Marriage,Relationship Quality,and Sleep among U.S. Older Adults

Material and Time Resources

Marriage brings material resources and time investments in each other. Thus, the married tend to gain greater financial well-being and companionship through increasing time in shared activities, such as eating, leisure activities, and sexuality, compared to the unmarried (Waite 1995; Waite and Gallagher 2001). The better material conditions of the married reduce exposure to many kinds of risks and stress and increase their ability to deal with stressful or unexpected events. Material resources tend to buffer against or decrease health consequences of unexpected events. Many of these benefits also accrue to cohabiting couples, although perhaps to a lesser extent (Waite and Gallagher 2001). Recent research has also pointed to the benefits of marriage for the order and cleanliness of the household through greater role clarity and specialization compared to other living arrangements (Cornwell 2014), suggesting a favorable physical environment as another resource that is available for married people.

The greater resources that tend to be available to the married can be deployed to achieve better physical and mental health, which may improve sleep outcomes. Married individuals have better physical health outcomes than the unmarried, including fewer chronic conditions and mobility limitations at midlife and lower levels of both depressive symptoms (Hughes and Waite 2009) and loneliness (Hawkley et al. 2008). Because many chronic diseases or conditions affect sleep quantity and quality, older adults with fewer medical problems are less likely to report insomnia symptoms (Foley et al. 2004). In addition, marriage reduces risky health behaviors. Having a spouse to look after you matters for health behaviors. Prior studies observe a reduction of excess alcohol consumption and tobacco use associated with marriage (Duncan, Wilkerson, and England 2006; Fleming, White, and Catalano 2010; Leonard and Rothbard 1999). Clinical and epidemiological studies indicate that regular alcohol abuse and smoking can disrupt sleep (Roehrs and Roth 2001; Zhang et al. 2006). The lower level of risky health behaviors may also lead to better physical outcomes at older ages, which in turn may promote better sleep. Taken together, the spousal relationship is expected to improve older adults’ sleep outcomes because the material and time resources it brings into the relationship promote better physical health and health-related behaviors.

Furthermore, the resource model suggests that the spousal relationship directly influences the housing environment in which older adults live. Married people tend to live in cleaner, quieter, tidier houses in better repair than the unmarried (Cornwell 2014). A disordered home environment can directly affect the residents’ sleep. In addition, married people are wealthier than the unmarried (Waite 1995; Wilmoth and Koso 2002) such that they tend to live in better neighborhoods. A large literature has documented the importance of neighborhood social conditions for individual health (Ross and Mirowsky 2001), including sleep (Hill et al. 2009). As such, the environment of the home itself and of the neighborhood in which it is situated may be an important mechanism through which marriage promotes better sleep.

Emotional and Social Resources

A second set of resources brought by marriage is emotional support and social integration. Aside from material, time, and physical resources, the better health outcomes of married people also come from the provision of a sense of companionship and belonging. Married people are much more likely to have a confidant than the unmarried and to have someone who listens to them and is concerned with their health and well-being. For a married man, this person is usually his wife (Umberson 1992). Being able to talk to another person about things that are bothering oneself reduces the effects of stress for both mental and physical health (Holt-Lunstadt, Birmingham, and Jones 2008; Thoits 2011). Because individuals’ social networks shrink with age, older adults often have relatively few confidants (McPherson, Smith-Lovin, and Brashears 2006; Schnittker 2007). Older adults are less close to their network members, on average, and have fewer nonprimary group ties than younger adults (Cornwell, Laumann, and Schumm 2008). Thus, older adults tend to be more dependent on their spouse or partner for social support and companionship, and reliance on marriage or marriage-related networks increases with age (Curran, McLanahan, and Knab 2003). As the primary focus of social relationships gradually shifts with age, spousal relationships play a vital role in providing instrumental and emotional support and a sense of social integration. In fact, a large literature has pointed to the importance of spousal relationships for the health of older adults (Warner and Kelley-Moore 2012). As for younger adults, those who are married or live with a partner are in general healthier and happier and live longer due in part to the emotional support and companionship that marriage provides (Hughes and Waite 2002; Lund et al. 2002; Michael et al. 2001; Sarwari et al. 1998).

The provision of emotional support makes marriage an important determinant of psychological well-being (Brown 2000; Gove, Hughes, and Style 1983; Kim and McKenry 2002), which in turn promotes good sleep at night. A spouse or partner also offers companionship, which tends to reduce feelings of loneliness, depressive symptoms, and social isolation (Hawkley et al. 2008). Clinical and epidemiological studies demonstrate that loneliness undermines sleep quality (Cacioppo et al. 2002; Kurina et al. 2011). The companionship provided by a spouse may support better sleep. Furthermore, a spouse or partner may provide emotional support, a sense of security, or information that buffers the effects of stressful events (Thoits 2011), which seems to be particularly the case for mental health outcomes (Pearlin and Johnson 1977). While a lack of sleep may contribute to depression, it is well documented that depression leads to sleep problems (Lopresti, Hood, and Drummond 2013). Additionally, given that insomnia is an important symptom of poor mental health (Harvey 2001), this may be an especially important pathway from marriage to sleep outcomes: marriage may affect sleep via its influence on psychological well-being and mental health. Finally, a high level of social integration is tied to greater social control, which can promote positive health behaviors as well as better sleep habits (Umberson 1987), which may also promote better sleep outcomes of older adults.

Self-reported Sleep Characteristics

Survey measures of sleep characteristics included self-reported hours of sleep and sleep quality. First, each respondent was asked to report his or her usual bedtime and wake-up time separately for weekdays and weekends: “What time do you usually go to bed and start trying to fall asleep?” and “What time do you usually wake up?” ² Using this information, we calculated each respondent’s usual sleep duration for weekdays and weekends separately. Next, we multiplied weekday sleep duration by 5 and weekend sleep duration by 2 and divided by 7 to obtain an estimate of self-reported average weekly total sleep time. Second, NSHAP also included four questions on sleep quality, each asking about the frequency of a sleep characteristic. These four questions, drawn from the sleep questions module in the Health and Retirement Study (HRS), were “How often do you have trouble falling asleep?” “How often do you have trouble with waking up during the night?” “How often do you have trouble with waking up too early and not being able to fall asleep again?” and “How often do you feel really rested when you wake up in the morning?” Response categories for these questions were “most of the time” = 2, “sometimes” = 1, and “rarely or never” = 0. Factor analysis showed that these four items loaded onto a single factor. We created a troubled sleep scale, ranging from 0 to 8, by summing all items (after reverse coding the question about feeling really rested). We standardized the score based on our final sample. The troubled sleep scale has a Cronbach’s alpha of .66.

Actigraph-estimated Sleep Characteristics

NSHAP used actigraphy to measure sleep characteristics of respondents. Actigraphy is a valid and unobtrusive approach to measuring sleep without disrupting routine behavior; sleep characteristics measured by actigraphy are moderately to highly correlated with those measured by polysomnography (Ancoli-Israel et al. 2003; Blackwell et al. 2008). The Actiwatch (Philips/Respironics, Andover, MA) records intensity and frequency of movement using a piezoelectric linear accelerometer with 15-second epochs. The Actiwatch continually registers wrist movements, and the sum of all wrist movements during each 15-second epoch is saved as an activity score. Data from the Actiwatch were downloaded and analyzed using the manufacturer’s Actiware software Version 5.59 (Philips/Respironics 2010). In the present study, we focused on four actigraph-estimated sleep characteristics that have often been used in previous studies (Chen et al. 2015; Lauderdale et al. 2014): (1) total sleep time (defined as the total duration of all epochs scored as sleep within the major sleep interval, that is, the time from the first epoch scored as sleep to the last epoch scored as sleep for the primary sleep interval in each 24 hours); (2) percentage sleep (defined as the percentage of the sleep interval that is actual sleep); (3) sleep fragmentation index (ranging from 0 to 100), an indicator of sleep disruption that is the sum of two percentages: the percentage of the sleep interval spent moving and the percentage of immobile periods (i.e., contiguous epochs with no movement) that are no longer than one minute; and (4) wake after sleep onset (defined as the total minutes awake during the sleep interval). Percentage sleep, sleep fragmentation, and wake after sleep onset are sleep quality indicators as measured by the actigraph. For each of the actigraph-estimated sleep characteristics, we calculated an average value over three nights for each respondent.

Marital Status and Relationship Quality

Each respondent was asked whether he or she lived with a spouse or unmarried partner in the household. The answer was used to create a binary indicator of marital status. Since cohabitation was rare in this population (n = 25, 3.5%), we combine the cohabiting with the married and simply refer to them all as “married.” We also performed sensitivity analyses that omitted the cohabiting. We followed the approach developed by Warner and Kelley-Moore (2012) and modified by Galinsky and Waite (2014) to create two factor scales of marital quality. These two scales were derived from nine items. Each married respondent answered a series of questions assessing his or her marital relationship. These are (1) “How often can you open up to your spouse if you need to talk about your worries?” (2) “How often can you rely on your spouse for help if you have a problem?” (3) “How often does your spouse make too many demands on you?” (4) “How often does your spouse criticize you?” and (5) “How often does your spouse get on your nerves?” The four response categories ranged from hardly ever or “rarely” = 1 to “often” = 4.

In addition, all partnered or married respondents were asked (6) how close they felt their relationship with their spouse was, with answer options “not very close,” “somewhat close,” “very close,” and “extremely close;” (7) how happy their relationship with their spouse was, with answers ranging from “very unhappy” (coded as ‘1’) to “very happy” (coded as ‘7’); (8) how they liked to spend their free time, with options including “doing things together with the spouse,” “doing some things together,” and “doing things separately”; and (9) how emotionally satisfying they found their sexual relationship with their spouse or partner to be, with options of “extremely satisfied,” “very satisfied,” “moderately satisfied,” “slightly satisfied,” and “not at all satisfied” (Kim and Waite 2014).

To form relationship quality scales, we performed exploratory factor analysis, which loaded these nine items into two factors. We referred to the first factor as positive marital relationship (α = .68) and the second factor as negative marital relationship (α = .58). The positive relationship scale included six items: (1) rely on partner, (2) open up to partner, (3) relationship happiness, (4) relationship closeness, (5) emotional satisfaction, and (6) time with partner. The negative relationship scale included three items: (1) partner criticizing, (2) making demands, and (3) getting on nerves. Two factor scales were derived from the factor analysis.

Physical Health Indicators

We have included three indicators of older adults’ physical health: self-rated health, a modified Charlson comorbidity index, and a frailty scale. First, the self-rated health was assessed with the question, “Would you say your health is excellent, very good, good, fair, or poor?” We treated the variable as a continuous measure. Higher values indicated better health. Second, NSHAP included a wealth of information on respondents’ chronic conditions. During the interview, each participant was asked, “Has a medical doctor told you have (had) [condition]?” The list of conditions included heart attack/myocardial infarction, congestive heart failure, procedure for coronary artery disease, stroke/cerebrovascular disease, diabetes, leukemia/lymphoma, metastatic cancer, COPD/asthma, rheumatoid arthritis, and Alzheimer’s dementia or another form of dementia. These chronic conditions were combined into a modified Charlson comorbidity index following the procedure described by Vasilopoulos et al. (2014). The original Charlson comorbidity index was developed to assess the severity of comorbid conditions based on its power to predict mortality. Conditions that were associated with higher mortality were given more weights in the calculation of the index score (Charlson et al. 1987). The index can be produced by review of medical record or questionnaire. The questionnaire version of the Charlson index is highly correlated with the medical record–based version and allows for the assessment of the comorbidity of the general population of older adults (Katz et al. 1996). Our modified index ranged from 0 to 16. Higher values indicated more serious conditions with higher mortality risk. Finally, there are several functional assessments in the NSHAP survey. Respondents were asked how often over the past week they felt (a) everything was an effort and (b) they could not get going. Response categories included “rarely/none of the time,” “some of the time,” “occasionally,” and “most of the time.” Respondents who answered “occasionally” or “most of the time” to at least one of the questions were coded as having exhaustion. Respondents were also asked to report their frequency of participation in vigorous activity or exercise (defined as 30 minutes or more of sports, exercise classes, or physical labor). Answer choices included “five or more times per week,” “three or four times per week,” “one to two times per week,” “one to three times per month,” “less than one time per month,” and “never.” Respondents were considered as low physical activity if they reported “one to three times per month,” “less than one time per month,” or “never.” In addition to the two survey questions, NHSAP included two direct assessments of respondents’ functional status: gait speed assessment and chair stand assessment. Detailed information of these two tests was described elsewhere (Huisingh-Scheetz et al. 2014). Drawing information from these survey questions and assessments, we created a four-point frailty scale (Huisingh-Scheetz et al. 2014). Higher scores indicated a greater level of frailty.

Psychosocial Factors

The present study focused on four psychosocial factors: self-rated mental health, anxiety, depressive symptoms, and loneliness (Payne et al. 2014). First, respondents were asked to rate their mental health as “excellent,” “very good,” “good,” “fair,” or “poor.” We treated this as a continuous variable, recoding it so that higher values indicated better mental health. Second, NSHAP included seven anxiety items from the Hospital Anxiety and Depression Scale (Zigmond and Snaith 1983). Higher values indicated high level of anxiety. Next, NSHAP included a shortened version of the Center for Epidemiologic Studies Depression Scale (CESD-10). The CESD-10 score ranges from 0 to 30 and is a widely used instrument for the identification of major depression in older adults (Irwin, Artim, and Oxman 1999). Finally, NSHAP included items from the short loneliness scale developed by Hughes et al. (2004). Detailed analysis of these scales can be found in a recent study by Payne et al. (2014).

Household Environment

Last, we included an indicator of household disorder. At the end of the interview, every interviewer was instructed to rate the respondent’s housing conditions on the following dimensions: cleanliness, neatness and tidiness, noise, spaciousness, clutter, and odor. Ratings ranged from 1 to 5 for each dimension. We constructed a household disorder scale by summing all these items (positive items were reverse coded). This scale is a modified version of household disorder scale developed by Cornwell (2014) and has high reliability.

Other Covariates

Respondent’s education was categorized as less than high school, high school or equivalent, some college education, or bachelor’s degree or higher. Race-ethnicity distinguished white, African American, Hispanic, and other. A dichotomous variable indicated whether the respondent was retired at the time of the interview. Log household income and log household assets were used as indicators of respondents’ economic standing. We also controlled for gender and age. ³

Analytic Strategy

We began with weighted descriptive statistics of older adults in the NSHAP sample. Next, we examined the relationship between marital status, relationship quality, and sleep characteristics adjusted for social and demographic characteristics using survey weighted ordinary least square regressions and clustered by families to account for the nesting of individuals within households. Because members of the same household (in this case, a husband and a wife) influence each other, this might violate the assumption of independent observations in traditional linear models, which results in underestimation of standard errors. However, multilevel models were not ideal for NSHAP data for two main reasons. First, NSHAP data are not multilevel by design. Household is not a unit in the sample selection process. In the second wave, the survey team sampled eligible older adults, and if the selected respondent lived with a spouse or a partner, he or she was also invited to participate in the survey. As such, NSHAP did not have weights for households. Furthermore, since the sampling unit was individuals, not households, the average number of Level 1 observations per Level 2 unit was very small. Here we had 727 older adults residing in 585 households. This resulted in an average of only 1.2 observations per household, leading to “data sparseness” when fitting the multilevel model. The problem of data sparseness can lead to overestimation of standard errors and reduce the ability to detect significant associations between explanatory variables and outcome variables (Clarke 2008; Clarke and Weaton 2007). Because of these issues, instead of fitting the data using multilevel models, we employed survey weighted least square regression models with clustering by households.

We performed two sets of analyses. The first set assessed the associations between marital status and self-reported and actigraph-estimated sleep characteristics. The second set of analyses examined the associations between relationship quality (as measured by positive relationship and negative relationship scales) and sleep characteristics among those married or partnered. We used multiple imputation to account for potential biases resulting from missing data in the control variables and mediators. Multiple imputation involved replacing missing values with predictions based on other observed variables using Monte Carlo techniques (Rubin 1987). In contrast to single imputation, which replaces each missing value with a predicted value, multiple imputation replaces several missing values with repeated imputations using available covariates, creating several complete data sets. The combined results produced better estimates of the missing values that created uncertainty around the missing data (Allison 2001; Hawkley et al. 2014). All regression analyses were survey weighted using Stata 13 (StataCorp 2011).