Parents’ Marital Quality and Children’s Transition to Adulthood

Abstract

Unique longitudinal measures from Nepal allow us to link both mothers’ and fathers’ reports of their martial relationships with a subsequent long-term record of their children’s behaviors. We focus on educational attainment and marriage timing because these two dimensions of the transition to adulthood have wide-ranging, long lasting consequences. We find that children whose parents report strong marital affection and less spousal conflict attain higher levels of education and marry later than children whose parents do not. Furthermore, these findings are independent of one another and of multiple factors known to influence children’s educational attainment and marriage timing. These intriguing results support theories pointing toward the long-term intergenerational consequences of variations in multiple dimensions of parents’ marriages.

The emotional bond between spouses is a critical component of family life, yet we know little about the long-term consequences of this component for the next generation. Marriage is a key dimension of social networks in nearly all societies (Coleman 1994; Ogburn and Tibbets 1934; Thornton and Fricke 1987) and emotional dimensions are widely recognized as core features of marriage, regardless of whether emotions were historically considered important for marriage formation (Ahearn 2001; Fuller and Narasimhan 2008; Pasupathi 2002). Research has established intergenerational links between parents’ marital experiences (such as divorce) and multiple dimensions of children’s transitions to adulthood, yet little investigates emotional dimensions of the parental marriage (for exceptions see Rijken and Liefbroer 2009 and Orthner et al. 2009).

In this paper we advance this topic by investigating the association of the emotional bond between mothers and fathers with the subsequent pace of their children’s transition to adulthood using more than a decade of intergenerational panel data from Nepal. We focus on two key dimensions of this transition: educational attainment and the entrance into marriage. Educational attainment has many important consequences for adulthood, particularly in work and earnings trajectories (Card 1999), but also in health, marital, and childbearing experiences (Baker et al. 2011; Brand and Davis 2011; Hout 2012). Likewise, marriage timing also has many important consequences, with earlier than average marriage associated with lower earnings, higher risk of divorce, poor mental health, higher rates of maternal and child death, and higher risk of domestic violence (Harris, Lee, and De Leone 2010; Loughran and Zissimopoulos 2009; Santhya et al. 2010; UNICEF 2001).

Research reveals that parental family factors constitute a powerful, long-lasting influence on both their children’s educational attainment and marriage behavior. For example, longitudinal studies demonstrate that the parental family in early childhood shapes cognitive development and children’s educational attainment (Cavanagh, Schiller, and Riegle-Crumb 2006; Fomby and Bosick 2013; Fomby and Cherlin 2007; Heard 2007; Sun and Li 2007). Similarly, studies also show the parental family influences both the pace at which children exit the parental home and at which they form their own unions (Goldscheider and Goldscheider 1993; Goldscheider and Waite 1993; McLanahan and Bumpass 1988). We add to the literature in two major ways. First, we examine the emotional dimension of the parental home environment as opposed to family events (such as divorce). Second, we explicitly link these measures of parental marital quality in childhood to trajectories of both children’s educational attainment and their marriage timing over 12 years. Also novel, we investigate these associations in rural South Asia, a setting quite different from the Western European and U.S. setting for most prior research on these associations.

We have multiple related research questions. Independent of other factors known to influence children’s education and marriage timing, do children whose parents have a stronger positive emotional bond (express more love) (1) have higher educational attainment and/or (2) marry later than those whose parents have a weaker positive emotional bond? And (3) do children whose parents have more marital conflict exit school and marry earlier than those whose parents have less marital conflict? Our theoretical framework and analyses examine parental affection and conflict as different, potentially contrasting, dimensions of the parental marital relationship. In the end, this study motivates increased research attention to the long-term intergenerational consequences of marital quality.

Theoretical Framework

To guide our investigation into the associations between parental marital relationships and their children’s transition to adulthood we construct a multi-component theoretical framework. We draw on research from Western Europe and North America to identify likely mechanisms creating these associations, but apply this reasoning to a very different context – Nepal. Therefore, we begin with background information on Nepal to contextualize both the specific dimensions of marital relationships we consider and the theoretical connection between those dimensions and children’s transition to adulthood. We then build upon that to present some setting-specific mechanisms likely to link parents’ marital quality to their children’s transition to adulthood generally, and both educational attainment and marriage timing specifically. Finally, we discuss the known determinants of children’s educational attainment and marriage timing that may also shape the emotional dimensions of the parental relationship.

Marriage in Nepal

Marriage in Nepal is universal. Until recently, marriages occurred at young ages for both men and women with a non-trivial amount of child marriage, but it has increased in recent decades for both men and women (Bajacharya and Bhandari 2014). For example, the median age at marriage for women increased from 16.4 in 1996 to 17.5 in 2011. Currently, the median age at first marriage among men aged 25–49 is 21.7 years among men age 25–49 (Ministry of Health 2017). Thus, women in Nepal marry about 4 years earlier than men. This age pattern is similar to that seen in India, although later than in Bangladesh (e.g., median age at marriage for women aged 25–49 in 2005/2006 was 17.0 for Nepal, 16.8 for India, 14.5 for Bangladesh) (Macro Intl. 2007). These relatively young ages at marriage make Nepal quite different from the U.S. and Western Europe.

Nepal’s history of parentally arranged marriage is another important difference from the U.S. and European populations. Marriages were characterized by high levels of parental involvement in many ethnic groups, but also high heterogeneity across ethnic groups in the ways that they were and are arranged (Acharya and Bennett 1981; Ahearn 2001; Fricke 1986). Individuals are becoming increasingly involved in the choice of spouse (Allendorf and Pandian 2016, Allendorf et al. 2017; Ghimire et al. 2006). In the data used here, no spouses of the 1936–45 marriage cohort had participated in the selection of their spouse, but half of those in the 1986–95 marriage cohort had done so (Ghimire et al. 2006). “Love marriages,” where parents have essentially no say in their children’s marriage partner are also increasingly prevalent, although they remain uncommon. In most Nepalese ethnic groups wives move to live with the husbands’ family after marriage, although this is less widespread today than in the past (Bennett 1983; Pearlman, et al. 2017).

Finally, premarital cohabitation and divorce, both common in North America and Western Europe, are increasing but are still quite rare in Nepal. There remains no evidence of population-scale premarital cohabitation, but divorce is rapidly becoming more common (Jennings 2016).

Education in Nepal

The spread of schools and accompanying increased proportion of childhood spent in school are key factors producing these changes in marriage and other family related behaviors (Brauner-Otto 2012; Ji 2013; Yabiku 2005). The dramatic increase in the availability and quality of schools created a watershed change in mass education, increasing both school enrollment and educational attainment, with wide-ranging demographic consequences (Axinn and Barber 2001; Brauner-Otto 2012; Ghimire, et al. 2006). From 2011 to 2016 the median years of schooling for those aged 15–49 increased from 3.5 to 5 for women and from 7.4 to 8 for men (Ministry of Health 2017). In our study area, by 1996 all children aged 5 and 6 had been to school for at least one day (Beutel and Axinn 2002). School location and quality has also changed over this time period. In 1996 only 88% of households in Nepal lived within a 30 minute walk of a school, but 95% did so by 2011 (Ministry of Education 2015). The percent of teachers with required qualifications and training increased from 15% in 2001 to 98% in 2012 (Ibid.).

The educational system in Nepal is structured such that primary school begins at age 5 and the completion of secondary school occurs after grade 10. Early childhood care and education (ECCE) has also become a greater priority in the Nepalese education plan and enrollment has similarly increased. For example, roughly 40% of eligible children were enrolled in ECCE in 2006 but over 70% were by 2012 (Ibid.). Upon completion of grade 10 students are eligible to take an exam to earn their School Leaving Certificate (SLC), the equivalent of a high school diploma in the U.S. Fewer than 3% of ever married women had received an SLC in 1996, 18.4% of women aged 15–49 had done so by 2011, and 24.4% by 2016 (Ministry of Health 2017). For men, the percent aged 15–49 with an SLC increased slightly from 31.6% to 36.8% from 2011 to 2016 (Ministry of Health 2017).

Parental Marital Quality

Research consistently finds that, across settings, marital quality is comprised of multiple related, often contrasting, but independent dimensions (Allendorf and Ghimire 2013; Bradbury, Fincham and Beach 2000; Umberson et al. 2005). The positive emotional bond (affection or love) is commonly identified as one of the most important (Allendorf 2009; Allendorf and Ghimire 2013; Axinn, Ghimire, and Smith-Greenaway 2017; Goode 1970). This has long been the case for research on Western European marriages which usually treats love and affection as a prerequisite and the singular core defining feature of marriage (Goode 1959; Hamon and Ingoldsby 2003; Thornton et al. 2007). The recognition of the centrality of love in non-Western marriages is more recent, as arranged marriages were often characterized as loveless, empty, patriarchal, and without choice (Khandelwal 2009; Pasupathi 2002). However, recent work criticizes this ethnocentric characterization recognizing love, in particular that between husbands and wives, as a universal psychological phenomenon, not something unique to Western marriage (Coontz 2005; Khandelwal 2009). In fact, Hinduism often glorifies the idea of love between the sexes as is evident from the variety of mythical love stories, such as Kamasutra, that abound in Sanskrit literature (Vātsyāyana 2009). It is clear that the positive emotional bond between husbands and wives is part of daily life for married couples in Nepal and that there is variation in the degree of this bond (Ahearn 2001; Allendorf 2009; Fricke 1986; Macfarlane 1976; Pasupathi 2002).

Negative dimensions of marriage, such as conflict between spouses, are also apparent in families in Nepal (Allendorf and Ghimire 2013; Hoelter, Axinn, and Ghimire 2004; Jennings 2014). Domestic violence, disagreements between spouses, and worries about the marriage have all been identified as prevalent in marriages in Nepal (Allendorf and Ghimire 2013; Ghimire et al. 2015; Hoelter et al. 2004). An important but missing component of the literature is understanding how love and conflict between spouses influence their children’s lives. The combination of measures of both positive emotional bond and spousal conflict (obtained from both members of the couple) in a long-term, intergenerational panel study provides us the rare opportunity to provide insight into the associations with these different, often contrasting, dimensions of marital quality.

Mechanisms Linking Parents’ Marital Quality to Children’s Transitions to Adulthood

Previous research indicates multiple mechanisms may link parents’ marital quality to their children’s life-courses (for a review see Orthner et al. 2009). Here we consider three likely to be operating in the Nepal-specific context: parental investments in childrearing, general social psychological links between parents and children, and children’s motivations to leave the parental home. This study does not aim to identify which mechanism is at work, and in all likelihood all three, and possibly others, are operating simultaneously. Rather, we examine the overarching relationship between parental marital quality and children’s transition to adulthood.

Parents’ investments in childrearing.

Positive parental marital quality may shape childrearing practices, with potentially strong consequences for educational attainment. As processes of entering marriage change, some scholars argue that married couples with higher levels of affection for each other may be more likely to spend more resources on their children than on their extended kin (Caldwell 1982; Degler 1980). In this view, as childbearing and rearing become an expression of marital love and affection, couples invest more time and resources into their children and begin trading low investments in many children for high investments in a small number of “high quality” children (Becker 1991; Caldwell 1982; Willis 1973). Parents’ investments in the “quality” of their children may increase children’s educational attainment, particularly as children’s educational resources have been identified as the leading form of parental investment in the quality of their children (Becker 1991; Caldwell 1982; Willis 1973). Therefore, we expect positive emotional bonds between parents to be positively related to their children’s educational attainment. Further, the social science literature provides formidable evidence that enrollment in school delays marriage among both men and women and across a wide range of settings, including in Asia (Caldwell et al. 1988; Thornton et al. 2007; Thornton and Lin 1994; Yabiku 2005). Because high educational attainment requires school enrollment, this mechanism may also slow children’s marriages.

Social psychological links between parents and children.

Social psychological perspectives on parenting behavior and intergenerational transmission suggest that marital relationships may shape multiple aspects of parents’ lives, including their parenting (Cherlin, Kiernan and Chase-Lansdale 1995; Shek 1998). This has been described as a “spill over” from the parental marital relationship to the parent-child relationship (e.g. Malinen et al. 2010), as shaping children’s ‘taste’ for family (Rijken and Liefbroer 2009), or simply as a fundamental component of the socialization process. Empirical research demonstrates this across the life course and for both behavioral and psychological outcomes such as children’s behavior problems and their family related attitudes in adulthood (Goldberg and Carlson 2014; Cunningham and Thornton 2006).

We expect that these positive dimensions of the parental marital relationship may also “spill over” to have other consequences for their children’s lives, including their children’s marital relationships. In settings historically characterized by young and arranged marriages, like Nepal, this positive “spill over” may delay children’s marriage. Exposure to parents’ positive marital relationships may set different expectations for the younger generation’s marital relationships such that young people and their families spend more time selecting a spouse before entering marriage (Bandura 1986). This is particularly likely as an increasing fraction of young people participate in the selection of their own spouse (Allendorf 2017; Ghimire et al. 2006). This mechanism could lead to delayed timing of marriage among young people with parents who have a relatively more positive relationship. Additionally, because early marriage is a key reason for truncated educational attainment in a setting like Nepal (Buetel and Axinn 2002), it is also possible that this mechanism links more positive parental marital relationships to higher educational attainment.

Children’s motivations to move away.

Children’s own motivations to move away from the parental home may also link parents’ marital quality to children’s transition to adulthood, especially their marriage timing (Goldscheider and Goldscheider 1993; Goldscheider and Waite 1993; Thornton et al. 2007). The quality of the parental home has been identified as influencing the pace of children’s moving out (Goldscheider and Goldscheider 1993). As with other dimensions (e.g., household goods) of the parental home that may make it more pleasant, positive mother-father relationships are expected to reduce children’s motivation to exit the parental home. Said another way, the positive parental marital relationship may delay children’s marriage because family members are content in their current situation (Rijken and Liefbroer 2009). This reduced motivation to leave the parental home is then expected to reduce the pace of entry into marriage because marriage is an important way children secure living arrangements away from parents (Goldscheider and Waite 1993; Thornton et al. 2007).

This is particularly likely in Nepal where, in contrast to settings like the U.S. in which young people often enjoy many different options for alternative living arrangements outside the parental home, marriage is the most common long-term, viable option (Fricke 1986; Macfarlane 1976; Pearlman et al. 2017; Yabiku 2004). It is rare that young people live alone, with housemates, or in school or work dormitories. Moreover, young people who test these alternative arrangements do not spend much time in them and transition rapidly to marriage. It is also possible that by continuing to live in the parental home, children are able to focus on education, achieving higher levels of educational attainment.

Negative dimensions of marital quality.

Because marital quality is multidimensional we note that while positive dimensions, such as love, may be operating as described above, negative dimensions, such as conflict, may also be simultaneously influencing children (e.g. Amato and DeBoer 2001; Amato and Sobolewski 2001; Cherlin et al. 1995; Rijken and Liefbroer 2009; Wolfinger 2003). For example, negative dimensions of the parental relationship may lower the quality of the parental home thereby increasing the motivation to leave the parental home. As described above, in this context marriage is the most likely long-term path away from the parental home and marriage truncates education attainment. Therefore, in this context, negative parental marital relationships may speed children’s marriages and lower educational attainment.

Because children are increasingly involved in the selection of their spouse, parental marital quality is likely related to children’s marriage timing even in this setting of marital arrangement. Children whose parents do not love each other or who have a high conflict marriage may be more willing to accept a less desirable spouse. Or, children in higher conflict homes may have lower expectations for positive affect from a partner, enabling earlier marriage by allowing a broader range of potential spouses to be acceptable.

Factors that May Shape Both Parental Marriage Quality and Children’s Transition to Adulthood

In order to estimate the relationship between parental marital quality and children’s transition to adulthood in an observational study design, we must consider other aspects of the parental family known to shape both children’s transitions to adulthood and parents’ emotional bond (Kalmijn 1991, 1994, 2015; Mare 1991; Sweeney 2002). As they themselves are not the focus of this manuscript we discuss them only briefly below.

Empirical research in Western and non-Western settings has documented substantial variation in the transition to adulthood by race, ethnicity, and religion (Goldscheider and Goldscheider 1993; Goode 1959; Gullickson and Torch 2014; Kuo and Kelly 2016; Page and Stevens 2005). In Nepal, these three characteristics are intricately connected, typically manifest themselves in terms of caste-ethnicity, and have been linked to variation in family formation behavior (Bennett 1983; Fricke 1986; Pearce, Brauner-Otto, and Ji 2015).

Parents’ marital and non-family experiences also shape the quality of their relationship and their children’s transitions to adulthood. Parental age at marriage has consistently been found to influence children’s transition to adulthood across geographic settings (Allendorf and Ghimire 2013; Goldscheider and Waite 1993; Thornton and Lin 1994). In a setting like Nepal, marital arrangement in the parents’ generation is also likely important as it may influence the level of spousal choice that they give their children, so that parents who had more choice may also give their children greater choice, leading the children to delay their entrance into marriage. Parents’ childbearing may also be important (Axinn, Clarkberg and Thornton 1994; Barber and Axinn 1998) as those who are strongly pro-natalist are likely to socialize their children to marry more quickly (Barber and Axinn 1998), but must also divide their investments in children across more children and the number of children couples have may influence their marital quality. Finally, parents with more experiences outside the family, for instance with living away from parents, and going to school, have likely been more influenced by Western ideas of education and courtship prior to marriage, which may influence their martial quality, their investments in their children, and their children’s marriage timing (Allendorf and Ghimire 2013; Ghimire et al. 2015; Goldscheider and Waite 1993; Thornton and Lin 1994; Yabiku 2005).

Prior research has also demonstrated mothers’ and fathers’ communities in childhood and later in life influence their children’s transitions to adulthood (Thornton and Lin 1994; Yabiku 2004, 2006). Communities shape both the opportunity structure individuals face and have independent influences on individuals’ attitudes (Barber 2004; Caldwell, et al. 1988; Coleman 1994; Pearce, et al. 2015). Schools can be particularly important because mechanisms link access to schools to educational attainment and delayed marriage, independently of school attendance (Axinn and Barber 2001; Brauner-Otto 2012; Thornton et al. 2007; Yabiku 2004).

Methods

Data

To test our predictions we use data from the Chitwan Valley Family Study (CVFS) conducted in the Western Chitwan Valley in Nepal. We are able to use the CVFS for this investigation because separate, individual interviews measuring marital relationship quality were conducted independently with mothers and fathers and their children were subsequently followed for 12 years to document their education and marital behaviors. In 1996 the CVFS collected information from residents of a systematic sample of 151 neighborhoods, or tols, in the Western Chitwan Valley. Tols are distinct clusters of 5–15 households, typically located at crossroads and surrounded by fields. Every resident between the ages of 15 and 59 in the sampled neighborhoods and their spouses were interviewed.

In 2008, original 1996 respondents were re-interviewed at which time women completed Life History Calendars (LHCs) with yearly information on children’s school enrollment for all children. These data are the source for annual measures of children’s education after the 1996 baseline interviews. Following the 1996 interviews, the CVFS began collecting a prospective monthly demographic event registry for every individual interviewed in 1996, even if they left Chitwan, yielding 126 months of event measures. These data are our source for monthly measures of the children’s marriage behavior after the 1996 baseline interviews.

We study the intergenerational link between parents’ marital quality and children’s transition to adulthood among children whose parents were married and interviewed in 1996. The CVFS interviewed both husbands and wives with an overall initial response rate of 97%. Response rates remained high over the data collection period with a low of 93% (Axinn, Ghimire, and Williams 2012).

To study educational attainment we examine children age 16 or younger, alive and living with their parents in 1996. We limit the sample to these children to ensure they experienced the parental marital quality as measured. Students typically finish secondary school around age 16 so we exclude children older than 16 in 1996. The final analysis sample is 2714 children living with 1168 sets of parents. We also estimate models including children born before 1980 and those born after 1996 and find similar if not stronger relationships than shown here.

To study marriage we examine children who were unmarried but aged 15–24 in 1996 (i.e., interviewed in 1996 and subsequently exposed to the risk of first marriage). We exclude children if they were already married in 1996 to ensure proper temporal ordering in our analysis. Analysis of 15–19 year olds, for whom marriage by 1996 was exceptionally rare, yields results substantively similar to those show here. The final analysis sample is 667 children living with 437 sets of parents.¹

Measures

Parents’ Marital Quality

To measure parents’ marital quality we create two indexes based on the separate reports from mothers and fathers in 1996. Recall that all currently married respondents were interviewed and fathers and mothers were interviewed at the same time in separate places. This was done specifically to ensure that one spouse was not present and influencing the interview of the other spouse. First, consider our measure of the positive emotional bond between spouses. Fathers and mothers were each asked: “How much do you love (“maya”) your (husband/wife)? Very much, some, a little, or not at all?” The item is coded 1 for “a little” or “not at all,” 2 for “some,” and 3 for “very much” so that a higher value indicates higher parental marital affection.² In other research in this setting, with a wider range of measures of marital quality, responses to this question loaded strongly onto a factor of “positive satisfaction” (Allendorf and Ghimire 2013). Because husbands and wives were interviewed simultaneously, but separately, we treat fathers’ and mothers’ responses as independent measures of the same underlying concept—the positive dimension of marital quality—and combine the two measures by calculating the mean. We also test models with each parents’ response separately and an index that adds the responses for both parents. Results for the additive measure are even larger than those shown here and effect estimates are stronger for fathers’ than mothers’ reports. Descriptive statistics for this and other variables used in the analyses are shown in Table 1. Average love/affection between spouses is fairly high with a mean of 2 across analysis samples and about 80% of children having a mother and father who each said they love their spouse “some” or “very much”.

Table 1.

Descriptive statistics for all variables used in analyses, Chitwan Valley Family Study.

	Education analysis. N=2714 children.				Marriage analysis. N=667 children.
Variable	MIN	MAX	MEAN	STH	MIN	MAX	MEAN	STH
Parental Marriage
Positive emotional bond/”love” (3=love very much, 2=love some, 1=love a little /not at all)	1	3	2.02	0.48	1	3	2.00	0.43
Spousal violence (either parent reported being beaten by spouse)	0	1	0.18	0.38	0	1	0.17	0.38
Child’s Transition to Adulthood
Education: Ever dropped out	0	1	0.13	0.34
Ever married					0	1	0.76	0.43
Controls. Factors Known to Shape Children’s Transition to Adulthood
Caste-ethnicity
Chhetri-Bahun	0	1	0.46	0.50	0	1	0.55	0.50
Dalit	0	1	0.11	0.31	0	1	0.08	0.28
Newar	0	1	0.06	0.24	0	1	0.08	0.27
Hill Janajati	0	1	0.14	0.35	0	1	0.11	0.31
Terai Janajatia	0	1	0.22	0.42	0	1	0.17	0.38
Parents’ non-family experiences before marriage (2, 1, or no parents)
Went to school	0	2	1.15	0.77	0	2	0.87	0.72
Non-family living	0	2	0.33	0.48	0	2	0.22	0.42
Characteristics of parental marriage formation
Parents’ had arranged marriage (1=parents, 5=self; mean of both parents’ responses)	1	5	2.18	1.36	1	5	1.84	1.19
Parents’ age at marriage (mean of both parents’ responses)	6	36	18.43	2.85	9.5	25.5	17.45	2.78
Year parents married	1945	1996	1981	8.55	1950	1989	1970	6.39
Household structure in 1996
Number of grandparents in the household	0	2	0.50	0.78	0	2	0.20	0.51
Number of children in the household	1	12	4.43	1.96	1	12	5.21	2.00
Number of siblings who died	0	5	0.37	0.73	0	5	0.57	0.88
Household wealth in 1996
Household owns land house is on	0	1	0.90	0.30	0	1	0.89	0.31
Number of livestock owned	0	23	4.14	3.42	0	41	5.29	4.05
Number of consumer durables owned (e.g., radios, TVs, bicycles, plows, etc.)	0	8	2.85	1.46	0	8	2.06	1.44
Current (1995) community context
Proportion of teachers with degrees at each school (geoweighted logged)b	223.95	267.66	247.09	10.32	223.95	266.27	248.97	9.96
Respondent’s age in 1996	2	16	15.04	1.65	15	24	17.42	2.18
Female	0	1	0.50	0.50	0	1	0.49	0.50
Time in hazard	0	17	10.62	1.86	2	126	72.18	42.44

^aIncludes 7 respondents with other ethnic categorization in the models of marriage.

^bVariable is a neighborhood level, specially designed geoweighted sums that incorporate all schools in study site--see Brauner-Otto 2012 for further details.

To measure marital conflict we create a measure of extreme conflict: whether either parent reported that his/her spouse had ever beaten him/her. In the 1996 interview respondents were asked “Has your (husband/wife) ever beaten you?” with response options yes or no. For both analysis samples about 18 percent of children have parents who reported being victims of spousal violence. Of them, roughly 80% have only mothers who reported that their husbands beat them, 10% have only fathers who said their wives beat them, and 8% of children had parents who both reported experiencing physical violence. The correlation between the two indicators of parental marital quality is −0.13 and −0.10 (p < 0.001) for the two analysis samples.

Of course, marriages may have conflict that does not result in violence. We also test models that use measures of arguably more mild marital conflict: how often they have disagreements with their spouse and how often they are criticized by their spouse. Results are similar to those presented here. However, disagreements and criticism, even frequent ones, are likely to be a part of any marriage and the more relevant question is whether such disagreements are dealt with positively, in a context of affection and understanding, or whether they lead to dysfunctional outcomes such as bitterness or domestic violence. As such, we only present the models with the more extreme measure of conflict.

Some studies provide evidence that marital quality changes over time and that different dimensions of marital quality may be more or less variable (James 2014, 2015). Specifically, there appears to be an overall decrease in marital happiness/satisfaction over marriage duration, but that decrease is more rapid for those who start at a lower level (James 2014, 2015; Whiteman et al. 2007). Marital conflict has been found to be more consistent over time, even with different starting levels (James 2015; Whiteman et al. 2007). Additionally, marital quality is generally thought to be most volatile in the early years of marriage and some research provides evidence that marital quality is quite stable over time (Bradbury 1998; Johnson et al., 1992; Johnson and Booth, 1998).

Despite the potential for marital quality to change over time, previous empirical research has looked at the relationship between marital quality at one point in time and various behaviors across the life course. Most relevant for our analytic approach, studies have linked marital quality to mortality and divorce risks over the subsequent 5 to 16 years (Bulanda et al. 2016; Lawrence et al. 2018; Whiteman et al. 2007). This work is consistent with the expectation that even a single point in time measure of parental marital quality may indicate variance associated with long term consequences.

On the other hand, from a measurement error perspective, a single point in time measure is likely characterized by error that increases over time. To test the sensitivity of our results to the potential measurement error resulting from the lag between the 1996 interview and the child’s behavior we estimate models that include an interaction term between parental marital quality and time since 1996. We also conduct robustness tests that limit the length of time between the measurement of parents’ marital quality and children’s behavior. We discuss these results below.

Transition to Adulthood: Education

To investigate educational attainment we estimate hazard models of the rate of stopping school. We use the 12 years of education data in the mothers’ LHCs collected in 2008 to operationalize the yearly hazard of stopping school in discrete-time. The hazard starts in 1996 or the year the child first attended if he/she started school after 1996. We create a time-varying, dichotomous variable equal to 1 the first year the child stops attending school and 0 in years prior. Children are censored in 2008 or the year they turned 16 (when children typically finish high school). Fewer than 5% of eligible children had never attended school and are therefore excluded from the analysis sample.

The CVFS data provide many alternative measures of children’s educational attainment. The hazard of exit from schooling has strong statistical and comparative properties (Beutel and Axinn 2002), however, only 13% of children stopped attending school during the observation period. We test the robustness of our findings from this primary specification in two ways. First, we estimate models including person-years after age 16. Second, we use a different analysis sample and estimate models of whether children over age 16 in 2008 had obtained an SLC as the dependent variable: 45% of children aged 17 to 27 in 2008 had received their SLC. In both cases results are substantively identical to those shown here.

Transition to Adulthood: Marriage Timing

To investigate the marriage transition we estimate hazard models of the rate of entrance into first marriage following the 1996 interview. We use 126 months of data on marriage to operationalize the monthly hazard of marrying in discrete-time between 1996 and 2008. We create a time-varying, dichotomous variable equal to 1 the month the respondent marries and 0 in months prior. Three-quarters of the sample, 72 percent of men and 87 percent of women, marry at some point in the prospective data.

Factors that May Shape both Parental Marriage Quality and Children’s Transition to Adulthood

Though we use carefully constructed longitudinal data to make sure the temporal order of our measures are consistent with our hypotheses, prior factors may produce a spurious association between measured parental marital quality and children’s behavior. The breadth of the CVFS allows us to include setting-specific measures of these factors that capture experiences and characteristics which occurred before parental marital quality and children’s behaviors were measured. Unless otherwise stated the measure is time invariant, is a mean of the mother’s and father’s reports, and measured at the parent level.

We differentiate among the five groups representing major caste-ethnic and religious differences in our study site in Nepal: Chhetri-Bahun (high-caste Hindus), Dalit (low-caste Hindus), Newar, Hill Janajati (Tibeto Burmese descendants of Hill residents), and Terai Janajati (indigenous to northern India and the southern part of Nepal). Caste-ethnic group is the same for all household members.

We create categorical measures for parents’ education and living away from family equal to 2 if both parents had the experience before marriage (measured in 1996), 1 if one parent did, and 0 if neither parent had the experience. For 92% of the children when only one parent went to school it was the father.

We create three measures of parents’ experiences with marriage: parental age at marriage, marital arrangement, and the year the parents’ married. On average, fathers are almost 5 years older than mothers but, consistent with what we know about assortative mating, mothers’ and fathers’ ages are correlated (Mare 1991). Following previous research, we create a scale from 1 to 5 to indicate the level of participation in spouse choice from having no choice of spouse (1) to having complete choice (5) (Ghimire et al. 2006; Jennings, Axinn, and Ghimire 2012). As expected given previous research on marital arrangement (Ghimire et al. 2006; Ghimire et al. 2015), fathers reported more involvement in the choice of their spouse than mothers.

Next we include household level measures of household characteristics including member composition in 1996 (the number of grandparents, number of siblings, and number of siblings who had died) and the quality of the parental home environment operationalized in terms of household affluence in 1996. In rural Nepal incomes are low and, because many farming related assets create affluence that increases the comfort of the parental home, we measure ownership of specific assets rather than income: a dichotomous measure for whether the household owned the land on which their house sits and counts of the number of livestock and consumer durables the household owned. Ownership of the home, livestock (which supply food) and consumer durables (radios, TVs, bicycles, etc.) each enhance the comfort of the parental home and are not dependent on one another (e.g. those who do not own their home may have significant numbers of livestock).³

To measure community context we focus on schools as they are the most salient dimension of community context these young people navigate during the years they are in our analyses. We include a neighborhood-level, geographically-weighted measure of school quality that incorporates all of the schools open in 1995 (see Brauner-Otto 2012; Brauner-Otto, Axinn, and Ghimire 2007 for more on geographic-specific measures of community context in this setting). The specific school characteristic we control for, the proportion of teachers with advanced degrees, was identified in the literature as important in keeping students, particularly girls, enrolled in school longer (Card and Krueger 1992; Lloyd, Mensch, and Clark 2000; Mensch and Lloyd 1988) and found to be related to family formation behaviors (Brauner-Otto 2012). This is a neighborhood level measure.

Finally, in all our models we include measures of children’s gender and year of birth. To account for the duration of the exposure to dropping out of school we control for years since started school, years squared, and years cubed. To account for exposure to marriage risk we use a measure of months since the first monthly interview and months squared.

In addition to the above mentioned controls we examine many other theoretically motivated measures.⁴ However, as these measures are not statistically significant and including them does not change our estimated relationship between parental marital quality and children’s behaviors, for parsimony we do not include them in the analyses shown.

Analytic Strategy

To investigate children’s behavior we treat education and marriage as transitions occurring over time, from being in school to stopping school and from being never married to marrying for the first time. We use event-history techniques to estimate these discrete-time hazard models (Allison 1982). Because the outcomes in question has only one destination state and are measured as dichotomous variables, logistic regression is an appropriate estimation technique (Allison 1982). Person-years of exposure are the unit of analysis for education and person-months for marriage. We start the hazard of stopping school in 1996 or the first year a student is enrolled if they have not yet started in 1996 and the hazard of marriage the first month of the prospective data collection. Individuals are clustered within households and within neighborhoods so we estimate three-level hazard models.⁵ In the CVFS data neighborhood clustering generally accounts for very little of the overall variance and in these models the neighborhood Interclass Correlation Coefficient (ICC) was 1% or lower.

We investigate potential gender differences by also estimating separate models for sons and daughters and pooled son-daughter models with interaction terms. We do not find any significant gender differences in our key independent variables so we present the pooled models.

Results

Education

Table 2 shows models of the relationship between parents’ marriage quality and children’s education (stop attending school). The hazard model coefficients displayed are the multiplicative effects on the odds of marriage—the odds ratios. We start with the positive dimension of the parental marriage. Looking at our most parsimonious model (Model 1) we see that parental emotional bond, “love,” has a negative relationship with the hazard of stopping school. Children whose parents reported loving each other more were less likely to stop attending school, i.e. had higher educational attainment, than children whose parents reported less love. In this model we also see that older children were more likely to stop school and there was no difference in the hazard for boys and girls.

Table 2.

Hazard models of dropping out of school for 2714 children. Three level (neighborhood, household, individual) discrete time hazard models estimated with logistic regression.

	1	2	3	4	5
Parental Marriage
Positive emotional bond/”love” (3=love very much, 2=love some, 1=love a little/not at all)	0.53***	0.69*			0.71*
	(−3.24)	(−2.16)			(−1.99)
Spousal violence (either parent reported being beaten by spouse)			2.66***	1.47*	1.42*
			(4.43)	(2.10)	(1.92)
Controls
Ethnicity (reference group: Chhetri-Bahun)
Dalit		4.88***		4.56***	4.59***
		(6.07)		(5.80)	(5.84)
Newar		0.85		0.78	0.81
		(−0.35)		(−0.53)	(−0.46)
Hill Janajati		2.21**		2.15**	2.16**
		(3.03)		(2.93)	(2.96)
Terai Janajati		2.42***		2.45***	2.39***
		(3.59)		(3.65)	(3.57)
Parents’ non-family experiences before marriage (2, 1, no parents experienced before marriage)
Went to school		0.54***		0.54***	0.54***
		(−4.76)		(−4.75)	(−4.71)
Non-family living		0.82		0.80	0.82
		(−1.09)		(−1.24)	(−1.09)
Characteristics of parental marriage formation
Parents’ had arranged marriage (1=parents, 5=self; mean of both parents’ responses)		1.09		1.10	1.09
		(1.53)		(1.58)	(1.47)
Parents’ age at marriage (mean of both parents’ responses)		0.97		0.98	0.98
		(−0.84)		(−0.59)	(−0.75)
Year parents married		1.03*		1.03*	1.03*
		(1.80)		(1.79)	(1.82)
Household structure in 1996
Number of grandparents in the household		0.84		0.86	0.85
		(−1.38)		(−1.21)	(−1.29)
Number of children in the household		1.16**		1.16**	1.16**
		(2.77)		(2.79)	(2.73)
Number of siblings who died		0.94		0.94	0.94
		(−0.57)		(−0.56)	(−0.55)
Household wealth in 1996
Household owns land house is on		0.72		0.71	0.70
		(−1.37)		(−1.47)	(−1.49)
Number of livestock owned		0.98		0.98	0.98
		(−0.83)		(−0.62)	(−0.69)
Number of consumer durables owned (e.g., radios, TVs, bicycles, plows, etc.)		0.64***		0.64***	0.64***
		(−6.47)		(−6.53)	(−6.41)
Current (1995) community context
Proportion of teachers with degrees at each school (geoweighted logged)		0.99		0.99	0.99
		(−1.17)		(−0.92)	(−1.11)
Female	1.00	1.01	1.00	1.02	1.01
	(−0.03)	(0.08)	(−0.03)	(0.12)	(0.08)
Year child was born	0.93***	0.93***	0.93***	0.93***	0.93***
	(−3.80)	(−3.20)	(−3.71)	(−3.20)	(−3.24)
Years in school	1.73**	1.77**	1.73**	1.77**	1.77**
	(2.47)	(2.58)	(2.45)	(2.58)	(2.57)
Years in school squared	0.98	0.97	0.98	0.97	0.97
	(−0.64)	(−0.96)	(−0.66)	(−1.00)	(−0.96)
Years in school cubed	1.00	1.00	1.00	1.00	1.00
	(0.49)	(1.05)	(0.54)	(1.12)	(1.07)
Neighborhood-level random effect estimate	0.95	0.12	0.91	0.12	0.11
standard error	0.29	0.12	0.29	0.13	0.12
Household-level random effect	2.90	1.38	2.90	1.36	1.37
standard error	0.56	0.33	0.56	0.33	0.33

Note: Table shows odds multipliers with asymptotic z-ratios in parentheses. All models also include a constant term. N=20618 person-years, 151 neighborhoods, 1133 households.

^*p<0.05

^**p<0.01

^***p<0.001 one-tailed tests

In Model 2 we add in our measures of factors known to influence parental marital quality and children’s outcomes. The relationship between parental marital quality and children’s education is robust to the inclusion of these additional controls for the ethnicity, parents’ experiences before birth, their marriage, household characteristics, and the current educational context. As expected, many of these factors are related to children’s education themselves. Chhetri-Bahun, the ethnic group which has historically held the highest social status and privilege, had a lower hazard of stopping school than Dalits, Hill Janajati, or Terai Janajati. Additionally, we find that children whose parents had gone to school and who lived in wealthier households (as measured by consumer durables) had a lower hazard of stopping school. Children with more siblings had a higher hazard.

In Model 3 we turn to our measure of marital conflict and find it is positively related to the hazard of stopping school and that relationship is robust to controlling for other factors (Model 4). Finally, in Model 5, we include both measures of parental marital quality and find that the relationship between parental love and children’s education is independent of the amount of marital conflict the parents report.

One potential limitation is that these measures of marital quality refer to one time period and that it is possible, and in fact probable, that the parental marriage changes over time, making these measures less accurate as time passes. This potential is exacerbated by looking over a 12 year period. Models with an interaction term between parental love and years since 1996 provide some evidence of this: the sign of this interaction term is consistent with the idea that the 1996 measure of marital quality is less reflective of the marriage as time goes on (i.e. that measurement error increases over time). But, the interaction term itself was not statistically significant. Of course, this finding is also consistent with other hypotheses (for example, a diminishing cumulative effect model).⁶

Marriage Timing.

Next we turn to our analysis of marriage behavior (Table 3). Model 1 is the most basic model and shows that high parental affection was significantly related to children’s marriage timing, delaying the transition to marriage. The more positive the relationship between the parents the later their children married.

Table 3.

Hazard of marriage timing for 667 men and women aged 15–24 in 1996 in Chitwan, Nepal. Three-level hierarchical logistic regression models (neighborhood, household, person).

	1	2	3	4	5
Parental Marriage
Positive emotional bond/”love” (3=love very much, 2=love some, 1=love a little/not at all)	0.78*	0.78*			0.81*
	(−2.09)	(−2.06)			(−1.80)
Spousal violence (either parent reported being beaten by spouse)			1.33*	1.38**	1.34*
			(2.16)	(2.45)	(2.21)
Factors Known to Shape Children’s Marriage Timing
Ethnicity (reference group: Chhetri-Bahun)
Dalit		1.04		1.01	1.01
		(0.21)		(0.05)	(0.05)
Newar		0.86		0.84	0.85
		(−0.72)		(−0.87)	(−0.83)
Hill Janajati		1.09		1.11	1.08
		(0.47)		(0.57)	(0.42)
Terai Janajatia		0.72*		0.71*	0.71*
		(−2.07)		(−2.21)	(−2.24)
Parents’ non-family experiences before marriage (2,1,no parents experienced before marriage)
Went to school		0.82**		0.81**	0.82**
		(−2.40)		(−2.62)	(−2.48)
Non-family living		0.83		0.82	0.82
		(−1.34)		(−1.49)	(−1.40)
Characteristics of parental marriage formation
Parents’ had arranged marriage (1=parents, 5=self; mean of both parents’ responses)		0.99		0.99	1.00
		(−0.11)		(−0.12)	(−0.10)
Parents’ age at marriage (mean of both parents’ responses)		0.99		0.99	0.99
		(−0.46)		(−0.31)	(−0.38)
Year parents married		1.01		1.01	1.01
		(0.77)		(1.03)	(0.92)
Household structure in 1996
Number of grandparents in the household		0.97		0.95	0.97
		(−0.32)		(−0.51)	(−0.33)
Number of children in the household		0.97		0.98	0.97
		(−0.84)		(−0.64)	(−0.79)
Number of siblings who died		0.98		0.97	0.97
		(−0.29)		(−0.43)	(−0.39)
Household wealth in 1996
Household owns land house is on		1.15		1.19	1.18
		(0.75)		(0.96)	(0.92)
Number of livestock owned		1.00		1.00	1.00
		(−0.11)		(−0.01)	(0.04)
Number of consumer durables owned (e.g., radios, TVs, bicycles, plows, etc.)		0.97		0.98	0.98
		(−0.65)		(−0.49)	(−0.44)
Current (1995) community context
Proportion of teachers with degrees at each school (geoweighted logged)		0.98**		0.99**	0.98**
		(−2.88)		(−2.70)	(−2.88)
Female	1.80***	1.81***	1.78***	1.79***	1.80***
	(5.34)	(5.32)	(5.32)	(5.26)	(5.30)
Year respondent born	0.91***	0.90***	0.91***	0.91***	0.91***
	(−4.04)	(−4.10)	(−4.01)	(−4.04)	(−3.94)
Time in hazard	1.02***	1.02***	1.02***	1.02***	1.02***
	(3.20)	(3.31)	(3.13)	(3.27)	(3.30)
Time in hazard squared	1.00*	1.00*	1.00*	1.00*	1.00*
	(−1.75)	(−1.79)	(−1.75)	(−1.81)	(−1.82)
Neighborhood-level random effect estimate	0.05	0.00	0.05	0.00	0.00
standard error	0.05	0.00	0.05	0.05	0.05
Household-level random effect	0.13	0.15	0.08	0.10	0.11
standard error	0.11	0.10	0.11	0.11	0.11

Note: Table shows odds multipliers with asymptotic z-ratios in parentheses. All models also include a constant term. N=48144 person-months, 141 neighborhoods, 432 households.

^aIncludes 7 respondents with other ethnic categorization in the models of marriage.

^*p<0.05

^**p<0.01

^***p<0.001 one-tailed tests

In Model 2 we add measures of factors previous research has found to be related to children’s marriage timing. We find that, even as some of these factors were related to children’s marriage timing (in particular parents’ education and the current educational context), the relationship between parents’ emotional bond and children’s marriage timing remains unchanged.

Gender is significantly related to the hazard of marriage with daughters marrying earlier than sons, but we find no significant differences in the relationship between parents’ emotional bond and children’s marriage timing by gender. This finding is entirely consistent with previous research on intergenerational influences on marriage timing, which in spite of strong gender differences in the overall pace of marriage, almost never identifies significant gender differences in associations between parents’ characteristics and the pace of marriage across many settings (Marini 1978; Thornton and Lin 1994; Thornton et al. 2007).

Next we examine the relationship between negative dimensions of parents’ marital quality and their children’s marriage timing (Model 3 showing the base model and Model 4 including additional measures). As we predict in this setting, parental marital conflict sped-up entry into marriage and this relationship is independent of the other factors known to influence children’s marriage timing. Finally, in Model 5 we include measures of both dimensions of the parental marriage quality and see that they maintain their strong, independent effects.

Conclusion

Decades of empirical research provide evidence that children’s transition to adulthood is influenced by their parents’ experiences (e.g. McLanahan and Bumpass 1988; Thornton et al. 2007) and that the emotional quality of a marriage has wide reaching consequences (Cherlin et al. 1995; Malinen et al. 2010). Yet we know little about how this emotional dimension of the parental home influences children’s transition behaviors. Though this is an important theoretical question, rarely do we have the empirical tools to investigate it. Because the CVFS includes measures of the emotional dimensions of parents’ marital quality in an intergenerational panel study covering more than 10 years we are able to investigate the association between variations in emotional dimensions of parents’ marriage and subsequent variations in their children’s education and the timing of their children’s marriages.

The relationship between mothers and fathers has a substantial, enduring association with the pace at which their children subsequently leave school and marry. Parents’ marital quality is multidimensional, encompassing both affection/love and conflict/violence as independent dimensions of the relationship. We find that children with parents who reported more affection/love for one another stayed in school longer and married later, whereas those whose parents reported spousal violence left school earlier and married earlier (i.e., left the parental home earlier). These findings, based on simple measures of marital quality, are what we expect based our setting-specific framework and on the literature documenting distinct dimensions of marital quality (Allendorf and Ghimire 2013; Bradbury, Fincham and Beach 2000; Umberson et al. 2005). Couples with a more positive emotional bond may be supporting, encouraging, and enabling of their children’s education because they are investing more in the “quality” of their children rather than “quantity.” Regarding marriage timing, these findings support multiple possible mechanisms including the possibility that parents’ love for one another may spill over to their children in such a way that the family spends more time looking for a spouse, perhaps because they have higher expectations, or that the love between the parents creates a comforting home environment that young people are reticent to leave.

Parental marital conflict may be related to children’s transition to adulthood behaviors because conflict between parents lowers the quality of the parental home increasing young peoples’ motivations to leave home. Because marriage is virtually the only acceptable long-term way for young people to leave the parental home in this setting, parental conflict, then, speeds up leaving school and entry into marriage.

Note that we find these strong relationships even though our measures of parental marital quality are simplistic. Although multi-item, repeated measurement over time might be desirable, this study is an important contribution because we have independent measurement from each parent (as opposed to children’s reports of their parents’ marriage) and these single items still reveal important insights. Research using more thorough measures of marital quality is crucial as it will likely yield evidence of even stronger long-term, intergenerational influences and will certainly reveal more information on the nature of these relationships.

Of course, the social context of our investigation is specific. Divorce remains rare in Nepal, particularly in the parental generation, so that parental marriages remain intact even when marital quality may be quite low. Independent living and premarital cohabitation are also rare, so marriage is a child’s primary long-term opportunity to leave their parental home. These setting-specific differences produce different expectations for the consequence of variations in parental marital quality. This study adds to the existing literature by emphasizing the importance of setting specific studies and hypotheses. At the same time, our study also provides evidence of the long-lasting influence parental relationships have on their children across a wide range of settings.

Documenting the intriguing finding that the response to a specific family/parental condition depends on the social context raises many important research questions. High priority among these is whether the different findings are due to different mechanisms or a result of different socially acceptable behavioral options available to individuals. Future research that can investigate the specific mechanisms across dramatically different settings is necessary to separate these possibilities.

The wide-ranging, intergenerational consequences of the emotional dimensions of parental relationships go beyond intriguing. Across the globe, including in South Asia and many other highly populated regions, we are witnessing dramatic transitions in marriage. Age at marriage has increased throughout South Asia (Yeung, Desai, and Jones 2018). There is widespread change from “arranged” marriage toward “love” marriage along with hybrid marital arrangements where both parents and children participate in spouse choice (Allendorf 2013, 2017; Allendorf and Pandian 2016; Ji 2013; Netting 2010; Rindfuss and Morgan 1983). And, divorce is rapidly becoming more common (Jennings 2016). Together these factors are changing the marital context within which childrearing takes place. Our findings are consistent with the conclusion that these changes in the nature of parental marriages are likely to have long-term consequences for their children across multiple dimensions of social life.

At the same time, there is still considerable variation in the institution of marriage within South Asia. Differences in economic and educational contexts as well as important cultural legacies may explain some of this variation (Desai and Andrist 2010; Yeung et al. 2018). The findings presented here demonstrate that research on the connection between parents’ marital relationships and their children is a high priority. This may help us better understand the consequences of the variability in marriage in this region.

The strong associations with subsequent education and marriage documented here provide evidence that it is also possible that the emotional dynamics within parental marriages may influence their children in other domains of social life including work, family, recreation, and ideation. Thus, we argue the findings described here should motivate a broad range of research into the emotional basis of behavior and the high potential for intergenerational consequences for multiple dimensions of children’s lives, potentially shaping all aspects of their health and wellbeing.