Marital Experiences and Depression in an Arranged Marriage Setting

Abstract

Understanding the consequences of marital experiences for individual mental health provides insight into how social relationships shape individual wellbeing. Using newly available, clinically validated diagnostic interviews with more than 10,000 respondents integrated with the longitudinal Chitwan Valley Family Study (CVFS), we assess the associations between marital experiences, intimate partner violence (IPV), and mental health and how they differ by gender in a setting of universal marriage—Nepal. Particularly novel, we integrate measures of arranged marriage, IPV, and marital quality into a single comprehensive analysis of the marital experiences shaping subsequent depression. This study reveals that becoming married can be positively associated with major depressive disorder (MDD) for women. IPV is a strong and independent risk factor for depression, but it only mediates a small portion of the consequences of marriage on depression. Among women, having no say at all in the selection of a spouse is also a strong and independent risk factor for depression, and IPV can only mediate a small portion of the consequences of arranged marriage on depression. We also investigate the associations between the positive (i.e., husband-wife emotional bond) and negative (i.e., spousal criticism and disagreement) dimensions of marital quality and depression. Frequent spousal disagreement significantly increases depression for women, but strong husband-wife emotional bond is not significantly associated with depression. Overall, the associations between marital experiences and mental health should be understood as contingent on both gender and the social contexts of marriage. Depending on these factors, specific marital experiences have the potential to increase transitions to depression, not just protect from depression.

Introduction

The study of associations between social relationships and individual mental health is as old as the field of Sociology itself (Durkheim 2005 [1897]). Decades of evidence now indicate that close, intimate, sexual relationships—especially marriage—powerfully shape mental health (Simon 2002; Simon and Barrett 2010; Williams 2003). Theories of social support suggest that relationships like marriage can help to reduce the risks of experiencing mental disorders in general, and depression specifically (House, Landis, and Umberson 1988). Yet, the consequences of marital experiences may depend on the existence of conflict within the relationship, the setting-specific context of marriage, and the gender of the individual. Research on mental disorders points to traumatic experiences as a key source of personal stress that triggers disorder, with traumatic experiences initiated in relationships like marriage—intimate partner violence (IPV)¹—being among the most severe (Fergusson, Horwood, and Ridder 2005; Trevillion et al. 2012). We integrate these important streams of research by explicitly linking them to investigate how IPV influences the consequences of marital experiences for individual mental health.

Of course, careful investigation of marriage reveals it is a highly gendered social relationship (Bernard 1982). The specific gendered differences in the consequences of marriage on mental health vary by social context, changing across time and place (Scott et al. 2010; Williams 2003). In wealthy Western settings, evidence indicates that both men and women benefit from marriage, but men often benefit significantly more than women (Budig and England 2001; Desai and Waite 1991; England 2017; Waite 1995; Waite and Gallagher 2000). These benefits extend to lower incidence of depression among married men and women than among the single, widowed, or divorced (Simon 2002; Williams 2003). Investigation of gendered aspects of marriage in other settings reveals that women derive very little benefit from arranged marriage, where women have little or no input into the decision to marry or choice of spouse (Allendorf 2017; Kolenda 1987). Although associations between marital experiences and mental health, and gender differences in these associations, are well-documented in western countries, little is known about these associations in an arranged marriage society with high gender segregation. We advance the understanding of the gender differences in the link between marital experiences and mental health focusing on an arranged marriage society—Nepal—in South Asia, where we conduct the first population-scale, longitudinal investigation of the gendered consequences of marital experiences for major depressive disorder (MDD).

We harness clinically validated measures of the diagnosis of depression, rather than mood screening and assessment scales (e.g., Kessler Scales, Center for Epidemiologic Studies Depression Scale), to investigate these associations. Most prior research on population-scale associations between marital experiences and depression involves survey questions that screen for non-specific distress² (e.g., depression and anxiety) and do not provide information on behavioral impairment. Consequently, they may greatly over-estimate presence of mood disorders and are not equivalent to clinically diagnosed major depressive disorder (MDD; or depression hereafter) (Murphy 2011).³ We use instead the World Mental Health Composite International Diagnostic Interview (CIDI) (referred to as the “diagnostic interview”), an instrument which mimics how a clinician assesses psychopathology and which is specifically designed to capture all of the behavioral impairments used for DSM-IV diagnoses of depression (Kessler and Üstün 2004). Use of this diagnostic measure eliminates much of the measurement error in prior studies linking marriage and depression, greatly improving our understanding of the strength and gender-based variation in the association between marriage and depression.

This advance is possible because of the recent integration of clinically validated diagnostic measures of depression into a long-term family panel study of marital experiences. The Chitwan Valley Family Study, launched in rural Nepal in 1995, has followed these families for more than two decades, tracking men and women over time and wherever they move as they marry, become widowed, or divorce. The setting is characterized by near universal marriage and rare divorce, reducing selection into and out of marriages based on mental health. This study includes measurement of each spouse’s level of participation in decisions about their marriage as well as individual reports of physical violence within marriage from both women and men. The reporting of IPV from both genders provides the opportunity to examine gendered responses, because IPV is often collected only among women, especially in developing countries. In addition, the measurements were collected during simultaneous interviews specifically designed to provide privacy to wives and husbands. The panel study added lifetime retrospective reports of episodes of depressive symptoms and impairments using the World Mental Health Composite International Diagnostic Interview in 2016–18. The study features a special version of the diagnostic interview that links retrospective symptom reports to full life histories of memory anchoring events (including marital events) using a Life History Calendar to enhance the accuracy of these reports (Axinn, Chardoul, et al. 2020). Together these special resources provide the means to estimate gender differences in the associations between marital experiences and mental health and how IPV influences the consequences of marital experiences for men and women in an arranged marriage setting. The results provide important new insights into the extent to which hypotheses constructed from evidence in western settings can be generalized to a vastly different social setting.

Theoretical Framework

Here we explain the key dimensions of sociological reasoning that motivate our approach to the construction of empirical models summarizing the gendered connections among marriage, IPV, and depression. We begin with a discussion of two distinct but coexisting characteristics of marriage—social support and conflict. Much is known about these potentially opposing dimensions of marital relationships in settings like the U.S. and Europe—we apply that information to the construction of new reasoning specific to a setting of arranged marriage. Next, we consider the gendered nature of marriage in general, and in an arranged marriage society specifically. Finally, we use a life course perspective to derive a modeling strategy and predictions for the interactions between gender and the sequence of marriage choice, marital experiences, and IPV.

Social Support and Conflict in Marriage

There are two distinct streams of literature on the consequences of marriage for individuals. One holds that marriage produces positive states of health and well-being. The other argues that marriage can be an institution of exploitation, resulting in marital conflict and traumatic experiences. These two seemingly incompatible streams of research are seldom reconciled and integrated. However, doing so reveals the complicated nature of marriage and its varying consequences for mental health.

Social Support

Reasoning regarding social support provides a strong theoretical foundation for understanding the benefits of marriage and the negative consequences of marriage dissolution for health and well-being. The history of this reasoning can be traced to Durkheim’s classic conclusion that the absence of social integration disconnects the individual from relationships to others, producing “anomie” and ultimately leading down a path to suicide (Durkheim 2005 [1897]). Following this line of argument, less socially integrated individuals are likely to be less healthy, psychologically and/or physically, and more likely to die (House, Landis, and Umberson 1988). Long-term intimate partnerships, particularly marriage, are among the most potent sources of social support. Thus, as a form of social integration, marriage is a particularly important form of social relationship with positive consequences for health and wellbeing.

Marriage can function as a protective barrier against stress. Marriage does not prevent external events from occurring, but it can help people manage the psychological distress that such events create (Pearlin and Johnson 1977). Decades of empirical evidence consistently shows that married people have lower psychological distress than the single, cohabiting, divorced, and widowed (Marcussen 2005; Marks and Lambert 1998; Simon 2002) and are less likely to develop mental disorders (Duncan, Wilkerson, and England 2006; Hope, Rodgers, and Power 1999; Kendler et al. 2016; Scott et al. 2010). Because of this, the dissolution of marriage (i.e., divorce or widowhood) may not only result in the loss of social support, but also create a substantial adverse event itself. Prior empirical studies indicate that both divorce and widowhood lead to a significant increase in psychological distress and mental disorders (Edwards et al. 2018; Lee et al. 2017; Simon 2002; Umberson, Wortman, and Kessler 1992).

Unfortunately, there is relatively little evidence about the extent to which these supportive features of marriage are universal or setting specific, including the extent to which parentally arranged marriage conveys as much social support as individual choice marriage. Therefore, it is a high priority to investigate the role of marital relationships in mental health in settings outside North America and Europe. Nepal is a good example of such a setting.

Marriage in Nepal is universal and frequently arranged by parents. Both anthropological and historical evidence from Nepal document centuries of universal marriage, young ages at marriage, marital arrangement, and near zero premarital cohabitation or divorce (Bennett 1983; Fricke 1986). This began to change in the 1970s. Women married in the 1970s had a mean age at marriage of 14, while women married in the 1990s had a mean age at marriage of 17 (Yabiku 2004). Men’s mean ages of marriage were older, but show a similar change toward delay from the 1970s to the 1990s (Yabiku 2004). Marriage continues to be delayed across recent cohorts (Allendorf et al. 2017, 2019; Allendorf and Thornton 2015). Though young adults from high caste Hindu and higher education background are more likely to delay marriages, most young men and women marry before age 30 (Brauner-Otto, Axinn, and Ghimire 2020; Jennings, Axinn, and Ghimire 2012; Yabiku 2004).⁴ Divorce has also become more common since the 1990s when virtually no Nepalese marriage ended in divorce (Jennings 2017). These two changes combined to change lived experiences from what was a small number of years single, many years married, and a small number of years widowed, to what is now many more years single and growing years lived divorced.

The practice of very young ages at marriage was supported by nearly universal parental choice of spouses, but that is also changing in more recent cohorts (Ghimire et al. 2006). Individuals are becoming increasingly involved in the choice of spouse. However, men are much more likely to be involved in decision-making about marriage than women in Nepal (Ghimire et al. 2006). “Love marriages,” where parents have essentially no say in their children’s marriage partner are also becoming more common, although they are still rare. Historically in most Nepalese ethnic groups, wives moved to live with the husbands’ family after marriage and this is still common, although less so than in the past (Bennett 1983; Pearlman et al. 2017).

In this arranged marriage setting, there are still many reasons to anticipate that entering marriage will enhance social support. The Hindu philosophy of parentally arranged marriage, on which much of this practice in South Asia is based, holds that true, enduring love is not born of mutual attraction, but rather of shared life experience including the daily struggles to care for a home and family (Khandelwal 2009; Pasupathi 2002). In fact, Hinduism identifies multiple different forms of love, affection, and mutual support that characterize marital relationships (Axinn, Ghimire and Smith-Greenaway 2017; Ramachandran 2010; Vātsyāyana 2009). Parental arrangement in this context is thought to improve the matchmaking between women and men, with the aim of yielding a more supportive long-term trajectory for the couple. The parental arrangement often also explicitly formalizes ties between the two parental families, creating family-level alliances (Fricke 1986), which may also add to the total social support each spouse receives.

On the other hand, it is also possible that with no individual choice in the marriage, one or both spouses may not hold a social commitment to care for the other. Rather, individuals could view the marriage as a social relationship entered by the parents, not themselves, yielding less within-couple mutual support. This possibility could remove the within-couple social support associated with marriage in an arranged marriage setting.

Conflict

The consequences of marital experiences may depend on the existence of conflict within the relationship. Marital relationships characterized by high conflict may harm mental health, rather than improve it (Horwitz, McLaughlin, and White 1998; Umberson et al. 1996; Williams 2003). This is most likely when the conflict produces traumatic experiences, including IPV. A voluminous literature documents the adverse consequences of a traumatic experience for mental health (e.g., Ellsberg 2008; Trevillion et al. 2012). Experiences of trauma increase the odds of many different mental disorders, especially depression (Fergusson et al. 2005; Hill et al. 2010; Trevillion et al. 2012). This robust association has been replicated in many settings, even the rural Asian settings we explore (Axinn et al. 2013).

Different from some traumatic experiences, which appear to build individual resilience to subsequence mental disorders (Bromet et al. 2018), IPV is among the traumatic experiences with the strongest adverse consequences for subsequent mental disorders (Breiding, Black, and Ryan 2008; Bromet et al. 2018; Campbell 2002; Ellsberg 2008; Trevillion et al. 2012). Thus, marriage may only protect people from depression if it does not include IPV; if marriage exposes people to the risk of IPV, it may increase depression.

Again, evidence on these topics comes primarily from western settings, such as North America and Europe, which do not practice arranged marriage. Evidence from arranged marriage settings indicates the highest levels and most severe forms of IPV may be more common within arranged marriage (Naved et al. 2006; Paudel 2007; Puri, Tamang, and Shah 2011). In settings like Nepal, in which arranged marriage is common and divorce is rare, IPV may generate particularly strong adverse consequences for mental health, because victims have virtually no way to escape.

Even though IPV is prevalent in arranged marriage settings, there is no accurate and reliable estimation of the magnitude. The measurement of IPV is often plagued by underreporting. An ethnographic study in Bangladesh found that 66% of physically abused women were silent about their experiences (Naved et al. 2006). Fear of jeopardizing family honor, victim’s own reputation, and children’s future are often the main reasons for keeping silent (Naved et al. 2006). In order to compensate for the unavoidable shortcoming of the measurement of IPV, we consider a broader evaluation of marital relationships, which encompasses both negative and positive dimensions.

Research on marital relationships in all settings, including Nepal, demonstrates that marital relationships are multidimensional, simultaneously including both positive and negative dimensions (Allendorf and Ghimire 2013; Amato et al. 2003). The positive dimensions involve high satisfaction for varied aspects of marital relationships, such as the level of love in marriage (Allendorf and Ghimire 2013). The negative dimensions include both the physical violence described above and verbal forms of interpersonal conflict such as disagreements or criticism (Allendorf and Ghimire 2013). Prior evidence consistently documents that marital quality is a crucial moderator in the link between marital experiences and mental disorders. Marriage of high quality usually improves mental health, while a marriage of low quality may produce negative consequences for mental well-being (Horwitz et al. 1998; Williams 2003).

Investigations of both negative and positive dimensions of marital relationships benefit from considering measures of broader forms of conflict and support. For instance, disagreements or criticisms may capture a wider range of conflict than physical violence alone; husband-wife emotional bonds may reflect the core support within couples, which is deeply rooted in mutual care and trust. Thus, if marriage exposes people to the risk of high levels of marital conflict in general, it may increase depression. However, if marriage involves high levels of emotional bonding, it may decrease depression.

Gendered Dimensions of Marital and Traumatic Experiences

Marriage is highly gendered, making the potential consequences much different for men than for women. Socialization theorists argue that women place more value than men on intimate relationships, so those relationships are more crucial for women’s self-identity, self-conception, and mental health across the life course (Chodorow 1999; Simon, Eder, and Evans 1992; Thorne 1993). Application of this kind of socialization framework to the study of gender differences in consequences of marital experiences emphasizes gendered differences in response to emotions (Pollak and Thoits 1989). For example, women may learn to express distress through internalizing emotional problems, such as symptoms of distress, depression, and anxiety, while men may not (Aneshensel, Rutter, and Lachenbruch 1991; Horwitz, White, and Howell-White 1996). Empirical evidence from the U.S. population is consistent with this reasoning (Simon 2002; Simon and Barrett 2010), but there is relatively little empirical research on these issues from settings that are vastly different from the U.S. (for a few exceptions see Liu, Li, and Feldman 2013; Strohschein and Ram 2017).

Social role theorists argue that marriage is more advantageous for men, due to the unrewarding and stressful nature of female social roles in marriage such as unpaid household labor and childcare burden (Gove 1972; Gove and Tudor 1973). Based on sex role theory, Bernard (1982) argued that in every marriage there are actually two marriages—his and hers—and that marriage is more beneficial for men than for women. Bernard further speculated that the shifts in marital and gender roles toward gender equality could lead to a “future of marriage” that provides equal advantages to women and men. The growing gender equality in families, education institutions, and the workplace in the past decades may reduce current cohorts of women’s dependence on intimate relationships for self-identity and financial security, compared to earlier cohorts of women (Cotter, England, and Hermsen 2013; DiPrete and Buchmann 2013; England et al. 2007; England and Li 2006). In contrast, other settings with substantial patriarchal and patrilineal histories, where fewer changes have occurred over time in gendered roles in marriage, education, and work, are likely to remain highly gendered in response to marital experiences.

Empirical evidence leaves little doubt of important gender differences in the response to marital experiences. In general, marriage improves health and wellbeing for both men and women (Simon 2002; Waite 1995; Waite and Gallagher 2000; Williams 2003), but the benefits are often significantly better for men than for women (Duncan et al. 2006; Kessler et al. 1993; Marks and Lambert 1998). Studies also document that both divorce and widowhood reduce health and wellbeing for both men and women (Dupre, Beck, and Meadows 2009; Sasson and Umberson 2014; Strohschein et al. 2005; Umberson et al. 1992). The negative consequences of marital disruption can be particularly severe for both men and women, including substance abuse, mood disorders, risk taking behaviors and elevated rates of death (Edwards et al. 2018; Hewitt and Turrell 2011; Lillard and Waite 1995; Simon 2002). Again, unfortunately, we know much more about these consequences in North America and Europe than we do in other settings.

As with most settings, marriages in Nepal are generally characterized by important gender role differences. Not only do women bear children, in Nepal women are primarily responsible for care of the children and household work (including farming) while married men work for pay or farm (often producing goods for market). Nepalese women rarely inherit property and often have no direct control of household wealth or financial assets (Acharya and Bennett 1981). There is important variability in gender roles across and within ethnic groups in Nepal, so there is some variation in women’s abilities to achieve substantial decision-making authority (Bennett 1983). Nevertheless, women’s roles in Nepal are generally characterized as substantially less egalitarian than their counterparts in North America and Europe (Acharya and Bennett 1981; March 1986).

IPV has long been documented as a gendered experience in western settings, which is rooted in structural social inequalities and embedded in power-laden intimate relationships (Sweet 2019). Gendered and sexual stereotypes, structural exclusions, and institutional discrimination together create the conditions for violence (Anderson 2005; Armstrong, Hamilton, and Sweeney 2006). Research has long demonstrated that all forms of abuse between intimate partners are commonly used by men against women (Breiding et al. 2015; Kimmel 2002). In addition, the consequences of IPV may be gender-asymmetry. This is because “IPV perpetrated by women and men is qualitatively different in terms of the motivations for the assault, the meanings associated with the assault, and the resources (bodily, economically, psychological) that men bring into conflicts with intimate partners” (Anderson 2002:853).

However, less is known about IPV in Nepal than in North America and Europe, but there are good reasons to expect the gender differences in experiences of IPV are even greater in settings like Nepal. Prior studies indicate that South Asia has the highest levels and most severe forms of IPV in the world, with women much more likely to have these traumatic experiences than men (Abramsky et al. 2011; Naved et al. 2006; Paudel 2007; Puri et al. 2011). Ethnographic evidence from the Northern region of South Asia (including North India, Bangladesh, Pakistan and Nepal) suggests that the gender inequality in marital relationships may support high levels of IPV within marriage (Krishnan 2005; Schuler et al. 1996). This region has practiced parentally arranged marriage for centuries, but as this system changes toward more involvement from the couple who are to be married, women remain much less likely to choose their own spouses than men (Ghimire et al. 2006)—a form of gender difference in entry into marriage is likely to produce gender differences in experiences of IPV. Besides the gendered exposure to IPV, among those who experienced IPV, the consequences for mental health may be greater among women than among men. The gendered socialization around the use and legitimacy of violence and cultural notions of men’s authority within families place wives in a disadvantaged position in violent confrontations with husbands (Anderson 2002; Anderson and Umberson 2001). Moreover, in terms of resources and social support used to cope with traumatic IPV experiences, Nepalese women are also disadvantaged relative to Nepalese men. For instance, population-scale research on kinship systems within South Asia suggests that the North Indian kinship systems have a high tendency for new wives to live with husbands’ families, often physically far away from their natal families (Dyson and Moore 1983). It is documented that IPV victims mainly seek help from informal social networks rather than from formal institutions in South Asia, usually from parents, siblings, and neighbors and seldom from in-laws (Naved et al. 2006). Living far away from natal families may impede the help-seeking behaviors of women. The prevalence of parentally arranged marriage and patrilocal residence may combine to expose women to an increased risk of conflict and IPV. Therefore, the risk of IPV experience within marriage may be particularly high and the adverse consequences may be particularly strong for women in some parts of South Asia including Nepal.

The Life Course of Marital Experiences and Consequences for Depression

To understand the influences of marital experiences on mental health outcomes and its intertwinement with IPV experiences in any specific setting, we need to consider the sequence of those important life events from a life course perspective, considering how they evolve over time (Elder, Johnson, and Crosnoe 2003; George 2013). As individuals move through life, courtship precedes marriage, entry into marriage precedes experiences of a marital relationship, and experiences of a marital relationship precede experiences of divorce or widowhood. This temporal order in marital experiences shapes our reasoning about the potential consequences of those experiences for depression. Applied to a specific setting—Nepal in our case—this means the social context of courtship is crucial to deriving a prediction for the consequences of marriage, and so forth through these linked experiences.

Hypothesis 1. Given the Nepal-specific context of marriage, we expect it is possible that marriage increases the risks of depression for women.

This is because spouses are often selected by parents, especially among women (Ghimire et al. 2006); IPV is prevalent within marriage (Puri et al. 2011); divorce is rarely an option (Jennings 2014, 2017); and women’s roles within marriage are less rewarding, more stressful and can be highly gender-segregated (Bennett 1983).

Hypothesis 2. We also expect that divorce and widowhood will increase the hazards of onset of depression for both men and women. Again, it is likely that the increased risk of depression following divorce or widowhood is higher for Nepalese women than men.

Both social and economic forces may play a role in producing this stronger negative outcome for women. Socially, the cultural history of widowhood in South Asia is particularly adverse, consistent with the ancient practice of wives committing suicide on their husbands’ funeral piers. Though nothing this extreme is now practiced in Nepal or elsewhere, the cultural legacy is one of significant isolation of widowed women. Divorce is rarely an option for an unhappy marriage because it is a socially stigmatized experience especially for women (Jennings 2017). Economically, because women rarely own major assets, work for pay, or control inheritance, and are disadvantaged in inheritance law and have been limited in their legal ability to file for divorce, it is likely their financial trajectories following these events are more adverse than for men (Acharya and Bennett 1981; Allendorf 2007; Gilbert 1992; March 1986). Thus, in this social and economic context, it is possible marital events such as divorce and widowhood have a stronger impact on the risks of depression for women than for men (see Axinn, Zhang, et al. 2020 for more details).

Hypothesis 3. We expect that among married people, when marriage is completely arranged by parents, they will have a higher risk of depression than when marriage involves some involvement from the individual who is getting married. Again, this might be more salient among women than men.

In Nepal, almost all young adults will get married, and most of them marry at young ages, most often before age 25. Even though arranged marriage is declining in Nepal in recent decades, the majority of marriages are still arranged by parents, especially for women (Allendorf 2013, 2017). However, the timing of marriage is an important confounder to consider, because it is associated with the level of marital arrangement by parents (Ghimire et al. 2006). This means that an individual’s level of involvement in selecting a spouse cannot be empirically separated from the timing of the transition from being single to being married. As a consequence, empirical models investigating this important life course transition must consider the possibility that effects of marriage timing and choice of spouse are not independent. Also, we recognize the possibility that parentally arranged marriage could enhance total social support, as reasoned above, but balanced with the evidence that indicates arranged marriage increases both the incidence and severity of IPV, we predict overall that it increases the risk of depression.

Hypothesis 4. We expect that the association between marriage and risk of depression is mediated by IPV. Again, this mediation might be more salient among women than men.

In all settings, transition into marriage necessarily precedes transitions to widowhood or divorce. In Nepal, it also precedes IPV. This is unlike other settings, in which dating and premarital cohabitation are a common part of the courtship process so individuals are exposed to the risk of IPV before they marry (Abramsky et al. 2011; Anderson 2010; Miller et al. 2011). There is no documentation of premarital cohabitation in Nepal. Although there is some evidence of romantic interaction, such as love letters, there is no evidence of dating behavior in which couples spend significant time by themselves but physically together before marriage (Ahearn 2001). This means that in the Nepalese social context transition to marriage happens first in the life course, before exposure to the risk of IPV. This fact eliminates the possibility that people are selected into marriage based on their prior IPV histories and supports the possibility that IPV may be a mediator between marriage and depression rather than a moderator.

Hypothesis 5. We expect that the association between arranged marriage and risk of depression is mediated by IPV, and this mediation may be more salient among women than men.

Transitions into arranged marriage also happen first in the life course before exposure to the risk of IPV. In addition, women usually have less autonomy in determining their marriage than men, and new wives often move to live with husbands and in-laws in Nepal.

Methods

Data

We use data from the Chitwan Valley Family Study (CVFS), launched in 1995 with a sample of 151 neighborhoods, fully representative of Western Chitwan Valley in Nepal. Chitwan Valley is in the south of Nepal, encompassing about one third of the land area and one half of the population of Chitwan district. To investigate marital experiences and depression, we linked data from multiple surveys of the CVFS.

Nepal-Specific Diagnostic Interview

To accurately document the associations between marital experiences and subsequent major depressive disorder (MDD) as defined by the DSM-IV (American Psychiatric Association 2000), we use the Nepal-specific version of the World Mental Health Survey Initiative’s Composite International Diagnostic Interview. This diagnostic interview uses dozens of measures of depression episode-specific symptoms, including mood, behavior, and functioning, to achieve a clinically accurate diagnosis of depression (Kessler et al. 2004; Kessler and Üstün 2004). Over a five-year period, a team of scientists spanning Anthropology, Psychology, Psychiatry, Sociology, and Survey Methodology created a Nepal-specific version of the diagnostic interview for a subset of the most prevalent psychiatric disorders in Nepal, including depression (Ghimire et al. 2013). This special version of the diagnostic interview was administered to the full CVFS sample aged 15 to 59 in 2016–18. Also important, this version of the diagnostic interview links retrospective symptom reports to full life histories of memorable experiences using a Life History Calendar to enhance the accuracy of these reports (Axinn, Chardoul, et al. 2020). This carefully constructed tool provides the means to differentiate between episodes of depression that occur before and after specific marital events. Following the standard practice for this diagnostic interview, professional interviewers were rigorously trained in administration of the diagnostic interview using computer-assisted personal interviewing. They then went to respondents’ homes, obtained privacy and consent, and administered the diagnostic interview. The modules included were depression, mania, panic disorder, generalized anxiety disorder, post-traumatic stress disorder, intermittent explosive disorder, and alcohol use disorders. Clinical validation of the Nepal diagnostic interview against the gold-standard of the clinician-administered Structured Clinical Interview for DSM-IV (SCID-IV) demonstrates high concordance, comparable to validation studies of the U.S. and European diagnostic interview instruments (Axinn, Chardoul, et al. 2020).

Household Registry Data

Prospective monthly household registry data with monthly event accuracy forms the backbone of the CVFS. This ongoing registry has been collecting measures of births, deaths, marriages, divorces, and pregnancies for all individuals in 151 neighborhoods since February 1997 (245 months of data are reported here). This household registry data includes all individuals and households in the 151 neighborhoods, following residents of original baseline households no matter where they move and adding residents of new households moved into the neighborhoods, and residents of households that were in the baseline but subsequently moved out of Chitwan. This unique household registry system has a high retention rate of original respondents (over 90%; Axinn, Ghimire, and Williams 2013). We constructed the marital history of each respondent (aged 15 to age 59 in 2016–18) from the monthly marital status records in the registry.

Individual Life History Calendar Data

To provide complete marital histories from respondents who joined the CVFS sample as adults, we added measures from individual life histories. The CVFS collected three waves (1996, 2008, and 2016) of individual Life History Calendar (LHC) data, which were conducted using a specially designed LHC (Axinn, Pearce, and Ghimire 1999). These LHC interviews collected a complete retrospective demographic history of individuals (e.g., marital and childbearing events), spanning each individual’s entire life. These LHC measures provide the information needed to supplement the registry data and construct a complete marital history for all CVFS members interviewed in 2016–18.

Individual Interview Survey in 2008

The 2008 CVFS individual interview survey followed standard CVFS practices, interviewing husbands and wives in private, including separate simultaneous interviews, to maximize privacy. This interview included measures of each spouse’s experiences of being hit by the other spouse. This was done in the context of measures of the full history of their marriage, including level of participation in decisions about their marriage and multiple items measuring marital relationship quality. A total of 6,084 individuals were interviewed in both this 2008 survey and the Nepal-specific diagnostic interview, and among them 4,121 individuals were ever married in 2008.

Household Census and Relationship Grid Data in 2008

To acquire information on living arrangements within households for CVFS respondents, we draw on relationship information from the 2008 Household Census and Relationship Grid Data. This relationship grid measured each household member’s relationships with every other household members (parent, child, sibling, spouse, or other). This relationship grid data was collected for a complete census which listed all household members who ate and slept in the household for more than half the time during the previous six months.

Agriculture and Consumption Survey in 2006

To measure household economic conditions, we use the 2006 Agriculture and Consumption Survey. This survey was designed to gather information about farming, livestock, household assets, non-farming income, and consumption. This data includes information from 2,361 households and achieved a response rate of 97.8%.

Analytic Samples

We use two complementary types of analytic designs: a retrospective design with multiple marriage cohorts and a prospective design focusing on the cohort currently married in 2008. The retrospective design maximizes both persons and person-years of exposure to both marital events and depression, which provides more power on the test of IPV as a potential mediator of the associations between marriage and depression. The prospective design provides an important opportunity to focus specifically on arranged marriage, IPV, and a broader conceptualization of marital relationships.

The retrospective analytic sample was created by linking the Nepal-specific diagnostic interview with the household registry data, and three waves of LHC data (1996, 2008, and 2016). The final sample size of the retrospective sample is 10,548 persons (232,900 person-years).

The prospective analytic sample was created by linking the 2008 individual interview data for those who were currently married in 2008 to the Nepal-specific diagnostic interview, the household registry data, the household census and relationship grid data, the household agriculture and consumption data, and the three waves of LHC data (1996, 2008, and 2016). Starting from 2008, we tracked marital events of those currently married in 2008 to predict their first onset of depression after those events. Because higher-order marriages are rare in Nepal, we focused on those who were in first marriages in 2008. This prospective cohort also allows us to investigate a different, complimentary measure of IPV (explained below), and some dimensions of marital quality in the context of measures of marital arrangement. The final sample size of the prospective sample is 3,912 persons (115,772 person-years). We present the measures for the retrospective analysis first, and then turn to the measures for the prospective analyses.

Measures for Retrospective Analyses

Depression

We measured depression by the lifetime diagnosis of DSM-IV major depressive disorder (MDD) and the age of onset of the first episode of depression. A clinical validation study of these measures demonstrates high accuracy of these measures of depression (Axinn, Chardoul, et al. 2020). Shown in Table 1, 21.75% of women and 6.80% of men were diagnosed with lifetime depression in the retrospective analytic sample.

Table 1.

Sample description of retrospective analytic sample in 2016–18 (N=10548).¹

	Women (N=5710)		Men (N=4838)
	Mean or Percent	SE	Mean or Percent	SE
Lifetime major depressive disorder
Yes	21.75	0.55	6.80	0.36
No	78.25	0.55	93.20	0.36
Marital experiences
Married	76.65	0.60	66.34	0.73
Never married	19.92	0.57	32.40	0.72
Widowed	2.60	0.23	0.35	0.09
Divorced	0.83	0.13	0.90	0.15
IPV experiences
Yes	5.64	0.31	0.37	0.09
No	94.36	0.31	99.63	0.09
Duration of marriage (exclude never married)2	15.78	0.17	15.41	0.19
Ever had first birth
Yes	70.18	0.61	61.45	0.70
No	29.82	0.61	38.55	0.70
Lifetime any other mental disorders3
Yes	2.77	0.22	6.20	0.35
No	97.24	0.22	93.80	0.35
Age	32.04	15.17	33.27	17.31
Birth cohort
1992–2001	37.72	0.64	35.59	0.69
1982–1991	29.84	0.61	25.94	0.63
1972–1981	17.83	0.51	21.21	0.59
1957–1971	14.61	0.47	17.26	0.54
Ethnicity
Brahmin/Chhetri	43.75	0.66	43.51	0.71
Hill Janajati	19.79	0.53	19.90	0.57
Dalit	11.77	0.43	12.82	0.48
Newar	6.09	0.32	5.89	0.34
Terai Janajati	18.60	0.51	17.88	0.55
Education4
No education	19.60	0.53	5.42	0.33
Some education	44.20	0.53	53.80	0.72
SLC	13.22	0.45	17.11	0.54
IA or BA or above	22.98	0.56	23.67	0.61

Note:

^1.We reported the descriptive statistics at the time of the Nepal-specific Diagnostic Interview (2016–18).

²The duration of marriage is calculated based on the sample without the never married respondents (N=7992). We include this measure in supplementary analyses (see Appendix Table S2).

^3.Any other mental disorder before the first marriage includes alcohol abuse disorder (ALA), alcohol dependence disorder (ALD), generalized anxiety disorder (GAD), post-traumatic stress disorder (PTSD), panic disorder (PDS), intermittent explosive disorder (IED), and broad bipolar (BIPOLAR).

^4.SLC= Schooling Leaving Certificate. IA=Intermediate Arts. BA=Bachelor of Arts. The SLC is awarded to those scoring highly enough on a nationally standardized exam offered after the successful completion of 10th grade. IA is an academic diploma awarded by a high school or junior college after the completion of 12th grade or equivalent.

Marital Experiences

We used time-varying marital status changes from age 10 onward to the onset of the first depression for those diagnosed with lifetime depression (or until the age when respondents received the diagnostic interview for those not diagnosed with lifetime depression). Annual marital status was measured by a set of dichotomous variables: currently married, never married (reference group), widowed, and divorced. In 2016–18, 76.65% of women were in marriage, 19.92% were never married, 2.60% widowed, and 0.83% divorced; for men, 66.34% were in marriage, 32.40% never married, 0.35% widowed, and 0.90% divorced. Note that the young age of the sample (10–59) limits exposure to widowhood.

Intimate Partner Violence (IPV)

The measure of IPV used in the retrospective analyses is time-varying, derived through two questions about potentially traumatic experiences in the diagnostic interview—(1) “Were you ever badly beaten up by a spouse or romantic partner?”; and (2) “When was the first time it happened?”. These diagnostic interview measures have been used in more than 30 different countries to document IPV (Bromet et al. 2018) and were carefully translated to the Nepal setting (Ghimire et al. 2013)—nevertheless, prevalence using this measure is low.⁵ As shown in Table 1, 5.64% of women and 0.37% of men ever had IPV in the retrospective analytic sample. To provide a more thorough investigation of the association between IPV and depression possible, we re-examine these associations in the prospective design (below) with a complimentary measure of IPV.

Covariates

Had first birth in the previous year.

In Nepal, couples often have their first birth shortly after their first marriage. The positive association between pregnancy and depression among women is well-documented in previous literature (Beck 2001; O’hara and Swain 1996). To parse out the independent association between marriage and depression, we constructed a time-varying covariate to indicate whether respondents had a first birth in the previous year. Similar to construction of time-varying marital experiences, we used the household registry data as the master data and used the LHC to complete any missing values in person-years among respondents. To demonstrate the distribution of this measure, we show whether respondents ever had their first birth before the diagnostic interview (2016–18) in Table 1. Approximately 70.18% of women and 61.45% of men had their first birth before 2016–18.

Lifetime onset of any other mental disorders.

To exclude the possibility of comorbidity of other mental disorders, we created a time-varying binary measure to capture the onset of any other mental disorder before the first marriage, which includes alcohol abuse disorder, alcohol dependence, generalized anxiety disorder, post-traumatic stress disorder, panic disorder, intermittent explosive disorder, and broad bipolar disorder. In 2016–18, 2.77% of women and 6.20% of men had any other mental disorder than depression before their first marriage.

Age.

Age is associated with both marital experiences and depression, so we included a time-varying continuous measure (in years) for the respondent’s age between 10 and 59. Shown in Table 1, the mean age of respondents in 2016–18 is 32.04 for women and 33.27 for men.

Birth cohort.

Because Nepalese society has experienced great transformation in marriage and related factors (Ghimire et al. 2006), respondents of different birth cohorts are likely to have different marriage patterns and IPV experiences. Thus, we also controlled for respondent’s birth cohort, coded in four categories (1992–2001, 1982–1991, 1972–1981, and 1957–1971 [reference group]), with the youngest cohort accounting for the highest percentage for both women (37.72%) and men (35.59%).

Ethnicity.

Although Nepalese society is ethnically complex, and the study population is diverse, it is also a sub-population of Nepal with more limited variations than the entire country. We used a standard categorization of ethnic groups into five categories: Brahmin/Chhetri (high caste Hindus), Dalit (low caste Hindus), Hill Janajati (Hill Tibeto-Burmese), Newar, and Terai Janajati (indigenous to Chitwan) to capture this diversity (Blaikie 1980; Fricke 1986). Brahmin/Chhetri, the most privileged ethnic group in Nepal, is omitted from models as the reference group. Shown in Table 1, 43.75% of women and 43.51% of men are Brahmin/Chhetri in the retrospective analytic sample.

Education.

We summarized variability in education with a four categorical measure—no education (reference group), some education (1–10 years), Schooling Leaving Certificate (SLC), and Intermediate Arts (IA) or Bachelor of Arts (BA) or above. The SLC is awarded to those scoring highly enough on a nationally standardized exam offered after the successful completion of 10th grade. IA is an academic diploma awarded by a high school or junior college after the completion of 12th grade or equivalent. Because of the time-invariant nature of this indicator, the value assigned at the time of the specific-Nepal diagnostic interview (2016–18) is carried through the entire hazard analysis. Shown in Table 1, 13.22% (22.98%) of women and 17.11% (23.67%) of men had a SLC (IA or BA or above) in 2016–18.

Measures for Prospective Analyses

Here we switch to describing measures for a cohort of people all currently married in 2008 to take special advantage of the CVFS measures of marital relationships. We used the same measures of depression and marital events as above (in the retrospective analytic sample). Shown in Table 2, in this prospective sample, 28.54% of women and 6.52% of men were diagnosed with lifetime depression. All of the respondents were married in 2008 and by 2016–18, for women, 96.51% of them were still married, 2.75% widowed, and 0.74% divorced; for men, 98.91% of them were still married, 0.57% widowed, and 0.51% divorced.

Table 2.

Sample description: 2008 survey cohort prospective analytic sample in 2016–18 (N=3912).¹

	Women (N=2088)		Men (N=1824)
	Mean or Percent	SE	Mean or Percent	SE
Lifetime major depressive disorder
Yes	28.54	0.99	6.52	0.58
No	71.46	0.99	93.48	0.58
Marital experiences
Married	96.51	0.44	98.91	0.27
Widowed	2.75	0.39	0.57	0.18
Divorced	0.74	0.21	0.51	0.18
Arranged marriage
Had no say in marriage	61.59	1.06	29.66	1.07
Had a say in marriage	38.41	1.06	70.34	1.07
IPV experiences2
Yes	14.61	0.77	2.96	0.40
No	85.39	0.77	97.04	0.40
Criticism from spouse
Never	47.89	1.09	43.42	1.16
Seldom	40.57	1.07	48.41	1.17
Sometimes/Frequently	11.54	0.70	8.17	0.64
Disagreement with spouse
Never	25.29	0.95	26.64	1.04
Seldom	55.46	1.09	61.07	1.14
Sometimes/Frequently	19.25	0.86	12.28	0.77
Husband-Wife emotional bond (maya)
Very much	37.12	1.06	36.57	1.13
Some, Little or not at all	62.88	1.06	63.43	1.13
Duration of marriage	23.58	0.23	20.71	0.24
Ever had first birth
Yes	97.66	0.33	98.57	0.28
No	2.35	0.33	1.43	0.28
Lifetime any other mental disorder3
Yes	1.77	0.29	2.80	0.39
No	98.23	0.29	97.20	0.39
Age	41.37	19.26	43.43	18.84
Birth cohort
1982–1991	35.92	1.05	23.52	0.99
1972–1981	34.20	1.04	40.95	1.15
1957–1971	29.89	1.00	35.53	1.12
Ethnicity
Brahmin/Chhetri	46.07	1.16	44.63	1.16
Hill Janajati	17.29	0.83	17.60	0.89
Dalit	11.35	0.69	12.17	0.77
Newar	6.27	0.53	6.36	0.57
Terai Janajati	19.01	0.86	19.24	0.92
Education4
No education	38.75	1.07	9.98	0.70
Some education	46.93	1.09	63.16	1.13
SLC	7.66	0.58	12.88	0.78
IA or BA or above	6.66	0.55	13.98	0.81
Household annual income (rupees)
25,000 or less	12.64	0.73	13.43	0.80
25,000 – 50,000	13.65	0.75	13.05	0.79
50,000 – 100,000	29.21	1.00	28.89	1.06
100,000 – 250,000	27.54	0.98	27.19	1.04
250,000 or more	16.96	0.82	17.43	0.89
Household asset index5	0.40	0.04	0.42	0.05
Household agriculture index6	0.57	0.05	0.57	0.05
Living in a three-generation household
Yes	42.39	1.08	44.19	1.16
No	57.61	1.08	55.81	1.16
Pray at a temple
Never	14.56	0.77	18.53	0.91
Once a month or less	64.08	1.05	63.05	1.13
More than once a month	21.36	0.90	18.42	0.91

Note:

^1.The IPV experiences, husband-wife emotional bond, criticism from spouse, disagreement with spouse, pray at a temple, and living in a three-generation household were reported in 2008. Household annual income, household asset index, and household agriculture index were reported in 2006. The other measures were reported at the time of the Nepal-specific Diagnostic Interview in 2016–18.

^2.IPV experiences for the 2008 survey cohort prospective analytic sample were measured prospectively, different from the retrospective analytic sample.

^3.Any other mental disorder before marriage includes alcohol abuse disorder (ALA), alcohol dependence (ALD), generalized anxiety disorder (GAD), posttraumatic stress disorder (PTSD), panic disorder (PDS), intermittent explosive disorder (IED), and broad bipolar (BIPOLAR).

^4.SLC= Schooling Leaving Certificate. IA=Intermediate Arts. BA=Bachelor of Arts. The SLC is awarded to those scoring highly enough on a nationally standardized exam offered after the successful completion of 10th grade. IA is an academic diploma awarded by a high school or junior college after the completion of 12th grade or equivalent.

^5.The household asset index is composed of three components: the total land area that the household owned, the ownership of the house plot, and the index of housing quality (the number of stories, wall materials, roof materials, and floor materials). The three components were first standardized (mean of 0 and standard deviation of 1) at the household level and then were summed together to construct an index of household assets.

^6.The household agriculture index is composed of three components: the total land area that the household farmed, the number of farm animals owned, and the number of poultry owned. The three components were first standardized (mean of 0 and standard deviation of 1) at the household level and then were summed together to construct an index of household agriculture.

^7.Because household economic status and living arrangement are quite homogenous within the same neighborhood and ethnicity background in Nepal, we imputed the missing values of household annual income (n=506), household asset index (n=506), household agriculture index (n=506), and living in a three-generation household (n=205) through linear, ordered logistic, and logistic regressions with neighborhood ID, household ID, education, gender, and ethnicity as predictors.

Arranged Marriage

The prospective, ever married only approach also allows us to investigate the level of individual involvement in the selection of a spouse. In Nepal, husbands and wives often experience different levels of involvement in arrangement of the same marriage (Ghimire et al. 2006). Two questions in the 2008 individual interview data provide these measures: (1) “In your case, who selected your spouse? Your parents/relatives, yourself, or both?”; and (2) “Although both of you may have decided, one of you may have had a little more influence than the other. Who had more influence in choosing your spouse? You or your parents/relatives?”. Following previous research, responses to these two items can be used to create a five-point scale of individual involvement in the choice of a spouse or a dichotomous indicator to capture marriages entirely chosen by the individual versus those that involved arrangement by parents (Ghimire et al. 2006). The dichotomous measure is coded with those who decided their marriage completely by their parents as 1 and those who decided their marriage completely or partially by themselves as 0. For robustness check, we examined all possible different coding methods of this measure and the detailed results are reported later. Because of the time-invariant nature of this indicator, the value assigned at the time of the 2008 interview is carried through the entire hazard analysis. Shown in Table 2, 61.59% of women and 29.66% of men had their parents/relatives completely decide their marriage.

The indicators of IPV in this prospective analytic sample are complimentary to those used in the retrospective analytic sample. This measure was derived from a question in the 2008 individual interview data asking every married respondent (in privacy), “Has your (current) (husband/wife) ever beaten you?”. If the respondent answered affirmatively, this response was categorized as having experienced domestic violence coded as “1” (otherwise coded as “0”) (Ghimire, Axinn, and Smith-Greenaway 2015). Because of the time-invariant nature of this IPV indicator, the value assigned at the time of the 2008 interview is carried through the entire hazard analysis. Demonstrated in Table 2, 14.61% of women, and 2.96% of men reported ever being beaten by their current spouse in 2008.

Marital Conflict

We used two measures—criticism from and disagreement with spouse—to capture marital conflicts. Criticism from spouse was derived from a question in the 2008 individual interview asking, “How often does your (husband/wife) criticize you?”. In the same interview, disagreement with spouse was derived from a question asking, “How often do you have disagreements with your (husband/wife)?”. Both responses were measured on a four-category ordinary scale: frequently, sometimes, seldom, and never (reference group). We simplified the measures by combining “frequently” and “sometimes” together, because each of these categories is rare. Shown in Table 2, 11.54% of women and 8.17% of men were criticized by their spouse sometimes or frequently. About 19.25% of women and 12.28% of men had disagreement with their spouse sometimes or frequently.

Husband-Wife Emotional Bond

We also controlled for a positive dimension of marital relationship. Following previously published research with the CVFS, we used an indicator—husband-wife emotional bond—as a proxy to capture the positive dimensions of marital quality. The indicator was derived from a structured survey question in the 2008 individual interview data—“How much do you love your (husband/wife)? Very much, some, a little, or not at all?”. This structured survey question was designed through a mixed method approach in an iterative process through a series of steps. More details are documented elsewhere (e.g., Axinn, Ghimire, and Smith-Greenaway 2017). Consistent with prior research, we coded the responses “very much” as 1, and “some”, “a little”, or “not at all” as 0. Because of the time-invariant nature of the indicator, the value assigned at the time of the 2008 interview is carried forward through the entire hazard analysis. 37.12% of women and 36.57% of men reported that they loved their spouse very much.

Covariates

We controlled for the covariates in the prospective analyses in the retrospective analyses. As for birth cohort, we excluded the 1992–2001 birth cohort in the prospective analyses, since most of them were not in the restricted age range in the 2008 individual interview.

Duration of the first marriage.

Because the associations between marital experiences and depression might be confounded by the timing of marriage, we controlled for the duration of the first marriage. Because the age of the first marriage, age, and duration of the first marriage are colinear, we operationalized the timing of marriage with duration since the first marriage. We also tested including the age of the first marriage as a covariate without controlling for the duration of the first marriage, and the main results are consistent. In 2016–18, the mean duration of the first marriage is 23.58 years of women and 20.71 years of men.

Household economic conditions.

Household economic conditions are also a crucial confounder for the association of marital behaviors and IPV with depression. Following previously published research with the CVFS, we constructed a household annual income measure, a household asset index, and a household agriculture index to capture household economic conditions (Ghimire, Zhang, and Williams 2019), derived from survey questions in the 2006 Agriculture and Consumption Survey. The annual household income in 2005 was constructed based on a series of questions asking respondents to narrow down their household income in a certain range gradually. This measure is a five-category ordinal indicator in rupees (in 2005, 1 dollar≈45 rupees): 25,000 or less (reference group); 25,000~50,000; 50,000~100,000; 100,000~250,000; 250,000 or more. The household asset index is composed of three components: the total land area that the household owned, the ownership of a house plot, and the index of housing quality.⁶ The household agriculture index is also composed of three components: the total land area that the household farmed (which is frequently different than land area owned), the number of farm animals owned, and the number of poultry owned. For the two indices, we first standardized (mean of 0 and standard deviation of 1) the three components at the household level and then summed them together to construct an index.

Household living arrangements.

The North Indian kinship systems have a long history of new wives living with in-laws (Dyson and Moore 1983). Living with in-laws and large extended families can be a stressful experience because daughters-in-law are usually at a disadvantaged position in the household power hierarchy (Bennett 1983). The intricacies of the interpersonal dynamics in extended families may also engender stress or conflict within the household, which may be harmful to each family member’s mental health. We created a dichotomous indicator to capture household living arrangements—living in a three-generation household—based on the relationships of household members from the 2008 Household Census and Relationship Grid data.⁷

Religion and religiosity.

Religion is a crucial aspect of social life in Nepal, which is associated with marital behaviors (Axinn and Yabiku 2001; Ghimire et al. 2006) and mental health outcomes (Scott et al. 2020). Religion is also closely intertwined with ethnicity in Nepal. The majority of Nepalese are Hindus (Brahmins/Chettris and Dalits), with the second most common religions being Buddhism (Hill Janajati) or a mix of Hinduism and Buddhism (Newar), and the least common other religions being local spirit worshipping groups (Terai Janajati). Beyond controlling for ethnicity in the prospective analysis, we also created a measure of religiosity based on data from the 2008 CVFS individual survey. Respondents were asked to report “How often do you pray at a Temple?”. The responses were coded into three categories: more than once a month, once a month or less, and never (reference group).

Analytic Strategy

We use hazard models to estimate the time-varying risk of the first onset of lifetime depression. The data are precise to the year, we use a discrete-time approach, and this approach is described in detail elsewhere (Allison 1982; Peterson 1991; Yamaguchi 1991). The unit of analysis is the person-year of exposure to the risk of depression onset. To estimate the discrete-time hazard model, we use logistic regression in the following form:

$$In\left(\frac{p}{1-p}\right)=\alpha +\sum {\beta }_k{X}_k$$

Where p is the yearly probability of the first onset of lifetime depression, $\frac{p}{1-p}$ is the odds of the first onset of lifetime depression, $a$is a constant intercept, and ${\beta }_k$ is the estimated coefficient of explanatory variables $X_k$.

Answering the questions posed in this study involves the examination of mediation and interaction effects in non-linear models. Due to the issue of scalar identification, the conventional techniques (i.e., compare coefficients across groups) are flawed for non-linear models (see Long and Mustillo 2018 for more details). Marginal effects are a useful method for quantifying effects of predictors because they are in the natural metric of the outcome measure and they avoid identification problems when comparing regression coefficients across non-linear models (e.g., logit or probit) (Long and Mustillo 2018; Mize, Doan, and Long 2019; Wooldridge 2010). Average marginal effects (AMEs) can be interpreted as the average of the changes in the predicated probability of outcome with respect to a unit change in the predictor over the whole population. Thus, we transformed the coefficients from discrete-time hazard models into AMEs.

The examination of interaction effects is realized by comparing the AMEs of predictors across two groups of people (e.g., gender). The investigation of mediation effects is achieved by comparing AMEs of predictors across models with and without the mediators (e.g., IPV). The decomposition of total effects in non-linear probability models is not as straightforward as in linear models (Winship and Mare 1984). In nested non-linear probability models, uncontrolled and controlled coefficients can differ not only because of confounding but also because of a rescaling of the model that arises whenever the mediator variable has an independent effect on the dependent variable. Several solutions have been proposed to deal with this issue (e.g., Y standardization, the use of average marginal effects, a decomposition method for binary response models developed by [Erikson et al. 2005], and a decomposition method developed by [Karlson, Holm, and Breen 2012; who refer to it as Karlson-Holm-Breen decomposition method]). We present results based on the method of average marginal effects, but as a robustness check, we also use the Karlson-Holm-Breen decomposition method to replicate the retrospective results (for details see Appendix Tables).

Furthermore, to test the inequality of AMEs across two non-linear models, we use the delta method to acquire the covariance matrix and then tested the inequality of AMEs based on the Wald statistic (see Mize, Doan, and Long 2019 for more details). The covariance matrix across two non-linear models can be obtained using the delta method (Agresti 2013:72–77), bootstrapping (Efron and Tibshirani 1993), or simulation (King, Tomz, and Wittenberg 2000). Because these methods produce similar results, we choose the delta method for its faster computation (Dowd, Greene, and Norton 2014). We performed all analyses in STATA 16.

Retrospective Results

In the retrospective analyses (see Table 3), we first examine the association between marriage and the first onset of lifetime depression in women and men, controlling for essential covariates. Then, we include IPV experiences in the hazard models to test whether IPV experiences are associated with the first onset of lifetime depression independent of marital experiences. In Table 4, we examine the mediation effects of IPV in the pathway between marriage entry and the first onset of lifetime depression in men and women. In Table 5, we examine the gender differences in the effects of marital experiences and IPV on the first onset of lifetime depression.

Table 3.

Estimates of marriage, intimate partner violence (IPV) for subsequent onset of major depressive disorder (MDD), Average Marginal Effects (%) from discrete-time hazard models (standard errors in parentheses, N=10548).

Model	Women		Men
	A1 AMEs (%)	B1 AMEs (%)	A2 AMEs (%)	B2 AMEs (%)
Marital experiences
Ref: Never married
Ever married	0.99***	0.96***	0.03	0.03
	(0.09)	(0.09)	(0.06)	(0.06)
IPV Experiences
Ref: No
Yes		1.10***		0.64
		(0.22)		(0.67)
Covariates
Had first birth in the previous year
Ref: No
Yes	0.47*	0.47**	0.09	0.10
	(0.18)	(0.18)	(0.11)	(0.12)
Lifetime any other mental disorder1
Ref: No
Yes	28.24***	27.56***	0.49*	0.51*
	(2.66)	(2.65)	(0.22)	(0.23)
Ethnicity
Ref: Brahmin/Chetri
Dalit	0.36**	0.30**	0.11	0.11
	0.12)	(0.11)	(0.06)	(0.06)
Hill Janajati	0.02	0.02	0.04	0.04
	(0.08)	(0.08)	(0.05)	(0.05)
Terai Janajati	−0.03	−0.03	−0.04	−0.04
	(0.09)	(0.09)	(0.04)	(0.05)
Newar	−0.08	−0.10	−0.05	−0.05
	(0.13)	(0.12)	(0.07)	(0.07)
Education2
Ref: No education
Some education	0.00	0.03	−0.05	−0.04
	(0.09)	(0.09)	(0.08)	(0.08)
SLC	−0.17	−0.12	−0.11	−0.11
	(0.13)	(0.13)	(0.09)	(0.09)
IA or BA or above	−0.38***	−0.33**	−0.20*	−0.20*
	(0.12)	(0.12)	(0.09)	(0.09)
Age	0.12***	0.11***	0.05***	0.05***
	(0.02)	(0.02)	(0.01)	(0.01)
Age squared	−0.00***	−0.00***	−0.00***	−0.00***
	(0.00)	(0.00)	(0.00)	(0.00)
Birth cohort
Ref: 1957–1971
1972–1981	0.54***	0.53***	0.80***	0.83***
	(0.16)	(0.15)	(0.14)	(0.15)
1982–1991	−0.05	−0.03	0.16***	0.16***
	(0.10)	(0.10)	(0.05)	(0.05)
1992–2001	0.05	0.06	0.01	0.01
	(0.08)	(0.08)	(0.03)	(0.03)
Persons	5710	5710	4838	4838
Log likelihood (ll)	−6419.66	−6399.17	−2171.82	−2170.86
AIC	12871.32	12832.34	4375.64	4375.75
BIC	13026.06	12996.75	4530.19	4539.95

Note:

^1.Any other mental disorder before the first marriage includes alcohol abuse disorder (ALA), alcohol dependence disorder (ALD), generalized anxiety disorder (GAD), post-traumatic stress disorder (PTSD), panic disorder (PDS), intermittent explosive disorder (IED), and broad bipolar (BIPOLAR).

^2.SLC= Schooling Leaving Certificate. IA=Intermediate Arts. BA=Bachelor of Arts. The SLC is awarded to those scoring highly enough on a nationally standardized exam offered after the successful completion of 10th grade. IA is an academic diploma awarded by a high school or junior college after the completion of 12th grade or equivalent.

^{3. *}p<0.05

^**p<0.01

^***p<0.001.

Table 4.

Average Marginal Effects (%) of marriage on major depressive disorder (MDD) before and after controlling for IPV, controlling for covariates (Standard errors in parentheses, N=10548).

	Women			Men
	A1 AMEs (%)	B1 AMEs (%)	A1-B1 AMEs (%)	A2 AMEs (%)	B2 AMEs (%)	A2-B2 AMEs (%)
Average marginal effects
Marital experiences
Ref: Never married
Ever married	0.993***	0.962***	0.031***	0.035	0.035	0.000
	(0.092)	(0.091)	(0.007)	(0.056)	(0.058)	(0.002)
Covariates	Yes	Yes	Yes	Yes	Yes	Yes

Note:

^1.SEs are estimated with the seemingly unrelated estimation approach (Mize, Doan, and Long 2019). An alternate approach, non-parametric bootstrapping, generates similar results.

^{2. *}p<0.05

^**p<0.01

^***p<0.001.

Table 5.

Gender differences in the estimates of marriage, intimate partner violence (IPV) for subsequent onset of major depressive disorder (MDD), Average Marginal Effects (%) from discrete-time hazard models (standard errors in parentheses, N=10548).

Model	Women	Men	Gender Difference
	A1 AMEs (%)	A2 AMEs (%)	A1-A2 AMEs (%)
Marital experiences1
Ref: Never married
Married	0.893***	0.015	0.878***
	(0.089)	(0.056)	(0.105)
Widowed	11.554***	4.642**	6.912*
	(1.493)	(1.637)	(2.215)
Divorced	2.872***	0.244	2.627**
	(0.534)	(0.180)	(0.563)
IPV experiences
Ref: No
Yes	0.983***	0.656	0.327
	(0.209)	(0.682)	(0.713)
Covariates	Yes	Yes	Yes

Note:

^1.We used a more detailed categorical measure for marital experiences in this table.

^2.SEs are estimated with the seemingly unrelated estimation approach (Mize, Doan, and Long 2019). An alternate approach, non-parametric bootstrapping, generates similar results.

^{3. *}p<0.05

^**p<0.01

^***p<0.001.

Marriage, IPV, and Major Depressive Disorder

Table 3 presents the results of hazard model estimates of the association between marriage, IPV experiences, and the first onset of lifetime major depressive disorder. Model A1 and Model A2 in Table 3 present the association between getting married and the first onset of lifetime major depressive disorder in women and men.

Compared with singlehood, marriage has a strong, positive, statistically significant relationship with depression among women. Marriage significantly increases the probability of depression by 0.99% for women; however, it only increases the probability of depression by 0.03% for men (albeit, not significant). This finding is consistent with evidence documented elsewhere (Axinn, Zhang, et al. 2020). Axinn, Zhang, and colleagues (2020) explained that the population of Nepal has somewhat different experiences of marriage than populations of North America, Western Europe, and the European diaspora. Throughout the European diaspora, in population level studies, transitions to marriage are typically associated with either lower probability of depression or no association with depression (Scott et al. 2010). In Nepal, however, marriage increases the probability of subsequent depression among women.

In Model B1 and Model B2, we added IPV experiences to Model A1 and Model A2, revealing that the observed association between marriage and depression decreases slightly in magnitude but remains highly significant for women. We also find that IPV experiences have strong, positive, statistically significant associations with depression in women. For instance, having IPV experiences increases the probability of depression among women by 1.10%. Note that the association among men is in a smaller magnitude and not statistically significant in this sample.

In order to compare the changes in associations between marriage and depression before and after controlling for IPV experiences, we present the results of mediation analysis in Table 4.⁸ For women, on average, the probability of depression increases by 0.99% after getting married. After controlling for IPV experiences, this average increase is reduced to 0.96%. Approximately 0.03% of risk is explained through IPV experiences, which only counts for 3.12% of the total risk. As for men, the mediation effects of IPV are really small and insignificant. These results imply that IPV is a significant mediator between marriage and depression among women, but it explains a very small proportion of the total association, indicating that there are other mechanisms explaining the strong positive association between marriage and depression in Nepal.

Gender, Marital Experiences, IPV, and Major Depressive Disorder

Table 5 displays the link between marital experiences and the first onset of lifetime major depressive disorder controlling for IPV experiences and essential covariates for women and men. Table 5 also shows the gender differences in the associations of marital experiences and IPV with depression.

For women, transitions into marriage, widowhood, and divorce all significantly increase the probability of depression by 0.89%, 11.55%, and 2.87%, with widowhood being the strongest predictor.⁹ However, marital experiences are associated with depression in a much smaller magnitude for men. Only transitions into widowhood significantly increase the probability of depression for men, by 4.64%.

The statistical test of AMEs reveals significant gender differences in the link between marital experiences and depression. Transitions into marriage increase the probability of depression for women about 0.88% more than that for men, with the increase in men close to zero. The increase in the probability of depression resulting from widowhood for women is more than twice as high as the increase for men. Transitions into divorce increase the probability of depression for women about 2.63% more than that for men, with the increase in men about 0.24%.

We also tested the gender differences in the link between IPV experiences and depression but find no statistically significant gender differences. It is likely that experiencing IPV within marriage has similar consequences for the risk of depression among both men and women. However, we need to interpret this finding with caution. IPV experiences are much less common for men than for women, which may reduce the confidence in conclusions based on statistical estimation.

Prospective Results

We now change to a prospective approach that allows us to examine the consequences of an arranged marriage for depression directly, investigate a complementary measure of IPV, and expand beyond the IPV experiences to a broader evaluation of the marital relationship.

Arranged Marriage, IPV, and Major Depressive Disorder

In Table 6, we examine the association between an arranged marriage and depression, re-examine the consequences of IPV experiences for depression, and investigate the link between positive and negative dimensions of marital quality and depression, controlling for essential covariates, in women and men.

Table 6.

Estimates of arranged marriage, intimate partner violence (IPV) for subsequent onset of major depressive disorder (MDD), Average Marginal Effects (%) from discrete-time hazard models (standard errors in parentheses, N=3912).

Model	Women			Men
	A1 AMEs (%)	B1 AMEs (%)	C1 AMEs (%)	A2 AMEs (%)	B2 AMEs (%)	C2 AMEs (%)
Arranged marriage
Ref: Had a say in marriage
Had no say in marriage	0.20*	0.19*	0.22*	0.04	0.03	0.03
	(0.09)	(0.09)	(0.09)	(0.05)	(0.05)	(0.05)
IPV experiences
Ref: No
Yes		0.46**	0.25		0.08	0.04
		(0.15)	(0.14)		(0.14)	(0.12)
Criticism from spouse
Ref: Never
Seldom			0.05			0.02
			(0.10)			(0.04)
Sometimes/Frequently			0.08			−0.01
			(0.14)			(0.07)
Disagreement with spouse
Ref: Never
Seldom			0.13			0.01
			(0.10)			(0.04)
Sometimes/Frequently			0.56***			0.16
			(0.16)			(0.08)
Husband-Wife emotional bond (maya)
Ref: Some, Little or not at all
Very much			0.03			0.01
			(0.09)			(0.04)
Covariates
Duration of marriage1	–0.01	–0.01	−0.01	0.00	0.00	0.00
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
Had a first birth in the previous year
Ref: No
Yes	1.14**	1.09**	1.10**	0.13	0.14	0.14
	(0.36)	(0.35)	(0.35)	(0.14)	(0.15)	(0.15)
Lifetime any other mental disorder
Ref: No
Yes	24.19***	25.27***	24.55***	1.20	1.26	1.17
	(5.40)	(5.40)	(5.30)	(0.81)	(0.85)	(0.80)
Ethnicity
Ref: Brahmin/Chetri
Dalit	0.21	0.17	0.16	−0.03	−0.03	−0.03
	(0.16)	(0.15)	(0.15)	(0.06)	(0.06)	(0.06)
Hill Janajati	–0.16	–0.16	−0.17	−0.02	−0.02	0.02
	(0.11)	(0.12)	(0.12)	(0.06)	(0.05)	(0.05)
Terai Janajati	–0.03	–0.03	−0.04	−0.07	−0.07	−0.08
	(0.13)	(0.13)	(0.13)	(0.05)	(0.05)	(0.05)
Newar	0.19	0.16	0.14	0.02	0.01	−0.01
	(0.20)	(0.19)	(0.19)	(0.08)	(0.08)	(0.08)
Education
Ref: No education
Some education	–0.05	–0.04	−0.04	−0.04	−0.04	−0.04
	(0.11)	(0.11)	(0.11)	(0.08)	(0.07)	(0.08)
SLC	–0.33	–0.31	−0.31	−0.05	−0.04	−0.05
	(0.17)	(0.17)	(0.17)	(0.09)	(0.09)	(0.09)
IA or BA or above	–0.59***	–0.57***	−0.57***	−0.19*	−0.18*	−0.19*
	(0.16)	(0.16)	(0.16)	(0.08)	(0.08)	(0.08)
Household annual income (rupees)
Ref: 25,000 or less
25,000 – 50,000	0.11	0.11	0.11	0.04	0.04	0.04
	(0.17)	(0.17)	(0.16)	(0.07)	(0.07)	(0.07)
50,000 – 100,000	0.04	0.06	0.08	0.05	0.05	0.04
	(0.14)	(0.14)	(0.14)	(0.06)	(0.06)	(0.06)
100,000 – 250,000	–0.18	–0.16	−0.12	−0.02	−0.02	−0.02
	(0.14)	(0.14)	(0.14)	(0.06)	(0.06)	(0.06)
250,000 or more	–0.21	–0.21	−0.16	−0.02	−0.02	−0.02
	(0.15)	(0.15)	(0.15)	(0.06)	(0.06)	(0.06)
Household asset index	–0.00	0.00	0.01	−0.00	−0.00	−0.00
	(0.03)	(0.02)	(0.03)	(0.01)	(0.01)	(0.01)
Household agriculture index	–0.02	–0.01	−0.01	−0.01	−0.01	−0.01
	(0.02)	(0.02)	(0.02)	(0.01)	(0.01)	(0.01)
Living in a three-generation household
Ref: No
Yes	–0.05	–0.04	−0.05	0.02	0.02	0.02
	(0.09)	(0.09)	(0.09)	(0.04)	(0.04)	(0.04)
Pray at a Temple
Ref: Never
Once a month or less	0.00	0.02	0.03	−0.01	−0.01	−0.01
	(0.13)	(0.13)	(0.35)	(0.05)	(0.05)	(0.05)
More than once a month	–0.16	–0.14	−0.13	0.04	0.04	0.04
	(0.15)	(0.14)	(0.14)	(0.06)	(0.06)	(0.01)
Age	0.18***	0.17***	0.17***	0.03**	0.03**	0.03**
	(0.03)	(0.03)	(0.03)	(0.01)	(0.01)	(0.01)
Age squared	–0.00***	–0.00***	−0.00***	−0.00*	−0.00*	−0.00*
	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)
Birth cohort
Ref: 1957–1971
1982–1991	–0.04	–0.03	0.01	0.07	0.07	0.08
	(0.13)	(0.13)	(0.13)	(0.07)	(0.07)	(0.07)
1972–1981	0.08	0.09	0.08	0.03	0.03	0.03
	(0.11)	(0.11)	(0.11)	(0.04)	(0.04)	(0.04)
Persons	2088	2088	2088	1824	1824	1824
Log likelihood (ll)	–3042.62	–3036.71	−3025.85	−762.43	−762.20	−758.95
AIC	6135.25	6125.41	6113.71	1574.87	1576.41	1579.89
BIC	6359.14	6358.25	6391.32	1799.29	1809.81	1858.18

Note:

^1.We present the results excluding the duration of marriage in Table S4. The main results are consistent.

^{2. *}p<0.05

^**p<0.01

^***p<0.001.

For women, an arranged marriage has a positive, statistically significant relationship with depression. For instance, Model A1 shows that an arranged marriage increases the probability of depression by 0.20% among women, compared with a marriage in which the woman had at least some say in the choice of her spouse.¹⁰ However, we do not observe a statistically significant relationship between a parentally arranged marriage and depression among men (Model A2), even though the direction is positive.

Consistent with the results in Table 3, the different, complimentary measure of IPV experiences is also positively associated with depression in women. In Model B1, having IPV experiences increases the probability of depression by 0.46% among women. The increase in the probability of depression due to IPV experiences is smaller in the prospective analyses than that in the retrospective analyses. This may be because mild and moderate IPV experiences are more likely to be underreported in the retrospective measurement (Naved et al. 2006), and the observed stronger positive association in the retrospective analyses is driven by those who experienced the most severe IPV traumatic experiences. In addition, consistent with results in the retrospective analyses, we do not observe a significant association between IPV experiences and depression among men and the magnitude of the AMEs is small.

To examine whether IPV experiences mediate the pathway between arranged marriage and depression, we present the mediation results in Table 7. For women, after controlling for IPV experiences, the increase in the probability of depression due to an arranged marriage is reduced from 0.20% to 0.19%. Approximately 0.005% of the risk is explained through IPV experiences (albeit, insignificant), which only counts for 2.50% of the total risk. As for men, the mediation effect of IPV is close to zero and insignificant. IPV is an insignificant mediator between arranged marriage and depression among women. This indicates that there are other mechanisms explaining the strong, positive association between arranged marriage and depression in Nepal.

Table 7.

Average Marginal Effects (%) of arranged marriage on major depressive disorder (MDD) before and after controlling for IPV, controlling for covariates (Standard errors in parentheses, N=3912).

	Women			Men
	A1 AMEs (%)	B1 AMEs (%)	A1-B1 AMEs (%)	A2 AMEs (%)	B2 AMEs (%)	A2-B2 AMEs (%)
Average marginal effects
Arranged marriage
Ref: Had a say in marriage
Had no say in marriage	0.200*	0.194	0.005	0.032	0.032	0.000
	(0.099)	(0.099)	(0.003)	(0.047)	(0.047)	(0.001)
Covariates	Yes	Yes	Yes	Yes	Yes	Yes

Note:

^1.SEs are estimated with the seemingly unrelated estimation approach (Mize, Doan, and Long 2019). An alternate approach, non-parametric bootstrapping, generates similar results.

^{2. *}p<0.05

^**p<0.01

^***p<0.001.

The timing of marriage is an important confounder to consider, both because length of time married defines exposure to the risk of IPV in Nepal and because timing of marriage is associated with the level of marital arrangement by parents (Ghimire et al. 2006). We also tested excluding the duration of the first marriage, and the main results are consistent. We present results of these models in the appendix as a reference for the changes in the associations of marital and IPV experiences with depression (see in Table S4).

Gender, Arranged Marriage, IPV, and Major Depressive Disorder

Table 8 shows the gender differences in the associations of arranged marriage and IPV experiences with depression. An arranged marriage leads to a higher increase in the probability of depression among women (0.20%) than among men (0.03%), though the statistical test on gender differences is insignificant. In contrast, we find significant gender differences in the association between marital experiences and depression in the retrospective analyses (Table 5). These findings suggest that marital experiences are indeed gendered experiences in Nepal, but we do not have sufficient statistical confidence to conclude that an arranged marriage, completely determined by parents, have gendered consequences for both men and women.

Table 8.

Gender differences in the estimates of arranged marriage, intimate partner violence (IPV) for subsequent onset of major depressive disorder (MDD), Average Marginal Effects (%) from discrete-time hazard models (standard errors in parentheses, N=3912).

Model	Women	Men	Gender Difference
	B1 AMEs (%)	B2 AMEs (%)	B1-B2 AMEs (%)
Average marginal effects
Arranged marriage
Ref: Had a say in marriage
Had no say in marriage	0.195*	0.032	0.163
	(0.09)	(0.042)	(0.113)
IPV experiences
Ref: No
Yes	0.456**	0.082	0.374
	(0.147)	(0.137)	(0.210)
Covariates	Yes	Yes	Yes

Note:

^1.SEs are estimated with the seemingly unrelated estimation approach (Mize, Doan, and Long 2019). An alternate approach, non-parametric bootstrapping, generates similar results.

^{2. *}p<0.05

^**p<0.01

^***p<0.001.

In addition, consistent with results in the retrospective analyses, we find no statistically significant gender differences in the consequences of IPV experiences for depression. This finding further indicates that IPV experiences within marriage may have similar consequences for the risk of depression among both men and women. However, men are much less likely to be victims of IPV. Again, we need to interpret the results with the consideration that this low incidence of IPV occurrence among men reduces the statistical confidence of the estimation.

Marital Conflict, Emotional Bond, and Major Depressive Disorder

As a supplementary analysis, we examine the association between marital quality (including both positive and negative dimensions) and depression. Model C1 and C2 in Table 6 include criticism from a spouse, disagreement with a spouse, and husband-wife emotional bond.

For women, those who sometimes/frequently disagree with their spouse are 0.56% more likely to have depression than those who never disagree with their spouse. Because both criticism and disagreement reflect negative aspects of marital relationships, we also examined them separately, controlling for the other covariates (see Table S5). Frequent criticism from and disagreement with spouse significantly increase the probability of depression for women by 0.34% and 0.60% without controlling for each other (see Table S5). However, after including both in the same model, we only observe a significant independent association between disagreement with spouse and depression. Husband-wife emotional bond is not significantly associated with depression, even when not controlling for criticism and disagreement with spouse (see Table 6 and Table S5).

In addition, the AME of IPV experiences for women is reduced from 0.46% to 0.25% after controlling for both positive and negative dimensions of marital relationships. Table 9 presents the test of the decrease in the AMEs of IPV experiences for both men and women. The reduction in the AMEs of IPV experiences is statistically significant for women. As for men, we do not find significant associations between positive and negative dimensions of marital relationships and depression. Even though we observe a decrease in the AMEs of IPV experiences in men, the reduction is not statistically significant.

Table 9.

Average Marginal Effects (%) of IPV on major depressive disorder (MDD) before and after controlling for disagreement, criticism, love with spouse, controlling for covariates (Standard errors in parentheses, N=3912).

	Women			Men
	B1 AMEs (%)	C1 AMEs (%)	B1-C1 AMEs (%)	B2 AMEs (%)	C2 AMEs (%)	B2-C2 AMEs (%)
IPV experiences
Ref: No
Yes	0.456**	0.249	0.207***	0.082	0.045	0.038
	(0.147)	(0.137)	(0.051)	(0.137)	(0.122)	(0.033)
Covariates	Yes	Yes	Yes	Yes	Yes	Yes

Note:

^1.SEs are estimated with the seemingly unrelated estimation approach (Mize, Doan, and Long 2019). An alternate approach, non-parametric bootstrapping, generates similar results.

^{2. *}p<0.05

^**p<0.01

^***p<0.001.

Taken together, these findings suggest that a broader measurement of marital conflicts, such as disagreements with a spouse, has a more consistent association with depression and can explain a large portion of the positive association between IPV experiences and depression, especially among women.¹¹ Secondly, women are more vulnerable than men to negative experiences—from physical violence to broader marital conflicts—in marital relationships.

Discussion

This study is the first to investigate the complex consequences of marital relationship and IPV experiences for individual mental health in a non-European diaspora social setting—a setting in which widespread arranged marriage persists and women and men have highly gendered marital experiences (Allendorf 2017; Kolenda 1987). Different from the majority of previous studies, which used screening or assessment scales to capture non-specific distress, this study used population-scale DSM-IV diagnoses of major depressive disorder (MDD). This clinically validated diagnostic measure of mental disorder greatly reduces measurement error in estimation of significant adverse mental health outcomes. Another strength of this study is the combination of prospective and retrospective research designs. Each design has limitations, but together they present a comprehensive empirical documentation of key associations.

Especially noteworthy in this setting, we find that any marital transition, including becoming married, increases the probability of depression. In the U.S. and Europe, empirical studies consistently show that marriage is associated with enhanced mental health for men and either enhanced or no change in mental health for women (Scott et al. 2010; Simon 2002; Williams 2003). However, we find a reversed association among women and no significant association among men in Nepal. We also show that divorce and widowhood (versus in marriage) are associated with an increased probability of depression in both men and women. This finding is consistent with previous research in western countries (Edwards et al. 2018; Simon 2002; Williams 2003). These findings of associations between marital experiences and first onset of depression and the gender differences are also documented elsewhere (see Axinn, Zhang, et al. 2020 for more details).

Both the retrospective and prospective analyses consistently demonstrate that intimate partner violence (IPV) experiences within marriage lead to large, significant increases in the risk of depression among women. Overall, there is little doubt that being the victim of physical violence perpetrated by one’s partner increases transitions to a diagnosis of major depressive disorder. Perhaps this is true among both women and men, but equally clearly this powerfully adverse experience is much more likely among women. Our finding here is consistent with prior research (e.g., Breiding et al. 2008; Bromet et al. 2018; Campbell 2002; Ellsberg 2008; Trevillion et al. 2012). But the replication of this result from a much different setting, with numerous cultural differences, points toward the universal adverse consequences of violence within intimate partner relationships.

Using the strengths of our retrospective design, results indicate that IPV experiences mediate the pathway between marriage and depression among women. IPV experiences explain a small portion of the strong, positive association between marriage and depression, but there are other factors producing adverse consequences of marriage in Nepal, especially for women.

The distinctive characteristics of marriage in Nepal and sharp differences in gender roles might provide a potential explanation for the gender differences in associations with depression observed in this study. In Nepal, spouses are often selected by parents, especially for women. Even as individuals become increasingly involved in the choice of a spouse, men have more freedom in the choice of a spouse than women (Allendorf 2013; Ghimire et al. 2006). In addition to a lower level of participation in the decision of whom they marry, Nepalese women and men face highly gendered family roles after getting married (Bennett 1983). The roles of women within marriage include childbearing, childrearing, and household work are gender-segregated. Also, women have less control of household wealth or financial assets and inheritance of property (Acharya and Bennett 1981). Additionally, as we show here, women are more likely than men to experience IPV within marriage. However, in Nepal divorce is rarely an option to leave an unhappy marriage (Jennings 2014, 2017). Altogether, these cumulative disadvantages of women in marriage make marriages more harmful for Nepalese women’s mental health than men’s mental health.

Important for understanding this setting, and settings like it, our prospective analyses reveal that an entirely parentally arranged marriage is associated with an increased risk of the first onset of lifetime depression among women. Even though individuals have become increasingly involved in the choice of spouse in Nepal (Allendorf 2017; Ghimire et al. 2006), many of these changes are modest, and the majority of recent marriages are still entirely arranged by parents for women. Scholars argue that the practice of an arranged marriage is shifting rather than declining with individuals joining their parents in choosing spouses rather than displacing their parents in the decision process (Allendorf and Pandian 2016). However, even the shifting of arranged marriage has not happened evenly across gender, with men being much more likely to be involved in decision-making about marriage than women in Nepal (Ghimire et al. 2006). We find evidence that parentally arranged marriage increases the risk of depression for all, and most clearly for women. Less input into these marital decisions has the potential to harm mental health after marriage.

Our prospective analysis provides considerable insight into these powerful adverse consequences of parentally arranged marriages for women. First, in spite of the strong association between IPV experience and depression for women, the association between arranged marriage and depression is almost entirely independent of experiences with IPV. By itself, this is something of a stunning indication of the mental health consequences for married people with no choice in the selection of a spouse. Second, the prospective design allows us to investigate the multidimensional nature of marital relationships, beyond reports of physical violence to include more general statements of disagreements. This allows us to document that not only are more general disagreements a powerful factor in increased depression for married women, but that consideration of this broader measure of relationship adversity substantially reduces observed associations with IPV. Third, even as this result is consistent with the argument that adverse marital quality should be conceptualized more broadly than physical violence to capture health and wellbeing consequences, this broader measurement also fails to reduce the association between parentally arranged marriage and depression for women. To be clear about the importance of this finding, recall that our measures of depression go beyond mood screening to assess behavioral impairments and are clinically validated to diagnose depression. For women, marriage with no individual input into the choice of a spouse has substantial long-term mental health consequences. These findings are not only crucial to our understanding of the links between marital experiences and mental health, they are equally crucial for parents and husbands to understand the long-term consequences of their approach to marriage.

Our findings should be interpreted with the following limitations in consideration. A first notable limitation of this study is its age restriction of respondents from 10 to 59. As life expectancy increases, age 59 becomes the beginning of higher exposure to the risk of widowhood. Therefore, both the number of cases and person-years of widowhood reported in our sample are relatively low. However, the fact that we document a strong, robust association between widowhood and subsequent depression when relying on few cases and shorter person-years suggests that we may still observe a strong relationship if we had a more complete reporting of widowhood that better captured respondents’ marital status over time (Shin, Kim, and Park 2018).

A second limitation is that our study focuses on only one direction of the potentially bidirectional association of marital and IPV experiences with mental health. Prior research demonstrates that people might be selected into marriage or out of marriage based on their mental health (Burt et al. 2010; Edwards et al. 2018; Yamaguchi and Kandel 1985); people might also be selected into IPV experiences because their mental health triggers stress and conflicts within marriage (Anderson 2002). The near universal marriage and rare divorce social setting reduces the probability of selection into and out of marriage based on mental health. Furthermore, our hazard model approach excludes the person-years after the occurrence of depression, so the person-years of the respondents to whom the IPV or marital experiences occurred after the onset of depression were eliminated from the models. Although this approach does not provide insight into the influence of depression on subsequent experiences of IPV or marital transitions, it does resolve the potential for these bi-directional associations to confuse the estimated association of IPV or marital experiences with subsequent depression. Additionally, as a robustness check, we also performed fixed-effects hazard models with adjusted first difference estimation (Tauchmann 2019)¹² for the retrospective analyses (see Table S6) to control for the unobserved confounding. The positive associations of marriage and IPV with depression remain positive and significant. Nevertheless, given the association that we document here, models of the potentially complex reciprocal association remain an important priority for future research.

A third limitation is that the measures of IPV experiences are limited to a single item and with a focus on physical violence in each of our retrospective and prospective analyses. This approach is likely to underestimate the prevalence of IPV. It is possible that we overestimate the magnitude of the association between IPV and depression in this study, because severe IPV experiences are more likely to be reported (Naved et al. 2006). However, the fact that we document a strong, robust association between two types of IPV measures and subsequent depression, and IPV only mediates a small portion of negative consequences of (arranged) marriage, leave little doubt that IPV experiences have an independent consequence for depression. Additionally, the strong link between broad marital conflicts and depression indicates that increased measurement of IPV and marital quality are also important avenues for future research on the links between intimate relationships and health.

Despite these limitations, this study motivates us to rethink key assumptions (such as the positive social support from becoming married) considering the contextualized nature and meaning of marriage in specific social settings. There is ample evidence that marriage has the potential to deliver positive mental health benefits to both men and women (Simon 2002; Williams 2003). Across many settings, however, it is equally clear these benefits are less for women (Scott et al. 2010). The evidence we present from Nepal demonstrates that in some settings marriage has the potential to produce the opposite—it can create significantly higher odds of transitioning to a diagnosis of major depressive disorder. We find two highly gendered experiences likely to contribute to this adverse result. One is absolute zero voice in the choice of a spouse. This experience may produce higher probability of depression among both men and women, but in Nepal it happens much more frequently to women. The other is intimate partner violence (IPV) or broader marital conflicts (e.g., disagreement). Again, this experience may produce higher probability of depression among both men and women, but in Nepal it happens much more frequently to women. Increasing women’s voice in the choice of a husband and reducing the prevalence of IPV both have the potential to reduce the prevalence of depression.

From a sociological perspective, however, these adverse experiences explain relatively little of the total association between (arranged) marriage and depression among women. Marriage in Nepal is embedded in a setting with a long history of patriarchy and rigid gender segregation; therefore, women and men may have many highly gendered experiences within marital relationships. Besides IPV experiences, other highly gendered adverse experiences likely play roles in the link between (arranged) marriage and depression among women. These adverse experiences may encompass parenthood, financial dependence, and unequal decision-making in sexual experiences. The results we present are independent of the risk factor of childbirth for depression, but gendered roles and responsibilities of parenthood are still potential reasons. Decision-making in sex, sexual assaults, and/or forced intercourse is also a highly gendered experience, and all signs point toward substantial prevalence within monogamous relationships like marriage (Axinn, Bardos and West 2018). Additionally, in a setting like Nepal, generally wives have little control of family assets and other economic resources and are highly financially dependent on their husbands (Acharya and Bennett 1981), which may also contribute to the adverse consequences of marriage for women. More research is needed to understand the role that sexual assaults, financial dependence, parenthood or other highly gendered adverse experiences play in connecting marriage to depression. However, the overarching sociological issue is that although marriage can be a source of social support, the science of these investigations must explicitly recognize the potential for adverse experiences within marriage to create equally powerful declines in mental health and overall wellbeing.