Few studies have examined the causes and consequences of marital dissolution in non-Western settings. This article explores the fundamental factors that may predict marital dissolution in a mainly agrarian setting in South Asia, where collectivism has historically been valued over individualism and where life is centered on the family. Using event history analyses with retrospective life history data from the Chitwan Valley Family Study conducted in rural Nepal, I explore the possible predictors of marital dissolution. Results suggest that couples in which wives married at older ages and chose their spouse in conjunction with their parents face lower risk of marital dissolution, while wives’ work increases the risk. Moreover, couples married for longer durations and couples who have more children face lower risks of marital dissolution. The influences of many of these factors have changed over the last few decades, pointing toward the important role of changing social context on marital trajectories.
A solid foundation of research has explored the causes and consequences of divorce in Western settings (e.g., Amato 2000; Becker et al. 1977; Coontz 2007; Hannan et al. 1977; McLanahan and Bumpass 1988; Morgan and Rindfuss 1985), but less evidence is available to facilitate an understanding of divorce in settings where limited personal freedom, a high value placed on collectivism and family obligations, and ingrained patriarchal norms are prevalent. Without a foundation of empirical research to establish the fundamental factors influencing marital dissolution in such settings, we cannot begin to understand the intricacies of the causes of marital dissolution or the consequences of it. The present investigation focuses on one such setting: rural Nepal. The characteristics of this population allow the opportunity to develop new insight into the process of marital dissolution and how that process has changed simultaneously with pervasive changes to the social context.
In rural Nepal, recent social changes have meant that people are now more exposed to activities and ideas outside their immediate communities than in the past (Axinn and Yabiku 2001; Barber and Axinn 2004; Thornton 2005). Even as family life has been changing along with these changes to the social infrastructure (Axinn and Barber 2001; Axinn and Yabiku 2001; Ghimire et al. 2006), marital dissolution remains uncommon in this setting, similar to other settings in South Asia (Dommaraju 2016; Dommaraju and Jones 2011; Goode 1993). This does not preclude that a nontrivial number of women and families have ever experienced marital dissolution (Dommaraju and Jones 2011). The potential for the rapidly changing social context to impact marital trajectories, combined with a lack of understanding about marital stability and instability in this kind of setting, makes it pertinent to investigate the causes of marital dissolution and changes in those causes over time.
Background and Setting
The study area for this investigation is in the southern region of Chitwan, Nepal. Chitwan is mainly rural—a characteristic that it shares with two-thirds of the population of South Asia (Population Reference Bureau 2015). Like most of Nepal, the population of Chitwan is dependent on subsistence agriculture (Axinn and Ghimire 2011). It was only as recently as the 1950s that Chitwan was settled, having previously been covered with forest and infested with malaria (Axinn and Yabiku 2001). Over these past few decades, a major hub—the city of Narayanghat—developed, schools were built, and transportation and communication systems were expanded (Axinn and Yabiku 2001).
In general, marriage in Nepal occurs at early ages and is relatively universal (Yabiku 2005). Most marriages have historically been arranged, at least in part, by parents and/or other relatives (Ghimire et al. 2006). However, this practice has been loosening recently, and it is now more common than in the past for young people to participate in choosing their marriage partner (Ghimire et al. 2006; Niraula 1995).
Ethnicity plays an important role in marriage practices and in gendered expectations regarding marriage. Ethnicity in Nepal is complex, multifaceted, and related to both caste and religion. (For detailed descriptions of the different ethnic groups, see Bennett (1983); Cameron (1998); Fricke (1986), and Guneratne (2002).) Although Nepal is an ethnically and a religiously diverse country, the majority of Nepalese people identify as Hindu (84 % of respondents in the study used here, as of 2008), and the country has historically been governed by high-caste Hindus: the Brahmins and Chhetris (Guneratne 2002; Levine 1987). These families of high standing are motivated to protect their prestige, and they therefore hold their daughters to stricter expectations for following Hindu customs. For example, Brahmin and Chhetri peoples have had a history of arranging marriages for their daughters prior to their daughters’ first menstruation in order to ensure that pregnancy will occur only within marriage (Bennett 1983; Dommaraju 2016; Niraula and Morgan 1996). Other ethnic groups are not held as strictly to Hindu customs (Cameron 1998; Fricke 1986; Guneratne 2002), and their marital practices tend to be less rigid.
Marital Dissolution in Nepal
Under Hindu decree, after a marriage occurs, it is bound for life and indissoluble (Holden 2008). Women face greater obstacles in dissolving a marriage than men. A divorced woman is considered to be “impure” for any man other than her first husband, as evidenced by the stigma against widow remarriage in South Asia (Niraula and Morgan 1996). Additionally, women are disadvantaged in inheritance laws and in their legal ability to file for divorce (Allendorf 2007; Gilbert 1992). Moreover, they have few opportunities to support themselves financially, especially in rural areas like Chitwan (Allendorf 2007). Still, records from the District Court of Chitwan indicate that divorces are most often filed by wives.1 This is likely due to the history of polygamy in Nepal, allowing men to take a second wife without formally dissolving their first marriage (Deuba and Rana 2001). A woman who wants to remarry, on the other hand, must dissolve her first marriage. For women without prospects for remarriage, separation may be a more desirable option than divorce because separation can allow for continued support from her husband and his family.
Despite the disincentives to divorce, court records reveal that the number of people filing for divorce is on the rise (see Fig. 1), suggesting an increased willingness over time to publicly seek divorce. In the sections that follow, I discuss the fundamental factors that may lead couples to dissolve their marriages, and then I empirically investigate the influence of these fundamental factors on marital dissolution and how those associations have changed over time.
Few empirical studies have examined the causes of divorce among South Asian populations (Bose and South 2003; Dommaraju 2016; Ranga Rao and Sekhar 2002). However, in addition to the extensive research among Western populations, some studies have focused on populations in East and Islamic Southeast Asia. In those Asian settings, divorce rates were high in the earlier part of the twentieth century and then fell as the social context and marital practices changed (Cammack and Heaton 2011; Guest 1992; Hirschman and Teerawichitchainan 2003; Jones et al. 1994; Thornton and Lin 1994; Yang and Yen 2011). Those settings are distinct from South Asia, where divorce rates have been consistently low (Dommaraju 2016; Dommaraju and Jones 2011; Jones 2010). Because of the different cultures, family practices, religious beliefs, economies, and politics, we cannot directly compare the influences on marital dissolution in South Asia with those in other settings (Goode 1993). Nonetheless, the existing knowledge from these other settings provides a source from which we can glean insight into marital dissolution in South Asia.
Some of the most reliable correlates of divorce are age at marriage, marital duration, education, work, and fertility. Specifically, those who marry at older ages and those married for longer durations are less likely to experience a marital dissolution (Becker et al. 1977; Bose and South 2003; Hirschman and Teerawichitchainan 2003; Jones 1994; Morgan and Rindfuss 1985; South 2001; Trent and South 1989; Yang and Yen 2011). Couples in which wives have greater educational attainment may be less likely to dissolve their marriages (Bose and South 2003; Guest 1992; Hirschman and Teerawichitchainan 2003; Jones 1994; Martin and Bumpass 1989; Stanley et al. 2006), although some work has shown that wives’ education can be positively associated with the likelihood of a dissolution (Heaton 1990; Teachman 2002; Thornton and Lin 1994). Findings regarding wives’ employment are more consistent, revealing that those couples in which wives work, work more hours, or make relatively more money face greater risk of marital dissolution (Kalmijn et al. 2007; Schoen et al. 2002; South 2001; South and Spitze 1986; Thornton and Lin 1994). Finally, the presence and greater number of children tends to suppress the likelihood of marital dissolution (Becker et al. 1977; Guest 1992; Morgan and Rindfuss 1985; Morgan et al. 1988; Waite and Lillard 1991).
The influence of these factors on marital dissolution depends on the preexisting social context (Goode 1993). In the following paragraphs, I combine knowledge from existing research with data from fieldwork to explore possible predictors of marital dissolution in Nepal and how those predictors may have changed over time.
With the prevalence of arranged marriage, level of spouse choice is likely to play an important role in marital dissolution in Nepal. In such a setting, those who choose their own spouse may place lower value on the family and greater value on personal fulfillment. These same values may influence their behavior if they become unsatisfied in the marriage. Moreover, those whose parents were not involved in choosing their spouse may lack family support in maintaining their marriage. In some cases, the parents might even encourage their children to divorce so that they can marry a spouse of the parents’ choosing. One local 27-year-old Nepali woman spoke of this possibility: “If [parents] encourage their son to marry another woman, then he will absolutely do it because he can’t go against the will of his parents. So there’ll be the possibility of getting divorced.”2 For these reasons, choosing one’s own spouse without family input may increase the risk of experiencing a marital dissolution.
On the other hand, choosing one’s own spouse may be more protective of marital stability than having no choice. Although spouses of arranged marriages may be well matched based on ethnicity and social status, they may not be well matched on an emotional level. In Islamic Southeast Asia, divorce rates fell drastically after people began choosing their own spouse, partly because people no longer needed to end their first marriage in order to marry the person they wanted to marry (Jones 1994). Likewise, in Nepal, respondents talk of young people being pressured to marry a partner that they did not want to marry. One 35-year-old Nepali man stated that this kind of marriage “can be broken easily.” If these mechanisms are at play, a person who chooses their own spouse may face lower risk of marital dissolution than a person whose parents arranged their marriage.
Couples may thrive most when their marriages are initiated with a combination of parental approval and their own choice. In these cases, they reap the benefits not only of family support in promoting the success of their marriage but also of emotional compatibility with their spouse. Hence, marriages in which both the spouse and their parents had input may face the lowest risk of marital dissolution.
In other settings, marrying at older ages tends to reduce the odds of marital dissolution. This outcome is attributed to the maturity and greater preparation for long-term commitment that comes with age (Becker et al. 1977; Chang and Jones 1990; Morgan and Rindfuss 1985). Moreover, people who marry later may have had more time to get to know their partner prior to marrying.
In Nepal, age at marriage is intertwined with marital arrangement (Ghimire et al. 2006; Jones 1994). Arranged marriages tend to occur at young ages (Bennett 1983). Arranged marriages that occur at later ages may be associated with greater difficulty in finding a suitable spouse and may be reflective of having less-desirable qualities for marriage. For the same reasons that it was difficult to find a suitable match, a person marrying at a relatively older age may face difficulty maintaining their marriage. Marriages in which the spouses were involved in selecting each other, on the other hand, may benefit from delay to older ages. In these cases, the spouses may have had more time to get to know each other and ensure emotional compatibility, thereby lowering their risk of marital dissolution.
Marital duration is also likely to play an important role in couples’ rate of marital dissolution. In general, couples who endure the early years of marriage are more likely to continue to stay together, and couples who decide to split are more likely to do so early on (Becker et al. 1977; Morgan and Rindfuss 1985). In Nepal, too, couples married for longer durations likely face lower risk of marital dissolution.
Education has been found to have a strong influence on marriage rates in Nepal (Yabiku 2005), and is also likely to affect rates of marital dissolution. In the United States, wives’ greater education can increase the likelihood of marital dissolution via greater ability to find employment that provides them with their own income (Ono 1998; South and Spitze 1986). However, the work that women perform in rural Nepal often involves farm labor and does not require formal education. Yet, wives’ education may affect marital stability via its effect on egalitarianism in marriages (Greenstein and Davis 2006): in this setting, as wives become more educated, the gap between spouses’ education tends to shrink (Stash and Hannum 2001). Increased educational homogeny has been found to enhance marital quality and increase barriers to marital dissolution in the United States (Tzeng 1992). This mechanism may be operating in Nepal as well, as a 27-year-old Nepali woman pointed out, “Sometimes [the] wife is not educated and the husband is educated, so they don’t understand each other and quarrel begins in their family and (finally) leads to a divorce.” Hence, wives’ increased education may reduce couples’ risk of marital dissolution.
As women gain opportunities to earn an income, divorce rates are likely to increase (Goode 1993; Ruggles 1997). In this setting, much of the work that women perform is wage labor and involves women in households with little or no land working on the farm of landowning households for compensation, which is then typically contributed to their household (Cameron 1998; Stash 1999). Because this work is not steady and because the earnings are not necessarily saved for personal benefit, wage work may not reduce wives’ barriers to marital dissolution.
Although few women have experience in steady, salaried employment in rural Nepal, those who do work in salaried jobs may be at greater risk of marital instability. The minority of women who hold salaried jobs may be selective of women who value self-sufficiency and independence. Moreover, as a result of their steady income, these women may perceive that they are capable of supporting themselves financially. Furthermore, salaried work provides greater opportunity to interact with people outside a woman’s immediate community, increasing the likelihood of meeting an alternative romantic partner (South and Lloyd 1995). Thus, wives’ experience in salaried work may reduce barriers to, and raise the risk of, marital dissolution.
In Nepal, children have an important value for their parents, both in their ability to economically contribute to the household and in their role in old-age security and Hindu religious rituals (Bose and South 2003; Niraula 1995). Like in the United States (Thornton and Young-DeMarco 2001), attitudes in Nepal are more disapproving of marital dissolution when couples have children. A 36-year-old Nepali woman conveyed this feeling, saying that, “. . . [I]f they [couples] already have children, then there are problems [in the case of divorce]. The situation of the children can be very bad.” The addition of each child can add value to the marriage and concern for the negative consequences of divorce, thus increasing barriers to marital dissolution.
Change Over Time
Changes to the social context, which have occurred rapidly over recent decades in this Nepalese setting (Axinn and Yabiku 2001), can affect family behaviors (Goode 1993). In Nepal, marital arrangement by parents and relatives has become less common, and spouses who exercise choice in their marriage are likely to face less pushback from their relatives more recently. These changes may result in happier marriages and lower odds of marital dissolution over time for spouses who had full choice or, to a lesser extent, shared choice in their marriage partners.
Those marrying at older ages may also face a lower risk of marital dissolution in more recent decades. Given that arranged marriage has been declining, older age at marriage may become more attributable to greater value placed on getting to know one’s partner before marriage. Similarly, because education has become more available, delaying marrying to older ages is likely increasingly associated with greater education at the time of marriage than in the past. These mechanisms may improve marital quality (South and Spitze 1986; Tzeng 1992) and increase barriers to marital dissolution over time.
Moreover, Nepalese couples may be more likely to dissolve their marriages at shorter durations in more recent years. In Southeast Asia, there is evidence that Muslims, for whom divorce is more common than non-Muslims, tend to dissolve their marriages at earlier marital durations than non-Muslims (Dommaraju and Jones 2011). Similarly, in Nepal, as new behaviors reflecting independent choice become more common and less stigmatized, couples may feel less pressure to invest time in maintaining unhappy marriages.
Additionally, education may become more connected to employment opportunities over time. If true, wives with greater education might increasingly perceive that they can support themselves, thereby increasing the odds of marital dissolution. However, the expansion of education in recent years may be leading wage work to become less lucrative for women. Thus, couples in which wives work for wages may face declining risk of marital dissolution over time. On the other hand, salaried work is uncommon in Chitwan, even in recent years, and women who enter these jobs are likely to continue to be selective of independent-minded women. Therefore, the influence of salaried employment may be stable over time.
Finally, as Nepalese fertility rates have declined and families have become smaller (Yabiku 2005), each child may hold even greater value to the parent. Thus, the suppressing influence that each additional child has on their parents’ marital dissolution may be more pronounced in more recent years.
Data and Sample
This investigation employs data from the Chitwan Valley Family Study (CVFS). The study began in 1996, when respondents were drawn through clustered sampling: 151 neighborhoods in Chitwan were randomly sampled, and each member of those neighborhoods between the ages of 15 and 59 (and their spouses) was interviewed. I use data from interviews conducted in 2008 with a sample that includes respondents from the 1996 sample and their spouses, all household members ages 12 to 34 and their spouses (if married) or parents (if unmarried), and all other neighborhood members ages 35 to 59. These interviews included a structured portion as well as a less-structured portion that used life history calendars to collect information on important events throughout respondents’ lives, such as school attendance, work, marriage, and marital dissolution (Axinn et al. 1999).
The analytic sample is drawn from all ever-married female respondents. I employ data from only women—not their husbands—because the retrospective nature of the data does not allow information from women to be linked to information from their ex-husbands. Although data come from interviews with women (wives), the unit of analysis can be conceptualized as “couples”: couples experience marital events (including dissolution) together. A benefit of these retrospective data, spanning from the beginning of the marriages, is that they eliminate concerns about left-censoring. The analytic sample includes a total of 3,413 couples; 17 couples were excluded because of missing values on at least one of the independent or control measures.
Following previous research on divorce (e.g., Bose and South 2003; Hirschman and Teerawichitchainan 2003; Morgan et al. 1988; Waite and Lillard 1991), the analysis focuses on dissolution of (wives’) first marriages. In Nepal, nearly everyone experiences first marriage (Yabiku 2005), but remarriage is rare.3 Moreover, remarriages may be selective of people who are prone to a greater likelihood of dissolution (Becker et al. 1977; Cherlin 1978; Martin and Bumpass 1989).
I operationalize the concept of marital dissolution by combining the events of separation and divorce. This is a common approach (Hirschman and Teerawichitchainan 2003; Morgan et al. 1988; Schoen 1992; South 2001) because there often is a temporal lag in the time from separation to divorce. This strategy allows the pinpointing of the time at which the marriage was first disrupted, which is especially important in this setting, where separation often occurs without a divorce to follow (Dommaraju and Jones 2011). On the other hand, separation is not a prerequisite for divorce in this setting, and many marital dissolutions are the result of immediate divorce. Of the couples in the analytic sample who experienced marital dissolution, only about one-third initially experienced separation.
This dependent measure is coded from the wives’ life history calendar and indicates the yearly hazard of marital dissolution. Coding this measure from wives’ reports is appropriate because husbands are less likely to report an event as a marital dissolution: husbands can be married to multiple women simultaneously. The measure of marital dissolution is coded as 0 in every year the couple is married and 1 in the first year in which the couple is separated (for at least a year) or divorced, after which the couple ceases to contribute to couple-years of exposure to the risk of marital dissolution.4
I examine the influences of both wives’ individual and couples’ shared experiences on couples’ odds of marital dissolution. All time-varying measures are precise to the year and are lagged by one year.
I first investigate couples’ marital experiences. Wife’s age at marriage is a time-invariant measure, coded in continuous years. In the upcoming logistic regression analyses, the measure is centered on its mean (17.46) to facilitate interpretation of interaction effects. Wife’s participation in spouse selection is time-invariant and coded from the following item in the structured survey: “People marry in different ways. Sometimes our parents or relatives decide whom we should marry, and sometimes we decide ourselves. In your case, who selected your (first) spouse? Your parents/relatives, yourself, or both?” I code three dummy variables to reflect that (1) the wife selected her husband herself (full choice), (2) both the wife and her parents/relatives selected her husband (shared choice), or (3) the wife’s parents/relatives selected her spouse (no choice). Marital duration, a time-varying measure, is coded in years to indicate the number of years since the couple was married. This measure is included as a linear term, which is appropriate based on the observed shape of the association between marital duration and marital dissolution in these data.
Next, I investigate wives’ experiences in activities outside the home (nonfamily experiences). I code wife’s education as a series of time-invariant dummy measures to indicate accumulated years of school enrollment at the time of marriage. These four dummy measures indicate whether the wife (1) never attended school, (2) attended school for 1 to 6 years, (3) attended school for 7 to 10 years, or (4) attended school for 11 or more years. A wife is considered to have attended a year of school if she was enrolled for at least part of the year. The analyses also account for a time-varying measure of whether the wife was enrolled in school in the couple-year of observation. This dummy measure is coded 1 in the couples-years in which the wife was enrolled for at least a partial year, and 0 otherwise. I also investigate two indicators of wife’s work experience: wage work and salaried work. These measures are time-varying and indicate whether the wife was working in the specified type of job, coded as 1 in the couple-years that the wife was working, and 0 otherwise.
I also investigate the influence of couples’ fertility. This time-varying measure is coded as the number of children that the couple had in each couple-year. A model (not shown here; see Jennings forthcoming) employing a series of dummy variables revealed that the odds of marital dissolution declined with the first, second, and third child, and then plateaued. I therefore top-code this measure at three or more children.
To investigate change over time, I use a measure to reflect the progression of historic time. This time-varying measure is coded into decades, indicating (1) before 1970, (2) 1970–1979, (3) 1980–1989, (4) 1990–1999, and (5) 2000–2008.
I control for other factors that could influence couples’ rate of dissolution. First, I control for wife’s childhood community context with a time-invariant measure that reflects the number of services (bus stop, health center, school, employer, or market) within a one-hour walk from the wife’s home until she was 12 years old. Second, I control for wife’s ethnicity. Because there are very few intercaste couples in the CVFS (Jennings 2014), wife’s ethnicity is generally indicative of both spouses’ ethnicity. Ethnicity is coded into four time-invariant dummy variables, indicating that they identify as Brahmin/Chhetri, Dalit, Hill indigenous, or Terai indigenous. Third, I control for the couples’ marital cohabitation, with a time-varying dummy variable, indicating that the couple lived together for at least six months of the couple-year of observation. Finally, I control for marital cohort, coded as five time-invariant dummy variables, indicating the decade in which the couple married. These measures indicate whether the couple was married (1) before 1970, (2) between 1970 and 1979, (3) between 1980 and 1989, (4) between 1990 and 1999, or (5) between 2000 and 2008.
The analyses use discrete-time, multilevel event history models with logistic regression to investigate the yearly risk of marital dissolution. The models adjust standard errors for clustering within neighborhoods to account for the clustered sampling design of the CVFS. Couples who are exposed to the risk of marital dissolution are defined as those in which wives are in their first marriage and no older than 50 years. I include those couples in the risk set until the wife is age 50 because marital dissolution becomes extremely rare after that age. Widowhood is treated as a competing risk, so that couples in which the husband dies cease to contribute couple-years to the hazard. The analyses use 58,696 couple-years of observation.
I discuss the results as odds ratios, which indicate the odds of marital dissolution in each yearly interval. Odds ratios can be easily transformed into the percentage change in the odds associated with each unit change in the respective independent variable by subtracting 1 from the odds ratio and multiplying by 100 (Thornton et al. 2007:352–353). Because so few marital dissolutions occur in each yearly interval, the yearly odds of marital dissolution are comparable with the rate of marital dissolution. For this reason, I discuss the rate of a marital dissolution as interchangeable with the odds of marital dissolution.
Six percent of the analytic sample experienced marital dissolution during the period of observation. Although this is a small proportion, it presents a large enough incidence of marital dissolution to allow for the use of logistic regression with event history analysis (King and Zeng 2001). Moreover, with such a small number of events, tests of significance are likely to be conservative.
Table 1 displays descriptive statistics for the analytic sample. I begin by noting a few results of the time-varying measures. First, although the average marital duration was almost 13 years, marriages that ended in dissolution lasted an average of 8.64 years (not shown in table). Next, the table shows that wives were enrolled in school for only 3 % of the couple-years of observation, but 25 % of the 3,413 wives were enrolled in school for at least a year (or partial year) during the period of observation (not shown in table). Similarly, wives worked for wages and salaried jobs during 32 % and 2 % of the couples-years of observation, respectively, while 43 % worked for wages, and 9 % worked in a salaried job during at least one year (latter percentages not shown in table). Finally, couples had an average of 1.94 children on the top-coded measure used in the analyses. The average number of children was 2.61 when this measure is not top-coded.
Table 2 displays odds ratio results from logistic regression analyses. Model 1 investigates the influences of marital characteristics. The model predicts that couples in which wives were older when they married experienced lower odds of marital dissolution. The coefficient of 0.91 indicates that with each additional year of wives’ age at marriage the rate of marital dissolution decreased by about 9 %. Because the odds ratio has a multiplicative influence on marital dissolution, this translates into about 25 % lower odds, or slower rate, of dissolution for couples who married when the wife was age 20 compared with couples who married when the wife was age 17.5
The model also predicts that couples in which wives shared spouse choice with their parents experienced lower odds of marital dissolution relative to couples in which wives had no choice in their spouse. Specifically, couples in which wives shared spouse choice experienced about a 48 % slower rate of marital dissolution than couples in which wives had no choice. Couples in which the wife had full choice in her spouse did not experience a significantly different rate of marital dissolution compared with couples in which the wife had no choice. I also tested a model in which full choice was treated as the reference category (not shown). That model revealed that sharing choice with parents significantly reduced couples’ rate of dissolution, but having no choice did not significantly influence the rate of marital dissolution relative to having full choice. Finally, Model 1 predicts that couples experienced lower odds of marital dissolution as marital duration increased. Couples experienced about a 7 % slower rate of dissolution with each additional year of marriage.
Many of the control measures also exert a significant influence on couples’ rate of marital dissolution. The measure of wives’ childhood community context negatively influenced couples’ rate of marital dissolution. Proximity to a greater number of services in childhood might have allowed women greater exposure to ideas, behaviors, and opportunities that reduce the risk of marital dissolution. Relative to Brahmins and Chhetris, each other ethnic group exhibits an increased rate of marital dissolution, which is likely due to the different marital practices across the groups, with Brahmins and Chhetris placing more emphasis on the purity of wives and the indissolubility of marriage than other groups (Bennett 1983; Stash and Hannum 2001). Next, couples who married in the 1980s experienced a significantly faster rate of marital dissolution relative to couples that married before 1970.6 One possible explanation for this cohort effect is that legal changes that were made in the mid-1970s, granting women alimony and custody rights (Manzione 2001), may have had an important impact on the cohort marrying in the 1980s. I more closely investigate the possible influence of change over time in Table 3.
Model 2 of Table 2 investigates the interaction between age at marriage and spouse choice. The model predicts that wives’ full choice in their spouse did not significantly moderate the influence of wives’ age at marriage, relative to wives who had no choice. However, there is a significant moderating influence of wives’ shared choice. Multiplying the odds ratios of the interaction term for shared choice and the main effect of age at marriage (0.79 × 0.92) produces an odds ratio of 0.73, suggesting that the rate of marital dissolution slowed by about 28 % with each unit increase in wives’ age at marriage for couples in which the wives shared spouse choice with their parents, relative to couples in which wives had no choice.
Model 3 of Table 2 investigates the influence of wives’ education on couples’ odds of marital dissolution, net of marital characteristics. The model does not reveal a significant influence of accumulated years of schooling at the time of marriage or of current school enrollment.
Model 4 investigates the influence of wives’ work experience on couples’ odds of marital dissolution. Theories based on human capital and women’s autonomy predict that education and work may operate in similar ways and through similar mechanisms to influence marital dissolution (Ono 1998; Schoen et al. 2002). Hence, Model 4 excludes variables reflecting wives’ schooling. The model predicts that couples in which wives were working for wages experienced about a 1.61 times faster rate of marital dissolution than couples in which wives were not working for wages. Couples experienced a rate of marital dissolution that was about 2.23 times faster in couple-years in which wives were working in a salaried job.
Model 5 tests the independent influences of wives’ schooling and work. The influence of wives’ wage and salaried work experiences remains significant, indicating that work experience influenced marital dissolution independent of wives’ schooling. Measures of schooling remain nonsignificant. Because of the effect that education can have on delaying marriage (Goode 1993; Jones 1994; Yabiku 2005), I also tested models replicating Models 3 and 5 but without inclusion of the variable indicating age at marriage (not shown). In those models, having been enrolled in school for 11 or more years at the time of marriage was revealed to have a significantly negative influence (at p < .05) on the odds of marital dissolution relative to having never attended school. Moreover, in models excluding the control for proximity to services (including schools) in childhood (not shown), coefficients associated with 1 to 6 years and more than 11 years of school enrollment at the time of marriage were found to be significant and negative.
I conducted sensitivity analyses (not shown) to address potential concerns about reverse causality in the association between work and marital dissolution (Rogers 1999). In these models, measures of wives’ work were lagged by three years. Wives’ experience of working for wages or in a salaried job maintained strong, significant influences, thus providing support for the hypothesis that it is indeed the experiences of wives’ work that led couples to dissolve (rather than the anticipation of dissolution that led wives to seek work).
Model 6 investigates the influence of couples’ fertility, net of the marital characteristics and nonfamily experiences tested in previous models. This model predicts that couples experienced lower odds of dissolution with each additional child (up to three children). The odds ratio of 0.51 indicates about a 49 % slower rate of marital dissolution with each child. Because fertility is correlated with marital duration, the influence of marital duration is diluted in this model.
Sensitivity analyses were performed to further investigate the influence of children (not shown). First, I restricted the sample to couple-years in which couples had at least one child (3,091 couples had at least one child during the period of observation). The results for the measure of number of children were similar to those shown in Model 6, although slightly diluted. Then, because of the prevalence of son preference in the region, I tested a model accounting for number of sons instead of total number of children to investigate whether the number of sons has a stronger influence (Bose and South 2003). Number of sons exerted an influence on marital dissolution similar to the influence of total number of children. Finally, including both number of children and number of sons in the model revealed that number of children maintained a significant influence, while number of sons did not.
Table 3 shows results of tests of the moderating influence of historic time, or calendar decade. In Model 1, I replicate the final model from Table 2 but with the addition of this measure of decade. To avoid overparameterization, I exclude measures of marital cohort from these models that include the measure of decade.7 In Model 1, decade is not significantly associated with couples’ risk of marital dissolution. Model 2 adds interactions between decade and the fundamental predictors of marital dissolution that were revealed to have a significant influence in Table 2.8 This model does not account for the interaction between number of children and decade because this would dilute the observed influence of the interaction between marital duration and decade. Marital dissolutions are distributed across decades, with 23 dissolutions occurring before 1970, 32 occurring in the 1970s, 44 occurring in the 1980s, 68 occurring in the 1990s, and 52 occurring between 2000 and 2008.
Model 2 of Table 3 predicts that the measure of decade significantly moderates the influences of wives’ age at marriage, marital duration, and wives’ wage work experience on marital dissolution. To illustrate, I plotted predicted probabilities with unadjusted models for each separate interaction. Figure 2 displays the moderating influence that decade has on the association between wives’ age at marriage and couples’ marital dissolution, with marital duration held at its mean. The predicted probabilities of marital dissolution associated with low (1 standard deviation below the mean for the analytic sample; 13.74), moderate (the mean; 17.46), and high (1 standard deviation above the mean; 21.18) values of age at marriage are plotted against decade. The high value shows the steepest slope, indicating that marrying at an older age had an increasingly suppressing influence on marital dissolution over time. Marrying at the mean (moderate) age was associated with a less rapidly declining probability of marital dissolution over time, and marrying at a younger (low) age was associated with almost no change over time in the probability of marital dissolution. In fact, Fig. 2 illustrates that the couples in which wives married at older (high) ages faced the highest probability of marital dissolution in the past, but around 1980 (decade = 2) the probability of dissolution for those marrying at older ages had declined to the lowest of the three categories. Hence, the influence of delayed marriage on marital dissolution has changed drastically over time.
Figure 3 plots predicted probabilities for the interaction between marital duration and decade, revealing that low (1 standard deviation below the mean for the analytic sample; 3.45 years), moderate (the mean; 12.84 years), and high (1 standard deviation above the mean; 22.23 years) values of marital duration were all associated with declining probability of marital dissolution over time. Those at the high end of marital duration consistently faced the lowest probability of dissolution.
Finally, Fig. 4 illustrates predicted probabilities associated with the interaction between wage work and decade, with marital duration held at its mean. This figure indicates that working for wages was associated with declining probability of marital dissolution over time. However, wives who were not working for wages faced a more rapidly declining probability of marital dissolution over time than those who were working for wages.
Model 3 of Table 3 investigates the role of decade in moderating the influence of number of children on marital dissolution, without inclusion of an interaction term for marital duration and decade. In this model, the influence of number of children on marital dissolution is not significantly moderated by decade.
This investigation offers some of the first empirical evidence of the fundamental predictors of marital dissolution in South Asia, where patriarchal norms are prevalent and people’s lives tend to be centered on the family (Dommaraju and Jones 2011). This investigation has revealed that wives’ older age at marriage and shared (with their parents) spouse choice reduced couples’ odds of marital dissolution. Similarly, couples married for longer durations and couples who had more children faced reduced odds of dissolution. Couples in which wives were working, on the other hand, faced increased odds of marital dissolution. The second part of the present investigation focused on change over time, given that this rural Nepalese setting has experienced rapid social changes during the past few decades that have altered people’s lives and social environment. Findings suggest that the influences of many of the fundamental predictors of marital dissolution have been changing alongside these social changes.
The practice of parents and relatives arranging marriages for younger generations is prevalent not only in Nepal but also in other settings (Dommaraju 2016; Ghimire et al. 2006; Mathur et al. 2003). Among Muslims in Southeast Asia, a move toward more “love marriages” and fewer arranged marriages led to a decline in divorce rates (Jones 1994). It is not surprising that the present investigation revealed a slightly different picture, with wives’ shared spouse choice found to reduce couples’ rate of marital dissolution and no evidence that having full choice lowers the odds of dissolution relative to having had no choice. Other research among this population has revealed that young people’s participation in choosing their spouse can enhance marital quality (Allendorf and Ghimire 2013). The present results suggest that couples thrive most when there is a combination of family support and spouses’ independent choice in the match. Family pressure remains strong in this setting: couples who enter a marriage without parental support may be no better off than couples whose spouses are chosen entirely by their relatives. These influences of marital arrangement were not impacted by the passing of time.
I found that couples faced lower risk of marital dissolution when wives married at later ages—a finding not dissimilar from other work in the region (Bose and South 2003). Still, this result was unexpected, given that Nepalese women marrying at younger ages are more likely to have had a marriage at least partly arranged by their parents. In fact, age at marriage was moderated by marital arrangement: couples were at lower risk of marital dissolution as wives’ age at marriage increased if wives shared spouse choice with their parents. The spouses in these shared choice marriages may have had more time to get to know one another and ensure emotional compatibility before marrying. Hence, waiting until later to marry may enhance marital quality if both the wife’s family and the wife, herself, are in favor of the spousal match.
Age at marriage was also significantly moderated by the passing of time: delaying marriage was associated with decreased odds of marital dissolution over time. In the United States, some research has found that younger age at marriage is associated with increased risk of dissolution over time (Martin and Bumpass 1989; Raley and Bumpass 2003), although other research has not found evidence of change over time (Teachman 2002). In this setting, where opportunities to interact informally with potential spouses have increased, people marrying at older ages in recent years may have a greater opportunity to get to know their spouse prior to marriage, thereby increasing their marital quality and reducing their risk of dissolution.
As these marriages endured, the odds of dissolution declined, similar to other settings (Heaton et al. 2001; South 2001). As expected, this effect has become more pronounced in more recent decades. Couples may perceive less stigma in opting to dissolve their marriages than in the past and, therefore, may be less motivated to invest time in maintaining unsatisfactory marriages. Moreover, it is also possible that, as couples are becoming increasingly likely to have known each other at the time of marriage, disagreements and conflict—found to be predictive of marital dissolution in this setting (Jennings 2014)—may arise earlier on.
I found that a couple’s number of children suppressed marital dissolution as well. As expected, with each additional child (up to three children), couples’ risk of marital dissolution declined. This result cannot be explained by the absence of children or by infertility/infecundity, which is historically a common reason for marital dissolution in this region (Bose and South 2003; Stone 1978). Instead, each additional child appears to impose barriers to marital dissolution. (For a more detailed analysis of childbearing and marital dissolution among this sample, see Jennings (forthcoming).) Interactions with decade revealed no evidence that additional children contributed to increasing marital stability over time.
In other settings, wives’ higher educational attainment sometimes increases marital dissolution—a finding that has often been explained by the acquisition of human capital (Ono 1998; South and Spitze 1986). In this setting, given that the majority of female labor tends to involve unskilled farm work, it is not entirely surprising that there is limited evidence of an influence of wives’ accumulated schooling. Yet, sensitivity analyses (not shown) revealed evidence that wives’ schooling can suppress marital dissolution, when either wife’s age at marriage (which may be delayed if she spends more time in school) or wives’ childhood proximity to services (including schools) is not accounted for in the model. Although the fully adjusted models shown in Table 2 do not reveal significant effects of schooling, the results from these sensitivity analyses suggest that we should not rule out the possibility that wives’ increased schooling may improve marital quality via increased educational homogeny between spouses (Tzeng 1992).
One of the most unexpected results from this investigation is the strong influence that wives’ wage work experience had on couples’ risk of marital dissolution. Wives’ labor force participation is found to increase the rate of marital dissolution in other settings (Schoen et al. 2002; South 2001; South and Spitze 1986). However, wage work in rural Nepal is unsteady and likely seasonal—based on the harvest—and payment is typically contributed to the household, rather than saved for the wife’s own benefit (Cameron 1998; Stash 1999). The unexpectedly strong influence of wives’ wage work suggests that the experience of earning compensation on their own, regardless of employment stability or personal savings, may be sufficient to reduce barriers to marital dissolution.
On the other hand, it is not surprising that couples were found to face an increased rate of dissolution when wives were working in salaried jobs. Women working in this kind of job have greater opportunity to meet people outside their immediate communities, heightening the risk that they meet an alternative romantic partner (South and Lloyd 1995). Moreover, wives who hold steady jobs may feel particularly confident in their ability to support themselves outside marriage. In fact, the minority of wives working in salaried jobs are likely to be selective of wives who value independence.
I found evidence that the risk of marital dissolution has declined over time for wives working for wages, as expected, suggesting that the wages and unsteadiness associated with this kind of work have become less conducive to wives’ ability or perceived ability to be independent. As time has progressed, independent living may require more money and a more steady income than wage labor provides. The influence of salaried work on marital dissolution, however, was not significantly moderated by the passing of time.
A number of limitations in this investigation are important to mention. First, the relatively small proportion of married couples experiencing dissolution suggests that these results should be generalized with caution. Second, given the retrospective nature of the data, this investigation was unable to account for or explore effects of husbands’ characteristics on couples’ risk of marital dissolution. Third, this investigation was unable to directly account for many religious, legal, and familial influences that are unique to this Nepalese setting. Because of the extent of these sociocultural differences, using research from other settings to frame this study may have led to misidentification of the mechanisms linking the factors investigated here to marital dissolution.
There is great value in understanding the process of marital dissolution, even among populations in which marital dissolution is uncommon. Goode (1993:319) conveyed this sentiment, suggesting that analyses among such populations will not only reveal the contributors to marital breakdown but will also provide greater insight into marital stability. Furthermore, it is crucial to recognize that low incidence of marital dissolution among a population is not necessarily indicative of happier marriages (Jones 1994). Instead, economic, social, and legal obstacles may prevent marital dissolution from being a viable option for many individuals who are in unhappy marriages. This study offers insights into the factors that can either augment or reduce the barriers to marital dissolution, and how barriers have changed as the social context and organization of people’s lives have been changing.
I am grateful for support from the Population Studies Center at University of Michigan (Grant Nos. R24 HD041028 and T32 HD007339), from the Carolina Population Center at the University of North Carolina (Grant Nos. T32 HD007168 and R24 HD050924), from the National Science Foundation (Grant No. OISE 0729709), and from the Harvard Center for Population and Development Studies. I would like to thank the Institute for Social and Environmental Research in Chitwan, Nepal for collecting the data used here; William Axinn, Abigail Stewart, Jennifer Barber, Dirgha Ghimire, Rachael Pierotti, William Story, Lisa Pearce, Prem Bhandari, and Jessica Pearlman for helpful comments on earlier versions of this article; and Cathy Sun for assisting with data management. All errors and omissions remain the responsibility of the author.
Of the 921 divorces that were filed in the Chitwan District Court between 2000 and 2009, 99.6 % were filed by wives (up from 92.4 % of the 249 divorces filed in the 1990s).
Quotes presented in this article are from in-depth interviews conducted in the fall of 2010. Respondents included 30 women and men between the ages of 18 and 45. These respondents were residing in neighborhoods of varying distances from the local city, Narayanghat, and represent the different ethnic groups in the area.
Only 7 % of ever-married women ages 40 and older in the 2008 CVFS sample had been married more than once.
The measure of marital dissolution indicates marital breakdown, and separation due to temporary migration is not coded as a marital dissolution.
Because of possible concerns about linearity, I also tested models with a series of three dummy variables to indicate age at marriage, each representing approximately one-third of the sample: under age 17, ages 17–19, and ages 20 and older. These models revealed that the odds of marital dissolution decline with increasing age at marriage, relative to marrying before age 17. To test whether women who marry at the high end of the age range are skewing the results, I also estimated a model in which women who married at or after age 30 (a total of 21 women in the sample) were dropped from the analytic sample. Results were nearly identical to those shown in Model 1 of Table 2.
This does not exactly parallel the trends shown in Fig. 1. Those data reflect numbers from Chitwan court records, while these data are from self-reports of either separation or divorce among women who may not have been residing in Chitwan at the time of marital dissolution. Moreover, court records are more likely to be sensitive to relaxation of laws restricting divorce (Goode 1993), and thus it is not surprising that self-reports are more stable.
Interactions with a measure indicating marital cohort (not shown), rather than calendar decade, reveal similar but weaker results compared with those shown in Table 3.
Interactions between variables reflecting years of schooling and decade (not shown) revealed no significant associations, regardless of the inclusion of measures for age at marriage or number of services within an hour’s walk during childhood.