Policy Research Working Paper 10441 The Effects of Communal Violence on Women’s Marital Outcomes Devaki Ghose Divya Pandey Development Economics Development Research Group May 2023 Policy Research Working Paper 10441 Abstract This study examines the effects of communal violence on study and synthetic control methods suggest that these women’s marital outcomes. Using individual-level survey effects are prominent two years after the riots and have data from India and a difference-in-differences approach, increased over time. Women who married after the riots the study shows that women’s age of marriage decreased, and also had fewer years of education and poorer social and the probability of getting married before age 18 increased economic status, such as a lower probability of employment after the Hindu-Muslim riots in Gujarat in 2002. Event and lower autonomy in household decision-making. This paper is a product of the Development Research Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at dghose@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team The Effects of Communal Violence on Women’s Marital Outcomes Devaki Ghose and Divya Pandey∗ Please find the most up-to-date version here Keywords: Gender, Early Marriage, Religion, Communal Violence, Riots JEL Classification: D74, J12, J16, O12, Z12 ∗ Affiliations: Ghose, Development Research Group, The World Bank, Washington DC USA; Pandey, In- ternational Initiative for Impact Evaluation (3ie), Delhi Office. We are immensely grateful to Anjini Kochar, Sheetal Sekhri, S. Anukriti, Tarun Jain, and participants of the Gender Brown Bag at Inclusion Economics for their thoughtful comments and feedback. We thank Jan Oledan for excellent research assistance. This paper is the third chapter of Divya Pandey’s dissertation (available here, and the author acknowledges the Bankard Fund for Political Economy for their financial support). The findings, interpretations, and conclu- sions expressed in this paper are entirely those of the authors and do not necessarily represent the views of 3ie and the World Bank and its affiliated organizations or those of the Executive Directors or the countries they represent. Any errors are the authors’ responsibility. 1 Introduction The disruptive impacts of violent conflicts are seldom gender-neutral. Women suffer in very specific ways during times of conflict as existing law and order break down, making them more susceptible to sexual violence and abuse. The Hindu-Muslim riots of 2002 in the Indian state of Gujarat were one such event in which not only were several killed, injured, and displaced, sexual violence against women had become its key feature (Jha and Farasat, 2016). While the direct consequences of violent conflicts on women include rape and other forms of abuse, its indirect consequences could involve changes in their marital outcomes and agency due to changes in socioeconomic and demographic characteristics of the conflict- affected regions (La Mattina, 2017; Gutierrez and Gallegos, 2016). This paper examines how women’s marriage age and post-marital socioeconomic status evolved in Gujarat after the 2002 riots. To examine the effects of the Gujarat riots of 2002 on women’s marital outcomes, we use a retrospective state-year level dataset from the fourth round of the National Family Health Surveys (2015-2016) and a difference-in-differences strategy. Our empirical strategy relies on variation over time in marriages that took place before and after 2002 and the state- wise variation in the occurrence of riots. We also use the event study approach to observe differential trends in women’s age at marriage between Gujarat and the other Indian states. The difference-in-differences estimates suggest that age at marriage declined for women in Gujarat relative to other Indian states after the riots of 2002. The probability of women getting married before 18 increased by 4.4 percentage points or 14 percent after the riots in Gujarat compared with the control states. Also, the probability of getting married before the age of 18 is relatively higher among Muslim women who were some of the primary victims of sexual violence during the riots. Our event study estimates suggest that the effects are prominent two years after the intercommunal violence, and the gap in terms of women’s age at marriage between Gujarat and the other Indian states has since increased. Our results are robust to synthetic control methods or restricting the control group to states that share their borders with Gujarat. We conduct a few robustness checks to allay concerns about other shocks within Gujarat driving our results. First, to address concerns that our results could be driven by the floods that severely affected some of the districts of Gujarat in 2005, we dropped the flood-affected districts from our sample. Dropping these districts does not significantly alter our results. Second, the massive earthquake that struck Gujarat in 2001 could have affected marital outcomes in the state (Das and Dasgupta, 2020). To address this challenge, we dropped districts in Gujarat from our sample affected by the earthquake and found that our results 2 remain unchanged. Finally, to address concerns regarding changes in political power driving our results, we restrict our sample to marriages that took place before the riots of 2002 and estimate the effect of party change in Gujarat in 1995 using our differences-in-differences strategy. We find no negative effect of changes in political power on women’s age at marriage. Next, to understand the implications of early marriage for women’s post-marital socioe- conomic status, we check education and autonomy outcomes for women married after the riots compared to women married before the riots. Existing literature has established that marrying young or exposure to conflict can decrease women’s bargaining power within the marriage (Field and Ambrus, 2008; Jensen and Thornton, 2003). We find that women mar- ried after the riots had fewer years of education, were more likely to experience domestic violence, had less control over their income, and had lower autonomy regarding household purchases and visits. Several mechanisms could explain our results. First, in the Indian culture, rape victims face the risks of being socially ostracized, which could worsen their position in the marriage market. Therefore, the fear inflicted by sexual violence on victims and their families could lead to early marriages of women. Second, distortions in sex ratios due to disproportionately higher mortality for men or women could also affect women’s position in the marriage market by altering their bargaining power (Becker, 1981; Teso, 2018; Abramitzky et al., 2011). Third, negative income shocks caused by the riots can also alter women’s marriage age due to the payment of dowry or bride prices at the time of marriage (Corno et al., 2020; Khanna and Kochhar, 2020). In this paper, we do not attempt to disentangle these different mechanisms but establish the effects of the riots on women’s age at marriage and explore its implications for women’s socioeconomic status and autonomy within the household. This study fits into the broad literature on the effects of violent conflicts. Several studies have analyzed the effect of ethnic conflict on economic growth (Abadie and Gardeazabal, 2003); human capital formation (Camacho, 2008; Duque, 2017); long-term biases in terms of lending behavior and judicial results (Hjort, 2014; Shayo and Zussman, 2011, 2017); and women’s employment rates and agency (Goldin and Olivetti, 2013; Zárate-Barrera and Ro- gall, 2020). This paper adds to the literature by examining the effects of intercommunal Hindu-Muslim riots on women’s marital outcomes in India. While there is a relatively large literature on the determinants of Hindu-Muslim riots (Jha, 2013, 2014; Nellis et al., 2016; van Noort and Goyal, 2020; Bohlken and Sergenti, 2010; Mitra and Ray, 2014; Field et al., 2008) and some work on the effects of these riots on electoral votes and lending (Iyer and Shrivastava, 2018; Ticku, 2015; Fisman et al., 2020), our paper is the first to study women’s marriage market outcomes following riots. This is particularly important and equally challenging to study as sexual violence against women 3 was a key feature of the Hindu-Muslim riots but abysmally under-reported (Khanna, 2008; Srivastava, 2011; Sen and Jasani, 2021). As pointed out by Srivastava (2011): “FIRs of women survivors were not lodged. In fact, the police refrained from registering rape as an individual crime. By clubbing rape with complaints of murder, sexual violence was made invisible across the state". Given how widespread the under-reporting or complete lack of crime reporting was, rather than studying reported crimes, we focus instead on women’s socioeconomic outcomes, such as marriage age, education, and autonomy inside a household that can be severely undermined through violence against women. Our paper also interacts with the literature on the determinants of early marriages of women. Most studies examine the effects of income shock associated with natural disasters on women’s age at marriage (Corno and Voena, 2016; Corno et al., 2020; Khanna and Kochhar, 2020). Corno et al. (2020) show that the interaction of social norms such as bride price and dowry and local income shocks caused by droughts increase the incidence of child marriage in Sub-Saharan Africa and reduce the same in India. While income shocks associated with natural disasters are important drivers of marriage market outcomes, violence against women during communal riots can also determine the incidence of child marriages. Our paper seeks to contribute to this literature by examining the effect of intercommunal conflict on women’s marital outcomes. Further, we also show that intercommunal conflict can hurt women’s socioeconomic outcomes, such as education and bargaining power within the marriage. Finally, this paper relates to the literature that examines the relationship between early marriages and women’s agency. Existing studies have shown that early marriages adversely impact women’s empowerment outcomes, such as educational attainment, health, economic condition, fertility and childbearing age, bargaining power, attitudes towards equitable gen- der relations, and children’s human capital (Field and Ambrus, 2008; Dahl, 2010; Sunder, 2019; Chari et al., 2017; Sekhri and Debnath, 2014; Jensen and Thornton, 2003; Raj et al., 2009; Yount et al., 2018; Crandall et al., 2016). While our paper does not directly isolate the effect of early marriage on women’s agency, our results suggest that communal vio- lence reduces women’s age at marriage and potentially adversely affects their socioeconomic well-being. The rest of this paper is structured as follows. Section 2 discusses the context. Sections 3 and 4 discuss the data and the empirical strategy. Sections 5 and 6 discuss our main results and their robustness. Section 7 concludes. 4 2 Background: Gujarat Riots of 2002 India has had a history of communal riots whereby violence was perpetrated by both Hindus and Muslims. The Hindu-Muslim violence peaked during the partition of India based on religion in 1947 and then settled down to periodic clashes centered on religious festivals and places of worship. These clashes are often determined by the state and nature of inter-communal civic engagements in the case of India (Varshney, 2008; Varshney and Gubler, 2012; Wilkinson, 2009). The Gujarat riots of 2002 erupted after a carriage carrying Hindu activists was set on fire in Godhra on February 27, 2002. While several theories circulate on who set the train on fire, the dominant one is that a Muslim mob set the train on fire due to an altercation between a Muslim tea seller and the Hindu activists and possibly the attempted molestation of the tea seller’s daughter (Khanna, 2008; Simpson, 2017). Several of the 58 people killed on the train were women and children. Followed by this incident, there was three-day-long violence perpetrated by Hindu mobs against the minority Muslim population. According to official figures, the riots killed 790 Muslims and 254 Hindus. However, unofficial statistics suggest that close to 2000 Muslims were killed over the next three months, and many others were injured (Amnesty Interna- tional, 2005). More than 200,000 Muslims were displaced because of the looting and burning of their homes and businesses (Srivastava, 2011). There were further outbreaks of statewide violence over the next year. Sexual violence against women was a key feature of the Gujarat riots of 2002. Several women were raped and killed. Survivors reported sexual violence such as rapes, gang rapes, forced nudity, torching, mutilation of the stomach and sexual organs, and insertion of objects into bodies (Hameed et al., 2002). According to Jha and Farasat (2016), Muslim women were not the only victims of sexual violence in the Gujarat carnage of 2002; Hindus were also targeted in some cases. Many of these cases were not reported to the police because of the fear of further attacks, stigma, and ostracism. During episodes of communal violence, women are often targeted because they are viewed as transmitters of the culture of a particular community. Exposure to such brutalities can have deep physical and emotional effects on the victims, their families, their communities, and their marital decisions. The fear of sexual violence and stigmatization can lead to early marriages of women. In this paper, we study the effects of the Gujarat riots of 2002 on women’s marital outcomes. 5 3 Data This study uses data from the fourth round of the National Family Health Surveys (NFHS-IV), based on the Demographic and Health Surveys (DHS), conducted in India dur- ing 2014-15. These are nationwide repeated cross-sectional surveys that are representative at the state level. The survey reports information on the year of marriage, age at mar- riage, age at cohabitation, age at the consummation of marriage, and age at first birth of ever-married women between ages 15 and 49. We construct a retrospective state-year-level dataset of marriages between 1990 and 2014 across Indian states using this information. This allows us to compare outcomes for women across Indian states married before and after the Gujarat riots of 2002. For a sub-sample of women, the survey also reports information on their socioeconomic statuses, such as their employment status, exposure to intimate partner violence, and their decision-making power, which we use to analyze the effect of the Gujarat riots on women’s post-marital outcomes. The variables used for analyzing women’s socioeconomic outcomes post riots include women’s years of education and indicators for whether they experienced physical or emotional violence by their husbands. We also include indicators for women’s employment status, autonomy over their income, whether they participate in decisions related to household purchases, and visiting relatives and friends. Table 1 reports the sample means of key variables for women married before 2002 across Gujarat and other Indian states. There are some differences between the characteristics of women married before 2002 in Gujarat and other Indian states. For example, in terms of religion, Gujarat has a higher fraction of the Hindu population and a lower fraction of Muslims. Gujarat is also wealthier on average than the other states. The average marriage age for women from Gujarat or other Indian states is approximately 18 years. While the fraction of women married before 18 is lower in Gujarat compared with other Indian states, the difference between the two groups is not statistically significant. 4 Empirical Strategy We use a difference-in-differences strategy to isolate the effect of the Hindu-Muslim riots on women’s marital outcomes. In particular, we rely on spatial variation due to the riots that erupted in the Indian state of Gujarat and the temporal variation in marriages before and after the riots of 2002. 6 4.1 Main Specification We first estimate the following equation: ′ Yist = α + βP ostt ∗ Riots + Xist θ + Y earF E + DistrictF E + ϵist (1) where Yist is the outcome of interest such as age at marriage, an indicator for marriage before age 18, age at cohabitation, years of education, a measure of autonomy or domestic violence for woman i in state s married in the year t; P ostt equals 1 if the woman was married after the year 2002 and 0 otherwise; Riots is an indicator for residence in the state of Gujarat;1 and Xist is a vector of controls that includes indicators for wealth, religion, caste, residence in a rural area, the gender of the household head, and household’s wealth. We also include marriage year fixed effects and district fixed effects. The coefficient of interest, β , measures, depending on the specification, the changes in women’s marital outcomes in Gujarat relative to the rest of India/ neighboring states/ synthetic Gujarat post the 2002 riots. The difference-in-differences estimate relies on the assumption of parallel trends. We check for differential trends in women’s age at marriage and the rate of child marriages by estimating the following event-study equation: ∑ 2014 ′ Yist = α + βt Y eard ∗ Riots + Xist θ + Y earF E + DistrictF E + ϵist (2) t=1991,̸=2002 where Y eard are year-specific dummies and the omitted year is the year 2002; βt represent the coefficients on interactions between the year dummies and the Riots dummy. Another advantage of estimating the difference-in-differences coefficients by year is that it enables us to investigate how women’s marital outcomes evolved over time after the riots. 4.2 Choice of treatment and control groups As discussed in section 4.1 we take Gujarat as the treated state and the rest of India or neighboring states or a synthetic amalgamation of states as the control. In principle, one can think of a within Gujarat analysis where we analyze whether women’s marriage market outcomes are correlated with the intensity of riots. There are two major issues with such an analysis: The first main issue is the high mobility of women across districts within the same 1 Since the survey only provides information on the state of residence, we are unable to account for women who migrated to another state due to marriage. However, this should not significantly affect our results because interstate migration in India is very low (Biswas, 2017). According to the 2011 census, the share of intra-state migration was about 88% and the share of inter-state migration was about 12%. 7 state, specifically for marriage. According to the 2001 Indian Census data, while the leading cause of migration for men is work, it is marriage for women: 70% of all women who migrate do so for marriage (Kone et al., 2018). In table 2, we see that, among those women who migrate for marriage, more than 70% cross district borders within their states; for Gujarat, this number is more than 80%. Since the available data only records the residence district of the women after marriage, a within Gujarat district-level analysis will be unable to account for the spillover from treated to control districts and vice versa. The second issue is the widespread prevalence of riots across Gujarat. According to contemporary news articles (Outlook India, 2002, 2022), 16 of 24 districts in Gujarat were affected, namely, Ahmadabad, Anand, Banas Kantha, Bharuch, Bhavnagar, Dohad, Gand- hinagar, Junagadh, Kheda, Mahesana, Panch Mahals, Patan, Rajkot, Sabar Kantha, Surat, and Vadodara. Varshney (2020) highlights that the cities of Ahmadabad, Vadodara, and Godhra (in the Panch Mahal district) experienced the worst violence. Given the widespread violence across districts of Gujarat and the high inter-district within state mobility of women for marriage, considering the whole of Gujarat as the affected region offers the most credible analysis. 4.3 Heterogeneity Next, we examine the effects of the riots by religion. The following equation identifies the effect of the riots by religion: Yist = α + γP ostt ∗ Riots ∗ M uslimi + βP ostt ∗ Riots + δP ostt ∗ M uslimi ′ (3) +ϕRiots ∗ M uslimi + Xi θ + Y earF E + DistrictF E + ϵist where M uslimi is equal to 1 if woman i belongs to the Muslim community. The coefficient of interest, γ identifies the additional effect of the riots on Muslim women’s marital outcomes relative to the outcomes of women of all other regions. We cluster all our standard errors at the state level or the state-year level in alternative specifications. 5 Results In this section, we discuss the effects of the Hindu-Muslim riots of Gujarat in 2002 on women’s age at marriage, and their post-marital social and economic status. 8 5.1 Women’s Marriage Age Using data from NFHS-IV and specification 1, Table 3 presents the changes in women’s age at marriage, probability of child marriage, age at cohabitation, age at consummation, and age at first birth post the 2002 riots compared to the rest of India. Columns (1) and (2) suggest that women’s age at marriage declined by 0.33 years or 4 months after the riots, and the probability of marriage before the age of 18 increased by 4.4 percentage points or 14 percent compared with the rest of India or the control mean. Columns (3) and (4) in Table 3 indicate that the age at cohabitation and age at first sex also declined after the riots. The estimates in the first four columns of Table 3 are significant at the 1 percent level. According to column (5) of Table 3, while statistically insignificant, the effect of violence on women’s age at first birth is negative. Figures 1(a) and 1(b) present coefficients using the event study specification in 2. Figure 1(a) suggests that the age at marriage decreased for women married after the Gujarat riots of 2002. Figure 1(b) suggests that the probability of child marriage increased for women married after 2002. Nevertheless, these effects are not immediate and are prominent after 2004. The coefficients on the years before the riots suggest some differences in a few years between women’s marriage age in Gujarat and the other Indian states even before 2002. Therefore, we will explore the same results using synthetic difference-in-difference and restrict our analysis to only bordering states in section 6. To put our findings in perspective, we compare our estimates to other studies which investigate the effect of natural disasters on marriage markets. Our estimate of a 4-month decrease in age at marriage and a 14 percent increase in the incidence of child marriages is in line with the findings of other studies that find a decline in women’s age at marriage due to natural disasters. Khanna and Kochhar (2020) find about a 4 months decline in age at marriage and a 9.5 percent increase in the incidence of child marriage due to floods, and Corno et al. (2020) find a 3 percent increase in child marriages in Sub-Saharan Africa due to droughts. 5.2 Marriage Age by Religion In Table 4, we present results by religion using specification 3. According to column (2) of 4, Muslim women are 3.1 percentage points more likely to get married before the age of 18 as a result of the riots. The results show that all the other outcomes, such as age at marriage, age at cohabitation, age at first sex, and age at first birth, declined for both Hindu and Muslim women, and there are no significant differences between the two groups. The additional effects of religion, even though negative and substantial for Muslim women, are 9 statistically insignificant. 5.3 Women’s Education In this section, we examine the effects of communal violence in Gujarat on women’s edu- cation outcomes. Table 5 presents estimates of the effect of the riots on women’s education using specification 1. According to column (1) of Table 5, women who were married after the riots in Gujarat have 5 percent fewer years of education than women in the rest of the country or the control average. In terms of years, women have 0.38 fewer years of education in Gujarat. Columns (3) and (4) show that women who married after the riots in Gujarat are 1.8 and 1.5 percentage points less likely to complete secondary and higher education. Our results are consistent with the literature that shows that women married at a younger age are more likely to enter marriages with fewer years of education (Field and Ambrus, 2008). 5.4 Post-Marital Social and Economic Outcomes Next, we examine post-marital social and economic outcomes of women who got married after the Gujarat riots of 2002 using specification 1. Columns (2) and (3) of Table 6 suggest that women who married after the riots in Gujarat were more likely to suffer emotional and physical violence from their husbands. Column (5) suggests that women are less likely to make decisions about visits to friends and relatives. In Table 7, we present the results on women’s economic status after marriage. Columns (1) and (2) of Table 7 suggest that women in Gujarat who were married after 2002 were less likely to be employed and were also less likely to control their income. Column (3) suggests that these women also had lower decision-making power in big household purchases. These findings are consistent with the literature that established that early marriage of women is associated with lower autonomy and bargaining power within the marriage (Jensen and Thornton, 2003), although communal violence can directly affect women’s sense of autonomy through changing notions of safety even outside marriage. 6 Robustness Checks 6.1 Gujarat and the Bordering States As evident from Table 1, Gujarat is significantly different from other Indian states re- garding cultural, social, and economic factors, which are important determinants of women’s 10 age at marriage and agency. Many of these factors are unobserved or lack credible proxies. In this section, we restrict our control group to the states that share their borders with the riots that affected Gujarat. The approach restricting the analysis to neighboring states has also been adopted in the previous literature to ensure that the control group consists of regions most similar to the treatment region (Anukriti, 2018). Figures A.1 and A.2 show the event-study coefficients using specification 2 after re- stricting the control group to the bordering states of Gujarat, which are Madhya Pradesh, Maharashtra, and Rajasthan. Figure A.1 shows no significant differences in age at marriage between Gujarat and its neighboring states before the riots of 2002. However, women’s age at marriage is consistently declining in Gujarat relative to its neighboring states, 2 years after the riots. According to figure A.2, the probability of women getting married before 18 increased after the Gujarat riots. 6.2 Synthetic Controls Approach The results from the event study approach (Figures 1(a) and 1(b)) suggest that there were some differential trends between Gujarat and the control states even before the riots of 2002 for a few years. To address this concern, we examine the effects of the Gujarat riots by employing the synthetic controls approach suggested by Abadie and Gardeazabal (2003). Using a combination of other Indian states or the donor pool, we first construct a synthetic Gujarat that best approximates the characteristics of women in Gujarat before the riots of 2002. We then use the post-riot outcomes for synthetic Gujarat to estimate the outcomes that would have been observed in Gujarat in the absence of the riots of 2002. Next, we compare the outcomes of synthetic Gujarat to actual Gujarat after 2002. We collapse the individual-level data in a state-year panel for the synthetic controls method. To approximate the outcomes (age at marriage and proportion of child marriages) at the state level, we use the proportion of women who reside in rural areas, reside in house- holds with male heads, belong to rich or poor households, are Hindus, and are scheduled castes/scheduled tribes/other backward classes. The results from the synthetic control meth- ods are presented in figures B.1 and B.2. Figure B.1 suggests that the age at marriage in Gujarat is identical to that in synthetic Gujarat before the riots of 2002 and begins to de- crease two years after the riots. These findings are similar to what we find using the event study method in figure 1(a). The probability of child marriage also increased after 2002 (Figure B.2). 11 6.3 Alternative Difference-in-difference estimator An emerging literature documents several problems associated with using the traditional difference-in-difference methods. However, since most of these problems arise in the case of staggered treatment timings, we have used the traditional difference-in-difference methods for our main estimation. The majority of the new methods are mostly applicable only with staggered treatment timing, with a few exceptions, such as Borusyak et al. (2021), which can be used in a cross-sectional setting without staggered adoption. In a non-staggered setting with never-treated units (such as all other states than Gujarat), the problem of negative weighting with forbidden comparison groups arises in very specific cases (Borusyak et al., 2021). In appendix C, we still repeat our main estimation using the Borusyak et al. (2021) method and show the effects of women’s marriage age, and the probability of child marriage remain very close to our main findings. This holds whether we use the rest of India as control (figures C.1 and C.2) or only the neighboring states of Gujarat as control (figures C.3 and C.4). 6.4 Gujarat Floods Some districts of Gujarat were severely affected by floods in 2005, and the existing literature suggests that natural disasters can affect marriage markets (Corno et al., 2020; Khanna and Kochhar, 2020).2 We drop flood-affected districts to address concerns regarding the floods driving our results. Our results are not significantly affected by dropping the flood- affected districts, which allays concerns about weather shocks driving our results (Figures D.1 and D.2; Table E.3). 6.5 Gujarat Earthquake The massive earthquake that struck the district of Kutch in Gujarat in January 2001 could be confounding our results. According to (Das and Dasgupta, 2020), the Gujarat earthquake reduced men’s and women’s ages at marriage. To address this concern, we drop districts that were affected by the earthquake of 2001 but not by the riots of 2002.3 Our results in table E.1 suggest that the effect of the Gujarat riots on women’s age at child marriage and child marriage remains large and statistically significant. 2 The flood-affected districts of Gujarat in 2005 were Vadodara, Ahmedabad, Navsari, Surat, Surendrana- gar, Anand, Kheda. 3 The earthquake-affected districts were Kutch, Ahmedabad, Jamnagar, Patan, Rajkot, and Surendrana- gar. We only drop Kachhch, Jamnagar, and Surendranagar since they do not overlap with the riots. 12 6.6 Change in Political Power Changes in political power during our analysis period could also confound our results. There was a change in political power in the year 1995 in Gujarat. Social norms, such as age at marriage within the state, could have been affected by the coming of a new political party based on the party’s ideologies. To address this concern, we restrict our sample to marriages before 2002 (before the riots) and estimate equation 1 using 1995 as the P ost period to estimate the effect of a change in political power on women’s age at marriage. Our results in table E.2 suggest no negative effect of change in the political party on women’s age at marriage, which alleviates concerns about the change in the party within Gujarat driving our results. 7 Conclusion In this paper, we have established that women’s age of marriage decreased, and the probability of child marriage increased in Gujarat after the riots of 2002 compared to the rest of India, neighboring states, or synthetic Gujarat. We show that years after the riots, women in Gujarat continued to be married off early and were less likely to finish secondary and higher education. Early marriage is often associated with lower education, worse household employment outcomes and bargaining power. Our findings show that post-2002, women in Gujarat have not only a lower probability of employment but also lower autonomy in social and economic decisions within the household. Our findings have a few limitations in terms of the inability to correlate our outcomes to the intensity of riots within Gujarat and the exact mechanisms that drive these results. First, due to the high migration rates of women for marriage across districts within Gujarat and the widespread prevalence of riots, we do not know whether the riots had worse effects in the directly affected areas. Second, the severe under-reporting of crimes related to sexual violence as documented in the literature (Palermo et al., 2014; Iyer et al., 2012; Sahay, 2021) restricts the paper’s ability to investigate the key mechanisms via which communal violence affects women’s martial outcomes. Given the effects of communal violence on women’s socioeconomic outcomes that we document, we believe that future research can enhance our understanding of the mechanisms that drive changes in women’s marriage market outcomes due to inter-communal conflict. This is especially challenging since violence against women is severely under-reported, and the social stigma associated with sexual violence is hard to measure. 13 References Abadie, A. and Gardeazabal, J. (2003). The Economic Costs of Conflict: A Case Study of the Basque Country. American Economic Review 93(1):113–132. Abramitzky, R., Delavande, A. and Vasconcelos, L. (2011). Marrying Up: The Role of Sex Ratio in Assortative Matching. American Economic Journal: Applied Economics 3(3):124–57. Amnesty International (2005). India: Justice, The Victim - Gujarat State Fails To Protect Women From Violence. ASA: https://www.refworld.org/docid/45c1cb702.html. Anukriti, S. (2018). Financial Incentives and the Fertility-Sex Ratio Trade-Off. American Economic Journal: Applied Economics 10(2):27–57. Becker, G.S. (1981). A Treatise on the Family. National Bureau of Economic Research, Inc. Biswas, N. (2017). Policy or Perception? Womens Political Liberalization and Sex-selection in India. Bohlken, A.T. and Sergenti, E.J. (2010). Economic growth and ethnic violence: An empirical investigation of Hindu Muslim riots in India. Journal of Peace Research 47(5):589–600. Borusyak, K., Jaravel, X. and Spiess, J. (2021). Revisiting event study designs: Robust and efficient estimation. arXiv preprint arXiv:2108.12419. Camacho, A. (2008). Stress and Birth Weight: Evidence from Terrorist Attacks. American Economic Review 98(2):511–15. Chari, A., Heath, R., Maertens, A. and Fatima, F. (2017). The causal effect of maternal age at marriage on child wellbeing: Evidence from India. Journal of Development Economics 127:42–55. Corno, L., Hildebrandt, N. and Voena, A. (2020). Age of Marriage, Weather Shocks, and the Direction of Marriage Payments. Econometrica 88(3):879–915. Corno, L. and Voena, A. (2016). Selling daughters: age of marriage, income shocks and the bride price tradition. IFS Working Papers, Institute for Fiscal Studies. Crandall, A., VanderEnde, K., Cheong, Y.F., Dodell, S. and Yount, K.M. (2016). Womens age at first marriage and postmarital agency in Egypt. Social Science Research 57:148–160. Dahl, G.B. (2010). Early Teen Marriage and Future Poverty. Demography 47(3):689718. 14 Das, S. and Dasgupta, S. (2020). Marriage Market Responses in the Wake of a Natural Disaster in India. Available at SSRN : https://ssrn.com/abstract=3266117. Duque, V. (2017). Early-life conditions and child development: Evidence from a violent conflict. SSM - Population Health 3:121–131. Field, E. and Ambrus, A. (2008). Early Marriage, Age of Menarche, and Female Schooling Attainment in Bangladesh. Journal of Political Economy 116(5):881–930. Field, E., Levinson, M., Pande, R. and Visaria, S. (2008). Segregation, Rent Control, and Riots: The Economics of Religious Conflict in an Indian City. American Economic Review: Papers and Proceedings 98(2):505–510. Fisman, R., Sarkar, A., Skrastins, J. and Vig, V. (2020). Experience of Communal Conflicts and Intergroup Lending. Journal of Political Economy 128(9):3346–3375. Goldin, C. and Olivetti, C. (2013). Shocking Labor Supply: A Reassessment of the Role of World War II on Women’s Labor Supply. American Economic Review 103(3):257–62. Gutierrez, I.A. and Gallegos, J.V. (2016). The Effect of Civil Conflict on Domestic Violence: The Case of Peru. RAND Working Paper Series WR-1168. Hameed, S., Manorama, R., Ghose, M., George, S., Naqvi, F. and Thekaekara, M. (2002). Citizens’ Initiative. Survivors Speak: How has the Gujarat Massacre Affected Minority Women? Fact-finding by a Women’s Panel. Outlook India : https://www.outlookindia. com/website/story/the-survivors-speak/215433. Hjort, J. (2014). Ethnic divisions and production in firms. The Quarterly Journal of Eco- nomics 129(4):1899–1946. Iyer, L., Mani, A., Mishra, P. and Topalova, P. (2012). The power of political voice: women’s political representation and crime in India. American Economic Journal: Applied Eco- nomics 4(4):165–193. Iyer, S. and Shrivastava, A. (2018). Religious riots and electoral politics in India. Journal of Development Economics 131:104–122. Jensen, R. and Thornton, R. (2003). Early Female Marriage in the Developing World. Gender Development 11:9–19. Jha, P. and Farasat, W. (2016). Splintered Justice: Living the Horror of Mass Communal Violence in Bhagalpur and Gujarat. Three Essays Collective. 15 Jha, S. (2013). Trade, institutions, and ethnic tolerance: Evidence from South Asia. Amer- ican Political Science Review 107(4):806–832. Jha, S. (2014). Unfinished business: Historic complementarities, political competition and ethnic violence in Gujarat. Journal of Economic Behavior & Organization 104:18–36. Khanna, M. and Kochhar, N. (2020). Do Marriage Markets Respond to a Natural Disaster? The Impact of Flooding of River Kosi in India. Available at SSRN : https://ssrn.com/ abstract=3644052. Khanna, R. (2008). Communal Violence in Gujarat, India: Impact of Sexual Violence and Responsibilities of the Health Care System. Reproductive Health Matters 16(31):142–152. Kone, Z.L., Liu, M.Y., Mattoo, A., Ozden, C. and Sharma, S. (2018). Internal borders and migration in India. Journal of Economic Geography 18(4):729–759. La Mattina, G. (2017). Civil conflict, domestic violence and intra-household bargaining in post-genocide Rwanda. Journal of Development Economics 124:168–198. Mitra, A. and Ray, D. (2014). Implications of an economic theory of conflict: Hindu-Muslim violence in India. Journal of Political Economy 122(4):719–765. Nellis, G., Weaver, M. and Rosenzweig, S.C. (2016). Do parties matter for ethnic violence? Evidence from India. Quarterly Journal of Political Science 11(3):249–277. Outlook India (2002). Mapping The Violence. Outlook India : https://www.outlookindia. com/website/story/mapping-the-violence/217990. Outlook India (2022). History of Communal Violence in Gujarat. Outlook India : https:// www.outlookindia.com/website/story/history-of-communal-violence-in-gujarat/217988. Palermo, T., Bleck, J. and Peterman, A. (2014). Tip of the Iceberg: Reporting and Gender- Based Violence in Developing Countries. American Journal of Epidemiology 179(5):602– 612. Raj, A., Saggurti, N., Balaiahand, D. and Silverman, J.G. (2009). Prevalence of child marriage and its effect on fertility and fertility-control outcomes of young women in India: a cross-sectional, observational study. The Lancet 373(9678):18831889. Sahay, A. (2021). The Silenced Women: Can Public Activism Stimulate Reporting of Vio- lence against Women? The World Bank : http://hdl.handle.net/10986/35245. 16 Sekhri, S. and Debnath, S. (2014). Intergenerational Consequences of Early Age Mar- riages of Girls: Effect on Childrens Human Capital. The Journal of Development Studies 50(12):1670–1686. Sen, A. and Jasani, R. (2021). Urban Hopes, Sexual Horrors. The Copenhagen Journal of Asian Studies 39(1). Shayo, M. and Zussman, A. (2011). Judicial Ingroup Bias in the Shadow of Terrorism. The Quarterly Journal of Economics 126(3):1447–1484. Shayo, M. and Zussman, A. (2017). Conflict and the Persistence of Ethnic Bias. American Economic Journal: Applied Economics 9(4):137–165. Simpson, E. (2017). Communal Violence, Forced Migration and the State: Gujarat since 2002, by Sanjeevini Badigar Lokhande. South Asia: Journal of South Asian Studies 40(2):421–422. Srivastava, A. (2011). ‘Engendering’ Communal Violence: An Note On The 2002 Gujarat Communal Riots. Proceedings of the Indian History Congress 72:1400–1404. Sunder, N. (2019). Marriage Age, Social Status, and Intergenerational Effects in Uganda. Demography 56(6):2123–2146. Teso, E. (2018). The Long-Term Effect of Demographic Shocks on the Evolution of Gender Roles: Evidence from the transatlantic Slave Trade. Journal of the European Economic Association 17(2):497–534. Ticku, R. (2015). Riot Rewards? Religious Conflict and Electoral Outcomes. Religious Conflict and Electoral Outcomes (May 1, 2016). Graduate Institute of International and Development Studies Working Paper (19). van Noort, S. and Goyal, T. (2020). Is Ethnic Violence Self-Perpetuating? Quasi- Experimental Evidence from Hindu-Muslim Riots in India. Quasi-Experimental Evidence from Hindu-Muslim Riots in India (April 20, 2020) . Varshney, A. (2008). Ethnic Conflict and Civic Life. Yale University Press. Varshney, A. (2020). Understanding Gujarat Violence. Social Science Research Council https://items.ssrc.org/from-our-archives/understanding-gujarat-violence/. Varshney, A. and Gubler, J.R. (2012). Does the State Promote Communal Violence for Electoral Reasons? India Review 11(3):191–199. 17 Wilkinson, S.I. (2009). Riots. Annual Review of Political Science 12(1):329–343. Yount, K.M., Crandall, A. and Cheong, Y.F. (2018). Womens Age at First Marriage and Long-Term Economic Empowerment in Egypt. World Development 102:124–134. Zárate-Barrera, T. and Rogall, T. (2020). Yes they can: Genocide, political participation, and female empowerment. HiCN Working Paper No. 338 338. 18 Figures Figure 1. Women’s Marriage Age/Probability of Women’s Marriage Before Age 18 Post the Gujarat riots of 2002 (a) Womens Marriage Age .5 Diff-in-diff Coefficients -.5 -1 0 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Marriage Year Data Source: NFHS-4 (b) Probability of Womens Marriage Age Before Age 18 .15 .1 Diff-in-diff Coefficients 0 .05 -.05 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Marriage Year Data Source: NFHS-4 Note : This figure plots the difference-in-differences estimates from specification 2 using NFHS-IV. In sub- figure (a), the outcome variable is women’s age at marriage. In subfigure (b), the outcome variable is an indicator for women’s marriage before age 18. Standard errors were clustered at the state-year level for both subfigures. 19 Tables Table 1. Sample Characteristics of Women Married Before 2002 (1) (2) (2) - (1) Other States Gujarat Control Variables: Hindu 0.75 0.92 0.16*** Muslim 0.12 0.07 -0.05** SC/ST/OBC 0.75 0.74 -0.01 Male Household Head 0.87 0.89 0.02* Wealth 2.90 3.35 0.45** Rural Resident 0.71 0.64 -0.07*** Outcome Variables: Women’s Age at Marriage 17.72 17.98 0.26 Women’ Child Marriage 0.49 0.43 -0.05 Women’ Age at Cohabitation 17.97 18.24 0.27 Women’ Age at First Intercourse 18.03 18.13 0.09 Women’s Age at First Birth 20.37 20.59 0.22* Women’s Years of Education 4.71 5.17 0.46 Husband Controls 0.48 0.47 -0.02 Husband is Emotionally Violent 0.14 0.11 -0.03** Husband is Physically Violent 0.33 0.21 -0.12*** Makes Health Decisions 0.78 0.78 -0.01 Employed 0.90 0.86 -0.04*** Controls Own Income 0.84 0.79 -0.05*** Makes Household Purchase Decisions 0.78 0.79 0.02* Makes Decisions About Visits 0.78 0.81 0.02* Men’s Age at Marriage 22.35 21.40 -0.95** No. of women married before 2002 174,226 6,104 No. of women married after 2002 227,593 7,518 Note : This table reports simple differences in means of the characteristics of women living in Gujarat and other Indian states using data from the NFHS-IV. The total sample represents the observations used for the age at marriage and education analysis, and the sub-sample represents observations used for analyzing women’s socioeconomic status. * Significant at 10%; ** Significant at 5%; *** Significant at 1%. 20 Table 2. Summary Statistics: Migration for Marriage Purposes State No. of migrants Percentage Out of district, same state Percentage Out of state Percentage Gujarat 1,937,395 4 1,560,660 81 376,735 19 Neighbors of Gujarat 10,416,132 24 8,062,937 77 2,353,195 23 Rest of India 31,629,202 72 23,066,868 73 8,562,334 27 Total 43,982,728 100 32,690,404 74 11,292,264 26 Note: This table reports the number and percentages of women migrating out of state and out of the district (same state) in Gujarat, its neighboring states, and the rest of India. Data source is the 2001 Indian Census. Table 3. Women’s Marriage Outcomes Post the Gujarat Riots of 2002 (1) (2) (3) (4) (5) Age at Married Before Age at Age at Age at Marriage Age 18 Cohabitation First Sex First Birth Post X Riot -0.330∗∗∗ 0.044∗∗∗ -0.271∗∗∗ -0.398∗∗∗ -0.030 (0.085) (0.016) (0.074) (0.062) (0.083) Observations 415441 425254 405697 418219 389744 R-squared 0.248 0.164 0.216 0.228 0.128 Control Average 19.57 0.312 19.68 19.52 21.34 District Fixed Effects Yes Yes Yes Yes Yes Year Fixed Effects Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Note : This table reports coefficients from specification 1. Standard errors are clustered at the state level. Control Variables include indicators for religion, sc/sct/obc, rural residence, wealth quintiles, and male household head. * Significant at 10%; ** Significant at 5%; *** Significant at 1%. 21 Table 4. Women’s Marriage Outcomes, by Religion Post the Gujarat Riots of 2002 (1) (2) (3) (4) (5) Age at Married Before Age at Age at Age at Marriage Age 18 Cohabitation First Sex First Birth Post X Riot -0.326∗∗∗ 0.043∗∗∗ -0.261∗∗∗ -0.395∗∗∗ -0.021 (0.081) (0.015) (0.070) (0.056) (0.080) Post X Riot X Muslim -0.155 0.031∗ -0.149 -0.065 -0.160 (0.234) (0.018) (0.243) (0.205) (0.253) Observations 415441 425254 405697 418219 389744 R-squared 0.247 0.164 0.216 0.228 0.128 Control Average 19.47 0.323 19.49 19.41 21.04 District Fixed Effects Yes Yes Yes Yes Yes Year Fixed Effects Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Note : This table reports coefficients from specification 3. Standard errors are clustered at the state level. Control Variables include indicators for religion, sc/sct/obc, rural residence, wealth quintiles, and male household head. * Significant at 10%; ** Significant at 5%; *** Significant at 1%. Table 5. Women’s Educational Outcomes Post the Gujarat Riots of 2002 (1) (2) (3) (4) Years of At least Completed At Least Completed Completed Education Primary School Secondary School Higher Education Post X Riot -0.388∗∗∗ -0.016 -0.018∗∗ -0.016∗ (0.096) (0.010) (0.008) (0.008) Observations 423957 423957 423957 423957 R-squared 0.483 0.330 0.369 0.211 Control Average 7.405 0.781 0.648 0.130 District Fixed Effects Yes Yes Yes Yes Year Fixed Effects Yes Yes Yes Yes Controls Yes Yes Yes Yes Note : This table reports coefficients from specification 1. Standard errors are clustered at the state level. Control Variables include indicators for religion, sc/sct/obc, rural residence, wealth quintiles, and male household head. * Significant at 10%; ** Significant at 5%; *** Significant at 1%. 22 Table 6. Women’s Social Status After Marriage Post the Gujarat Riots of 2002 (1) (2) (3) (4) (5) Controlled by Emotional Physical Makes Health Makes Decisions Husband Violence Violence Decisions about Visits Post X Riot 0.003 0.028∗∗∗ 0.045∗∗∗ -0.001 -0.019∗∗ (0.004) (0.003) (0.005) (0.006) (0.007) Observations 55133 55162 55162 70700 70700 R-squared 0.135 0.0557 0.124 0.0643 0.0760 Control Average 0.474 0.118 0.280 0.740 0.730 District Fixed Effects Yes Yes Yes Yes Yes Year Fixed Effects Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Note : This table reports coefficients from specification 1. Standard errors are clustered at the state level. Control Variables include indicators for religion, sc/sct/obc, rural residence, wealth quintiles, and male household head. * Significant at 10%; ** Significant at 5%; *** Significant at 1%. Table 7. Women’s Economic Status After Marriage Post the Gujarat Riots of 2002 (1) (2) (3) Employed in Controls own Makes Household Last 12 months Income Purchase Decisions Post X Riot -0.017∗∗∗ -0.022∗∗∗ -0.050∗∗∗ (0.002) (0.007) (0.004) Observations 423957 16043 70700 R-squared 0.0208 0.108 0.0687 Control Average 0.871 0.817 0.717 District Fixed Effects Yes Yes Yes Year Fixed Effects Yes Yes Yes Controls Yes Yes Yes Note : This table reports coefficients from specification 1. Standard errors are clustered at the state level. Control Variables include indicators for religion, sc/sct/obc, rural residence, wealth quintiles, and male household head. * Significant at 10%; ** Significant at 5%; *** Significant at 1%. 23 A Event Study Approach Using Gujarat and Its Bor- dering States Only Figure A.1. Women’s Marriage Age Post the Gujarat Riots of 2002 .5 Diff-in-diff Coefficients -.5 -1 0 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Marriage Year Data Source: NFHS-4 Note : This figure plots the difference-in-differences estimates from specification 2 using NFHS-IV. The outcome variable is women’s age at marriage. The control states only include the bordering states of Gujarat, which are Maharashtra, Rajasthan and Madhya Pradesh. Standard errors were clustered at the state-year level. 24 Figure A.2. Women’s Marriage Before Age 18 Post the Gujarat Riots of 2002 .2 Diff-in-diff Coefficients 0 -.1 .1 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Marriage Year Data Source: NFHS-4 Note : This figure plots the difference-in-differences estimates from specification 2 using NFHS-IV. The outcome variable is an indicator of women’s marriage before age 18. The con- trol states only include the bordering states of Gujarat, which are Maharashtra, Rajasthan and Madhya Pradesh. Standard errors were clustered at the state-year level 25 B Synthetic Controls Figure B.1. Synthetic Controls: Women’s Age at Marriage Post the Gujarat Riots of 2002 21 20 age_marriage 19 18 17 1990 1995 2000 2005 2010 2015 year gujarat synthetic gujarat Note : This figure plots the estimates using synthetic controls approach and NFHS-IV. The outcome variable is women’s average age at marriage. 26 Figure B.2. Synthetic Controls: Effect of Gujarat Riots on Women’s Marriage Before Age 18 .5 .4 child_marriage .3 .2 .1 1990 1995 2000 2005 2010 2015 year gujarat synthetic gujarat Note : This figure plots the estimates using synthetic controls approach and NFHS-IV. The outcome variable is the fraction of women who got married before age 18. 27 C Alternative Difference-in-difference Estimates Figure C.1. Women’s Marriage Age Post the Gujarat Riots of 2002: Borusyak et al. (2021) Estimation .5 Diff-in-diff Coefficients -.5 -1 0 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Marriage Year Data Source: NFHS-4 Note : This figure plots the difference-in-differences estimates from specification 2 using NFHS-IV. The outcome variable is women’s age at marriage. The control states include all other states of India. Standard errors were clustered at the state-year level. 28 Figure C.2. Probability of Women’s Marriage Before Age 18 Post the Gujarat Riots of 2002: Borusyak et al. (2021) Estimation .1 Diff-in-diff Coefficients -.05 0 -.1 .05 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Marriage Year Data Source: NFHS-4 Note : This figure plots the difference-in-differences estimates from specification 2 using NFHS-IV. The outcome variable is an indicator of women’s marriage before age 18. The control states include all other states of India. The control states only include the bordering states of Gujarat, which are Maharashtra, Rajasthan and Madhya Pradesh. Standard errors were clustered at the state-year level 29 Figure C.3. Women’s Marriage Age Post the Gujarat Riots of 2002: Borusyak et al. (2021) Estimation Using Neighboring States as Control 1 .5 Diff-in-diff Coefficients -.5 0 -1 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Marriage Year Data Source: NFHS-4 Note : This figure plots the difference-in-differences estimates from specification 2 using NFHS-IV. The outcome variable is women’s age at marriage. The control states only include the bordering states of Gujarat, which are Maharashtra, Rajasthan and Madhya Pradesh. Standard errors were clustered at the state-year level. Standard errors were clustered at the state-year level. 30 Figure C.4. Women’s Marriage Before Age 18 Post the Gujarat Riots of 2002: Borusyak et al. (2021) Estimation Using Neighboring States as Control .2 Diff-in-diff Coefficients 0 -.1 .1 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Marriage Year Data Source: NFHS-4 Note : This figure plots the difference-in-differences estimates from specification 2 using NFHS-IV. The outcome variable is an indicator of women’s marriage before age 18. The con- trol states only include the bordering states of Gujarat, which are Maharashtra, Rajasthan and Madhya Pradesh. Standard errors were clustered at the state-year level 31 D Dropping Flood Districts Figure D.1. Women’s Marriage Age Post the Gujarat Riots of 2002 .5 Diff-in-diff Coefficients 0 -.5 -1 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Marriage Year Data Source: NFHS-4 Note : This figure plots the difference-in-differences estimates from specification 2 using NFHS-IV dropping all flood affected districts. The outcome variable is women’s age at marriage. The control states only include the bordering states of Gujarat, which are Maha- rashtra, Rajasthan and Madhya Pradesh. Standard errors were clustered at the state-year level. 32 Figure D.2. Probability of Women’s Marriage Before Age 18 Post the Gujarat Riots of 2002 .2 Diff-in-diff Coefficients .1 0 -.1 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Marriage Year Data Source: NFHS-4 Note : This figure plots the difference-in-differences estimates from specification 2 using NFHS-IV dropping all flood affected districts. The outcome variable is an indicator of women’s marriage before age 18. The control states only include the bordering states of Gujarat, which are Maharashtra, Rajasthan and Madhya Pradesh. Standard errors were clustered at the state-year level 33 E Additional Tables Table E.1. Women’s Marriage Age (Dropping Earthquake Districts) Post the Gujarat Riots of 2002 (1) (2) Age at Married Before Marriage Age 18 Post X Riot -0.279∗∗∗ 0.061∗∗∗ (0.102) (0.016) Observations 93826 95530 R-squared 0.219 0.151 Control Average 18.63 0.388 District Fixed Effects Yes Yes Year Fixed Effects Yes Yes Controls Yes Yes Note : This table reports coefficients from specification 1 only including Gujarat and its neighboring states. Standard errors are clustered at the state-year level. Control Variables include indicators for religion, sc/sct/obc, rural residence, wealth quintiles, and male houshold head. * Significant at 10%; ** Significant at 5%; *** Significant at 1%. 34 Table E.2. Effect of Political Power Change on Women’s Marriage Age (1) (2) Age at Married Before Marriage Age 18 Post 1995 X Gujarat 0.180∗ 0.013 (0.107) (0.017) Observations 42046 43237 R-squared 0.152 0.102 Control Average 16.96 0.560 District Fixed Effects Yes Yes Year Fixed Effects Yes Yes Controls Yes Yes Note : This table reports coefficients from specification 1 only including Gujarat and its neighboring states. Standard errors are clustered at the state-year level. Control Variables include indicators for religion, sc/sct/obc, rural residence, wealth quintiles, and male household head. * Significant at 10%; ** Significant at 5%; *** Significant at 1%. Table E.3. Women’s Marriage (Dropping Flood Districts) Post the Gujarat Riots of 2002 (1) (2) Age at Married Before Marriage Age 18 Post X Riot -0.346∗∗∗ 0.068∗∗∗ (0.089) (0.016) Observations 91781 93416 R-squared 0.219 0.151 Control Average 18.63 0.388 District Fixed Effects Yes Yes Year Fixed Effects Yes Yes Controls Yes Yes Note : This table reports coefficients from specification 1 only including Gujarat and its neighboring states. Standard errors are clustered at the state-year level. Control Variables include indicators for religion, sc/sct/obc, rural residence, wealth quintiles, and male household head. * Significant at 10%; ** Significant at 5%; *** Significant at 1%. 35