5 Forced Migration and Attitudes Towards Domestic Violence: Evidence from Turkey

Civil conflicts and wars have many dire consequences, including forcing people to flee their homes. By the end of 2014, 59.5 million people had been displaced due to conflict, persecution, or human rights violations (UNHCR 2015). Among them, a large share (38.2 million) were internally displaced persons (IDPs) who had moved away from their homes but remained in their country of citizenship. Although forced migration affects millions of people, its effects are relatively understudied (Ruiz and Vargas-Silva 2013). In particular, we know little about the effects of conflict-induced displacement on gender norms and attitudes.

In this chapter, I explore the long-term consequences of internal forced migration for women’s attitudes towards domestic violence in Turkey. According to official figures, between 1984 and 1999 around one million people¹ were internally displaced due to the conflict between the Turkish state and the Kurdistan Workers’ Party (PKK) (HÜNEE 2005). Many of the migrants were Kurdish, lived in rural areas, and had to move to urban centres in the region or in the western parts of the country. As of 2009, fewer than 20 per cent of IDPs had returned to their original residence (IDMC 2013).

Theoretically, forced migration may affect women’s attitudes to domestic violence through a number of mechanisms. Household bargaining models predict that a change in women’s access to economic opportunities may affect domestic violence. Forced migrants may lack necessary skills or networks to obtain employment at their destination. If such resources are rarer for women or it is more important for them to gain access to economic opportunities in the urban sector, it is likely that forced migration from rural to urban areas will reduce their bargaining power in the household. In Turkey, employment opportunities for low-skilled women are rarer in the urban sector than in the rural, while for men the difference is not as large (Aran et al. 2009; Tunali 2003). Therefore forced rural-to-urban migration may have led to a fall in women’s intra-household bargaining power, which may have increased the incidence and acceptability of domestic violence. Alternatively, being forced to move from rural to urban areas may forcibly expose migrants to different cultural norms. Traditional gender roles may come into conflict with the cultural norms at the destination, causing conflict within the household. Moreover, the psychological and social consequences of being forced to migrate may increase the incidence of domestic violence and affect attitudes towards it. Ultimately, whether forced migration affects domestic violence, and if so through what mechanisms, are empirical questions.

To examine whether women who were forced to migrate during the conflict have different attitudes towards domestic violence, I use two different data sources. First, I use the 2008 and 2013 waves of the Turkish Demographic and Health Survey (TDHS). These surveys were targeted at a representative sample of women² and collected information on their health outcomes and basic socio-economic characteristics. In addition, information was recorded on their attitudes towards domestic violence and their migration histories. This enables me to test whether women who migrated due to the conflict have different attitudes towards domestic violence. In particular, I estimate a triple-difference model, exploiting differences in ethnicity (Turkish vs Kurdish women), region of birth/childhood (the conflict region vs the rest of the country), and timing of migration (whether they migrated during the conflict period (1984–99) or not). I find that Kurdish women who migrated from their homes during the conflict are more likely to believe that a husband is justified in beating his wife in at least one of the scenarios they were asked about. Controlling for observable characteristics such as age, family background, parental education, and birthplace, I find that women who were forced to migrate are 16 percentage points more inclined to believe domestic violence is acceptable. Examining the possible mechanisms behind this effect, I find some evidence consistent with the bargaining power hypothesis: forced migration increased the likelihood that the respondent’s husband was working while she was not. Moreover, spouses of migrant women were more likely to have tried to control their wives by, for example, limiting their movements or social interactions. This may be in line with the idea that forced migration threatens the traditional male gender role more than the female role, and causes conflict in the household.

As another, suggestive, piece of evidence, I use a novel database that was compiled by one of the largest NGOs working to support women affected by domestic violence in eastern Turkey.³ The data contains information on all women who applied to the NGO for help in dealing with domestic violence between October 2009 and December 2011. In the sample, there are more than 2,278 Kurdish applicants who were subjected to physical or sexual intrafamily violence. Among them, about 15 per cent reported that they had been forced to migrate due to security concerns, and there is a robust relationship between being a forced migrant and the duration as well as the extent of the domestic violence they experienced before seeking help from the NGO. In particular, women who were forced migrants were more likely to report that the abusive relationship had been going on for more than ten years and that the abuse had started at the beginning of the relationship. They were also less likely to have sought legal aid or alternative help, to have been forced to have sex against their will, and to have had a miscarriage due to domestic violence.⁴ These findings suggest that women who had been forced to migrate had suffered domestic violence for longer and possibly of greater severity before deciding to seek help from the NGO (and were no more likely to have sought help elsewhere).

Taken together, the findings from the TDHS and the applicants’ data imply that forced migration ensuing from the Kurdish-Turkish conflict changed women’s attitudes towards domestic violence, making them more tolerant of it. Further research is needed to fully understand the mechanisms driving the effects of forced migration on attitudes towards and the incidence of domestic violence.

The forced migration studied in this chapter was caused by the conflict between Turkey and Kurdish insurgent groups, mainly the PKK. The PKK was founded in 1978 under the leadership of Abdullah Öcalan, and transformed into a paramilitary group in 1984. They started a guerrilla war, benefiting from the difficult terrain in the region for cover. In the ensuing fighting, Turkey both used its army and recruited local Kurdish villagers to create a paramilitary organization named ‘village guards’ (Ozar et al. 2013). Between 1984 and the late 1990s, an estimated 40,000 people lost their lives, and many villages were evacuated and/or destroyed in the largely Kurdish eastern and south-eastern parts of the country (Beriker-Atiyas 1997). The intensity of the fighting diminished with the capture of Öcalan in 1999, and in September 1999 a ceasefire was declared. The focus of this chapter will be on the forced migrations that took place in this period. The fighting resumed after 2004, and continued with lower intensity than in the 1990s until its re-escalation in 2015.

During the height of the conflict, a state of emergency was declared by the Turkish state in eleven provinces in the east and south-east of the country. These provinces were Bingöl, Diyarbakır, Elazığ, Hakkari, Mardin, Siirt, Tunceli, and Van, and the neighbouring provinces of Adıyaman, Bitlis, and Muş. The state of emergency came to be known as OHAL (from olağanüstü hal (extraordinary state) in Turkish), and the region came to be known as the OHAL region. Most of the forced migration took place within this area as inhabitants of many rural settlements were evicted.⁵ While most of the displaced were from rural areas, there was also forced migration from some towns (e.g., Lice, Kulp, Cizre, and Şırnak) to other urban areas, and the majority of the displaced population was Kurdish (Kurban et al. 2007). A report by the Turkish Parliament Investigation Commission in 1998 highlighted three main reasons for the displacement: (1) the collapse of agriculture and animal husbandry due to the clashes; (2) the PKK’s eviction of villagers who were cooperating with the state (i.e. becoming village guards); (3) the eviction by the security forces of villagers who refused to become village guards. Many of the migrants settled in urban centres in the region and some moved to western parts of the country. As of 2009, fewer than 20 per cent of IDPs had returned to their original residence (IDMC 2013).

For a long time the number of people forced to migrate during the conflict was disputed. Official government reports put this number at around 300,000, while many NGOs claimed it was as high as three million (IDMC 2013). In 2005 the Turkish government commissioned a comprehensive study to estimate the correct number. According to this report (HÜNEE 2005), between 925,000 and 1.2 million people were internally displaced due to the conflict. The same study showed that forced migrants on average had low socio-economic status and low schooling, and the majority (94 per cent) of them were either not employed or working in the informal sector (i.e. in jobs with no social security).

The forced migration that was caused by the Turkish-Kurdish conflict may have affected migrants’ attitudes towards domestic violence through a number of mechanisms.

In theoretical models of household bargaining, spousal violence can be a means to increase the abuser’s bargaining power. An improvement in women’s access to economic opportunities (such as employment or earnings) may decrease or increase the incidence of domestic violence, depending on the initial allocation of bargaining power within the couple and whether the reservation utility of the woman or her spouse is binding (Anderson and Genicot 2015; Bloch and Rao 2002; Eswaran and Malhotra 2011; Tauchen et al. 1991). This implies that a change in the economic opportunities of women relative to their spouses is likely to affect the incidence of domestic violence. Empirical literature testing these predictions has studied how women’s employment or earning opportunities can influence the incidence of domestic violence across a variety of settings (e.g. Aizer 2010; Alesina et al. 2016; Amaral et al. 2015; Andenberg et al. 2016; Angelucci 2008; Bobonis et al. 2013; Chin 2012; Heath 2014; Heise and Kotsadam 2015; Hidrobo and Fernald 2013; van den Berg and Tertilt 2015). Broadly speaking, the evidence suggests that an increase in women’s bargaining power reduces domestic violence in high-income settings, while it leads to an increase in domestic violence in low-income settings. If forced migration affects women’s access to economic opportunities (relative to men’s), it may change the likelihood of domestic violence and conflict within the family, which may in turn affect attitudes towards it.⁶ In the context of the Turkish-Kurdish conflict, the forced migration occurred in the poorest parts of the country, among households that had limited income and education.⁷ Moreover, in Turkey, women with low education typically have fewer employment opportunities in the urban sector relative to the rural sector, while for men the reverse is true on average (Aran et al. 2009; cf. evidence presented in Table A4). Thus forced migration in Turkey may have reduced the economic opportunities of women relative to men.

Another mechanism through which forced migration may affect attitudes is the migrants’ exposure to different cultural norms. In the Turkish context, many forced migrants had to move from rural to urban areas, which on average have more progressive gender norms.⁸ This may have brought traditional gender norms into conflict with more modern norms. Threats to the traditional male gender role may have led to an increase in the incidence of domestic violence (Atkinson et al. 2005). On the other hand, if migrants over time adopt the norms of their destinations, they may have lower tolerance of domestic violence compared with individuals from similar ex ante backgrounds who were not forced to migrate.⁹

Another alternative mechanism might be related to the long-run psychological consequences of being exposed to conflict. In psychology, social cognition theories propose that exposure to community violence may culminate in the normalization of the use of aggressive behaviours (e.g., Fowler et al. 2009). In economics, empirical studies have demonstrated a positive association between exposure to conflict and individuals’ propensity to commit violent or criminal acts later on (Couttenier et al. 2016; Miguel et al. 2011; Rohlfs 2010), and the long-run psychological consequences have been highlighted as one of the potential mechanisms behind such effects.¹⁰

Last but not least, forced migration may result in the breaking apart of existing social ties and the loss of social capital. Without the support of their family and friends, women may have difficulty in leaving abusive relationships, which may increase the incidence and acceptability of violence.

The data used in this study comes from two sources: (1) the TDHS; (2) data on applicants to one of the largest NGOs in Turkey providing support for women affected by domestic violence (henceforth referred to as ‘applicant data’).

I use data from the 2008 and 2013 waves of the TDHS. In the 2008 wave, a representative sample of women aged 15–50 who had been married at least once was surveyed. In TDHS 2013, all women were sampled, regardless of their marriage history.¹¹ In addition to the respondents’ basic demographic and socio-economic characteristics, the surveys collected information on their migration history and domestic violence attitudes. For migration, respondents were asked to report their birth province, the place where they had grown up (their main residence until age 12), and every location they had lived in for at least six months after the age of 12. If they had moved from one place to another, the timing of migration was recorded. On the basis of this information one can identify whether a respondent is from an OHAL province and if she migrated from there (or within the province from one location to another) during the conflict years (1984–99). Of the 16,216 women in the data, 2,860 were born and/or grew up in one of the OHAL provinces. Among them, 752 migrated from an OHAL province during the conflict period, and 2,108 did not. Respondents were asked to report whether they thought that a husband was justified in beating his wife if she (1) neglected their children’s needs, (2) argued with her husband, (3) refused to have sex with him, (4) burned the food, (5) wasted money, (6) did not cook, or (7) neglected household chores.¹² On the basis of these, I define three summary indices: first, a dummy variable equal to one if the respondent thought it was acceptable for a husband to beat his wife in any one of these scenarios; second, the fraction of cases in which the respondent thought domestic violence was justified; and third, the first principal component of responses to the individual scenarios. On average, 20 per cent of respondents thought men were justified in beating their wives in at least one of the scenarios. For detailed descriptive statistics on these and background characteristics of the respondents, see Gulesci (2017: Table 1).

The second data set was collected by one of the main NGOs offering support to women who are victims of violence in eastern and south-eastern Turkey—the provinces affected by the conflict. Depending on their needs, the NGO may offer the women legal, medical, psychological, or economic (e.g. in finding a job) assistance, as well as a chance to meet and forge bonds with other women who have been victims of violence.¹³ The data contains information on all women who applied to the NGO for assistance between October 2009 and December 2011. Upon application (prior to receiving any support from the NGO) they were asked to complete a brief questionnaire about why they were applying, what type of abuse they had faced, and some information on their background (e.g., their age, education level, and work status). Importantly for the analysis, the survey contained questions on the applicant’s migration status, which enables me to identify whether the applicant was forced to migrate due to the conflict.

I limit the sample to women who applied to the NGO seeking help because they were subject to physical or sexual domestic (intrahousehold) violence. This is clearly a selected sample. It only contains women who chose to report and seek help for any violence they experienced. I use this data to provide suggestive evidence about forced migrants’ attitudes towards domestic violence relative to other applicants. In particular, I test whether forced migrants endured more domestic violence, in terms of duration or severity, before deciding to seek help. There are 3,582 cases in total of women who applied to the NGO to seek assistance in dealing with physical or sexual domestic violence. Of these, 2,278 were Kurdish and 349 had been forced to migrate.¹⁴ The average woman in the sample is 37 years old, but there is a lot of age variation among them (the youngest applicant in the sample is 15, and the oldest 80 years old). I study the correlation between being a forced migrant and indicators of the duration and extent of the domestic violence experienced by the applicants, and any assistance they may have sought prior to their application to the NGO. For detailed descriptive statistics on this dataset, see Gulesci (2017: Table 2).

To identify the effects of forced migration on women’s attitudes towards domestic violence in the TDHS, I estimate the following model:

where y_ipt is the domestic violence attitude of respondent i from province p whose migration status was revealed in period t; K_i and C_p are dummy variables for whether respondent i is Kurdish and from a conflict province respectively; M_it is a dummy variable if respondent i migrated during the conflict period (1984–99); X_i is a matrix of control variables;¹⁵ θ_p and S_t are province of birth and survey wave fixed effects respectively. Standard errors are clustered by birth province to account for the fact that women born in the same province will have correlated outcomes. The parameter of interest is λ, which gives the triple difference between women from the conflict region (vs the rest of the country), who migrated during the conflict period (vs did not migrate in this period), who are Kurdish (vs Turkish). The key identifying assumption is that, conditional on the control variables, this triple difference should pick up the effect of being forced to migrate because of the conflict. On the basis of the discussion of the historical and political context in Section 2, it is reasonable to assume that Kurdish women who moved from the conflict provinces during this period were most likely forced to do so because of the conflict.

In the applicant data, I estimate:

where y_ip is the outcome of interest for applicant i from province p; F_i is a dummy variable equal to one if the respondent is a forced migrant; X_i is a vector of demographic and socio-economic controls;¹⁶ and θ_p is the province of application fixed effects. Standard errors are clustered by province. The parameter of interest is σ, the difference between forced migrants and the rest of the applicants, controlling for other observable characteristics (including whether the respondent is a migrant in general).

Table 5.1 presents the results of estimating the triple-difference specification in [1] on women’s attitudes towards domestic violence in the TDHS data.¹⁷ In Column (1), the dependent variable is an indicator for whether the respondent thought a man was justified in beating his wife in any of the scenarios she was asked about. The estimates imply that Kurdish women from the conflict region who migrated during the conflict are 16 percentage points more likely to think that domestic violence is justified in at least one of these scenarios. This is a very large effect, both compared to the sample mean (20 per cent) and relative to the subsample of Kurdish women who did not migrate during the conflict (32 per cent). In Column (2), the dependent variable is the fraction of scenarios in which the respondent thought domestic violence was acceptable. The outcome is higher by 0.06 for Kurdish women who were forced to migrate during the conflict, which is a large effect relative to the sample mean (0.08) or compared to Kurdish women who did not migrate during the conflict (0.15). Finally, Column (3) shows the finding is robust to aggregating the responses using an alternative way—the first principal component of the individual questions on domestic violence. Overall, results in Table 5.3 imply that more than twenty years after the beginning and ten years after the end of the conflict in question, forced migrant women were more likely to find physical domestic violence justified.¹⁸

In order to assess the robustness of the estimates and the validity of the identification strategy, I explore two approaches. First, I conduct a placebo test where I estimate the triple-difference specification in [1] for women who migrated before the conflict period (i.e. prior to 1984). In other words, I substitute the dummy variable M_it with one that is equal to one if respondent i migrated before the conflict period. If the findings are driven by the forced migration ensuing from the conflict, then I should not find any significant differences in the domestic violence attitudes of Kurdish women who migrated before the conflict relative to those who did not and relative to Turkish women. The results are reported in Gulesci (2017: Table A1). The coefficient estimates for the triple-interaction term (λ) are negative and insignificant for all outcome variables, which builds confidence that the effects demonstrated before were driven by the forced migration and not by any other underlying differences across the comparison groups.

Second, as an alternative specification, I restrict the sample to Kurdish respondents and estimate the difference-in-difference between those who are from the OHAL provinces and those who migrated during the conflict period. Formally, I estimate:

In this specification, the coefficient of interest is τ, the difference-in-difference between Kurdish women who were from the OHAL provinces versus the rest of the country, and those who migrated during the conflict relative to those who did not. The results are reported in Gulesci (2017: Tables A2 and A3). The estimated effects are qualitatively similar to the triple-difference approach in specification [1], but due to the smaller sample size some of the estimates lose precision. For example, in Table A2, the estimate for τ is not precise for the first principal component (Column (3)), but it is positive and significant for the other two indicators.

To return to the discussion in Section 3, a potential mechanism behind this change in women’s attitudes might be that forced migration lowers the bargaining power of women relative to men and thus makes domestic violence more expected and acceptable in their households. To test this mechanism explicitly, I estimate specification [1] on the employment and schooling level of women and their husbands (thus limiting the sample to married women only). Results in Table 5.2 show that while forced migration did not have a significant effect on the employment probability of women in the long run (Column (1)), it did widen the gap between them and their husbands in terms of employment probability. Column (3) shows that Kurdish women who migrated during the conflict are 11 percentage points more likely to be in a relationship where their spouse is working while they themselves are not. They are also 12 percentage points less likely to be classified as above middle-wealth class according to the wealth ranking provided by the TDHS. These effects are in line with forced migration leading to a change in intrahousehold bargaining power that favours the men in the family. In Turkey, on average, women’s labour force participation is higher in the rural sector compared with the urban sector (Aran et al. 2009; Tunali 2003). Moreover, Tunali (2003) shows that women’s unemployment rate is higher in the urban sector, which suggests that employment rates are even lower.¹⁹ Given these patterns, forced migration in this context is likely to have widened the gap in employment opportunities of women relative to men, lowering their bargaining position. While this is in line with the bargaining power mechanism discussed in Section 3, it does not imply that this is the only mechanism that may be at work.

Finally, I explore whether forced migration increased the likelihood of the respondent’s spouse behaving in a controlling manner. In the TDHS, every respondent who was (or had been) married was asked if her current or (if divorced or widowed) last partner had ever tried to prevent her from seeing her friends, limited her contact with her family, insisted on knowing where she was at all times, distrusted her with money, or accused her of being unfaithful. Table 5.3 shows the effect of forced migration on respondents’ likelihood of having experienced such behaviours. Forced migration is associated with a 10 percentage point increase in the likelihood that respondents’ husbands exhibited at least one of these behaviours (Column (6)). This effect is mainly driven by an increase in their partners’ likelihood of insisting on knowing where she was at all times and preventing her from seeing her friends. This increase in men’s attempts to control women’s actions may be in line with a number of mechanisms that were discussed in Section 3. For example, forced migration may have brought traditional male gender roles into conflict with more modern roles, which may have induced men to become more controlling. Further research and more detailed data are needed to fully understand the key mechanisms driving the effect of forced migration on domestic violence attitudes.

As an additional, suggestive piece of evidence, I explore the differences between forced migrant women and others who sought help from the shelter NGO in dealing with the consequences of domestic violence. Table 5.4 presents the results of estimating specification [2] on the duration of domestic violence experienced by each applicant and previous help sought prior to applying to the NGO. The estimates in Column (1) show that among Kurdish women who applied to the NGO for assistance, forced migrants—with controls for being a migrant in general—were seven percentage points more likely to have been in a relationship for over ten years and experienced domestic violence since the beginning of this relationship. This finding is robust to controlling for age at marriage, being forced into the marriage, housing conditions, and literacy level of the applicants (Column (2)). This suggests that forced migrant women suffered domestic violence for a longer period before deciding to seek assistance from the NGO. The rest of the table shows that forced migrants are also less likely to have sought any treatment or filed a legal complaint, and no more likely to have complained to other institutions or people, prior to seeking help from the shelter. This rules out the possibility that the reason why forced migrant women applied later was that they were receiving assistance from alternative institutions or people.

Table 5.5 shows the association between forced migration and some indicators of the extent of domestic violence experienced by the applicants. The first two columns show that forced migrants were five to six percentage points more likely to have been forced to have sex against their will—significant only when controls are added in Column (2). This corresponds to an increase of 12 per cent in the outcome relative to the sample mean. In Columns (3–4), the dependent variable is whether the applicant reported having had a miscarriage caused by the domestic violence she experienced. This likelihood is four percentage points higher among forced migrants, a large effect considering that in the entire sample only 8 per cent of women reported having had a miscarriage due to domestic violence. Taken together, the findings suggest that forced migrant women had endured domestic violence for longer and of greater severity before deciding to seek help from the NGO.

In this chapter, I studied the effects of forced migration caused by the Turkish-Kurdish conflict from the mid-1980s to the end of the 1990s on migrating women’s attitudes towards domestic violence. Evidence from two separate data sources (the TDHS and a database of applicants to a women’s shelter NGO) suggests that women who were forced to leave their homes due to the conflict are more tolerant towards domestic violence. In the TDHS data, forced migrants were more likely to report that a husband was justified in beating his wife and to have experienced controlling behaviours by their husbands. Moreover, among the applicants to the shelter NGO, forced migrants had endured domestic violence for longer and of greater severity before deciding to seek assistance.

There may be multiple mechanisms that drive the impact of forced migration on attitudes towards domestic violence in this context. One possible mechanism is a fall in the bargaining position of women within the household, and a general worsening of their economic status. Moreover, forced migration increased the incidence of controlling behaviours by men, suggesting that it may have threatened traditional gender roles. Future research is needed on the extent to which other mechanisms—for example, the loss of social networks, or the psychological consequences of conflict exposure—may influence the way in which forced migration shapes migrants’ attitudes towards domestic violence. Understanding the relative importance of these mechanisms across contexts is essential for designing effective policies to address the long-run consequences of forced migration for domestic violence.