14
Journal of Health Economics 29 (2010) 191–204 Contents lists available at ScienceDirect Journal of Health Economics journal homepage: www.elsevier.com/locate/econbase The puzzle of Muslim advantage in child survival in India Sonia Bhalotra a , Christine Valente b,, Arthur van Soest c a University of Bristol, UK b University of Nottingham, School of Economics, University Park, Nottingham NG7 2RD, UK c Tilburg University, The Netherlands article info Article history: Received 11 February 2009 Received in revised form 15 October 2009 Accepted 4 November 2009 Available online 10 November 2009 JEL classification: O12, I12, J15, J16, J18 Keywords: Religion Caste Gender Child survival India abstract The socioeconomic status of Indian Muslims is, on average, considerably lower than that of upper-caste Hindus. Muslims nevertheless exhibit substantially higher child survival rates, and have done for decades. This paper analyses this seeming puzzle. A decomposition of the survival differential confirms that some compositional effects favour Muslims but that, overall, differences in characteristics and especially the Muslim deficit in parental education predict a Muslim disadvantage. The results of this study contribute to a recent literature that debates the importance of socioeconomic status (SES) in determining health and survival. They augment a growing literature on the role of religion or culture as encapsulating important unobservable behaviours or endowments that influence health, indeed, enough to reverse the SES gradient that is commonly observed. © 2009 Elsevier B.V. All rights reserved. 1. Introduction Hindus and Muslims have cohabited in India for centuries, with Muslims ruling most of the Indian subcontinent from the early 16th to the mid-19th centuries. However, today, their socioeconomic condition is thought to be not much better than that of low-caste Hindus, who have a long history of deprivation (Government of India, 2006). Despite being, on average, less educated and poorer, Indian Muslims exhibit a substantial advantage in child survival over high-caste Hindus. This paper analyses this seeming puzzle. It shows that the Muslim advantage is large, persistent, and hard to explain. A number of recent studies document socioeconomic status (SES) gradients in health and survival, across countries, across SES groups within country, and within groups over time; for a survey see Cutler et al. (2006). Previous analyses of health inequalities along ethnic or religious lines tend to start out with a differen- tial consistent with SES differences, as is the case, for example, with black–white differences in health in the United States. While Corresponding author. Tel.: +44 (0)115 846 6393. E-mail addresses: [email protected] (S. Bhalotra), [email protected], [email protected] (C. Valente), [email protected] (A. van Soest). it has been recognised that unhealthy behaviours like smoking or drinking may vary positively with SES (e.g. Rogers et al., 2000, p. 245), these are seldom large enough to alter the raw differential in favour of the lower-SES group. The case of Muslims in India is, in this respect, most unusual. By age five, the Muslim survival advantage over Hindus is as high as 2.31%-points, which is about 17% of baseline mortality risk amongst Hindus. Restricting the comparison to upper-caste Hin- dus, who enjoy unambiguously higher social status than Muslims, the differential is 1.30%-points, or about 10% of baseline mortality risk. Based on the total number of births recorded in 2000 (Census of India, 2001), and on the proportions of high- and low-caste chil- dren born that year (obtained from representative survey data used in this paper), this differential translates into an annual 127,955 (244,535) excess under-5 deaths amongst high-caste (low-caste) Hindus. To put the size of this differential in perspective, con- sider that the more widely discussed gender differential in under-5 mortality is 0.30%-points, which is less than a fourth of the dif- ferential between Muslims and high-caste Hindus. The average annual rate of decrease in under-5 mortality risk between 1960 and 2001 in India was 0.61%-points p.a., which is about half the differential. The Muslim–Hindu survival differential is not a new or an isolated phenomenon. It is evident for most of the last half century and across most of India. It has nevertheless claimed lit- tle public or academic attention. Although it is flagged by Shariff 0167-6296/$ – see front matter © 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.jhealeco.2009.11.002

The puzzle of Muslim advantage in child survival in India

Embed Size (px)

Citation preview

  • Journal of Health Economics 29 (2010) 191204

    Contents lists available at ScienceDirect

    Journal of Health Economics

    journa l homepage: www.e lsev ier .com/

    The pu val

    Sonia Bha University ofb University ofc Tilburg Unive

    a r t i c l

    Article history:Received 11 FeReceived in reAccepted 4 NoAvailable onlin

    JEL classicatioO12, I12, J15, J

    Keywords:ReligionCasteGenderChild survivalIndia

    uslimubstle. A dbut tredicport

    aturents th

    1. Introdu

    Hindus aMuslims ruto the mid-condition isHindus, whIndia, 2006Indian Musover high-cshows thatexplain.

    A numb(SES) gradiegroups withsee Cutleralong ethnitial consistwith black

    CorresponE-mail add

    c.valente@[email protected] (A

    0167-6296/$ doi:10.1016/j.ction

    nd Muslims have cohabited in India for centuries, withlingmost of the Indian subcontinent from the early 16th19th centuries. However, today, their socioeconomicthought to be not much better than that of low-casteo have a long history of deprivation (Government of). Despite being, on average, less educated and poorer,lims exhibit a substantial advantage in child survivalaste Hindus. This paper analyses this seeming puzzle. Itthe Muslim advantage is large, persistent, and hard to

    er of recent studies document socioeconomic statusnts in health and survival, across countries, across SESin country, and within groups over time; for a survey

    et al. (2006). Previous analyses of health inequalitiesc or religious lines tend to start out with a differen-ent with SES differences, as is the case, for example,white differences in health in the United States. While

    ding author. Tel.: +44 (0)115 846 6393.resses: [email protected] (S. Bhalotra),tingham.ac.uk, [email protected] (C. Valente),. van Soest).

    it has been recognised that unhealthy behaviours like smoking ordrinking may vary positively with SES (e.g. Rogers et al., 2000, p.245), these are seldom large enough to alter the raw differential infavour of the lower-SES group. The case of Muslims in India is, inthis respect, most unusual.

    By age ve, the Muslim survival advantage over Hindus is ashigh as 2.31%-points, which is about 17% of baseline mortality riskamongst Hindus. Restricting the comparison to upper-caste Hin-dus, who enjoy unambiguously higher social status than Muslims,the differential is 1.30%-points, or about 10% of baseline mortalityrisk. Based on the total number of births recorded in 2000 (Censusof India, 2001), and on the proportions of high- and low-caste chil-dren born that year (obtained from representative surveydata usedin this paper), this differential translates into an annual 127,955(244,535) excess under-5 deaths amongst high-caste (low-caste)Hindus. To put the size of this differential in perspective, con-sider that themorewidely discussed gender differential in under-5mortality is 0.30%-points, which is less than a fourth of the dif-ferential between Muslims and high-caste Hindus. The averageannual rate of decrease in under-5 mortality risk between 1960and 2001 in India was 0.61%-points p.a., which is about half thedifferential. The MuslimHindu survival differential is not a newor an isolated phenomenon. It is evident for most of the last halfcentury and across most of India. It has nevertheless claimed lit-tle public or academic attention. Although it is agged by Shariff

    see front matter 2009 Elsevier B.V. All rights reserved.jhealeco.2009.11.002zzle of Muslim advantage in child survi

    alotraa, Christine Valenteb,, Arthur van Soestc

    Bristol, UKNottingham, School of Economics, University Park, Nottingham NG7 2RD, UKrsity, The Netherlands

    e i n f o

    bruary 2009vised form 15 October 2009vember 2009e 10 November 2009

    n:16, J18

    a b s t r a c t

    The socioeconomic status of Indian MHindus.Muslims nevertheless exhibit sThis paper analyses this seeming puzzcompositional effects favour MuslimsMuslim decit in parental education pto a recent literature that debates the imsurvival. They augment a growing literunobservablebehavioursor endowmethat is commonly observed.locate /econbase

    in India

    s is, on average, considerably lower than that of upper-casteantially higher child survival rates, and have done for decades.ecomposition of the survival differential conrms that somehat, overall, differences in characteristics and especially thet a Muslim disadvantage. The results of this study contributeance of socioeconomic status (SES) in determining health andon the role of religion or culture as encapsulating important

    at inuencehealth, indeed, enough to reverse theSESgradient

    2009 Elsevier B.V. All rights reserved.

  • 192 S. Bhalotra et al. / Journal of Health Economics 29 (2010) 191204

    (1995), Bhat and Zavier (2005), Bhalotra and van Soest (2008)and Deolalikar (forthcoming), we know of no previous researchthat tries to explain this phenomenon. This paper lls this gap,using micro200,000 Indsummarisedescribed inof this seemused. Baselin Section 6Section 8 co

    A descriimportant itage of Musperiod (theof the diffebehavioursand less wlims exhibi(male/femaThis suggesdate explanalso show svival advantheir advanareas.

    We ndraw data iseffects (whcharacteristics) and stainvestigatelevel incomand infrastrtial. Usingnone of thecaste Hinduidentify so(such as ththan balandus (primathat thereinversely cmunities ddespite theIndian socithis possibthe Muslimexplained bthan half oThis suggeacross SESplained adcommon exesis that adifferential

    Since nunutritionalmunity diffand wastinsmall commdifferencesDecomposiperform lesdus, then

    when they do better, as is the case relative to low-caste Hin-dus, their advantage is mostly unexplained. The latter is againconsistent with Muslims owning unobservable traits that favour

    .se oursy toity raricheal hf delmyantribothetenar effendogst hthe

    gateaveoursresu

    hat say bividut thescuss. AlticityWilsn effeor ex(Gu

    . Theee, fos acrossey arrovarks.

    kgro

    slims9.9

    with2001tra assioneak

    a, 20indued thmenlikaren shik a

    elaticentMushougion, iposdata on more than 0.6 million children born to aboutian women during 19602006. The rest of this sections our approach and ndings. The data and context areSection2. Section3presents therst systematicproleing puzzle. Section 4 presents the estimation methodsine results are discussed in Section 5 and extensions. Section 7 considers differences in nutritional status.ncludes.ptive proling of the religion differential yields somensights. More than two-thirds of the survival advan-lims over high-caste Hindus is apparent in the neonatalrst 30 days after birth), suggesting that explanationsrential may have more to do with customs, attitudes,, maternal health, delivery and early feeding practices,ith access to health and nutrition after birth. Mus-t lower son preference in terms of a lower sex ratiole) at birth and a smaller gender gap in child mortality.ts better survival chances amongst girls as a candi-ation of the overall Muslim advantage, even if theyome advantage for boys. While Muslims exhibit a sur-tage over low-caste Hindus in rural and urban areas,tage over high-caste Hindus is only signicant in rural

    that the Muslim survival advantage apparent in thescarcely diminished by controlling for neighbourhoodich we model as cluster xed effects), socioeconomicics of the household (education, wealth, demograph-te and cohort-specic unobservables. We specicallystate-level health and development expenditure, state-e and inequality, and village-level health servicesucture, but they do not explain the religion differen-decomposition techniques, we show that, in general,Muslim advantage in under-5 mortality over high-s can be explained by characteristics. Although we

    me factors that contribute to a Muslim advantageeir greater urbanisation) these contributions are moreced by characteristics that favour high-caste Hin-rily their better education). In general, it is possibleare omitted variables that improve survival and areorrelated with SESfor example, if high SES com-rink or smoke more they may exhibit poorer healthir higher SES. The layering of religion and caste inety provides us with an opportunity to investigateility. We do this by considering the extent to which

    advantage over low-caste Hindus (of lower SES) isy the same set of characteristics. We nd that lessf their advantage in under-5 mortality is explained.sts that omitted variables that favour Muslims cutgroups. In other words, as Muslims have an unex-vantage over two groups of Hindus with little incept for their religion, we cannot rule out the hypoth-ttitudes and practices related to religion cause the.tritional status tends to be closely tied to SES andstatus predicts mortality, we also investigate com-erentials in two indicators of net nutrition, stuntingg, for children 03 years old (Section 7). We ndunity differentials, especially for wasting. Stuntingcorrespond to SES differences across communities.

    tion of these differences shows that, when Muslimss well, as is the case compared with high-caste Hin-most of their disadvantage is explained. However,

    healthThe

    behavithe kemortalgatingmaternplace oendogadiet coSomeand anas theitially eAmongexplaininvestiship, hbehavi

    Thends tvival mthe indvices athis dimunityin ethn2004;religioysed, fEuropehealthture, sexplainrates ature thdisappnetwo

    2. Bac

    Muup fromparedIndia,(Bhalocommitively wof Indicaste Heducatlim wo(Deolahas be(Bhaumsmall rare conSES ofYet, alteducatsimilarndings motivate consideration of what attitudes,or unobserved traits Muslims might have, unlockingwhich could make an enormous impact on averagetes in India. We take one step in this direction, investi-r specications of the child survival model that includeealth and diet, mothers employment, antenatal care,ivery, early initiation of breastfeeding and indicators forndsonpreference.Maternalheight andnon-vegetarianute signicantly to explaining the Muslim advantage.r variables like mothers employment, breastfeedingtal care also contribute towards an explanation butcts are individually insignicant and they are poten-enous, we do not put much weight on these ndings.ypotheses that we believe may hold the potential topersistent Muslim advantage but that we are unable toto our satisfaction, are that Muslims enjoy closer kin-a lower degree of son preference, and have healthier.lts contribute to a recent literature in economics whichocioeconomic status as a determinant of health or sur-e less important than other factors, such as attitudes atal level (Fuchs, 2004) or medical technology and ser-aggregate level (Cutler et al., 2006). This paper extendsion to incorporate the importance of culture or com-hough there is a surge of interest amongst economistseffects, especially in education (e.g. Fryer and Levitt,

    on et al., 2005), there remains limited research oncts. The effects of religion on fertility have been anal-ample, for India (Bhat and Zavier, 2005) and historicalinnane, 2005), but there is little work on religion andre is some relevant work in the sociological litera-r example, Dwyer et al. (1990) who nd that religionsubstantial share of the variation in cancer mortalityUS counties. In their review of the sociological litera-gue that religion effects appear to work through sociall of unhealthy behaviours and the benets of social

    und and data

    constituted 13.4% of the Indian population in 2001,% in 1951. Their total fertility rate was 3.06, as com-2.47 for Hindu women, a 24% differential (Census of) and they have shorter birth intervals than Hindusnd van Soest, 2008). The Sachar Committee Reported by the Indian Prime Minister documents their rela-social, economic and educational status (Government06, henceforth GOI). Muslims are poorer than upper-s, especially in urban areas (GOI). They have been lessan upper-caste Hindus for decades and while Mus-have exhibited some catch-up, Muslim men have not

    , forthcoming). The educational deprivation of Muslimsown to drive their disadvantage in the labour marketnd Chakrabarty, 2006). Their political representation isve to their population share and the areas in which theyrated receive poorer public services (GOI). Overall, thelims is not much better than that of low-caste Hindus.h there are reserved places for the low castes in highern public sector jobs and in state legislatures, there is noitive discrimination in favour of Muslims. These facts

  • S. Bhalotra et al. / Journal of Health Economics 29 (2010) 191204 193

    all make their relative success in averting child mortality quiteremarkable.1

    To investigate the mortality differential, we stack three roundsof the National Family Health Survey of India (NFHS) conductedin 1992/1993, 1998/1999, and 2005/2006 (see IIPS, 1995; IIPS andORC Macro, 2000; IIPS and Macro International, 2007). These sur-veys interviewed women aged 1549 (1349 in NFHS-1) at thetime of the survey and obtained complete fertility histories, includ-ing the dates of live births and of any child deaths. The surveyscontain information on relevant individual and household char-acteristics, and the rst two rounds also include information onvillage characteristics including health infrastructure. Births in theoriginal sample occur during 19542006. We restrict the sampleto mothers who are normal residents in the dwelling in whichthey are interviewed. We drop children born before 1960 (0.08%of the sample), as the sample sizes are very small for these years.We right-truncate the sample to ensure that all children analysedhave full exposure to the relevant mortality risk; for instance, forunder-5 mortality, we remove children less than 60 months oldat the timehave ever hwhom infoalso drop thholds of relrefer to thethe religionfully exposemothers.

    The largare an advaare a minorbias in datefurtherbackerror is simour results.fertility histWe allow founder-5) topossible issscarce anddren in thesdata becauThis resultsof childrenbe poorer achildren bonature of thbirth.

    1 In this papThere are alsohistory as the

    2 It is standsingletons as dthe statistics. Aand amongst H

    3 Due to sareligions othepopulation. Wof the other recompared to 6

    4 For examprepresentativeexample 1968because olderfrom the samp

    A potential drawback of our data is that variables are measuredat the time of the survey and not at the time of birth of the indexchild. This creates no problems for variables like parental educa-tion which do not change between the index childs birth and thesurvey datebecause thelage infraststatus can clem by condwhenever twere an imalter the sto

    The heignet nutritioare measurreported by35 years brounds comchildren ag

    adesthropd asple wtatistionsionse chafor tt 5 ywe aes liktiveana

    al m

    role

    secty in wral/u

    s a siors offereis.

    ligio

    natag bears,hildMus006, 12.s; se

    f basw-catha

    pointinduindu006.06]%relatof interview (17.43% of births). We drop mothers whoad a multiple birth (3.11% of births)2 and mothers forrmation on caste is missing (0.38% of all births). Wee 11.49% of births in the survey that occur in house-igions other than Muslim or Hindu, and from now onmortality differential between Muslim and Hindus asdifferential.3 The largest sample analysed (for childrend to neonatal risk) has 653,496 live births of 197,952

    e samples generated by using the full history of birthsntage given that mortality is a rare event and Muslimsity group. A potential problem with these data is recalls of birth and death, which is expected to be greater thein time theeventoccurred. To theextent that any recallilar across communities this will not matter much forMoreover, Beckett et al. (2001) nd that recall error inories is not a big problem, except for some age heaping.r age heaping by dening indicators of mortality (e.g.include deaths in the last month (e.g. 60th). A secondue is that the further back one goes in time, the morethe less representative of the complete cohorts of chil-e birth years is the sample of births in the birth-historyse only mothers younger than 50 were interviewed.in the early years including a disproportionate shareborn to young mothers.4 These mothers are likely tond have higher fertility than the average mother ofrn at the same time. We account for this triangulare data structure since we condition on mothers age at

    er, low-caste Hindus refer to scheduled castes and tribes (SC and ST).castes within the Muslim community but they do not have the sameHindu lower castes and do not qualify for positive discrimination.ard practice in the demographic literature to restrict the analysis toeath risks are many times higher for multiple births and can skewmongst Muslims 1.48% of live births are multiple (twin, triplet, etc.)indus the corresponding gure is 1.29%.

    mpling design, the non-weighted share of births in households ofr than Muslim or Hindu is much larger than their share in the totalhen the number of births is corrected for sampling design, the shareligions is closer to gures from the population census: 4.2% of births.1% of the population of all ages in 2001.le, in the 1998 round, births that occur near the survey datewill comely from mothers aged 15 to 49. However, births in earlier years, for, come disproportionately from women who gave birth early. This ismothers, for instance age 25 in 1968,were 55 in 1998 and so excludedle by design.

    hya PrThe anreporte

    Sammary spopulaRegresanalysmonththe lastively,variabldescripsampleneonat

    3. A p

    Thisthe wayear, ruthere iindicatgion dianalys

    3.1. Re

    Neoof dyinve yeple of c14.42%19602HindusMuslim(17% oover logreater1.30%-caste Hcaste HIndia, 21.72 [3points. Migration is not a problem for the religion variablemother carries her religion with her. However, vil-

    ructure changes over time and the familys rural/urbanhange on account ofmigration.Weovercome this prob-ucting the analysis for shorter samples of recent birthsime-varying variables are included as regressors. If itportant source of bias then the shorter samples wouldry (suffering much less from this bias) but they do not.hts and weights of children are indicators of the childsnal status (e.g. Micklewright and Ismail, 2001). Theseed by surveyors at the time of the survey rather thanthe mother. They are only available for children bornefore the survey. To render the samples in the threeparable, we restrict attention to height and weight ofed 03 and exclude the states Andhra Pradesh, Mad-h, Tamil Nadu, West Bengal, and Himachal Pradesh.ometric data are standardized by age and gender andz-scores.eights included in the surveys are used to obtain sum-

    tics that are representative for the all India childrenborn to mothers aged 1549 at the time of the survey.are also weighted using these weights. The sample wenges because, for example, we drop births in the lasthe analysis of neonatal mortality but we drop births inears when we investigate under-5 mortality. Alterna-re restricted to a shorter sample once we incorporatee child height and weight or village infrastructure. Allstatistics are, unless otherwise stated, for the largestlysed, which is the sample of children fully exposed toortality risk in the pooled sample.

    of the religion differential

    ion describes the religion differential in mortality andhich it varies with caste, age, gender, birth order, birthrban and state location. We further consider whether

    milar religion differential in height- and weight-basedf nutritional status. This section also summarises reli-nces in some of the explanatory variables used in the

    n differentials in mortality

    l, infant and under-5 mortality are dened as the risktween birth and the age of one month, one year, andrespectively. After dropping other religions, the sam-ren analysed in this paper contains 84.58% Hindus andlims. Across births in the data, which span the period, averageunder-5mortality is15.93%amongst low-caste59% amongst high-caste Hindus, and 11.29% amongste Table 1. Muslims have an advantage of 2.31% pointseline mortality risk) over all Hindus. Their advantageste Hindus, at 4.64%-points (29.14%) is, unsurprisingly,n their advantage over high-caste Hindus, which iss (10.34%). The latter is the real puzzle since upper-s are clearly better off than Muslims, whereas lowers are, by many indicators, worse off (Government of

    ). For neonatal [infant] mortality, the raw differential is-points relative to low-caste Hindus and 0.90 [1.31]%-ive to high-caste Hindus.

  • 194 S. Bhalotra et al. / Journal of Health Economics 29 (2010) 191204

    Table 1Mortality rates and differentials by community.

    (1) Low caste (%) (2) Differential LC-M (%-points)(as % of (1))

    (3) High caste (%) (4) Differential HC-M (%-points)as % of (3))

    (5) Muslim (%)

    Neonatal 6.79 1.72 5.98 0.90 5.08(25.30) (15.12)

    Infant 11.03 3.06 9.29 1.31 7.98(27.69) (14.08)

    Under-5 15.93 4.64 12.59 1.30 11.29(29.14) (10.34)

    LC is low-caste Hindu (SC and ST), HC is high-caste Hindu, and M is Muslim. Sample of children fullyknown (N=653,496 for neonates, N=629,058 for infants, N=522,377 for under 5s). All differentials are

    3.1.1. The religion mortality differential by age of exposure,gender and

    In propohigh-caste Hof the differlishedwithconstant fradvantage wexposure, cwith the low

    Averaginvival advandriven by h0.53%-pointHindus is gran advantagwith lowerwith age (elow-caste HMuslim advfor girls. Byrates amon0.08%-pointthese patterfacts are sting that higlower castethat they exdegree of sobeen notedratios (boysfemale deforthcomin2008). A loilluminate tments in glater life anand girls. Sfamilies tha

    Table 2Mortality by c

    Male

    Female

    See notes to Tsample size is

    Table 3y rate

    rder

    s to Ta; andrentiaial for

    lth,s.Muslim/high-caste differential increases monotonically

    irth order, whereas the Muslim/low-caste differential isear in birth order (Table 3). In addition, high-caste Hinduswer children than Muslims (Appendix Table 1), and Muslimuse contraceptive methods less often than Hindus (37%

    49.2% in NFHS-2, IIPS and ORC Macro, 2000, and 45.7% ver-.8% in NFHS-3, IIPS and IIPS and Macro International, 2007).together, these patterns are consistent with Muslims havingr taste for fertility than high-caste Hindus.

    The religion mortality differential across time, sector and

    Muslim advantage is not a recent phenomenon, beingnt early in the sample period: see Fig. 1 and Bhat and Zavier.5 Annual averages of religion-specic mortality rates in ourdata are subject to considerable sampling variation, but thisothed when comparing decadal averages. We nd a Muslim5 survival advantage over all Hindus of 1.91%-points (8.9%Hindu under-5 mortality rate) for births occurring duringbirth orderrtional terms, the mortality advantage with respect toindus is decreasing with age of exposure (Table 1): 70%

    ence between Muslims and high-caste Hindus is estab-in the rstmonth after birth, and the difference remainsom infancy up until age ve. In contrast, the Muslimith respect to low-caste Hindus is increasing in age of

    onsistent with the higher SES of Muslims as compareder castes.g across the three religion groups, the under-5 sur-tage of boys over girls is 0.30%-points. This is entirelyigh-caste Hindus amongst whom the differential iss (Table 2). The Muslim advantage over upper-casteeater for girl survival, even though Muslims also showe in boy survival. At birth, girls are, by nature, endowedmortality risks than boys and their advantage is eroded.g. Waldron, 1983). While the Muslim advantage overindus increases with age for both boys and girls, theantage over high-caste Hindus only increases with ageage ve, there is no gender difference in mortality

    gst low-caste Hindus, and girls exhibit an advantage ofs amongst Muslims. In Section 5.1 we shall see thatns persist after conditioning on other covariates. Theseriking and consistent with previous studies suggest-her caste Hindus exhibit greater son preference thanHindus (e.g. Drze and Sen, 1997). Our data indicatehibit greater son preference than Muslims too. A lowern preference amongstMuslims than amongHindus hasin other contexts in India. Muslims exhibit lower sex/girls) at birth (e.g. Barooah and Iyer, 2005), a smallercit in educational enrolment (e.g. Bhalotra and Zamora,g) and a smaller gender gap in height growth (Bhalotra,wer degree of son preference amongst Muslims canhe puzzle of interest in twoways. First, greater invest-irls in childhood result in better maternal health ind this is advantageous for the survival of both boysecond, if girls face better survival chances in Muslimn in Hindu families then, at any given level of mater-

    Mortalit

    Birth o

    1

    2

    3

    4

    See note117,490All diffedifferent

    nal heaMuslim

    Thewith bnonlinhave fewomenversussus 57Takena highe

    3.1.2.state

    Theappare(2005)surveyis smounder-of theommunity and gender.

    Neonatal Infant Under-5

    Low caste 7.41 11.33 15.93High caste 6.40 9.42 12.34Muslim 5.60 8.26 11.33

    Low caste 6.13 10.71 15.94High caste 5.52 9.15 12.87Muslim 4.51 7.68 11.25

    able 1. For neonates, the female sample size is 313,461 and the male340,035.

    19601970terms to 1.6

    5 Their Tablveys of 1963/11981 and 1991998/1999. In2005/2006, whyears.While dsampling erroreligion differeexposed to the relevant mortality risk and for whom caste status issignicant at the 1% level.

    s by community and birth order.

    Neonatal Infant Under-5

    Low caste 8.63 12.64 16.67High caste 7.14 9.94 11.74Muslim 6.62 9.19 11.21

    Low caste 6.24 10.44 15.36High caste 5.11 7.98 10.25Muslim 4.53 7.20 9.78

    Low caste 5.43 9.38 14.64High caste 4.96 8.01 10.58Muslim 4.24 6.82 9.31

    Low caste 6.36 11.01 16.50High caste 6.18 10.40 14.45Muslim 4.70 7.92 11.42

    ble 1. For neonates, sample sizes are, respectively, 197,952; 166,382;171,672 for rst-, second-, third- and fourth and above birth order.ls are signicant at the 1% level, except for the high-caste Hindurst-born children (p-value =0.0226).

    this will tend to contribute to the overall advantage ofwhich decreased in absolute but not in proportional4%-points (16.24%) in 19902001.

    e 6, p. 389 reports the relevant means from the National Sample Sur-964 and 1965/1966, the Sample Registration Survey of 1979, Census1, and the National Family Health Surveys (NFHS) of 1992/1993 andthis paper, we use the NFHS surveys for 1992/1993, 1998/1999, andich contain information on births and child deaths over a span of 42

    ata onmortality from surveys such as theNFHS can be subject to largers, SRS and Census data are not likely to suffer from this problem. Thence investigated here is apparent in these other data sets.

  • S. Bhalotra et al. / Journal of Health Economics 29 (2010) 191204 195

    Table 4Mortality rate

    Urban

    Rural

    See notes to T443,497 for rurfor the Muslimany usual leve

    Althoughobserved inreveals thatHindus in rreligion diffTheMuslimregion; it isin 19 of 26is notable gsonprefereand the Nor

    3.2. Religio

    Stuntingtional statuheight-for-athe second rneous insulWHO convechild is morence populato predict malthough, atcoexist with

    The avebelow the r

    6 The U.S. Nby the Worldproduce the z-

    munities (Table 5). Height outcomes are particularly poor: theaverage low-caste Hindu child is stunted, the averageMuslim childis almost so, and even the average high-caste Hindu child is 1.8standard deviations below the reference median. Stunting ratesare 52.0%, 46.7%, and 43.9% respectively. Indian children fare betterin terms of weight-for-height, with wasting rates at 20.3%, 16.9%

    .3%. The ranking of the three communities by stunting isent with their SES ranking. Muslims exhibit slightly lowerg ratl.7 Ostatuno(Tathaostggesors oby a cighey th

    on ansia (ummof elth,nderantagver, tande. Cr thaagewith reFig. 1.

    s by community and rural/urban location.

    Neonatal Infant Under-5

    Low caste 5.13 8.25 11.48High caste 4.32 6.54 8.51Muslim 3.97 6.40 8.84

    Low caste 7.13 11.61 16.86High caste 6.55 10.24 14.03Muslim 5.68 8.84 12.66

    able 1. For neonates, sample sizes are 209,999 for urban areas andal areas. All differentials are signicant at the1% level or belowexcept/high-caste differentials in urban areas, which are not signicant atl.

    the Muslim advantage over low-caste Hindus is

    and 17consistwastinis smaltionalrevealsHindusshorterbeing mhere suindicattritionin the hto die bnutritiSouth A

    To sby agenal heaand gedisadvMoreoorderor morsmalleadvanttage wboth rural and urban sectors, disaggregation by sectorMuslims only do signicantly better than upper-caste

    ural areas (Table 4). For this reason, we investigate theerential for all-India aswell as for the rural sample only.advantage is not drivenby special circumstances inoneapparent in 11 of 26 states for high-caste Hindus andstates for low-caste Hindus (see Appendix Table 3). Itiven our observations regarding religion differences innce that theMuslimadvantage is least visible in the Easttheast, where Hindus have more matriarchal societies.

    n differentials in nutritional status

    and wasting are commonly used indicators of nutri-s for children under the age of ve. The rst refers toge and indicates cumulative retardation of growth, andefers toweight-for-height,which reects contempora-ts tohealth (e.g.Martorell andHabicht, 1986). Followingntions, both indicators are dened as equal to one if thee than two standard deviations below the NCHS refer-tion median.6 Stunting and wasting have been shownortality at the individual level (e.g. Katz et al., 1989)the population level, wemay see lownutritional statushigh mortality (e.g. Klasen, 2003).

    rage childs height-for-age and weight-for-height iseference population median in each of the three com-

    ational Center for Health Statistics (NCHS) standard, recommendedHealth Organization (WHO) until recently, was the standard used toscores provided in the rst two NFHS waves.

    3.3. Religio

    All-Indiaare in AppesummariseamongstMunicant in rand fatherscated than(all) Hinduspoverty ratwithin rurainequality iin rural areathat the avepositional eaverage, tostandard (soMuslims ev

    SinceMuwe considetality. Statehave lowersition effecthe state av

    7 Regressioncaste and relifever and, in uare Hindu chiles but the differentialwith respect to high-caste Hindusverall, we do not see a clear Muslim advantage in nutri-s the way we do in survival. Disaggregation by genderboygirl differences amongst Muslims and low-casteble 6). However, amongst high-caste Hindu, girls aren boys. This is a further indication of son preferencemarked in this relatively well-off group. The ndingst that mortality is not systematically related to otherf health. In particular, although the incidence ofmalnu-umulative indicator (height) is lower amongst childrenr SES group (Hindus), they are nevertheless more likelye age of ve. This echoes the contrary patterns of mal-d mortality in comparisons of sub-Saharan Africa ande.g. Klasen, 2003).arise, variation in community differences in survival

    xposure suggests that they may be related to mater-and community differences in survival by birth orderindicate that at least some of the high-caste Hindue may stem from their stronger preference for sons.he Hindu disadvantage increases with the childs birthis particularly large for children of birth order fourommunity differences in nutritional status are muchn community differences in survival. Muslims have noith respect to stunting (height) andonly a small advan-spect to wasting (weight).

    n differences in the independent variables

    means of variables used in the analysis by religionndix Table 1. Some relevant religion differences are

    d here. The sex ratio (male/female) at birth is lowestslims, and this difference ismarginally statistically sig-

    elation to high-casteHindus. High-casteHindumotherstend to be the most educated. Muslims are more edu-low-caste Hindus. Muslims are more urbanised thanand (not shown in these descriptive statistics) their

    e relative to Hindus is higher within urban areas thanl areas (Government of India, 2006, p. 153). In this way,n SES betweenMuslims andHindus appears to be lowers. The higher fertility of Muslims has the consequencerage child is of higher birth order but it also exerts com-ffects associated with Muslim children being born, onolder mothers and later in (calendar) time. Overall, thecioeconomic) predictors ofmortality risk donot favouren if some compositional effects do.slims are unevenly distributed across the Indian states,red the state-level relationship of religion and mor-s with a higher proportion of Muslims appear tounder-5 mortality, although this may be a compo-

    t, with the lower mortality risks of Muslims drivingerage (Appendix Fig. 1). The analysis to follow allows

    s of disease probabilities on parental education and indicators forgion suggest that Muslim children are signicantly more prone torban areas, also to diarrhoea (in the 2 weeks before the survey) thandren; see Bhalotra (in press).

  • 196 S. Bhalotra et al. / Journal of Health Economics 29 (2010) 191204

    Table 5Malnutrition by community.

    Low-caste Hindu p-Values of Wald test LC-M High-caste Hindu p-Values of Wald test HC-M Muslim

    Mean S.D. Mean S.D. Mean S.D.

    Height-for-ageS.D. from the reference median (z) 2.054 1.588 0.0003 1.788 1.566 0.0000 1.927 1.650% stunted (z

  • S. Bhalotra et al. / Journal of Health Economics 29 (2010) 191204 197

    with H indexing Hindus (either low- or high caste) and M indexingMuslims. Y J (J=H and M) is the average probability of child deathat the relevant age, XJ

    iis the row vector of independent variables

    of observation i in group J, J is a vector of logit coefcient esti-mates including an intercept and NJ is the number of observationsin group J. Therst term in Eq. (2) is themortality differentialwhichwe would see given the different characteristics of the two groupsif Muslims behaved like Hindus (i.e. with parameters set equal toH for both groups). It is an estimate of the extent to which the gapwould close if Hinduswere assigned the characteristics ofMuslims.We could just as well estimate this term forcing the responses ofthe two groeters of thesecond termtion in moras reectinrates, attitu

    The chartributions oto match osimilar sizetive to thesamples weMuslim samtions. Moreregressors icharacteristprevious onminimisedables in eaobtained.11

    choice of th(Oaxaca and

    5. Results

    5.1. Religioequation

    Logit estand the corare in Appeof newborntage inall thHindus. Bycommunitynicant disfor rst-borsteeply fortality; partilower for ththat the esteffects of mtion. Resultevidence o

    11 The total cmatched withimented withrobustness ofdifferent subsdecompositiondecompositionpaper, there isavailable upon

    ity amongst high-caste Hindus, conrming ndings reported inSection 3.1.

    Mortality risk tends to decrease monotonically with mothersage at birth until age 25 (for high-caste Hindus) or later. Chil-dren born in March and October/November, when temperaturesare more moderate, tend to have better survival chances. Theseeffects are strongest in the low-caste group, possibly indicatingthat vulnerability to the epidemiological environment is positivelyassociated with poverty. The benecial effects of parents educa-tion increase with child age, consistent with an increasing role forenvironmen

    of proupion ae modisacomm. Stamonientsw-can surtrenrefe

    ity anity

    .

    selin

    his seals borta(wh

    for nrespity wsionsnchmtes me oove

    Muslierenedict%-pouslimageffectMuslreate. Firhichl tecrs arome

    lternaucatiomothis forwhen

    e 7.ups to be represented or benchmarked by the param-Muslim equation, M. We present both estimates. Thein Eq. (2) picks up the residual or unexplained varia-

    tality between the two groups. This may be interpretedg group-specic cultural norms, information, discountdes or indeed any omitted variables.acteristics effect can be further decomposed into con-f (groups of) covariates. For this purpose, it is necessarybservations from both groups to obtain samples ofs. Since decomposition results are potentially sensi-matching procedure, 100 low- or high-caste Hindure drawn randomly to be matched with the (smaller)ple, and the reported results are means across simula-

    over, for the contribution of each variable, the order ofn the equation matters since the contribution of eachic is calculated conditional on the contribution of thees (Fairlie, 2006, p. 4). This potential arbitrariness isby randomising the ordering of the independent vari-ch replication and reporting the average results thusThe detailed decomposition is not sensitive to the

    e omitted category for dummies included in the modelRansom, 1999).

    n differences in the parameters of the mortality

    imates for under-5 mortality are in Appendix Table 5responding estimates for neonatal and infant mortalityndix Table 4. Consistent with the biological advantagegirls, they have a signicant neonatal survival advan-ree communities,which is smallest amongsthigh-castethe age of ve, the advantage of girls is eroded in everyand amongst high-caste Hindus it changes into a sig-

    advantage. Under-5 mortality odds tend to be lowestns and then to increase with birth order, though lessMuslims. This is different for neonatal and infant mor-cularly for neonatalmortality, themortality chances aree second and third child than for the rst child. Note

    imated birth order effects are purged of the (correlated)aternal age at birth since this is included in the equa-s for gender and birth order tie in with long-standingf greater son preference and lower desired fertil-

    haracteristics effect is neither sensitive to the choice of sample to bethe smaller group, nor to the order of covariates. We have exper-different matched samples and orders of covariates to check theour baseline results to these two sources of arbitrariness. Usingamples of the larger group has almost no effect on the detailedresults. Changing the order of covariates affects the results of singles. However, when averaging over 100 replications, as we do in thisvery little difference over 10 different sets of 100 replications (detailsrequest).

    effectsthree geducatbecaus

    TheacrossHindustality acoefcthat loment ilinearat anymortalcommutrends

    5.2. Ba

    In tferentieach mTable 7matesand 7,mortalconcluare beestimaM). Sobut the

    5.2.1.Diff

    ties prof 0.36that Madvantdirect eorder.their gfertilitytime, wmedicamotheshow s

    12 An ain the edwith the13 This

    reversessee Tablt and care in the survival technology. In general, theaternal and maternal education are similar across thes. As is commonly found, the coefcients on mothersre somewhat larger than on fathers education, possiblythers are the principal care-givers.12

    dvantage associated with living in a rural area is similarunities although it is a bit smaller amongst high-caste

    te of residence is amore signicant determinant ofmor-gst Hindus than amongst Muslims. The year dummy, which are jointly signicant in all regressions, suggestste Hindus have experienced the smallest improve-vival over time, and Muslims the largest. State-specicds are jointly signicant for the two Hindu groupsrence age, but they are only signicant for under-5mongst Muslims. Overall, there is signicant between-variation in state-level xed effects and state-specic

    e decomposition results

    ction, we present decompositions of the mortality dif-etween Muslims and high- and low-caste Hindus forlity indicator. Estimates for under-5 mortality are inole sample) and Appendix Table 8 (rural sample). Esti-eonatal and infant mortality are in Appendix Tables 6ectively. The discussion will mostly focus on under-5here the paradox is most pronounced, but the essentialare similar for neonatal and infant mortality. Resultsarked rst on one parameter set (the Hindu sample

    H) and then on the other (the Muslim sample estimatesf the results are sensitive to the choice of benchmark,

    rall conclusions of the analysis are not.

    ms versus high-caste Hindusces in average characteristics between the communi-a Muslim disadvantage relative to high-caste Hindusints, explaining none of the 1.34% points advantages exhibit. The characteristics that drive the predicted

    of Hindus are their better parental education and theof their lower fertility, expressedas loweraveragebirthims gain some advantage over Hindus on account ofr urbanisation and two indirect effects of their higherst, the average Muslim child is born later in calendar

    means she benets from secular improvements inhnology and institutional quality, and second, Muslime, on average, older at birth. State-specic trends alsofavour for Muslims.13

    tive explanationwith our sample designmight bemeasurement errorn of the father, since this information is obtained from the interviewer.the case where high-caste Hindus are the reference group. This resultMuslims are the reference group; the other results remain similar;

  • 198 S. Bhalotra et al. / Journal of Health Economics 29 (2010) 191204

    Table 7Decomposition of the HinduMuslim Under-5 mortality differential, whole sample.

    Low-caste/Muslim High-caste/Muslim

    Benchmark LC M HC

    %-Points % ofYH YM

    z-Statistic %-Points % ofYH YM

    z-Statistic %-

    YH YM 4.65 100.00 4.65 100.00 1Explaineda 1.56 33.62 1.25 26.83 0Unexplainedb 3.09 66.38 3.40 73.17 1

    Detailed contributionsGender 0.00 0.00 0.06 0.00 0.00 0.00 0Birth order 0.22 4.71 5.59 0.15 3.17 3.14 0Birth month 0.00 0.05 0.26 0.01 0.18 0.64 0Mothers a 0Fathers ed 0Mothers e 0Rural 0Birth year 0State 0State trend 0

    LC is low-caste mort

    a 1NH

    NHi=1

    F( contr

    be predicted b

    b [YH YM] edicte

    the differentia ertaintwo sets of est total dis 4.65% points ned ovcasts this in % than 1

    Overall,tages ofMuare overwhtion. The suremains a p

    5.2.2. MuslSo as to

    religion andsus religioncompare Mhigh-caste Hlims are, b(Governmeabout a thirbe explaineThis is consservable tra

    5.2.3. IsolatAlthoug

    urbanisatio

    14 In contrasdus is explainrequest). Two/low-caste diffindications thour data. Secoconsistentwitdiscriminationrich with talesit is entirely plservices.

    thatant ipositain cwegge at birth 0.43 9.25 8.08 0.28 6.00 4.76ucation 0.16 3.48 6.81 0.14 3.02 4.95ducation 0.30 6.53 6.71 0.30 6.44 6.36

    0.56 12.09 7.14 0.43 9.34 5.540.05 1.07 0.71 0.04 0.94 0.13

    1.06 22.88 2.29 0.90 19.28 1.33s 1.34 28.86 2.99 1.09 23.37 1.47

    Hindus, HC is high-caste Hindus and M is Muslim. YH YM is the difference in the

    XHi

    J ) 1NM

    NMi=1

    F(XMi

    J ), where J is the group used as benchmark, is the sum of the

    ased on differences in characteristics.

    [1

    NH

    NHi=1

    F(XHi

    J ) 1NM

    NMi=1

    F(XMi

    J )

    ]is the difference between the actual and pr

    l between low-caste Hindus and Muslims while the panel in the right-hand side pimates, one for each benchmark (refer p. 14). The rst cell in each panel shows thehigher than that of Muslims. The rst column shows the %-point difference explai

    terms, and the third column presents the z-statistic. A value of the z-statistic larger

    the decomposition shows that compositional advan-slimsonaccount of their locationor their higher fertilityelmed by the effects of their lower levels of educa-bstantial survival advantage that they exhibit thereforeuzzle with the current (conventional) specication.

    earliersignicdecomtics agclosestims versus low-caste Hindusdetach omitted variables correlated with SES fromgain a better understanding of the role of SES ver-(i.e. unobservables associated with religion), we alsouslims with low-caste Hindus. As discussed earlier,indus are distinctly better off than Muslims but Mus-

    y many indicators, better off than low-caste Hindusnt of India, 2006). The decomposition shows that onlyd of the Muslim advantage over low-caste Hindus cand by the more favourable characteristics of Muslims.14

    istent with the hypothesis that Muslims have an unob-it that is heath-improving.

    ing the rural sampleh the all-India decomposition showed that their greatern confers an advantage upon Muslims, we observed

    t, about 56% of the high-caste Hindu advantage over low-caste Hin-ed by differences in the same covariates (results are available uponpoints are worth noting. First, covariates explain more of the high-erential than that of the Hindu/Muslim differential, reinforcing otherat Muslims have some advantage in survival that is unobservable innd, covariates do not completely explain the caste differential. This ish unmeasured differences in SES between the high and low castes butagainst low-caste households may also contribute. Indian folklore isof low-caste households being denied access to the village well andausible that low-caste households have more limited access to health

    uses H rathable to expl

    6. Extensio

    The prelower survisurvival) chrelated witof identifyieffect of thethe local cobe insufciepublic healtfavour survis any similanthropom

    6.1. Neighb

    There isbourhood oconditionsal., 2003). Cbourhood thfood and hyinfrastructuronmental.M

    Points % ofYH YM

    z-Statistic %-Points % ofYH YM

    z-Statistic

    .34 100.00 1.34 100.00

    .36 26.67 0.37 27.86

    .70 126.67 1.71 127.86

    .00 0.14 1.66 0.00 0.01 0.10

    .36 26.75 10.81 0.24 18.02 4.27

    .01 0.70 1.27 0.00 0.32 0.29

    .09 6.65 2.69 0.01 0.90 0.23

    .57 42.57 14.59 0.39 29.01 5.52

    .34 25.34 14.19 0.32 23.83 5.47

    .25 18.62 8.24 0.24 17.90 5.25

    .43 31.85 4.66 0.70 52.00 1.80

    .05 3.85 0.24 0.01 0.66 0.02

    .21 15.72 0.86 0.38 28.65 0.62ality rate of the Hindu community H and the Muslim community M.

    ibutions of each variable. In words, this is the differential that would

    d differentials. The left-hand side panel shows the decomposition of

    s to high-caste Hindus and Muslims. Within each panel, we presentifferential. For example. low-caste Hindus have a mortality risk thaterall and by each covariate (or set of covariates). The second column.96 in absolute terms denotes a statistically signicant contribution.

    the Muslim advantage over high-caste Hindus is onlyn the rural sample (Section 3). We therefore repeat theion isolating rural households. In general, characteris-ompletely fail to explain the Muslim advantage. Theet is that in the comparisonwithhigh-casteHindus that

    er than M, birth year, state effects and state trends areain 10.6% of the Muslim advantage (Appendix Table 8).

    ns

    vious section shows that upper-caste Hindus sufferval rates for their children despite having better (foraracteristics and this suggests omitted variables cor-h religion. We now extend the covariates in the hopeng relevant omitted variables. Is the religion effect anfamilys religion or is it an effect that operates through

    mmunity? Might the controls for family SES used so farnt? Might Muslims live in states or villages with betterh facilities? What behaviours might Muslims have thatival? In the next section, we consider whether therear unexplained variation in community differences inetric indicators that favours Muslims.

    ourhood effects

    a growing recognition of the importance of neigh-r local community in determining individual outcomes(e.g., Madise et al., 1999; Diez Roux, 2001; Kabir ethild health, for example, may be inuenced by neigh-rough commoncultural beliefs andpractices regardinggiene for mothers and children, differential access tore or health services, or exposure to different envi-To investigate the possible conation of religion and

  • S. Bhalotra et al. / Journal of Health Economics 29 (2010) 191204 199

    neighbourhood effects, we estimate a model with neighbourhoodxed effects. Incorporating such effects is possible because the dataare stratied into local clusters; they contain 8620 clusters withwithin-cluster variation in under-5 mortality; 1274 clusters hadno variation. The average cluster has 52 children. The xed effectslogitmodel is estimated using the conditionalmaximum likelihoodestimator of Chamberlain (1980). This estimator requires a largenumber of clusters but remains consistent if the number of chil-dren per cluster does not tend to innity, which is important in ourcase since the number of children in many clusters is quite small.The estimator does not give consistent estimates of the separatelocal neighbourhood effects and can therefore not be used for thedecompositions. We therefore only consider models for the pooledsample with rst order religion/caste effects, not separate modelsfor the three groups.

    Detailed results are in Appendix Table 9, for otherwise identicalmodels withthe covariatof a Muslimon the correin the modeodds of a lowout and 1.1insensitiveis to controeffects we rect effectslocal comm

    6.2. Househ

    So farweand motherorder, whicof dependeatically weinstance, becontrol forted regressboth wealthments in chis why wecontains inat the timetion samplesurvey. Thethe samplenot yet beeing the meuse the rsindicator of

    15 Indian woto give birth. Tever, the meaMuslims andsystematicallysystematicallylive in at the tibe an artefact16 Assets and

    Thomas, 199517 We also co

    tage in neonatsurvey. Wealthupon request.

    tile of the wealth distribution that the individual household fallsinto.18

    Children face signicantly lower odds of dying as they movebeyond the second quintile of the distribution. The extendeddecomposition shows that wealth contributes to explaining theMuslim survival advantage over low-caste Hindus but makes nocontribution to theMuslim-high-castemortalitydifferential. In thissmaller, more recent sample, the overall contribution of character-istics to explaining the Muslim advantage is even more negativethan in the full sample (Appendix Tables 10A and 10B).

    6.3. Public s

    The baseof birth favseems plau

    ssociawaplainme oith stinglth anper

    efcheselargpopue wo.Wet rurh-caant cdix Te pus ande imilabias onvey.rableordere thlysealysidixme fc timth adl advta. Hnt agron 7,addeors f

    icallicycle,let, oreffecttrendsxampseemNFHSand without neighbourhood effects. Conditional upones listed above, relative to high-caste Hindus, the oddschild dying before the age of ve are 0.848 (t-value

    sponding coefcient: 7.77) and 0.850 (t-value 5.93)ls without and with neighbourhood effects, while the-casteHindu child dying are 1.128 (t-value 7.82)with-

    66 (t-value 8.91) with cluster effects. It is striking howthis conditional distribution of risk across communitieslling for cluster effects. This implies that the religionnd are effects of the religion of the family and not indi-of factors correlated with the share of Muslims in theunity.15

    old wealth

    have controlled for socioeconomic statususing fatherss education (and the mothers age at birth and birthh proxy the stage of the lifecycle and the numbernts). This may be insufcient if Muslims are system-althier than Hindus for a given educational level, forcause they own more physical capital. We thereforeownership of assets.16 Wealth is endogenous if omit-ors such as ability and social connections inuenceand mortality or if mortality, by inuencing invest-

    ildren, determines investments in physical capital. Thisdid not include it in our baseline models. The NFHSformation on a range of assets owned by householdsof the survey only. We therefore restrict the estima-to children born no more than 10 years before eachequation is estimated for under-5 mortality risk, sois further restricted to remove children who have

    n exposed to this risk for the full 5 years.17 Follow-thod popularised by Filmer and Pritchett (2001), wet principal component of a set of assets to create anwealth and include in themodel dummies for the quin-

    men sometimes migrate to their maternal home for a few monthshis will lead to local community being measured with error. How-

    surement error would have to be systematically different betweenHindus to deliver a bias. For example, if migrating Muslim womenmoved at birth to areas with better facilities and/or Hindu womenmoved at birth to areas with worse facilities than in the areas theyme of the survey, then the Muslim advantage in child survival couldof the data. We feel this is not very plausible.

    income appear in reduced form models of health (Strauss and).nsidered the contribution of wealth to explaining the Muslim advan-al mortality amongst children born no more than 5 years before each

    does not explain any part of this differential. Results are available

    vival ahealthalso exand soture wby lookon heaincomeGini cotor of tthe 15Indiasbecausdesignbut thaand higsignicAppen

    Sincvillageeffectivthe avation wthe surcompaare recno moto anathis an(Appenand sospeci(1), wisurvivaraw darepresethe two(colum

    Theindicat

    18 Specfridge, bush toi19 The

    specic20 For e

    such as abetweenervices

    line decomposition (Table 7) showed that (later) yearour Muslim children relative to high-caste Hindus. Itsible that this reects secular improvements in sur-ated with the quality and spread of medical facilities,reness and overall prosperity. These same factors maythe higher mortality risk associated with rural areas,

    f the inter-state variation in intercepts which we cap-ate dummies. We investigate this for under-5 mortalityat the effects of (log) real per capita state expenditured development projects, controlling for (log) real statecapita and for rural and urban income inequality usingients.19 We add to Eq. (1), for under-5 mortality, a vec-state-level variables. These data are only available for

    er Indian states, which account for more than 95% oflation, but result in a loss of about15.5%of observationsmen in smaller states are over-sampled in the NFHSnd that state expenditure and income are insignicantal inequality increases childmortality amongstMuslimsste Hindus. Together, the state-level variables make noontribution to explaining themortality differential (seeables 11A and 11B).blic expenditure may not be evenly distributed acrossincreases in expenditure do not directly translate intoprovement in services, we also consider indicators oflity of health facilities at the village level. This informa-ly collected for rural areas in the rst two rounds ofSince the data in these two rounds are not to be strictly,20 we use only NFHS-2 for this exercise. As facilitiesd only at the time of the survey, we keep children bornan six years before the interview. This makes it hardunder-5 mortality and allow 5 years exposure and sos is conducted only for neonatal and infant mortalityTables 12A, 12B and 12C). Given the short time-spanairly small state-specic samples, we drop the state-e trends but, otherwise, the estimatedmodel is as in Eq.ded village-level regressors. In this sample, the Muslimantage over high-caste Hindus is not signicant in theowever, even equal survival chances of these groupspuzzle because differences in the characteristics of

    upspredict aMuslimdisadvantageofup to0.77%-pointsAppendix Table 12B).d village variables are the log of the village population,or an all-weather road, a pharmacy, a mahilamandal

    y, we use a PCA index based on dummies for ownership of a radio,motorbike, car, television, andwhether the household has electricity,pit toilet.s of these variables are identied because the model includes staterather than interactions of state and time dummies.le, some unexpected patterns appear when comparing these data,ing reduction in the percentage of villages with a hospital or a clinic-1 and NFHS-2.

  • 200 S. Bhalotra et al. / Journal of Health Economics 29 (2010) 191204

    (womens council), an anganwadi (community childcare centre), aprimary health centre, a primary health sub-centre, a hospital, anddispensary or clinic.21 The contributions of the village variables toexplaining the differential are insignicant even if some are large inmagnitude. Thiswas anticipated by our estimateswith xed villageeffects on the pooled sample; see Section 6.1. Using more aggre-gated variables that are comparable in NFHS-1 and NFHS-2, andthus using both datasets, we also found no systematic evidence ofdifferential access tohealth services after controlling for village sizein Bhalotra et al. (forthcoming).

    6.4. Maternal health, diet, preferences, behaviours, kinship

    We observed in Section 3 that more than two-thirds of the sur-vival advantageofMuslimsoverhigh-casteHindus is realized in therstmonthmay be reladelivery anincorporateset of indicadescriptivethe specicresults.

    We usetal care (i.eanti-tetanuhome, a privfed and if sh24h; whethvegetarian;any employ(whether ma proxy forfromher reperence in febirths (malsex ratio re

    Antenatlished inpuhealth transtature aredisposition2009). In gof the mothhighly prevto the fact tlims tobe veHindu womlim womentime away fingantenatchildren to

    21 For neonalim regressionthe presence oamongst low-mortality to thwadi, which isthis suggests tnative explanactual accessexclusion, or hdownwards.22 The indica

    the time of the

    employment (in agriculture or informal work) is associated withless healthy behaviours and an increase in infant mortality. If Mus-lims practice greater endogamy and have stronger kin networksthen they may be better placed to avoid child death in difculttimes. Endogamy is associated with lower marriage payments andtheperceptionof thegirl child as less of a burden, aswell aswith theavailability of extended care for the mother and the newborn childfrom the natal family (e.g. Robinson, 2007). It may also make forbetter information sharing amongst women, and stronger insur-ance networks.

    In Section 3, we provided evidence suggesting that Hindusexhibit greaence mortawith excessupon the av

    effec. If sot of glth todue

    andu hoin

    in adentheigew mmeas, bedix Twhyfferea tent to loss limorefererporomers eet throssus t

    on we orrianifull

    eferende

    s in mariabg thludeandselint the

    e of the useectiveotherecic

    l sex rparityet to zof life. This suggests that explanations of thedifferentialted to customs, attitudes, behaviours, maternal health,d early feeding practices. In this section, we attempt tothese variables in the analysis. We describe the richtors used, then explain why they are relevant, presentstatistics, discuss the limitations of the data, describeations employed and then proceed to summarise the

    indicators for whether the mother completed antena-., whether she had at least three antenatal checks, ones injection, and one course of iron tablets); gave birth atate facility or a public facility;whether she ever breast-e did, whether she initiated breastfeeding within 2 orer she has low BMI (

  • S. Bhalotra et al. / Journal of Health Economics 29 (2010) 191204 201

    Table 8Decomposition of the HinduMuslim stunting differential.

    Low-caste/Muslim High-caste/Muslim

    Benchmark HC M

    %-Point % of z-Statistic %-Point % of z-Statistic

    YH YM 2Explained 2Unexplained 0Detailed con

    Gender 0Birth orde 1Birth mon 0Mothers a 0Fathers ed 1Mothers e 1Rural 0Birth year 0State 0

    Children no m FHS rNadu, West Be

    Table 9Decomposition

    Hi

    Benchmark

    YH YMExplainedUnexplained

    Detailed conGenderBirth ordeBirth monMothers aFathers edMothers eRuralBirth yearState

    Children no mNadu, West Be

    cedes NFHSand estimatbreastfeedi14A and 14height, vegewas only gaarately, usinuse under-5mortality. Aon a 35 yearisk of unde

    We ndmortality rihad low BMstature, hascomplete aIn additionhigher thanson prefereinsignicanbias. The saing and placLC M

    %-Point % ofYH YM

    z-Statistic %-Point % ofYH YM

    z-Statistic

    5.25 100.0 5.25 100.01.53 29.15 1.23 23.433.72 70.85 4.02 76.57

    tributions0.00 0.00 0.02 0.00 0.08 0.33

    r 0.28 5.26 2.75 0.28 5.41 2.21th 0.22 4.10 2.72 0.57 10.87 5.89ge at birth 0.32 6.14 2.36 0.36 6.79 2.07ucation 0.10 1.97 1.52 0.24 4.61 2.90ducation 0.71 13.60 4.99 0.51 9.65 3.16

    0.42 7.99 1.36 0.79 15.10 2.650.58 11.11 6.06 0.54 10.33 4.130.08 1.45 0.19 0.11 2.16 0.18

    ore than 36 months old for states where height and weight were measured in all Nngal, and Himachal Pradesh. See also notes to Table 7.

    of the HinduMuslim wasting differential.

    Low-caste/MuslimLC M HC

    %-Point % ofYH YM

    z-Statistic %-Point % ofYH YM

    z-Statistic %-

    3.43 100.0 3.43 100.0 00.70 20.44 0.31 8.94 02.73 79.56 3.13 91.06 0

    tributions0.01 0.17 0.39 0.00 0.10 0.19 0

    r 0.01 0.27 0.11 0.07 2.02 0.66 0th 0.19 5.58 2.70 0.06 1.88 0.75 0ge at birth 0.10 2.84 0.80 0.06 1.76 0.42 0ucation 0.02 0.55 0.35 0.04 1.21 0.63 0ducation 0.19 5.55 1.48 0.25 7.37 1.66 0

    0.30 8.67 1.14 0.16 4.76 0.65 00.03 0.84 0.26 0.14 3.99 1.19 00.53 15.55 1.70 0.26 7.44 0.55 0

    ore than 36 months old for states where height and weight were measured in all NFHS rngal, and Himachal Pradesh. See also notes to Table 7.

    -2 by at most 3 years or NFHS-3 by at most 5 years,e the effects of sibling-varying antenatal care, delivery,ng, BMI, and mothers employment (Appendix TablesB). We do not use NFHS-1 because it has no data ontarianism or BMI. However information on endogamythered in NFHS-1, forcing us to estimate its effects sep-g all births in this one round. For the long samples, wemortality and, for the sample of latest births, neonatals discussed earlier, we cannotmodel under-5mortalityrs sample and adequately allow for full exposure to ther-5 mortality.some evidence, in some communities, that childhoodsk is elevated when the mother has a vegetarian diet,I when pregnant at the time of the survey, is of shortemployment outside the home, exhibits failure to

    ntenatal care, or did not breastfeed the index child., we nd that the mortality risk for a female child isfor a male child when their mother exhibits higher

    nce. Place of delivery and timing of breastfeeding havet or unexpected effects that suggest an omitted variablemple of births for which antenatal care, breastfeed-e of delivery variables were collected is comparatively

    small and, pculated in thTherefore, wvariables in12% [44%] oever, a signconsideringbreastfeediing for theMvegetarianadvantagecaste Hinduthe puzzletality amonmortality adiet, and sfor up to 4castes.26

    26 Wealsoesalcohol and toYH YM YH YM

    .81 100.00 2.81 100.00

    .63 93.60 2.11 75.25

    .18 6.40 0.70 24.75

    .01 0.53 1.80 0.01 0.18 0.45

    .04 37.07 6.03 0.68 24.05 2.22

    .22 7.79 4.11 0.47 16.70 4.86

    .31 10.96 3.19 0.26 9.19 1.69

    .57 55.88 7.00 1.63 58.00 3.89

    .58 56.32 11.01 0.97 34.52 3.03

    .28 10.11 2.50 0.46 16.24 2.63

    .73 26.10 9.51 0.90 31.90 7.16

    .48 17.21 2.62 0.04 1.27 0.09ounds, i.e., all states except Andhra Pradesh, Madhya Pradesh, Tamil

    gh-caste/MuslimM

    Point % ofYH YM

    z-Statistic %-Point % ofYH YM

    z-Statistic

    .40 100.0 0.40 100.0

    .56 138.8 0.27 66.67

    .96 238.8 0.67 166.7

    .01 2.89 1.24 0.02 3.82 0.77

    .28 70.20 2.02 0.11 27.75 0.46

    .03 7.64 0.70 0.07 18.43 0.91

    .02 4.56 0.22 0.07 16.60 0.48

    .40 98.75 2.29 0.23 55.83 0.70

    .39 96.80 3.36 0.52 127.7 2.09

    .08 20.69 0.93 0.10 24.63 0.65

    .08 18.83 1.29 0.20 50.01 2.02

    .36 88.20 2.44 0.41 102.0 1.38

    ounds, i.e., all states except Andhra Pradesh, Madhya Pradesh, Tamil

    robably for this reason, most of the contributions cal-e corresponding decomposition have low signicance.e do not put too much weight on these ndings. The

    cluded in this specication can only account for up tof the Muslim advantage over high [low] castes. How-icant contribution of maternal work emerges whenthe Muslim advantage over low-caste Hindus, and

    ng practices appear to play a signicant role in account-uslim advantage over high castes.Maternal height and

    diet contribute signicantly to explaining the Muslimin under-5 mortality, especially with respect to high-s. Our measure of son preference tends to reinforce

    because, although it contributes to lower female mor-gst Muslims, it contributes even more to lower malemongst Hindus. All in all, including maternal height,on preference in the analysis allows us to account6% [52%] of the Muslim advantage over high [low]

    timateda specication including indicators for themother consumingbacco. These variables do not contribute to the differential.

  • 202 S. Bhalotra et al. / Journal of Health Economics 29 (2010) 191204

    The data conrm that Muslims marry relatives more often (seeAppendix Table 1), but children of marriages between relativestend to suffer higher death risks (see Appendix Table 15A), so thattaking endogamy into account reinforces the puzzle (see AppendixTable 15B). This is not surprising, since endogamy may be captur-ing much more than close kinship. Consanguinity is often foundto be positively correlated with higher child mortality, althoughthe evidence is mixed and little is known about the respectiverole of genetic, socio-demographic, and economic factors when acorrelation is observed (Dorsten et al., 1999). The data needed to

    identify stronger support networks amongstMuslims are not avail-able to us but this would be a fruitful avenue for future research. Arecent study nds some evidence of separate social interactionsamongst Muslims and Hindus in rural Bangladesh (Munshi andMyaux, 2006).

    To summarise, except for maternal vegetarian diet and height,controlling forhouseholdwealth, access tohealth services andpub-lic infrastructure, and a range of more unconventional variablespertaining to the mothers health, attitudes and behaviours makes,if at all, a small contribution to understanding the puzzle of why

    Table 10Summary of decomposition results.

    Explained variable Specication details Sample size Low caste High caste

    Benchmark LC M HC M

    (1) Under-5 mortality Standard set of regressors 518,585 Differential 4.65 4.65 1.34 1.34Explaineda 1.56 1.25 0.36 0.37

    (2) Rural sample 349,662 Differential 4.25 4.25 1.44 1.44Standard set of regressors exceptrural indicator

    Explaineda 1.28 0.54 0.15 0.56

    (3) Children born no more than 10 yearsbefore the survey. Standard set ofregressors plus wealth quintile

    151,068 Differential 3.87 3.87 1.02 1.02Explaineda 1.48 0.67 0.62 0.89

    (4) Fifteen largest Indian states only 436,151 Differential 4.75 4.75 1.41 1.41Standard set of regressors plus statemacroeconomic variables

    Explaineda 1.55 1.24 0.36 0.38

    (5) NFHS-2 and NFHS-3 284,479 Differential 4.77 4.77 1.30 1.30Standard set of regressors plusmother height, diet, andself-reported son preference

    Explaineda 2.30 2.50 0.41 0.60

    (6) NFHS-1. Standard set of regressorsplus endogamy variables

    175,829 Differential 4.89 4.89 1.16 1.16Explaineda 0.89 0.08 1.55 1.37

    (7) Neonatal mortality Standard set of regressors 648,615 Differential 1.72 1.72 0.92 0.92Explaineda 0.62 0.56 0.00 0.01

    (8) NFHS-2. Children born no more than6 years before the survey. RuralsampleStandard set of regressors exceptstate-specic linear trends, plusvillage characteristics

    36,382 Differential 1.47 1.47 0.60 0.60Explaineda 0.62 0.45 0.77 0.47

    (9) NFHS-2 and NFHS-3. Latest births ifoccurred to the respondent no morethan 3 years before NFHS-2 and nomore than 5 years before NFHS-3Standard set of regressors exceptstate-specic linear trends, plusantenatal, delivery, breastfeeding,maternal work and maternal BMIdummies

    44,462 Differential 0.78 0.78 0.35 0.35

    Explaineda 0.34 0.26 0.25 0.04

    (10) Infant mortality Standard set of regressors 624,444 Differential 3.06 3.06 1.33 1.33Explaineda 1.18 0.87 0.07 0.12

    (11) As in specication (8) 31,204 Differential 2.31 2.31 0.41 0.41Explaineda 1.04 0.13 0.72 0.43

    (12) Stunting Children up to 36 months old. Fivestates excluded for consistencyacross rounds

    43,192 Differential 5.25 5.25 2.81 2.81Explaineda 1.53 1.23 2.63 2.11

    (13) Wasting As in specication (12) 43,354 Differential 3.43 3.43 0.40 0.40

    All samples in therwdata. The stand ers agof birth of the rted hlow-caste Hin

    a 1NH

    NHi=1

    F(clude only children fully exposed to the relevant mortality risk. Unless specied oard set of regressors includes indicators for gender, birth order, birth month, mothchild, state of residence, and state-specic linear time trends. The sample size repodus, HC is high-caste Hindus and M is Muslim. See also notes to Table 7.

    XHi

    J ) 1NM

    NMi=1

    F(XMi

    J ) where J is the group used as benchmark.Explaineda 0.70 0.31 0.56 0.27ise, all samples are drawn from pooled NFHS-1, NFHS-2 and NFHS-3e at birth, fathers education, mothers education, rural location, yearere is the sum of all observations in community-specic logits. LC is

  • S. Bhalotra et al. / Journal of Health Economics 29 (2010) 191204 203

    mortality amongst children ofMuslims is lower than amongst chil-dren of high-caste Hindus. To some extent this is no doubt becausethe proxies available for health, attitudes and behaviours are par-tial and not immune to endogeneity issues. We expect that the realweight of these variables is considerably greater than we estimate.

    7. Nutritional status: stunting and wasting

    Commundescribed inwe do notrelatively ssurvey roundecompositof communshows thattage over M

    The chareducation, bonly partlysation, birthover low-cferential inMothers edcaste Hindunamely, higfathers eduspecic to Mrelative tocase compavantage is erelative to lis consistenhas some u

    Now corelatively w0.40%-pointhigher by uwho are fabirth order.20% of theirupon unobsMuslims shHindus in amortality.

    8. Conclus

    Our maiand analysebetter healmuch largerchild survivWe have esfactors thatvival but thdominate sMuslim advhalf of their

    27 In this shoThis is a reeparedwithMurapidly (see De

    show that the religion differential is unchanged if we control com-prehensively for characteristics of the villages or towns in whichhouseholds of different religions live. Of all the more unusual vari-ables we haalone appeatage over hover a third

    Potentiaer soge cuours,onphealtdictae rige daf simily cato eexp

    rks thsha

    on, 2l aspey, anndogporteightoutions.articong

    at, inl adv

    igh-calityst hist hiigher

    evfac. Ited)

    xperial deof

    of hitracey (SEfact ming te livelikelysurvre coial foattitith coity an

    wled

    ia Bresearlierity differences in anthropometric decits wereSection 3. The set of regressors is as in Eq. (1) althoughinclude state-specic trends because this sample ishort, consisting of three years for each of the threeds. The logit estimates are in Appendix Table 16, andion results in Tables 8 and 9. For stunting, the rankingities is consistent with their SES. The decompositionbetween 75% and 94% of the high-caste Hindu advan-uslims is due to differences in characteristics (Table 8).acteristics that favour high-caste Hindus are parentalirth order, and month of birth, and their advantage isoffset by a Muslim advantage on account of urbani-year, and state of residence. The Muslim advantage

    aste Hindus is more enigmatic: only 29% of the dif-stunting is explained by differences in characteristics.ucation, age at birth and birth year disadvantage lowers and overwhelm factors that disadvantage Muslims,her birth order, distribution of month of birth andcation.27 Thus there appear to be omitted variablesuslim children that favour their height performance

    low-caste Hindus. When Muslims do worse, as is thered with high-caste Hindus, then most of their disad-xplained. However, when they do better, as is the caseow-caste Hindus, their advantage is unexplained. Thist with our earlier nding that the Muslim communitynobservable traits that favour health.nsider wasting, the indicator on which Muslims doell. Although wasting amongst Muslims is lower bys, differences in characteristics suggest it should bep to 0.56%-points, compared with high-caste Hindus,voured again by better parental education and lowerMuslimsdobetter than the low-caste groupbut atmostadvantage is explained by characteristics and this restservables captured by the state dummies (Table 9). Soow an unexplained advantage relative to both castes ofvoiding low weight-for-age just as they do in avoiding

    ion

    n purpose in this paper has been to highlight, describethe hitherto neglected fact that Muslims in India have

    th outcomes than Hindus. This religion differential isthan themorewidely recognised gender differential inal in India and it appears to have persisted for decades.timated a range of specications and identied someendow Muslims with an advantage in health and sur-eir lower-SES and, in particular, education, tends too that conventional variables can explain none of theantage over high-caste Hindus and, typically, less thanadvantage over low-caste Hindus (see Table 10). We

    rt sample, Muslim fathers are less educated than low-caste Hindus.ction of the more rapid educational progress of low caste as com-slimmen over time. In contrast,Muslimwomenhave advanced fairlyolalikar, forthcoming).

    strongmarriabehavilower sbetterlims iswith thhave thtance oprimarappearbut onenetwople, theRobinsseveradeliverture (eself-remore lto rulelimitat

    In pence amnd thsurvivathan h5 mortamongamonghave h

    Thefactorsof life(observthey eof foetfertilitytermsof conpovertare inassumaveragmorehigherrates apotentand inrisk, wmortal

    Ackno

    Sonundertol. Eave experimented with, non-vegetarian diet and heightr to be signicant contributors to the Muslim advan-igh-caste Hindus, with non-vegetarian diet explainingof their advantage in under-5 mortality.l explanations of the Muslim advantage include theircial networks, which may be accounted for by theirlture and their minority status in India; their healthiersome of which are associated with religion; and their

    referencewhich, amongst other things,may explain theh of Muslim mothers. Personal hygiene amongst Mus-tedby the requirement towashbeforeprayer combinedour of praying ve times a day and while we do notta to analyse this, historical records suggest the impor-ple hygiene in lowering mortality when mortality isused by infectious disease (e.g. Miller, 2008). Muslims

    njoy closer kinship than Hindus, of which endogamy isression. Stronger kinship tends to result in better socialat may inuence child mortality through, for exam-

    ring of information or child care amongst women (e.g.007). The data at hand allowus to investigate the role ofcts of behaviours around birth (antenatal care, place ofd breastfeeding practices), one aspect of marriage cul-amy), and of one proxy for son preference based on thed ideal sons to total children ratio. These do not shedon the puzzle at hand, but further research is neededthe possibility that this lack of evidence is due to data

    ular, we nd indirect evidence that lower son prefer-st Muslims may be part of the explanation. Indeed, wea plain logit model of neonatal or infant mortality, theantage of the female child is higher amongst Muslimsaste Hindus, whilst, in a logit model explaining under-, the risk associated with being a girl is only signicantgh castes. Furthermore, the sex ratio at birth is highergh-caste Hindus. Finally, Muslim women are found tostature, which in part captures better health.

    idence points to the importance of early lifetors that inuence child health by the rst monthis therefore plausible that Muslims not only havebetter survival chances of newborn children but thatence, for the same reasons, (unobserved) lower risksath. This would contribute to explaining the higherMuslim women which, so far, has been explained ingher desired fertility and greater aversion to the useption. However, selection may turn this around. IfS) dominates in the risk of foetal loss and Muslimsore likely than upper-caste Hindus to suffer it then,

    hat it is the frailest children who die in utero; thebirth amongst Muslims will tend to be less frail andto survive. This may contribute to explaining the

    ival chances of Muslim children given that survivalnventionally measured as a fraction of live births. Ther differences between communities in maternal healthudes to pregnancy to create differences in foetal deathnsequences for observed differences in both childhoodd fertility, has not been investigated in the literature.

    gements

    halotra acknowledges funding from ESRC and DFIDrch grant RES-167-25-0236 held at the CMPO in Bris-versions of this paper were presented at a DFID and

  • 204 S. Bhalotra et al. / Journal of Health Economics 29 (2010) 191204

    CMPO funded workshop on Child Health in Developing Countriesin Bristol in June 2005 and at the European Society of PopulationEconomics Conference in Seville in June 2009. The authors wouldlike to thank Timothy Besley and Robin Burgess for sharing theirstate-level data. We are grateful to the Editor and two anonymousreferees for helpful comments.

    Supplementary results.

    Supplementary results associatedwith this article can be found,in the online version, at doi:10.1016/j.jhealeco.2009.11.002.

    References

    Barooah,V.K., Iyer, S., 2005.Vidya,VedaandVarna: the inuenceof religionandcasteon educati

    Basant, R., 200Economic

    Bauer, T., GolhHealth Eco

    Beckett, M., DrospectiveJournal of

    Bhalotra, S., 20les of the

    Bhalotra, S., inof Develop

    Bhalotra, S., Rhealth in d

    Bhalotra, S., Vin India. InUniversity

    Bhalotra, S., vin India: d274290.

    Bhalotra, S., ZaA., BasantDelhi.

    Bhat, M., ZaviMuslims. E

    Bhaumik, S.K.,caste andWilliam D

    Bozzoli, C.G., Ddisease. D

    Census of Indigov.in/DatF10.pdf.

    Chamberlain,nomic Stu

    Cutler, D., Deaof Econom

    Deolalikar, A.BSharif, A.,Press, Delh

    DiezRoux,A., 2Journal of

    Dorsten, L.E., Hhood mor(2), 2632

    Drze, J., Sen, Adon Press,

    Dwyer, J.W., Cand afliaBehaviour

    Fairlie, R., 2006. An extension of the BlinderOaxaca decomposition technique tologit and probit models. IZA Discussion Paper No. 1917. Institute for the Studyof Labor (IZA), Bonn.

    Filmer, D., Pritchett, L.H., 2001. Estimating wealth effects without expendituredata-or tears: an application to educational enrolments in states of India.Demography 38, 115132.

    Fryer, R.G., Levitt, S.D., 2004. Understanding the blackwhite test score gap in therst two years of school. Review of Economics and Statistics 86, 447464.

    Fuchs, V.R., 2004. Reections on the socio-economic correlates of health. Journal ofHealth Economics 23, 653661.

    Government of India, 2006. Social, Economic and Educational Status of the MuslimCommunity of India: A Report. Prime Ministers High Level Committee.

    Guinnane, T., 2005. The fertility transition in Europe. European Society of PopulationEconomics (Keynote address).

    IIPS, 1995. National Family Health Survey (MCH and Family Planning), India199293. International Institute for Population Sciences (IIPS), Mumbai.

    IIPS and ORC Macro, 2000. National Family Health Survey (NFHS-2) 199899 India.International Institute for Population Sciences (IIPS), Mumbai.

    IIPS and Macro International, 2007. National Family Health Survey (NFHS-3) India06. International Institute for Population Sciences (IIPS), Mumbai.., 200of Bic/boco.N., Tir, B.,y old1.West,g antemiol., 200gyounssmenll, R., Halkne3: Meted. Pleright,dardsWealt., 200ricanK., MlopmR., Ranew ofS., Sen-healtN.J., Ms betwlation

    n, R., 2PoliticR.G.,vioraYork.

    A., 19and295J., Thsions.omics, I.,s in mtality:York,

    D., Bus ethnon in rural India. Journal of Development Studies 41 (8), 13691404.7. Social economic and educational conditions of Indian Muslims.and Political Weekly 42 (10), 828832.mann, S., Sinning, M., 2007. Gender differences in smoking behavior.nomics 16, 895909.aVanzo, J., Sastry, N., Panis, C., Peterson, C., 2001. The quality of ret-data: an examination of long-term recall in a developing country.

    Human Resources 36 (3), 593625.08. Gender Inequality in Height Growth: Evidence from Cohort Pro-Heights of Men and Women, Mimeograph. University of Bristol.press. Fataluctuations: cyclicality in infantmortality in India. Journalment Economics (published online in May).awlings, S., 2009. Gradients of the intergenerational transmission ofeveloping countries. IZA Discussion Paper 4353. Bonn.alente, C., van Soest, A., forthcoming. Religion and childhood death: Sharif, A., Basant, R. (Eds.), Handbook of Muslims in India. OxfordPress, Delhi.an Soest, A., 2008. Birth-spacing, fertility and neonatal mortalityynamics, frailty, and fecundity. Journal of Econometrics 143 (2),

    mora, B., forthcoming. Social divisions in education in India. In: Sharif,, R. (Eds.), Handbook of Muslims in India. Oxford University Press,

    er, F., 2005. Role of religion in fertility decline: the case of Indianconomic and Political Weekly 40 (5), 385402.Chakrabarty, M., 2006. Earnings inequality in India: has the rise ofreligion based politics in India had an impact? Working Paper 819.avidson Institute, University of Michigan.eaton, A., Quintana-Domeque, C., 2009. Adult height and childhood

    emography 46 (4), 647669.a, 2001. Fertility Tables. Available online at http://www.censusindia.a Products/Data Highlights/Data Highlights link/data highlights F9

    G., 1980. Analysis of covariance with qualitative data. Review of Eco-dies 67, 225238.ton, A., Lleras-Muney, A., 2006. The determinants ofmortality. Journalic Perspectives 20 (3), 97120.., forthcoming. How do Indian Muslims fare on social indicators? In:Basant, R. (Eds.), Handbook of Muslims in India. Oxford Universityi.001. Investigatingneighborhoodandareaeffectsonhealth.AmericanPublic Health 91 (11), 17831789.otchkiss, L., King, T.M., 1999. The effect of inbreeding on early child-

    tality: twelve generations of an Amish settlement. Demography 3671.., 1997. Indian Development: Selected Regional Perspectives. Claren-Oxford.larke, L.L., Miller, M.K., 1990. The effect of religious concentrationtion on county cancer mortality rates. Journal of Health and Social31 (June (2)), 185202.

    2005Jann, B

    tionc/bo

    Kabir, ZHojetus b779

    Katz, J.,uatinEpid

    Klasen, SdyinAsse

    MartoreIn: Fvol.2nd

    Micklewstanand

    Miller, GAme

    Munshi,Deve

    Oaxaca,Revi

    Osmani,of ill

    Madise,statuPopu

    Robinsoand

    Rogers,BehaNew

    Shariff,dus2947

    Strauss,deciEcon

    WaldronenceMorNew

    Wilson,land6. Fairlie: Stata Module to Generate Nonlinear Decomposi-nary Outcome Differentials. Available from http://ideas.repec.org/de/s456727.html.shelman, C., Aguero-Torres, H., Chowdhury, A.M.R., Windblad, B.,2003. Gender and ruralurban differences in reported health sta-er people in Bangladesh. Archives of Gerontology and Geriatrics 37,

    K., Tarwotjo, I., Sommer, A., 1989. The importance of age in eval-hropometric indices for predicting mortality. American Journal ofogy 130 (6), 12191226.3. Malnourished and surviving in South Asia, better nourished andg inAfrica:what canexplain this puzzle? In: FAO (Ed.),Measurement,t of Food Deprivation and Undernutrition. FAO, Rome.abicht, J.-P., 1986. Growth in early childhood in developing countries.r, F., Tanner, J.M. (Eds.), Human Growth: A Comprehensive Treatise,hodology and Ecological, Genetic, and Nutritional Effects on Growth,num Press, New York.J., Ismail, S., 2001.What can child anthropometrics reveal about livingand public policy: an illustration from central Asia. Review of Incomeh 47 (1), 6580.8. Womens suffrage, political responsiveness and child survival inhistory. Quarterly Journal of Economics 123 (3), 12871327.yaux, J., 2006. Social norms and the fertility transition. Journal ofent Economics 80 (1), 138.som, M., 1999. Identication in detailed wage decompositions. TheEconomics and Statistics 81 (1), 154157., A.K., 2003. The hidden penalties of gender inequality: fetal originsh. Economics and Human Biology 1, 105121.atthews, Z., Margetts, B., 1999. Heterogeneity of child nutritionaleen households: a comparison of six sub-Saharan African countries.Studies 53 (3), 331343.

    007. IndianMuslims: the varied dimensions ofmarginality. Economical Weekly 42 (10), 839843.Hummer, R.A., Nam, C.B., 2000. Living and dying in the USA.l, health and social differences in adult mortality. Academic Press,

    95. Socio-economic and demographic differentials between Hin-Muslims in India. Economic and Political Weekly 30 (46),

    3.omas, D., 1995. Human resources: empirical models of householdIn: Behrman, J.R., Srinivasan, T.N. (Eds.), Handbook of Development, vol. IIIA. North Holland, Amsterdam.1983. The role of genetic and biological factors in sex differ-

    ortality. In: Lopez, A.D., Ruzicka, L.T. (Eds.), Sex Differentials inTrends, Determinants and Consequences. Dept. of Demography,NY.rgess, S., Briggs, A., 2005. The dynamics of school attainment of Eng-ic minorities. Working Paper 05/130. CMPO, University of Bristol.

    The puzzle of Muslim advantage in child survival in IndiaIntroductionBackground and dataA profile of the religion differentialReligion differentials in mortalityThe religion mortality differential by age of exposure, gender and birth orderThe religion mortality differential across time, sector and state

    Religion differentials in nutritional statusReligion differences in the independent variables

    MethodsResultsReligion differences in the parameters of the mortality equationBaseline decomposition resultsMuslims versus high-caste HindusMuslims versus low-caste HindusIsolating the rural sample

    ExtensionsNeighbourhood effectsHousehold wealthPublic servicesMaternal health, diet, preferences, behaviours, kinship

    Nutritional status: stunting and wastingConclusionAcknowledgementsReferencesReferences