For Better or Worse: The Health Implications of Marriage Separation Due to Migration in Rural China

Abstract

Massive rural-to-urban migration in China has led to spatial separation of millions of married couples. In this article, we examine the question of whether the well-documented health benefits of marriage extend to left-behind individuals in rural China who are spatially separated from their spouses. Using longitudinal data that span 16 years (China Health and Nutrition Survey 1991, 1993, 1997, 2000, 2004, 2006), we compare the self-reported health trajectories of adults across different marital statuses while taking into account the physical location of their spouses. Our results suggest a clear health disadvantage of married individuals whose spouses are absent compared with those whose spouses are living in the same household. Further, longer spousal absence is more harmful to an individual’s health. Finally, spousal absence and longer physical separation from their spouses induce stronger health deficits for married men than for married women, suggesting that a gendered process is at work.

Introduction

Research has consistently documented that marriage is positively associated with health and well-being and that marital dissolution may lead to a health decline (Hughes and Waite 2009; Liu and Umberson 2008; Waite and Gallagher 2000; Williams and Umberson 2004). Notwithstanding selection mechanisms, this health advantage occurs primarily because marriage provides additional economic resources, enhances social integration, and promotes psychological well-being and a healthier lifestyle (Waite and Gallagher 2000). What remains an open question, however, is whether the presumed benefits of marriage prevail when one spouse does not live at home or is frequently absent. Do those who are married with spouses absent experience any disadvantages in health relative to their married counterparts whose spouses are living at home?

These research questions are rarely asked in the United States, partly because spousal absence is typically temporary (see Fuller 2010). Although this is an unconventional family type in the United States, it is a quintessential way of life for millions of married couples in China. Deemed “a new demographic phenomenon in China” (Liang and Ma 2004:467), massive rural-to-urban migration began in the 1980s and has resulted in a remarkable growth of a “floating population.” These internal “temporary” migrants include an estimated 150 million people, which is more than one-tenth of China’s total population (Chan and Zhang 1999; Fan 2003). For married rural couples, it is common for one spouse (more often the husband than wife) to seek employment in urban areas while the other stays behind to tend to agricultural work and family needs (Pan et al. 2012; Zuo 2007).

Does marriage separation due to migration have any impact on married individuals’ health? This question has been partially addressed in the migration and health literature, which has predominantly focused on migrant health, but pays little attention to the health of spouses remaining at home (left behind) (see review by Lu 2012). Our paper fills an important gap in the literature and crosscuts the research on marriage and migration and their links to health by focusing on the dimension of physical health, as measured by self-reported health. The comparisons of married individuals who live with their spouse and with those who live without their spouses, as well as comparisons with unmarried individuals, help shed light on the various pathways through which marriage and spousal separation influence health. Using longitudinal data that span 16 years (China Health and Nutrition Survey 1991, 1993, 1997, 2000, 2004, 2006), we examined the differences in physical health trajectories by marital status and spousal separation in rural China, and subsequently determined whether the duration of spousal separation affected the physical health of the left-behind spouse. Given the long-standing observations on gender differences in marriage, migration, and health linkages, we paid close attention to potential gender differences in these processes.

Marriage and Migration in Rural China

China offers a compelling context in which to study the health implications of marriage and physical separation from the spouse. Marriage patterns in China were characterized as “near universal,” according to the 1982 census (Zeng et al. 1985). Since the 1980s, China’s marriage rate has declined, and its divorce rate has steadily increased (Xu and Ye 2002; Ye and Lin 1998; Zeng and Wu 2000). However, despite this increase in China’s divorce rate in recent years (from 0.85 per 1,000 married persons in 1979 to 2.62 per 1,000 married persons in 2007; see Wang and Zhou 2010), it is still considerably lower than in other developed countries, such as the United States (e.g., 19 per 1,000 married persons in 2010; see Cruz 2013). The universal marriage culture remains strong in rural China. The 2010 Chinese census estimated that marriage rates for rural men and women aged 35–39 were as high as 97.2 % and 99.5 %, respectively (Liu et al. 2014).

At the same time, marriages with one spouse living apart from his or her spouse, an uncommon living arrangement in the United States, has been rapidly growing in China. Since the 1980s, economic reform in China has brought out an astounding tidal wave of rural-to-urban migration that resulted from an agricultural labor surplus and rising employment opportunities in the cities. A recent report from the National Bureau of Statistics of China estimated that about269 million migrants have moved from rural to urban areas in 2013 (National Bureau of Statistics of China 2014). Living costs in the cities are often prohibitive for migrant workers to support a family there. Furthermore, with a strict migration policy imposed by the state government and the hukou (household registration) system in place, it is often difficult for family members to join migrants in the destination cities. Although no national prevalence estimates for this phenomenon are available, a recent study in Beijing (Fan et al. 2011) suggested that around one-half of the married migrants did not live with their spouses and children in 2009. Migrants themselves are treated as second-class citizens and are excluded from many of the social welfare benefits to which urban residents are entitled (e.g., education for school-age children, health care) (Pan et al. 2012; Solinger 1999). In addition, the land policy in China also ties those with rural hukou to rural areas and limits their mobility to the cities. Under these circumstances, temporary migration often turns out to be a prolonged family separation.

Migration and health researchers often argue that family disruption due to migration has deleterious effects on migrants’ physical health (Hu et al. 2008; Lu 2010). Studies in developing countries also have investigated the negative consequences of out-migration on children’s well-being and the ways it may undermine intergenerational support to aging parents (Gao et al. 2010; Guo et al. 2009; Morooka and Liang 2009; Schmeer 2009). However, the issue of how out-migration affects the well-being of the left-behind spouse has received virtually no attention (see Lu 2012). In our study, we take a significant step forward in bridging the literature of migration, marriage, and health to enhance our scientific understanding about the implications of spatial marriage separation due to migration on the physical health of the left-behind spouse.

Spatial Marriage Separation and Physical Health

Some studies in the Chinese aging literature have documented the health benefits of living with a spouse and the adverse effects of transition to widowhood for older adults (Korinek et al. 2011; Li et al. 2009; Yi et al. 2002). Because research on the links between marriage and health in the general population in China is sparse, our literature review focuses heavily on U.S.-based studies. We apply the U.S.-based theoretical models of marital resource and stress processes to build our research hypotheses, while paying close attention to the context of rural China.

Empirical research on the relationship between marriage and physical health in the United States has consistently demonstrated that being married is positively associated with a range of physical health outcomes, including self-reported physical health, chronic conditions, activity limitations, and mortality (Hughes and Waite 2009; Waite and Gallagher 2000). Despite the possibility of selection of healthier people into marriage, recent studies using longitudinal data have generally concluded that even after a number of selection factors are controlled for, married individuals still have significantly better physical health than the unmarried, and those who experience marital dissolution have worse physical health trajectories than the continually married (Williams and Umberson 2004; Wu and Hart 2002).

Moreover, the literature increasingly recognizes heterogeneity within groups of unmarried and married individuals. Advances in the literature not only distinguish those who are unmarried as divorced, separated, widowed, and never married but also highlight cohabitation as a separate union status (Musick and Bumpass 2012; Wu and Hart 2002; Zimmerman and Easterlin 2006). Within the married group, recent studies have documented that the benefits of marriage are far from universal and are largely contingent on the length and quality of the marriage (Hawkins and Booth 2005; Williams 2003).

Our central focus on marital status and spousal absence is a further push on the issue of heterogeneity within the married group—a phenomenon that is rarely studied because it is not a common living arrangement in the United States. One exception is a U.S. study by Fuller (2010). Using data from the National Health Interview Survey (NHIS), he found that married individuals who do not live with their spouses (so-called commuters) reported similar health statuses yet had poorer health behaviors than married individuals who lived with their spouses. However, this study design of using cross-sectional data did not offer much information on the history of physical separation between commuter couples, nor did it take into account how certain married couples are selected into the commuter relationship.

Despite the paucity of relevant empirical evidence, theories regarding the process through which marriage presumably promotes physical health provide a conceptual foundation to expect variation in health for those who are married but living separately from a spouse, those who are married and living with a spouse, and those who are unmarried in the context of rural China.

The Marital Resource Model

One of the most recognized theories explaining the association between marriage and physical health in the Western literature is the marital resource model, which argues that marriage provides economic, social, and psychological resources that consequently enhance physical health (Waite and Gallagher 2000). First, marriage leads to an increase in economic resources through specialization, economies of scale, and the pooling of wealth (Becker 1991), which are known to influence various proximate determinants of physical health, including nutrition, access to health care, and health behaviors (Waite and Gallagher 2000). Because married couples who live apart do not share a residence, they do not have access to the benefit of economies of scale (e.g., sharing bills), a situation that is similar for unmarried individuals. However, earnings sent home by the migrant spouse could boost available family income for the left-behind spouse. In this case, similar to the married person whose spouse lives at home, married people whose spouses migrate may still enjoy the increased economic resources from pooling income. In this sense, they should have greater access to economic resources (and thus better health) than the unmarried.

Second, marriage is associated with increasing access to social and emotional support and a sense of personal control and responsibility, which subsequently lead to better mental health and healthy behaviors and thus promote physical health (Liu and Umberson 2008; Waite and Gallagher 2000). The married person whose spouse lives away from home may not have full access to the social and psychological resources provided by marriage; physical separation may reduce couples’ sexual intimacy and emotional closeness, which may, in turn, undermine marital quality and reduce the levels of social support from the long-distance spouse in comparison with married couples who live together. Marital quality and social support are both key factors linking to physical health (Waite and Gallagher 2000; Williams 2003). Therefore, we expect that because of their limited access to resources normally conferred to married couples, those married with spouses absent will experience worse physical health trajectories than their married counterparts whose spouses live at.

On the other hand, some researchers have argued that the benefits of marriage come from its institutionalized nature (e.g., Waite and Gallagher 2000). In this sense, compared with unmarried persons, the married person whose spouse lives away from home may still have somewhat greater access to some social and psychological resources (e.g., emotional support) as long as the “long-term contract” of marriage continues to be honored. For example, in the U.S. literature, researchers have suggested that spatial separation does not significantly reduce the physical health advantages of married persons relative to unmarried persons (e.g., cohabiting couples) because of the enhanced level of social support experienced by married couples, the long-term commitment of marriage, and its provision of access to other social institutions (Waite and Gallagher 2000).

Thus, similar to the processes of marriage in the United States, we expect that marriage in rural China would also provide additional resources that are inaccessible to the unmarried. Similar to the United States, the importance of marriage as a central institution in Chinese society has been well recognized (see review by Riley 1994). Historically, a marriage is not simply a conjugal relationship between two persons; it is also a family-to-family contract, placing far more emphasis on family obligations than individual happiness (Xu and Whyte 1990). Indeed, migration of rural-to-urban laborers is often seen as a “family-based strategy mediated by an implicit contract,” and these migrants continue to maintain strong ties with the family (Murphy 2007:13). Therefore, it is reasonable to expect that those married with absent spouses would still experience better physical health trajectories than unmarried persons because of their greater access to marital resources (although perhaps to a lesser extent than those who are married and living with their spouse) in rural China.

On the other hand, the conjugal bond has historically been considered less important than the intergenerational ties between parents and children in the Chinese kinship system, with little emphasis on intimacy and affection (see the review by Pimentel 2000). In this sense, marriages may not provide as much emotional support in rural China as in Western society. In addition, because the divorce rate is low and culturally stigmatized in rural China, those who are married are potentially a more heterogeneous group. Because the literature has documented that marriages of poorer quality lead to worse health, presumably those who are married in rural China make up a higher proportion of those who are in unhappy marriages but stay in the marriage because of strong cultural sanction against divorce. Thus the psychological costs of marriage could outweigh the benefits of additional resources. Taken together, the benefits associated with marital resources may be less salient in the context of rural China.

The Stress Model

Another branch of U.S. literature on marriage and health does not focus on the benefits of marriage per se, but rather on the stress processes of marital dissolution. According to the stress process model, the process of marital dissolution creates many stresses (e.g., economic stress, loss of a confidant) for both men and women, which have psychological and social consequences that harm physical health (Williams and Umberson 2004). Stress may affect physical health either by indirectly undermining psychological well-being and promoting unhealthy behaviors (e.g., smoking, drinking) or by directly stimulating the production of stress hormones and evoking physiological responses (Kiecolt-Glaser and Newton 2001; Umberson 1987). This line of research contends that marital dissolution may lead to a decline in physical health because of increased stress during the process of marital dissolution (Williams and Umberson 2004).

Although spatial marriage separation due to migration is distinctly different from divorce or widowhood, it is reasonable to expect that this kind of separation could potentially strain a marriage. The stress model suggests that, similar to divorce and widowhood (although to a lesser extent), physical separation from the spouse could induce stress and thus negatively affect physical health (Williams and Umberson 2004). Indeed, a recent study of left-behind rural wives in Hunan province in China found that they reported high levels of depression, stress, and lower quality of life compared with non-left-behind wives (Yi et al. 2014). At the same time, stress related to unmarried status may be even greater in rural China than in the United States, and thus those who are married but not living with their spouses may still have a health advantage compared with the unmarried group. Given that marriage is still nearly universal in China, social pressure to get and stay married is tremendous. The prevalent use of derogatory terms such as “leftover” women (Sheng Nu) and “bare branches” (Guang Gun) reflects the cultural humiliation associated with an unmarried status. Thus, the negative health implication associated with being unmarried could be greater in rural China than in the United States.

Taken together, the marital resource model and the stress processes model as well as information on the context of rural China lead us to expect the following:

Hypothesis 1: Those who are married but live apart from their spouses will have worse physical health compared with those who are married and live with their spouses.

Hypothesis 2: Those who are married but live apart from their spouses will have better physical health compared with the unmarried, including those who are divorced, widowed, or never married.

Does Duration of Marital Separation Matter?

Both the marital resource model and the stress process model indicate the importance of marital status duration in deciding health impact, although they point to different predictions with regard to the direction of the effect of marital separation duration. According to the marital resource model, the benefits of marriage are instrumental, and day-to-day interaction is essential for the promotion of healthy behavior and psychological well-being. Access and lack of access to marital resources may have long-term persistent or cumulative consequences on individuals’ health (Waite and Gallagher 2000). If this is the case, temporary separation may not have ill effects, but a longer duration of migratory separation may take a toll on the health of the left-behind spouse. In addition, those who are left behind in rural areas face a heavy burden in attending to both agricultural fieldwork and caregiving responsibilities for the left-behind children and elderly by themselves, which could lead to a role overload and potentially be harmful to their health. Previous research has suggested that some migrant workers in China eventually started new families in their destination area and abandoned their families of origin, despite the absence of a formal divorce (Yang et al. 2009). Thus, the long-term absence of a spouse could resemble marital dissolution and have negative consequences for health trajectories, whereas a short-term absence may not carry such consequences. Therefore, we predict the following:

Hypothesis 3a: Long-term separation from the spouse takes a heavier toll on physical health than short-term separation.

In contrast, the stress model suggests that the process of marital separation and dissolution is stressful and may lead to a transitory change in life context, which may temporarily hurt physical health (Williams and Umberson 2004). As time passes, though, both the migrant and left-behind spouse would be more settled and would adjust to long-distance married life. The stress model thus suggests that spousal separation may have temporary negative effects on physical health that would decrease with longer duration of spousal separation. Therefore, we propose a competing hypothesis about the effect of marital separation duration:

Hypothesis 3b: Long-term separation from the spouse takes a lesser toll on physical health than short-term separation.

Does Gender Matter?

Gender has been a key focus of the marriage, migration, and health literature. Among rural-to-urban migrants in China, the percentage of married males is much higher than that of females, although there are tremendous regional differences (Wang 2000). Married women’s migration abilities are often constrained by familial roles and domestic activities. Since the 1980s, accompanying the trend of rural-to-urban migration, the feminization of agriculture has also become increasingly common (Jacka 1997; Yang 2000). Rural women often have to shoulder the so-called triple burden: heavy toil of agricultural labor, housework, and caring for children and aging parents and parents-in-law, which could induce both physical and psychological strain and damage women’s health. For married women who migrate alone for economic gain, because of highly gendered social roles, norms, and attitudes, their husbands could be considered failures for not doing “their part to provide adequate support and protection for their families” (Yang and Guo 1999:947). The social stigma of the left-behind status for men thus could be damaging to their health.

Previous research in the United States has also highlighted major gender differences in the relationship between marriage and health (Simon 2002; Umberson 1992; William and Umberson 2004). A long-time sociological tenet is that marriage promotes men’s health more than women’s health because wives are more likely than husbands to maintain social connections, provide social support, and regulate their spouses’ health behaviors (Umberson 1987, 1992). Many empirical studies have confirmed this view and have documented that the gaps between the unmarried and married groups in terms of mental and physical health as well as mortality are greater for men than for women and that marital dissolution affects men more adversely than women (Hughes and Waite 2009; Liu 2009; Williams 2003; Williams and Umberson 2004). However, other studies have suggested either no gender differences or a stronger marital effect on women’s health than on men’s health. Indeed, recent studies have suggested that gender differences in health benefits through marriage may have diminished over time because the meanings and implications of gender and marriage have changed substantially (Liu and Umberson 2008).

Given the highly salient gender expectations and gendered normative structure in China (see Hershatter 2000), we expect men to reap more physical health benefits from marriage due to gendered family roles within the marriage. Similar to marriages in the United States, wives in China usually provide emotional support to their husbands and regulate their husbands’ health behaviors, whereas husbands are less likely to do so for their wives. Therefore, in the case of spousal absence, men’s lives (and physical health) may be more profoundly affected by physical separation from their spouses than would the lives of the women. Moreover, left-behind wives may have more to gain economically from their spouses’ migration than would left-behind husbands. Migrant occupations are highly segmented by gender, with men mainly engaged in the industry or service sectors, and women were engaged in the service and retailing sectors; consequently, women may generally receive poorer pay than men (Jacka and Geotano 2004). The increased economic resources from remittances may buffer some negative effects of physical separation on physical health more so for women than for men.

At the same time, we recognize that marriage in rural China differs from marriage in Western culture. The traditional patriarchal family system remains strong in rural China. The Confucian doctrines prescribe a domination of husbands over wives and elders over the younger generations, thus potentially making marriage oppressive rather than protective for some rural women. Some scholars documented that young married women in rural China may have lower status and a lower level of social support than unmarried women (e.g., Zhang 2010). In that sense, spousal absence may not render much negative health implications for women.

Taking all these findings together, we expect the following:

Hypothesis 4: Spousal absence is more detrimental to men’s physical health than to women’s.

Data and Sample

We used data from the China Health and Nutrition Survey (CHNS) for our analyses. The CHNS is a longitudinal survey collected by the Carolina Population Center at the University of North Carolina, the Institute of Nutrition and Food Hygiene, and the Chinese Academy of Preventive Medicine in Beijing. The CHNS was designed to examine the influences of social and economic transformations in China on population, nutrition, demographics, and health status. The data were collected on individual, household, and community levels, and covered nine provinces and autonomous regions in China: Liaoning, Heilongjian, Shandong, Jiangsu, Henan, Hubei, Hunan, Guangxi, and Guizhou.¹ One-third of the Chinese population (approximately 450 million people in 1989) lives in these provinces, which vary substantially in geography and economic development. Stratified multistage cluster design was applied to the process of sampling (a detailed description of the CHNS design can be found at http://www.cpc.unc.edu/projects/china). Although the CHNS sample is not representative of the Chinese population, previous studies showed that characteristics of the CHNS households and individuals were comparable to those from national samples (see Du et al. 2002; Entwisle and Chen 2002; Short et al. 2000). The CHNS has released eight waves of data so far (1989, 1991, 1993, 1997, 2000, 2004, 2006, and 2009). In the current study, we used the 1991–2006 data waves of the survey because neither the 1989 nor 2009 waves collected information on self-rated health. We restricted our analyses to individuals aged 21 to 65 in every wave of the study in rural China. On average, an individual was observed more than three times in the panel. Our working sample consisted of 4,851 individuals in 1991; 4,741 in 1993; 5,366 in 1997; 4,940 in 2000; 5,423 in 2004; and 5,378 in 2006.² Among them, 985 individuals died by the end of 2006. Loss to follow-up rate ranged from 2 % to 8 % across the years. Altogether, this yielded 30,699 person-year records. We later explain in detail our approach to handling potential bias introduced by attrition and death in the analyses.

Measures

Our dependent variable was self-rated physical health and was based on the following survey question: “How would you describe your health compared to that of other people your age?” Responses ranged from 1 to 4, indicating excellent to poor health. We recoded this variable so that higher values indicate better health. We used this measure of self-reported physical health as our main dependent variable; this measure has been used extensively in United States and international research and has been consistently documented to be a valid measure of physical health (Farmer and Ferraro 1997; Hays et al. 1996; Johnson and Wolinsky 1993) and a strong predictor of survival and mortality, controlling for objective measures of health, such as physicians’ examinations, medical records, or extensive health histories (Idler and Angel 1990; Idler and Benyamini 1997).

We measured both marital status and duration of physical separation. We used the standard question in each wave of the survey on marital status, along with a question that indicated whether the spouse currently lived in the house, to create a five-category variable: married with spouse at home, married with spouse absent (reference), never married, divorced/separated, and widowed.³ To capture duration of spousal absence, we constructed a categorical variable that captured the duration of spousal absence: spouse never being away (reference), spouse being away in one wave, and spouse being away for more than one wave of the survey.

We controlled for three types of covariates: demographic covariates, socioeconomic resources, and health behaviors. Basic demographic covariates included gender, region (coastal, northeast, inland, mountainous south), and household composition (presence of children aged 0– 5 years, children aged 6–14 years, and older adults 60 years and older, as proxies for caregiving responsibilities). We used three variables to measure socioeconomic resources: education, logged per capita family income (family income adjusted for household size), and an index for the household’s aggregate asset ownership, weighted according to the number and approximate value of select household consumer durables owned by household members (for detailed documentation, see Korinek et al. 2006). We used smoking and drinking as proxies for health behaviors. All demographic covariates were time-invariant, and all variables for socioeconomic resources and health behaviors were time-varying. Descriptive statistics of all the variables used in the analysis are included in Table 1.

Table 1.

Descriptive statistics of all variables in the analyses (N = 30,699)

	Total Sample (30,669)		Female (15,760)		Male (14,939)
	Mean	SD	Mean	SD	Mean	SD
Self-reported Health (1–4)	2.779	0.729	2.725	0.732	2.835	0.722
Marital Status (& spousal location)
Married, spouse living at home (married, spouse living at home =   1, else = 0)	0.838	0.369	0.838	0.369	0.838	0.369
Married, spouse not at home (married, spouse not at home =1,   else = 0)	0.035	0.184	0.052	0.222	0.017	0.130
Never married (never married = 1, else = 0)	0.089	0.285	0.063	0.243	0.117	0.322
Divorced/separated (Divorced/separated = 1, else = 0)	0.006	0.075	0.003	0.059	0.008	0.090
Widowed (widowed = 1, else = 0)	0.033	0.178	0.044	0.206	0.020	0.141
Duration (for those who are married)
Spouse not away	0.891	0.312	0.846	0.361	0.938	0.242
Spouse away in one wave	0.082	0.274	0.108	0.310	0.055	0.226
Spouse away in two or more waves	0.028	0.164	0.046	0.210	0.008	0.090
Age (in years)	41.903	11.743	41.872	11.585	41.935	11.908
Sex (male = 1, female = 0)	0.487	0.500
Household composition
Presence of children aged 0–5 years	0.244	0.430	0.251	0.434	0.236	0.425
Presence of children aged 6–14 years	0.338	0.473	0.346	0.476	0.330	0.470
Presence of adults over the age of 60 years	0.253	0.435	0.264	0.441	0.242	0.428
Region
Northeast (northeast = 1, else = 0)	0.176	0.381	0.176	0.380	0.176	0.380
Coastal (coastal = 1, else = 0)	0.234	0.423	0.239	0.426	0.228	0.420
Inland (inland = 1, else = 0)	0.337	0.473	0.336	0.472	0.337	0.473
Mountainous south (south = 1, else = 0)	0.254	0.435	0.250	0.433	0.259	0.438
Died (died in 1991–2006= 1, not died = 0)	0.032	0.176	0.023	0.149	0.042	0.201
Higher Education
(higher education =1, primary education or less =0)	0.470	0.499	0.373	0.484	0.573	0.495
Health Behavior
Smoking (smoking = 1, not smoking = 0)	0.332	0.471	0.039	0.195	0.642	0.480
Drink (drinking = 1, not drinking = 0)	0.364	0.481	0.093	0.291	0.651	0.477
Economic Resources
Logged per capita family income	8.060	1.438	8.067	1.412	8.052	1.465
Household asset index (quintile)	2.810	1.431	2.814	1.430	2.806	1.431

Source: CHNS 1991–2006.

[TYPESETTER: Align values on the decimal point.]

Methods

To take advantage of six waves of longitudinal data, we used growth curve modeling techniques (Singer and Willett 2003) to estimate how marriage (including differences between those whose spouses were absent and those whose spouses were present) may shape typical self-rated physical health trajectories across age. Age was the analysis time metric. This analytic approach allowed us to take into account that individuals start with different levels of self-rated physical health and that each individual could experience a different rate of change in self-rated physical health across age as a function of marital status and spousal residence. Average self-rated physical health level (i.e., intercept of health trajectories) and rate of change (i.e., age slope of health trajectories) were modeled as individually varying growth parameters (i.e., random effects). The structural parameters from this part of the model provided the basis for assessing the effects of marriage (including separation from spouse) on an average level and rate of change in self-rated physical health trajectories. One of the major advantages of growth curve modeling in comparison with traditional regression modeling is its ability to distinguish the two levels (i.e., within- and between-individual) of heterogeneity in estimating self-rated physical health trajectories shaped by marital status (Singer and Willet 2003). Because our sample was derived from repeated measures of six waves of longitudinal panel data, the observations were nested. The growth curve models took into account the unobserved heterogeneity related to the nested distribution—which was often a source of selection bias—by allowing random-effects variation across individuals. With time-varying covariates, growth curve models also allowed respondents to serve as their own controls, thus eliminating both within-individual and between-individual confounds. The linear growth curve model we estimated can be specified as follows:

$$\begin{array}{cc}\hfill Y_{\mathit{ij}}& ={\pi }_{0i}+{\pi }_{1i}\left({\mathit{Age}}_{1i}-42\right)+{\pi }_{2i}{\left({\mathit{Age}}_{\mathit{ij}}-42\right)}^2+{\mathbf{Z}}_{\mathit{ij}}^{\prime }A+{\epsilon }_{\mathit{ij}}\hfill \\ \hfill {\pi }_{0i}& ={\alpha }_0+X_0^{\prime }{B}_0+{\xi }_{0i}\hfill \\ \hfill {\pi }_{1i}& ={\alpha }_1+X_1^{\prime }{B}_1+{\xi }_{1i}\hfill \\ \hfill {\pi }_{2i}& ={\alpha }_2,\hfill \end{array}$$ (1)

where Y_ak represents the outcome variable (i.e., the self-rated health of individual i at wave j), and j = 1, 2 … 6 indexes CHNS Waves 1 to 6 (starting from the 1991 wave). The term Age_ij is the main analysis time-scale variable and represents the age of individual i at wave j. We centered age on 42 years so that the intercept reflected the level of self-rated physical health at the average age of 42. The term ε_ij is the level 1 residual, and terms ξ_0i and ξ_1i are individual-specific (level 2) residuals. The terms π_0i and π_1i represent the ith individual’s intercept and age slope (i.e., random coefficients). We also included the quadratic age slope (with the coefficient indicated by π_2i and estimated as a fixed effect) to take into account the nonlinear change of self-rated health with age. All time-invariant covariates (e.g., gender) indicated by $X_0^{\prime }$ and $X_1^{\prime }$ were included at level 2 to predict intercept and age slope, respectively. All time-varying covariates (e.g., marital status, income) indicated by the vector of Z’ were included at level 1 of the model; A, B₀, and B₁ are their population average (i.e., fixed) effects of the covariates on average level and changing rate of self-rated physical health trajectories. We also included the interaction of marital status (time-varying) by age in level 1 of the model to test whether the effect of marital status on health varied by age. Because our preliminary analyses (not shown) suggested that the age-squared interactions with marital status were not significant, we did not include them in our final model.

We conducted two sets of analyses to examine the effects of marital status and duration of absence, respectively. In the analysis of marital status, we included the full sample. In the analysis of the duration effect of spouse absence, we limited our sample to those who were married when the survey was conducted. For both the full sample and subsample analyses, we estimated a sequence of growth curve models, with a basic model that included only marital status variables and controls for basic demographic characteristics, then added socioeconomic resources and health behavior variables. Because adding those variables step by step did not show major differences in the results, we reported the final models with all covariates controlled. We further stratified the analysis by gender to assess potential gender differences.⁴

Growth curve analysis allows for data that are unbalanced in time by including all individuals for the estimation of trajectories, regardless of their attrition status or the number of waves contributed to the person-year data set (Raudenbush and Bryk 2002). This largely reduced the number of cases lost to follow-up and alleviated the sample selection problem common to other regression models that exclude cases lost to follow-up. To further control for potential selection effects due to sample attrition, we adopted a straightforward approach by entering two time-invariant dummy variables, which indicated the deceased and lost-to-follow-up identities (the latter dropped because of insignificance) in level tw2o of the model (see Chen et al. 2010). All statistical analyses were performed using STATA 10 (StataCorp 2007).

Results From Growth Curve Models

Results from growth curve models for the full sample are shown in Table 2. For interpretation, the main effects of marital status reflect the effects of marital status on the intercept (i.e., average level at age 42, the centered age value) of self-reported physical health trajectories. The interaction effects of marital status by age reflect the effects of marital status on the rate of change of self-reported physical health trajectories (i.e., how health trajectory change across age differs from across marital status groups). Model 1 shows that when individual- and household-level sociodemographic characteristics are controlled, those married with a spouse living at home, on average, reported a better physical health level at age 42 than those whose spouse was absent: at age 42, they were on average 0.053 points higher in self-reported physical health (on a scale of 1 to 4). The nonsignificant age interaction effect of married with spouse living at home suggests that the spouse’s physical location do not affect the rate of change in respondents’ health trajectories across age. In other words, differences between married with spouse absent and married with spouse at home are stable and persistent across age. The never married reported a better average level of health (0.068 higher) at age 42 than the married with spouse absent, but their rate of health decline for these two groups do not significantly differ from each other. The widowed and divorced reported similar average levels of health at age 42 as those who were married with spouse absent, but the rate of decline in health trajectories across age for the divorced group is slower (b = 0.012).

Table 2.

Growth curve models of marital status on self reported health in rural China, CHNS 1991–2006

Fixed Effects	Total Sample	Females	Males
Intercept	2.527***	2.516***	2.565***
Marital Status	(0.042)	(0.057)	(0.067)
Married spouse living at home	0.053*	0.037	0.091*
	(0.022)	(0.026)	(0.043)
Never married	0.068*	0.114*	0.079
	(0.032)	(0.048)	(0.051)
Divorced/separated	−0.042	−0.049	−0.005
	(0.058)	(0.099)	(0.078)
Widowed	0.007	−0.049	0.097
(ref. = married, spouse not at home)	(0.052)	(0.065)	(0.089)
Age (centered on the mean)	−0.017***	−0.018***	−0.015***
	(0.002)	(0.003)	(0.004)
Age (centered), squared	−0.000***	−0.000***	−0.000*
	(0.000)	(0.000)	(0.000)
Married Spouse Living at Home × Age	0.000	−0.000	0.000
	(0.002)	(0.003)	(0.004)
Never Married × Age	0.001	0.002	0.000
	(0.003)	(0.003)	(0.004)
Divorced/Separated × Age	0.012*	0.011	0.011
	(0.006)	(0.009)	(0.007)
Widowed × Age	0.003	0.007	−0.003
	(0.004)	(0.005)	(0.006)
Sex (female = 0)	0.062***
	(0.013)
Education (primary school or less = 0)	0.016	−0.010	0.038**
	(0.010)	(0.015)	(0.014)
Presence of Children Aged 0–5	−0.008	−0.001	−0.015
	(0.011)	(0.015)	(0.015)
Presence of Children Aged 6–14	−0.011	−0.002	−0.023
	(0.009)	(0.012)	(0.013)
Presence of Adults Over Age 60	−0.023*	0.000	−0.038**
	(0.010)	(0.015)	(0.015)
Region: Northeast	0.165***	0.153***	0.179***
	(0.015)	(0.021)	(0.021)
Coastal	0.254***	0.227***	0.277***
	(0.014)	(0.020)	(0.020)
Inland (ref. =mountainous south)	0.089***	0.085***	0.091***
	(0.013)	(0.018)	(0.018)
Died (died in 1996–2001 = 1)	−0.135***	−0.123**	−0.142***
	(0.026)	(0.042)	(0.033)
Smoking	0.022	0.006	0.026*
	(0.012)	(0.032)	(0.013)
Drinking	0.061***	0.011	0.079***
	(0.011)	(0.020)	(0.013)
Logged per Capita Family Income	0.010**	0.010*	0.009*
	(0.003)	(0.005)	(0.004)
Household Assets Index	0.007*	0.008	0.005
	(0.003)	(0.005)	(0.005)
Random Effects–Variance Components
Level 1: Within-person	0.410***	0.418***	0.402***
	(0.004)	(0.006)	(0.006)
Level 2: In intercept	0.063***	0.064***	0.062***
	(0.004)	(0.005)	(0.005)
Level 2: In linear growth rate	0.000***	0.000***	0.000***
	(0.000)	(0.000)	(0.000)
Goodness-of-Fit
BIC (smaller is better)	64,295.64	33,370.92	31,246.76
Observations	30,699	15,760	14,939
Number of Groups	9,635	4,892	4,743

Note: Standard errors are shown in parentheses.

^*p < .05;

^**p < .01;

^***p < .001 (two-tailed test)

[TYPESETTER: Align values with decimals on the decimal point. Align values without decimal points on the left.]

To simultaneously show the effect of marital status and presence of spouse on both the average level of health and rate of decline in health, we display health trajectories in Fig. 1 by using the coefficients in Table 2 and holding all the control variables at their means (for continuous variables) and modes (for dummy variables). Figure 1 shows a clear health disadvantage for those who were married but separated from their spouses spatially compared with those whose spouses were present. This health disadvantage is persistent across age. Consistent with existing research in other contexts, the widowed and divorced have the worst health, on average, although the differences between them and the married with spouses absent are not statistically different at age 42. The divorced group shows the lowest average level of health, but their rate of decline is less steep compared with other groups. The better health of the divorced relative to other groups at older ages may be related to a selection process. Given that divorce is still a relatively rare phenomenon (0.6 % of our sample) and is even more uncommon among older adults in rural China, we refrain from overinterpreting the “crossover” observed around age 50 in Fig. 1.

Fig. 1.

Predicted self-reported health trajectories by marital status and living arrangement, CHNS 1991–2006

It is noteworthy that the never-married group does not seem to have a health disadvantage when compared with those who were married and living with their spouses—and may have better health at an older age. Our additional analysis (not shown) suggest that the difference in self-reported physical health trajectories between the never married and the married who live with their spouses is not statistically significant (p > .05). Previous research has suggested that the evidence of a health advantage of the married group in comparison to the never-married group is mixed, with some findings suggesting better health of the married group than the never-married group and others finding no significant differences between these two groups (see Umberson and Williams 1999; Williams 1992).

Findings for other covariates are similar. Self-reported physical health declines with age. Men reported better physical health than did women. People in northeast, coastal, and inland regions reported better physical health than people in the mountainous south. The number of old adults living at home is negatively associated with respondents’ self-reported physical health. As expected, income and wealth have health-promoting effects (i.e., positive coefficients). However, the effects of smoking and drinking are also positive in direction. Although this may seem paradoxical, these results are consistent with earlier studies showing that in countries experiencing rapid economic development and epidemiological transition, those with better economic resources were more likely to make unhealthier behavior choices (including unhealthy diets, characterized by higher fat levels and more added sugar), more likely to be obese, more likely to smoke and drink, and more likely to be sedentary (Chen et al. 2010; Du et al. 2002; Kim et al. 2004; Popkin 1998). This is consistent with the positive association between socioeconomic status (SES) and health behavior indicators in the CHNS sample. Because we were unable to include other proximate determinants of health in the model, such as nutrition, access to health care, and exposure to stressors, we view the positive effects of smoking and drinking as a further demonstration of the residual SES effect.

Consistent with the literature, married women in the rural CHNS sample are almost three times more likely to experience spousal absence as married men (see Table 1). How does such a gendered pattern influence their health trajectories? We further explored this question by conducting the same analyses presented in Table 2 in male and female subsamples. Although the effect of spousal absence on the average health level at age 42 (i.e., main effect of marital status) is nonsignificant in the female subsamples, the coefficient size for the male sample is three times larger than that for the female sample (0.091 vs. 0.037; Table 2). Moreover, the coefficient for the main effect of a spouse living at home is statistically significant only in the male sample. Interestingly, never-married women reported the highest average level of health at age 42 compared with other groups, whereas no such effect is present for men.

Although the preceding results clearly suggest physical health disadvantages for those married but separated from their spouses relative to those married with their spouses at home (and even relative to some unmarried groups, such as the never married), we further investigated whether the duration of spousal absence has any effect on health trajectories. In the next stage, we limited our sample to those who were married and differentiated among those whose spouses were always present (reference), absent in one wave of the study, and away for two or more waves. We present the results both in the full sample and gender subsamples in Table 3. The results suggest that short-term separation (one wave of spousal absence) is not associated with a health disadvantage. However, in the total and male samples, those whose spouses were absent in two or more waves experienced worse health trajectories than married individuals whose spouses were always present. As Fig. 2 demonstrates, those whose spouses were away for two waves or longer reported worse physical health after their late 20s and also experienced a more rapid decline in health (p < .10). The effect is much more salient for men than women, as indicated by the larger gap between the health trajectories of those whose spouses were away for more than one wave and those whose spouses were away for only one wave of the study (see Fig. 3). We further discuss the implication of the gender difference in the next section.

Table 3.

Growth curve models of duration of spousal absence on self reported health, married rural subsample, CHNS 1991–2006

Fixed Effects	Total Sample	Females	Males
Intercept	2.560***	2.524***	2.652***
Duration of Spousal Absence	(0.040)	(0.055)	(0.058)
Spouse away in one wave	0.012	0.036	−0.039
	(0.019)	(0.024)	(0.032)
Spouse away in two or more waves	−0.070*	−0.044	−0.157*
(ref. = spouse not away)	(0.033)	(0.037)	(0.079)
Age (centered on the mean)	−0.017***	−0.019***	−0.015***
	(0.001)	(0.001)	(0.001)
Age (centered), square	−0.000***	−0.000***	−0.000***
	(0.000)	(0.000)	(0.000)
Spouse Away in One Wave × Age	0.002	0.003	0.001
	(0.002)	(0.002)	(0.003)
Spouse Away in Two or More Waves × Age	−0.005	−0.002	−0.008
	(0.003)	(0.003)	(0.006)
Sex (female = 0)	0.075***
	(0.014)
High Education	0.004	−0.017	0.024
	(0.011)	(0.015)	(0.015)
Presence of Children Aged 0–5	−0.004	0.001	−0.006
	(0.011)	(0.016)	(0.016)
Presence of Children Aged 6–14	−0.014	0.001	−0.031*
	(0.009)	(0.013)	(0.014)
Presence of Adults Over Age 60	−0.011	0.002	−0.016
	(0.011)	(0.016)	(0.016)
Region: Northeast	0.164***	0.153***	0.177***
	(0.016)	(0.022)	(0.022)
Coastal	0.250***	0.231***	0.267***
	(0.015)	(0.022)	(0.022)
Inland (ref. = mountainous south)	0.082***	0.087***	0.075***
	(0.014)	(0.019)	(0.019)
Died (died in 1996–2001 = 1)	−0.140***	−0.129**	−0.148***
	(0.028)	(0.047)	(0.035)
Logged per Capita Family Income	0.009**	0.010*	0.008
	(0.003)	(0.005)	(0.005)
Household Assets Index	0.006	0.007	0.004
	(0.004)	(0.005)	(0.005)
Smoking	0.022	0.000	0.029*
	(0.013)	(0.034)	(0.014)
Drinking	0.060***	0.013	0.079***
	(0.011)	(0.021)	(0.014)
Random Effects–Variance Components	0.412***	0.420***	0.403***
Level 1: Within-person	(0.004)	(0.006)	(0.006)
	0.065***	0.067***	0.062***
Level 2: In intercept	(0.004)	(0.005)	(0.005)
	0.000***	0.000***	0.000***
Level 2: In linear growth rate	(0.000)	(0.000)	(0.000)
Goodness-of-Fit
BIC (smaller is better)	56,154.13	29,726.63	26,693.04
Observations	26,785	14,017	12,768
Number of Groups	8,152	4,144	4,008

Note: Standard errors are shown in parentheses.

^*p < .05;

^**p < .01;

^***p < .001 (two-tailed test)

[TYPESETTER: Align values with decimals on the decimal point. Align values without decimal points on the left.]

Fig. 2.

predicted self-reported health trajectories by duration of absence of spouse, married sample, CHNS 1991–2006

Fig. 3.

Predicted self-reported health trajectories by duration of absence of spouse by sex, CHNS 1991–2006

Discussion and Conclusion

Do the health profiles of married people who do not live together resemble those of unmarried individuals or married couples who live together? The common phenomenon of physical separation between spouses due to migration in China, made prevalent by massive migration trends, highlights the emerging importance of this question in the context of rural China.

Our study offers a clear contribution to the literature by distinguishing between married individuals living with their spouses and married individuals who are physically separated from their spouses, with the latter suffering from a significant health deficit more so than some unmarried groups. In general, our results suggest that those who are married but living apart from their spouses have persistently worse physical health compared with those who are married and living with their spouses (consistent with Hypothesis 1), and they seem even more disadvantaged than some of the unmarried groups, such as the never married (inconsistent with Hypothesis 2, as we discuss later). This is not a trivial issue. In rural China, marriage separation due to migration is far more common than divorce. The findings shed light on the link between marriage and health by demonstrating that marriage benefits do not accrue to the same extent for those who are spatially separated from their spouses. Marriage with both spouses living together may promote social and emotional support and reduce psychological distress during stressful events. In contrast, married people who live separately from their spouses may not enjoy the same level of marital resources because of reduced intimacy due to geographical distance. For example, on the emotional side, physical separation from the spouse could lead to increased stress and decreased emotional support. On the instrumental side, those who are left behind have to take on new roles and responsibilities. Davin (1999) discussed a heightened level of role strain for both married men and women who were left behind in rural villages in China: husbands had to learn tasks that were traditionally seen as women’s work (e.g., cooking, laundry), and women became the head of the household and made day-to-day decisions about their land, home, and children. All these factors may harm the health of married couples who live apart.

Further, when we took into account the duration of separation, we found that the longer the separation, the more detrimental the health consequences (supporting Hypothesis 3a rather than Hypothesis 3b). This evidence supports the marital resource model, suggesting a cumulative effect of marital status on health. Unlike marital dissolution, which is more often associated with a loss of income, marriage separation due to migration does not necessarily harm earnings and may be associated with an increase in income. Spousal migration for employment has often ben explained as an economic diversification strategy of the household, particularly in developing countries (Sana and Massey 2005). Migrant spouses typically send earnings home and continue to be active members of the household economy. Thus, the economic benefits of spousal migration could alleviate some of the negative health consequences of spousal separation, but perhaps only in the short term. This may explain why a short-term absence of the spouse (one wave of the survey) may not render any health deficiencies.

Finally, our study also found that spousal absence, especially longer duration of absence, yielded stronger negative effects for married men than for married women (consistent with Hypotheses 3a and 4). This finding is consistent with the general literature showing that marriage enhances men’s health more than women’s and that the adverse effects of marital dissolution are stronger for men than women (Williams and Umberson 2004). Because women often act as kin keepers, men gain more from marriage in terms of emotional and social support, with marriage acting as a buffer against stress and a deterrent against risky behavior. The absence of a wife could mean a significant loss of such support. Women, however, are more resilient in that regard. Historically, Chinese women have fostered bonds and social networks on their own, which could provide them with much-needed help in times of emotional stress (see Wolf 1972). Men, on the other hand, could suffer from additional stress because of the stigma associated with their left-behind status.

Although we consider our study a contribution to the general literature, the preceding discussion highlights the unique context of rural China in shaping the relationship between marriage and health. Most of our findings are in line with marital resource theory, but one notable deviation is that never-married women are not disadvantaged compared with married women (spouse present or not present) in terms of average level of health. This finding resonates with the well-documented literature on the low status of married women in rural China (Riley 1994). Under a highly hierarchical family system that is typically patrilocal, patrilineal, and patriarchal, it is conceivable that the cost and burden that comes with marriage outweighs its benefits. A recent study on marriage and suicide in rural China found that instead of strengthening familial and social integration, as suggested by Western theories and statistics, marriage is not a protective factor for rural Chinese women (Zhang 2010). In addition, because of the cultural sanction against divorce and less emphasis on intimacy and emotional gains in marriage, the married may be more heterogeneous and thus not advantaged, on average, when compared with their never-married counterparts.

The current study is not without its limitations. First, although the longitudinal nature of the CHNS allowed us to profile individual health trajectories, it did not follow up on those who migrated out of the household. Thus, our study is limited to those who stayed behind in rural areas over the years. At the same time, exclusion of migrants is an advantage in that the selection issue of primary concern to studies of migration and health is of less importance for our study. Second, the CHNS data did not provide information on the amount of migrants’ earnings or frequency of remittances, thus limiting our ability to directly assess the economic benefits brought by spousal migration. Third, our data do not include measures of social support and emotional support. The health behavior variables (smoking and drinking) also essentially capture unmeasured aspects of SES resulting from the developmental context of China (i.e., wealthier people are more likely to smoke and drink). Future research is needed to identify other, better-measured mechanisms to explain the identified health differences. Additionally, we used self-reported physical health as a health outcome variable in the analyses. Although the CHNS data provide rich information on physical health, the health items changed from one wave of the survey to the other. Therefore, we were unable to use other physical health measures in our growth curve models. Finally, most of our key findings on differences in physical health trajectories between married persons with and without spousal absence are mainly related to the average levels (i.e., the intercept) of the health trajectories instead of the changing rates (i.e., the slope) of the trajectories. Our results indicate that the differences between the married persons with spouse present and those with spouse absent are stable and persistent across age. Is this finding a likely result from a selection process? Our answer is a definite no. If any selection process were at work, we would expect that those with a spouse in poorer health would be less likely to leave home and migrate to other cities because of their health care responsibilities for the sick spouse. In this sense, those married with their spouse absent would be more likely to be selective of people in better health. In this sense, our findings of major health disadvantages for married persons with spouses absent compared with married persons with spouses living at home are conservative. Indeed, the finding of persistent health disadvantage of those married with spouse absent across age is consistent with the marital resource model, suggesting that a lack of access to marital resources associated with day-to-day interactions may have long-term persistent consequences on individuals’ health.

Despite its limitations, our study provides a useful understanding of how marriage with spouses living apart shapes individual health trajectories in the context of rural China. Given that left-behind spouses often become sole providers of child care and older adult care, it is imperative for us to understand their needs, stresses, and role strain as part of the general process of how migration influences individuals’ health, particularly in a context where migration can be construed as a strategy of economic diversification in many rural households. Although marriage separation due to migration is qualitatively different from divorce, its disruptive health effects are evident from this study. Extended periods of spousal absence could also lead to a weakened marital relationship and eventually martial dissolution. With the trend of divorce on the rise in China, the findings of this study could serve as a warning about health risks, as migration, urbanization, and economic development are gradually changing normative and cultural expectations about marriage in China.