that excludes childless women.We also conduct extensive analysis on the child sample, which includes 256,782 children who are bornto 84,609 mothers in the female sample. The information was gathered from birth histories and istherefore representative of children who are alive as well as those who died; it also includes childrenliving within the household and those who left. Summary statistics for the child sample is presentedin Panel B of Table 1.

The basis of our analysis is a comparison between the demographic and socio-economic differencesbetween three mutually exclusive subgroups of India’s population: Muslims, Hindu Dalits and HinduNon-Dalits. For the sake of convenience, we refer to these three groups as Muslims, Dalits and Non-Dalits respectively. We focus on these mutually exclusive categories for two reasons. First, while 30%of Hindus are Dalits, the proportion of Muslims reporting Dalit status is approximately 2–5% (SacharCommittee Report 2006). In our sample, only 452 Muslim women reported themselves as Dalits. Weincluded this group in our Muslim sample, and excluded them from the Dalit sample. A second reasonis that among Hindus, the criteria for being a Dalit are widely recognized, since the Indian governmentmaintains official lists of “Scheduled Castes” and “Scheduled Tribes”. The overlap between Dalit and

3The first NFHS survey was conducted in 1992-93, and the second in 1998-99. All three surveys were conducted inconjunction with the Ministry of Health and Family Welfare (MOHFW),Government of India. Funding for the surveywas provided by the United States Agency for International Development (USAID), the United Kingdom Departmentfor International Development (DFID), the Bill and Melinda Gates Foundation, UNICEF, UNFPA, and the Governmentof India. Technical assistance was provided by Macro International Maryland, USA.

4The average age of childless women in this sample is 24 years and the standard deviation is 7.38 years. Many ofthese are women who have married, but not yet moved to the homes of their husbands. In India, the date of marriageand the date of the move to a husband’s house may be several years apart.

3

Non-Dalit is thus close to zero. The criteria for being a Dalit for non-Hindus are much less clear. Inthe case of inter-caste marriages, we follow Hindu civil law and assign the woman the caste of herhusband.

2.1 Descriptive statistics

We begin our analysis with a comparison of the three aforementioned groups along several key demo-graphic variables. Table 2 presents two comparisons of Muslims with (a) Non-Dalits, and (b) Dalits.Since the focus is on child mortality, we focus on the sample of women who have ever had children, aswell as the sample of ever-born children.5 Unconditional estimates, or rather simple differences in themeans of fertility and mortality variable are presented in Table 2. Note that compared to Non-DalitsHindus, Muslims are more likely to experience the death of an ever-born boy, and that the differenceis statistically significant. There is however, no statistically significant difference in the deaths of ever-born girls. Muslims also report higher levels of fertility, lower rates of permanent contraception, greaterpreferences for ideal numbers of boys and girls, and a lower age at first marriage. Muslims also appearto have higher female-male sex-ratios than Non-Dalits and appear to be less averse to having girlsrelative to boys.6 In all these cases, the differences are statistically significant. The low female-maleratio at birth may at least in part be attributed to the fact that the practice of sex selective abortionis higher among upper-caste Hindus. This group has been shown to engage in this practice more thanother groups (Arnold, Kishor and Roy 2002). However, since the NFHS data do not contain informa-tion on the prevalence and practice of abortion, this is one aspect that we are not able to investigatecomprehensively and remains beyond the scope of this paper. However, we do acknowledge, readily,the importance of these practices for the subject of this research.

Comparisons between Muslims and Dalits are also highlighted in Table 2. Muslims show lower levelsof male as well as female mortality, higher levels of fertility, lower levels of contraceptive use andlower ages at marriage. There is however, no statistically significant difference in sex-ratios of childrenever-born or children alive.

These patterns are also seen in the child sample. More Muslim children are reported to have died thanDalit children, and there are more girls alive than boys. There is also evidence that Muslim deathsoccur earlier than Non-Dalit deaths. The difference is statistically significant and the point estimateis around 2 months. In the sections below, we focus exclusively on the inter-group differences in

5The proportion of childless women does not significantly differ between Hindus and Muslims. 9.8 percent of Hindusand 9.7 percent of Muslims report that they are childless. The difference, 0.1 has a t-statistic of 0.20 and a p-value of0.838.

6Sex-ratios are measured at the level of the cluster/village rather than at the level of a woman. Since a sex-ratiois defined as the number of females relative to the number of males, it can only be constructed for a woman who hashad at least one male birth. When aggregating at the level of the village or cluster (which was the NFHS primarysampling unit) however, this problem is alleviated, since it is an average of female and male deaths for a broad groupof women. In results not shown here, we also examine the sex-ratios of ever-born children for Dalits, Non-Dalits andMuslims for women over the age of 40, a group that has presumably completed child-bearing. We found that thesesex-ratios were far less distorted, and were on average at 1068 for all three groups. There was no statistically significantdifference between them. We find this most interesting, but are unable to analyze these differences in sex-ratios acrosscohorts more extensively. This is mainly because the estimation of sex-ratios for particular age-groups results in a lossof precision for the estimates. This is further complicated by the possibility of measurement error in reporting of birthsincreases with the age of the respondent.

4

child mortality and examine whether these differences persist once we control for other socio-economic,locational and nutrition-related variables.

2.2 Determinants of child mortality

Our empirical analysis of the determinants of mortality is based on a probit model. For each child i, wedefine two dummy variables that take value 1 if the child died before the age of 1 and 5, respectively,and 0 otherwise. We regress each of these on dummy variables that indicate whether the child’s parentsare Dalit Hindus or Non-Dalit Hindus (Muslim is therefore the omitted category). We also includea vector of observable characteristics Xi that capture potential risk factors. As before, we refer tothe two groups in the analysis as simply “Dalits” and “Non-Dalits”, respectively. Our working sampleinitially includes 218,769 children born to 79,118 mothers in the NFHS. Of the 218,769 children whowere included in our sample, 25,784 (10.4 percent of the total) had died. The children’s ages rangefrom 0–37.7 For children who died, the average age at death was 20 months.

Our control variables include a variety of other individual, family and regional characteristics. Amongindividual characteristics, we include a child’s age, gender and a dummy variable that indicates whetheror not the child is first-born. We further include information on the child’s parents; father’s age,mother’s age and age at marriage, and dummy variables for whether a child’s mother and father hadcompleted primary school. We control for economic status using household land holdings (in acres)and a household wealth index that has been tested in a large number of countries and has been shownto be consistent with expenditure and income measures (Rustein 1999, Filmer and Pritchett 2001).8

We also control for whether the family resides in a rural area. Furthermore, we add to the regressionspecifications information on nutritional intakes and access to basic health care services. Becausereligious communities in India have a variety of dietary restrictions and prescriptions, we includedummy indicators for whether any meat, dairy or plant-protein was consumed in the 24 hours priorto the survey. To address the possibility that some groups may also receive better access to healthcare than others, we include cluster-level averages of vaccination rates for measles, polio, tuberculosisand DPT. The reliance on aggregates instead of individual-level or household-level variables for theseindicators is particularly important in addressing the potential endogeneity issues in the relationshipbetween mortality and vaccinations. For similar reasons, we also include cluster-level averages offemale labor force participation rates. Finally, we add state fixed-effects to control for the fact thatsome Indian states just have historically had distinctive demographic patterns (Dyson and Moore 1983).The subset of observations that has information on all these characteristics gives us a final sample of

7For those children who had died, the age variable was coded as age-at-death.8The wealth index is constructed by combining information on 33 household assets and housing characteristics such as

ownership of consumer items, type of dwelling, source of water, and availability of electricity into a single wealth index.These 33 assets are as follows: household electrification, type of windows, drinking water source, type of toilet facility,type of flooring, material of exterior walls, type of roofing, cooking fuel, house ownership, number of household membersper sleeping room, ownership of a bank account, ownership of a mattress, a pressure cooker, a chair, a cot/bed, a table,an electric fan, a radio/transmitter, a black and white television, a color television, a sewing machine, a mobile phone,another other telephone, a computer, a refrigerator, a watch or clock, a bicycle, a motorcycle or scooter, an animal-drawncart, a car, a water-pump, a thresher, and a tractor. Each household asset is assigned a weight (factor score) that isgenerated through principal components analysis. The resulting asset scores are standardized in relation to a normalhousehold and is then assigned a score for each asset, and scores were summed for each household; individuals are rankedaccording to the score of the household in which they reside.

5

195,080 children.

Marginal effects from the basic probit model are presented in Tables 3 and 3. Three specifications areestimated for each of the two dependent variables (death before the age of 1, Table 3; and death beforethe age of 5, Table 3). The first is the simplest version, with no control variables. The second includesthe full set of controls. The third is the includes the full set of controls, but is restricted to childrenwhose mothers are older than 40, a subgroup for which mothers have in all likelihood completed theirchild-bearing. All models include state fixed-effects. Note that in the specification with the full setof controls (models 2, 3, 5 and 6 in both Tables 3 and 3) the variables that are of most interest tous – Dalit and Non-Dalit – take a positive sign and are statistically significant at the 1 percent level,indicating that individuals in both these groups face higher mortality risks than Muslims. The controlvariables take the expected signs and will be discussed in the next section.

These results permit us to evaluate a range of mechanisms that can drive the lower observed mortalityamong Muslims. These are discussed in the next section.

2.2.1 Socio-economic status

A vast literature in economics and the social sciences in general confirms that socio-economic status,as measured by income or wealth, parental education (particularly mother’s education) and location(rural versus urban) is one of the primary determinants of the risks of child mortality (Strauss andThomas 1995). The first confounding factor we need to investigate is thus socio-economic status:households from Muslim communities might simply benefit from favorable socio-economic conditionsthat are conducive to lower mortality rates among children. However, this hypothesis does not seemto be supported by the data; the Sachar Committee Report (2005), for example, argues that bothMuslims and Dalits are substantially poorer than their upper-caste Hindu counterparts. Comparisonsusing our dataset are presented in Appendix Table A. Relative to Non-Dalits, Muslims have lower ratesof schooling, lower ownership of farmland and a higher chance of falling below the poverty line. In manyrespects, Muslims have similar socio-economic characteristics to Hindu Dalits. Thus, in all likelihood,socio-economic status cannot by itself explain the lower mortality of Muslim children compared to theirNon-Dalits counterparts.

Our main specifications control for socio-economic status (see Tables 3 and 3, columns 2, 3, 5 and 6).The coefficients on socio-economic controls are as expected: mortality risks are lower when mothersmarried older but or younger at time of birth, are educated and households are wealthier (columns 2in Tables 3 and 3). Religion, while its influence is reduced when adding education controls, continuesto have a statistically significant effect on mortality risks.

2.2.2 Nutrition

A second possible mechanism is nutrition. It is well-known that levels of nutrition in India are low(Dreze and Sen, 2001). In our own dataset, it is documented that 46 per cent of children are underweightand 59 per cent of pregnant women, and 63 per cent of lactating women are anemic (IIPS, 2007).Appendix Table B indicates systematic differences in nutrition and preventive health behavior between

6

Muslims and Hindus. While Hindus are less likely to eat meat and dairy, they are more likely to bevaccinated against measles, tuberculosis, polio and DPT. Furthermore, Muslim women are found tohave significantly higher levels of hemoglobin. As noted earlier, adding nutrition variables to our mainprobit specification indicates that eating meat (which can also be an indication that the household iswell-off) is associated with lower mortality. Elevated hemoglobin levels are also correlated with lowermortality risks. Finally, vaccination rates do not have consistent signs since rates can be driven byeither supply or demand for vaccination. However, whether or not we control for these variables, thecorrelation between religion and child mortality is barely affected (columns 3 in Tables 3 and 3).

2.2.3 Female labor force participation

Finally, the results so far eliminate the possibility that differences in mortality rates across groups aredriven by differences in female labor force participation rates. In Appendix Table C, we indeed findthat Muslim women work less than Hindu women and are also less likely to be self-employed or workaway from home. This can translate into more time dedicated to child care and ultimately result inlower mortality risks for children.

To account for female labor force participation, and since female labor-force participation may beendogenous, we do not include a direct control for female labor force participation in our regressionsand rely instead on a community level average. Controlling for aggregate measures of female labor forceparticipation at the cluster level, as shown in columns 4 and 5 in Tables 3 and 3, barely affects thecoefficient on the Dalit or Non-Dalit variables. Admittedly, equilibrium female labor-force participationmeasures are endogenous to the outcome of interest, but adding them as controls does not affect therobust association between religion and mortality.

A word is in order about the geographic variation in mortality risks. We do not present coefficientsfrom the specification with state-level fixed effects here, but we found that geography did in fact playa role, with lowest levels of mortality in the South, particularly in the states of Goa and Kerala, andhigher levels of mortality in the North, particularly in Delhi, Uttar Pradesh, Bihar, Arunachal Pradesh,Tripura, Assam and Jharkhand. This is consistent with the broad differences in human developmentthat have been noted for India in the past (Dyson and Moore 1983, Dreze and Sen 2001). 9 Finally,other econometric specifications (linear probability models or Cox proportional hazard models) deliverqualitatively similar results (results not shown here).

The association between religion and mortality is apparently robust to various specifications. The restof the paper is devoted to unbundling the possible channels that could drive the observed correlation.We now turn to alternative explanations for the religious differences in child mortality.

9There is no evidence that the size of the Muslim population explains these effects. In a separate analysis that isnot shown here, we also explore the regional patterns of mortality, and more precisely the heterogeneity of the Muslim-infant mortality association across regions. Among our explanatory variables, we include interactions of the variableMuslim with dummy variables corresponding to different regions. The results indicate that the interaction, and theoverall regional effect is most significant in the states and Union Territories of the North West (Jammu and Kashmir,Himachal Pradesh, Uttaranchal, Haryana, Punjab, Delhi, Rajasthan and Uttar Pradesh). The regional effect is in factinsignificant in the South (Goa, Karnataka, Kerala and Tamil Nadu) as well as in the North-East (Sikkim, ArunachalPradesh, Nagaland, Manipur, Mizoram, Tripura, Meghalaya, Assam, West Bengal and Jharkhand). In other regions, theMuslim effect as well as the regional effect remain significant, but less so than in the Northwestern states. The strongerdisparities between Hindus and Muslims in the North and in particular the North West is consistent with the findings ofprevious studies (Dyson and Moore 1983, Dreze and Sen 2001, Jejeebhoy and Sathar 2001).

7

3 Religion, Fertility and Child Mortality

We investigate an alternative explanation for the differences in child mortality by first noting that thesegroups also vary in their fertility. Evidence from both developed and developing countries suggeststhat early child bearing, short (previous and subsequent) birth intervals, and high overall fertility canall have strong effects on infant and early child mortality.10

3.1 Religious differences in fertility

As was already noted in the introduction, it is well-known that Hindus and Muslims show very differentlevels of fertility. Muslim women on average bear larger numbers of children at earlier ages than womenfrom other religious groups. Table 2 lends support to this hypothesis. Muslim women report higherfertility as well as higher levels of “ideal” fertility, for both boys and girls. They are also substantially lesslikely to use both temporary and permanent methods of contraception. The prevalence of sterilization– the dominant method of contraception in India – is about half the level among Muslims as Non-Dalits, and the use of temporary methods of contraception is about a third lower. Estimates of thesedifferences are statistically significant at the 1 percent level.

These patterns are also seen in the results of a regression of observed fertility on religion with the sampleof all women. This analysis is conducted on the sample of all women, including childless women, whoare older than 40 years of age. Summary statistics for this sample of women are presented in Table 1.Estimation on this limited sample avoids issues of selection that may be caused by differences in ages atchild-bearing, and varying rates of childlessness across the three groups. Results are presented in table4 with and without a set of control variables that include a woman’s age, age at marriage, education,husband’s age, husband’s education, and socio-economic status, as measured by land-holdings, a wealthindex and rural location. Note that the variable Non-Dalit has a negative and significant coefficientin all the regressions (column 2, 3 and 4). The coefficient for Dalit becomes negative and statisticallyin the full specification, with all controls and fixed effects (column 4). We present one specificationomitting woman’s age and age at marriage on the grounds of the possible endogeneity of the twovariables (column 3), but even here, we conclude that on average, Muslims show higher fertility thanNon-Dalit Hindus.

Higher levels of fertility among Muslims is not however, prima facie consistent with the finding oflower infant and child mortality documented earlier in this paper, and is in fact the exact oppositeof what we would expect based on the findings of mortality alone, since higher mortality is typicallyassociated with higher fertility. Before digging into this puzzle further, we look at economic, culturaland religious foundations for fertility differences across religious groups. Several factors might explainhigher fertility among Muslim families. We review them in turn.

Socio-economic status The relationship between fertility and socio-economic status is the subject ofa vast literature. Increases in income are typically associated with increased education, literacy, femalelabor force participation, and chances of child survival. All of these factors decrease the benefits of

10See Commission on Behavioral and Social Sciences and Education (1998) for a review of the literature.

8

additional children and raise the costs (Becker 1991). The impact of socio-economic status on fertility isevident in our simple regression of female fertility on indicators of religion as well as numerous measuresof socio-economic status (Table 4). Many of the coefficients for the measures of socio-economic statustake the requisite signs. Women who delayed marriage, who have some education, and who residein wealthier households have fewer children. All regressions include state-level fixed effects to controlfor differences in state-level factors that may affect fertility levels. Socio-economic conditions thus donot entirely explain the larger fertility outcomes of Muslim women compared to Dalit and Non-DalitHindus. Our results echo those of Almond, Elund, and Milligan (2009) who find that sex-ratios amongimmigrant communities in Canada differ significantly across religious groups. They find relativelynormal sex-ratios in the Muslim and Christian community, but highly skewed sex-ratios among otherreligious groups and argue that observable socio-economic characteristics do not account for thesedifferences.

Minority group status A second reason that might explain higher Muslim fertility is the “minoritygroup status” hypothesis (Goldscheider and Uhlenberg 1969). This is the idea that their political andsocial insecurity as a minority group creates incentives for higher fertility than the Hindu majoritygroup. This is likely to be particularly important where the minority feels threatened by the majoritycommunity in political, economic or social spheres (Van Heek 1966: 125-38, Stinner and Mader 1975:53-9). This is also likely where identification with religion can be used for economic gain and rent-seeking activities. For example, greater numbers may mean better representation for minority groupsin government, and greater voice in the allocations in budgets for education, health, water and so forth.

We concede that these factors may be drivers of higher fertility among Muslims, but we do not believethis is an adequate explanation for the patterns seen in the data. This is for three main reasons. First,in the context of a country like India, there are many different minority groups including not onlyMuslims but also Sikhs, Buddhists, Christians, etc. Muslim fertility remains higher than the averagefertility in these other minority groups which are typically smaller and have even greater incentive toboost their group size. Second, in including state-level fixed effects in our specification, we controlfor many of the historical, institutional and cultural factors that are likely to affect the variations ofpolitics across states. Third, in results not shown here we included estimates of the size of the Muslimpopulation (current as well as historical) as dependent variables and found that the results were notsignificantly different than the results we report here with the inclusion of state fixed effects. So whilewe concede that the minority group hypothesis may be an explanation for higher Muslim fertility inIndia, it is not the only important mechanism.

Religion The final explanation for why we might observe higher Muslim fertility is the effect ofreligion itself. A number of empirical studies have argued that the particular philosophical contentof Islam affects demographic behaviour (Youssef 1978, Coulson and Hinchcliffe 1978, Qureshi 1980,Gellner 1981, Gallagher and Searle 1983, Lutz 1983, Caldwell 1986, Obermeyer 1992, Jeffery andJeffery 1997). Three main mechanisms have been highlighted. First, is Islam’s position on birthcontrol and abortion. Some have argued that the general opposition to both practices induces higherfertility (Youssef 1978). The extent to which Islam influences the decision to control births however, is

9

strongly debated and varies across different schools of Islamic jurisprudence (Obermeyer 1992). Someschools do permit abortion up to the time when the foetus is regarded as being ‘ensouled’, a definitionwhich varies to include the 40th, 80th or 120th day of pregnancy, depending on the school, afterwhich abortion is prohibited by all schools (Obermeyer 1992, Musallam 1983). A second mechanismthrough which Islam may influence fertility is its positions on marriage. In the Koran, Muslim malesare encouraged to marry, and the early and universal remarriage of widowed and divorced women ishighly encouraged (Youssef 1978, p. 88). According to Sharia law - the basis of laws applicable toMuslims in India since 1937 - a man may take multiple wives, and in some cases make a unilateraldecision to divorce (Coulson and Hinchcliffe 1978, pp. 37-38). This creates incentives for women tohave large numbers of children since they rely on children to secure their status in their husband’shomes, or provide financial and social security in the event of divorce (Youssef 1978).

Such arguments on the relationship between religion and fertility must however, be viewed with cau-tion. The issues are strongly debated within Islam, and a clear consensus is lacking. Moreover, inIndia, Hindu and Muslim marriage patterns display similarities: early marriage is widespread in mostcommunities, and polygyny is relatively uncommon among Muslims (Indian National Social ActionForum Manual 1994). The variations in patterns of marriage are known to be greater across regionsrather than religions (Jejeebhoy and Sathar 2001).

An alternate explanation for the observed differences in fertility is that Muslim families have a lowerlevel of “daughter aversion”, i.e. a disinclination to have daughters that is driven by the higher costsof raising them (Borooah and Iyer 2005a). Furthermore, we postulate that the costs of raising girlsdiffer significantly across religious groups. These differences are evident in the norms and regulationsof marriage. An Islamic marriage or the nikah, is defined as a civil contract (Azim 1997).11 Parentsand guardians exercise control over the selection of marriage partners, but within-family marriages arecommon (between cousins) and a dower or “bride price” is paid to the bride or her guardian (Youssef1978, p.78). The Koran recognizes the possibility of divorce and encourages remarriage of divorcedor widowed women (Qureshi 1980: 564, Youssef 1978: 88, Coulson and Hinchcliffe 1978: 37-38).Hindu marriage presents a stark contrast. While there are indeed significant regional variations, Hindumarriages are typically between strangers, and involve dowries (Edlund 1999, Bhat and Zavier 2003,Rao 1993, Bloch and Rao 2002, Botticini and Siow 2003). Bride-givers are regarded as below bride-takers. Women do not have property rights over these dowries and they are not returned to her or herfamily in the event of divorce or widowhood. Opportunities for remarriage in the event of divorce orwidowhood are limited and rare. A daughter’s marriage is thus typically described as kanyadaan, whichcan be translated as the “donation” of a daughter. Such a donation is believed to benefit Hindu familiesboth socially and religiously (Niraula and Morgan 1996). Such distinctions between the “contractual”versus the “donational” notion of marriage in Islam compared to Hinduism may have implications forthe relative costs and benefits of having sons and daughters. Thus, if having daughters comes at ahigher cost to Hindus that it does to Muslims, one can expect optimal family size to be lower for theformer religious group.

We conclude from this discussion that Muslims likely have higher levels of desired fertility than both11For a Muslim marriage to be legally valid, it needs to meet four conditions: proposal by one party; acceptance by the

other; the presence of a sufficient number of witnesses (two in Sunni law); and a formal expression of both the proposaland the acceptance at the same meeting (Azim 1997).

10

Dalits and Non-Dalits. Our analysis suggests that this is not entirely driven by differences in socio-economic status or location. Other factors, such as minority status and varying levels of son-preferenceor daughter-aversion, may play a role. In this paper, we are agnostic about the strength of suchmechanisms and acknowledge that there may be other factors. We simply underscore the fact thatMuslims prefer larger families than Hindus and this is not attributable to observable characteristics inour data.

3.2 Fertility, birth order, gender preference and child mortality

Our explanation for the differences in infant mortality rates between religious groups centers on theconjecture that optimal household size varies across groups.

Birth order and religious differences in child mortality Based on the findings and the dis-cussion of the previous section, we take it as given that Muslims have higher fertility, and a taste forlarger families. Thus, the likelihood that a child is unplanned and consistute a “household size shock”is lower among Muslim households. As a consequence, one expects lower risk for unplanned children;this effect is likely to be most pronounced among higher birth-order children, regardless of their gender,since optimal household size for both Hindus and Muslims is presumably larger than one. This leadsto the prediction that religious differences in infant and child mortality will be lowest among first-bornchildren.

In order to formally test our claim, we return to the probit model in Section 2 and include interactionterms between Dalit (and Non-Dalit) dummy variables and birth order of the child. More specifically,we interact the religious or caste group with a dummy variable indicating whether the child is the firstborn. Since we predict that religious preferences in child mortality becomes starker for higher birthorder children, we expect the sign on the interaction between religion and birth order to be negative.This prediction is confirmed in Table 5, columns 1 and 5. We also present results for a sample of womenwho are likely to have completed child-bearing (columns 4 and 8). The results are qualitatively similar,although no longer significant at conventional confidence levels for the interaction Dalit×FirstBorn.

Thus, the religious differences in child mortality are concentrated among higher order births, lendingsupport to the view that differences in optimal family size might be driving our findings.

Gender and religious differences in child mortality Finally we look at the gender differencesin child mortality, and how birth order affects these differences. Consistently across specifications, agender-bias is detected whereby boys have a survival advantage compared to girls across the board(columns 2 and 6 in Table 5). However, the triple interaction FirstBorn × NonDalit × Female

indicates that first born girls of Non Dalit communities do have a survival disadvantage (columns 3and 7). The coefficient is positive and significant at the 10 percent level, although not robust when werestrict our sample to women over the age of 40 (columns 4 and 8).

This finding is consistent with our postulated view that girls of Muslim communities have a survivaladvantage compared to Hindu girls and that the effect is statistically significant for first births, sinceoptimal family sizes are substantially lower among Hindus than they are among Muslims.

11

Overall, our findings suggest that the main mechanism that accounts for religious differences in childmortality is the combination of smaller optimal family sizes for Hindu households and a preference forboys. In other words, the observed differences in child mortality across communities could be explainedby Hindu households wanting small families with a majority of boys. Note that this explanation makesno assumptions about differences in son-preference or daughter-aversion across Hindus and Muslims.The effects documented here would however be reinforced if Muslims genuinely face lower costs ofraising daughters, and are less averse to having daughters. We believe that this is an interesting andimportant topic for future research.

4 Conclusion

In this paper, we use recent data from the 2006 National Family Health Survey to explore the rela-tionship between religion, fertility and mortality outcomes across two major religious groups – Hindusand Muslims – in India. We find that while Muslims are poorer and have more children than Hindus,they also exhibit lower infant mortality rates. Socio-economic status, location and policy variables donot adequately explain this mortality difference. We provide a novel explanation. We argue that sinceMuslims prefer larger families, they are more tolerant of having daughters, particularly early on intheir fertility history. Hindus on the other hand, desire small families, and will want to have sons asearly as possible. This results in skewed sex-ratios among first- and second-born children in the Hinducommunity. Our empirical analysis supports these conclusions. Muslim girls, particularly those whoare first-born show the lower mortality risks than their wealthier counterparts in the Non-Dalit Hinducommunity. We conclude from this that the interaction between religion and demographic behavior inIndia is worthy of greater attention from both academics and policy-makers.

References

[1] Al Ghazzali, M. H. 1909. The Alchemy of Happiness. Translated from the Hindustani by ClaudField, London: John Murray.

[2] Agnihotri, S. 2000. Sex Ratio Patterns in the Indian Population. New Delhi: Sage Publications.

[3] Almond, D., Edlund, L., and Milligan, K. 2009. “Son Preference and the Persistence of Culture:Evidence from Asian Immigrants to Canada.” NBER Working Paper 15391.

[4] Arnold, F., Kishor, S. and T. K. Roy. 2002. “Sex-Selective Abortions in India.” Population andDevelopment Review 28 (4): 759-785.

[5] Azim, S. 1997. Muslim Women: Emerging Identity. New Delhi: Rawat Publications.

[6] Basu, A. 1997. “The ’Politicization’ of Fertility to Achieve Non-Demographic Objectives.” Popu-lation Studies 51: 5-18.

[7] Bhat, M and F. Zavier. 2003. “Fertility Decline and Gender Bias in Northern India.” Demography40 (4): 637- 657.

12

[8] Bloch, F. and V. Rao. 2002. “Terror as a Bargaining Instrument: A Case Study of Dowry Violencein Rural India.” American Economic Review 93(4): 1385-98.

[9] Borooah, V. 2004. “Illuminating the Politics of Demography: A Study of Inter Community FertilityDifferences in India.” European Journal of Political Economy, 20: 551-578.

[10] Borooah, V and S. Iyer. 2005a. ‘Religion, Literacy, and the Female-to-Male Ratio in India’, Eco-nomic and Political Weekly, Special Issue on Religion and Fertility, January.

[11] Borooah, V and S. Iyer. 2005b. “Vidya, Veda and Varna: The Influence of Religion and Caste onEducation in Rural India.” The Journal of Development Studies 81(4): 1369-1404.

[12] Borooah, V and S. Iyer. 2005c. “The Decomposition of Inter-group Differences in a Logit Model:Extending the Oaxaca-Blinder Approach with an Application to School Enrolment in Rural India.”Journal of Economic and Social Measurement 30(4): 279-293.

[13] Borooah, V. and S. Iyer. 2004. “Religion and Fertility in India: The Role of Son Preferenceand Daughter Aversion.” Cambridge Working Papers in Economics 436, Faculty of Economics,University of Cambridge.

[14] Botticini, M and A. Siow. 2003. “Why Dowries?” American Economic Review 93(4): 1385-98

[15] Caldwell J. C., P. H. Reddy and P. Caldwell. 1983. “The Causes of Marriage Change in SouthIndia.” Population Studies 37 (3): 343-361.

[16] Chen, L. C. 1982. “Where have the Women Gone? Insights from Bangladesh on the Low Sex Ratioof India’s Population.” Economic and Political Weekly, 17.

[17] Commission on Behavioral and Social Sciences and Education. 1998. From Death to Birth: Mor-tality Decline and Reproductive Change, Montgomery and Cohen (ed.), Washington DC: NationalAcademy Press.

[18] Coulson, N. and D. Hinchcliffe. 1978. “Women and Law Reform in Contemporary Islam”, in Womenin the Muslim World, edited by L. Beck and N. Keddie, Cambridge: Harvard University Press.

[19] Das Gupta, M. 2005. “Explaining Asia’s ‘Missing Women: A Look at the Data”, Population andDevelopment Review 31(3): 529-535.

[20] Deolalikar, A. B. and V. Rao. 1998. “The Demand for Dowries and Bride Characteristics in Mar-riage: Empirical Estimates for Rural South-Central India.” in Gender, Population and Develop-ment, edited by Krishnaraj, M., Sudarshan, R. and Shariff, A. Delhi: Oxford University Press.

[21] Deshpande, C. R. 1978. Transmission of the Mahabharata Tradition. Vyasa and Vyasids. Simla:Institute of Advanced Study.

[22] Dharmalingam, A. and S. P. Morgan. 2004. “Muslim-Hindu Fertility Differences in India.” Demog-raphy 41(3): 529-545.

[23] Dreze J. and M. Murthi. 2001. “Fertility, Education and Development: Evidence from India.”Population and Development Review 27(1): 33-63.

13

[24] Dreze, J. and A.K. Sen. 1996. Economic Development and Social Opportunity. New Delhi: OxfordUniversity Press.

[25] Dyson, T. and M. Moore. 1983. “On Kinship Structure, Female Autonomy, and DemographicBehaviour in India.” Population and Development Review 9(1): 35-60.

[26] Edlund, L. E. 1999. “Son Preference, Sex Ratios and Marriage Preference.” Journal of PoliticalEconomy 107: 1275-1304.

[27] Filmer, D. and L. Pritchett. 1999. “The Effect of Household Wealth on Educational Attainment:Evidence From 35 Countries.” Population and Development Review 25(1):85-120

[28] Goldschielder, C. and P. Uhlenberg. 1969. “Minority Group Status and Fertility”, American Journalof Sociology, 74 (January): 261-272.

[29] Government of India. 2006. Social, Economic and Educational Status of the Muslim Communityof India, Prime Minister’s High Level Committee, Cabinet Secretariat, Government of India.

[30] International Institute of Population Sciences, Mumbai and ORC Macro International. 2000. India1998-99 National Family Health Survey (NFHS-2). IIPS, Mumbai.

[31] International Institute of Population Sciences, Mumbai and ORC Macro International. 2007. India2005-06 National Family Health Survey (NFHS-3). IIPS, Mumbai.

[32] Iyer S. 2002. Demography and Religion in India. Oxford University Press: Delhi.

[33] Jacoby, H. G. and G. Mansuri. 2005. ’Watta Satta: Exchange Marriage and Women’s Welfare inRural Pakistan’, mimeograph, The World Bank.

[34] Jeffery R. and P. Jeffery. 1997. Population, Gender and Politics: Demographic Change in RuralNorth India. Cambridge University Press: Cambridge.

[35] Jejeebhoy, S and Z. Sathar. 2001. "Women’s autonomy in India and Pakistan: The influence ofreligion and region." Population and Development Review 27(4): 687–712.

[36] Jensen, R. 2003. “Equal Treatment, Unequal Outcomes: Generating Sex-Inequality Through Fer-tility Behavior.” mimeograph, Watson Institute.

[37] Kishor, S. 1993. “May God Give Sons to All: Gender and Child Mortality in India.” AmericanSociological Review 58 (2): 247-265.

[38] Klasen, S. 1994. "Missing Women Reconsidered." World Development 22: 1061-1071.

[39] Landes, D. S. 1998. The Wealth and Poverty of Nations. London: Little, Brown and Company.

[40] Manski, C., and J. Mayshar. 2002. “Private and social incentives for fertility: Israeli puzzles.”NBER Working Paper 8984.

[41] Minault, G. 1998. Secluded Scholars. Delhi: Oxford University Press.

14

[42] Moulasha, K and G. R. Rao. 1999. “Religion-Specific Differentials in Fertility and Family Plan-ning.” Economic and Political Weekly 34 (42): 3047-51.

[43] Murthi, M., A.C. Guio, and J. Dreze. 1995. “Mortality, Fertility, and Gender Bias in India: ADistrict-Level Analysis.” Population and Development Review 21(4): 745-82.

[44] Niraula, B. B. and S. P. Morgan. 1996. “Son and Daughter Preferences in Benighat, Nepal: Im-plications for Fertility Transition.” Social Biology 42: 256-73.

[45] Obermeyer C. M. 1992. “Islam, Women and Politics: The Demography of Arab Countries.” Pop-ulation and Development Review 18(1): 33-60.

[46] Oster, E. 2005. “Hepatitis B and the Case of the Missing Women.” Journal of Political Economy113(6): 1163-1216.

[47] Qureshi, S. 1980. “Islam and Development: The Zia Regime in Pakistan.” World Development 8:563-575.

[48] Radhakrishnan, S. 1923. Indian Philosophy. Volume 1. Reprinted in 1997 by Oxford UniversityPress, USA.

[49] Radhakrishnan, S. 1947. Religion and Society. London: George Allen and Unwin.

[50] Rao V. 1993. “The Rising Price of Husbands: A Hedonic Analysis of Dowry Increases in RuralIndia.” Journal of Political Economy 101: 666-677.

[51] Rustein, S. 1999. “Wealth versus expenditure: Comparison between the DHS wealth index andhousehold expenditures in four departments of Guatemala” Calverton, Maryland: ORC Macro.

[52] Shamasastry, R 1951. Translation into English of Kautilya’s Arthasastra. Mysore: Sri RaghuveerPrinting Press.

[53] Sen, B. 1998. “Why Does Dowry Still Persist in India? An Economic Analysis Using HumanCapital.” in South Asians and the Dowry Problem, ed by Menski, W., Trentham Books; London:University of London, School of Oriental and African Studies.

[54] Sen, A. K. 1992. “Missing Women.” British Medical Journal 304 (March): 586-587.

[55] Sen, A.K. 2003. “Missing Women - Revisited.” British Medical Journal 327 (December): 1297-1298.

[56] Shariff, A. 1999. India Human Development Report, New Delhi: Oxford University Press.

[57] Strauss, J. and D. Thomas. 1995. “Human resources: Empirical modeling of household and familydecisions.” In Handbook of Development Economics, edition 1, volume 3, chapter 34, edited byHollis Chenery and T.N. Srinivasan. Elsevier.

[58] Suryanarayana Murthy, C. 2000. Sri Lalita Sahasranama with Introduction and Commentary.Mumbai: Bharatiya Vidya Bhavan Publications.

[59] United Nations. 1961. Mysore Population Study. New York: Department of Social and EconomicAffairs.

15

[60] Visaria, P. 1971. The Sex Ratio of the Population of India, Census of India, Monograph No. 10(New Delhi: Manager of Publications).

[61] Yerushalmy, J. 1938. “The Effect of Order of Birth and Age of Parents Upon Neonatal Mortality.”American Sociological Review 3(6): 868-872.

[62] Youssef, N. H. 1978. “The Status and Fertility Patterns of Muslim Women” in Women in theMuslim World, edited by L. Beck and N. Keddie, Cambridge: Harvard University Press.

16

Table 1: Summary statistics for variables used in regressionsVariable Mean Std. Dev. Mean Std. Dev. Mean Std. Dev.

Panel (A): Female Sample All women Childless women excl. Age ≥ 40TotalChildren 2.738 1.926 3.034 1.789 3.867 2.100Dalit 0.275 0.446 0.286 0.452 0.258 0.438Non-Dalit 0.528 0.499 0.564 0.496 0.593 0.491Woman’s age 32.307 8.48 32.883 8.068 43.752 2.818Woman’s age at marriage 17.956 3.925 93724 17.828 3.801 18.253Woman never attended school 0.394 0.489 0.401 0.49 0.458 0.498Husband’s age 37.643 9.493 87623 38.58 9.142 49.583Husband never attended school 0.222 0.416 0.221 0.415 0.236 0.425Land (in acres) 4.510 5.511 4.706 5.811 4.521 5.404Wealth index 0.001 0.099 0.002 0.1 0.022 0.103Rural 0.533 0.499 0.562 0.496 0.523 0.499

Panel (B): Child SampleDalit 0.305 0.460NonDalit 0.524 0.499Female 0.479 0.500Firstborn 0.331 0.470Dalit X Female 0.147 0.354NonDalit X Female 0.249 0.433Firstborn X Female 0.141 0.348Firstborn X Dalit 0.092 0.289Firstborn X NonDalit 0.190 0.392Firstborn X Female X Dalit 0.041 0.198Firstborn X Female X NonDalit 0.078 0.268Mother’s age 34.931 7.723Mother’s age at marriage 17.08 3.469Mother never attended school 0.514 0.5Father’s age 40.646 8.856Father never attended school 0.281 0.449Land (acres) 3.415 17.733Wealth index -0.016 0.097Rural 0.618 0.486Ate meat 24 hours before the survey 0.026 0.16Ate plant protein 24 hours before the survey 0.03 0.171Ate dairy 24 hours before the survey 0.019 0.136Hemoglobin level 116.099 17.669Average measles vaccinations 0.547 0.262Average BCG vaccinations 0.795 0.234Average polio vaccinations 0.885 0.152Average DPT vaccinations 0.761 0.245Average female LFP 0.376 0.237

Table 1: Summary statistics for variables used in the regressions.

17

Tab

le2:

Differen

cesin

dem

ographic

outcom

esforHindus,

Muslim

san

dDalits

Non

-DalitHindus

Muslim

sDifferen

ceDalitHindus

Muslim

sDifferen

ce(s

td.

err.

)(s

td.

err.

)

FemaleSam

ple:

Fraction

ofever-bornbo

yswho

died

0.036

0.039

-0.0026∗

0.047

0.039

0.0082

∗∗∗

(-0.

0014

)(0

.001

6)

Fraction

ofever-borngirlswho

died

0.032

0.033

-0.0012

0.041

0.033

0.0080

∗∗∗

(-0.

0012

)(0

.001

4)

Sex-ratioof

child

renalive

980.8

1053.4

72.56∗

∗∗1033.8

1053.4

19.59

(20.

12)

(24.

92)

Sex-ratioof

child

renever-born

972.7

1029.8

57.05∗

∗∗1007.6

1029.8

22.20

(18.

72)

(22.

28)

Num

berof

child

renbo

rn2.9

3.65

-0.76∗

∗∗3.34

3.65

0.31

∗∗∗

(-0.

02)

(0.0

25)

Sterilized

0.44

0.26

0.18

∗∗∗

0.39

0.26

0.13

∗∗∗

(-0.

0058

)(0

.006

0)

Femalecontraceptiveuse

0.62

0.45

0.17

∗∗∗

0.52

0.45

0.062∗

∗∗

(-0.

0058

)(0

.006

3)

Idealn

umbe

rof

girls

1.02

1.22

-0.20∗

∗∗1.23

1.22

0.014∗

(-0.

0054

)(-

0.00

77)

Idealn

umbe

rof

boys

1.31

1.6

-0.28∗

∗∗1.56

1.6

-0.035

∗∗∗

(-0.

0067

)(-

0.00

88)

Ideals

ex-ratio

0.85

0.84

0.0095

∗∗0.86

0.84

0.017∗

∗∗

(-0.

0041

)(0

.004

7)

Age

atfirst

marriage

18.2

17.1

1.12

∗∗∗

17.3

17.1

0.25

∗∗∗

(-0.

04)

(-0.

041)

ChildSam

ple:

Died

0.091

0.093

-0.002

0.12

0.093

0.023∗

∗∗

(0.0

01)

(0.0

019)

Female

0.48

0.49

-0.009

∗∗∗

0.48

0.49

-0.0007

(0.003)

(0.003)

Age

atdeath

19.7

17.6

2.17

∗∗21.1

17.6

3.54

∗∗

(0.8

7)(0

.85)

Tab

le2:

Notes:(i)Individu

alda

taareba

sedon

atotalsampleof

5121

7Hindu

wom

enan

d92

69Muslim

wom

enwho

aremarried

andha

veha

dat

leaston

emalebirthpriorto

thesurvey;(ii)

The

sex-ratiois

defin

edpe

rclusteras

the(num

berof

females)/(num

berof

males)×

1000

andis

calculated

asan

averag

eover

clusters;(

iii)Stan

dard

errors

inpa

renthe

ses:∗p<

0.05,∗∗

p<

0.01

,∗∗∗

p<

0.001

.

18

Table 3A: Probit Regressions (child sample)

Died before age 1Mother’s age ≥ 40

(1) (2) (3) (4) (5)Dalit 0.0182∗∗∗ 0.0106∗∗∗ 0.0107∗∗∗ 0.00991∗∗∗ 0.0139∗∗∗

(0.0157) (0.0159) (0.0160) (0.0162) (0.0293)

NonDalit 0.00411∗∗ 0.00828∗∗∗ 0.00858∗∗∗ 0.00803∗∗∗ 0.0101∗∗∗

(0.0145) (0.0149) (0.0151) (0.0152) (0.0272)

Female -0.0110∗∗∗ -0.0107∗∗∗ -0.0107∗∗∗ -0.0107∗∗∗ -0.0111∗∗∗

(0.00856) (0.00861) (0.00862) (0.00862) (0.0145)

Firstborn -0.0561∗∗∗ -0.0501∗∗∗ -0.0499∗∗∗ -0.0499∗∗∗ -0.0506∗∗∗

(0.0108) (0.0110) (0.0110) (0.0110) (0.0194)

Mother’s age 0.000673∗∗∗ 0.000649∗∗∗ 0.000642∗∗∗ 0.00139∗∗∗

(0.00132) (0.00133) (0.00133) (0.00338)

Mother’s age at marriage -0.00213∗∗∗ -0.00208∗∗∗ -0.00208∗∗∗ -0.00221∗∗∗

(0.00175) (0.00176) (0.00176) (0.00282)

Mother never attended school 0.0131∗∗∗ 0.0128∗∗∗ 0.0126∗∗∗ 0.0151∗∗∗

(0.0127) (0.0127) (0.0127) (0.0228)

Father’s age -0.000179 -0.000191 -0.000181 -0.0000890(0.00114) (0.00114) (0.00114) (0.00183)

Father never attended school 0.00103 0.000721 0.000695 -0.00184(0.0118) (0.0119) (0.0119) (0.0207)

Land (acres) 0.000163 0.000143 0.0000701 -0.0586∗∗

(0.0854) (0.0850) (0.0853) (0.180)

Wealth index -0.150∗∗∗ -0.140∗∗∗ -0.137∗∗∗ -0.182∗∗∗

(0.0817) (0.0828) (0.0831) (0.143)

Rural -0.000831 -0.000816 -0.00182 -0.00287(0.0131) (0.0131) (0.0134) (0.0238)

AteMeat -0.0135∗∗∗ -0.0135∗∗∗ 0.00289(0.0359) (0.0359) (0.135)

AtePlantProtein -0.00573 -0.00569 0.0175(0.0302) (0.0303) (0.111)

AteDairy -0.00624 -0.00637 -0.0419∗∗

(0.0388) (0.0387) (0.186)

Hemelevel -0.0235∗∗∗ -0.0237∗∗∗ -0.00864(0.0280) (0.0280) (0.0490)

Average measles vaccinations -0.00330 -0.00312 0.00104(0.0325) (0.0325) (0.0549)

Average polio vaccinations 0.0125∗∗ 0.0125∗∗ 0.00100(0.0429) (0.0429) (0.0761)

Average DPT vaccinations -0.0105∗∗ -0.0106∗∗ -0.000414(0.0367) (0.0367) (0.0651)

Average female LFP 0.00829∗∗∗ 0.00856(0.0245) (0.0419)

N 204841 204841 204841 204841 69151Pseudo R-squared 0.0369 0.0475 0.0484 0.0485 0.0478Chi-squared 3270.7 4126.7 4228.5 4229.5 1448.2State fixed-effects Yes Yes Yes Yes Yes

Table 3: Marginal effects of probit model. Note: (1) ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

19

Table 3B: Probit Regressions (child sample)

Died before age 5Mother’s age ≥ 40

(1) (2) (3) (4) (5)Dalit 0.0301∗∗∗ 0.0165∗∗∗ 0.0176∗∗∗ 0.0166∗∗∗ 0.0194∗∗∗

(0.0152) (0.0154) (0.0155) (0.0156) (0.0267)

NonDalit -0.00129 0.00667∗∗∗ 0.00838∗∗∗ 0.00768∗∗∗ 0.0109∗∗

(0.0141) (0.0145) (0.0146) (0.0147) (0.0250)

Female -0.00726∗∗∗ -0.00749∗∗∗ -0.00760∗∗∗ -0.00760∗∗∗ -0.00600∗∗∗

(0.00832) (0.00838) (0.00839) (0.00839) (0.0136)

Firstborn -0.0537∗∗∗ -0.0451∗∗∗ -0.0443∗∗∗ -0.0443∗∗∗ -0.0601∗∗∗

(0.0109) (0.0111) (0.0111) (0.0112) (0.0182)

Mother’s age -0.000607∗∗ -0.000504∗∗ -0.000513∗∗ 0.00272∗∗∗

(0.00134) (0.00134) (0.00134) (0.00316)

Mother’s age at marriage -0.00182∗∗∗ -0.00186∗∗∗ -0.00186∗∗∗ -0.00271∗∗∗

(0.00173) (0.00173) (0.00173) (0.00264)

Mother never attended school 0.0199∗∗∗ 0.0192∗∗∗ 0.0189∗∗∗ 0.0192∗∗∗

(0.0122) (0.0122) (0.0123) (0.0208)

Father’s age -0.000288 -0.000291 -0.000279 -0.000378(0.00110) (0.00110) (0.00111) (0.00170)

Father never attended school 0.00484∗∗ 0.00428∗∗ 0.00425∗∗ 0.00241(0.0114) (0.0115) (0.0115) (0.0191)

Land (acres) 0.00715 0.00667 0.00657 -0.0268(0.0873) (0.0870) (0.0871) (0.145)

Wealth index -0.280∗∗∗ -0.260∗∗∗ -0.256∗∗∗ -0.306∗∗∗

(0.0796) (0.0806) (0.0809) (0.132)

Rural -0.00130 -0.00134 -0.00261 -0.00697∗

(0.0126) (0.0126) (0.0129) (0.0218)

AteMeat -0.00274 -0.00280 0.00500(0.0394) (0.0394) (0.133)

AtePlantProtein 0.0194∗∗∗ 0.0195∗∗∗ 0.0152(0.0337) (0.0337) (0.109)

AteDairy 0.0146∗ 0.0144∗ 0.00333(0.0431) (0.0430) (0.147)

Hemelevel -0.0391∗∗∗ -0.0395∗∗∗ -0.0119(0.0269) (0.0269) (0.0448)

Average measles vaccinations -0.0131∗∗ -0.0129∗∗ -0.00521(0.0311) (0.0311) (0.0503)

Average polio vaccinations 0.0221∗∗∗ 0.0220∗∗∗ 0.00985(0.0416) (0.0416) (0.0719)

Average DPT vaccinations -0.0179∗∗∗ -0.0179∗∗∗ -0.00227(0.0354) (0.0354) (0.0605)

Average female LFP 0.0105∗∗ 0.00845(0.0239) (0.0393)

N 172416 172416 172416 172416 68284Pseudo R-squared 0.0318 0.0478 0.0491 0.0491 0.0601Chi-squared 3018.0 4625.8 4754.0 4755.5 2091.6State fixed-effects Yes Yes Yes Yes Yes

Table 3: Marginal effects of probit model. Note: (1) ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

20

Table 4: Fertility Regressions, Dependent Variable: Number of children born

(1) (2) (3) (4)Dalit 0.118 -0.0187 -0.163 -0.248∗

(0.145) (0.152) (0.130) (0.133)

Hindu Non-Dalit -0.813∗∗∗ -0.770∗∗∗ -0.570∗∗∗ -0.610∗∗∗

(0.156) (0.154) (0.121) (0.127)

Age 0.0561∗∗∗ 0.0755∗∗∗

(0.00420) (0.00485)

Age at marriage -0.157∗∗∗ -0.118∗∗∗

(0.00838) (0.00475)

Never attended school 0.289∗∗∗ 0.257∗∗∗

(0.0452) (0.0461)

Completed primary school -0.646∗∗∗ -0.352∗∗∗

(0.0588) (0.0513)

Husband’s age 0.0163∗∗∗ -0.0166∗∗∗

(0.00337) (0.00340)

Husband never attended school 0.167∗∗ 0.168∗∗

(0.0643) (0.0620)

Land (in acres) 0.000332 -0.000180(0.00391) (0.00390)

Wealth index -3.312∗∗∗ -3.356∗∗∗

(0.374) (0.339)

Rural population 0.237∗∗∗ 0.169∗∗

(0.0646) (0.0621)

Constant 4.214∗∗∗ 4.664∗∗∗ 3.303∗∗∗ 3.790∗∗∗

(0.0557) (0.252) (0.183) (0.172)

Observations 19917 19917 19917 19917Adjusted R2 0.173 0.272 0.295 0.344State Fixed-Effects Yes Yes Yes Yes

Table 4: Fertility Regressions. Note: ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001.

21

Tab

le5:

Probit

Regressionswithgender

andbirth-order

effects

(childsample)

Diedbeforeage1

Diedbeforeage5

Mom

age≥

40Mom

age≥

40(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Dalit

0.0125

∗∗∗

0.0116

∗∗∗

0.0145

∗∗∗

0.0192

∗∗∗

0.0199

∗∗∗

0.0198

∗∗∗

0.0239

∗∗∗

0.0271

∗∗∗

(0.0

202)

(0.0

174)

(0.0

221)

(0.0

391)

(0.0

198)

(0.0

167)

(0.0

213)

(0.0

361)

Non

Dalit

0.00801∗

∗∗0.0110

∗∗∗

0.0128

∗∗∗

0.0150

∗∗∗

0.00760∗

∗0.0137

∗∗∗

0.0163

∗∗∗

0.0181

∗∗∗

(0.0

190)

(0.0

163)

(0.0

208)

(0.0

367)

(0.0

186)

(0.0

157)

(0.0

202)

(0.0

340)

Female

-0.00893

∗∗∗

-0.0108∗

∗∗-0.0112∗

∗∗-0.0120∗

∗-0.00538

-0.00783

∗∗∗

-0.00842

∗∗-0.00461

(0.0

209)

(0.0

0862

)(0

.023

2)(0

.039

7)(0

.020

6)(0

.008

39)

(0.0

225)

(0.0

367)

Firstbo

rn-0.0499∗

∗∗-0.0397∗

∗∗-0.0452∗

∗∗-0.0403∗

∗∗-0.0442∗

∗∗-0.0181∗

∗∗-0.0282∗

∗∗-0.0492∗

∗∗

(0.0

110)

(0.0

276)

(0.0

384)

(0.0

699)

(0.0

112)

(0.0

285)

(0.0

400)

(0.0

676)

DalitX

Female

-0.00526

-0.00574

-0.00522

-0.00666

-0.00809

∗-0.00916

(0.0

257)

(0.0

284)

(0.0

483)

(0.0

251)

(0.0

273)

(0.0

445)

Non

DalitX

Female

0.0000117

-0.00384

-0.00294

0.000153

-0.00540

-0.00601

(0.0

242)

(0.0

269)

(0.0

454)

(0.0

238)

(0.0

261)

(0.0

423)

DalitX

Firstbo

rn-0.00975

∗∗-0.0103∗

-0.0132

-0.0221∗

∗∗-0.0244∗

∗∗-0.0158

(0.0

336)

(0.0

461)

(0.0

833)

(0.0

343)

(0.0

476)

(0.0

800)

Non

DalitX

Firstbo

rn-0.0164∗

∗∗-0.0215∗

∗∗-0.0319∗

∗∗-0.0366∗

∗∗-0.0425∗

∗∗-0.0352∗

∗∗

(0.0

313)

(0.0

433)

(0.0

784)

(0.0

320)

(0.0

447)

(0.0

754)

Firstbo

rnX

Dalit

XFe

male

0.00191

-0.0106

0.00769

-0.00905

(0.0

681)

(0.1

24)

(0.0

695)

(0.1

16)

Firstbo

rnX

Non

Dalit

XFe

male

0.0166

∗0.0295

0.0219

∗0.0201

(0.0

638)

(0.1

15)

(0.0

651)

(0.1

08)

Firstbo

rnX

Female

0.0160

∗∗0.0157

0.0256

∗∗0.0238

(0.0

563)

(0.1

02)

(0.0

577)

(0.0

961)

N204841

204841

204841

69151

172416

172416

172416

68284

Chi-squ

ared

4238.2

4245.3

4366.9

1490.3

4761.9

4772.0

4869.5

2120.5

lincom(F

BXDal,F

BXDalXFe

m)

-0.0694

-0.179

-0.108

-0.153

0.0495

0.0912

0.0501

0.0834

lincom(F

BXNon

Dal,F

BXNon

DalXFe

m)

-0.0630

-0.0708

-0.157

-0.118

0.0461

0.0824

0.0467

0.0764

Con

trols

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Statefix

ed-effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Tab

le5:

Margina

leffe

ctsof

prob

itmod

el.Note:∗p<

0.05,∗∗

p<

0.01,∗∗∗p<

0.001.

22

Tab

leA:Differen

cesin

socioeconom

icou

tcom

esof

Hindus,

Muslim

san

dDalits

Non

DalitHindus

Muslim

sDifferen

ceDalitHindus

Muslim

sDifferen

ce(std.err)

(std.err)

Femalecompletionof

prim

aryscho

ol0.52

0.34

0.18

∗∗∗

0.32

0.34

0.19

∗∗∗

(0.0

052)

(0.0

038)

Femalecompletionof

second

aryscho

ol0.12

0.032

0.084∗

∗∗0.030

0.032

0.086∗

∗∗

(0.0

032)

(0.0

022)

Malecompletionof

prim

aryscho

ol0.69

0.48

0.21

∗∗∗

0.50

0.48

0.20

∗∗∗

(0.0

049)

(0.0

037)

Malecompletionof

second

aryscho

ol0.21

0.085

0.12

∗∗∗

0.093

0.085

0.12

∗∗∗

(0.0

041)

(0.0

029)

Land

(acres)

5.32

3.15

2.17

∗∗∗

4.50

3.15

0.62

(0.6

01)

(0.4

5)

Hou

seho

ldbe

low

povertylin

e0.53

0.57

-0.045

∗∗∗

0.57

0.57

-0.046

∗∗∗

(0.0

16)

(0.0

12)

Wom

encurrentlyworking

0.35

0.24

0.11

∗∗∗

0.46

0.24

0.22

∗∗∗

(0.0

049)

(0.0

055)

Wom

enself-em

ployed

0.072

0.04

0.032∗

∗∗0.080

0.04

0.040∗

∗∗

(0.0

026)

(0.0

029)

Wom

enworking

away

from

home

0.31

0.16

0.15

∗∗∗

0.46

0.16

0.30

∗∗∗

(0.0

047)

(0.0

053)

Tab

leA:Notes:(i)Individu

alda

taareba

sedon

atotalsampleof

5121

7Hindu

wom

enan

d92

69Muslim

wom

enwho

aremarried

andha

veha

dat

leaston

emalebirthpriorto

thesurvey;(ii)

Stan

dard

errors

inpa

renthe

ses:∗p<

0.05,∗∗

p<

0.01,∗∗∗p<

0.001.

23

Tab

leB:Differen

cesin

nutritionan

dhealthindicatorsof

Hindus,

Muslim

san

dDalits

Non

DalitHindus

Muslim

sDifferen

ceDalitHindus

Muslim

sDifferen

ce(s

td.

err)

(std

.er

r)

Eat

Meat

0.026

0.055

-0.029

∗∗∗

0.036

0.055

-0.019

∗∗∗

(-0.

0019

)(0

.002

4)

Eat

Plant

protein

0.037

0.035

0.0024

0.037

0.035

0.00023

(-0.

0021

)(-

0.00

23)

Eat

Dairy

0.019

0.04

-0.022

∗∗∗

0.026

0.04

-0.014

∗∗∗

(-0.

0016

)(0

.002

1)

Wom

en’s

hemoglobinlevel

117

116.2

0.89

∗∗∗

114.9

116.2

-1.26∗

∗∗

(-0.

19)

(0.2

2)

Vaccina

tedagainstmeasles

(cluster

average)

0.62

0.46

0.16

∗∗∗

0.50

0.46

0.044∗

∗∗

(-0.

0063

)(0

.006

8)

Vaccina

tedagainstBCG

(cluster

average)

0.83

0.68

0.16

∗∗∗

0.72

0.68

0.049∗

∗∗

(-0.

0051

)(0

.006

2)

Vaccina

tedagainstpo

lio(cluster

average)

0.92

0.86

0.059∗

∗∗0.85

0.86

-0.0065∗

∗∗

(-0.

0038

)(0

.004

8)

Vaccina

tedagainstDPT

(cluster

average)

0.81

0.63

0.17

∗∗∗

0.69

0.63

0.060∗

∗∗

(-0.

0054

)(0

.006

4)

Tab

leB:N

otes:(i)Cluster

levela

veragesrefle

ctaverag

esof

Muslim

s,Non

-Dalitsan

dDalitsin

3850

clusters;(

ii)Stan

dard

errors

inpa

renthe

ses:

∗p<

0.05,∗∗

p<

0.01,∗∗∗p<

0.001

Tab

leC:Differen

cesin

femalelabor

forceparticipationratesforHindus,

Muslim

san

dDalits

Non

DalitHindus

Muslim

sDifferen

ceDalitHindus

Muslim

sDifferen

ce(s

td.

err)

(std

.er

r)

Wom

encurrentlyworking

(cluster

average)

0.35

0.24

0.11

∗∗∗

0.46

0.24

0.22

∗∗∗

(0.0

049)

(0.0

05)

Wom

enself-em

ployed

(cluster

average)

0.072

0.04

0.032∗

∗∗0.080

0.04

0.040∗

∗∗

(0.0

026)

(0.0

029)

Wom

enworking

away

from

home(cluster

average)

0.31

0.16

0.15

∗∗∗

0.46

0.16

0.30

∗∗∗

(0.0

047)

(0.0

051)

Tab

leC:N

otes:(i)Cluster

levela

veragesrefle

ctaverag

esof

Muslim

s,Non

-Dalitsan

dDalitsin

3850

clusters;(

ii)Stan

dard

errors

inpa

renthe

ses:

∗p<

0.05,∗∗

p<

0.01,∗∗∗p<

0.001

24