Women’s Agency and the Quality of Family Relationships in India

Abstract

The role of family context in determining women’s agency has been addressed through kinship patterns, household structure, and domestic violence. This study suggests that another aspect of family context – family relationship quality – can also influence women’s agency. Data from the Women’s Reproductive Histories Survey, collected in Madhya Pradesh, India, are used to examine whether family relationship quality is a determinant of women’s agency. Results show that women with higher quality relationships with husbands and parents-in-law do have greater agency. Further, family relationship quality is just as influential as other well known determinants of agency, including education and employment.

Women’s empowerment has received a great deal of attention as both a means and an end of international development. Education, micro-credit, and land rights have all been promoted as ways to help women become the beneficiaries of development. At the same time, these policies and programs are seen as ways to help women become more powerful agents of development. Empowered women appear better able to secure the well-being of themselves and their families. While the importance of women’s empowerment to development is increasingly accepted, however, the complexities of the links among empowerment and its determinants and consequences are still not well understood.

One element that has not been adequately addressed is the quality of family relationships. By quality, I refer to the range of companionate aspects of relationships, including love, affection, and support. Love and support found in high quality family relationships may be an important source of women’s empowerment. Higher quality relationships may create an environment where women are both more willing to assert themselves and family members above the woman in the customary family hierarchy are more willing to take her choices into account. In such a context, husbands, mother-in-laws, or other senior family members, may be better acquainted and concerned with her wishes and take young women’s preferences more seriously. Further, family members wanting to maintain a loving or at least amicable relationship may be a powerful incentive for them to give women greater scope to make decisions. Thus, this paper addresses the question of whether high quality family relationships are a source of women’s empowerment.

Before proceeding further, it is important to first make a brief statement about terminology. There are several terms for what is often the same, but blurry concept of women’s empowerment. As given by Kabeer (2001a) and reinforced in an influential review by Malhotra and Schuler (2005), empowerment is defined as an increase in women’s ability to make choices about their lives and environment. This definition focuses on women’s agency – women’s ability to exercise power by making choices – while also specifying that empowerment is a process of change where women gain agency over time. When using cross-sectional data, the term women’s autonomy is often used to refer to the same concept of agency found in women’s empowerment without the element of change over time (Jejeebhoy 2000; Malhotra and Schuler 2005). Some critique the use of the term autonomy because the accompanying emphasis on acting alone discounts the interdependence of family life and women’s own preferences to make choices with others (Kabeer 2001b; White 1992). Thus, from now on, I use the term women’s agency defined as women’s ability to make choices. I use agency to signal a focus on women’s ability to make choices without including an emphasis on change over time or women acting alone.

Women’s Agency and the Family

The quality of family relationships is an aspect of family context that is largely missing from work that explores the determinants of women’s agency. In some cases, family context in general is neglected, as seen in much of the research on intrahousehold allocation in economics. The early unitary model of household bargaining assumed that a household has a single utility function, implying that the household is functionally equivalent to an individual (Becker 1965). More recently, collective models open up the black box of the household and contend that individual’s preferences within a household can and do vary (Haddad 1997). However, these models base an individual’s bargaining power – or agency – largely, if not entirely, on an individual’s ability to be economically independent (Haddad 1997; Quisumbing and de la Briere 2000). Thus, a woman’s bargaining power is usually measured by her individual income and ownership of assets. Some newer models of intrahousehold allocation now include social norms, bargaining skills, and contextual factors, like divorce laws, as additional determinants of bargaining power (Agarwal 1997; McElroy and Horney 1981; Quisumbing 2003). However, to my knowledge, none of these models take into account how the quality of family relationships may shape who makes household decisions.

Kinship and Household Structure

Some anthropologists and sociologists incorporate family context by examining whether kinship structures affect women’s status and agency (e.g. Dyson and Moore 1983; Dube 1997; Mason 1998). Much of this work incorporates the family context as a determinant of women’s agency across communities. Thus, this work compares levels of autonomy in women in one area characterized by a particular kinship structure to women in another community within a different kinship structure. For instance, Dyson and Moore’s (1983) classic article contends that the difference in women’s agency between North and South India is largely determined by kinship structure. In the North, they suggest that exogamous marriage patterns and cooperation among men related by blood only lead to low female agency. Conversely, preferred endogamy and cooperation between men related by blood and marriage lead to relatively high female agency in the South.

Other studies examine variation within a community by examining how kinship structures affect women’s agency at the household level. Thus, this work incorporates family context as a determinant of women’s agency in so far as individual women occupy different locations in the kinship structure at the household level (e.g. Deshmukh-Ranadive 2005; Bennet 1983). In the South Asian context, women’s placement within the joint family system is often examined in terms of how it determines women’s agency. Within the joint family system married women gain power as they move from being daughters-in-law to mothers-in-law. As mothers-in-law they have established their value and security by continuing the family line through having sons and have their own daughters-in-law to command. Following the patterns within the joint family system, several studies of women’s agency take women’s position in the household hierarchy into account (Bloom et al. 2001; Mason 1998; Mullany et al. 2005; Mumtaz and Salway 2005; Malhotra and Mather 1997; Allendorf 2007; Balk 1997; Jejeebhoy and Sathar 2001). Nearly all quantitative studies include women’s age, which is a marker of woman’s stage in the trajectory from new daughter-in-law to established mother-in-law. Several studies also include measures of household structure or women’s place in that structure, such as whether she resides with a mother-in-law or in a nuclear household. These studies reinforce that women who reside with older in-laws have less agency than women who reside in nuclear households or are at the top of the generational hierarchy of a joint household.

This study builds on this earlier work on family context and women’s agency by examining whether another aspect of family context also influences women’s agency. Specifically, it examines whether variation in the quality of family relationships within particular household and kinship structures affects women’s agency. The work described above draws on patterns across kinship and household structures – women on average have greater agency when they reside in family structures that are supportive of women’s power or when they occupy positions in the family that have greater power. This paper builds on this previous work by suggesting that variation in the quality of the family relationships within these kinship and household structures also affects women’s agency. Previous research shows that a woman who is a daughter-in-law in a joint family has less agency than a woman who is a mother-in-law. I go on to hypothesize that a daughter-in-law who has high quality family relationships will have more agency than a daughter-in-law with poor quality family relationships.

Some literature does hint at the potential importance of the quality of family relationships within the marital family. For example, after noting that emotions in general have been neglected in understanding demographic outcomes, Basu (2006) speculates that women’s agency may be one area where emotions, and in particular, love play a key role. She notes that:

“Perhaps, once a suitable survey instrument has been designed, it will be found that love, if only it can be sustained, is as empowering as other ‘demographic’ variables like education and economic independence. It is not so surprising that in much popular fiction, the stereotypical South Asian mother’s biggest nightmare is that her son will fall in love with his wife. The fictional mother-in-law knows that this is the surest way of dismantling her control over household affairs” (Basu 2006, p 117).

Some ethnographic work similarly suggests that family relationship quality may be a source of agency. White (1992) suggests that the quality of women’s family relationships affected the extent to which they control assets in rural Bangladesh. She describes the case of Minu whose “undoubted power” stemmed from the fact that her husband “adored her from the first” and this love gave her confidence. Also in Bangladesh, Kabeer (2000) speculates that the extent to which women are able to negotiate with husbands and other family members to take on factory employment varied according to the amount of harmony in their family relationships.

The Role of Domestic Violence

There is one aspect of family context where relationship quality and women’s agency does explicitly come under scrutiny together – the area of domestic violence. Domestic violence is a dimension of relationship quality. In line with this categorization, domestic violence itself, or broader measures of conflict and fighting that may tap into domestic violence, are often included in measures of marital quality in Western contexts (Bryant et al. 2001; Amato and Booth 1995; Skinner et al. 2002). In the more relevant context of rural Nepal, Hoelter et al. (2004) developed measures of different dimensions of marital quality in which they too identified domestic violence as one dimension.

A handful of studies explore the link between women’s agency and one aspect of relationship quality by exploring the link between agency and domestic violence. Two relevant studies view domestic violence as a dimension of agency and, thus, evaluate the association between domestic violence and agency while exploring associations among dimensions of agency using data from India (Jejeebhoy 2000; Agarwala and Lynch 2006). Using bivariate associations they find that women who experience domestic violence have lower levels of agency. Two other studies reflect on the connection between domestic violence and agency while exploring the determinants of domestic violence. Unlike the studies discussed above, these studies use multivariate analyses that control for potential confounding factors. Hindin and Adair (2002) find in the Philippines that couples where the husband or wife dominate decision-making have the highest levels of abuse, while couples with joint decision-making have the lowest levels of abuse. In Bangladesh, Koenig et al. (2003) find that the effect of women’s agency on domestic violence varied by research site. In the more culturally conservative site, women’s individual agency was a risk factor for violence and women’s agency aggregated to the community level had no association with violence. In the less conservative site, women’s individual agency was a weak protective factor – with marginal statistical significance – and women’s community level agency was a clear protective factor. Both studies speculate that violence is a response to women’s agency. Hindin and Adair conclude that a backlash towards women’s agency should be explored, while Koenig et al. similarly speculate that violence is an initial response to women’s empowerment, but tapers off as women’s increased levels of agency become more accepted.

Taken together these studies have two main lessons for the connections between domestic violence and women’s agency. First, these studies show a mix of results. There is some evidence that there is no association between agency and violence, that there is a negative association, and that there is a positive association. These differences may be due to different cultural settings and changes over time in women’s agency. Thus, these results reinforce the potential importance of context and do not provide unmitigated support of my hypothesis. Second, the work focusing on domestic violence frames women’s agency as a determinant of domestic violence. I hypothesize the opposite, that domestic violence – and relationship quality more broadly with a focus on love and more positive aspects – is a determinant of agency. As is often the case with the complex social world, their framing reinforces that causal connections may go in both directions between women’s agency and relationship quality.

These studies are important reflections on the links between domestic violence and women’s agency. However, they do not reflect on how positive aspects of relationship quality or relationship quality overall influence women’s agency. Measures of domestic violence categorize relationships as ranging from neutral to negative, either violence or not violent. Not experiencing violence, however, is not the same as experiencing a loving, high quality relationship. Domestic violence and other aspects of relationship quality, especially positive aspects, are not well correlated and consistently load onto different dimensions of relationship quality in factor analyses (Hoelter et al., 2004; Johnson et al., 1986; Amato and Booth, 1995). Thus, measures of domestic violence are neither adequate proxies for positive dimensions of relationship quality, nor for relationship quality overall.

Methods

Data

The data for this analysis come from the Women’s Reproductive Histories Survey (WRHS), a survey of 2,444 currently married women aged 15–39 with at least one child in Madhya Pradesh, India. The survey was undertaken in 2002 by the International Center for Research on Women and the International Institute for Population Studies. Respondents were selected through a stratified cluster sample with a response rate of 97%. WRHS was designed to gather data on women’s experiences of abortion. Therefore, while it is a cross-sectional survey, it includes data pertaining to the time of survey, as well as a unique set of retrospective data. The retrospective data cover the time from the respondents’ marriage to the time of interview using a chronological framework anchored on key events in women’s lives, including marriage and births.

Measures

Women’s Agency

The measure of women’s agency is an index composed of several items on women’s ability to make decisions about their own mobility and spending at the time of interview. Women were asked about their mobility in reference to six places: the local market, nearby health center or health services provider, nearby temple or other religious place, home of relatives or friends, cinema or other place of entertainment, and outside of her village or town. The response options included 1) not being able to go at all; 2) going only with a companion; 3) going only with permission; and 4) can go without permission or after informing someone. The spending decision questions asked about women’s ability to purchase eight different items, including food, a small household item, medicine for a child, medicine for herself, a gift, a sari for herself, a small piece of jewelry for herself, and a big household item. The response options included 1) only with permission; 2) with consultation; and 3) without permission or informing.

Exploratory factor analysis was used to create an agency index from these measures of mobility and spending. Exploratory factor analysis allowed for an inductive identification of latent agency variables without presupposing the number of the latent variables or the weighting of the agency items. Given the ordinal and categorical nature of the variables and the survey design, a weighted, polychoric correlation matrix was used in the factor analysis (Kolenikov and Angeles 2009). One factor was retained as the measure of women’s agency. This first factor had an eigenvalue of 8.7 and explained 62% of the shared variance in the agency items. The screeplot test strongly indicated that only one factor should be retained. Further, the agency items all loaded very well onto the first factor with loadings generally well above 0.5 and little remaining uniqueness (table 1).

Table 1.

Loading of items on the agency index from the exploratory factor analysis.

	Agency Index
Mobility – market	0.79
Mobility – health facility	0.85
Mobility – religious place	0.74
Mobility – relative or friend’s home	0.65
Mobility – entertainment center	0.48
Mobility – outside village or town	0.47
Spending decision – food	0.86
Spending decision – small household item	0.89
Spending decision – medicine for child	0.88
Spending decision – medicine for self	0.87
Spending decision – gift	0.88
Spending decision – sari	0.87
Spending decision – jewelry	0.83
Spending decision – large household item	0.83
Eigenvalue	8.74
Variance explained	0.62

Relationship with husband

The quality of the relationship with the respondent’s husband is measured with three variables. The first variable comes from a question which asked women how their relations with their husband were, and followed with a second question on whether they got along well or whether there were difficulties and problems in the period just after her most recent pregnancy ended. Respondents could indicate that they had many, some, or very few difficulties with their husband. This measure is reduced to a dummy variable with a one indicating that the respondent had very few difficulties with her husband just after her most recent pregnancy ended and a zero indicating that she had many or some difficulties with her husband at that time.

The second variable measuring the quality of the marital relationship is a measure of domestic violence. Women were asked whether their husband had beaten or slapped them in the past twelve months. They were also asked whether their husband had threatened to abandon them or throw them out of the house in the last year. Almost all women whose husbands threatened to abandon them also experienced physical violence. Thus, responses from these two questions are combined to create a three category measure of domestic violence: 1) experienced no violence; 2) experienced physical violence only; and 3) husband threatened to abandon her. 92% of women in this third category who experienced the threat of abandonment or being thrown out of the house also experienced physical violence.

The third and final measure of the quality of the marital relationship is an indicator of who is most important to the respondent’s husband. Customarily in the joint family system, a husband’s primary loyalty should be to his parents (Das Gupta 1999). So, if a woman believes that she comes first for her husband it is an indication that they have a strong relationship (Derne 1995). Respondents were asked “most of the time, do you feel that your husband’s first loyalty is to you or to his parents?” The response options included the respondent, her husband’s parents, the respondent and the parents equally, and someone else. Five women reported that their husband’s first loyalty was to someone else and ten women reported that they did not know who came first for their husbands. Since so few respondents chose these two options these responses were grouped with the husband’s primary loyalty being to his parents. This variable was also further adjusted for family structure. Many of these women who responded that they came first for their husbands likely did so simply because their husband no longer had parents. In this case, the woman’s response is determined by the availability of family members, rather than the quality of her relationship with her husband. So, an additional category was created which includes all respondents whose husband did not have parents at the time of the interview. Thus, the final variable indicating who came first for the respondent’s husband includes four categories: 1) the husband’s parents; 2) the parents and the respondent equally; 3) the respondent; and 4) husband does not have (living) parents.

These three measures of marital quality are used separately in the following multivariate analysis. The possibility of combining these variables into a single measure of marital quality was explored using factor analysis. However, apart from the two violence items, the variables did not load satisfactorily onto a single factor. Thus, the violence variables are combined into a single measure, as described above, while the other measures are used independently. This result is consistent with previous research noted above, which finds that domestic violence is not well correlated with other aspects of marital quality.

Relationship with in-laws

There is one measure of the relationship with in-laws. Just as for husbands, respondents were asked how relations were with their in-laws just after their most recent pregnancy ended. This first question on relations was followed up with a probe on whether they got along well or whether there were difficulties and problem. Unlike the husband variable, this in-law relations variable also includes a category for those who do not have (living) in-laws. Thus, the measure of in-law relations includes the following categories: 1) respondent has many difficulties with in-laws; 2) respondent has some difficulties with in-laws; 3) respondent has very few difficulties with in-laws; and 4) respondent has no in-laws. Due to the small number of women with many difficulties with their in-laws, the first two categories are combined in the multivariate analysis.

Unfortunately, these indicators are not ideal measures of family relationship quality. Ideally, indicators of family relationship quality should directly measure key dimensions of relationship quality, such as love, understanding, frequency and manner of disagreements, and communication dynamics (Author 2009). Further, the two more global measures of relations with husbands and in-laws have a problematic distribution. Roughly three-quarters or more of women report that they have very few difficulties with their husbands or in-laws (table 2). Thus, these measures are not able to identify respondents with the very best relationships, such as those in the top fifth of the distribution. Evaluating the hypothesis that women with higher quality relationships have higher levels of agency requires identifying women with high quality relations. Instead, these measures allow for comparison of women with average relationship quality to women with very poor relationship quality. The lack of better measures of relationship quality is indicative of the neglect of relationship quality more broadly in the literature. This survey is unusual in including any measures of relationship quality that go beyond domestic violence.

Table 2.

Descriptive statistics for key variables. (The percentages in parentheses refer to a sample that is limited to women who have in-laws.)

	Women in all families n = 2,444 %	Women in nuclear families n = 1,340 %	Women in joint families n = 1,104 %	Differs significantly between joint and nuclear families
Agency index	mn:23.9 sd:7.6	mn:25.4 sd:7.9	mn:22.1 sd:6.9	Yes
Relations with husband
Many difficulties	1.7	1.8	1.6	No
Some difficulties	19.3	17.8	21.0
Very few difficulties	79.0	80.4	77.5
Husband’s violence
None	52.1	47.8	57.0	Yes
Physical violence only	26.6	27.9	25.1
Threatened abandonment	21.3	24.3	17.9
Comes first for husband
Parents	15.3 (18.7)	5.9 (8.9)	26.2	Yes
Parents & respondent equally	34.5 (42.2)	20.8 (31.4)	50.4
Respondent	32.0 (39.1)	39.4 (59.7)	23.4
Husband has no parents	18.3 (0)	34.0 (0)	0
Relations with in-laws
Many difficulties	1.6 (2.0)	1.4 (2.3)	1.8	Yes
Some difficulties	21.8 (27.9)	13.2 (22.3)	31.8
Very few difficulties	54.7 (70.1)	44.6 (75.4)	66.4
No in-laws	21.9 (0)	40.8 (0)	0
Household wealth factor	mn:0.0 sd:1.9	mn: −0.2 sd:1.8	mn:0.3 sd:2.0	Yes
Urban residence	23.0	24.7	21.1	No
Age
Less than 25	28.8	19.9	39.3	Yes
25–29	28.8	26.8	31.1
30–34	24.8	30.3	18.3
35–39	17.7	23.1	11.4
Education
No schooling	56.1	61.7	49.5	Yes
Less than SLC (<10 years)	30.3	28.0	33.0
SLC or more (10+ years)	13.6	10.3	17.5
Works outside household	35.4	45.4	23.8	Yes
Mobility after marriage
Too many restrictions	23.6	24.1	23.1	No
Several restrictions	26.8	27.6	25.8
A few restrictions	34.2	31.9	36.8
No restrictions	15.5	16.4	14.4
Spending after marriage
Only with permission	40.8	42.1	39.3	No
Usually with permission	22.5	21.5	23.6
Sometimes with permission	20.6	19.2	22.3
Without permission	16.1	17.2	14.9

Control variables

Contextual controls for urban residence and household wealth are included in the analysis. Other contextual controls, including caste, religion, husband’s occupation, and husband’s education were included in earlier models. These variables are not included in the final analysis because they were consistently insignificant and did not affect the results of the family relationship variables. Urban residence is a dummy variable indicating that the respondent lives in an urban area. The measure of household wealth is a factor obtained from a principle components analysis. The survey included several categorical and dichotomous measures of household wealth, including measures of housing materials and ownership of consumer durables and assets, which were reduced into a single factor with principle components analysis using a weighted polychoric correlation matrix (Kolenikov and Angeles 2009). The resulting wealth factor accounts for 53% of the total variance in the original wealth measures.

Four other sources of agency were also included as controls, including age, education, employment, and having two sons. The respondents’ age is included in five year groups. The measure of education includes categories for no schooling, less than a school leaving certificate (SLC), and a school leaving certificate or more. The school leaving certificate, which is obtained after completing ten years of schooling and passing a test, is roughly equivalent to completing high school in the United States. The measure of employment is a simple dummy measure indicating whether the respondent worked outside the household or not. 95% of women who worked were laborers. Thus, there is not enough occupational variation to include a more nuanced measure of employment. Finally, completing the familial obligation to bear sons and continue the family line is another potential source of agency, which may also improve family relations. Thus, a dummy variable indicating whether women had two sons is also included. Other potential sources of agency, including age at marriage (gauna), length of marriage, and difference in age compared to husband were included in earlier models. These variables were dropped from the final analysis because they did not affect the results of the family variables and were too collinear with age or consistently insignificant.

Finally, two measures of women’s level of agency at the time of marriage are also included as controls. The two measures of women’s agency at the time of marriage are ordinal variables that measure the amount of choice women had over spending and their mobility. For mobility, women were asked in reference to the time just after they were married, “What kind of restrictions did you face personally? Were you able to come and go as you pleased, or did you have to seek permission?” Response categories include 1) too many restrictions; 2) several restrictions; 3) a few restrictions; and 4) no restrictions. For spending, respondents were asked again in reference to the time just after they were married, “Were you able to spend money as you pleased, or did you have to seek permission?” The response categories included 1) only with permission; 2) usually with permission; 3) sometimes with permission; and 4) without permission. Descriptive statistics for all of the variables used in the analysis appear in table 2.

Analysis

All analyses are run separately by family structure since the effect of the family relationship variables are likely to vary by the different family contexts. Moreover, interpretation of the results is furthered by better understanding the family context that women live in. Family structure is a key determinant of household power relations. In joint households, where women reside with their in-laws, the father and mother-in-law are customarily the most important decision makers in the household. Thus, in joint households, women’s relations with their in-laws are likely to play a more important role than in nuclear households. By comparison, in nuclear households where the husband is the customary decision maker, the marital relationship may play a bigger role in determining women’s participation in household decisions.

Thus, respondents are divided into two categories of family structure: joint and nuclear. Families are categorized as joint if the respondent resided with her father-in-law, mother-in-law, brother-in-law, or sister-in-law. If none of these in-laws were present, women are categorized as residing in a nuclear family. A handful of women who did not live with in-laws resided with natal family members, including six women who lived with their fathers, six who lived with their mothers, and one who lived with both of her own parents. Residing with natal family members is substantially different than residing with in-laws – the traditional joint family household. Yet the number of women residing with natal family members is too small to comprise a category by itself. Therefore, this handful of women are grouped in with the nuclear households.

A multivariate linear regression is used to test the hypothesis that women with greater family relationship quality have higher levels of agency (table 3). The standard errors in these models are adjusted to control for clustering of respondents by the primary sampling unit and district in the survey. As noted above, all analyses are presented separately for women in nuclear and joint families. Further, for each family type, two models are presented. The first model includes the family relationship variables and a set of controls that include basic contextual controls and other sources of agency. The second model adds the measures of women’s agency at the time of marriage.

Table 3.

Coefficients and standard errors from multivariate regression models of women’s agency.

	Women in nuclear families n = 1,340				Women in joint families n = 1,104
	Model 1		Model 2		Model 1		Model 2
	β	SE	β	SE	β	SE	β	SE
Relations with husband
Many or some difficulties (ref)	0		0		0		0
Very few difficulties	1.4*	(0.62)	1.5**	(0.60)	−0.0	(0.51)	0.8	(0.52)
Comes first for husband
Parents (ref)	0		0		0		0
Parents & respondent equally	0.9	(0.88)	0.6	(0.88)	1.0*	(0.44)	0.8*	(0.38)
Respondent	3.1**	(0.83)	2.5**	(0.80)	3.4**	(0.73)	3.4**	(0.63)
Husband has no parents	2.5*	(1.12)	2.2†	(1.11)
Husband’s violence
None (ref)	0		0		0		0
Physical violence only	−1.3*	(0.55)	−0.1	(0.51)	−0.8†	(0.40)	0.3	(0.43)
Threatened abandonment	−2.1**	(0.60)	−0.4	(0.63)	−1.3*	(0.54)	0.9	(0.54)
Relations with in-laws
Many or some difficulties (ref)	0		0		0		0
Very few difficulties	0.4	(0.80)	0.9	(0.79)	2.1**	(0.49)	1.7**	(0.44)
No in-laws	0.7	(1.17)	0.9	(1.18)
Household wealth factor	0.5*	(0.21)	0.4†	(0.20)	−0.1	(0.23)	−0.0	(0.17)
Urban residence	4.9**	(0.91)	4.2**	(0.85)	4.8**	(1.02)	3.8**	(0.76)
Age
Less than 25 (ref)	0		0		0		0
25–29	1.2*	(0.56)	0.9	(0.53)	1.4**	(1.02)	1.3**	(0.44)
30–34	1.7**	(0.53)	1.4*	(0.51)	3.4**	(0.65)	3.4**	(0.58)
35–39	1.5*	(0.68)	1.4*	(0.63)	4.2**	(0.72)	3.9**	(0.61)
Education
No schooling (ref)	0		0		0		0
Less than SLC (<10 years)	0.9†	(0.48)	0.7	(0.47)	0.6	(0.49)	0.8	(0.42)
SLC or more (10+ years)	3.4**	(0.79)	2.2**	(0.79)	3.2**	(0.72)	2.3**	(0.60)
Works outside household	2.0**	(0.49)	1.8**	(0.48)	1.6**	(0.47)	1. 7**	(0.38)
Has two or more sons	0.7	(0.44)	1.1*	(0.43)	−0.0	(0.43)	0.1	(0.36)
Mobility after marriage
Too many restrictions (ref)			0				0
Several restrictions			0.9	(0.57)			2.5**	(0.42)
A few restrictions			2.0**	(0.58)			2.4**	(0.51)
No restrictions			2.8**	(0.81)			3.1**	(0.81)
Spending after marriage
Only with permission (ref)			0				0
Usually with permission			−0.2	(0.46)			1.1*	(0.47)
Sometimes with permission			2.7**	(0.70)			3.6**	(0.57)
Without permission			5.0**	(0.77)			5.7**	(0.85)
Constant	18.4**	(1.12)	15.3**	(1.20)	16.1**	(0.57)	11.8**	(0.58)
R2	0.26		0.36		0.29		0.42

^†p<0.10

^*p<0.05

^**p<0.01, two tailed

The measures of agency at the time of marriage are added to control for reverse causation. I hypothesize that higher quality relationships increase women’s agency. Thus, I expect to find that women with higher quality relationships have higher levels of agency. However, such a correlation may also be due to causation going in the opposite direction. Having greater agency may make a woman more likely to have a high quality relationship. For example, women with greater agency may be more likely to choose their spouse and thus may be more likely to have a higher quality marital relationship (at least in the beginning of a marriage) since they are likely to choose someone they have a good relationship with. Even after controlling for agency at the time of marriage, it is still possible that women who are given greater freedom by their family members are also more likely to perceive those relationships as being of a higher quality. In other words, women love and appreciate family members who let them do what they want. So, some amount of reverse causation may still be an issue however.

Results

Relationship with Husband

In nuclear families, women who have very few difficulties with their husbands have greater agency (table 3). Women in nuclear families who have very few difficulties with their husband score significantly higher on the agency index, with 1.5 more points on average than women who have many or some difficulties. In joint families, however, there is no association. In joint families, the coefficient for very few difficulties with husband is 0.8 and not significant. This result suggests that having very few difficulties with husbands has a positive effect on women’s agency in nuclear families, but not in joint families. However, the differences in these coefficients by family structure are not significant. In a pooled model with women from both family types, an interaction term for very few difficulties with husband by family structure is not statistically significant (results not shown).

The husband’s primary loyalty has a strong effect on women’s agency in both nuclear and joint families. Women in nuclear families who come first for their husbands score 2.5 points higher on the agency index than women whose husband’s parents come first. Women in nuclear families whose husbands have no parents score nearly as high as women who have the primary loyalty of their husbands. They score 2.2 more points on the agency index than women whose husband’s parents come first. Thus, it appears that having no parents-in-law has the same positive impact on agency as having the husband’s primary loyalty. This result highlights the role that in-laws can play in decision-making, even when they do not reside in the same household. Women whose husband’s have equal loyalty to both them and his parents score 0.6 points more on the agency index than women who husband’s primary loyalty is to his parents only, but the result is not significant.

Similar results are seen for women in joint families. Women in joint families who come first for their husbands score 3.4 points higher on the agency index compared to women whose husband’s primary loyalty is to his parents. Women in joint families whose husband’s primary loyalty is to them and his parents equally score 0.8 points higher on the agency index. The point estimates for women in joint families are higher than for women in nuclear families, suggesting that husband’s loyalty has a larger effect on women’s agency in joint families. However, this difference in the coefficients is not significant. In a pooled model, the interaction term for husband’s loyalty by family structure is not statistically significant (results not shown).

Unlike the other two measures of marital quality, domestic violence has no effect on women’s agency. Interestingly, in model one, husband’s violence at first appears to have a negative effect – women who report physical violence or a threat of abandonment score lower on the agency index according to model one. However, in model two, in which agency at the time of marriage is included as an additional control, husband’s violence no longer has an effect. For women in nuclear families, the coefficient for reporting physical violence only is −0.1 and for threats of abandonment it is −0.4. Not surprisingly, neither coefficient is significantly different from zero. For women in joint families, the coefficient for experiencing physical violence is 0.3 and for reporting threats of abandonment it is 0.9. Again, neither coefficient is significantly different from zero.

Relationship with In-Laws

The quality of in-law relations has a positive and significant effect on agency for women in joint families, but not for women in nuclear families (table 3). In joint families, women who have very few difficulties with their in-laws score 1.7 points higher on the agency index than those who have many or some difficulties with their in-laws. This coefficient is significant with a p-value below 0.01. In nuclear families, women who have very few difficulties with their in-laws score nearly one point higher on the agency index, but this result is not significantly different from zero. However, as with husband relations, the difference in the effect of in-law relations between joint and nuclear families is not significant. In a pooled model, an interaction term for in-law relations by family structure is not statistically significant (results not shown).

Comparative Influence of Family Relationship Quality

The results described above show that the quality of family relationships is a source of women’s agency. Thus, a key follow up question is how important is family relationship quality? Is relationship quality more or less influential than other determinants of agency? To address this question, I compare the amount of variance explained by the quality of family relationship variables and the relative size of the coefficients for the relationship quality variables to those of the other sources of agency. The contextual variables and other sources of agency presented in table three – as well as those dropped from the analysis – are not an exhaustive list of potential sources of agency. They do comprise the most commonly addressed sources of agency, however, particularly with the inclusion of education and employment.

Relationship quality, as measured here, explains 4.5% of the variation in women’s agency overall (table 4). Specifically, relationship quality accounts for 2.9% of the variation in women’s agency among women residing in nuclear families and 6.6% of the variation among women residing in joint families. These percentages are calculated by subtracting the r-squared value from model two, which includes all controls and measures of agency at time of marriage, from the r-squared value from the same model without the measures of relationship quality.

Table 4.

Percent of variation in women’s agency uniquely explained by different sources.

	Women in all families %	Women in nuclear families %	Women in joint families %
Relationship quality	4.5	2.9	6.6
Urban residence	2.8	3.0	2.6
Age	1.9	0.4	3.6
Works outside household	1.1	1.2	0.9
Education	0.6	0.4	0. 8
Has two or more sons	0.3	0.5	0.0

Explaining 4.5% of the variation in women’s agency may appear like a small contribution, however, this is larger than all other determinants of agency (table 4). Commonly addressed sources of agency, including women’s age, education, employment, having sons, and urban residence explain even less of the variation in women’s agency. In nuclear families, urban residence explains 3.0% of the variation in agency, employment outside the household accounts for 1.2%, and age, education, and having two sons all account for less than a half of one percent. Similarly, in joint families, urban residence explains 2.6% of the variation in agency, age explains 3.6%, and education and employment both explain less than one percent.

The size of the coefficients of the relationship quality variables, particularly who comes first for the husband, also compare favorably with other sources of agency. By and large, according to adjusted Wald tests, the coefficients for relationship quality are not significantly larger than those for other sources of agency (results not shown). However, the point estimates for the coefficients suggest that relationship quality is just as influential, if not more influential, than other sources of agency. For example, the point estimate of the coefficient for the respondent coming first in their husband’s estimation is larger than those for having a secondary education and working outside the household in both nuclear and joint families. One clear exception to this pattern, however, is domestic violence. Domestic violence is not an influential determinant of agency since it has no effect on women’s agency.

Discussion and Conclusion

Despite the lack of ideal measures, it appears that as hypothesized, women with higher quality family relationships do have higher levels of agency. In both joint and nuclear families, women who have better relations with their husbands have greater agency. In joint families, women who have better relations with their in-laws also have greater agency. In particular, being the first one in their husband’s loyalties, above and beyond their husband’s parents, has the strongest effect on women’s agency, suggesting that husband’s influence plays a key role in women’s agency. Further, as a determinant of women’s agency, the quality of family relationships is comparable to other well-known determinants of agency, including education and working outside the home.

There is a clear exception however. Women’s reporting of domestic violence has no association with their agency after controlling for their agency at the time of marriage. This null result may be due in part to the dual role that violence can play. Domestic violence is certainly an indicator of poor family relations, which, as theorized here, should lower women’s agency. However, at the same time, domestic violence may be a response to women asserting their choices. If domestic violence simultaneously occurs in both directions – both working to reduce agency and punish women whose agency oversteps the bounds approved by family members – then women’s experience of domestic violence may have no association on average with agency as seen here.

More broadly, this study suggests that positive aspects of family relationship quality do influence women’s lives. The strength of family bonds, especially the marital bond, plays an important role in determining women’s ability to make decisions in the family. This suggests that the fears of Basu’s “fictional mothers-in-law” are not just fictional. This also reinforces the importance of incorporating a more holistic view of relationship quality into future research. The burgeoning literature on domestic violence reflects on an important aspect of family relations and women’s lives, but it is not the only aspect that should be examined. While women certainly experience violence and abuse in the family, they also experience love and affection. These positive family relationships can be a positive and empowering force in women’s lives.