We study the impact of marriages resulting from bride kidnapping on infant birth weight. Bride kidnapping—a form of forced marriage—implies that women are abducted by men and have little choice other than to marry their kidnappers. Given this lack of choice over the spouse, we expect adverse consequences for women in such marriages. Remarkable survey data from the Central Asian nation of Kyrgyzstan enable exploration of differential birth outcomes for women in kidnap-based and other types of marriage using both OLS and IV estimation. We find that children born to mothers in kidnap-based marriages have lower birth weight compared with children born to other mothers. The largest difference is between kidnap-based and arranged marriages: the magnitude of the birth weight loss is in the range of 2 % to 6 % of average birth weight. Our finding is one of the first statistically sound estimates of the impact of forced marriage and implies not only adverse consequences for the women involved but potentially also for their children.
Around the world, people usually marry either through a so-called love marriage or an arranged marriage. In a love marriage, the two members of a couple identify one another following a search process. The groom typically proposes to the bride, and the bride accepts. In an arranged marriage, this search activity is not undertaken by the groom and bride themselves but by their parents, their relatives, or a contracted external party. Each potential spouse tends to consider multiple marriage options, identified by the search agents, and chooses among them (Batabyal 2001). In economic terms, the partners compare each other’s characteristics and evaluate their potential gains from a (love or arranged) marriage. If the expected gains to both partners from marriage are greater than the expected gains from continued search for a mate, then they contract marriage (Becker 1973).
In some societies, forced marriage constitutes a third option. The fundamental difference from the other marriage types is that one spouse—generally the bride–or occasionally both spouses have no choice regarding their marriage partner. This lack of choice presumably reduces the expected gains from marriage, which should lead to adverse outcomes in forced marriages. The literature is surprisingly scant on this topic. There is some evidence for high prevalence of depression, other emotional and health issues, divorce, self-harm, and even suicide in forced marriages (Ali et al. 2009; Sabbe et al. 2014; Samad 2010; Samad and Eade 2003), but we are not aware of any study that has quantitatively measured the impact of forced marriage. We aim to fill this gap. Specifically, we analyze the impact of forced marriage on infant birth weight.1 Birth weight is known to be strongly linked to maternal psychological and physical health and cannot easily be altered by nonmaternal inputs (Almond and Currie 2011; Currie 2009; Kramer 1987). This measure is also of high policy relevance. It is a primary marker of infant health and is related to long-run outcomes such as adult height, completed education, and earnings (Behrman and Rosenzweig 2004; Black et al. 2007).
We hypothesize that the impact of forced marriage on infant birth weight is negative. Although we do not identify the channels through which forced marriage affects birth weight in this article, we anticipate that the impact works through psychological stress. Spouses in forced marriages are obligated to live with mates whom they did not choose and with whom they do not necessarily share common interests and attitudes. This living arrangement is likely to induce psychological stress. Psychological stress experienced by women during pregnancy, in turn, affects birth outcomes and reduces birth weight (Black et al. 2007; Duncan et al. 2016; Mansour and Rees 2012; Persson and Rossin-Slater forthcoming; Torche 2011).
We study a specific form of forced marriage: marriages resulting from bride kidnapping. We focus on Kyrgyzstan, a small country with a population of 5.8 million located in Central Asia, although this is not the only country where bride kidnapping is practiced. According to ethnographic reports, various forms of kidnapping (also referred to as abduction or capture) have existed in societies across Europe, Asia, Africa, Australia, and the Americas in the past (Ayres 1974; Barnes 1999). In many places, kidnapping no longer seems common, but it is still practiced in such countries as Armenia, Ethiopia, Kazakhstan, and South Africa (Dito 2015; Edling 2012; Nkosi and Buthelezi 2013; Rice 2014; Werner 2009).
Kyrgyzstan has a reputation for being reformist and pluralistic and for placing fewer constraints on women than most of its former Soviet neighbors and countries farther south, such as Afghanistan or Pakistan. Women hold high positions in society and politics; recently, a woman, Roza Otunbayeva, even served as president. At the same time, a significant percentage of women—an estimated 16 % to 23 % (Nedoluzhko and Agadjanian 2015; UNFPA 2016; Za Reformy i Rezul’tat2015)— are married following kidnapping. They are abducted by men and have little choice but to marry their kidnappers.
The extent of force involved in the kidnapping appears to vary widely (Amsler and Kleinbach 1999; Kleinbach et al. 2005). One extreme is fully nonconsensual abduction, in which the groom kidnaps the bride through physical force. Another extreme is elopement or staged abduction, in which the groom and the bride agree on the kidnapping beforehand—for example, in the case of parental disapproval of their marriage plans. The data on the type of marriage used in our analysis are self-reported and do not distinguish between different degrees of force involved in kidnapping. We acknowledge that depending on whether self-reported kidnap-based marriages include both consensual and nonconsensual marriages or only the most violent kidnappings, our point estimates may be an underestimate or an overestimate of the adverse consequences of forced marriage. Thus, our primary interest is to determine whether there is any negative impact of kidnap-based marriages on birth weight. We place less importance on the exact magnitude of this impact.
We focus on rural areas and urban settlements but exclude the two largest cities in Kyrgyzstan (Bishkek and Osh) for two reasons. First, kidnap-based marriage is not common in cities. A recent, representative survey measures its prevalence in Bishkek and Osh to be 8.3 % and 14.2 %, respectively, compared with a national prevalence of 22.1 % (UNFPA 2016). Second, many kidnap-based marriages in the cities are reported to be staged rather than forced (Handrahan 2000) and are therefore not the type of kidnapping with which we are concerned.
Background: Kidnapping of Brides in Kyrgyzstan
Bride kidnapping, called ala kachuu in Kyrgyz (literally, “to take and run away”), is the act of abducting a woman to marry her. In Kyrgyzstan, kidnapping is illegal and punishable with a prison sentence of 3 to 7 years, or up to 10 years if the kidnapped bride is below the minimum legal marriage age of 18 years.2 However, because few people report cases of kidnapping to the authorities, prosecutions are rare. The practice of kidnapping is essentially limited to ethnic Kyrgyz, who account for 71 % of the population (Minnesota Population Center 2013). With very few exceptions, other ethnic groups do not engage in kidnapping but marry through either love or arranged marriages (Handrahan 2004). That Kyrgyz (and Kazakhs in southern Kazakhstan) tolerate kidnapping, while more conservative Uzbeks and Tajiks do not, is one of the puzzles that remain to be explained. Religion does not serve as an explanation, given that the vast majority of Kyrgyzstan, including the Kyrgyz, Uzbek, and Tajik population groups, is Muslim.
Today, ala kachuu usually involves a potential groom and his male friends taking a young woman into a car and then transporting her to his home. The woman might be kidnapped from her house or another location, such as a school or her workplace. In the groom’s home, she is pressured by the groom’s female relatives to write a letter of “consent” to her family and put on a marriage scarf. After she puts the marriage scarf over her hair, she accepts the marriage (Borbieva 2012). This process can last from a few hours to several days. Eventually, the groom’s family goes to visit the bride’s parents to “apologize,” to hand over their daughter’s letter, and, traditionally, to offer sheep and other gifts. Such offers are considered the bride price (kalym). In principle, it is possible for the woman to resist kidnapping. Amsler and Kleinbach (1999) estimated that 17 % of kidnappings do not result in marriage, either because the woman herself or her family resist. Kleinbach et al. (2005) provided an estimate of rejected kidnappings of only 8 %.
Some researchers have attempted to explain why most kidnapped women marry their kidnapper and why the women’s parents agree to this practice. Werner (2009) argued that the act of kidnapping damages the woman’s reputation and that subsequent acceptance of marriage helps to restore that reputation. This is because, after kidnapping, these women are no longer assumed to be virgins. In addition, they might be perceived as stubborn and belligerent if they resist the marriage. Taken together, kidnapped women become less attractive to other potential suitors. In a related explanation, Borbieva (2012) emphasized the role of marriage as a social institution. She noted that women and men in Kyrgyzstan are expected to marry young and without long courtships for several reasons, including lack of social acceptance of premarital sexual activity, the expectation for women to have many children, and, most importantly, the function of marriage as transition to full adult status. Families of women with relatively poor prospects on the marriage market therefore should be particularly inclined to accede to kidnapping. Finally, Handrahan (2004) emphasized that kidnapping is often used as an act of ethnic definition. A “woman who rejects kidnapping is seen to be rejecting not only a Kyrgyz tradition but also Kyrgyz ethnicity” (Handrahan 2004:222). In turn, rejection might create conflict within the community because it implies renunciation of a common practice, especially where kidnapping is widespread.
Data on the prevalence of kidnapping and marriages resulting from kidnapping are rare but increasingly available. Given the sensitivity of the subject, a reporting bias seems likely. However, the stigma of having been kidnapped does not seem to be large among the Kyrgyz, and people are willing to discuss it openly, even with strangers (such as the authors). Nedoluzhko and Agadjanian (2015) conducted a nationally representative survey in 2011–2012 among slightly more than 2,000 households. They reported that one-third of marriages among Kyrgyz were the result of kidnappings; one-half of these marriages were nonconsensual. A similar prevalence is reported by UNFPA (2016), based on a recent survey of nearly 5,000 households. For all married Kyrgyzstani women without ethnic distinction, 22.1 % of marriages are kidnap-based marriages. This translates into 31 % of married Kyrgyz women, assuming no kidnap-based marriages among the other ethnic groups. One-quarter of kidnap-based marriages was reported to be nonconsensual. A lower prevalence is reported in a study of roughly 1,600 women for the International Crime Victims Unit Survey conducted by the local NGO Za Reformy i Rezul’tat (2015), which found that 16 % of married women were kidnapped for the purpose of marriage, or 22.5 % of married Kyrgyz women. One-third of the kidnapped women did not agree in advance to their kidnapping.
The data we use in our analysis are from the Life in Kyrgyzstan (LIK) survey (Brück et al. 2014), which is a panel survey conducted annually between 2010 and 2013 and again in 2016. Two of the authors were involved in designing and implementing the survey. The sampling technique was stratified, two-stage random sampling based on the 2009 Population Census with probabilities proportional to population size. The strata were formed by Bishkek (capital city), Osh (the major city in southern Kyrgyzstan), and the rural and urban areas of the seven provinces (oblasts). The data are representative at the national, urban/rural, and North/South levels.
We use data from 2011 because this wave provides most of the information on respondents’ marriage and fertility. Some of the control variables were collected in earlier (employment history) or later (maternal height and personality) waves, and we assign them to our sample. The 2011 wave contains information on 8,066 adults (aged 18 and older) in 2,863 households. Female respondents were asked whether they had ever been married. If they answered affirmatively, they were questioned about the evolution of their marriage. The exact wording for this question was, How did this marriage come about? The answer options were (1) love marriage, (2) arranged marriage, and (3) bride kidnapping.3 We are unable to distinguish between different degrees of force involved in the kidnapping.
Married female respondents were also asked about their own and their husband’s ages at marriage, whether they had ever given birth, and detailed information on their children (live births only). Children’s birth weights were recorded based on the mother’s recall. During both the Soviet Union era and in today’s Kyrgyzstan, birth weight has been recorded on a so-called health card for each child, which has been used to document all medical information during a child’s development. Health cards used to be kept at hospitals or health centers but are increasingly stored at home. We expect that younger women more accurately recall their children’s birth weight. For this reason, we split our sample of women at the median age (43 years) and conduct the following analysis only for those children born to women aged 18–43 years. Henceforth, we refer to this group as the “younger generation.”
We restrict the sample to Kyrgyz women who have ever been married and live outside the two major cities. Among this sample (1,763 women in total), 23.7 % reported that they contracted marriage through the process of kidnapping, 53.7 % reported love marriage, and 22.6 % reported arranged marriage. Among the younger generation of women, the respective shares are 22.0 %, 61.1 %, and 16.9 %; among the older generation (aged 44+), the shares are 25.5 %, 46.3 %, and 28.2 %, respectively.4 Fig. 1 shows the incidence of the different marriage types by year of marriage. We include only the marriage period 1965–2010 in this figure because there are fewer than 10 observations for 2011 and for the years before 1965 in our data set. The share of love marriages increased over this period, and the share of kidnap-based marriages decreased, especially in the most recent years. These trends are in line with Nedoluzhko and Agadjanian (2015), although the low incidence of reported kidnap-based marriage in 1965–1975 makes it difficult to determine a clear time trend.
Compared with the numbers in Nedoluzhko and Agadjanian (2015) and UNFPA (2016), the share of kidnap-based marriages among all marriages observed in our data is low. The sample of Nedoluzhko and Agadjanian (2015) is an age group comparable with our younger generation. There, one-third of all marriages of Kyrgyz women were the result of kidnappings. We measure slightly more than one-fifth. The sample in UNFPA (2016) covers women aged 15 and older; again, the prevalence is approximately one-third among the Kyrgyz. We measure one-quarter for the younger and older generations together. It is impossible to know whether the misreporting is in our data or in those of the other authors. Given that the UNFPA (2016) sample is much larger than ours and that the estimated prevalences in Nedoluzhko and Agadjanian (2015) and UNFPA (2016) are similar, we suspect that some kidnap-based marriages are not reported as such in our data set. We return to this issue later.
Child characteristics (X) capture the sex and the birth order of child i. They additionally capture whether child i is a twin (no other multiple births were observed). We include indicators for the child’s birth month and birth year to control for seasonal variations in birth weight and time trends. Mother characteristics (M) include her age at child i’s birth and her educational attainment. A quadratic term for her age at the child’s birth is also included to control for a nonlinear relationship between the mother’s age at birth and the child’s birth weight. We include indicators for the year of marriage. We also control for whether the mother speaks both Kyrgyz and Russian (as an alternative indicator of education) and for maternal height (in cm), which is an important determinant of birth weight (Kramer 1987).5 We include a dummy variable for residing at an altitude above 2,500m because birth weight was found to be reduced at high altitudes, presumably due to lower levels of oxygen (Giussani et al. 2001; Jensen and Moore 1997; Unger et al. 1988). τ are district fixed effects, which help to control for heterogeneity between districts, such as different poverty rates, the availability of medical facilities, or gender-specific norms. These fixed effects capture both the predominantly rural districts (rayon) and the urban municipalities (gorod).6 They are at the same level of subnational administration. ε is the error term assumed to be orthogonal to birth weight.
The coefficient of interest is β1. We aim at estimating the causal effect of kidnap-based marriage on birth weight, but kidnap-based marriage is unlikely to be a random event. It seems plausible that there is selection on the marriage market such that a particular type of women is kidnapped, and/or a specific group of men engage in kidnapping. Estimates of β1 would be biased if women and men ended up in kidnap-based marriages based on certain characteristics for which we do not control and that also affect the birth weight of children.
An obvious characteristic that may differ between kidnap-based marriages and other types of marriage is education. If, for example, the least-educated ended up in kidnap-based marriages, this could have consequences on children’s birth weight through the level of health care or the quality of nutrition among pregnant women. Table S1 in Online Resource 1 provides a cross-tabulation of the educational attainment of the spouses in the younger generation sample and type of marriage. Educational attainment is relatively similar for wives as well as husbands in kidnap-based and arranged marriages; it differs between kidnap-based and love marriages and between arranged and love marriages. A substantially higher share of spouses has a university degree in love marriages than in the other types of marriage. Clearly, this type of selection is not problematic because we can control for educational attainment of children’s mothers and fathers. Other types of selection are much more difficult to observe and to take into account in the estimation. For example, it could be that women who appear particularly unhealthy (and therefore presumably less fertile) are least likely to be kidnapped. Conversely, women known to be obedient and subordinate may face the greatest kidnapping risk because they are not expected to resist marriage. If healthy appearance and personality have an effect on birth weight, we may get biased estimates if we do not control for these variables.
We employ two strategies to reduce the potential omitted variable bias. First, we include additional control variables. It is important that these variables refer to the time of childbirth, and not to the time of the survey, because many characteristics change over time. Several potential variables (e.g., maternal body mass index) are measured only at the time of the survey and hence cannot be used. We do control for educational attainment of the child’s father, which is assumed to be fixed at the time of childbirth. Furthermore, we control for the mother’s personality, which is largely if not completely time-invariant during adulthood (Borghans et al. 2008). The LIK contains the 21-item version of the Big Five Inventory (Rammstedt and John 2005), and we compute five distinct personality traits through factor analysis. Finally, we control for the mother’s employment status in the year before childbirth. The LIK holds labor market information going back to the year 1989. For women who gave birth after 1989,7 we can distinguish between women who worked in agriculture, worked in nonagriculture, were enrolled in an education program, and were not in the labor force (housewife or maternity leave) in the year before giving birth. Two of the three additional controls are available only for a subset of the sample, which reduces the number of observations when we include them.
We cannot be sure that the additional controls capture all sources of selection into kidnap-based marriage. Thus, as a second strategy, we run a two-stage least-squares (2SLS) estimation for Eq. (1), using an instrument for kidnap-based marriage. The anthropological literature explains that the family of the groom plays a crucial role in his decision to kidnap. The groom’s parents and siblings usually actively help plan the kidnapping (Amsler and Kleinbach 1999; Borbieva 2012). Even when the groom’s family is not informed in advance of the kidnapping, their members become engaged when the kidnapped woman enters the groom’s house. The female relatives of the groom are expected to persuade the prospective bride to place the marriage scarf on her head. We argue that in areas with a high prevalence of past kidnappings, individual families are more likely to support this practice and may pressure prospective grooms to kidnap (Ismailbekova 2014). An unmarried woman living in such an area is more likely to be kidnapped and acquiesce to the marriage than a woman living in an area with a low prevalence of past kidnappings. We calculate the district-level share of kidnapping among Kyrgyz women in the older generation (aged 44 or older) and use this information as an instrument for kidnappings of Kyrgyz women in the younger generation.8 The district-level share of kidnapping is the share of kidnap-based marriages among all marriages of the older generation per district.
We assume that the district-level share of kidnapping among older women provides an exogenous source of variation for the marriage selection of younger women in the district. It is unlikely that the older generation chooses to live in a particular location because of the marriage market that their children face. Whether a woman in the younger generation still lives in the same province9 where she was born is not significantly related to the district-level share of kidnapping (regression results unreported). Kyrgyz people are not very mobile (except for migration to Bishkek and temporary labor migration to Russia), and marriage markets tend to be local. In our sample including the younger and older generations, 89 % were born in the same province where they lived at the time of the survey. Thus, the geographical circumstances into which a Kyrgyz woman is born are highly likely to predetermine her individual probability of having a kidnap-based marriage. Our identifying assumption in the upcoming estimation is that the district-level share affects the likelihood of kidnap-based marriage of an individual woman but does not have an independent influence on her children’s birth weight after we control for kidnapping. We elaborate on potential threats to this assumption in the section, Plausibility of the Identifying Assumption. As we show there, the assumption is plausibly valid.
Our sample consists of observations from 45 distinct districts. In some districts, the number of observations among the older generation of women is very low, which makes calculation of district-level shares for kidnap-based marriages among all marriages unreliable. Therefore, we exclude districts with fewer than nine women in the older generation. Consequently, we lose information on 4 of 45 districts and calculate the district-level share of kidnapping for 41 districts, which have between 9 and 90 older-generation women. The district-level share ranges between 0 and 0.9 (Fig. S1, Online Resource 1). In our sample of the younger generation, the mean of this district-level share is 0.25 (with a standard deviation of 0.20), and the median is 0.23.
Table 1 provides summary statistics on the main variables used, distinguished by marriage type. In the upper panel of the table, we report summary statistics for children born to the mothers in the younger generation; the lower panel describes the mothers themselves. Some children (21 in kidnap-based marriages, 15 in arranged marriages, and 28 in love marriages) seem to be born out of wedlock; the year of birth is reportedly before the year of marriage. We include these children in our analysis.10
Our sample comprises 791 women who gave birth to 2,191 children. Mean birth weight is significantly lower by 110g for children born in kidnap-based marriages than for children born in arranged marriages. In contrast, children born in kidnap-based marriages are only 30g smaller than children born in love marriages. There are no differences in the share of male children across the marriage types or in the share of children who are the firstborn, thus indicating that the average number of children is comparable: 3.0 for kidnap-based marriages, 3.0 for arranged marriages, and 2.7 for love marriages.
Kidnapped mothers differ from mothers in arranged and love marriages on three counts. First, kidnapped mothers are slightly younger at the time of first and subsequent births, reflecting the fact that kidnapped mothers are younger at the time of marriage than mothers in arranged and love marriages.11 Second, educational attainment of mothers differs across marriage types. Whereas the difference between kidnap-based and arranged marriages is in the share of mothers who obtain a basic or secondary school degree, the difference between kidnap-based and love marriages lies in the share of mothers who go to university. It is important to note that all types of marriage are rare at very young ages; most marriages occur between 17 and 25 years of age. Hence, very young girls almost never marry. Instead, these are young women who have finished most if not all of their education.12 Third, kidnapped mothers have slightly lower height than mothers in love marriages.
Table 2 provides ordinary least squares (OLS) estimates for the relationship between kidnap-based marriage and birth weight. Column 1 estimates Eq. (1) without district fixed effects, and column 2 includes district fixed effects. By including district fixed effects, we compare children in kidnap-based marriage and those in non-kidnap-based marriages within the same district. Columns 3–5 include additional controls. Ceteris paribus, mothers in kidnap-based marriages give birth to children with a lower birth weight. The birth weight loss is estimated to be between 40g and 60g (1 % to 2 % of average birth weight of 3,187g) across the columns but is statistically significant only when we do not include district fixed effects.
As a next step, we compare children in kidnap-based marriages with children in love marriages (Table 3, panel a) and with children in arranged marriages (Table 3, panel b) separately. We control for all variables as before. This exercise reveals that the difference in birth weight is large and statistically significant between kidnap-based and arranged marriages but not between kidnap-based and love marriages—and therefore, given the many love marriages, is not statistically significant between kidnap-based and all non-kidnap-based marriages in Table 2. Children in kidnap-based marriages are between 80g and 125g (2 % to 4 % of average birth weight) smaller than children in arranged marriages. This difference is statistically significant (at the 10 % level) in all columns in Table 3 except for column 3, in which we control for education of the child’s father. This loss of significance is due to smaller sample size rather than inclusion of an additional control variable. Running the estimation of column 2—that is, without controlling for father’s education—for this reduced sample leads to a coefficient of –78.59, with a standard error of 52.95. This coefficient is very close to that in column 3 and also is statistically insignificant, suggesting that controlling for father’s education has little effect. Controlling for mother’s personality and employment in the year before giving birth (columns 4 and 5) also has only a modest impact.
As explained earlier, there is room for omitted variable bias in these OLS estimations, making instrumental variable (IV) estimation for Eq. (1) necessary. Note that the identifying variation in the IV is at the district level rather than at the mother level as before. District fixed effects thus need to be excluded. The IV estimate gives the local average treatment effect. It captures the effect of kidnap-based marriage on birth weight for mothers who ended up in kidnap-based marriages because they lived in districts with high kidnapping prevalence but would not otherwise be in a kidnap-based marriage. The number of observations is slightly lower in the IV estimation than in Tables 2 and 3 because we calculated the instrument for only 41 of 45 districts.13
Table 4 reports the results for the first and second stages of the IV estimation, comparing kidnap-based with all non-kidnap-based marriages in columns 1 and 2, with love marriages in columns 3 and 4, and with arranged marriages in columns 5 and 6. Control variables are omitted from the table. The district-level share of kidnapping in the older generation strongly predicts a younger generation woman’s likelihood of being in a kidnap-based marriage. A 10 percentage point increase in the district-level share increases the likelihood of marriage through kidnapping of an individual woman by 6.4, 6.0, or 10.5 percentage points, depending on the group with which we compare kidnap-based marriages. The instrument is not weak: the F statistic of the test for the significance of the excluded instrument is 113, 71, or 76, all at a p value of .00.
We perform an endogeneity test. If the null hypothesis that kidnap-based marriage is exogenous cannot be rejected, we would conclude that the OLS and IV models estimate the same coefficient. The test compares the respective IV estimate in Table 4 with the OLS estimate without district fixed effects in Tables 2 or 3. The null hypothesis that kidnap-based marriage is exogenous cannot be rejected. The robust regression F statistic is 0.816 at a p value of .372 for comparing kidnap-based with all non-kidnap-based marriages; 0.234 at a p value of .613 for comparing kidnap-based with love marriages; and 0.415 at a p value of .523 for comparing kidnap-based with arranged marriages.14
The birth weight loss is much higher in the IV estimation than in the OLS estimation. In column 2, it amounts to 199g (6 % of average birth weight) at a p value of .18. When we use the sample of kidnap-based and love marriages (column 4), the birth weight loss is 117g (4 % of average birth weight), but this estimate is not statistically significant (p value = .40). Comparing kidnap-based and arranged marriages (column 6), the effect of kidnap-based marriage on birth weight is –194g (6 % of average birth weight). This effect is marginally significant (p value = .11). Given that IV estimation leads to larger standard errors than OLS estimation by construction, even if the instrument has sufficient power as in our case, we regard the estimate in column 6 as reasonably precise.15
The main message from the OLS estimation is confirmed. The large and precisely estimated difference in birth weight is between kidnap-based and arranged marriages, not between kidnap-based and love marriages. This finding is not straightforward to explain, but we suggest a number of possible reasons. First, we suspect that some kidnap-based marriages are misreported in our data set (see the Data section). If so, misreported kidnappings are unlikely to be categorized as arranged marriages because these are a particular, well-defined type of marriage. Rather, we would expect unreported kidnap-based marriages to be categorized as love marriages. This, in turn, would make the group of love marriages a contaminated group, containing marriages resulting from kidnappings. Second, couples marry young in Kyrgyzstan (see Table 1) and without long courtships. This is true for all types of marriage. Given the young age at marriage, grooms and brides have very little experience in selecting a mate. As Batabyal (2001) noted, one of the main justifications for arranged marriage is imperfect information among young people, which inhibits them from finding a suitable spouse. Thus, the matchmaking process may work best in arranged marriages in which more parties than just the couple are involved. If true, we expect fewer marital arguments and difficulties in arranged marriages compared with the other marriage types. That said, evidence from a somewhat similar culture—western China—does not support this idea (Xiaohe and Whyte 1990). Third, when arguments and difficulties occur, family support for the couple is most likely forthcoming in an arranged marriage. Families have a great interest in making an arranged marriage work because they negotiated it. This argument was put forward in the context of South Asians living in the UK (Samad and Eade 2003), but it matches the Central Asian context.
As a test for the validity of our results, we implement a placebo test. Maternal height is determined before women get married and should therefore be unaffected by the type of marriage.16 We regress maternal height on the marriage type, controlling for women’s age at marriage, educational attainment, language proficiency, living at high altitude, and district fixed effects (only in the OLS estimation). Table 5 confirms that kidnap-based marriage is not significantly related with maternal height. Although all coefficients are negative, they are far from being precisely estimated.
We conduct heterogeneity analyses and restrict these to the sample of kidnap-based and arranged marriages because birth weight differs most between these types of marriage. We test whether the effect of kidnap-based marriage is different for boys than for girls and for firstborn children compared with children of higher birth order (Table 6). Control variables are included but not shown in the table. In columns 1 and 2, we interact the dummy variable for the child’s sex with the kidnap dummy variable. There is modest son preference in Kyrgyzstan (Brainerd 2013), and we investigate whether this preference is stronger in kidnap-based than in arranged marriages. If so, the effect of kidnap-based marriage should be stronger for girls than for boys. Instead, we find that boys are more negatively affected than girls (insignificantly in the IV). Hence, there is no evidence for higher son preference in kidnap-based marriages. The stronger effect for boys may be explained with the fact that boys tend to be more vulnerable in the womb. Other studies have found similar results: maternal stress factors have an adverse effect on fetal growth in males but not in females (Black et al. 2016; Kirchengast and Hartmann 2009). In columns 3 and 4, we interact the dummy variable for the firstborn child with the kidnap dummy variable to investigate whether the negative effect of kidnap-based marriage decreases with the duration of marriage—presumably because spouses may learn to live harmoniously—and hence from firstborn children to children of higher birth order. We do not find evidence for such a development. In fact, firstborn children seem to be less negatively affected by kidnap-based marriage than later-born children. We leave this question to future research.
Plausibility of the Identifying Assumption
As outlined in the Empirical Strategy section, the identifying assumption for our IV estimation is that the district-level share of kidnapping affects the likelihood of kidnap-based marriage of an individual woman but does not have an independent influence on her children’s birth weight after we control for kidnapping. This may seem questionable if the prevalence of kidnap-based marriage is correlated with other district-level characteristics, such as the level of poverty, which also affect children’s birth weight. We address this concern in the following way. Because the Kyrgyz are the only ethnic group in Kyrgyzstan that has kidnap-based marriages, and because different ethnicities tend to live close to one another rather than in distinct communities, we test for a relationship between the district-level share of kidnapping and birth weight using data from other ethnic groups. If the district-level share captured information other than simply the prevalence of kidnap-based marriages, a correlation between the district-level share and children’s birth weight should emerge for the non-Kyrgyz. Fig. S1 in Online Resource 1 illustrates the frequency with which children born to non-Kyrgyz mothers of the age group 18–43 live in districts with varying prevalence of kidnap-based marriages. Even though the non-Kyrgyz in our sample do not reside in the districts with the highest kidnapping prevalence, and they live in only 23 of the 41 districts for which we calculated the district-level share, we observe substantial variation in the district-level share for this group. We then regress the birth weight of non-Kyrgyz children (sample mean: 3,220g) on the district-level share of kidnapping. Table 7 illustrates that there is no robust and no significant association between the district-level share and children’s birth weight for the non-Kyrgyz. Across the columns, the sign of the coefficient for the district-level share of kidnapping changes and the standard errors are very large.17
This test rests on two assumptions: (1) that the 23 districts used here are not systematically different from the other districts included in our data set; and (2) that there are no district-level differences captured by the instrument, such as gender-specific norms, that affect the Kyrgyz but do not affect the other ethnic groups.18 Possibly, these assumptions do not hold. With regard to the first concern, we repeat the estimation of Eq. (1) for the subset of the 23 districts included in Table 7. In the OLS estimation (including district fixed effects), children born to mothers in kidnap-based marriages are found to be smaller than children born to mothers in other types of marriage. The birth weight loss amounts to 70g (significant at the 10 % level) for the comparison with all non-kidnap-based marriages, 58g (significant at the 10 % level) for the comparison with love marriages, and 128g (significant at the 5 % level) for the comparison with arranged marriages. In the IV estimation, only the comparison of kidnap-based with arranged marriage shows a statistically significant (at the 10 % level) birth weight difference: –249g. These numbers indicate that the effect of kidnap-based marriage on birth weight in the subset of districts included in Table 7 is similar (slightly larger) to that in the total sample of districts. In sum, there is no reason to believe that this subset of districts is systematically different such that the analysis in Table 7 would be invalidated.
With regard to the second concern, we analyze whether birth weight (of the first child), as well as a number of variables that may be related to birth weight, vary with the district-level share of kidnapping for Kyrgyz but not for non-Kyrgyz mothers. This analysis aims at determining whether district characteristics, captured by the instrument, disadvantage the Kyrgyz over the non-Kyrgyz. Beside birth weight, we use information on whether a woman smokes, the average number of doctor consultations per woman in the last 12 months, whether a woman (alone or together with her husband) has main decision-making power over issues of child health, and average household income (in the Kyrgyz currency, som).19 For the sample of Kyrgyz and non-Kyrgyz mothers of the younger generation, we regress these variables on the district-level share of kidnapping, a Kyrgyz ethnicity dummy variable, and an interaction term. We control for age and education. Our main interest is the coefficient on the interaction term. We restrict the analysis to the 23 districts used in Table 7. The results are reported in Table 8. The interaction term is statistically significant only for doctor visits, which may be an indication of district-level differences captured by the instrument that vary by ethnicity: these differences are in favor of the Kyrgyz. We are thus cautiously confident that the instrument does not capture district characteristics that put ethnic Kyrgyz at a disadvantage.
Our identifying assumption could still be violated if the district-level prevalence of kidnap-based marriage affected a number of premarriage characteristics of Kyrgyz women that influence children’s birth weight. Health, education, and family background are such potential characteristics. We have data on women’s height, their highest education degree, and whether their fathers worked in agriculture. These variables potentially influence children’s birth weight, and two of the three are determined before women’s marriage age. An exception is schooling, which may continue after marriage. However, basic education, which consists of four years of primary school and the first five years of secondary school, is finished for all women before their marriage age. After basic education, women can continue with two more years of secondary school, potentially followed by tertiary education, or with technical school. The likelihood of these alternative tracks may vary with the prevalence of kidnap-based marriage per district. Thus, we analyze whether having only basic education is related to the district-level share of kidnapping for Kyrgyz women. We do the same for the other variables mentioned. The results in Table 9 show no association between the three variables and the prevalence of kidnap-based marriage.
In this study, we analyze the effects of marriage resulting from bride kidnapping—a form of forced marriage practiced in Kyrgyzstan and several other countries—on infant birth weight. We show that kidnap-based marriage negatively affects the birth weight of infants. Although the effect is small and statistically insignificant when we compare kidnap-based marriages with love marriages, it becomes relatively large and significant when we compare kidnap-based marriages with arranged marriages. Infants born in kidnap-based marriages weigh 80g to 190g (2 % to 6 % of average birth weight) less than infants born in arranged marriages, everything else equal. This difference is not negligible. It is in line with the very limited empirical evidence on violence against women and birth outcomes. Most critically, Aizer (2011) found that admission to hospital for assault during pregnancy reduces birth weight by 163g among women in California.
Our results are inevitably subject to some limitations. First, given the small sample size and the self-reporting of kidnap-based marriage, the estimated effect should not be regarded as definitive. Vital statistics records would be preferable to household survey data, but such records do not include information on the evolution of marriage. Second, marriage resulting from kidnapping is a specific form of forced marriage. Thus, the results obtained here cannot be generalized to all forms of forced marriage. Third, the results suffer from the fetal selection problem (Currie 2009). The LIK, as most other surveys, records information only on live-born children and is silent on stillborn children. If women in kidnap-based marriages had systematically more or less stillborn children than women in other types of marriage, the obtained estimates would be biased, although most likely in such a way as to understate the estimated effects.
Despite these limitations, we believe these findings to be the first statistically sound estimate of the adverse impact of forced marriage. In further research, important follow-up questions can be addressed. What is the channel through which forced marriage affects the birth weight of infants? Besides psychological stress, prenatal care and nutrition also seem plausible. What are the long-term consequences for children born to mothers in forced marriages? For policy recommendations, it is crucial to determine the extent to which birth weight loss translates into impairments in health, education, and labor market outcomes. Finally, it would be important to more accurately assess the intensity of trauma experienced by women in forced marriages, and hence to differentiate outcomes according to trauma intensity.
We thank four anonymous referees, Giovanna d’Adda, Victor Agadjanian, Kathryn Anderson, Bezawit Beyene Chichaibelu, Christopher Edling, Damir Esenaliev, Bernd Fitzenberger, Urakorn Fuderich, Adrian Garcia-Mosqueira, Robert Garlick, Priscilla Hermida, Aliya Ibragimova, Joshua Jacobs, Olga Kozlova, Bohdan Krawchenko, Natalia Kyui, Friederike Lenel, Sabine Liebenehm, Nathan Light, Mieke Meurs, Luciano Mauro, Huon Morton, Akylai Muktarbek kyzy, Dan Oldman, Kani Omurzakova, Sultan Omurzakov, Klara Sabirianova Peter, Daniel Schnitzlein, Cathy Starkweather, Artem Streltsov, Sebastian Vollmer, and Marlene Waske for helpful comments. We also received crucial feedback from participants of workshops, seminars, and conferences at the American University of Central Asia, El Colegio de Mexico, Humboldt University of Berlin, Leibniz Universität Hannover, University of Hamburg, University of Göttingen, University of Freiburg, and the New Economics School. We are grateful to the German Research Centre for Geosciences in Potsdam for providing us with altitude data. All errors, omissions, and faulty interpretations remain our own.
Because of the small size of our sample, we are unable to investigate other birth outcomes. We have too few observations for rare events, such as the incidence of low birth weight or neonatal and infant mortality.
These penalties are determined in a bride kidnapping bill that came into law in January 2013. Bride kidnapping had in fact been illegal since the 1920s, when the Soviet Union established laws banning it (Werner 2009).
It may seem awkward to ask such a question in a survey. When drafting the questionnaire, we consulted Nathan Light, an anthropologist who worked on the marriage market in Kyrgyzstan. He assured us that Kyrgyz people do not regard this information to be sensitive and that they talk about kidnappings openly. He encouraged us to ask the question as bluntly as we did. We know from conversations with Victor Agadjanian that questions about kidnapping were not problematic in their survey, either.
We only consider information about first marriages. The women in our younger generation sample were not necessarily still married to their first husbands at the time of data collection. A small share (2.1 %) of these women are married to a second husband, 7.6 % are divorced, and 2.2 % are widowed. We ignore the type of second marriage as well as children born during second marriages.
We would like to control for gestation duration, but as is typical for household surveys (see also Mansour and Rees 2012), this information is unavailable. Another important determinant of birth weight is smoking during pregnancy. Information on whether the women in our sample smoked at the time of pregnancy is also unavailable. However, very few women smoke in Kyrgyzstan; only 2.6 % of women in the younger generation smoked at the time of the survey.
There are 44 districts and 24 municipalities in the country; without Bishkek and Osh, these are 40 districts and 23 municipalities. Our sample contains observations from 35 districts and 10 municipalities.
The younger generation in our sample was born between 1968 and 1992. The oldest child of a mother in this younger generation was born in 1986.
We are grateful to Priscilla Hermida of Pontificia Universidad Católica del Ecuador for this suggestion.
Unfortunately, the LIK does not contain information on whether respondents still live in the same district where they were born.
Excluding them (in unreported regressions) has little effect on the empirical findings.
In kidnap-based and arranged marriages, the first child is born, on average, 1.3 years after marriage; in love marriages, an estimated 1.0 years after marriage. This timing is inexact because we lack information about month of marriage.
We have heard anecdotal evidence that secondary school and university graduation ceremonies are a popular time for kidnapping. Families of the groom often promise to pay for the bride’s education, so marrying after graduation reduces costs for the groom’s family. The results presented in Nedoluzhko and Agadjanian (2015) are consistent with this claim: being in school more than halves one’s risk of incurring either a forced or staged kidnapping.
The reduced sample size does not modify the results by much: running the OLS estimation with district fixed effects for the 41 districts included in Table 4 leads to a kidnapping coefficient of –43g (insignificant) for the sample of kidnap-based and love marriages, and –86g (significant at the 10 % level) for the sample of kidnap-based and arranged marriages.
We experimented with different functional forms in the first-stage estimation. First, we ran the first stage as a probit model and used the resulting predicted probabilities as instruments in the second stage. Second, we categorized the district-level share of kidnapping into deciles and included these deciles as instruments. Finally, we added a squared term of the district-level share as an instrument. The F statistic of the excluded instruments remained high, between 31 and 90. The null hypothesis could never be rejected (p value between .38 and .92).
In alternate estimations, we added district-level population size, employment rate, and unemployment rate as controls (separately and together) in the IV estimation. Our results did not change qualitatively. None of these variables entered significantly in the first stage. The district-level share of kidnapping still predicted individual kidnap-based marriage highly significantly and with a high F statistic for the excluded instrument. The effect of kidnap-based marriage on birth weight in the second stage remained negative and was imprecisely estimated.
We also have information on the prevalence of chronic health conditions, such as diabetes, heart disease, or high and low blood pressure, among our sample women. None of these are related with the type of marriage, either. However, we do not know whether these conditions were determined before marriage age or developed after marriage. Thus, they do not serve as a strict placebo test.
We combine all non-Kyrgyz ethnic groups. The same result emerges when we restrict consideration to Uzbeks, the largest minority group.
We are grateful to an anonymous referee for pointing this out.
In 2011, 1 US$ was worth about 45 Kyrgyzstani som.