1Sociodemographic studies have found that international migration can result in an increase in divorce (e.g. Andersson and Scott, 2010; Frank and Wildsmith, 2005; Hill, 2004; Landale and Ogena, 1995). Two explanations prevail. First, the act of moving is a stressful life event, resulting in a greater likelihood of divorce (Boyle et al., 2008), and this stress associated with moving might increase when international borders are crossed. Second, migration policies have become stricter, making it difficult to migrate as a family. Consequently, more families are geographically separated and faced with the challenge of arranging family life transnationally. While living transnationally might be unproblematic for some, for others it could cause marital stress and eventually result in divorce.
2Many studies evaluate the extent to which immigrants follow family formation or dissolution patterns that are similar to those of native counterparts in destination countries. Yet these studies are inconclusive about whether it is the act of migrating that leads to higher divorce rates because this would require a comparison with divorce rates of non-migrants in the origin country (Clark et al., 2009; Glick, 2010). Data for such comparisons are few and far between, as data collection typically takes place in destination countries (with some exceptions: e.g. Frank and Wildsmith, 2005; Hill, 2004). The current study compares divorce rates of Ghanaian couples with and without international migration experience. Ghana has high rates of both international migration (Twum-Baah, 2005) and divorce (Tabutin and Schoumaker, 2004). Although these findings could indicate a relationship between the two, anthropological studies argue that marital relationships in some parts of Africa – and Ghana is no exception – are historically flexible due to the effects of matriliny, the existence of polygyny, and wider sociopolitical conditions (Boni, 2001; Clark, 1994; Fortes, 1950; Manuh, 1999; Oppong, 1970, 1980). It is therefore important to compare migrants with their counterparts who stay in the country of origin.
3Migration between Africa and Europe includes independent male and female migration (Grillo and Mazzucato, 2008). Furthermore, increasingly strict migration laws make it difficult for couples to migrate together, so transnational couples, where one partner migrates and the other stays in the country of origin, are increasingly common. The analysis presented here therefore also compares transnational couples where the husband or wife migrates. By accounting for such couple configurations, this study pays particular attention to the different effects of male and female migration experiences, as previous studies have found gender differences in the effects of migration, such as changes in gender roles that affect men and women differently (Gallo, 2006; Hill, 2004; Jolly and Reeves, 2005).
4The case of sub-Saharan African migration studied here makes a new contribution to a scholarly literature that has predominantly focused on migration between Latin America or Asia and the United States, or on former guest workers and migrants from former colonies in Europe (e.g. Constable, 2003; Frank and Wildsmith, 2005; Glick, 2010; Hill, 2004; Landale and Ogena, 1995). This has resulted in a knowledge gap concerning “new” migrant groups, despite the fact that these groups constitute a significant proportion of existing migration systems. The contexts of migration in sub-Saharan Africa are distinctive, firstly because spousal separation is commonly practiced in many West African countries, and secondly because family reunification policies of their destination countries are more restrictive than they were in the 1970s and 1980s, when many guest workers reunited with their families (Mazzucato and Schans, 2011).
5We examine the relationship between migration and divorce by means of discrete-time event history analysis, using life histories collected from Ghanaian migrants, returnees, migrant spouses, and non-migrants in 2009. Data from the MAFE-Ghana dataset are used, with data collected in the Netherlands, the United Kingdom, and Ghana. The following section discusses several strands of the literature that address migration and divorce, and anthropological literature on marital relationships in Africa and Ghana in particular.
I – Theoretical framework
Migration and divorce
6Much of the sociological and demographic literature on the dissolution of unions has indicated that stressful life events are strong predictors; and one major stressor is moving (Boyle et al., 2008). Studies that examine the relationship between moving and divorce often use traditional models that focus on male pioneers with their so-called “trailing” wives. They show that migration typically benefits the careers of men, and that women’s labour market status is negatively affected by migration, regardless of their motherhood status, income or occupational status prior to migration (Boyle et al., 2008). Marital stress increases, as does the risk of divorce, because migration makes women “suffer”. This is commonly explained by the fact that family migration is strongly influenced by traditional gender roles that prioritize the male’s economic well-being.
7However, these traditional demographic studies often do not take international migration into account. Sociological studies on immigrant families, for their part, explore the relationship between family life and migration, mainly from the perspective of assimilation and acculturation arguments (Glick, 2010). For example, Bean, Berg, and Hook (1996) found that marital disruption behaviour of second- and third-generation Mexicans was similar to that of non-Hispanic Whites in the United States, which they attributed to processes of assimilation. Similarly, Phillips and Sweeney (2006) found that a migration experience is a strong predictor of marital stability among Mexicans in the United States when compared with other immigrant groups or the native population. Yet these findings do not shed light on the effect of migration on divorce as they do not make comparisons with the non-migrant population in the origin country, nor do they consider the transnational context of international migration.
Divorce in a transnational context
8In the context of international migration, it is not exceptional for couples to opt for a transnational living arrangement, with one of the spouses, typically the husband, migrating while the other remains in the country of origin. Living transnationally can be a preferred option, and family reunification, whether in the host or home country, is not always a feasible or desirable outcome (Baizan et al., 2011; Mazzucato and Schans, 2011). Quantitative demographers and family sociologists have given relatively limited attention to this phenomenon, leading to recent calls for more transnational approaches within these quantitative disciplines (Glick, 2010; Mazzucato and Schans, 2011). Moreover, despite this increased interest in the relationship between migration and family life, the impact of migration on the probability of divorce has received scant attention (Glick, 2010).
9With an increase in the availability of bi-national datasets, the Mexican Migration Project (MMP) being a well-known and long-running example, more quantitative scholars have taken up the challenges of including a transnational perspective and making comparisons with the non-migrant population when examining the impact of migration. For example, Frank and Wildsmith (2005) concluded that migration as such is not a sufficient causal factor in explaining union dissolution among Mexicans in the United States. Rather, extensive periods in the United States increase the risk of union dissolution among these Mexican couples. Hill (2004), studying Mexican and Central American women migrating to the United States, also indicated that the duration of a stay abroad increases the risk of divorce. Hill hypothesized that the risk of divorce is elevated because migrant women are exposed to different normative values concerning divorce in the United States. Exposure to different normative contexts and its effect on couples’ marital stability has also been the topic of several qualitative studies (e.g. Hirsch, 2003; Zontini, 2010).
Divorce and gender norms
10Qualitative analyses of marital stability have focused mainly on how migration modifies ideas about gender norms. Migrants are often confronted with conflicting gender norms from home and host countries, as well as among the migrant community in the host country. These previous studies have shown that international migration affects gender relations, revealing that couples’ relationships can become stressed, strengthened or altered in the context of migration (Fouron and Schiller, 2001; Mahler, 2001).
11Differences in spouses’ gender expectations and attitudes can be important stress factors and increase the risk of divorce (Boyle et al., 2008; George, 2000; Jolly and Reeves, 2005; White, 1990). Hirsch (2003) found that after migration to the United States, some Mexican migrant women experienced greater freedom from the constraining gender norms prevalent in their home communities, and were more likely to experience marriage instability. Zontini’s (2010) ethnographic work shows that women often change the prevailing gender roles in response to migration, by becoming breadwinners for example. Men do not necessarily recognize these new gender roles; they may feel that they have been marginalized or that their masculinity is under threat, which can lead to spousal conflicts (Charsley, 2005; Gallo, 2006; George, 2000; Manuh, 1999).
12Notwithstanding, two knowledge gaps remain. First, the abovementioned studies focus on either male or female migration, yet couples can experience migration in a variety of ways. They can migrate together, simultaneously or successively, or they can become transnational couples, with the wife or husband migrating while the other spouse remains in the country of origin. These experiences might affect marriages differently. Second, migrants from sub-Saharan Africa have been largely overlooked. The present article aims to address this knowledge gap by examining a sub-Saharan African migrant group, that of Ghanaians who migrate internationally.
Migration and divorce in Ghana
13Studies on divorce in sub-Saharan Africa are scarce (Tabutin and Schoumaker, 2004). The few studies that estimate divorce rates in Ghana have recorded high numbers: Tabutin and Schoumaker (2004) mention that 35% of women’s first marriages end in divorce after 30 years of marriage, and Takyi and Gyimah (2007) estimate that in 2003, around 25% of ever-married women aged 15-49 years had divorced. Several anthropological studies on marriage instability in Ghana discuss these seemingly high rates and the cultural notions concerning family relationships. These notions are important to consider, as they can shape individuals’ marriage and divorce decisions. They might thus explain the relatively high prevalence of divorce.
14In many West African contexts, the norms concerning couples’ living arrangements do not require geographical proximity for family life, and in Ghana, multilocal residence is quite common (Clark, 1994; Coe, 2011; Fortes, 1950; Manuh, 1999; Oppong, 1970). Traditionally, men and women lived apart, each spouse with his or her own family (Fortes, 1950), and this multilocal residence was practiced in both matrilineal and patrilineal descent groups (Oppong, 1970). Spousal separation in such a context might affect marital relationships differently than in contexts where proximity is viewed as a necessity for family life.
15Several authors point to external factors as an explanation for the relatively high prevalence of divorce, such as the diffusion of Western norms and values concerning individualism; others point to specific sociocultural features of Ghana, such as the importance of lineage ties. The latter influence marital relationships in that loyalty to one’s lineage causes conjugal bonds to weaken; consequently, divorce occurs relatively easily and frequently (Bleek, 1987; Oppong, 1980; Van der Geest, 1976).
16The prevalence of divorce is said to be even higher among Ghanaians with matrilineal kinship ties (Bleek, 1987; Takyi and Gyimah, 2007), such the Akan who constitute the majority of international migrants. Akan women are said to enjoy greater autonomy than their patrilineal counterparts, although some studies indicate that their independence is decreasing due to processes of modernization. Clark (1994) argues that the difficulties associated with matrilineage stem from the fact that women must manage their marital households in addition to their work. Furthermore, husbands feel greater responsibility toward their own matrilineage, and their interest in their marital household decreases when their wives become more independent.
17In general, studies indicate that women in Ghana are quite independent, whatever their lineage. According to Oppong (1970), the majority of women in Ghana work outside the household and have done so traditionally. This, combined with the practice of separate residence, has led to relationships that are not necessarily egalitarian, but are characterized by the autonomy of spouses. Women’s greater autonomy is, in turn, associated with marital instability, as these women may feel better able to establish independent households and experience fewer moral obligations to remain in a marriage (Boyle et al., 2008).
Ghanaian couples and international migration
18Given these anthropological insights, two contesting hypotheses can be formulated. First, in line with sociological and demographic studies, migration is expected to increase divorce rates due to an increase in marital stress caused by migration. Yet, second, taking the Ghanaian context into account, migration is expected to have little or no effect on divorce rates because multilocal residence is relatively common among Ghanaian couples. In addition to these two contesting hypotheses, different outcomes are expected for couples with male, female, and joint migration, since men and women experience migration differently. Additionally, specific migration characteristics are expected to influence the divorce rates of migrant couples, such as the length of time spent apart (more time spent abroad increases the risk of divorce) (e.g. Frank and Wildsmith, 2005; Hill, 2004), the region of migration (migration to a Western context can create marital tensions, e.g. Charsley, 2005; Gallo, 2006; George, 2000; Manuh, 1999; Zontini, 2010), and whether the couple was already geographically separated at the start of marriage.
19In addition to the role of migration in explaining the probability of divorce, previous studies have identified other important predictors. The effect of the presence of children has been extensively researched, and most studies find that it discourages divorce (Boyle et al., 2008; Frank and Wildsmith, 2005; Hill, 2004; White, 1990). Studies on the effect of educational attainment on the risk of divorce have remained inconclusive (Amato, 2010; Takyi and Gyimah, 2007). Several scholars have found that higher levels of education decrease the risk of divorce (Boyle et al., 2008), while others have pointed to a reversed effect for women’s educational attainment (Frank and Wildsmith, 2005; Kalmijn et al., 2004). Previous studies have shown that several measures of low socioeconomic status can result in an increased risk of divorce (White, 1990), and lower ages at marriage are also associated with a higher risk (Boyle et al., 2008; White, 1990). Finally, couples who were in a relationship (either cohabiting or not) prior to their marriage are likely to be more stable (similar to the “trial marriage” theory) (Kulu and Boyle, 2010). These variables are controlled for in the analyses.
II – Data
20Our study is based on a longitudinal biographical dataset, collected in 2009 in Ghana, the United Kingdom, and the Netherlands as part of the MAFE-Ghana project. The survey was carried out in urban areas of Ghana (Accra and Kumasi), the Netherlands (Amsterdam, The Hague, and Almere), and the United Kingdom (London). Current migrants (in the Netherlands and the United Kingdom), non-migrants, returnees, and migrant spouses (in Ghana) were asked identical questions. Both in Europe and in Ghana, respondents were eligible if they were aged between 25 and 75 years and born in Ghana.
21In Ghana, data were collected using stratified random samples of households in the cities of Accra and Kumasi. First, a sampling frame was used to randomly select house-holds, and the sampling frame was then stratified to oversample households with return migrants. Next, individuals were selected from these households. All return migrants and migrant spouses, if they were currently living in the household, and one randomly chosen other eligible member were selected. In total, 1,243 life event history questionnaires containing retrospective life histories were collected in Ghana.
22No suitable sampling frame was available in the Netherlands or the United Kingdom due to large numbers of undocumented migrants, so quota sampling, setting quotas for age and gender, was used. Respondents were recruited in various ways (e.g. some were recruited at the underground station, at church, at other public places) and interviewers from a variety of sociodemographic backgrounds were hired to increase the likelihood of accessing respondents with different sociodemographic characteristics. In total, 422 Ghanaian migrants were surveyed in Europe (273 in the Netherlands and 149 in the United Kingdom). The three datasets combined yielded 1,665 respondents. The use of retrospective data made it possible to examine marriages formed over the past 60 years. As retrospective surveys about peoples’ attitudes are unreliable, direct measures of attitudinal variables, on gender norms for example, could not be used. Instead, this study focused on couples’ behaviour, distinguishing between husbands and wives.
23Couples’ retrospective data were used from the year of a respondent’s first marriage until divorce or, in case of censoring, up to the time of the survey (2008) or the death of a spouse. A sub-sample of couples who were married for at least one year was selected. Information concerning respondents and their spouses is asymmetrical because information on spouses was obtained through respondents. For respondents, detailed retrospective information concerning a wide variety of modules was available (e.g. housing histories, occupational histories, migration histories). For spouses, basic sociodemographic information referring to the situation at the start of the marriage and migration histories were derived through the network module of the questionnaire.
24We included migration periods of either spouse that lasted for at least one year. Complete information concerning start and end years of marriage and migration periods was also required. Additionally, polygamous couples, couples where the spouse was not Ghanaian, and couples where the wife migrated and her husband followed were excluded because these groups were too small to permit specific analysis. These restrictions yielded a dataset of 927 couples, of whom 144 divorced during the observation period. The first divorce occurred in 1954, and the last in 2008.
25In order to examine the relationship between divorce and migration, we analysed two models. In the first model, we studied to what extent a migration experience affects the probability of divorce, taking the full observation period and the full analytical sample into account. The observation period starts at the time of marriage, and ends when either the marriage ended in divorce or when censoring occurred. In the second model, we examined only couples who have experienced migration, and observed what happened during their migration period. To do so, we excluded couples without migration experience and we additionally changed our observation period. The observation period now starts at the time of the first migration after the couple was formed. We then studied whether, at the end of the migration period, the couple had experienced the event of divorce, or not. This means that the end of the observation period coincides with the end of the migration period, or when the couple divorced or when censoring occurred. Since all couples without migration experience were excluded from this model, the dataset for model 2 consisted of 442 couples, of whom 44 were divorced. Due to the small sample size, we controlled for only a limited number of variables (i.e. educational level of both spouses and the couples’ subjective wealth status). This two-pronged approach allowed us to study the effect of migration on the risk of divorce (Model 1), and to additionally examine how specific characteristics of the migration experience affect this risk (Model 2).
The probability of divorce
26A discrete-time proportional hazard event history model was used (Singer and Willett, 2003) to assess the probability of divorce. Couple-year datasets consisting of 12,481 and 3,775 couple-years were designed, for the first and second model, respectively. Using these datasets, complementary log-log models were used because the underlying survival process is continuous, even though the data were collected on an annual basis (Jenkins, 2005). Duration dependency was assessed in the models using the years of marriage and two polynomials of the years of marriage (a squared and a cubic term). The inclusion of these three terms fitted the data best. All time-varying variables were lagged by one year, following standard event-history procedures, which rest on the assumption that changes in the covariates in the previous year will affect the probability of divorce in the current year (Singer and Willett, 2003).
Couples’ migration experience and control variables
27In this study we paid particular attention to the gendered effects of migration. We did so by constructing a time-varying variable with information obtained from the migration histories of respondents and their spouses. The resulting couples’ migration experience variable consists of the following time-varying categories: 1 = couple-years without migration experience; 2 = couple-years where only the husband migrated; 3 = couple-years where only the wife migrated; 4 = couple-years where both spouses migrated, but the husband preceded the wife; and 5 = couple-years where both spouses migrated simultaneously. 
28The second model focused exclusively on couples with migration experience, which allowed the inclusion of migration-specific characteristics. First, two variables identifying the region of migration for the husband and the region of migration for the wife were taken into account. Both variables are time-variant, indicating whether the husband or wife migrated to Europe or North America or not (0 = no, 1 = yes). Second, the number of years couples spent living geographically separated across borders during their marriage, was included. Third, we included a variable indicating whether the couple was geographically separated across borders at the time the marriage started, referring to whether the husband and wife were living in the same country or not when they married (0 = no, 1 = yes).
29Marriage cohort refers to the period in which the couple got married, considering the following 5 time periods: 1 ≤ 1970, 2 = 1971-1980, 3 = 1981- 1990, 4 = 1991-2000, and 5 ≥ 2001. Husbands’ and wife’s educational levels are time-varying and indicate the spouses’ educational level in four categories, with 0 = no education, 1 = primary education, 2 = secondary education, and 3 = tertiary education.
30Respondents’ ethnicity is a time-constant variable that refers to whether the respondent is from a matrilineal ethnic group, e.g. Akan, or not. As religion strongly correlates with ethnicity (Akan are almost exclusively non-Muslims), only ethnicity was taken into account. Since income is difficult to reliably measure with a retrospective survey, we used respondents’ subjective wealth status as a measure for socioeconomic status. This variable varies over time and indicates the subjective wealth status of the respondent for each year. The following question was asked: “Would you say that during this period you had enough to live on?” This resulted in three response categories, 1 = absolutely, 2 = it depended, and 3 = not at all.
31Respondent’s age at the start of the marriage is added as time-constant covariate. Union before marriage is a time-constant variable that indicates whether the couple was in a consensual union before they married. Presence of children is a continuous time-varying variable that indicates whether the couple had children together. It refers to children born to both spouses, whether or not during the marriage. In some cases (n = 87), children were also born outside the marriage, with a different spouse; however, controlling for this did not result in a significant effect on the probability of divorce or an improved fit of the models. Appendix Table A.1 presents all of the variables used for Models 1 with all couples, as well as the variables for Model 2 including only couples with migration experience. For the time-varying variables, descriptive statistics are provided for the beginning of the observation period (the year of marriage or the first migration if it occurred after marriage) and the end (the year of divorce or the survey year in the case of censored observations).
III – Findings
32The cross-tabulations shown in Table 1 reveal that couples without migration experience and couples who experienced joint migration have similar divorce rates (19.0%, and 19.8%, respectively). Additionally, couples where only the wife migrated were also found to have a high divorce rate (13.6%). Couples where the husband migrated, either independently or as a pioneer (with his wife following), have a much lower prevalence of divorce (8.2% and 7.8%, respectively). The same pattern is observed when examining only couples with migration experience, among whom couples where the husband was the only migrant or the pioneer migrant divorce less frequently than couples where the wife migrated, either independently or jointly with her husband.
Divorce experience of Ghanaian couples with and without ever experiencing migration (time-constant)
Divorce experience of Ghanaian couples with and without ever experiencing migration (time-constant)Significance levels: * p < 0.1; ** p < 0.05; *** p < 0.01.
33We examined the proportion of divorcees in the full sample for each year. After 5 years, 5.0% of the sample are divorced; after 10 years, this percentage increases to 10.2%, and after 15 years, to 15.4%. In total, 30.8% of the couples in the sample divorced. This percentage is in line with previous findings that 35% of women’s first marriages in Ghana end in divorce (Tabutin and Schoumaker, 2004).
Couples’ migration experience and the probability of divorce
34A multivariate discrete-time event history model was estimated to investigate the effect of couples’ migration experience on the probability of divorce by including the variable of interest and the control variables. Table 2 presents the results of two complementary log-log regression models. Model 1A assesses the unconditional effect of the key variable of interest, and in Model 1B the control variables are included.
Discrete-time event history analyses of the risk of divorce for Ghanaian couples (complementary log-log estimates)(a),(b),(c),(d)
Discrete-time event history analyses of the risk of divorce for Ghanaian couples (complementary log-log estimates)(a),(b),(c),(d)(a) Duration of marriage was transformed using “duration of marriage/10” in order to avoid high coefficients for the cubic parameters.
(b) Ethnicity (respondent): 0 = non-Akan, 1 = Akan.
(c) Couple in union before marriage: 0 = no, 1 = yes.
(d) % divorced based on Kaplan-Meier estimates that take into account censored observations.
Significance levels: * p < 0.1; ** p < 0.05; *** p < 0.01.
35Model 1A shows that couples who only experienced female migration have a significantly higher risk of divorce than couples without migration experience. Similarly, the risk of divorce increases significantly when both partners migrated simultaneously. Couples where the husband migrated and couples where the wife followed her husband show no significant difference compared to couples without migration experience. These findings remain after including the control variables in Model 1B.
36Model 1B shows a slight decrease in the probability of divorce for the period 1981-1990 compared with the reference period before 1970. Although previous studies found decreasing levels of divorce in Ghana over time (e.g. Takyi and Gyimah, 2007), we found no significant effects for the other time periods.
37Considering the educational attainment of both spouses, there are no significant differences in the probability of divorce between women with no schooling compared to women with higher levels of education. For men, only primary education significantly increases the probability of divorce compared to having no schooling.
38Respondents’ age at the time of marriage does not affect the probability of divorce, which is surprising considering that most studies have found that younger ages at the start of the marriage increase the risk of divorce (Amato, 2010). Likewise, having been in a non-marital union prior to the marriage does not affect the risk of divorce. In the case of subjective wealth status, the risk of divorce increases for those who do not consider themselves well-off compared to those who do, which is consistent with the literature that suggests that lower socioeconomic status increases divorce risk (White, 1990). The presence of children in the marriage does not have a significant effect.
39Predicted probabilities of divorce were calculated from Model 1B in Table 2, for each of the five categories capturing couples’ migration experience. Average levels were used for the reference population (marriage cohort = 1981-1990; ethnicity = Akan; wife’s education = secondary; husband’s education = secondary; subjective wealth status = it depended; age marriage started = 27.33 years; couple in union before marriage = yes; number of children (mean at time of censoring/divorce) = 2.04).
40Figure 1 shows the results for marriage durations of 1-25 years. For all couples, probability of divorce is highest at around 10 years of marriage. Figure 1 shows similar divorce risks for couples without migration experience, couples where only the husband migrated or couples where the wife followed her husband. By contrast, much higher predicted probabilities are observed for couples where the wife migrated and couples where both partners migrated simultaneously. To summarize, couples’ international migration experience is an important predictor of the probability of divorce.
Risk of divorce by marriage duration, estimated from Model 1B
Risk of divorce by marriage duration, estimated from Model 1B
Migration characteristics and the probability of divorce
41The second multivariate models only considered couples with migration experience. Migration-specific characteristics were included as follows: in Model 2A, the migration experience of the couple was included; in Model 2B only couples where the wife experienced migration (i.e. excluding those where the husband experienced independent migration) were examined, to investigate the effect of the wife’s migration to Europe or North America compared with her migration to African/other countries; in Model 2C only couples where the husband experienced migration (i.e. excluding those where the wife experienced independent migration) were examined, which allowed us to study the effect of the husband’s migration to Europe or North America. Additionally, all Models 2 included the length of time spent geographically apart and whether the couple was geographically separated at the start of the marriage. The results are presented in Table 3 below.
Discrete-time event history analyses of the risk of divorce for Ghanaian couples with migration experience (complementary log-log estimates)(a),(b),(c)
Discrete-time event history analyses of the risk of divorce for Ghanaian couples with migration experience (complementary log-log estimates)(a),(b),(c)(a) Geographically separated at marriage: 0 = no, 1 = yes.
(b) Duration of marriage was transformed using “duration of marriage/10” in order to avoid high coefficients for the cubic parameters.
(c)% divorced based on Kaplan-Meier estimates that take into account censored observations.
Significance levels: * p < 0.1; ** p < 0.05; *** p < 0.01.
42Like Models 1A and 1B, Model 2A in Table 3 reveals that couples where the wife migrated independently have increased probabilities of divorce relative to couples where the husband migrated independently. The effect of joint migration experience disappears, and instead we find that couples where the wife followed her husband are less likely to divorce. Surprisingly, the probability of divorce decreases when couples spend more time living apart due to migration. This might reflect the fact that, for some couples, living apart together has become a stable and long-term arrangement. Whether or not the couple were geographically separated due to migration at the start of the marriage does not affect the probability of divorce.
43In Models 2B and 2C, the couples’ migration experience variable is replaced with two dummy variables, representing whether or not the husband and/or wife migrated to Europe/North America. For the husbands, there is no difference when considering migration to Europe/North America (Model 2C), while for the wives, migration to Europe or North America significantly increases the probability of divorce (Model 2B). Being geographically separated at the start of the marriage does increase the risk of divorce for couples where the wife migrated, but not for couples where the husband migrated. For couples where the husband migrated, we also find that a longer period spent apart decreases the risk of divorce, while this effect is not present for couples where the wife migrated.
44Since the sample size was much smaller for Models 2 than for Models 1, we controlled for a limited number of variables. The two variables capturing duration dependence reveal that the probability of divorce follows a U-shaped pattern in Model 2A, with an increased risk after the first years of migration, followed by a later decrease. There is no effect of duration in Model 2B and a linear effect in Model 2C, indicating that the risk of divorce increases after the first years of migration. Contrary to the previous models in Table 3, no effects of educational level are found for either the husband or the wife. For subjective wealth status, not having enough to live on (“not at all” response) significantly increases the probability of divorce, in Model 2A, in line with Models 1. This effect is present but not significant in Models 2B and 2C.
45The predicted probabilities of divorce for couples with migration experience are calculated based on Model 2A, using the same average levels for all covariates as for Model 1B. The average number of years spent apart (4.80 years) was used, and couples who were geographically separated at marriage were included. Couples where the wife migrated have a much higher probability of divorce than couples where the husband migrated. Couples where the wife followed her husband have a much lower probability of divorce, while for couples where both partners migrated simultaneously the probabilities are similar. This again indicates that which partner migrates is an important predictor of divorce probability.
IV – Discussion
46The current article aims to contribute to the literature on migration and divorce in several ways. Few studies have so far focused on this topic (Glick, 2010), and most existing research has compared immigrant populations with native populations in the migrant receiving country. Following in the footsteps of several other scholars (e.g. Frank and Wildsmith, 2005; Hill, 2004), we examine migration from a bi-national perspective by comparing migrants with their non-migrant counterparts from the same origin country. Moreover, previous studies investigated either male migration or female migration only, whereas our analysis examines the migration experience from the couple’s perspective, providing scope to assess the different ways in which a couple can experience migration. We also scrutinize the effects of migration more closely by considering certain migration characteristics pertaining to (a) the region of migration, (b) the time the couple spent living apart due to migration, and (c) whether the couple married while living in different countries. Our comparison of two destination regions is original in that most studies focus on one receiving context (generally the United States). This article also examines whether the effects of migration to a Western destination are different from those of migration to a non-Western country. Finally, while studies dealing with migration and family life have been dominated by migration from Latin America and Asia, this article adds a case study from sub-Saharan Africa.
47Our findings reveal the existence of gender differences in the impact of migration on divorce. Migration increases divorce risk when women migrate without their husbands, or when both partners migrate simultaneously. Based on these findings, several hypotheses can be put forward. In many African contexts, norms concerning marriage and gender roles have been characterized as highly flexible. It is possible that for couples where a husband migrates internationally, this situation is experienced in the same way as the traditional multilocal residency of spouses, and therefore has no effect on divorce risk.
48Women who migrate to a Western context are more prone to divorce than those who migrate to other African countries, but this is not the case for men. This might reflect the importance of the receiving context in terms of altering gender norms, or a desire to escape a (restrictive) marriage (Hill, 2004; Hirsch, 2003; Jolly and Reeves, 2005; Manuh, 1999; Zontini, 2010). Migration to Western countries can also create tensions between spouses due to changes in gender roles, for example, when the wife becomes the main breadwinner. Alternatively, these findings may be explained by other differences that exist between migration within Africa and migration to Europe or North America. When a family member migrates within Africa, people in the sending countries have lower expectations of the benefits to be reaped. This may in turn lead to fewer tensions between spouses. Likewise, easier movement within Africa due to cheaper travel or more relaxed boarder enforcement may provide more opportunities for face-to-face contact, which can reduce marital stress. Overall, the current findings demonstrate the importance of taking the destination context into account.
49The probability of divorce decreases for couples where the husband migrated and the wife followed. This contradicts previous studies conducted in a European context which found that marriages where the wife follows the husband are more unstable, likely due to marital stress caused by the wife’s loss of labour market status after migration (Boyle et al., 2008). Our different findings may be an indication that the labour force situation of the “trailing” Ghanaian wife does not necessarily deteriorate in the European context. Alternatively, marriages where the wife is a “trailing spouse”, may be more stable because the wife is in a vulnerable position due to her dependence on her husbands’ status (Kraler, 2010). However, further research is needed to examine the effect of the labour market status of both spouses before, during and after migration on the probability of divorce, as well as the potentially vulnerable position of trailing wives.
50Previous studies (e.g. Frank and Wildsmith, 2005; Hill, 2004) have indicated that a longer period of separation is related to an increase in divorce. For Ghanaian couples, we found a reverse effect: for equivalent marriage durations, longer periods apart decreased the risk of divorce. Importantly, though, in cases where the wife migrates, the wife’s region of migration is a more important predictor. When destination is controlled, the effect of duration is no longer significant, although it likely to be captured by the variable of wife’s migration region, as when a wife migrates to a Western context, couples spends more time apart, on average, and have a higher probability of divorce.
51The above finding that, in most cases, divorce rates of couples with migration experience are very similar to those without, except when the wife migrates, highlights the fact that high divorce rates are part and parcel of Ghanaian society and not necessarily brought on by international migration. The importance of local context was also taken into account in this study by including information about whether the respondent is part of a matrilineal lineage, which is claimed to increase the probability of divorce (Boni, 2001; Takyi and Gyimah, 2007). Although there are different pressures on a wife, depending on whether she is part of a matriliny or patriliny (Clark, 1994; Oppong, 1970), the findings show that matriliny is not associated with a higher probability of divorce.
52While we focused in this article on the effect of migration on the probability of divorce, we also found some surprising results for our control variables. Two results are particularly worth mentioning, as they seem to deviate from previous findings. First, we found limited effects for the husband’s educational level, and no effects for wife’s educational level. Although research on the relationship between educational levels and divorce remains inconclusive, the majority of studies associate higher educational attainment with higher risks of divorce (Frank and Wildsmith, 2005; Kalmijn et al., 2004; Takyi and Gyimah, 2007). Two divergent effects may perhaps be at play: the higher educated may indeed have more means to escape unhappy marriages, but at the same time, the higher educated may wait longer and make more considered choices when seeking a partner, leading to greater marriage stability. The combined presence of both positive and negative effects of education on divorce in our sample might explain the non-significant effects for education. Second, the known relationship between age at marriage and divorce, with those marrying young having higher risks, was not found in our study either. Future research is needed to further investigate these unexpected results.
53This study has highlighted the importance of a gender perspective in analysing the effects of international migration on divorce. There are two avenues of research that can help carry this type of analysis forward. First, surveys can collect more information to assess whether changes in gender relations are part of the explanation of the elevated risk of divorce for couples where a wife migrates. Such information includes the reasons for divorce and who initiated the divorce.
54Second, the relationship between divorce and migration may perhaps work in the reverse direction to the one considered in this study: that is, divorce causes people to migrate. As some qualitative studies indicate, divorce may be a source of social stigma, and as such will induce people to migrate. In that case, the findings reported here are likely to underestimate the relationship between migration and divorce as they report only one part of it. As this study did not account for unobserved heterogeneity, migrants may be more likely to divorce due to unobserved characteristics that give them a latent propensity to divorce. If this is indeed the case, our results are likely to overestimate the effect of migration on divorce. This study has included some pre-migration characteristics of couples in order to establish causality between migration and divorce. Future studies can carry this further by including more variables concerning the period before migration to investigate the extent to which divorce and migration are interrelated events.
55Despite these limitations, this study is one of the few that compares the probability of divorce between migrants and a non-migrant population. This was made possible due to the unique features of the MAFE-Ghana data. Through this comparison, we are able to better identify the relationship between migration and the probability of divorce. A couples’ perspective further refined our analyses, revealing that marital stability also depends on which spouse migrates. Finally, this article stresses the importance of taking the contexts of both home and host countries into account.
Descriptive statistics of the independent variable at the start and end points of the observation period(a),(b),(c),(d),(e),(f)
Descriptive statistics of the independent variable at the start and end points of the observation period(a),(b),(c),(d),(e),(f)(a) Mean and standard deviation. Time spent apart: 0-33 years. No missing data.
(b) Several control variables were omitted when considering only couples with migration experience to avoid overestimating the models with a small sample size.
(c) Mean and standard deviation. Duration of marriage: 1-55 years in the full sample; 1-39 years for couples with migration experience.
(d) Husband’s education: for the sample with migrants only, the categories “no schooling” and “primary education” were collapsed.
(e) Mean and standard deviation. Respondents’ age at marriage:13-66 years. No missing data.
(f) Mean and standard deviation. Number of children: 0-11. No missing data.