 In this chapter that's titled force migration and attitudes towards domestic violence, I explored the effects of force migration in the case of Turkey on women's attitudes towards domestic violence. And let me start by giving a more general motivation. So by now we have a large body of evidence showing that civil war and conflict can have long-run consequences on various economic outcomes and human capital accumulation. But we have relatively little evidence on the long-term effects of force migration, which affects millions of people in the world today. So the most recent official data I could find was from UNHCR going back to 2015 and even then there were 60 million official displaced people and slightly more than half of them were actually internally displaced people, which we have even less evidence from. So of course now those numbers are much bigger yet we know little about the long-run effects of force migration on economic outcomes. Also we have very little evidence on effects on cultural norms and attitudes. So the question that I try to explore in this chapter is can conflict induced force migration have effects on gender norms in the long-run? And if so, through what mechanisms could these effects take place? So I explored this in the case of the force migration that resulted from the conflict between the Turkish state and Kurdish militia in the mid-1980s until the end of the 90s. And during this period, of course, the type of force migration that might result from a conflict can vary from different contexts. So I think it's important to pin down exactly what type of migration I'm studying here. So the type of migration was mainly from rural to urban areas and targeted towards the Kurdish population. So during this period, and I will give much more detail about the context in a few slides, the main impact was that Kurdish villages were destroyed for various reasons and according to official figures around a million people were displaced and mainly from going from rural to urban centers. So to study the effects of this on attitudes towards domestic violence, I used two data sources. And the main one is using two waves of the Turkish demographic health survey. And I used the two most recent waves. There are four waves. I used 2008 and 2013. And the reason I use these waves is because there I have detailed information on three dimensions that are key to my identification strategy, which is based on a triple difference in difference along these dimensions. So the first one is ethnicity. And I used the information on the mother tongue of the respondents. And the main groups are Kurdish versus Turkish in the sample. The second one, the second and third one are based on this migration module, which wasn't done in the other waves of the DHS. But in these last two waves, there is a detailed module about the migration history, which records both the place of origin, the birthplace and also the place of childhood until age 12. And then every movement for any location where the respondents spent at least six months. So that enables me to identify the time and the direction of the movement, which is important for the identification strategy that I use in the analysis. I also have a unique data set that I use more for supportive evidence due to various limitations that I'll come back to. And this data I acquired it from an NGO, one of the largest NGOs in the country, that aims at helping women that suffer from domestic violence. Due to anonymity reasons, I'm not going to name that, but it's a large NGO and the data covers essentially the universe of applications that they received from 2009 and 2011. So of course, this has lots of limitations. That's why I'm only going to claim that it's some supportive evidence. But I think it's also important to look at it and I'll come back to that later. So, of course, the first question is, why should forced migration affect women's attitudes towards domestic violence? And I think Roberto did a great job in summarizing the literature on intrasol bargaining. And one potential mechanism might be related to that. And in a standard cooperative bargaining model, we would expect that if for any reason migration improved women's outside option, then that should lead to a drop in the rate of domestic violence and potentially longer and also about expectations and acceptability of domestic violence by both women and men. But in the data, I only have women. That's why I keep preparing to women's attitudes towards it. But on the other hand, we know there's a growing body of literature as Roberto summarized that non-corporative bargaining models seem to be, at least in the developing countries, settings more and more relevant. And in fact, if we model domestic violence under such a non-corporative bargaining situation, it's not so clear that an increasing women's outside option will necessarily lead to a reduction in domestic violence. And in fact, literature outside of economics and sociology has for a long time written about the male backlash theory that if women's autonomy improves, this might create a backlash from the men and that might result in more domestic violence. So theoretically, it's not clear which way the effect might go, but to the extent that if forced migration have had an impact on women's outside option relative to men, then it is likely that it might have had an impact on domestic violence and the attitude about domestic violence. So as I mentioned before, going back to the Turkish context, the main movement was from forcing populations to move from rural to urban centers. And both in my data and in general in other studies that have looked at female and male labor force participation rates, it seems that in Turkey, the stylized facts is that in rural areas, female labor force participation and employment rates are relatively higher compared to urban areas, and especially this is the case for women with low schooling. On the other hand, for men, we see a smaller difference between the rural and urban, but if anything, the difference is the other way around. So in the urban areas, male labor force participation rates and employment rates are higher. So unemployment rates for women have been shown to be higher again in the urban sector and especially for unskilled labor. So in that sense, it is not unreasonable to expect that this forced migration may have reduced women's employment opportunities once they were forced to move to the urban centers and most of them either don't have any formal schooling or have very little. So they were definitely in the unskilled labor. So if we go by just the simple rates of employment rates in rural urban areas for low-skilled labor, it is likely that women's employment rates were lower and in fact I will test and show some evidence in line with that in my dataset as well. So in that sense, it seems like the bargaining power mechanism might be active and that could be one potential channel. Another channel might be through cultural diffusion. So in Turkey, if we again look at the dataset, for example, the DHS data, we see that in urban areas on average respondents are less likely to say that domestic violence is acceptable compared to rural areas. So in some sense, in urban areas, on average, there is more, we can say there's more progressive gender norms about domestic violence. So in that sense, since this movement forced populations to go from villages to the urban centers, it might have led to a cultural diffusion whereby the migrants' norms eventually change in the long run and might become also more not accepting towards domestic violence. So this would go in a positive way, making domestic violence less acceptable. So what I find in the data is that I find that Kurdish women who migrated during the conflict from the conflict region, who were from the conflict region and migrated during the conflict, they are 10 to 15 years later when they are surveyed, they are significantly more likely to say that they think domestic violence is acceptable under at least one of the situations they are asked about. And the magnitude of the effect is really large. So it's around 50% relative to the group I use as the counterfactual in the difference in difference analysis. So it's a big increase in their rate of finding, reporting that domestic violence is acceptable. And when I try to look at the mechanisms, I find some evidence in line with the bargaining power mechanism. In particular, I find that forced migrant women using the same identification strategy, they are more likely to be in a relationship where their spouse is working but they are not working. So the gap in the employment rates between men and women in these households has increased and this is again in the very long run. So unfortunately I can't look at the immediate effect when the migration took place but I can only look at it at the time of the survey and it seems like there is still some significant difference between men and women's employment rates. And also there's more poverty still so they are on average below the middle class in terms of the wealth score that DHS assigns to households. So both the poverty has increased and also the women's bargaining position relative to men seems to have decreased. And I use the second dataset from the shelter NGO. So this dataset has, as I said, various limitations and the main one is that it's of course a selected sample of women who decided to apply to this shelter. So they decided to seek help for domestic violence. The advantage of the data is that it is first of all reported by the applicant to the center. So in terms of the reporting bias that we typically expect to see in domestic violence cases even in criminal records, it's relatively reasonable to think that what they report there is quite truthful and there's relatively detailed information about the case, the type of the violence and the duration of the relationship in which they were subjected to it. And when I compare, controlling for as many things as I have information on, the forced migrants to the rest of the applicants, I find that forced migrants have been in the relationship for a longer period before applying to the NGO. And they are also less likely to have sought other types of help compared to the others. And they also seem to suffer more extensive violence. So on the intensive margin, they have suffered both for a longer period and also more severely before deciding to come to the organization. So that, again, as I said, is suggestive. There are various shortcomings that we can discuss, but it supports the idea that the forced migration may have reduced their likelihood to not find domestic violence acceptable and to seek help quickly. So overall, both findings are in line with that. So in terms of the related literature, first of all, it's related to a large and growing literature on the interrelation between conflict, civil war and conflict, and domestic violence. And here there's little but growing literature that tries to estimate these effects. I think in particular, one paper by Julia Lamatina that was recently published in the Journal of Development Economics is quite closely related. And there she looks at the effects of the Rwandan genocide, the longer effects on, in terms of domestic violence rates. And she finds similar evidence. But there the mechanism is very different. She looks at the big effect on the sex ratios in the population and finds that where women's sex ratios have fallen down, the domestic violence rates have increased more, which seems to suggest that the bargaining power through the marriage market is more relevant. So in the case that I'm studying, the direct effect on sex ratios is likely to be not so large. The number of deaths was nowhere as large as in the case of Rwanda. But other, the bargaining power effect might again be there through the employment mechanism that I just described. And then it's also related to a larger literature that looks at the interconnection between women's bargaining power and domestic violence. So on the empirical side, as Roberto also mentioned, the evidence is like if we look at it globally, it's mixed. But if we try to make a general conclusion, we can, I think, say fairly that in developed countries or high income settings, typically when women's outside option improves, domestic violence seems to go down, whereas in low-income countries, it's the other way around. So Turkey is, of course, kind of in the middle. And in that sense, I think the two papers I mentioned here by Manuela Angelucci and Bobonis Atal from Mexico are probably like if we just go by rough GDP per capita rates are the most comparable. And there they find, on average, that an improvement in women's bargaining power, which they are exploiting the conditional cash transfer program that was targeted to the women in the household, on average, is associated with a fall in domestic violence. But the effect is very heterogeneous. And depending on the size of the transfer, Manuela Angelucci shows that it might be the other way around if the transfer is very large. But on average, it's in line with the thing, the results I'm finding also in the Turkish case. Of course, I'm not looking at the rate of domestic violence, but if we can think that the two are correlated, then it's similar. Sorry, here's mine. Finally, I think it's also related to a growing literature on the connection between the effects of migration and cultural diffusion of cultural norms. So there, of course, this is difficult to estimate the effects of migration on the migrant's attitudes, because migration is typically very selected and endogenously so. So migrants, people who decide to migrate presumably do so, also taking into account the existing cultural norms in the places they choose to go. But in the case of forced migration, I think this selection effect is less so, or if anything, it's not there at all. So in that sense, exploiting forced migration might be kind of a methodologically convenient way to also estimate the effects of it on the migrants' attitudes. Okay, so for the rest, I'm going to first give you a bit more context about the migration that I'm studying here, and then talk about the data results, so yeah, pretty standard. Okay, so in terms of the context, so the conflict here was mainly between the Turkish army and the state and the Kurdistan Workers' Party, or PKK. And even though PKK was founded before towards the end of the 70s, the violence really escalated in the mid-80s, and from 84 until 99, it continued at a very high intensity level. And then it kind of died down with the capture of Öcalan, Abdullah Öcalan, the leader of PKK during that period. But then, as we know recently, it's escalated again, but I'm not studying the recent wave. I'm looking at this period, and during this period, there were many villages that were either evacuated or destroyed. And according to the official parliamentary report from 98, this report cites three main reasons. The order that I listed them here is the order that's in the report, and I'm not sure if that order is correlated with the relevance of that reason by any means. But the three reasons mentioned in the report is one is the collapse of agriculture and animal husbandry, which are the main economic sources of income for the livelihoods there. The second one is that they're saying that PKK was evicting villagers and destroying villages that were cooperating with the state. And the state was also evicting villages that were not cooperating with the state or cooperating with the PKK. So which one of these was the main driver? As I said, there's no official figure on that, so I don't wanna make a judgment personally, but all three of these effects were there. And in the end, again, the official figure is there was between 900,000 and 1.2 million people that were displaced because of that. And again, recent information shows that even as recent as 2009, only 20% of them had returned back to their villages. So in terms of the location, so this is the map of Turkey and the provinces that are highlighted green is the area where the high intensity conflict was taking place, although there were also terrorist attacks and other types of maybe more minor conflicts in other parts, especially in the surrounding provinces. But these states were really where the conflict was happening and the most of the migration was also taking place. They were also declared by the government to be under a state of emergency all throughout this 8499 period, which gave the government more rights. And so the evacuations were really taking place there. So I'm going to compare essentially people from this area and migrated from this area with people from the other provinces. So the blue ones are the ones where, again, according to the official report, most of the migrants went and settled. But also in this area, as I said from the villages, a lot of them settled in the town, in the province centers. So the data that I'm using is coming from the DHS. And as I said, the main advantage here is that there was the migration module that enables me to do the analysis that I'm conducting. The respondents were, for the last way, all women age 15 to 50. And as many of you know, I'm sure the DHS surveys that age group because of reproductive modules, etc., are mainly targeted to that age group. And then in the 2008 wave, they only surveyed ever married women. I'm going to pull both data sources mainly because I need a lot of power to do the triple difference analysis. But I control for, of course, survey wave effects and also the marital status to make sure that the results are not driven by this restriction. And for my main outcome is I'm using this module on the respondents' attitudes towards domestic violence, which exists in most DHS waves. Some DHSs also ask about direct incidents of domestic violence. Unfortunately, in Turkey, this was not included. So I only have the attitudes towards domestic violence. And here, there were seven different situations that the respondents were asked about in 2008. And then in 2013, they limited that to the first five. Again, I'm going to use both, but the results are not really driven by the last two. So if I exclude those two and only focus on the five, I find similar effects. And the situations, I mean, I don't think they are directly important, but it's things like if she burns the food, if she neglects the children, etc. But I use all of these things to build indices in three different ways. The first index will be whether the respondents thought that domestic violence was acceptable under any scenario. And the second will be the fraction of scenarios. And the third will be a principal component. But I think it's fair to say that the first index, that it shouldn't be accepted under any scenario. It's probably the one that is most relevant. And finally, the ethnicity is the third important dimension that I look at. So these are just some descriptive statistics from the entire sample. And then if I limit the sample to people that grew up in these conflict provinces, and you can see that the sample goes down a lot. And then if I also look at the ethnic division within the sample, people from the conflict region, a larger share of them are Kurdish because that's where the Kurdish population is mainly based. And in the other regions, there are way fewer Kurdish respondents. And let me see. So in terms of the main outcome of interest, on average 20% of respondents reported that they think domestic violence is accepted under at least one of these situations. And you can see that the rate is higher for respondents from the conflict region. And especially when we look at people that migrated during, it's even higher. It's 38%. Of course, that could be due to lots of things. So it's important to control for as many predetermined covariates as I can. So I'm going to control for things like their parental education, whether their parents were related. So basically anything that I have that are likely to be predetermined to the migration, I will control for in the specification. And so the main specification that I estimate is, as I mentioned, is a triple difference and difference model, where the differences come from the ethnicity of the respondent, whether she grew up in the conflict region, and then if the migration happened between during this conflict period. Now, of course, one caveat is that it doesn't necessarily mean that it's a forced migration if it happened during the conflict period. So here, unfortunately, I don't have exact information on whether it's forced. It was not asked, but I'm assuming that people that migrated during this period are more likely to be forced. So in a way, you can think of it as like an intention to treat thing, but I don't have the means to only focus on forced migrations because I don't have information on that. I will control for respondents age and age squared, which was significantly different between the people that are migrating during this period versus not. So that's, again, because I'm looking at the effects from a survey 10 years later, of course, people that moved during that period are older. So I'm controlling for their age and age squared, their parents' education, et cetera. Right. So the parameter of interest will be the triple interaction between these three dimensions. So in the paper, I report all the controls, but here in the interest of you being able to view it, I only show you the triple interaction term. And you can see that. So here in this column, the outcome variable is whether respondent things, domestic violence is justified under any scenario. The point estimate is 16 percentage points. If we compare it to the mean in the full sample, it's a very large effect. So the mean was 20%, but perhaps the more relevant comparison group is Kurdish women who did not migrate during this period but are from this area. So for that sub-sample, the average level of the outcome is 32%, which still means that this is a 50% higher likelihood that they think, so Kurdish women who migrated during the period are 50% more likely to think that domestic violence is justified relative to Kurdish women who did not migrate during that period but are from the same area. Right. And then if I use like the other indices, this fraction of scenarios that she thinks domestic violence is justified or the first principle component, the percentage effects are similar relative to the mean. So I try to do a few types of placebo tests in the paper. Actually, this wasn't in the draft that I think was circulated, but it will be in the next round because these are more recent. So in particular, I tried to use the small sample that I had of people that moved before the forced migration and then compare people that moved before the conflict period and people that moved way before the conflict period. So kind of a placebo test and there I don't find any significant effect. So that kind of is reassuring I think because it shows that the previous effect that I showed you is not driven by some preexisting trends that was already taking place, which is in a way supporting my identifying assumption of parallel trends. If I look at the breakdown of the different scenarios, again as I said before I'm not sure exactly how to interpret them but it seems like the effect is more precisely estimated for these three scenarios about neglecting children's needs, arguing with their husbands and refusing to have sex with them. For the other scenarios, the coefficients are positive but usually imprecisely estimated, but they're all imprecisely estimated. But as I said, I think it's fair to say that all of them are positive and for most of them it's similar. Okay, I have 15 minutes. Now in terms of the mechanisms, as I said before, it is reasonable to expect that the migration may have reduced women's employment opportunities. So if I look in the overall sample at employment rates of women, I also see what previous literature from Turkey has shown, which is that women in urban areas their employment rate is lower on average compared to rural areas in Turkey. So in rural areas women's employment rate in my DHS sample is about 40%. In the urban areas it's 29%. And of course if we break this down by schooling level it's very interrelated and women that have low schooling their employment rates are way lower in the urban area relative to the rural whereas women that have high schooling there's no significant difference. And then for men, even though the differences are smaller, the reverse relationship is there. So on average men are more likely to be employed in urban sector but in the rural sector they are less likely to be employed. So I'm not trying to say anything about why this is, it's beyond my paper although it's interesting. But it's reasonable to expect that this forced migration may have reversed or affected the bargaining position within the household. So for that I look at the likelihood that their husband is working when they respond and it's not. I see that there's an 11 percentage point increase in this probability even though this effect is 10 to 15 years after the migration and it's a large effect relative to the sample mean. So that again is in line with the bargaining power being affected. Okay now in the remaining time let me also show you what I do with the data from the applicants to the women's shelters. So as I said this is one of the largest NGOs in the country and it is well known and so even though I don't try to claim that this is the universe of women that seek help, it is a large share of them. And I have around 3,000 applicants that came to the NGO to seek help between these two and a half years. It's from mid 2009 until the end of 2012 and I use this to test if forced migrants have remained in an abusive relationship for longer and also have suffered like more intensive domestic violence before at the time of the application. So these are some summary statistics from this data set which is really like not the most uplifting thing to show unfortunately but so the type of information on domestic violence are things like so I have one question about the duration which is if whether the relationship in which domestic violence was experienced started more than 10 years ago and also whether the domestic violence started at the beginning of the relationship. So I combine these two things to define a variable which is violence has been going on for 10 plus years and 30% of the applicants said yes to that and if I compare forced migrants to other applicants it's much higher amongst the forced migrants and the way I identify forced migrants here is different there is no migration history but there is a direct question have you ever migrated and then if yes what was the reason and one of the options is due to security reasons and discussing the way the data was recorded with the NGO workers that means it was forced and so I define forced migrants based on that. And then other outcomes is like whether she received any treatment physical or psychological treatment before coming to the center only 12% of the applicants said yes whether she filed a legal complaint very small only 4% said so and it was much lower again significantly lower amongst the forced migrants if she complained to anyone including friends, family or authorities only 52% of the applicants said yes to any of the potential places and again the rate is lower amongst forced migrants half of them were raped and again the rate is much higher amongst the forced migrants and finally there was one question about if she had a miscarriage due to domestic violence 8% of the applicants said yes and the rate was almost double amongst forced migrants to other applicants so in the analysis I control for I regressed these outcomes on being a forced migrant controlling for being a migrant in general I also control for additional things that I have information on such as the timing of marriage whether the marriage was forced marriage if it was a family marriage with a cousin or a family member whether the respondent has an independent source of income so even things that are potentially endogenous to forced migration I control for to see if these are driving them and it seems like with all of these the correlations are robust to controlling for so forced migration seems to be positively correlated with the likelihood of having been in the relationship for more than 10 years before coming to the NGO negatively correlated with receiving other treatment and negatively correlated with seeking other help which is kind of important because of course it could be that maybe they didn't come to the NGO but they went elsewhere so that doesn't seem, unfortunately that doesn't seem to be the case there is no significant at least correlation if anything it's negative between going to the police or friends or family being a forced marriage finally when I look at if they were forced to have sex against their will or if they had a miscarriage they are more likely to say yes to those as well the forced marriage and as I said the relationships are robust to controlling for the things I have information on so to conclude it seems like the conflict induced migration in this context led to an increase in the likelihood that women think that domestic violence is acceptable and this was a long-term effect 10 to 15 years later and one possible mechanism that I'm able to kind of test for is the bargaining power effect and I find evidence in line with this that women's bargaining power women's economic opportunities and therefore their bargaining power was reduced relative to men because of the forced migration from rural to urban areas in this context of course this doesn't mean that it will be it will work in the same way for all types of forced migration it depends a lot on the context and where they are forced to move from and to where but in this context this is what happened but overall I think we can draw a general conclusion that it is reasonable to expect that forced migration might lead to effects on gender norms and it might have bigger consequences in terms of women's well-being so the effects might be heterogeneous by gender and I think it's important to take this into account when designing and implementing policies on forced migration thank you