Introduction

There has been a growing awareness that a sizable proportion of US adults experience what researchers call work-family conflict, individuals’ perceptions that their work and family responsibilities interfere with each other (Greenhaus and Beutell 1985; Voydanoff 2005a). Work-family conflict is two directional: work-to-family conflict (WFC) and family-to-work conflict (FWC). Most research has focused on WFC, leaving FWC less often investigated (Bianchi and Milkie 2010). Yet, many employed adults today have some kind of family responsibility that they have to balance with their paid work (Williams 2010). Prior research has found that FWC is related to poor mental health, missing work, and dropping out of the labor force (see Bellavia and Frone 2005 for a review). Thus, more research is needed to identify the various kinds of family demands that may foster FWC in order to better understand employed adults’ experiences of stressfulness in balancing work and family life.

Prior research has largely focused on childcare responsibilities as major family demands that may foster FWC (e.g., Stevens et al. 2007). Children’s specific conditions that indicate higher care demands, such as a chronic illness, disability, difficult disposition, or emotional or behavioral disorder, have been identified as sources of FWC (Lewis et al. 2000; Voydanoff 2005b). More recent studies have examined caregiving to family members with health problems or a disability as a source of FWC (Glavin and Peters 2015; Kim et al. 2013). Even though one’s spouse—in married or cohabiting unions—could have a variety of problems that may generate stressful conditions at home, little research has examined one’s spouse’s problems as a key source of one’s FWC. This is not surprising because US culture does not encourage an open dialogue about one’s spouse’s problems as a burden that influences one’s capacity to be a good worker (Corrigan et al. 2003; Williams 2010). It is important to recognize those who may be struggling in silence with balancing their work life with high family demands due to their spouses’ problems.

Using data from the 2011 National Survey of Midlife Development in the United States (MIDUS) Refreshed Sample (the 2011 MIDUS-RS), the only data set (except for the 1995 MIDUS) that provides primary information relevant to the present analysis, we examined the association between respondents’ reports of their spouses’ problems and their own FWC among employed adults aged 25–61 who were married to or cohabiting with a partner. We examined three kinds of spousal problems: (a) poor physical health, (b) poor mental health, and (c) behavioral disorders, defined as disruptive behaviors that cause problems at work, at home and in social situations (US Department of Health and Human Services 2016). Guided by the concept of stress proliferation in the stress process model (Pearlin and Bierman 2013), we expected that their spouses’ problems were related to respondents’ higher FWC through proliferation of other stressors, especially strain-based stressors such as the respondents’ role overload, financial strain, and relationship strain. Because individuals’ vulnerability to the same stressor may vary by their social locations, we examined how the respondents’ gender and parental status may moderate the association between their spouses’ problems and their own FWC.

Determinants of FWC

The concept of FWC is rooted in role theory (Bellavia and Frone 2005). It defines FWC as a form of inter-role conflict, which involves the extent to which individuals feel that it is difficult to meet expectations in their work role because their family demands and obligations negatively influence their attitudes, energies, and capacities on the job (Greenhaus and Beutell 1985). Identifying the specific family demands or stressors is needed to understand determinants of FWC (Voydanoff 2005a). Following role theory’s emphasis on the importance of social roles as a source of stress, researchers have identified types of demands within specific roles, such as child demands, spouse demands, kin demands, and household demands (Voydanoff 2005a, b). Ecological and sociological perspectives, which emphasize that social roles are embedded in structural contexts of the household or the larger society, identify sources of FWC beyond specific roles, such as low household income (e.g., Grzywacz and Marks 2000; Schieman and Young 2011). In any of these theoretical approaches, researchers typically consider two types of demands. The first type is structural, time-based, or task-based demands, which focus on the amount of time or tasks, or the degree of involvement required for individuals to fulfill certain family responsibilities. To measure these types of demands, some researchers have used the amount of time spent on child care or household chores; other researchers have used other indicators as a proxy, such as the number or age of children, children’s illness, children’s difficult temperament, children’s emotional or behavioral disorders, or whether respondents had any caregiving responsibilities to family members (Hyde et al. 2004; Nomaguchi 2012; Stevens et al. 2007; VandenHeuvel 1997; Voydanoff 2005a, b). The second type is psychological, strain-based demands, which consist of perceptions of demands, such as feelings of being overloaded, perceived harsh treatments or low support from one’s spouse, or financial strain (Grzywacz and Marks 2000; Schieman and Young 2011; Seery et al. 2008; Voydanoff 2005b).

Spousal Problems as Family Demands Influencing Respondent’s FWC

In this paper, we sought to expand understanding of spouse demands by examining three kinds of respondents’ spouses’ problems as a potentially important source of the respondents’ FWC. First, we considered one’s spouse’s poor physical health. Past research has examined caregiving to one’s spouse for a serious illness or disability as a stressor that relates to FWC (Marks 1998; Pearlin et al. 1997). Focusing on those who were living in the San Francisco Bay Area and Los Angeles County, Pearlin et al. (1997) found that caregiving for a partner who had AIDS was related to higher FWC, although the authors called it work strain. Using data from the 1992–1993 Wisconsin Longitudinal Study (WLS), when respondents were 52–53 years old, Marks (1998) found that providing care to their spouse for health problems was positively related to respondents’ FWC. We extended these prior studies by using a broader definition of physical health problems than caregiving demands and a national sample of working-age US adults.

Second, we examined two other types of one’s spouse’s problems that have been rarely examined in the FWC literature: one’s spouse’s poor mental health and behavioral disorders. The term behavioral disorder is most commonly used to refer to children’s or adolescents’ disruptive behaviors that cause problems in school, at home, and in social situations which may manifest as inattention, hyperactivity, impulsivity, substance use, and criminal activity (US Department of Health and Human Services 2016). Although often neglected, these problems may continue into adulthood (US Public Health Service 2000). Although there has been an increase in awareness of poor mental health and behavioral disorders as medical issues, stigma against these conditions remains (Corrigan et al. 2003). When one spouse has a mental illness or behavioral disorder, it may be hard on the other spouse who is often a sole supporter of the ill spouse while managing paid work and household responsibilities.

To understand the link between their spouses’ problems and the respondents’ FWC, we used a concept of stress proliferation from the stress process model. The stress process model is a dominant framework in sociological research on stress (Pearlin and Bierman 2013). Work-family research has begun to use this model to understand determinants and consequences of FWC (e.g., Glavin and Peters 2015; Schieman and Young 2011). Stress proliferation refers to the expansion of stressors within and beyond a situation where the stressfulness initially originated from (Pearlin et al. 1997). The underlying idea is that ongoing difficulties in life do not exist in an isolated manner. A stressful experience tends to produce more challenges in people’s lives, creating a chain of stressors by spilling over into other domains of life (LeBlanc et al. 2015). Past research has found that stressfulness of caregiving tasks can proliferate into difficulties in the workplace (Glavin and Peters 2015; Pearlin et al. 1997). In this paper, we consider that one’s spouse’s problem may generate several stressful circumstances at home, which, in turn, relates to his or her experience of higher FWC. We discuss these processes in the following section.

One spouse’s problems may increase the other spouse’s paid work and household responsibilities. Health problems and behavioral disorders can undermine one’s ability to work for pay. Prior research has found that physical health problems (Pelkowski and Berger 2004), mental illness (Ettner et al. 1997), excessive alcohol use (Berger and Leigh 1988), and illegal drug use (DeSimone 2002) were negatively related to paid work hours. In addition, poor health and behavioral disorders can hinder one’s capacity to do household chores (Bair et al. 2008; Finney et al. 1991). When one spouse has a health problem or behavioral disorder, the other spouse may need to work longer hours to compensate for the lost income (O’Hara 2004; Siegel 2006) and also may need to pick up the slack in the household chores. Not everyone in this situation can increase his or her work hours, however. One spouse’s cutting paid work hours may mean unexpected reduction in his or her household income (Pelkowski and Berger 2004). Prior research has shown that mental illness is related to lower earnings (Ettner et al. 1997). Heavy drinking (Barrett 2002), illegal drug use (DeSimone 2002), and incarceration history (Schwartz-Soicher et al. 2011) have a negative association with earnings. Thus, employed adults whose spouse has problems may be more likely than those whose spouse does not have problems to experience lower household economic conditions and shoulder the breadwinning responsibility.

Empirical findings have been inconsistent as to how a respondent’s and his or her spouse’s time spent on paid work and household chores are related to the respondent’s FWC. Some studies have found that paid work hours were positively related to FWC (Grzywacz et al. 2002; Voydanoff 2005c; Young and Schieman 2012), whereas other studies did not find the association (Dilworth 2004; Mennino et al. 2005; Nomaguchi 2012). Young and Schieman (2012) found that their spouses’ paid work hours were negatively related to the respondents’ FWC, although Dilworth (2004) found there was no significant association. Past research has found that there is no association between hours spent on housework and FWC (Dilworth 2004; Stevens et al. 2007). How household income is related to FWC is unclear, too. Some studies found that low income or earnings were positively related to FWC (Nomaguchi 2012; Voydanoff 2007), whereas other studies found no association (Grzywacz and Marks 2000; Grzywacz et al. 2002; Mennino et al. 2005; Schieman and Young 2011). Little research has examined how respondents’ contribution to the household income relative to their spouses’ is related to the respondents’ FWC.

The weak associations between time allocation or household income and FWC are not surprising. Stress researchers have noted that the actual amount of paid work or household chores is not always a direct indicator of a stressor (Pearlin and Bierman 2013; Schieman and Young 2011). Rather, as Pearlin et al. (1997) noted, the intensity of a stressor tends to be better measured by asking individuals how they feel about what they do rather than asking them how many hours they spend on it. Past empirical research has suggested support for this argument. Voydanoff (2005c) found that strain-based demands were stronger predictors of FWC than time- or task-based demands. Similarly, Schieman and Young (2011) found that perceptions of economic conditions were a stronger predictor of FWC than household income. Following these arguments and prior findings, we focus on three measures of strain-based stressors as mediators of the link between respondents’ spouses’ problems and FWC: role overload, financial strain, and relationship strain.

Role overload involves the extent to which individuals feel as if demands on energy and stamina exceed their capacity (Pearlin and Bierman 2013). Their spouses’ poor physical health, poor mental health, and behavioral disorders may increase the respondents’ role overload, as past evidence suggests that increases in paid work or housework hours may lead to perceived role overload (Hecht 2001; Shultz et al. 2010). In their study of AIDS caregivers, Pearlin et al. (1997) found that increases in caregiving demands were related to more role overload. Pearlin et al. found no association between increases in role overload and increases in FWC. In contrast, using data from the 1995 MIDUS, Voydanoff (2005b) found that role overload, although she called it household demands, was a strong predictor of FWC.

Financial strain, also called economic hardship, is generally defined as individuals’ perceptions of the difficulties in paying bills and acquiring basic necessities (Schieman and Young 2011). This concept involves the extent to which individuals feel as if their household economic conditions did not achieve their expected level. Because health care is very expensive in US, adults whose family members have health problems often worry about their economic situation (McCarthy 2016). As medical expenses increase unexpectedly, financial burdens can place a major strain upon families (Amir et al. 2012). In addition, addicted spouses may spend a substantial amount of money on alcohol or drugs that couples cannot afford (Orford et al. 2010). Thus we expected that all three types of spouses’ problems—physical illness, mental illness, and behavioral disorders—would be related to respondents’ higher financial strain. Prior research has found that financial strain is positively related to FWC (Schieman and Young 2011).

Finally, we consider relationship strain. The literature of marital relationships has suggested that a spouse’s illnesses can have negative associations with relationship quality (Yorgason et al. 2008). The ill spouse’s lack of participation in household chores may foster the other spouse’s feelings of unfairness or dissatisfaction, which is positively related to the other spouse’s FWC (Stevens et al. 2007). Yet some studies, which focused on cancer patients and their spouses, found little effect of one spouse’s illness on the other spouse’s perception of relationship quality (Kuijer et al. 2001). Other research has shown that substance abuse (Cranford et al. 2011; Zavala and Spohn 2010), mental health issues (Whisman et al. 2004), or problems with the law (Turney 2015) were related to poor relationship quality measured as relationship satisfaction, supportiveness, and physical or emotional abuse. Prior research has shown that relationship strain is positively related to FWC (Grzywacz and Marks 2000; Voydanoff 2005b).

Variations by Gender and Parental Status

The stress process model contends that an individual’s vulnerability to the same level of stressor may vary depending on his or her location in stratified social statuses (Pearlin and Bierman 2013). We consider two distinct social statuses that might moderate the association between their spouses’ problems and respondents’ FWC: the respondents’ gender and parental status. Caregiving is considered “women’s work” in the US and men were less likely than women to expect to take on caregiving roles (Williams 2010). Having a spouse who needs care or cannot perform household tasks may be more likely to be perceived as burdensome by men than women and thus may create more stress for men than women (Zarit et al. 1986). Consistent with this view, Yorgason et al. (2008) found that a wife’s disability was related to her husband’s perception of lower marital quality, whereas a husband’s disability was not related to his wife’s perception of marital quality. Similarly, Marks (1998) found that the link between caregiving to a spouse and FWC was greater for men than women. In contrast, among spouse caregivers in the 2004 National Long-Term Care Survey, Lin et al. (2012) found that wives were more likely than husbands to report negative experiences of caregiving. With regard to variation by parental status, the association between their spouses’ problems and respondents’ FWC may be greater for those with children, especially younger ones. Children may require adults to make additional efforts to adjust to their spouses’ problems, including explaining their spouses’ problems to their children (Semple and McCance 2010). Lastly, the relationship between their spouses’ problems and respondents’ FWC may be stronger for fathers with young children than mothers or adults without children living in the household, because of unexpected double-burdens of care for both their spouse and children.

The stress process model also suggests that the degree to which proliferation of stressors may occur would vary by social locations, including gender and parental status (Pearlin and Bierman 2013). Because little prior research has examined one’s spouse’s problems as predictors of one’s FWC, it is difficult to make specific predictions. Still, we explored whether the primary mediation factors differed by gender and parental status.

The Present Study

On the basis of the foregoing discussion, we stated the following six hypotheses:

Hypothesis 1

Respondents’ reports of their spouses’ problems are positively related to their own FWC.

Hypothesis 2

The association between their spouses’ problems and respondents’ FWC is mediated modestly by structural, time- or task-based demands, such as the respondents’ and their spouses’ time allocation, household income, and share of the breadwinning responsibility.

Hypothesis 3

The association between their spouses’ problems and respondents’ FWC is mediated by strain-based demands including role overload, financial strain, and relationship strain.

Hypothesis 4

The association between their spouses’ problems and respondents’ FWC is greater for men than women.

Hypothesis 5

The relationship between their spouses’ problems and respondents’ FWC is greater for parents with young children compared to those without children.

Hypothesis 6

The relationship between their spouses’ problems and respondents’ FWC is greater for fathers with young children compared to mothers or men and women without young children.

All multivariate analyses controlled for respondents’ characteristics that may be related to both their spouses’ problems and the respondents’ FWC. Prior research has found that older age is positively related to poor physical health (Conn 2011), but negatively related to FWC (Hill et al. 2014). Researchers have also found racial-ethnic differences in prevalence of health problems and behavioral disorders (Caetano et al. 2012; Williams and Sternthal 2010) as well as levels of FWC (Grzywacz and Marks 2000). Prior work has found that cohabiting adults are more likely than married adults to have health problems and behavioral disorders (Brown et al. 2005), but are less likely to report FWC (McGinnity and Whelan 2009). Researchers also have found that education is negatively related to poor physical health (Ross and Wu 1995), but positively related to FWC (Minnotte et al. 2015). As past work suggested, respondents whose spouse has problems may have poor physical or mental health themselves (Evangelista et al. 2002; Marks 1998) and respondents with poor mental health may evaluate both their spouse’s health and their FWC negatively. Thus, we controlled for respondents’ physical and mental health. The MIDUS did not have information about respondents’ behavioral disorders.

Because we used cross-sectional data, we did not aim to address causal directions of these associations. Although we controlled for respondents’ mental health to reduce the possibility that their poor mental health might have influenced their perceptions of their spouses’ problems or their own FWC, the causal order might have been the opposite from what we discussed. For example, some research has suggested that financial strain and relationship strain could lead to poor mental health and behavioral disorders (e.g., Fairbairn and Testa 2017; Gudmunson et al. 2007; Mavandadi et al. 2014). We discuss these limitations in the “Discussion” section.

Method

Data

Data for the present analyses were drawn from the 2011 MIDUS-RS. The MIDUS, a longitudinal study of a nationally representative sample of non-institutionalized English-speaking adults aged 25 to 74 residing within the contiguous US, was conducted by the MacArthur Foundation Research Network on Successful Midlife Development (Ryff et al. 2016). The first, second, and third waves were collected in 1995–1996, 2002–2006, and 2011–2014 respectively. In the third wave, in addition to a follow-up to the 1995–1996 study (the 1995 MIDUS), a “refreshed” sample (the 2011 MIDUS-RS) of US adults with matching criteria to the 1995 MIDUS was collected. In the 2011 MIDUS-RS, respondents were first interviewed through either landline or mobile phone interviews and later were sent a self-administered questionnaire (SAQ) to complete. The 2011 MIDUS-RS was the best dataset with a nationally representative sample of US adults that provided information about a wide range of respondents’ spouses’ problems and question items that could be used to construct a FWC scale for respondents. Although the first two waves of MIDUS had similar information about spousal problems and FWC, we were unable to use the longitudinal data of the 1995 MIDUS for a few reasons. First, the interval between the first and the second waves was approximately nine years, which is too long to examine the associations between spousal problems in the first wave and respondents’ FWC in the second wave. Second, a substantial proportion of respondents in the analytical sample in the first wave were not in the second wave due to retirement or unemployment (24.7%) or relationship dissolution (17.3%).

For this paper, of the 3343 respondents in the 2011 MIDUS-RS, we first selected 2476 respondents who were married to or cohabiting with a partner. Next, following prior research (Grzywacz and Marks 2000), we included those under the age of 62 only (n = 1825). Then we selected those who were working for pay (n = 1434). Lastly, we restricted the sample to those who answered the SAQ, which resulted in the final sample of N = 980. Using Heckman’s (1979) method, we evaluated possible bias from selecting respondents with a completed SAQ. Those included in our analytical sample were more likely to be older and have higher levels of education, and were less likely to be Hispanic. We then estimated the probability of being selected into the analytical sample (λ) and included it in our regression models. We found that λ had no significant effects in our models nor did it alter any patterns of findings discussed below, which suggests that our results were not biased by our sample restriction.

Measures

Dependent Variable

Respondents’ FWC was measured as the mean of four-items. Respondents were asked how often they experienced the following in the past year: (a) “responsibilities at home reduce the effort you can devote to your job, (b) personal or family worries and problems distract you when you are at work, (c) activities and chores at home prevent you from getting the amount of sleep you need to do your job well, and (d) stress at home makes you irritable at work (1 = never, 2 = rarely, 3 = sometimes, 4 = most of the time, and 5 = all of the time).” This scale has been used by other studies (Grzywacz and Marks 2000; Voydanoff 2005a). Internal consistency measured by Cronbach’s alpha was α = 0.76, which was within the conventional standards of reliability (Tavakol and Dennick 2011).

Independent Variables

Three types of spousal problems were examined. Respondents’ spouses’ poor physical health was measured by the question, “In general, would you say your spouse’s physical health is 1 = excellent, 2 = very good, 3 = good, 4 = fair, or 5 = poor?” Similarly, respondents’ spouses’ poor mental health was measured by the question, “Would you say your spouse’s mental or emotional health is 1 = excellent, 2 = very good, 3 = good, 4 = fair, or 5 = poor?” Respondents’ spouses’ behavioral disorders were measured as a sum of six dichotomous variables which indicated whether a certain type of problem had happened to their spouse or partner in the previous 12 months (0 = no, and 1 = yes), including: (a) alcohol or substance problems, (b) financial problems (e.g., heavy debts), (c) problems at school or at work (e.g., failing grades, poor job performance), (d) difficulty finding or keeping a job, (e) legal problems (e.g., involved in law suits, police charges, traffic violations), and (f) difficulty getting along with others (α= 0.53) (range = 0–6). The alpha level was below the conventional standard level of reliability (between 0.70 and 0.90), thus we were cautious when interpreting findings for behavioral disorders. Still, these question items provided a rare opportunity of examining the association between one’s spouse’s behavioral disorders and his or her FWC with a nationally representative sample.

Mediating Variables

Four indicators of time-based demands were included. Respondent’s weekly work hours was measured as the number of hours of paid work at the respondent’s main job and any other jobs in a typical week. Respondent’s weekly housework hours was measured as the number of hours spent on household chores in a typical week. Spouse’s weekly work hours was measured as spouse’s hours of paid work at their main job and any other jobs in an average week. Spouse’s weekly housework hours was measured as typical hours spent on household chores. For respondents’ and their spouses’ work hours and housework, extreme cases were recoded into the 95th percentile by gender (Bianchi et al. 2000). In addition, household economic conditions and the respondent’s breadwinning responsibilities were included. Household income was measured using the total annual income of all those residing within the household in the previous year. Respondent’s share of household income was measured as respondents’ income relative to their spouses’ income earned over the past 12 months. As in Britt and Huston (2012), values were separated into three distinct categories including (a) < 40%, (b) 40–60%, and (c) > 60%.

Our primary mediating measures were four strain-based stressors that respondents reported. Role overload was measured as the mean of the four items (α= 0.69) which asked respondents, “How often has each of the following occurred at home in the past year? (a) you have too many demands made on you, (b) you control the amount of time you spend on tasks (reverse coded), (c) you have enough time to get everything done (reverse coded), or (d) you have a lot of interruptions (1 = never, 2 = rarely, 3 = sometimes, 4 = most of the time, and 5 = all of the time).” Financial strain was measured using the following question: “How difficult is it for you (and your family) to pay your monthly bills (1 = not at all difficult, 2 = not very difficult, 3 = somewhat difficult, and 4 = very difficult)?” Relationship strain was measured by two indicators. Perceived relationship strain was measured as the mean of four questions (α= 0.81). Respondents were asked how often their spouse (a) makes too many demands on them, (b) criticizes them, (c) lets them down when they were counting on him/her, and (d) gets on their nerves in the past year (1 = not at all, 2 = a little, 3 = some, and 4 = and a lot). Perceived housework unfairness was measured by the question: “How fair do you think your arrangement of household chores is to you (1 = very fair, 2 = somewhat fair, 3 = somewhat unfair, and 4 = very unfair)?”

Moderating Variables

The respondent’s gender was measured as a dichotomous variable (0 = men, 1 = women). Parental status was measured by four dummy variables indicating the age of youngest child living in the household including (a) no child (reference), (b) child under age 6, (c) child aged 6 to 17, and (d) child aged 18 or above.

Control Variables

Respondent’s age was measured in years. Race and ethnicity was measured as four dichotomous variables including White (reference), Black, Hispanic, and Other race. Union status was measured as a dichotomous variable (1 = cohabiting, and 0 = married) and was labeled as cohabitation in analyses. We included the number of children in the household as a control because it was related to the age of youngest child living in the household. Education was measured as an ordered variable (1 = no school or some grade school, 2 = eighth grade or junior high school, 3 = some high school, no diploma or GED, 4 = GED, 5 = graduated from high school, 6 = 1–2 years of college, no degree yet, 7 = 3 or more years of college, no degree yet, 8 = graduated from 2-year college or vocational school, or associates degree, 9 = graduated from a 4- or 5-year college, or bachelor’s degree, 10 = some graduate school, 11 = Master’s degree, and 12 = Ph.D., Ed.D., M.D., D.D.S, L.L.B., L.L.D, J.D. or other professional degree). Respondent’s poor physical health was measured by the question, “In general, would you say your physical health is 1 = excellent, 2 = very good, 3 = good, 4 = fair, or 5 = poor?” Similarly, respondent’s poor mental health was measured by the question, “Would you say your mental or emotional health is 1 = excellent, 2 = very good, 3 = good, 4 = fair, or 5 = poor?”

Analytic Approach

We first examined bivariate associations between their spouses’ problems and respondents’ FWC as well as mediator variables, using ordinary least squares (OLS) regression models or, for respondents’ share of household income only, a multinomial logistic (ML) regression model. Then we conducted multivariate analyses using OLS to examine whether the problem of one’s spouse were related to one’s own FWC controlling for other variables. We also explored ordered logistic regression models and found very similar patterns of results (not shown). We presented results using OLS regressions. DeMaris (2004) suggests OLS models over ordered logistic regressions when dependent variables have five or more categories. Many prior studies used OLS regression models (e.g., Glavin and Peters 2015; Grzywacz and Marks 2000; Marks 1998; Pearlin et al. 1997). We presented six models for each spousal problem. Model 1 examined the association between their spouses’ problems and respondents’ FWC with controls. Model 2 added time allocation variables, household income, and respondents’ share of household income. Model 3 added strain-based stressors, which were role overload, financial strain, and relationship strain, to Model 2 to examine whether these factors mediate the association between their spouses’ problems and the respondents’ FWC net of time respondents and their spouse spent on paid work and housework as well as household income and respondents’ relative contribution to household income. To examine variation by gender, Model 4 included interactions between respondents’ gender and their spouses’ problems on respondents’ FWC. To examine variations by parental status, Model 5 included interaction terms between respondents’ parental status and their spouses’ problems. Model 6 added a three-way interaction term among respondents’ gender, their parental status, and their spouses’ problems.

The mediation effect was examined using Sobel tests (MacKinnon et al. 2002). Sobel tests evaluate the effects of a mediating variable (MV) on the relationship between the independent variable (IV) and the dependent variable (DV). First, there must be significant associations between the IV and the DV, between the IV and the MV, and between the MV and the DV. Then, the effect of a mediating variable is τ − τ′ divided by τ, where τ is the IV coefficient in the relationship between the IV and the DV and τ′ is the IV coefficient in the relationship between the IV and the DV controlling for the effect of the MV. If the IV coefficient τ′ in the model including the MV did not differ significantly from zero, the association between the IV and DV is completely mediated by the MV. As MacKinnon et al. (2002) showed, τ − τ′ is analogous to αβ, where α represents the IV coefficient in the relationship between the IV and the MV, β represents the IV coefficient in the relationship between the MV and the DV. The Sobel test statistic (z) is calculated as αβ divided by \(\sqrt {{a^2}\sigma _{\beta }^{2}+{\beta ^2}\sigma _{\alpha }^{2}}\) where \(\sigma _{\beta }^{2}\) represents variance of β and \(\sigma _{\alpha }^{2}\) represents variance of α. By comparing the Sobel test statistics, we examined whether a specific MV in the present analysis would appear to play a key role in mediating the link between the IV (respondents’ spouses’ problems) and the DV (respondents’ FWC).

Some variables had a small percentage of missing cases. Missing data were handled using multivariate imputation by chained equations (MICE), using the Stata command ice (Royston 2006). MICE estimates missing data conditionally upon information from DV and IV (White et al. 2011). Estimations of missing data are distribution-specific (e.g., ordinal variables were modeled using logistic regression equations) and are constrained within the original values. Following Harel (2009), we calculated R 2 by transforming the R 2 values from each of the five imputed data sets into standard (z) scores, averaging these five z-scores, and transforming the averaged z-score into a R 2 value. All analyses used the weight variable provided by the 2011 MIDUS-RS to adjust for the sampling design and probability of selection into the 2011 MIDUS-RS including education, age, and race.

Results

Descriptive statistics for all variables were presented in Table 1. The mean age was 44.05 years old. A majority of respondents (84.2%) had children living in the household. About 8% were cohabiting. A majority were non-Hispanic White (85.7%). The mean education level was “some college.” The mean score for FWC was 2.23 (ranging 1–5). The mean spouse’s poor physical health score was 2.36 (ranging 1–5) and the mean spouse’s poor mental health score was 2.21 (ranging 1–5), whereas the mean spouse’s behavioral disorder score was 0.43 (ranging 0–6). About one in seven respondents (14.2%) reported their spouse having fair or poor physical health, one in nine (11.7%) reported their spouse having fair or poor mental health (not shown), and over a quarter (27.5%) reported their spouse having at least one type of behavioral disorders (not shown).

Table 1 Weighted descriptive statistics of variables (N = 980)
Full size table

Table 2 presents OLS or ML regression coefficients for the associations between respondents’ spouses’ problems and potential mediator variables as well as respondents’ FWC without including control or other mediating variables. All types of respondents’ spouses’ problems were negatively related to spouses’ hours spent on paid work and household income. Apart from spouse’s behavioral disorders and their housework, respondents’ spouses’ problems were not related to other time-based stressors. Respondents whose spouse had poor mental health were less likely to depend on their spouses’ contribution to household income. All three types of respondents’ spouses’ problems were related to higher levels of all four strain-based stressors that respondents experienced—role overload, financial strain, relationship strain, and unfairness in division of labor—except that their spouses’ behavioral disorders were not related to respondents’ perceptions of unfairness in division of labor.

Table 2 Ordinary-least-squares (OLS) or multinomial regressions models for the bivariate association between spousal problems, potential mediating variables, and respondents’ FWC
Full size table

All three types of respondents’ spouses’ problems were related to higher levels of respondents’ FWC at the bivariate level. The effect size of their spouses’ poor physical health on respondents’ FWC was 0.14 (i.e., 0.099/0.715), small but not negligible according to Cohen (1988). This means that, on average, each unit increase in respondents’ spouses’ poor physical health was related to an increase in respondents’ FWC by 0.14 standard deviations. For respondents’ spouses’ mental health, the effect size was medium (0.23). On average, each unit increase in their spouses’ poor mental health was related to an increase in respondents’ FWC by 0.23 standard deviations. Finally, the effect size for respondents’ spouses’ behavioral disorders was small to medium in size (0.18). Thus, each unit increase in their spouses’ behavioral disorders, on average, was related to an increase in respondents’ FWC by 0.18 standard deviations.

Spouse’s Physical Health and Respondent’s FWC

Turning to multivariate analyses, first we examined the association between their spouses’ poor physical health and respondents’ FWC (Table 3). Respondents’ spouses’ poor physical health was positively associated with their own FWC, after including control variables in the model (Model 1). Including respondents’ and their spouses’ hours spent in paid work and housework, household income, and respondents’ share of household income in the model (Model 2) added very little to the reduction of the coefficient for spouses’ poor physical health (about 15% altogether). When strain-based stressors were controlled for in Model 3, the coefficient for the effects of spouses’ poor physical health on respondents’ FWC was reduced by 73% from the first model. Role overload and relationship strain were both positively related to FWC. Supplemental analyses (not shown) suggested that financial strain was related to higher FWC when the model did not control for relationship strain. Sobel tests showed that, in Model 3, the link between their spouses’ poor physical health and respondents’ FWC was largely mediated by relationship strain (34.5%; z = 2.39, p < 0.05). Role overload and financial strain failed to meet the conditions for mediation. Supplemental analyses (not shown) suggested that role overload and financial strain met the conditions for mediation when relationship strain was not in the model. Models 4–6 evaluated the interaction between respondents’ gender and their spouses’ physical health, the interaction between respondents’ parental status and their spouses’ physical health, and the three-way interaction among respondents’ gender, their parental status, and their spouses’ physical health respectively. These interactions were not significant. We also examined whether the mediation factor differed by gender and parental status by conducting Sobel tests for men and women separately as well as for the four groups of parental status separately (data not shown). We found that the mediating role of relationship strain was significant for men (37.5%; z = 2.10, p < 0.05) but not for women (z = 1.36, p > 0.05), largely because the link between relationship strain and FWC was significant for men but not for women. There were no significant differences in the role of relationship strain as a mediator by parental status.

Table 3 OLS regression models predicting the association between spouse’s physical health problems and family-to-work conflict (N = 980)
Full size table

Spouse’s Mental Health and Respondents’ FWC

Next, the relationship between their spouses’ poor mental health and respondents’ FWC was assessed (Table 4). With control variables, respondents’ spouses’ poor mental health was significantly associated with their own FWC (Model 1). Adding respondents’ and their spouses’ time allocation variables and household income and respondents’ share of household income to Model 1 (Model 2) reduced the magnitude of the coefficient for respondents’ spouses’ poor mental health modestly (about 10% altogether). When indicators of strain-based stressors were included in Model 3, the size of the coefficient for spouses’ poor mental health was 82% lower from the first model and was no longer significant. Sobel tests within this model suggested that the association between their spouses’ poor mental health and respondents’ FWC was largely mediated by relationship strain (56.8%; z = 2.46, p < 0.05). Role overload also met the mediation conditions (46.8%; z = 2.09, p < 0.05). Supplemental analyses (not shown) suggested that financial strain met the conditions for mediation when relationship strain was not included in the model. There were no significant variations in these associations by gender and/or parental status (Models 4–6). We conducted supplemental analyses similar to those we did for respondents’ spouses’ poor physical health and calculated Sobel tests by gender and parental status separately (not shown). We found that, again, the mediating role of relationship strain was significant for men (70.5%; z = 2.14, p < 0.05) but not for women (z = 1.47, p > 0.05). Role overload met the mediation conditions for those without children in the household (50.6%; z = 2.15, p < 0.05) but not for those in the other three categories of parental status, largely because respondents’ spouses’ poor mental health was only related to higher role overload for households without children.

Table 4 OLS regression models predicting the association between spouse’s mental health problems and family-to-work conflict (N = 980)
Full size table

Spouse’s Behavioral Disorders and Respondent’s FWC

Finally, we examined the association between their spouses’ behavioral disorders and respondents’ FWC (Table 5). With control variables, their spouse’s behavioral disorders scale was significantly associated with respondents’ FWC (Model 1). Adding respondents’ and their spouses’ hours spent in paid work and housework, household income, and respondents’ share of household income reduced the coefficient for respondents’ spouses’ behavioral disorders about 10% (Model 2). When indicators of strain-based demands were included in Model 3, the size of the coefficient for spouses’ behavioral disorders was 94% lower from Model 1 and no longer significant. Sobel tests suggested that the association between their spouses’ behavioral disorders and respondents’ FWC was largely mediated by relationship strain (67.5%; z = 2.50, p < 0.05). Except for the three-way interaction (Model 6), there were no significant variations in these associations by gender and/or parental status (Models 4 and 5). The association between their spouses’ behavioral disorders and respondents’ FWC appears to be stronger for women who were in households with children aged 0–5 than for other groups. This finding was inconsistent with our hypothesis. Supplemental analyses (not shown) suggested that the mediating role of relationship strain was significant for men (32.4%; z = 1.96, p < 0.05) but not for women (z = 1.69, p > 0.05), whereas there were no significant differences in the role of relationship strain as a mediator by the categories of parental status.

Table 5 OLS regression models predicting the association between spouse’s behavioral disorders and family-to-work conflict (N = 980)
Full size table

Post Hoc Analyses

To assess robustness of the findings, we conducted series of supplemental analyses (not shown). First, we included the three aspects of respondents’ spouses’ problems—poor physical health, poor mental health, and behavioral disorders—in the same model. Only respondents’ spouses’ poor mental health was related to respondents’ FWC. The significant association disappeared when strain-based demands were included in the model. Sobel tests suggested that the association between their spouses’ poor mental health and respondents’ FWC was mediated by relationship strain (69.5%; z = 2.33, p < 0.05) with no gender/parental status differences. Second, we subdivided respondents’ spouses’ behavioral disorders into (a) poor economic ability (i.e., financial problems, problems at school or work, difficulty finding or keeping a job; α = 0.49) and (b) non-economic difficulties (i.e., alcohol or substances problems, legal problems, difficulty getting along with others; α = 0.33). We found that both types of their spouses’ behavioral disorders were positively related to respondents’ FWC when examined separately (not shown) but only poor economic ability was positively associated with respondents’ FWC when both indicators were modeled together. The association was, again, mediated by relationship strain (54.1%; z = 2.34, p < 0.05) with no gender/parental status differences. Third, we created a scale of spouse’s problems by summing standardized versions of physical health, mental health, and behavioral disorders and examined the same models discussed above. We found patterns of results that were similar to those found for respondents’ spouses’ mental health, including findings for mediation with relationship strain (49.1%; z = 2.40, p < 0.05) with no gender/parental status differences.

Discussion

Using a nationally representative sample of employed married or cohabiting adults, this paper examined the associations between three types of respondents’ spouses’ problems—poor physical health, poor mental health, and behavioral disorders—and respondents’ FWC. Drawing from the concept of stress proliferation in the stress process model (Pearlin and Bierman 2013), we conceptualized that these associations would be mainly mediated by strain-based stressors, including role overload, financial strain, and relationship strain. Further, we examined whether these associations might vary by respondents’ gender and parental status. We have several key findings.

We found that all three types of their spouses’ problems were positively related to respondents’ FWC. These findings expand prior research that found caregiving to an ill spouse was related to higher FWC among employed adults in particular groups, such as AIDS caregivers in San Francisco and Los Angeles or adults in their early 50s who were in the WLS (Marks 1998; Pearlin et al. 1997). In supplemental analyses, we found that their spouses’ mental health problems, which are rarely discussed in prior research on work-family conflict, have the strongest association with respondents’ FWC among the three types of spousal problems we examined. Future research is warranted to investigate the prevalence of having a spouse who has mental health problems among working-age US adults. One’s spouse’s behavioral disorders, another set of potential family demands that have been ignored in work-family research, too, warrant future research. In our sample, we found that more than one-quarter of respondents reported that their spouses had at least one of the six symptoms of behavioral disorders in the past 12 months. It is important to pay more attention to implications of having a spouse who has symptoms of behavioral disorders for employed adults’ ability to maintain their work-family balance.

Our findings suggest that the association between one’s spouse’s problems and one’s FWC is largely mediated by one’s perceptions of relationship strain. The three strain-based stressors, role overload, financial strain, and relationship strain, were closely related to one another. Each stressor by itself reduced the magnitude of the association between spousal problems and respondents’ FWC. Yet, when all of these three variables were in the model, relationship strain appeared to be the strongest mediator for the association between their spouses’ problems and respondents’ FWC. It is noteworthy that role overload was the strongest predictor of FWC among the three strains, but relationship strain was the strongest mediator for the link between their spouses’ problems and respondents’ FWC. These results suggest the importance of assessing mediation using statistical tests. All in all, these findings suggest that understanding what would buffer the link between spousal problems and relationship strain may be the key to diminish the stress proliferation process from respondents’ spouses’ problems to their own FWC.

Contrary to our predictions, the association between their spouses’ problems and respondents’ FWC varied little by respondents’ gender and parental status. We expected that this relationship would be greater for men than women, and especially for fathers with minor children, as men are less likely than women to expect to shoulder family responsibilities and thus may be more vulnerable to stressful circumstances caused by their spouses’ problems. Our hypothesis was not supported. Instead, we found that the association between their spouses’ behavioral problems and respondents’ FWC was greater for women living with children aged 0–5 than men or women in other parental status categories. We suggest that having a spouse who has behavioral disorders may be more stressful when there are younger children in the household. More research is needed to interpret why this is especially the case for mothers than fathers.

More robust gender differences were found in mediation variables. Relationship strain appeared to play a stronger role for men than women in mediating the link between their spouses’ problems and their own FWC. This was largely because the association between relationship strain and FWC was stronger for men than for women, which is inconsistent with prior research which suggests that the association between family conflict and FWC is stronger for women than men (Duxbury and Higgins 1991) or other research which suggests no gender differences in the association between spousal disagreement and FWC (Grzywacz and Marks 2000). Prior research has examined gender differences in levels of FWC and predictors of FWC (e.g., Grzywacz and Marks 2000), as we did in the present analysis. The present findings found gender differences in mediators, suggesting that more research is warranted to investigate gender differences in the mechanisms through which particular family demands are linked to FWC.

The present analysis has limitations that future research should address. First, measures of spousal problems should be improved. Future research should utilize measures that can capture the magnitude of the problems of one spouse and the degree of caregiving demands placed on the other spouse. In particular, the behavioral disorders scale we used has a weak internal reliability and thus warrants future improvement. Second, we used the concept of stress proliferation from the stress process model to conceptualize the association between their spouses’ problems and respondents’ FWC. Yet, because our data are cross-sectional, we were unable to empirically identify which stressor was the initial stressor. It is possible that relationship strain is the initial stressor: Spouses who are less satisfied about the quality of their relationship may see their spouses negatively and report that their spouses have problems. In addition, the observed associations among respondents’ perceptions of their spouses having problems, relationship strain, and FWC may be all shaped by a third variable that was not controlled for in the analysis, such as personality. Longitudinal data would allow researchers to identify the causal chain of stressors in the stress proliferation process and to eliminate omitted variable biases. Finally, as the stress process model emphasizes (Pearlin and Bierman 2013), future research should investigate how differences in financial, social, or psychological resources may buffer stress proliferation across spousal problems, relationship quality, and respondents’ FWC.

This study documents associations between one spouse’s problems, including physical health, mental health, and behavioral disorders, and the other spouse’s FWC, with one’s spouse’s mental health problems having the strongest association. These findings are important given that mental health problems as well as behavioral disorders tend to be neglected in policy discussions. Relationship strain is central as a mediating factor, especially when a man has a wife with problems. These findings suggest a merit of further investigation of their spouses problems as an important source of family responsibilities that employed adults may have and gender differences in the mechanisms through which one’s spouse’s problems are linked to stressfulness in balancing work and family life.