Muslim and Hindu Women’s Public and Private Behaviors: Gender, Family and Communalized Politics in India

Abstract

Prior research on fundamentalist religious movements has focused attention on the complicated relationship between gender, family and religion. Using data from a nationally representative survey of 30,000 Hindu and Muslim women, this study compares the daily public and private behaviors of women in India to examine how gender and family norms are shaped in the context of communalized identity politics. Building on the theoretical framework of “doing gender”, it argues that because communal identities are expressed through externally visible behaviors, greater religious differences are expected in external markers of gendered behaviors and family norms. Results indicate that Muslim women are more likely to engage in veiling and less likely to venture outside the home for recreation and employment. However, religious differences are absent when attention is directed at private behaviors such as household decision making power, gender segregation within households, and discrimination against daughters. Results underscore the multidimensionality of gender.

Introduction

Although demographers have long recognized the role of culture in shaping demographic behaviors like marriage, childbearing and intra-household relationships, in spite of its tremendous promise (Fricke 2003), incorporating culture in demographic analysis is fraught with challenges (Hammel 1990). Much of this challenge comes from the fact that unlike geography, culture is constantly changing and evolving while “the use of `culture’ in demography seems mired in structural-functional concepts that are about 40 years old, hardening rapidly, and showing every sign of fossilization” (Hammel 1990 p. 456). The role of religion and gender in shaping demographic outcomes is a particularly striking example. The terms `religion’ and `women’s status’ have both been used in the literature as shorthand for a complex set of constructs without adequate attention to identifying the parts that are relevant to the demographic discourse. While this omission is problematic when it comes to treating religion and gender in isolation, it becomes particularly salient when the two conflate as in the research on Muslim societies and gender.

In one of the earliest demographic arguments to link religion, women’s status and demographic outcomes, Caldwell (1986, P. 175) noted, “The central aspect of the relationship between Islam and mortality levels is undoubtedly the separate and distinctive position of women operating partly through their access to education but also in many other ways.” Since then a variety of studies have struggled to reconcile the relationship between religion and women’s status on the one hand and fertility(Morgan et al. 2002), mortality (Ghuman 2003; Kuhn 2010) or labor force participation (Spierings, Smits and Verloo 2009) on the other.

However, in spite of the increasing recognition in the literature that gender is a multidimensional phenomenon (Mason 1995; Presser and Sen 2000), this insight has not been integrated in the research on gender and demography. Similarly, in recent years it has become clear that it is not cultural norms propagated by religion per se, but rather the political economy of religion and its salience to individual lives that is relevant (McQuillan 2004), but once again demographic literature tends to use religion as a categorical variable instead of seeing its manifestation as being contingent on political economic conditions. This omission is particularly troublesome when gender is incorporated into the religion-demography nexus. A focus on religious norms tends to overlook the agency of women who are the primary actors in this demographic drama and is counter to the past two decades of research on gender and demography (Riley 1997).

In recent years, literature on gender and religion has been challenged by trends that do not conform to the monolithic conceptualizations of the relationship between the two. The dominant discourse on religion and gender argues that religious fundamentalism or politicized religion, particularly conservative versions of Islam, thrive on emphasizing different rights and obligations of men and women and are often complicit in women’s oppression (Inglehart and Norris 2003) and in demographic behaviors that may disempower women such as lower female labor force participation. However, this discourse frequently fails to recognize that women are often willing participants in fundamentalist religious movements (Bacchetta 2004; Blaydes and Linzer 2008; Bedi 2006; Mahmood 2001; Sarkar and Butalia 1995). We suggest that it would be difficult to explain women’s acquiescence to and in some cases willing participation in religious movements that may disempower them without taking into account the political conditions under which they operate.

This brief description raises a host of questions. Does religion shape gender relations in a society? If so, which aspects of gender are most affected? Is this relationship affected by the political economy of religion in a society? How does this interaction shape demographic outcomes such as sex-selective mortality and women’s labor force participation? These are some of the questions we seek to address through our focus on the ways in which religion and politics intertwine to shape Muslim women’s lives in India.

The outcomes of interest to our analysis include subjective reports of interpersonal behaviors such as practicing veiling and intra-household decision-making power, and measurable demographic outcomes such as gender differences in child mortality and female labor force participation. A vast literature indicates that regardless of religion, Indian women are subjected to patriarchal structures with unequal access to education, income, control over resources, and power in household decision-making (Agarwal 1994; Jain and Banerjee 1985; Jejeebhoy and Sathar 2001; Wadley 1994). However, the distinctive expression of gender is often shaped by a variety of historical forces (Avishai 2008; Chatterjee1989; Kandiyoti 1988; Ray and Korteweg 1999). We argue that the increasing politicization of gender in service to communal forces plays an important role in the experience of Indian Muslim women and shapes demographic behaviors as well as behaviors that are precursors to demographic outcomes. Although our discussion in this paper focuses on Muslim women, it is important to note that right-wing Hindu movements also tend to shape gendered behaviors and ideologies (Sarkar and Butalia 1995).

Much of the literature in this area focuses on women living in predominantly Muslim countries (Hussain 2010; Moghadam 2002) and on Muslim immigrant communities in Western countries (Killan 2006; Korteweg 2008; Meer, Dwyer and Modood 2010) for which there are no sizable comparison groups. India is an interesting site for research on the intersection of gender, religion and group identity formation. Although India is home to the second largest Muslim population in the world, estimated at about 150 million, Muslims are a minority that coexists with a Hindu majority and a plethora of smaller religious communities. Most importantly, the rising tide of fundamentalist Hinduism has targeted gender as an arena in which to fight communal battles (Mankekar 1997; Sunder Rajan 2000). Using data from a nationally representative survey of 41,554 households conducted in 2004–2005, we compare the public and the private lives of Hindu and Muslim women to examine gender norms in the context of politicized communal identities.

Theoretical Framework

Our theoretical approach draws upon two strands of literature - gender performance and the use of gender in the politics of identity creation. In the ensuing paragraphs, we first describe the theoretical underpinnings of each strand of literature and subsequently we discuss their implications in the Indian context.

Gender Performance

The first strand of literature has its roots in symbolic interactionism and focuses on the way in which gender is displayed or enacted through the day-to-day actions of individuals (Butler 1990; Connell 1987; Goffman 1976; Ridgeway and Smith-Lovin 1999; West 1996; West and Fenstermaker 1995; West and Zimmerman 1987). Goffman (1976) first articulated this perspective in his discussion of “gender display” in which he argued that men and women engage in a highly stylized mode of interaction where presence or absence of symmetry can indicate deference or dominance. This concept was further elaborated in a highly influential paper by West and Zimmerman (1987), titled “Doing Gender,” which argues that gender is not something one is but something one does and individuals create gender through day-to-day interactions. In a subsequent article titled “Doing Difference,” West and Fenstermaker (1995) expand their arguments about the enactment of gender to the enactment of differences based on a variety of forms of inequality, particularly class and race.

In spite of the initial promise, this approach has drawn considerable criticism from long-term practitioners working in the area of race, class and gender. Collins acerbically points out (Collins 1995, p. 493), “recasting racism, patriarchy and class exploitation solely in social constructionist terms reduces class, race and gender to performances, interactions between people embedded in a never ending stream of equivalent relations, all containing race, class and gender in some form, but a chain of equivalences, devoid of power relations.” The issue of power and the political economy within which it is exercised is central to the controversy surrounding Muslim women and veiling in the Western world (Bloul 1998; Fournier and Yurdakul 2006) where veiling is constructed as both a marker and a stake for collective identities.

Gender and Identity Creation

The second strand of literature that we draw upon deals with gender and the politics of identity formation. Building on pioneering work by Valentine Moghadam (1994), a number of scholars suggest that women’s continued - and sometimes reinforced - engagement in culturally distinctive gendered practices, such as veiling, segregation, and tightly controlled marriage patterns, is part of the process of group identity formation (Amrita Basu 1998; Bloul 1998; Chowdhry 2007; Jeffrey and Basu 1998). In this process, some aspects of gendered norms and behaviors become vehicles through which communal identities are articulated and women are singled out as the symbolic repository of group identity (Z. Hasan 1994; Moallem 2005; van Wichelen 2010).

Following these arguments, we suggest that the prevalence of culturally specific gender behaviors neither connotes a desire to subjugate women in all areas of life nor reflects women’s own sense of subordination (Shirazi 2001). A focus on gendered behavior as a vehicle for identity creation leads us to distinguish between different aspects of gender – specifically, behaviors that are publicly visible and relevant to expressing a separate and distinct identity, and behaviors that are located inside the home and shielded from prying eyes. This split between public and private is far from perfect and in some cases, public adherence to gender norms allows women to extract private concessions from the family allowing for greater freedom to study or participate in the labor force (Kariapper 2009; MacLeod 1992; Read and Bartkowski 2000).

Distinctions Between the Public and The Private

As noted above, a number of scholars have suggested that demographic behaviors such as fertility, mortality and women’s labor force participation differ between Muslim and non-Muslim communities and have suggested that women’s roles and status form the mechanism through which this relationship operates. However, in one of the few rigorous tests of this relationship, women’s status was not important as a mechanism through which religious differences in fertility operate (Morgan et al. 2002), mainly because differences in women’s autonomy between Hindus and Muslims in the same community were modest (also Jejeebhoy and Sathar 2001).

We suggest that this observation may be due to failure to theorize about the underlying relationships between religion, gender and demography. Demographers as well as gender scholars have made a persuasive argument that gender is a multi-dimensional construct (Kabeer 1999; Malhotra, Schuler and Boender 2002; Mason 1986; Mason 1995). If we accept this argument, there is no reason to expect that religion will affect all dimensions of this multifaceted construct in a similar fashion. In this paper we argue that the political economy of religion in India is such that it is important to distinguish between gendered behaviors that are public and gendered behaviors that are private. For example, discrimination against daughters, culminating in higher mortality for girls takes place within the household through subtle behaviors that are not easily visible to the outsiders and may be subject to different forces than women’s labor force participation which is easily visible. Thus, distinguishing between different aspects of gender may give us a better analytical leverage than undifferentiated emphasis on women’s autonomy.

Gender Performance: The Indian Discourse

The discussion of gender performance in the global literature reviewed above has an analog in Indian literature. In most parts of South Asia, regardless of religion, women’s seclusion from the public gaze is seen as a sign of superior social status (Balk 1997; Mandelbaum 1988; Papanek 1979; Srinivas 1977). This seclusion is established and maintained in many different ways. Physical shielding of one’s face, or veiling, is but one instance. Veiling includes using a sari or dupatta to cover one’s face (known as ghunghat among north Indian Hindus), or wearing a burqa among Muslims. Staying away from public places such as bazaars or movie theaters, not venturing outside the home unless accompanied, and not participating in the labor force are some other means of maintaining seclusion (Derné 1994; Mann1994).

An ethnographic account of Muslims in Alighar notes (Mann 1994):

In Alighar, a Muslim woman embodies the honour of her patrilineage … her behaviour can be a guide to the degree of honour that male members of her household maintain. Yet, she does not merely reflect household honour, a woman is the household honour (ghar ki izzat). Consequently it is not only important to maintain control over her behaviour and deportment, it is essential that access to people or circumstances whereby honour may be insulted or compromised is limited. (P. 132–33)

An account of interviews with Hindu men similarly notes (Derné 1994):

Shame [lajia] and shyness [sharma] are the ornaments of Indian women. These are the most important ornaments. What are gold and silver? Thieves can steal them. Shame and modesty are the only things that can't be stolen … and the meaning of this [shame] is to use the veil. (P. 209)

Two forces, however, lead to the breakdown of this ideal of feminine virtues. First, poor women have never had the freedom to remain secluded within their households (Sharma 1980). Second, as literature from India and elsewhere documents, with increasing incorporation into the global culture, performance of modernity often triumphs over performance of gender as a way of attaining social status (Srinivas 1977). Muslims of India, however, have also faced an additional threat from the rise of Hindu fundamentalism over the past three decades.

Gender and Identity Politics in India

Religious conflicts are not new to India. However, the rise of Hindu fundamentalist political parties like the Shiv Sena or the Bharatiya Janta Party over the past three decades represents a new turn in the politics of Independent India (Brass 1990; Frankel et al. 2000). In opposition to secular Nehruvian politics, these parties often seize control of key issues to articulate their political platform while appealing to Hindu nationalism (Hasan 2001). The plight of Muslim women has emerged as one of the key issues around which the Hindu fundamentalist platform has coalesced (Z. Hasan 1998; Pathak and Sunder Rajan 1989; Rastogi 2007).

In a 1985 case involving Shah Bano, a separated Muslim woman, the Supreme Court decreed that her husband pay maintenance. The husband argued that under Muslim personal law, his responsibility extended only to the sum initially agreed upon in the marriage contract signed several decades earlier, and to maintenance for three months of the traditional iddat period. The supreme court judgment, containing inflammatory language about a unified civil code, led to vociferous protests from some segments of the Muslim community, resulting in a legislation paradoxically titled Muslim Women (Protection of Rights on Divorce) Bill, which effectively shifted the responsibility for financial support of divorced women to their natal kin or the state welfare board (Pathak and Sunder Rajan 1989).

This process allowed the Hindu right to portray itself as the protector of Muslim women in its demands for a unified civil law, code words for abolishing Muslim personal law while ignoring the gendered injustices inherent in other legal systems, including the civil code (Engineer 1987). The reaction of the Muslim community to this co-option reveals the genuine dilemma in which many Muslim women activists find themselves. On the one side stand champions of Islamic virtues who represent the Muslim orthodoxy, and on the other side stand representatives of Hindu orthodoxy who are ready to use Muslim women’s interests to bolster their quest for hegemonic Hindu power.

The Shah Bano case is but one in a series of incidents in which gender has formed the nucleus of the construction of communal rhetoric (Amrita Basu 1998). Many other flashpoints in the Indian political process have added to these tensions. For example, in the aftermath of the Mumbai riots against Muslims that took place in December 1992 and January 1993, in which a largely Hindu police force stood on the sidelines tacitly encouraging the rioters, Muslim women’s groups found themselves cancelling anti-domestic violence programs for fear of providing additional ammunition to the police to harass Muslim men (Agnes 1994).

Construction of Gender Within a Communalized Polity

As we try to understand the impact of this political tug of war on the lives of Indian Muslim women, other episodes in Indian history, such as dissension around the age at marriage (Chatterjee 1989) and the practice of widow burning or sati (Mani 1990) offer interesting parallels.¹ In each instance, political forces construct a modest and pious notion of womanhood in which women’s engagement with obvious markers of identity, such as the veil or widow burning, come to represent a distinctive community identity. A striking example of this is documented in an incident in which a well-known Muslim actress, activist, and parliamentarian, Ms. Shabana Azmi, while receiving the prestigious International Gandhi Peace Prize noted that modesty does not demand covering women’s face. Her comments resulted in a sharp rebuke by prominent Muslim clerics (Outlook India 2006).

It would be simplistic to say that this view represents all of India’s Muslims; Muslim intelligentsia and grassroots women’s organizations represent a far more liberal view of gender (Z. Hasan 1994). However, their power vis-à-vis the more conservative sections of Muslim society has been severely undermined by historical events, such as riots in Gujarat in 2002, putting the issue of survival at the front and center of any discussions within the Muslim community, precluding attention to gender reforms (Agnes 2002). Moreover, because sexualized violence against women formed the core of this terror (Sarkar 2002), gender became the site of struggle and a vehicle through which community identity was constructed by both sides. With rising communalism, religious identities have sharpened, resulting in the politicization of gender identities (Z. Hasan 1994) and the co-option of women’s bodies as symbols and repositories of community and national identity (Chhachhi 1994).

Theoretical Framework

In trying to understand the role of gender performance and identity creation in the lives of Indian Muslim women, our theoretical arguments rest on two planks. First, we argue that women’s external behavior, clothing, and demeanor are used by women themselves, their husbands, families, and communities as symbols to communicate a host of meanings ranging from social status to distinctive communal identities. Second, these external symbols may not correspond with the internal dynamics of the household and power of women within their families, or lack thereof. These assumptions build on the growing literature documenting the multidimensional nature of gender relations (Collins et al. 1993; Narayan 2006; Presser and Sen 2000), which argues that women’s empowerment in one domain is not necessarily synonymous with empowerment in another domain.

In this instance, the orthogonal nature of external symbolism and internal household dynamics provides us with an interesting analytical handle for exploring gender norms in the context of a communalized polity. If increasing levels of communalism create the climate within which Muslim women become symbols of community identity, we are likely to observe more distinctive gender practices in public and less distinctive gender practices in arenas that are shielded from the external gaze.

Hence we address the following questions:

1. How strong are religious differences in different aspects of gendered behavior? Do they persist after controlling for background factors such as education and broader historical and contextual influences associated with place of residence?

2. Should we expect religion to shape public and private gendered behaviors in a similar fashion? If our arguments regarding specific types of gender performance as markers of Muslim identity creation hold, we should see religious differences in externally visible behaviors but not necessarily in internal household dynamics.

3. Are these gendered behaviors primarily symbolic in nature or do the processes that shape symbolic behaviors also shape demographic outcomes such as sex differentials in mortality and female labor force participation?

India Human Development (IHDS)

To examine differences in symbolic gendered behaviors between Hindus and Muslims, we use individual and household data collected from the multi-topic India Human Development Survey 2005 (IHDS). The IHDS is a nationally representative survey that includes 41,554 households in 1,503 villages and 971 urban neighborhoods (Desai, Vanneman and National Council of Applied Economic Research, 2005). It offers rich information on health, education, employment, economic status, marriage, fertility, social capital, and gender relations (Desai et al. 2010). Response rate for IHDS is about 92%. For more details about sample design and response rates see Desai et al. (2010). Female interviewers conducted face-to-face interviews with the respondents in local languages and asked them about a variety of issues related to gender roles and behaviors.

The portion of the survey that is of greatest interest to this paper comes from a module administered to 30,422 ever-married women aged 15–49. Given our focus on comparing Hindu and Muslim women, we have limited our analyses to these two groups, omitting women from minority religions, such as Sikhism, Christianity, Jainism, and tribal religions - about 6 percent of the sample.

This survey offers a unique opportunity to compare the lives of Hindu and Muslim women in a pan-Indian context while controlling for other differences between them. The Sachar Committee, appointed by the Government of India, documented considerable religious segmentation in Indian society (Government of India 2006). Table 1, based on the IHDS, corroborates this. Muslim women are more likely to live in urban areas, less likely to be educated, and more likely to come from families with low levels of education and have slightly higher fertility. Muslim families are also concentrated in the states of Jammu and Kashmir, Uttar Pradesh, Assam, and West Bengal (Government of India 2006). Consequently, it is important to control for place of residence and family’s socioeconomic conditions to rule out spurious influences. Sample size limitations do not allow us to explore the influence of state of residence and religion simultaneously, but the inclusion of state specific dummy variables implies a fixed-effects model within state in which state influences are controlled. The IHDS survey contains extensive information about family and household characteristics, allowing for these controls. Although there is considerable diversity both within the Hindu community between different castes and within the Muslim community, we combine all Hindus and all Muslims in this paper. Data limitations do not allow us to fully distinguish between different groups of Muslims. Although broad distinctions between Hindu groups is empirically possible, our analyses controlling for broad caste categories among Hindus did not suggest that any single category was responsible for the Hindu-Muslim differences we document. Indeed, for outcomes where religion matters, Muslims are different from every single Hindu caste category. Hence, for simplicity of presentation, we merge all caste categories among Hindus.

Table 1.

Socioeconomic Characteristics of Hindu and Muslim Ever-Married Women Age 15–49

Variable	Hindu	Muslim
Not practicing purdah/ghunghat	0.47	0.16
Go out with their husband for outings	0.52	0.40
Men and women eat together	0.48	0.51
No. of items for which respondent is the primary decision maker (max=5)	1.45	1.43
Participation in the labor force	0.56	0.41
Participation in wage work	0.27	0.11
Woman's Education (Completed Grade)	4.18	3.19
Woman's Education
No education	0.47	0.54
1–4 standards	0.08	0.10
5–9 standards	0.28	0.25
10–11 standards	0.09	0.06
Higher secondary & some college	0.04	0.02
College graduate	0.04	0.02
Age of the women	32.84	32.33
Not Currently Married	0.04	0.03
Husband not in the Household	0.08	0.09
Husband's Education	6.01	4.59
Husband Employed in Salaried Job	0.21	0.17
Household Size	5.49	6.15
No. of Married Women in the Household	1.32	1.30
No. of children living with respondent	2.27	2.91
Family owns or cultivates land	0.47	0.35
Log of total household income	10.15	10.13
Proportion of households with negative income	0.02	0.02
Log of family income excluding women's wages	9.96	10.02
Proportion of households with negative income (excluding women's earnings)	0.03	0.02
Place of Residence
Metro city	0.08	0.08
Other urban	0.20	0.29
More developed village	0.35	0.25
Less developed village	0.37	0.38
Sample Size	26,688	3,913

Our theoretical discussion distinguishes between aspects of gender performance that are visible to the outside world (public or external markers) and those that are located within the household (private or internal markers) and hence are not useful as external symbols. The bulk of research on gender norms has been qualitative, and we build on this literature by proposing quantitative measures of the public-private dichotomy of gender practices. We operationalize external markers of gendered behaviors by focusing on whether women practice veiling and whether women go on outings with their husbands. In order ensure symmetry of results, we have recoded variable on veiling to not practicing veiling. Thus a positive coefficient on both absence of veiling and going on outings reflects greater public visibility. We capture internal markers of gender norms with two measures: whether women eat with male family members and whether women respondents are primary decision makers for a variety of household decisions.

Our selection of external markers is grounded in literature suggesting that in the context of religious fundamentalism or politicized religion, women’s visibility in the public sphere may be reduced by the combination of physical separation through veiling and the withdrawal of women from public life (including waged work) to the private sphere (Moghadam 1994), although this relationship is not consistent across societies and social classes (Amin 1997; Bahramitash 2004; Gerami and Lehnerer 2001). Our selection of internal markers of gendered behaviors - namely, a focus on household decision making and eating meals together - is based on literature that suggests that women’s power in the household is strongly reflected in their active participation in household decision making and equality in day-to-day household behaviors (Narayan 2006; Presser and Sen 2000). Decision making as a marker of gender relations has been used in a variety of studies but a focus on eating together is new in this analysis. In many Indian families, gender and generational hierarchies are enacted by maintaining a distance between older male family members and younger daughters-in-law. This involves the separation of physical space as well as eating arrangements in which women may serve the food but it would be inappropriate for them to sit down with older male members to eat. We use this indicator to identify male-female segregation in the household, just as veiling or purdah is used to identify male-female segregation in public spaces. In addition to focusing on markers of gender that have tremendous personal significance to individuals and communities, we also focus on two outcomes that have received considerable attention in the demographic literature as having direct implications for women’s well-being: women’s participation in the labor force and sex differences in child survival. Descriptive statistics for markers of gender performance are provided in Table 2.

Table 2.

Gendered Outcomes for Hindu and Muslim Women

	Hindu	Muslim	Indicator
Ever-married women aged 15–49
Not practicing purdah/ghunghat	0.47	0.16	External
Go out with their husband for outings	0.52	0.40	External
Men and women eat together	0.48	0.51	Internal
No. of items for which respondent is the primary decision maker (max=5)	1.45	1.43	Internal
Participation in the labor force	0.56	0.41	External
Participation in wage work	0.27	0.11	External
Probability of Dying Between Age 1 and 5 (births taking place 5–15 years before survey interview)			Internal
Boys	0.014	0.016
Girls	0.022	0.017

We analyze different dependent variables using different techniques. We use logistic regression to analyze categorical variables with only two categories, such as engaging in veiling, going on family outings, or male and female members of the household eating together. The decision making index is a continuous index ranging from 0 to 5 and is analyzed with multivariate regression. Items in the index include: what to cook; number of children to have; and the purchase of expensive items. Additionally, the index contains two questions about taking children to the doctor and children’s marriage, thus it is restricted to women with children.

Results: Public And Private Symbolic Behaviors

Results from multivariate analyses are presented in Tables 3 to 6. In each analysis, we control for urban residence, state of residence, woman’s age, marital status and education, household income² and landownership, household structure, and husband’s characteristics. For tables 3 and 5, the bottom two rows indicate predicted outcomes for Hindus and Muslims holding all other variables at their mean value, allowing us to compare the magnitude of religious differences at the margin.

Table 3.

Impact of Selected Covariates on Markers of Gender Performance

	External				Internal
Variable	No Purdaha		Family Outinga		Men & Women Eat Togethera		Primary Decision Makerb
	Coef.	Z-statistic	Coef.	Z-statistic	Coef.	Z-statistic	Coef.	Z-Statistic
Muslim	−2.09	−23.01 **	−0.48	−8.45 **	0.35	5.77 **	0.00	0.03
No Husband in the Household	−0.38	−3.24 **	0.04	0.37	0.48	3.09 **	0.50	6.65 **
Not Married	0.82	4.18 **	−1.16	−6.50 **	−0.13	−0.63	1.42	13.64 **
Husband's Education	−0.02	−4.35 **	0.03	5.31 **	0.01	2.71 **	0.00	−0.90
Husband in Salaried Job	0.02	0.32	0.06	1.18	0.24	5.10 **	0.01	0.57
No. Persons in the Household	−0.08	−3.61 **	−0.06	−3.07 **	−0.10	−4.28 **	−0.05	−6.54 **
No. of Married Women in the
Household	0.01	0.10	0.06	0.99	−0.36	−5.18 **	−0.15	−5.47 **
Place of residence (omitted category: Metropolitan City)
Other urban	0.01	0.10	−0.44	−5.59 **	−0.65	−8.00 **	0.02	0.54
More developed village	−0.56	−6.03 **	−0.78	−8.89 **	−0.92	−10.12 **	−0.06	−1.50
Less developed village	−0.43	−4.62 **	−0.71	−7.93 **	−0.98	−10.60 **	−0.08	−2.20 *
Household cultivates land	−0.07	−1.41	−0.08	−1.59	0.06	1.25	−0.12	−4.12 **
Log of total income	0.08	2.96 **	0.19	7.54 **	0.01	0.57	−0.02	−1.66
Income less than 0	0.23	0.74	1.74	5.61 **	−0.13	−0.47	−0.07	−0.46
Age of the woman	0.02	5.86 **	−0.01	−5.19 **	0.00	0.92	0.01	11.86 **
No. of children living with respondent	0.03	0.98	0.03	1.19	0.03	1.17	0.05	4.35 **
Years of education of the woman	0.05	9.07 **	0.08	14.45 **	0.03	4.76 **	0.00	1.57
Constant	−0.31	−0.88	−0.77	−2.49 *	2.04	6.57 **	1.84	11.83 **
N	29,741		29,679		29,735		27,883
Chi-Sq (df=37)	6,586		2,856		4,046
R-Square							0.2705
Predicted probability of gendered behav ior
Hindu	0.49		0.55		0.50		1.45
Muslim	0.11		0.43		0.59		1.45

Note: All models include dummy variables for state of residence (not presented in the table). Predicte d probability of gendered behavior is from regressions holding all other variables at their mean values.

^aCoefficients are from logistic regression.

^bCoefficients are from linear regression.

^*p < .05;

^**p < 0.01

Table 6.

Within Household Fixed Effects Logit for Mortality Between Ages 1 and 5

	Coef.	Z-statistic
Daughter	0.55	350.77 **
No. of siblings alive at birth of child	−0.04	−32.07 **
Age of the child at interview (expected age for diseased child)	0.01	94.4 **
Daughter * Muslim	−0.73	−175.79 **
No. of women with more than one child borth 5–15 years prior to survey	19,305
No. of sibling groups used in analysis	491
Chi sq. (df=4)	233,770

Note: Analytic sample is limited to sibling groups in which at least one child died and one survived.

^*p < 0 .05;

^**p < 0.01.

Table 5.

Effect of Selected Covariates on Women's Employment

	All Work		Waged Work
Variable	Coef.	Z-Statistic	Coef.	Z-Statistic
Muslim	−0.57	−8.81 **	−0.83	−10.8 **
No Husband in the Household	−0.51	−4.18	−0.37	−2.807 **
Not Married	1.07	5.82 **	1.09	6.005 **
Husband’s Education	−0.04	−7.71 **	−0.06	−9.468 **
Husband in Salaried Job	−0.38	−6.89 **	−0.18	−2.504 *
No. Persons in the Household	−0.08	−3.81 **	0.02	0.992
No. of Married Women in the Household	−0.05	−0.80	−0.17	−2.3 *
Place of residence (omitted category: Metropolitan City)
Other urban	0.44	5.35 **	0.25	2.539 *
More developed village	1.47	16.55 **	0.97	9.289 **
Less developed village	1.68	18.39 **	1.14	10.6 **
Household cultivates land	1.15	21.97 **	−12.66	−12.10 **
Log of total income	−0.29	−9.65 **	−0.55	−16.75 **
Income less than 0	−2.20	−6.49 **	−4.87	−14.06 **
Age of the woman	0.03	11.02 **	0.02	5.853 **
No. of children living with respondent	0.21	8.34 **	0.05	1.71
Years of education of the woman	−0.05	−7.64 **	−0.05	−6.72 **
Constant	1.60	4.53 **	2.80	6.397 **
N	29,826	4.63	29,826	6.16
Chi-Sq (df=37)	4,690		2,914
Predicted Probability of Work Participation
Hindu	0.54		0.16
Muslim	0.40		0.08

Note: Results are from logistic regressions containing dummy variables for state of residence (not presented in table). Predicted probability is from regressions holding all other variables at their mean values.

^*p < 0.05;

^**p < 0.01.

Hindu-Muslim Differences in Gender Performance

Comparisons of Hindu and Muslim women on various markers of gender performance are presented in Table 3. The results show striking differences between Hindu and Muslim women in external markers of gender performance, but this relationship is absent for indicators of gender performance that are not easily visible to outsiders.

Adjusting for urban residence, state of residence, education, income, landownership, fertility, and other individual and household-level characteristics, these differences are even wider, with about 49 percent of Hindu women not practicing veiling compared with 11 percent for Muslim women with a comparable background³. Similarly, 55 percent of Hindu women go on family outings with their husband (and children), while 43 percent of the Muslim women do so. Both of these differences are large and statistically significant even after all the control variables are included.

However, when we focus on markers of gender performance that are not easily visible to outsiders, this relationship is absent. For example, Hindu and Muslim women’s score on the decision making index⁴ (reflecting the number of items for which women report being primary decision makers) is 1.45, indicating no religious difference. For men and women eating together, the findings show more gender integration among Muslim households; the predicted probability of eating together for Hindus is 0.50 compared to 0.59 for Muslims.

These results present an interesting picture. On behaviors that are visible to the outside world, Muslim women appear to be more likely to engage in the performance of segregation and modesty. On behaviors that are not visible to the external world, Muslim women’s lives are very similar to those of their Hindu sisters. Cross cousin and within family marriages are far more prevalent among Muslims than among Hindus -- 24% among Muslims as opposed to 10–12% among Hindus (Desai et al. 2010). In as much as within family marriages may be held accountable for greater gender equality in the household (Dyson and Moore 1983), this could explain greater intermingling of sexes within Muslim households than Hindu households.

Our analysis so far has controlled for socioeconomic background, but has assumed that upper- and lower-class families construct religious and communal identities similarly. However, as discussed earlier, upper-class families are beset by two competing demands. The ideology of “doing gender” (West and Zimmerman 1987) in the South Asian context demands that women present a modest and decorous façade to the world to buttress the claim by family/community to high culture; the ideology of “doing modernity” (Schein 1999) demands that the family present a liberal and progressive image to the world to establish their claims to a sophisticated and worldly outlook. In some cases, these contradictions are resolved by men donning the mask of modernity while women remain guardians of family virtue (Liechty 2003). In others, women’s external behaviors are altered in deference to a claim to modernity. Given the importance of gender performance for Muslims as a minority living in a communalized society, we expect that the balance would tilt in the favor of gender performance for Muslim women far more readily than that for Hindu women.

To examine this balance, we interact women’s education with religion in multivariate analyses. The results are presented in Table 4. The first finding of note is that the interaction between religion and education is significant and negative for the two outcomes that reflect visible performances of gender. However, this relationship is not statistically significant for markers of gender performance that are concentrated in the household.

Table 4.

Interaction between Religion and Education in Shaping Gendered Behaviors

Variable	No Purdaha	Family Outinga	Men & Women Eat Togethera	Primary Decision Makerb
Muslim	−1.613 **	−0.41 **	0.427 **	0.015
Education	0.060 **	0.09 **	0.029 **	0.005
Muslim * Education	−0.119 **	−0.02 *	−0.023	−0.004

Note: Models include all variables from Table 3.

^aCoefficients are from logistic regression.

^bCoefficients are from linear regression.

^*p < 0.05;

^**p < 0.01.

Higher education is associated with an increasing number of Hindu women abandoning the veil. Among educated Muslim women, however, the performance of gender triumphs over the performance of modernity: the practice of veiling increases with the level of education. For family outings, educated women are far more likely to go out with their husbands and families to the market or to a restaurant, but this increase is greater for Hindu women than for Muslim women. In contrast, for the two markers of gender on which Muslim women are on par with Hindu women, religious differences do not vary with education. In each case, however, educated women seem to experience lower gender segregation within the household and hold greater decision making power. These results parallel those of Jejeebhoy and Sathar (2001), who found slightly lower decision making, moderately lower freedom of movement, and somewhat greater freedom from domestic violence among Muslim women than among Hindu women.

Results: Demographic Outcomes

The preceding discussion has focused on symbolic or relational dimensions of women’s lives. However, it seems reasonable to ask whether symbolic aspects of gender have any implications for other dimensions of women’s lives. In the Indian context, two aspects of women’s day-to-day lives deserve particular attention: women’s participation in paid and unpaid work and higher than expected mortality among female children.

Indian women tend to have low rates of labor force participation (National Sample Survey Organization 2006). This low employment rate owes partly to pervasive underemployment in rural areas and difficulties in finding suitable employment, even for men. However, Muslim women’s low employment levels may reflect something more than a shared labor force disadvantage with their Hindu sisters (Das 2005; Lateef 1990). Some factors leading to low rates of labor force participation may well be structural (Das 2005). Surveys often do not capture home-based work very well. Because Muslim women often engage in artisanal activities, labor force surveys may omit this work. Education may also shape labor force participation, and the low educational level of Muslim women may account for their lower labor force participation. However, if our arguments hold — that is, if the role of public gender performance is particularly important for Muslim women — then Muslim women should be less likely to be in the labor force even after these structural factors are taken into account.

Discrimination against daughters is another aspect of vital importance to women’s lives, but one that is located within the household. India is among a small number of countries in which the sex ratio among children is skewed in favor of boys (Klasen and Wink 2003). This disadvantage is a function of higher mortality for girls than for boys, although sex-selective abortion has also played a role in recent times (Arnold, Kishor and Roy 2002). However, the mechanisms through which daughters come to be disadvantaged are diverse, resulting in considerable speculation among researchers about whether female survival disadvantage owes to discrimination in food intake or to discrimination in access to medical care (Alaka Basu 1989). The myriad day-to-day actions that create this disadvantage are rarely visible to outside observers, and if communal identities are created through externally visible behaviors, hidden discrimination against daughters does not serve this purpose. Hence, we suggest that there should be little difference between Hindu and Muslim girls in female survival disadvantage.

Religious Differences in Labor Force Participation

Analyses of female labor force participation in developing countries face a variety of methodological challenges and form the subject of a large body of literature (Donahoe 1999; Jain and Banerjee 1985). The definition of labor force participation (i.e., whether only waged work is counted or whether women’s work on family farms and family businesses is included) tremendously influences the proportion of women deemed to be in the labor force (Sathar and Desai 2000). Differences in labor force participation between Hindu and Muslim women may be particularly sensitive to these definitions (Das 2005). Hence, in this analysis, we focus on two definitions of labor force participation. The first measure of labor force participation is highly expansive and includes paid labor as well as unpaid work on family farms, in family businesses, and caring for livestock. The second relies only on waged work — that is, work as a daily farm laborer or manual laborer, or a monthly salaried worker who receives income in cash or in kind. We expect that if a focus on gender performance reveals lower labor force participation among Muslim women than among comparable Hindu women, this effect will be greater for waged work where women may be more likely to come in contact with Hindu employers or coworkers.

Results from a logistic regression with these two measures of labor force participation are presented in Table 5. The control variables in this analysis are more or less identical to those in the analysis reported in Table 3, with one major exception. Instead of controlling for total family income, we control for household income excluding women’s own wage income. Because income is endogenous to employment decisions, only the income generated by other household members is relevant to this analysis.

The results in Table 5 show that with all other factors held constant, Muslim women are less likely to participate in the labor force, regardless of which labor force participation variable we consider. However, the difference is greater for waged work than for the measure of overall labor force participation. The predicted probability of Muslim women’s participation in any type of work (waged work or work on family farms/business) is 0.40 compared to 0.54 for Hindu women. However, for waged work, the difference is two-fold with the probability of waged work being 0.08 for Muslim women and 0.16 for Hindu women.

It would be simplistic to attribute this difference in labor force participation to religious differences in preferences for seclusion and segregation, such as veiling. Although both Hindu and Muslim women who practice veiling are slightly less likely to be employed than women who do not, these differences are dwarfed by the religious differences within each category (results not reported here but available upon request). We suggest that labor force participation is part of a complex set of interactions between individuals, families, communities, and the environment in which structural forces and individual responses play a role (Ahmed-Ghosh 1994) where politicization of religious identities creates conditions under which women’s withdrawal takes place, as we discuss in conclusion, situation in Muslim majority Asian countries like Indonesia, Malaysia and Bangladesh is far more favorable to higher labor force participation among Muslim women.

Our arguments have revolved around increasing communalization in the Indian political space. However, our data are cross-sectional, precluding any discussion of trends. Women’s labor force participation is one arena in which historical trends are observable and support our contention. Sonya Rastogi (2007), in her research using data from National Sample Surveys between 1983 and 2000, found that with education, residence, and household characteristics held constant, the predicted probability of participation in wage labor for Hindu women increased slightly whereas that for Muslim women dropped, widening the gap between Hindu and Muslim women.

Religious Differences in Sex Differentials in Child Mortality

As mentioned earlier, higher than expected mortality among girls in India is well documented (Klasen and Wink 2003; Arnold, Kishor, and Roy 2002). Demographic research has suggested that neonatal mortality (i.e., mortality in the first month of life) and some of the subsequent early childhood mortality is attributable to birth trauma and unsafe delivery (National Academy 1997), something that may not be directly related to gender discrimination by parents. However, mortality between ages 1 and 5 is largely associated with malnutrition, exposure to disease, and lack of treatment (Mosley and Chen 1984). Hence, we compare excess mortality for girls aged 1–5 among Hindu and Muslim families to determine the extent of discrimination against daughters.

Because religion may also play an important role in determining access to health care for both boys and girls, resulting in a complicated relationship between religion, gender, and child survival (Guillot and Allendorf 2010), we control for all family- and community-level factors by using a fixed effects model that compares brothers and sisters within the same household. These models provide an estimate of differences in mortality between brothers and sisters from ages 1 to 5, and the interaction term between being Muslim and gender indicates the religious difference in this effect. Because family background factors do not differ between brothers and sisters, gender, child age (or age had s/he survived) and birth order are the only variables that enter the model. The dependent variable mortality between ages 1 and 5 is dichotomous; thus, the analysis relies on sibling groups in which there is some variation in the dependent variable, i.e. at least one child died and at least one survived, allowing for an analysis of the effect of the child’s gender on the probability of child mortality using STATA’s xtlogit command with fixed effects (Greene 1993).

The results, presented in Table 6, are striking. Whereas girls in Hindu families are 1.74 times as likely as their brothers to die between ages 1 and 5, Muslim girls are only about 0.83 times (1.74*0.48) as likely to die as their brothers. Both relationships are statistically significant. Research has documented that globally, in the absence of discrimination, boys are naturally more likely to succumb to illness than girls (Klasen and Wink 2003). Hence, higher female mortality among Hindu families is most likely attributable to the neglect of daughters. Although the neglect of daughters in India is well documented, only limited attention has been directed to the fact that the extent of this neglect among Hindu families is much greater than among Muslim families. Our results show that when it comes to caring for their daughters, Muslim families are far more egalitarian than Hindu families. Other statistics also provide hints of this phenomenon. For example, among 0- to 6-year-olds, the ratio of girls to boys is 986 per 1,000 among Muslims compared with 927 in India as a whole (Government of India 2006). These results complement other studies that have documented lower preferences for sons in Muslim families than in Hindu families (Bhat and Zavier 2003).

Alternative Explanations: Role of Discrimination and Violence

While our analysis has focused on the role of polarized identities in shaping differences in gendered behaviors and outcomes, an alternative explanation focusing on the role of discrimination and exclusion cannot be ruled out. Muslim women find it difficult to access public spaces because they may fear violence (Robinson 2010) and labor market discrimination may reduce their incentive to participate in the labor force. It is difficult for us to totally rule out this alternative explanation. But an examination of our data does not provide strong evidence to support it.

The IHDS asked questions about frequency of harassment experienced by young women in the local neighborhood or village. About 19 percent of Hindu women report that women experience some harassment in their neighborhood compared to 26 percent of Muslim women. However, controlling for frequency of harassment does not substantially change differences in Hindu and Muslim women’s public behaviors reported above. Similarly, when we examine the relationship between the labor market outcomes (wages) of Hindu and Muslim men in each state and state-specific differences in participation in paid work for Hindu and Muslim women, the regression line is nearly flat, indicating no relationship.

This does not imply that fear for Muslim women’s safety and worries about labor market discrimination may not play a role in observed differences in the behaviors of Hindu and Muslim women. This suggests that these concerns are not necessarily connected to local experiences but rather a generally pervasive concern that may affect all Muslim families in India, regardless of their immediate surroundings. If this is so, it fits in well with our argument that increased political tensions and associated sexualized violence against women creates an environment in which gender performance becomes a vehicle through which community identities and anxieties are expressed.

Discussion

Two sets of findings presented in this article are particularly noteworthy. First, the pervasive discourse about the disadvantages imposed on women by Islamic traditions remains unsubstantiated in the IHDS data. In our analyses, Muslim and Hindu women differ very little from each other on some dimensions of empowerment (e.g., household decision making power). On at least one dimension - sex differences in mortality - Muslim girls are far more advantaged than Hindu girls. Second, the one area in which Muslim women seem to follow a very different pattern of gendered behavior than their Hindu sisters is around publicly visible activities: compared with Hindu women, Muslim women are more likely to participate in veiling, less likely to go on family outings to places like fairs and movie theaters, and less likely to be employed. Accordingly, inasmuch as these gender performances are meant to assert a unique communal identity, they may also reinforce disempowering outcomes for women under the canopy of culture or tradition.

However, we argue that the public-private dichotomy in gender performance and the greater seclusion in publicly visible behaviors among Muslim women must be seen in the context of an increasingly communalized polity. In keeping with the growing literature on gender and identity politics (Jeffrey and Basu 1998; Moghadam 1994; Siapno 2002), this paper argues that under conditions where religion forms the axis of polarization, women often carry the burden of community identity. This results in a focus on symbols that define a modest and refined demeanor, project a unique and separate communal identity, and are visible to the outside, often hostile, world. Findings presented in this paper show that when it comes to these symbols — such as veiling, not going on family outings, and not participating in wage labor — Hindu and Muslim women differ substantially but there are few differences between them in areas that are not publicly visible.

As we begin to explore the implications of these observations, we stand on highly contentious grounds. We suspect that these results will provide some fodder to all viewpoints. Those who seek to establish that Islam, in itself, is not the vehicle through which oppression of women takes place will point to the remarkable similarities we found between Hindu and Muslim women in some regards and the considerably lower discrimination we observed in Muslim families than in Hindu families. Those who want to point to women’s exclusion from public spaces among Muslim communities and disempowering outcomes will also find evidence for their thesis in our results.

A more constructive approach, however, is to recognize that the functioning of gendered ideologies and behaviors is not solely a function of religion or religiosity. Historically specific contexts in which religious communities organize themselves determine specific expressions of gender within specific situations. Our results in Table 4 document that whereas education is associated with decline in veiling and greater performance of “modernity” for Hindu women, it is not the case for Muslims where performance of gender dominates performance of modernity. However, in other contexts other identities may triumph. For example, studies suggest that the process of national liberation in Bangladesh, where the “Bengali” identity was juxtaposed against the “Muslim” identity appropriated by the united Pakistani state, created a space in which Bangladeshi women’s movement could grow (Kabeer 1991). Some comparisons between India and Bangladesh are instructive. Between 1995 and 2002–2003, women’s labor force participation in Bangladesh grew sharply (World Bank 2008), while Muslim women’s labor force participation in India declined slightly (Rastogi 2007). Similarly only 84 percent of the young women aged 15–25 practice purdah in Bangladesh (Das 2010) compared with 94 percent of Muslim women in West Bengal and 75 percent of Hindu women in West Bengal recorded in IHDS. These comparisons should be treated cautiously because of small sample sizes. The experiences of Muslim-majority countries like Malaysia and Indonesia also provides interesting comparisons; the employment rate for women ages 15 and above in Indonesia has risen from 37 percent in 1980 to 51 percent in 2009; for the same years, women’s employment rate has risen from 40 percent to 46 percent in Malaysia (International Labor Organisation 2011). Malaysia is particularly interesting in that with a sizeable Hindu population, it is possible to compare gendered behaviors for the Muslim majority and Hindu minority population. Morgan et al. (2002) find that unlike in India, Muslim women in Malaysia have a small advantage over the Hindu women in autonomy.

A serious comparison of cross-national differences is beyond the scope of this paper and is hampered by the difficulties of finding comparable data. However, our observations highlight the challenges faced by the Indian Muslim community. A fragile reform movement among Indian Muslims is buffeted by the clash between Hindu fundamentalism on one side and Muslim orthodoxy on the other. It may be too much to expect that if Indian politics were to be less polarized along religious lines, gendered fault lines would disappear. However, it seems highly likely that a decrease in religious polarization would create a space in which women’s gendered interests would not have to be subordinated to the demands of communal identity.

This paper is one of the few empirical studies to combine two separate theoretical approaches – one focusing on day-to-day gender performances and the other dealing with politics of identity creation. Its findings have several important implications for methodological and substantive dimensions of the scholarship on gender and demography.

First our analysis detects theoretically meaningful differences in external behaviors of Hindu and Muslim women, but it also notes an overarching influence of patriarchy that affects both communities. For example, with other factors held constant, Hindu women are more likely than Muslim women to go on family outings with their husbands (55 percent versus 43 percent), but nearly half the women in each of these communities do not go out at all. This suggests that religion is only one axis along which social groups form a self-conscious identity.

Second, this paper expands the literature that emphasizes multidimensionality of gender (Narayan 2006; Presser and Sen 2000) by arguing that symbolic dimension of gender deserves greater attention than it has hitherto received in demographic studies. Although attention to the symbolic aspect of gender has a distinguished history within cultural studies literature (Mani 1990; Moghadam 1994; Mohanty 1991), quantitative studies of gender have found it difficult to incorporate, possibly owing to data limitations. For example, in an otherwise excellent article, Morgan et al. (2002) focus only on familial control and interpersonal relationships as they operationalize women’s autonomy. Gender influences not only in women’s relationships with their partners and other family members, but as we argue, it also operates to subordinate women’s welfare to that of group identity formation. So by focusing our attention on symbolic aspects of gender, we can expand the repertoire of explanatory variables available into demographers. In this paper, we have identified women’s labor force participation as a particular example, but a variety of demographic phenomena – particularly those surrounding marriage, non-marital childbearing and divorce – may lend themselves to this approach.

Third, at a broader level, these findings suggest that addition of symbolic behaviors may fruitfully expand a demographer’s toolkit in other arenas besides gender. A large number of demographic studies have struggled to establish relative importance of individual agency vis-à-vis social structures (Cleland and Wilson, 1987; Hammel, 1995). We suggest that a focus on symbolic behaviors may provide added leverage for including individual agency, where agency is not defined solely by narrow economic motives but is also not subsumed under an umbrella of social norms.

Appendix 1: Variable Definitions and Means

Variable	Description	Mean
Outcome Variables
No purdah/ghunghat	Proportion of women who do not veil	0.43
Mixed family outing	Whether respondent goes out alone with husband or with husband and family to the movies, restaurant, or market	0.50
Women and men eat together	Women eat main meal with men in household	0.48
No. of items for which respondent is the primary decision maker (max.=5)	Respondent has the most say in 5 areas of household decision making: (a) what to cook; (b) number of children to have; (c) purchase of expensive items; (d) treatment of sick child; and (e) whom children should marry	1.45
Participation in the labor force	Includes anytime of work, paid and unpaid, agricultural and nonagricultural, family owned and salaried/waged	0.54
Participation in waged work	Includes paid farm and non farm work and salaried/waged work	0.25
Mortality between age 1 and 4	Mortality between ages 1 and 5 for children born between 5 and 15 years prior to the interview	0.02
Independent Variables -- Individual
Level
Muslim	Hindu=0; Muslim=1	0.13
Not currently married	Current marital status of respondent is widowed/divorced/separated	0.04
Age of woman	Age at interview	32.78 (8.06)
Women’s education	Completed years of education	4.06 (4.6)
Husband not in the household	Either unmarried or has a migrant husband	0.04
Husband's education	Grades completed, 0 if no husband	5.83 (4.99)
Husband in a salaried job	Husband is employed in waged work where he receives monthly salary, 0 if no husband	0.21
Place of residence	Place of residence
	Metropolitan city (Omitted)	0.08
	Other urban	0.21
	More developed village	0.34
	Less developed village	0.37
Independent Variables -- Household
Level
State of residence	22 states included as dummy variables
Family owns or cultivates land	0=does not cultivate; 1=cultivator	0.45
Log of total household income	Household income from all sources	10.15 (1.63)
Proportion of households with negative income	Whether household has negative income, largely due to farm losses	0.02
Log of family income excluding women's wages	Household income excluding women's wage or salary earnings	9.97 (1.93)
Household size	No. of adults and children in the household	5.57
No. of children living with respondent	Includes girls and boys living with respondent	2.35
No. of married women in the Household	Proxy for extended household structure	1.32

Standard Deviations in parenthesis for continuous variables.