The higher growth rate and fertility of Muslims compared

Transcription

1 Muslim-Hindu Fertility Differences Evidence from National Family Health Survey-II This paper examines Muslim-Hindu differences in the desire for an additional child and the use of contraceptives. It uses data from the National Family Health Survey carried out in and employs multivariate and multilevel regression models in data analysis. Results show that Muslim-Hindu differences in the desire for additional children and use of contraceptives are pervasive across India and almost invariant across states and districts. This is consistent with the findings from our analysis of data from the first NFHS in However, Muslim-Hindu differences have narrowed between and It is thus argued that Muslim-Hindu fertility behaviour seems to be moving towards convergence. The pervasiveness of Muslim-Hindu differences in reproductive behaviour calls for complementary global explanations. A DHARMALINGAM, K NAVANEETHAM, S PHILIP MORGAN The higher growth rate and fertility of Muslims compared with Hindus in India is not a new phenomenon. Some right-wing Hindu religious groups have always attempted to make political capital out of the higher growth of Muslims. The late S Chandrasekhar, a former union minister for health and family planning, once observed: While the government has tried to make the family planning programme a national one above party and communal religious and caste politics, there has always been some criticism from certain right-wing caste or religious groups pointing to the disproportionate rates of growth of the other group. However, with the growth of right-wing political movements, the politicisation of fertility differentials between Muslim and Hindu populations intensified in the 1990s [Basu 1996, 1997; Jeffery and Jeffery 2005]. Although it is now a commonly held view that the high growth rate of Muslims in India results from their higher fertility, there are doubts about the existence of fertility differentials between Muslims and Hindus if proper allowance is made for differentials in compositional and structural aspects such as rural-urban distribution, education, poverty and status of women [Jeffery and Jeffery 1997, 2000, 2002; Iyer 2002a]. A number of national level studies in the last few years have looked at the net influence of religion on fertility [e g, Bhat 1996; Ramesh, et al 1996; Dreze and Murthi 2001]. Only a few studies however had religious differentials in fertility as their main focus [Bhat and Zavier 2004; Dharmalingam and Morgan 2004; Mishra 2004]. Bhat and Zavier (2004) used time series and cross-sectional census and survey data, and a number of fertility measures. They also pooled together data from the two National Family Health Surveys (NFHS-I and NFHS-II) for the years and , and employed multivariate statistical techniques to examine the net effect of religion. Their analysis showed that Muslims have always (both in pre- and post-independence India) had higher fertility than Hindus and more importantly the gap has been increasing since the 1950s (see also their paper in this issue): Mishra s (2004) analysis of the two NFHS also showed similar Muslim-Hindu differences in fertility and use of contraception. In a parallel study Dharmalingam and Morgan (2004) analysed the NFHS-I data using multi-level models. They showed that the Muslim-Hindu fertility differential was pervasive across India: in all major states with the exception of Madhya Pradesh an average Muslim woman was in general more likely to want to have another child for a given parity, and she was less likely to use contraception given that she did not want to have any more children. In this paper we update and compare the study with Dharmalingam and Morgan (2004) by using the NFHS-II data for Our analysis complements the pooled data analysis of Bhat and Zavier (2004) and Mishra (2004). The fertility and contraceptive measures used here are slightly different from the ones used in Bhat and Zavier (2004) and Mishra (2004); we also employ multilevel logistic regressions to model the desire for an additional child and current contraceptive use. Religion and Reproduction Substantive literature suggests that Muslim-Hindu differences in observed fertility might not be real (this section draws on Morgan et al 2002; Dharmalingam and Morgan 2004). If this is true, Muslim-Hindu fertility differences will disappear once Muslim-Hindu differences in socio-economic factors are taken into account. Some scholars have argued that there may not be any substantial differences in the Muslim and Hindu religious texts with respect to values or injunction on the timing and frequency of union formation and dissolution and contraception [Iyer 2002a]. In fact the observed fertility differentials before the 1970s were not due to Muslim-Hindu differences in polygamy or contraception but largely due to the differences in the duration of sexual abstinence following childbirth [United Nations 1961; Mandelbaum 1974; Visaria 1974; Srinivas and Ramaswamy 1977; Bhat and Zavier 2004]. Perhaps what matters for a higher Muslim fertility is how Islam is interpreted and adhered to. There is overwhelming evidence to show that the interpretation and implementation of values relating to marriage, divorce and contraception vary considerably Economic and Political Weekly January 29,

2 between countries that are predominantly Muslim [Coulson and Hinchcliffe 1978; Youssef 1978; Obermeyer 1992; Musallam 1983]. In India there are scholars who argue that Islam is not against birth control [Quraishi 2000; Enginneer 1994; Mahmood 1977; Khan 1979; also see Qureshi 1989; Omran 1992; Obermeyer 1992]. In Morgan et al s (2002) study of several settings in a number of Asian countries their Muslim respondents did not consistently report that contraception was against their religion. But as Bhat and Zavier (2004) argue while there may be differences in the way Islam is interpreted, what matters for group differences in fertility is whose or which interpretation is followed by the vast majority of Muslims. For example, Iran experienced a substantial decline in fertility following the favourable stand of the religious clergy there [Hoodfar and Assadpour 2000]. Bhat and Zavier (2004) believe that Indian Muslims seem to follow the interpretations that oppose contraception, in particular sterilisation a method that is widely and rigorously promoted and used in India. It thus appears that there is no consensus on how much influence Islam has on marriage and contraception in India. In addition to directly influencing fertility through the proximate determinants, religion can also impact fertility indirectly through socio-economic factors. Religion can differentially affect or be related to the distribution of individual and group characteristics which then influence reproductive behaviour [Jeffery and Jeffery 2002]. For instance, it is argued that the higher fertility among Muslims is due to their higher likelihood of illiteracy or low education, poverty, lower women s autonomy and nondiscrimination of daughters. There are at least two recent studies that support this line of argument [Jeffery and Jeffery 1997, 2002; Iyer 2002a; also see Alagarajan 2003]. Their micro studies from Uttar Pradesh and Karnataka showed that net of socio-economic differences there is very little fertility differential that can be explained by religion per se: the Muslim-Hindu fertility difference was related to such socio-economic characteristics as education, age, infrastructure, son-preference, female extended family, the change in social norms over time, and how members of different religions respond demographically to these influences [Iyer 2002b:34]. Similarly, Jeffery and Jeffery (1997:225) suggested that the Muslim-Hindu differences in fertility rates are the result of underlying differences in region, residence, class and schooling. In contrast, Morgan et al s (2002) comparative study across a number of Asian countries including India showed that religion had a net effect on fertility: Muslim women (compared to non-muslim referent groups) were more likely to desire another child and less likely to use contraception even after controlling for several major socio-economic characteristics. Similar findings were observed in the analysis of survey and census data for some Indian states [e g for Gujarat: Visaria and Jain 1995; for Karnataka: Rao, Kulkarni and Rayappa 1986; for Kerala: Bhat and Rajan 1990, Alagarajan 2003; Alagarajan and Kulkarni 2003; for Greater Bombay: Rele and Kanitkar 1976], and for India as a whole [Ramesh et al 1996; Bhat 1996; Dreze and Murthi 2001; Reddy 2003; Bhat and Zavier 2004; Dharmalingam and Morgan 2004; Mishra 2004]. It is widely believed that Muslim-Hindu differences in fertility behaviour are related to the lower autonomy of Muslim women rooted in patriarchy. For instance, Jack Caldwell (1986) proposed that the lack of autonomy, independence and socio-economic security among women was causally related to the higher than expected infant and child mortality in the predominantly Muslim Arab nations. This argument can be extended to their high fertility as well [Kirk 1968]. High fertility can provide protection and enable women to access family resources [Mason and Taj 1987; Cain, Khanam and Nahar 1979]. However, this view is challenged by the secular decline in fertility in many predominantly Muslim countries to lower levels (e g, Bangladesh, Indonesia, Malaysia and Iran). Morgan et al (2002) used data that allowed for direct measurement of women s autonomy across a number of settings in Asia. They found no links between women s autonomy and fertility among Muslims or non-muslims. Bhat and Zavier (2004) analysed the Indian National Family Health Survey II data on women s autonomy and son preference. Although Hindu women tend to have slightly more autonomy than their Muslim counterparts, son preference is stronger among Hindus than among Muslims. This led them to conclude that we should not expect these factors to be that important in explaining the differential fertility of Hindus and Muslims [Bhat and Zavier 2004:101]. Other explanations centre on religion as symbolising membership in minority and majority groups. Goldscheider and Uhlenberg 1969 [also see Chamie 1977] argue that groups that have access to social institutions for upward social mobility are likely to limit fertility as one strategy for such mobility. Access to education and to subsequent appropriate employment and remuneration motivates families to limit fertility and invest more heavily in fewer children. In contrast, groups that are denied or perceived to be denied such access to education and employment were much less likely to achieve a small family size. In other words, in the face of substantial obstacles to upward mobility the costs of children were less relevant. In the Indian context, Muslims may have more children and may be less willing to use contraceptives because they see less opportunity for upward mobility and less trustful of institutions that may provide for insurance or security. Dharmalingam and Morgan (2004; also see Morgan et al 2002) suggest an alternative or complementary dynamic that stresses intergroup dynamics and the role of family and demographic behaviour as markers of group identity and sources of solidarity. The pervasive religious tension in Indian political discourse and the currency of demographic arguments increase the plausibility Table 1: Women Who Were Currently Married, Not in Menopause, Aged 15-39, and Had At Least One Living Child, NFHS-II States Sample Size Women Who Wanted Another Child Muslims Per Cent HindusTotal Percentage of Percentage of Muslims Hindus South India Tamil Nadu Kerala Andhra Pradesh Karnataka West India Maharashtra Gujarat North India Bihar Uttar Pradesh Madhya Pradesh Rajasthan Jammu and Kashmir East India West Bengal Assam Economic and Political Weekly January 29, 2005

3 of this argument [Jeffery and Jeffery 1997, 2005; Basu 1996, 1997]. Moreover, Islam s opposition to specific contraceptive methods (e g, sterilisation) may limit their use among Muslims especially if the family planning programmes push terminal method [Basu 1996; Jeffery and Jeffrey 2005] and the government is Hindu controlled. Morgan et al (2002:534) say in the context of south Asia: To a great extent, religion is politics in this region of the world. Conflict between groups, drawn across explicitly religious lines, is central to south Asian cultures, and it is also relevant in southeast Asia. The position of women and women s roles in reproduction occupy central positions in religious, popular, and political discourse. It may be conjectured that the Muslim community response to these concerns, especially when they are a minority, is a stronger emphasis on aspects of family life conducive to childbearing or opposition to particular forms of birth control, for example sterilisation. Larger family sizes, greater demand for children, and higher risks of unwanted pregnancies could be the unintended outcomes of these group struggles. Table 2: Definition and Classification of Variables Used in the Analysis Variable Response Categories Description of Variable Want another child Yes, would like another child Whether the respondent would like to have another child (29.6 per cent); no, does not want any more or would prefer not to have any more children. children (70.4 per cent) Contraceptive use Yes (54.3 per cent); no (45.7 per cent) Whether the respondent or her husband was currently using any contraceptive method, including any modern or traditional methods. Age of women Younger than 30 years (57.2 per cent); Age of women in completed years at the time of survey; years (42.8 per cent) coded into two groups. Woman s education Illiterate (53.2 per cent); literate but not completed Highest level of education the woman had attained. middle school (20.5 per cent); middle school complete (10 per cent); high school complete and above (16.3 per cent) Woman s occupation Professional, managerial, technical and clerical workers The kind of work the women did most of the time. (2.6 per cent); sales and services (1.7 per cent); farming (23.1 per cent); skilled and unskilled manual (7.2 per cent); housework (65.2 per cent) Exposure to mass media High exposure (5.1 per cent); moderate exposure Exposure to mass was derived from the responses to three questions: (24.8 per cent); low exposure (70.1per cent) whether the woman listened to the radio at least once a week, whether she watched television at least once a week, and whether she went to a movie theater at least once a month. If a woman responded yes to all three questions, she was coded as having high exposure; if she responded yes to any two items, she was coded as having moderate exposure; and if she responded yes to either one item or no items, she was coded as having low exposure. Infant death None (78.8 per cent); one or more (21.2 per cent) Whether any of a woman s live births had died by the survey date. Woman s Autonomy Low (50.2 per cent); moderate (43.5 per cent); This has been constructed from the reported decision-making power of High (6.3 per cent) woman with respect to (a) what to cook; (b) obtaining healthcare; (c) purchase jewellery; (d) respondent staying with family; (e) how money will be spent; and freedom of movement with respect to (a) go to market; and (b) visit relatives. In the case of decision-making, the codes used are: own decision=2; joint decision=1; others=0. In the case of freedom of movement, the codes are: yes=2; no=0. The total autonomy scores range from 0 to 16. This has been classified as low (0-5), moderate (6-11) and high (12-16). Husband s occupation Professional, managerial, technical and clerical The kind of work that the husbands did most of the time. workers (12 per cent); sales and service workers (15.6 per cent); farming (37.2 per cent); skilled and unskilled manual (33.2 per cent); not working (1.9 per cent) Caste Scheduled caste (17.6 per cent); scheduled tribe Caste of the respondent. (12.8 per cent); others (69.6 per cent) Household economic status Poor (29.9 per cent);middle class (48.1 per cent); A composite index of household economic status or standard of living Rich (21.9 per cent) index was constructed by combining house type, toilet facility, source of lighting, fuel for cooking, source of drinking water, separate room for cooking, ownership of house, ownership of agricultural land, ownership of irrigated land, ownership of livestock and ownership of durable goods like car, television, telephone, etc. This index scores range from 0-14 for low economic status, for medium economic status and for high economic status. For details see IIPS (2000). Residence Urban (26.0 per cent); rural (74.0 per cent) Whether the household was in a rural or an urban area at the time of the survey. District-level economic status Poor (26.8 per cent);middle (37.2 per cent); This variable is based on the proportion of households that were poor in rich (36.0 per cent) a district. We derived three broad groups: those in which less than 20 per cent of their respondents households were poor (rich), per cent were poor (middle), or over 40 per cent were poor (poor). Proportion of Muslims at the Low (66.6 per cent); medium (15.2 per cent); This variable is based on the proportion of respondents in a district who district level high (18.2 per cent) were Muslims. We derived three broad groups of districts: those with less than 10 per cent Muslim respondents (low), per cent Muslims respondents (medium), or more than 20 per cent Muslim respondents (high). Note: The computations in this table are based on women who were currently married, not in menopause, aged 15-39, and had at least one living child at the time of survey. The figures in brackets in the second column are the percentages of respondents who gave the particular response for India as a whole. Economic and Political Weekly January 29,

4 In short, Muslims may desire more children and eschew sterilisation because this is what good Muslims do. These views and decisions mark and reaffirm fundamental identities. In the following sections we present our empirical results from the National Family Health Survey II. In the last section, we return to consider the various explanations presented above. Data We used data from the Indian National Family Health Survey II (NFHS-II) that was carried out across all the states in Although the survey covered women aged up to 49 years, our analysis included only those aged below 40 with at least one living child. As our main focus is on the reproductive behaviour of Muslim and Hindu women, we excluded women of other religious backgrounds. Our analysis focuses on only 13 major states with at least 5 per cent Muslim respondents in the sample. The sample size and Muslim percentages in the sample varied by state (Table 1). Only in Jammu and Kashmir were the majority of the respondents Muslims; in Assam and Kerala close to onethird were Muslims; one in five respondents were Muslims in West Bengal and about 15 per cent were Muslims in three states (Maharastra, Bihar and Uttar Pradesh); and in the remaining six states the percentage of Muslim respondents varied between 6 and 13. Analysis of the desire for additional child is based on the sample sizes presented in Table 1. In most states a greater proportion of Muslim women than Hindus want to have additional children. The largest Muslim-Hindu gap can be observed in Kerala; there is no difference in the states of Tamil Nadu and Gujarat; and in Uttar Pradesh and Madhya Pradesh marginally more Hindu women desire additional children. There are also variations among Muslims and Hindus across India. The highest proportion wanting to have additional children is observed in Bihar for both the Muslims and Hindus. But the lowest proportion among Muslims that intends to have an additional child is seen in Tamil Nadu, and that among the Hindus in West Bengal. The analysis of contraceptive use is based on a subsample of women who did not want any more children than they already had. Methodology Following the model of sequential decision-making, we measure fertility-relevant decisions as the intent to have an additional child, given the number of surviving children [Namboodiri 1972, 1983]. This mode of conceptualisation is based on the understanding that fertility decisions are conditional on the stage in the family-building process. Moreover, where fertility is controlled, stated intention has strong predictive power. The analysis thus focuses on Muslim-Hindu differences in: (i) whether a woman wanted to have another child (yes or no), and (ii) given that a women did not want another child whether she used any contraceptives, traditional or modern (yes or no). We employed logistic regressions to investigate both the desire for additional child and the current use of contraceptive. The basic logistic model had only religion as the covariate (Model 1 in Tables 3 and 4). Then, to investigate the influence of a number of confounding socio-economic factors, we progressively added additional covariates. This is reflected in the results for Models 1 and 2 in Table 3 and for Models 1 to 5 in Table 4. In Models 2 to 5 we added the independent variables presented in Table 2: Table 3: Logistic Regression Estimates of the Relationship between Religion and the Intention to Have Another Child, NFHS-II Two Living Children Three or More Living Children Total Muslims Model 1 Model 2 Total Muslims Model 1 Model 2 Sample Sample South India Tamil Nadu * 3.96** * Kerala ** 40.11** ** 8.05 Andhra Pradesh ** 3.90** ** 4.75* Karnataka ** 3.35** West India Maharashtra ** 3.03** ** 2.55** Gujarat North India Bihar ** 2.39** ** 1.43* Uttar Pradesh * Madhya Pradesh Rajasthan ** 2.33** Jammu and Kashmir East India West Bengal ** 5.31** * 1.75 Assam ** 2.82** ** 3.75** India ** 2.65** ** 1.41** Notes: Model 1 is a bivariate model with only one explanatory variable: religion. Model 2 includes control variables: women s age, education, occupation, exposure to mass media, experience of child death; women s autonomy; caste; husband s occupation; household economic status; place of residence; district level economic status; and proportion of Muslims at the district level. ** p<0.01; *p<0.05; p<0.10 Table 4: Logistic Regression Estimates of the Relationship between Religion and Current Contraceptive Use, NFHS-II Total Model 1 Model 2 Model 3 Model 4 Model 5 Sample South India Tamil Nadu Kerala ** 0.56* Andhra Pradesh Karnataka ** 0.52** 0.53** 0.44* 0.47** West India Maharashtra ** 0.48** 0.48** 0.53** 0.53** Gujarat North India Bihar ** 0.22** 0.22** 0.19** 0.19** Uttar Pradesh ** 0.60** 0.58** 0.50** 0.53** Madhya Pradesh * 0.68* 0.66* 0.49** 0.50** Rajasthan ** 0.39** 0.39** 0.31** 0.32** Jammu and Kashmir ** ** East India West Bengal ** 0.61** 0.59** 0.58** 0.53** Assam ** 0.62** 0.66** 0.70* 0.77 India ** 0.60** 0.61** 0.58** 0.59** Notes: Analysis is based on women who were currently married, not in menopause, aged 15-39, had at least one living child and not intend to have additional children. Model 1 is a bivariate model with only one explanatory variable: religion. Model 2 includes control variables: women s age, education, occupation, exposure to mass media, experience of child death; Model 3 includes all control variables in Model 2, plus women s autonomy; Model 4 includes all control variables in Model 3, plus caste, husband s occupation; household economic status; Model 5 includes all control variables in Model 4 plus place of residence; district level economic status; and proportion of Muslims at the district level. ** p<0.01; *p<0.05; p< Economic and Political Weekly January 29, 2005

5 individual characteristics of the respondents (age, education, occupation, a measure of exposure to mass media, and an infant s death), husband s occupation and household economic status, measures of women s autonomy and caste, and two district level measures (proportion of poor households and proportion of Muslim households in a district). Although the district level measures were based on the NFHS-II sample population, we believe that they provide reasonable estimates in the absence of more reliable district level information. To conserve space we present the bivariate (first) and the final model that includes all covariates in the subsequent tables. As in NFHS-I survey, NFHS-II also used a multistage cluster sampling approach for data collection. This sampling design produced a hierarchical data structure. For India as a whole, the data set had four levels of hierarchy (or aggregation): individual women (or households) at Level 1 are clustered within villages at Level 2 (2280 PSUs), which are clustered within districts at Level 3 (344 districts), which are then clustered within states at Level 4 (13 states). The 13 states included in the analysis constituted the all-india sample in this study. To obtain statistically efficient estimates, we had to account for the data s clustered and hierarchical nature [Brk and Raudenbush 1992; Cook and Weisberg 1982; Goldstein 1995; Longford 1993]. The random effects models that we used not only take into account the intracluster correlation arising from a multistage cluster sampling design but also control for unobserved factors that vary across these levels of aggregation and allowed us to assess the pervasiveness of Muslim-Hindu fertility differences [Amin et al 1997; Curtis, Diamond and McDonald 1993; Goldstein 1995; Sastry 1996; Steele, Diamond and Wang 1996]. The random effect logistic regression models that we estimated have four levels: individual women at Level 1, villages at Level 2, districts at Level 3, and states at Level 4. These models enabled us to decompose a parameter estimate into two parts: fixed and random. In other words, we could estimate the magnitude of the religion effect that is common (or fixed) throughout India and the magnitude that is variable at the state, district and village level [for details see Dharmalingam and Morgan 2004]. We used the multilevel software package MLwiN to estimate the fixed and random effects [Rasbash et al 2000]. To fit the random-effects logistic regression models we used Markov Chain Monte Carlo (MCMC) estimation using Gibbs sampling [Goldstein 1995]. Intent to Have an Additional Child In Table 3 we present the estimated effects in odds ratios that reflect the Muslim-Hindu differences in the intent to have another child for 13 major states in India. The first column panel gives the odds ratios for those with two living children, and the second column panel gives the same for those with three or more living children. Model 1 included only one covariate, religion. It provides the estimated gross Muslim-Hindu difference. But Model 2 included all the socio-economic and demographic covariates including religion and thus provides the net Muslim-Hindu differences in the intent to have another child. Looking at the result for those with two living children, it is clear that all the states with the exception of Madhya Pradesh have odds ratios that are substantially greater than one. In fact, the ratios are greater than two in nine states. This pattern indicates that Muslim women with two living children in all these states are more likely to desire to have another child compared to their Hindu counterparts. For example, in Maharashtra Muslim women with two living children are over three times as likely as Hindu women to want to have another child. In terms of broader geographical regions, it is interesting that the net Muslim-Hindu difference is greater in southern and eastern India than elsewhere the odds ratios in southern and eastern Indian states are greater than three. For India as a whole, a Muslim woman with two living children is on average about 2.65 times more likely to intend another child than is a Hindu woman. Moreover, the estimated gross and net Muslim-Hindu differences are of very similar magnitude. In fact, for many states the net effect is greater than the gross effect. This indicates that the Muslim-Hindu difference in the intent to have another child is independent of the differentials that may arise from the differences in socio-economic characteristics. For those with three or more living children the results are very similar to the ones observed for those with two more living children. For India as a whole, although Muslim women are more likely to intend to have another child, it was only by 41 per cent (compared to 2.65 times among women with two living children). Another difference is that for two states, Gujarat and Karnataka, there is no Muslim-Hindu difference in the desire for additional children. In fact, although the result is not statistically significant, Muslim women with three or more living children in Gujarat are less likely than their Hindu counterparts to intend another child. While the net Muslim-Hindu difference among women with two living children was not statistically significant only in three states (Gujarat, Madhya Pradesh and Jammu and Kashmir), the results were not statistically significant in seven states among women with three or more living children. Interestingly, three of the four south Indian states showed a significant Muslim- Hindu difference in the intention to have another child. Contraceptive Use In Table 4 we present results on the use of contraceptives among those not intending to have additional children. Model 1 shows the bivariate (or gross) effect of religion. In Models 2 to 5 we introduce additional covariates. Indian Muslim women who do not intend to have any more additional children are on average only 59 per cent as likely as the Hindu women to use contraceptives. This effect is very robust to inclusion of covariates in Models 1 to 5 (Table 4). Muslim women are less likely to use contraceptives in almost all states except Tamil Nadu where there is no Muslim-Hindu difference in contraceptive use. For two other states, Andhra Pradesh and Gujarat, although Muslim women are less likely to use contraceptives, the likelihood is not statistically significant. Thus as with the analysis of the desire for additional children, Muslim- Hindu difference in contraceptive use is pervasive across India. The Pervasiveness of Religious Differences So far Muslim-Hindu differences in the desire for additional children and contraceptive use were modelled and estimated for the 13 major states separately. We also controlled for a number of covariates at individual, household and district levels. To examine the pervasiveness of Muslim-Hindu differences we now present the results from multilevel logistic models for India as a whole [see the Methodology Section, and also Dharmalingam and Morgan 2004]. Multilevel models provide the estimated Economic and Political Weekly January 29,

6 effect of religion for India as a whole (known as fixed effect) and also estimates of variation in this effect (known as random effects) at three different levels of aggregations the state, district and village or PSU. The multilevel model estimates for religion are given in Table 5. Although we have presented the estimates for intercept and religion, we describe the estimates for religion as this is our focus. The fixed effect of religion on the intention to have an additional child indicates that a Muslim woman is on average 44 per cent more likely to want to have another child than a Hindu woman. When this is compared with the odds ratio in Table 3, it is clear that when the multilevel context is taken into account the net effect of religion on the intention for additional child is reduced but the magnitude is still substantial and statistically significant. The values of more importance for our purposes are the random effects for religion at the state and district level. At the state level the random effect coefficient was only 1.06 and statistically not significant. And at the district level the coefficient was This indicates that the effect of religion varies on average very little among states and districts. At the village (or PSU) level the random effect for religion was small (estimated odds ratio of 1.16) but it was statistically significant. The statistical significance is likely to have resulted from the large sample size for village level aggregation (n=2280) compared to the number of units of analysis at the state (n=13) and district levels (n=344). A comparison with the result from the analysis of NFHS-I shows that between and , the net effect of religion has declined by about one-third (1.44/2.23) (see Table 5 here, and Table 5 in Dharmalingam and Morgan 2004:541). However, the random effects at both the state and district levels in NFHS-I and NFHS-II are very small and statistically not significant. This suggests that Muslim-Hindu differences in the desire for additional children are pervasive across India, irrespective state or district level aggregation. The second panel in Table 5 provides the multilevel model estimates for use of contraceptive among those not intending to Table 5: Multilevel Model Estimates (Odd Ratio) for Contraceptive Use and the Desire for an Additional Child, NFHS-II Variables MCMC Estimates for the MCMC Estimates for Desire for an Additional Child Contraceptive Use Effect Significance Effect Significance Religion (fixed effect) Hindu Muslim 1.44 ** 0.53 Random effects at the state ** level (level 4) Religion Intercept Random effect at the district level (level 3) Religion Intercept 1.17 ** 1.11 ** Random effects at the village or PSU level (level 2) Religion 1.16 ** 1.20 ** Intercept ** Notes: Control variable included in the model are: women s age, education, occupation, exposure to mass media, experience of child death, women s autonomy, caste, husband s occupation, household economic status, place of residence; district level economic status; and proportion of Muslims at the district level. ** p<.01 have another child. For India as a whole, a Muslim woman was on average only 0.53 times as likely as her Hindu counterpart to use contraceptive, given that she did not want any more children. Again, as with the desire for additional children, the effect of religion does not vary at the state and district levels. The random effect coefficients were not large (1.14 at the state level and 1.24 at the district level) and statistically not significant. Although the random effect was significant at the village level (1.20) it was not large either. Muslim-Hindu difference in contraceptive has also declined. In NFHS-I the fixed effect of religion on contraceptive use was 0.44; but in NFHS-II this effect has improved to Thus Muslim-Hindu difference in contraceptive use has narrowed by a factor of about 20 per cent (0.53/0.44) (for NFHS-I figures see Table 5 in Dharmalingam and Morgan 2004) between and Thus the multilevel model estimates from NFHS-II lead us to conclusions similar to that we arrived at from the analysis of NFHS-I data: that Muslim-Hindu differences in the desire for additional child and use of contraceptives are invariant across states and districts. Discussion Our study does not give us great leverage to discriminate between possible interpretations of these patterns. Rather our demographic description identifies what needs to be explained [Dharmalingam and Morgan 2004:529-30]. In general, we document pervasive Muslim-Hindu differences in the intent to have additional children. Among women with two or more children, Muslim women (compared to Hindu ones) are much more likely to intend an additional child. Muslim women are also less likely to use contraception given an intention for no more children. These results are remarkably pervasive and clearly visible in the and data. However, there is evidence that the religious differences weakened over this decade. More specifically, our analysis of second National Family Health Survey has shown that Muslim-Hindu differences in fertility and contraceptive use are substantial and pervasive across India. A Muslim woman is about 50 per cent more likely to want to have an additional child, and is about half as likely to use contraceptives as a Hindu woman. This is net of observable socioeconomic characteristics at individual, household and district levels, and after taking into account the multilevel structure of data and reality. The findings from NFHS-II are consistent with that from NFHS-I that was carried out six years earlier in [Dharmalingam and Morgan 2004]. Not only did the proportions of women intending to have additional children among both Hindus and Muslims have declined between and , the Muslim-Hindu differences have also narrowed. Comparison of our multilevel model results for NFHS-I and NFHS-II showed that the greater likelihood of Muslim women to desire an additional child has decreased by about 20 per cent, and their likelihood to use contraceptive has increased by a similar magnitude. Additionally, the Muslim-Hindu differences in the desire for an additional child and in the use of contraceptives vary little across states and districts, signifying the pervasiveness of the Muslim-Hindu differences in the fertility behaviour. The conclusions we arrived at from the analysis of NFHS-I (Dharmalingam and Morgan 2004) hold for NFHS-II as well. 434 Economic and Political Weekly January 29, 2005

7 Muslim-Hindu differences in fertility behaviour are real, not due to differences in socio-economic characteristics between Muslims and Hindus in India. This difference is not due the Muslim-Hindu differences in women s autonomy either because Muslim and Hindu women do not differ substantially in their autonomy or because the influence of differences in autonomy on fertility is minimal [Morgan et al 2002; Iyer 2002b; Bhat and Zavier 2004]. We included individual level measures of women s autonomy in our models here; their inclusion did not reduce the magnitude of the effect of religion. The narrowing of the Muslim-Hindu differences between and shows assuming that this narrowing signifies the beginning of a long-term trend that whatever is the nature and magnitude of the influence of Islam on fertility behaviour, the influence seems to be changing to the evolving socio-economic and political contexts. To restate our conclusion arrived at elsewhere [Dharmalingam and Morgan 2004; also see Morgan et al 2002], a more plausible explanation for the substantial but narrowing Muslim-Hindu differences in fertility behaviour is likely to lie in minority-majority group dynamics. Specifically, the source of these differences lies in discrimination against the minority group or in the role of demographic behaviour as markers of group identity. The intrinsic messages of Islam on reproduction may have assumed a pro-natalist character in practice given the historical Muslim-Hindu animosities and the growth of right-wing Hindu nationalist political movements particularly since Indian independence. However, as our findings from the two National Family Health Surveys indicate the pronatalist reproductive behaviour of Indian Muslims might be converging toward Hindu behaviour. Morgan et al (2002) stressed that these differentials are indicators of socially constructed institutions and/or identities and are subject to change. They say: these religious and political effects favourable to higher levels of childbearing are nonetheless susceptible to change. The linking of largely ascribed characteristics to behavioural outcomes does not imply perpetual group differences. (W)e noted that Islam is open to various interpretations regarding the roles of women and the acceptability of contraception. The dominant Muslim position on these issues is not immutable, nor is the saliency of such differences in political debates. The way these forces manifest themselves may vary sharply between countries or across regions within countries where Muslims are minority members and where they are not (in addition) globalisation of the economy and of communication also raises the possibility of strong pan-islamic influences and responses. Post hoc we speculate that these demographic behaviour (higher family size desire and less contraceptive use) are becoming less salient markers of Indian Muslim identity. This hypothesis suggests that they would be emphasised less by religious leaders in the context of declining Muslim fertility worldwide perhaps due to the wider availability of non-terminal methods of birth control. This might already be happening in the Indian context. Reacting to the controversy over the recent census figures on Muslim growth rate, the Muslim Law board emphasised that the strength of Muslims is not in quantity but in quality, and that the only way forward is modernisation (Hindustan Times 2004; Indian Express 2004). The Board also made positive references to the low fertility and high contraceptive use in Iran. Our post hoc argument also suggests that either other new markers are replacing these demographic ones or that other existing markers are becoming more relevant for Muslim/Hindu identity. Comparative micro and macro research in multiple settings are needed to fully comprehend the dynamics of construction of Muslim/Hindu identity in India. EPW Address for correspondence: nava@cds.ac.in dharma@waikato.ac.nz pmorgan@soc.duke.edu References Alagarajan, M (2003): An Analysis of Fertility Differentials by Religion in Kerala State: A Test of the Interaction Hypothesis, Population Research and Policy Review 22, pp Alagarajan, M and P M Kulkarni (1998): Fertility Differentials by Religion in Kerala: A Period Parity Progression Ratio Analysis, Demography India 27(1): pp Amin, S, I Diamond and F Steele (1997): Contraception and Religiosity in Bangladesh in The Continuing Demographic Transition, edited by G W Jones, R M Douglas, J C Caldwell and R M D Souza, Clarendon Press, Oxford. Basu, A M (1996): The Demographics of Religious Fundamentalism in Unravelling the Nation: Sectarian Conflict and India s Secular Identity, (ed), K Basu and S Subrahmanyam, Penguin Books India, New Delhi. (1997): The Politicisation of Fertility to Achieve non-demographic Objectives, Population Studies 51, pp Bhat, P N M (1996): Contours of Fertility Decline in India: A District Level Study Based on the 1991 Census in Population Policy and Reproductive Health, (ed), K Srinivasan Hindustan, Delhi. Bhat, P N M and A J F Zavier (2004): Religion in Demographic Transition: The Case of Indian Muslims in Demographic Change, Health Inequality and Human Development in India, (ed), S I Rajan and K J James, Centre for Economic and Social Studies, Hyderabad. Bhat, P N M and S Irudaya Rajan (1990): Demographic Transition in Kerala revisited, Economic and Political Weekly 25(35 and 36): pp Bryk, A S and S W Raudenbush (1993): Hierarchical Linear Models, Sage, Newbury Park. Cain, M, S R Khanam and S Nahar (1979): Class, Patriarchy, and Women s Work in Bangladesh, Population and Development Review 5, pp Caldwell, J C (1986): Routes to Low Mortality in Poor Countries, Population and Development Review 12, pp Chamie, J (1977): Religious Differentials in Fertility: Lebanon, 1971, Population Studies 31(2): pp Chandrasekar, S (n d): India s Population Problems and Policies: Qualifying the Right to Reproduce? Cook, R D and S Weisberg (1982): Residuals and Influence in Regression, Chapman and Hall, London. Coulson, N and D Hinchcliffe (1978): Women and Law Reform in Contemporary Islam in Women in the Muslim World, (ed), L Beck and N Keddie, Harvard University Press, Cambridge, pp Curtis, S L, I Diamond and J W McDonald (1993): Birth Intervals and Family Effects on Postneonatal Mortality in Brazil, Demography 30(1): pp Dharmalingam, A and S Philip Morgan (2004): Pervasive Muslim-Hindu Fertility Differences in India, Demography 41(3): pp Dreze, J and M Murthi (2001): Fertility, Education and Development: Evidence from India, Population and Development Review, 27(1), pp Engineeer, A A (1994): Status of Muslim Women, Economic and Political Weekly, 29, pp Goldscheider, C and P R Uhlenberg (1969): Minority Group Status and Fertility, American Journal of Sociology, 74: pp Goldstein, H (1995): Multilevel Statistical Models, 2nd edition, Arnold, London. Hindustan Times (2004): Muslim Law Board Ready to Discuss Family Planning, September 13. Hoodfar, H and S Assadpour (2000): The Politics of Population Policy in the Islamic Republic of Iran, Studies in Family Planning 31(1): pp Indian Express (2004): Reason for Reform, editorial, September 15. International Institute for Population Sciences (IIPS) (2000): India: National Family Health Survey, , International Institute for Population Sciences, Mumbai, India. Economic and Political Weekly January 29,

8 Iyer, S (2002a): Demography and Religion in India, Oxford University Press, New Delhi. (2002b): Understanding Religion and the Economics of Fertility in India, Centre of South Asia Studies Occasional Paper 2, Centre of South Asia Studies, Cambridge. Jeffery, R and P Jeffery (1997): Population, Gender and Politics, Cambridge University Press, Cambridge. (2000): Religion and Fertility in India, Economic and Political Weekly 35: pp (2002): A Population Out of Control: Myths about Muslim Fertility in Contemporary India, World Development 30, pp (2005): Saffron Demography, The Common Wisdom, Aspirations and Uneven Governmentalities, Economic and Political Weekly, this issue. Khan, M E (1979): Family Planning among Muslims in India. Manohar, Delhi. Kirk, D (1968): Factors Affecting Moslem Natality in Population and Society: A Textbook of Readings, (ed), C B Nam, Houghton Mifflin Company, Boston, pp Kulkarni, P M (1996): Differentials in the Population Growth of Hindus and Muslims in India, , Monograph Series No 1, Population Foundation of India, New Delhi. Longford, N T (1993): Random Coefficient Models, Clarendon Press, Oxford. Mahmood, T (1977): Family Planning: The Muslim Viewpoint Vikas, New Delhi. Mandelbaum, D G (1974): Human Fertility in India: Social Components and Policy Perspectives, University of California, Berkeley. Mishra, V (2004): Muslim/non-Muslim Differentials in Fertility and Family Planning in India, East-West Center Working Papers: Population and Health Series, No 112, East-West Center, Hawaii, Honolulu. Morgan, S P, S Stash, H Smith and K O Mason (2002): Muslim and non- Muslim Differences in Female Autonomy and Fertility: Evidence from Four Asian Countries, Population and Development Review, 28, pp Musallam, B F (1983): Sex and Society in Islam: Birth Control before the 19th Century, Cambridge University Press, Cambridge. Namboodiri, N K (1972): Some Observations on the Economic Framework for Fertility Analysis, Population Studies 26, pp (1983): Sequential Fertility Decision-Making and the Life Course, In Determinants of Fertility in Developing Countries, vol 2, (ed), R Bulatao and R D Lee, Academic Press, New York. Obermeyer, C M (1994a): Reproductive Choice in Islam: Gender and State in Iran and Tunisia, Studies in Family Planning 25, pp (1994b): Religious Doctrine, State Ideology, and Reproductive Options in Islam in Power and Decision: The Social Control of Reproduction, (ed), G Sen and R C Snow, Harvard University, Centre for Population and Development Studies, Massachusetts, Cambridge, Harvard University, School of Public Health, Department of Population and International Health, Massachusetts, Boston, pp Omran, A R (1992): Family Planning in the Legacy of Islam, Routledge, London, New York. Qureshi, S (1989): Islam and Development: The Zia Regime in Pakistan World Development, 8, pp Quraishi, S Y (2000): Leave Your Heirs Rich, Indian Express, May 17. Ramesh, B M, S C Gulati and R D Retherford (1996): Contraceptive Use in India, , National Family Health Survey Subject Report No 2, International Institute for Population Sciences, Mumbai. Rao, N B, P M Kulkarni and P Hanumantha Rayappa (1986): Determinants of Fertility Decline: A Study of Rural Karnataka. South Asia, New Delhi. Rasbash, J, W Browne, M Healy, B Cameron and C Charlton (2000): MLwiN BETA Version , Multilevel Models Project, Institute of Education, London. Reddy, P H (2003): Religion, Population Growth, Fertility and Family Planning Practice in India, Economic and Political Weekly, 38, pp Rele, J R and T Kanitkar (1976): Fertility Differentials by Variables in The Economic and Social Support for High Fertility, (ed) by L T Ruzicka, Australian National University, Canberra. Sastry, N (1996): Community Characteristics, Individual and Household Attributes and Child Survival in Brazil, Demography 33(2), pp Srinivas, M N and E A Ramaswamy (1977): Culture and Human Fertility in India, Oxford University Press for Family Planning Foundation, Delhi. Steele, F, I Diamond and D Wang (1996): The Determinants of the Duration of Contraceptive Use in China: A Multilevel Multinomial Discrete- Hazards Modelling Approach, Demography 33(1), pp United Nations (1961): Mysore Population Study, United Nations Department of Social and Economic Affairs, New York. Visaria, L (1974): Religious Differentials in Fertility in Population in India s Development, , (ed), A Bose et al Vikas, Delhi. Visaria, P, LVisaria and A K Jain (1995): Contraceptive Use and Fertility in India: A Case Study of Gujarat, Sage, New Delhi. Youssef, N H (1978): The Status and Fertility Patterns of Muslim Women in Women in the Muslim World, (ed), L Beck and N Keddie, Harvard University Press, Cambridge, pp Economic and Political Weekly January 29, 2005