The Influence of Mothers-in-Law on Women in India Part 2

Curse of the Mummy-ji Figures Figure 1: Directed Acrylic Graph X (Pretreatment confounders) MIL co-residence (Treatment) Intermediate confounders, Z (Mobility, Autonomy) Close Peers (Mediator) Y (Outcome) Curse of the Mummy-ji Figure 2: Non-parametric relationship between ClosePeersicv and IV NOTES: These graphs plot the non-parametric relationship between a woman’s number of close outside peers in her village and our IV, (𝑃𝑟𝑜𝑝𝑐𝑣 ⋅ 𝑀𝐼𝐿𝑖 ). In the top left figure, we use the Epanechnikov kernel function and a locallinear estimator. In the top right figure, we use the Gaussian kernel function and a local-linear estimator. In the bottom left figure, we use the Epanechnikov kernel function and a local-constant estimator. The bottom right figure uses the Epanechnikov kernel function, a local-linear estimator, and improved AIC instead of cross-validation to compute optimal bandwidth. Curse of the Mummy-ji Tables Table 1: Summary statistics Variable Age Husband age SC ST OBC Upper caste Hindu Years of schooling Marriage duration (years) Age at marriage Live with MIL Own land Amount of land owned (acres) No. of living sons No. of living children Firstborn is a son Allowed to visit alone: Home of relatives/ friends Health facility Grocery store Short distance train/ bus Market Outside village/ community Wears ghunghat/ purdah Worked in the last 7 days Using modern contraception Has visited family planning clinic N (1) 671 644 671 671 671 671 671 671 655 655 671 638 671 671 671 671 Mean (2) 25.67 32.57 0.43 0.01 0.44 0.12 0.93 9.53 7.31 18.36 0.68 0.60 5.43 0.96 1.95 0.50 Std. Dev. (3) 2.65 9.45 0.50 0.12 0.50 0.32 0.25 4.47 3.61 2.52 0.47 0.49 2.72 0.76 0.92 0.50 Min (4) 18 18 0 0 0 0 0 0 0 6 0 0 1 0 1 0 Max (5) 30 73 1 1 1 1 1 15 21 28 1 1 26 4 5 1 671 671 671 671 671 671 671 666 670 0.12 0.14 0.16 0.08 0.19 0.21 0.88 0.14 0.18 0.32 0.35 0.37 0.27 0.39 0.40 0.32 0.35 0.38 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 671 0.35 0.48 NOTES: This table describes the characteristics of our sample. Columns (1)-(5) report, respectively, the number of observations, the mean, and the standard deviation, the minimum, and the maximum value for each variable. SC, ST, and OBC denote, respectively Scheduled Caste, Scheduled Tribe, and Other Backward Class. Curse of the Mummy-ji Table 2: Influence of the MIL on DIL's number of peers, OLS Co-residence with: Mother-in-law (1) -0.120** [0.045] Father-in-law # other women > age 18 # other 18-30 women # other 18-30 married women Control mean Mother-in-law Father-in-law # other women > age 18 # other 18-30 women # other 18-30 married women Control mean N 0.606 (2) (3) (4) (5) A. Outcome: # close peers in the village -0.114* -0.117** -0.128** -0.129** [0.062] [0.050] [0.049] [0.053] -0.0004 -0.020 -0.010 -0.011 [0.056] [0.041] [0.039] [0.039] 0.072** [0.029] 0.079** [0.033] 0.129*** [0.039] 0.606 0.606 0.606 0.606 B. Outcome: # close outside peers in the village -0.133*** -0.138** -0.137*** -0.133*** -0.134*** [0.035] [0.053] [0.044] [0.044] [0.045] 0.003 0.010 0.007 0.005 [0.048] [0.045] [0.045] [0.044] -0.028* [0.016] -0.032 [0.020] -0.035 [0.023] 0.361 0.361 0.361 0.361 0.361 671 653 653 653 653 NOTES: This table reports coefficients from specification (1). Each column is a separate OLS regression. The outcome variables in panels A and B are the DIL's number of close peers in the same village and the number of close peers that are not household members, respectively. In all cases, we control for the DIL's age, years of schooling, Hindu dummy, amount of land owned by the household, and fixed effects for her caste category (SC-ST, OBC, or Other caste) and village. In addition, we gradually add an indicator for residence with the FIL, the number of other women in the HH that are above 18, in the 18-30 age group, and in the married 18-30 group, as controls across columns. Control mean refers to the dependent variable mean for women who do not live with their MIL. Robust standard errors in brackets are clustered by village. *** 𝑝 < 0.01, ** 𝑝 < 0.05, * 𝑝 < 0.1. Curse of the Mummy-ji Table 3: Influence of the MIL on DIL's mobility, OLS regressions Outcome: DIL is usually allowed to visit the following places alone: Short Home of Health Grocery distance Outside village/ relatives/ friends facility store bus/ train Market community (1) (2) (3) (4) (5) (6) Lives with MIL N Control Mean Wears ghunghat/ purdah (7) -0.096** [0.036] -0.134*** [0.037] -0.157*** [0.038] -0.043* [0.021] -0.167*** [0.035] -0.083*** [0.026] 0.064*** [0.019] 671 0.218 671 0.255 671 0.310 671 0.125 671 0.329 671 0.296 671 0.838 NOTES: This table reports coefficients from specification (1). Each column is a separate regression. The outcome variables are indicators that equal one if the DIL is usually allowed to visit the respective places alone. In all cases, we control for the DIL's age, years of schooling, Hindu dummy, amount of land owned by the household, and fixed effects for her caste category (SC-ST, OBC, or Other caste) and village. Control mean refers to the dependent variable mean for women who do not live with their MIL. Robust standard errors in brackets are clustered by village. *** 𝑝 < 0.01, ** 𝑝 < 0.05, * 𝑝 < 0.1. Curse of the Mummy-ji \ Table 4: Test for selection into living with MIL Variables: Age Husband age Marriage Duration Years of schooling Spousal schooling gap SC ST OBC Hindu Amount of land owned (acres) Employed No. of living sons Live with MIL = 0 N Mean Live with MIL = 1 N Mean Difference (2)-(4) (1) (2) (3) (4) (5) 216 211 210 216 211 216 216 216 216 216 213 216 26.38 31.938 8.595 8.102 0.275 0.472 0.014 0.454 0.926 5.374 0.131 1.028 455 433 445 455 432 455 455 455 455 455 453 455 25.336 32.875 6.697 10.207 -0.155 0.429 0.033 0.431 0.938 5.460 0.146 0.925 1.043*** -0.937 1.899*** -2.105*** 0.430 0.044 -0.019 0.023 -0.013 -0.085 -0.014 0.103 NOTES: This table compares the average characteristics of women who do not live with the MIL (columns (1)-(2)) and women who do (columns (3)-(4)). Column (5) reports the difference in the sample mean for the two groups. SC, ST, and OBC denote, respectively Scheduled Caste, Scheduled Tribe, and Other Backward Class. *** 𝑝 < 0.01, ** 𝑝 < 0.05, * 𝑝 < 0.1. Curse of the Mummy-ji Table 5: Influence of the MIL on DIL's mobility if FIL is co-resident, OLS regressions Outcome: DIL is usually allowed to visit the following places alone: # Close outside peers (1) MIL N Control mean Home of relatives/ friends in village (2) Short distance bus/ train (5) Sample restriction: FIL co-resident -0.098** -0.131*** -0.054 [0.045] [0.047] [0.042] Health facility (3) Grocery store (4) Market (6) Outside village/ community (7) Wears ghunghat/ purdah (8) -0.168** [0.061] -0.091 [0.060] 0.020 [0.036] -0.234*** [0.081] -0.134** [0.059] 406 406 406 406 406 406 406 406 0.422 0.234 0.203 0.250 0.109 0.312 0.436 0.891 NOTES: This table reports coefficients from specification (1). Each column is a separate regression. The outcome variables are the same as those in Tables 2 and 3. The sample is restricted to households where the father-in-law (FIL) is co-resident. In all cases, we control for the DIL's age, years of schooling, Hindu dummy, amount of land owned by the household, and fixed effects for her caste category (SC-ST, OBC, or Other caste) and village. Control mean refers to the dependent variable mean for women who have a co-resident FIL but who do not live with their MIL. Robust standard errors in brackets are clustered by village. *** 𝑝 < 0.01, ** 𝑝 < 0.05, * 𝑝 < 0.1. Curse of the Mummy-ji Table 6: Heterogeneity in the influence of the MIL on DIL's number of peers, by MIL’s fertility preferences, OLS regressions A. Outcome: # close peers in the village Lives with MIL Control Mean Lives with MIL Control Mean N MIL disapproves of FP (1) MIL approves of FP (2) Ideal KidsMIL > Ideal KidsDIL (3) Ideal KidsMIL <= Ideal KidsDIL (4) Ideal SonsMIL > DIL’s sons (5) Ideal SonsMIL <= DIL sons (6) -0.160** [0.061] 0.556 -0.115* [0.060] 0.691 -0.117** [0.045] 0.572 0.0003 [0.145] 0.744 -0.127** [0.052] 0.573 0.074 [0.144] 0.733 B. Outcome: # close outside peers in the village -0.103** -0.169 -0.098** [0.041] [0.125] [0.043] 0.329 0.488 0.316 -0.164 [0.134] 0.533 -0.149*** [0.041] 0.348 -0.119* [0.064] 0.383 320 351 519 152 530 141 NOTES: This table reports coefficients from specification (1). Each column within a panel is a separate regression. The outcome variable is the number of close peers a woman has in her village in Panel A and the number of such peers outside the household in Panel B. Columns (1) and (2) split the sample by whether the MIL approves of using FP or not. Columns (3) and (4) compare the number of children the MIL would like the DIL to have (Ideal KidsMIL) and the DIL’s ideal number of children (Ideal KidsDIL). Columns (5) and (6) compare the number of sons the MIL would like the DIL to have (Ideal SonsMIL) and the DIL’s number of sons at the time of the survey (DIL sons). In all cases, we control for the DIL's age, years of schooling, Hindu dummy, amount of land owned by the household, and fixed effects for her caste category (SC-ST, OBC, or Other caste) and village. Control mean refers to the dependent variable mean for women who do not live with their MIL. Robust standard errors in brackets are clustered by village. *** 𝑝 < 0.01, ** 𝑝 < 0.05, * 𝑝 < 0.1. Curse of the Mummy-ji Table 7: The influence of peers on women’s access and use of family planning, OLS regressions # close outside peers N Control mean Has visited FP clinic (1) Uses modern method (2) Beliefs about FP use in village (3) Allowed to visit health facility with someone (4) 0.130** [0.054] 0.067 [0.039] 0.233 [0.188] 0.024*** [0.008] 671 0.303 670 0.164 671 2.295 671 0.971 NOTES: This table reports coefficients from specification (1). Each column is a separate regression. The key explanatory variable is a woman’s number of close peers who live in her village but not in her household. The outcome variables are: an indicator for whether a woman has visited a health facility for reproductive health, fertility, or family planning services in column (1); an indicator for whether a woman is using a modern method of contraception at the time of survey in column (2); a categorical variable that takes values 0 to 6 with higher values indicating a woman’s belief that more women in her village use family planning in column (3); and an indicator for whether a woman is usually allowed to visit a health facility with someone in column (4). In all cases, we control for the DIL's age, years of schooling, Hindu dummy, amount of land owned by the household, and fixed effects for her caste category (SC-ST, OBC, or Other caste) and village. Control mean refers to the dependent variable mean for women who do not have a close outside peer in their village. Robust standard errors in brackets are clustered by village. *** 𝑝 < 0.01, ** 𝑝 < 0.05, * 𝑝 < 0.1. Curse of the Mummy-ji Table 8: Mediation analysis ATE (1) 𝑀𝐼𝐿𝑖 N p-value: (3) vs (1) 𝑀𝐼𝐿𝑖 N p-value: (3) vs (1) 𝑀𝐼𝐿𝑖 N p-value: (3) vs (1) 𝑀𝐼𝐿𝑖 N p-value: (3) vs (1) -0.066* [0.037] 671 -0.005 [0.047] 670 -0.436** [0.180] 671 -0.020** [0.009] 671 ACDE (g-est) (2) -0.050 [0.038] (0.0407) 671 0.004 [0.047] (0.036) 670 Decline: (2) - (1) ATE (3) (4) A. Has visited FP clinic 24% -0.079* [0.040] ACDE (g-est) (5) -0.064 [0.040] 671 671 0.0002 B. Uses modern method -0.011 -0.002 [0.049] [0.050] 670 -0.478** [0.187] 19% 0.0002 670 0.0009 C. Beliefs about FP use in village 6% Decline: (5) - (4) (6) 0.458** [0.185] -0.412 [0.178]** (0.165)** 671 671 671 0.0002 D. Allowed to visit health facility with someone 15% -0.023** 0.020** -0.017 [0.010] [0.010] [0.009]* (0.010) 671 671 671 0.0002 0.0006 4% 0.0002 13% 0.0002 NOTES: The p-values test if the estimates of ACDE from sequential g-estimation are significantly different from the estimated ATE. In columns (4) and (5), the specification includes Caste x Village FE. Standard errors in brackets are clustered at the village level and in parentheses are bootstrapped. *** 𝑝 < 0.01, ** 𝑝 < 0.05, * 𝑝 < 0.1. Curse of the Mummy-ji Table 9: First-stage results from 2SLS regressions 𝑃𝑟𝑜𝑝𝑐𝑣 𝑃𝑟𝑜𝑝𝑐𝑣 ∗ 𝑀𝐼𝐿𝑖 𝑃𝑟𝑜𝑝𝑐𝑣 N First stage F-stat 𝑋𝑖 Caste FE Village FE Caste x village FE Outcome: No. of close outside peers in village 𝑃𝑟𝑜𝑝𝑐𝑣 1830 𝑃𝑟𝑜𝑝𝑐𝑣 1830,𝑚𝑎𝑟𝑟𝑖𝑒𝑑 (1) (2) (3) (4) (5) (6) -0.256*** [0.066] 0.162* [0.094] -0.214*** [0.068] -0.261*** [0.066] 0.134 [0.098] -0.232*** [0.066] -0.260*** [0.069] 0.130 [0.099] -0.229*** [0.068] 671 14.93 671 10.05 671 15.70 671 12.51 671 14.22 671 11.35 x x x x x x x x x x x x NOTES: This table reports coefficients from two versions of specification (5); in columns (1), (3), and (5), we only include 𝑃𝑟𝑜𝑝𝑐𝑣 ∗ 𝑀𝐼𝐿𝑖 and 𝑃𝑟𝑜𝑝𝑐𝑣 as explanatory variables, while the rest of the columns estimate the full version of specification (5). Each column is a separate regression. The outcome variable is a woman’s number of close peers who live in her village but not in her household. Across columns, we use the three definitions of the peer-pool described in the text. Robust standard errors in brackets are clustered at the village level. *** 𝑝 < 0.01, ** 𝑝 < 0.05, * 𝑝 < 0.1. Curse of the Mummy-ji Table 10: Second stage and reduced form results from 2SLS regressions A. Second stage # close outside peers B. Reduced form 𝑃𝑟𝑜𝑝𝑐𝑣 ∗ 𝑀𝐼𝐿𝑖 N Visited FP clinic (1) Uses modern method (2) Beliefs about FP use in village (3) Allowed to visit health facility with someone (4) 0.664* [0.345] 0.111 [0.350] 2.607*** [0.966] 0.103 [0.072] -0.152* [0.083] -0.024 [0.082] -0.596* [0.321] -0.024 [0.017] 671 670 671 671 NOTES: This table reports coefficients from specification (6) in Panel A and the reduced form estimates for our IV estimation in Panel B. Each coefficient is from a separate regression. The outcome variables are: an indicator for whether a woman has ever-visited a health facility for reproductive health, fertility, or family planning services in column (1); an indicator for whether a woman is using a modern method of contraception at the time of survey in column (2); a categorical variable that takes values 0 to 6 with higher values indicating a woman’s belief that more women in her village use family planning in column (3); and an indicator for whether a woman is usually allowed to visit a health facility with someone in column (4). Robust standard errors in brackets are clustered at the village level. *** 𝑝 < 0.01, ** 𝑝 < 0.05, * 𝑝 < 0.1. Curse of the Mummy-ji Table 11: Placebo tests A. Second stage # close outside peers B. Reduced form 𝑃𝑟𝑜𝑝𝑐𝑣 ∗ 𝑀𝐼𝐿𝑖 N Firstborn is a son (1) Involved in land dispute (2) Mother attended school (3) Father attended school (4) -0.204 [0.353] -0.038 [0.182] 0.320 [0.304] -0.489 [0.308] 0.047 [0.083] 0.009 [0.045] -0.073 [0.075] 0.112 [0.075] 451 671 671 671 NOTES: This table reports coefficients from specification (6) in Panel A and the reduced form estimates for our IV estimation in Panel B. Each column is a separate regression. The outcome variables are indicators that equal one if the firstborn child of the woman is a son in column (1), if her household is involved in a land dispute in column (2), if her mother attended school in column (3), and if her father attended school in column (4). Robust standard errors in brackets are clustered at the village level. *** 𝑝 < 0.01, ** 𝑝 < 0.05, * 𝑝 < 0.1. Curse of the Mummy-ji 1 With an estimated population of 200 million people in 2012, Uttar Pradesh would be the world’s fifth most populated country by itself. 2 Patrilocality refers to the practice of a married couple residing with or near the husband’s parents. Patrilineality is a kinship system in which an individual’s family membership derives from the father's lineage. 3 Other prominent studies on this topic include Merrill 2007; Simkhada et al. 2008; Char, Saavala, and Kulmala 2010; Shih and Pyke 2010; and Gangoli and Rew 2011. 4 Empirical evidence has extensively shown that in contexts where formal sources of information are missing, peers can play an important role in disseminating information about new health technologies (H. P. Kohler, Behrman, and Watkins 2001; Miguel and Kremer 2004; Godlonton and Thornton 2012), including family planning ( H.-P. Kohler, Behrman, and Watkins 2000, 2002; Behrman, Kohler, and Watkins 2002), employment opportunities (Munshi and Rosenzweig 2006), and agricultural technologies (Foster and Rosenzweig 1995; Conley and Udry 2010; Magnan et al. 2015), among other issues in developing countries (for a review, see Breza 2016). 5 Unlike other correlational studies on this topic, Varghese and Roy (2019) estimate the causal impacts of co-residence with the MIL on health during pregnancy. 6 Ethical approval to conduct the trial and all related study activities was received from the Northeastern University Institutional Review Board (protocol number 18-04-24) and from the University of Delhi Research Council. An informed consent form to participate in the study was provided to each woman that we contacted and only women who consented were recruited into Curse of the Mummy-ji the study. The trial was also registered at the American Economic Association Registry for randomized controlled trials on September 16, 2018 (AEARCTR-0003283). 7 Since we are interested in the MIL’s influence on women’s social networks, the relevant group for this article is the population of married women. Our sample-selection criteria mean that two types of married women are missing from our analysis: those who had no children and those who were older than 30 at the time of our baseline survey. The direction in which the omission of these two groups could bias our results is, a priori, unclear. On the one hand, childless married women are likely to have moved into their marital villages more recently than our sample women, and may, therefore, have fewer close outside peers and weaker bargaining power with respect to the MIL. On the other hand, women who are older than 30 are likely to have resided in their villages for a longer duration and, hence, may have more peers and greater bargaining power with respect to the MIL, in comparison to our sample women. 8 Specifically, close peers are individuals who are mentioned by the woman in response to the following question: “I would like to ask about the list of people, different from your husband and mother-in-law, with whom you talk about family planning, fertility, and reproductive matters and whose opinions are important to you. They are the people with whom you discuss your personal affairs or private concerns related to family planning, pregnancy, childbearing, and health.” 9 Chandrasekhar and Lewis (2011) show that a large downward bias can result when top-coding is used to sample peer networks. However, in our survey, all participants reported fewer than five close peers, besides their husband and MIL; in fact, the modal woman listed only one close peer. Thus, five appears to be an effective upper limit on network size in our sample. Curse of the Mummy-ji 10 We did not collect such information about general peers since the primary focus of our experiment was on women’s health, fertility, and family planning related networks. 11 The average number of close peers in the village among women who have at least one such peer is 1.14. 12 Although the estimates in Kandpal and Baylis (2019) are larger than ours, we note that our sample is younger and that Uttar Pradesh is a more conservative state relative to Uttarakhand, especially in terms of gender norms. 13 In our baseline data, all social network information about the peers is self-reported by the surveyed woman, and we are unable to construct the reciprocal links between women and their close peers, except for a few cases where the close peers turned out to be a part of our sample. It is plausible that these “out-degree” measures of social ties are less reliable as they cannot be corroborated by the “in-degree” measures, i.e., where others report having a link with the individual in question. Thus, it is worth noting that the variables used to characterize the close peers might induce some bias in our network depiction. 14 One concern with the inclusion of household landholdings as a control in specification (1) is that co-residence with the MIL may alter the amount of land the DIL’s sub-family owns. Therefore, in the Online Supplementary Appendix Tables A.1 and A.2 we show that the 𝛽 estimates from specification (1) are robust to the exclusion of the landholding variable. Nevertheless, we retain landownership as a control in our main tables as a proxy to control for a woman’s household economic status, which is likely to be a relevant determinant of her access to health services. Ideally, we would like to control for economic status using a covariate that is entirely independent of co-residence with MIL; however, we lack such data in our survey. Curse of the Mummy-ji 15 We do not make a causal claim here as the number of close peers in the household may be simultaneously determined with the number of close peers outside the household. 16 The influence of other adult women in the household becomes somewhat more restrictive when the MIL is absent from the household. This suggests that adult females-in-law in the household (e.g., older sisters-in-law) may become substitutes for the MIL upon her departure from the household as the enforcers of mobility restrictions on the younger DIL. Nevertheless, the negative influence of these adult females-in-law on the DIL’s number of close peers in the absence of the MIL is smaller than that of the MIL when she is present, implying that they are imperfect substitutes for the MIL. 17 The mobility index is the sum of the six indicator variables used in columns (1)-(6) of Table 3. The autonomy index is the sum of two indicator variables that equal one if the DIL has a say in decisions about her healthcare and about her visits to family or relatives. 18 We consider the area of landholdings as a post-treatment variable since co-residence with their MIL may influence the amount of land that the household owns. 19 We cannot estimate bootstrapped errors for columns (4)-(5) that control for caste-by-village fixed effects because we lack enough degrees of freedom due to our small sample size. 20 To assess the sensitivity of our estimates to the assumptions that we make about the pathways in Figure 1, we also estimated the ACDE under the “no intermediate confounders” assumption— the estimates are similar to those obtained after accounting for intermediate confounders using sequential g-estimation. Results are available upon request. 21 Following Acharya et al. (2016), the ACDE is identified under the following assumptions. First, conditional on pre-treatment covariates, there are not omitted variables for the effect of Curse of the Mummy-ji MIL-co-residence on the outcomes and no omitted variables for the effect of 𝐶𝑙𝑜𝑠𝑒𝑃𝑒𝑒𝑟𝑠𝑖 on the outcome, conditional on MIL-co-residence, pre-treatment covariates and intermediate confounders. Second, the effect of 𝐶𝑙𝑜𝑠𝑒𝑃𝑒𝑒𝑟𝑠𝑖 on the outcomes is independent of the intermediate confounders. 22 We divide the sample into three distinct caste groups: scheduled castes or tribes (SC-ST), OBC, and upper castes. As previously mentioned, 44 percent of the sample is OBC, 11 percent belongs to an upper caste, and the rest are SC-ST, with the ST share being negligible (1.5 percent). 23 Given the potential influence of co-residence with the MIL on the household’s landholdings, we show in Online Supplementary Appendix Table A.4 that our IV results are robust to the exclusion of this variable. 24 Our use of Propcv as a proxy for a woman’s potential local peer group is motivated by the work of Luke and Munshi (2006) who use a similar proxy to define a man’s available pool of brides in his marriage market. In their context, marriage is clan-based and exogamous (i.e., a man must marry outside his own ethnic clan or any related ethnic clan); so the authors create a concentration measure of local clan relatedness as an instrument for marriage in a location—if a higher fraction of clans in a marriage market are related to a man’s own clan, then his pool of eligible brides is smaller, and so on. 25 The exact definitions are as follows: 𝑃𝑟𝑜𝑝𝑐𝑣 1830 = 𝑁𝑐𝑣 1830 𝑁𝑣 1830 and 𝑃𝑟𝑜𝑝𝑐𝑣 = 𝑁𝑐𝑣 𝑁𝑣 , where 𝑁𝑣 1830 and 𝑁𝑐𝑣 1830 respectively denote the total number of 18-30-year-old women in the village and the number of such women who belong to the caste-group c, while 𝑁𝑣 and 𝑁𝑐𝑣 respectively denote the total number of women in the village and the total number of women who belong to castegroup c in the village. 26 Our use of co-residence with the MIL as a proxy for a woman’s access to her network is motivated by Posades and Vidal-Fernandez (2013), Debnath (2015), and Dhanaraj and Mahambare (2019) who employ a range of family structure-based instruments (e.g., death of the woman’s father-in-law, co-residence with the father-in-law, co-residence with other matriarchal figures (grandmothers), and joint family residence) as proxies for women’s access, mobility, potential for employment outside the home, and empowerment. 27 While the influence of peers on women’s family planning outcomes is likely to differ by peers’ beliefs and attitudes towards family planning as well as other observable characteristics, we are unable to robustly conduct heterogeneity analysis due to our relatively small sample. Moreover, peer characteristics are unlikely to be exogenous and peer influence is likely to be bidirectional—for instance, almost 79 percent of peers have both given family planning advice to and received family planning advice from our sample women. 28 The outcome variable is a categorical variable that takes values 0 to 6 with higher values indicating a woman’s belief that more women in her village use family planning. 29 Most of these interventions take place in Sub-Saharan Africa (Colleran and Mace, 2015; Institute of Reproductive Health, 2019).