Socioeconomic Status and Sleep among Couples

Abstract

Objective/Background:

Lower socioeconomic status (SES) is generally associated with poor sleep but little is known about how different SES indices are associated with sleep duration and quality, or about these relations longitudinally or in cohabiting couples. The main objective was to examine longitudinal associations between multiple SES and sleep parameters in cohabiting adults.

Participants:

Participants were cohabiting couples (N = 135) of women (M age = 37.2 years, SD = 5.93; 76% White/European American, 18% Black/African American) and men (M = 39.9 years, SD = 7.33; 78% White, 18% Black).

Methods:

Men and women participated twice with a 1-year lag. At Time (T1), participants reported on multiple SES indices including their income, perceived economic well-being, education, employment status, and occupation. Sleep at T1 and T2 was assessed with self-reports and actigraphs (sleep duration from onset to wake time, %sleep from onset to wake, long wake episodes).

Results:

Actor effects on actigraphy-assessed sleep parameters were evident for both men and women; low SES was associated with shorter duration and poor quality (%sleep, long wake episodes) sleep. These associations were most pronounced for income-to-needs ratio (men and women) and perceived economic well-being (women only). Partner effects were also evident such that men’s employment status was associated with women’s longer sleep duration and greater sleep quality (%sleep) whereas women’s employment predicted increased subjective sleep problems for men.

Conclusion:

Findings illustrate the need to consider multiple SES and sleep indices, as well as the family context in studies addressing linkages between SES and sleep.

Wide and pervasive disparities in health exist across individuals in the U.S. with notable differences based on socioeconomic status (SES). Lower SES has been consistently associated with physical (Chen & Miller, 2013) and mental (Meyer, Castro-Schilo, & Aguilar-Gaxiola, 2014) health problems. Likewise, a rapidly developing body of evidence reveals that low SES is associated with sleep problems (Grandner, Williams, Knutson, Roberts, & Jean-Louis, 2016; Whinnery, Jackson, Rattanaumpawan, & Grandner, 2014). Poor sleep, including short sleep duration and sleep difficulties, is prevalent among adults with large percentage of individuals reporting obtaining < 6 hours of sleep (CDC, 2010) and having poor quality sleep (National Sleep Foundation, 2015). Cohabiting partners have related sleep duration and quality (Gunn, Buysse, Hasler, Begley, & Troxel, 2015; Lee et al., 2018), however, less is known about which SES indices are most strongly associated with sleep parameters or how SES influences sleep within the couple context. Given the importance of sleep to well-being, it is imperative to identify associations among various SES and sleep variables and to explicate factors within the family context that may influence sleep duration and quality over time.

Socioeconomic status is a multidimensional construct (Chen & Paterson, 2006; Hout, 2008) that indicates different facets of economic well-being and adversity (Krieger, Williams, & Moss, 1997), with the most common indicators being income, educational attainment, and occupation (Duncan, Daly, McDonough, & Williams, 2002). Income, or income-based assessments of SES, are often used when examining links between SES and health. Educational attainment and occupation reflect social standing and opportunity for the individual to make advances along the social ladder and obtain social prestige (Braveman et al., 2005). Assessments of perceived economic hardship (e.g., difficulty making ends meet, financial cutbacks) are less commonly used, yet are important for understanding economic pressure and stress on individuals, regardless of their income (Chen & Patterson, 2006).

Income is the most widely used SES index and its associations with self-reported sleep duration (Stamatakis, Kaplan, & Roberts, 2007; Whinnery et al., 2014) and quality (Grandner et al., 2016; Moore, Adler, Williams, & Jackson, 2002) have been established. Furthermore, it has been documented that unemployment status (Burgard & Ailshire, 2009) and low educational attainment (Grandner, Petrov, Rattanaumpawan, Jackson, Platt, & Patel, 2013; Patel, Grandner, Xie, Branas, & Gooneratne, 2010) are associated with poor sleep quality. Additionally, unemployment status and lower status occupations, have been associated with sleep problems (Jackson, Redline, Kawachi, Williams, & Hu, 2013; Grandner et al., 2010). Many pertinent studies, including the aforementioned, rely upon short self-reported assessments of sleep that vary from a single item to validated questionnaires.

Fewer relevant studies have utilized objective assessments of sleep (actigraphy, polysomnography) and in those that did, SES was not often examined with multiple parameters. In studies that used objective sleep assessments, lower income-to-needs was associated with shorter sleep duration (El-Sheikh, Keiley, Bagley & Chen, 2015) and poor sleep quality (Friedman et al., 2007; Lauderdale et al., 2006). Likewise, lower SES (composite of educational attainment and income) was associated with poor sleep quality indexed by polysomnography (Mezick et al., 2008). In a few studies, perceptions of economic hardship have been associated with poor sleep among women (El-Sheikh et al., 2015; Hall et al., 2009).

Although various SES and sleep indices have been examined in the literature, those that assessed multiple parameters of SES and sleep in the same study are rare. Further, investigations of longitudinal relations between SES and sleep, especially those that control for autoregressive effects of sleep and thus examine changes in sleep over time in the context of socioeconomic adversity, are almost nonexistent. Furthermore, the vast majority of relevant studies utilize subjective assessments of sleep where they dichotomize sleep duration (e.g., < 6 hours of sleep vs > 6 hours of sleep; Whinnery et al., 2014) and quality (e.g., yes/no responses for difficulties falling asleep, night waking, etc.; Patel et al., 2010). These studies have provided important evidence and the present investigation builds on this literature by examining multiple sleep variables along a continuum, which allows for sensitive assessments and capitalizing on the full variability in the sleep parameters that are otherwise missed if the variables are categorized (MacCallum, Zhang, Preacher, & Ruckher, 2002). Consistent with recommendations for utilizing multiple methods and parameters (Sadeh, 2015), sleep duration was measured with actigraphy, and sleep quality was assessed by both actigraphy and self-reports.

In the U.S., 65% of individuals report being married or cohabiting with a partner (Manning, Brown, & Stykes, 2014) yet research examining dyadic influences on sleep is limited (Gunn et al, 2015; Troxel, 2010). Studies examining couples have shown that an individual’s sleep is related to the partner’s sleep duration and quality (Gunn et al., 2015; Kouros & El-Sheikh, 2017; Lee et al., 2018). Further, marital experiences of a couple including conflict and dissatisfaction can have a negative influence on their sleep (El-Sheikh, Kelly, Koss, & Rauer, 2015; Kane, Slatcher, Reynolds, Repetti, & Robles, 2014; Troxel, Robles, Hall, & Buysse, 2007). In addition, some studies have reported that work-related stressors are associated with poor sleep in couples (Arber, Bote, & Meadows, 2009; Maume, Sebastian, & Bardo, 2010). However, studies that have examined couples’ sleep in the context of the broader socio-cultural milieu including socioeconomic adversity are scarce.

Present Study

The aim of the study was to examine several indices of SES variables as predictors of multiple sleep parameters in individuals and their partners over one year. Due to the interdependent nature of couple data, an actor-partner interdependence model (APIM) was utilized to analyze the data. Autoregressive effects for sleep were included in all models to allow for the examination of change in the sleep outcome over time as predicted by SES. We expected that lower SES would predict short sleep duration and poor sleep quality longitudinally for individuals and their partners (actor and partner effects). No specific predictions were made regarding differences in associations across the specific SES and sleep parameters and this aspect of the study was considered exploratory.

Method

Participants

Data for the current study come from the second and third waves of a larger longitudinal investigation examining sleep, health, and adjustment across middle to late childhood (Auburn University Sleep Study, 2010 – 2012). Actigraphic assessments of men’s and women’s sleep, as well as assessment of some SES indices, did not occur at T1 of the larger study and thus this wave was not included in analyses. The study was approved by the university’s institutional review board. Participants were initially recruited from public schools in the Southeastern United States (Alabama) through letters sent home with children where interested families were asked to call the investigators to schedule lab visits and sleep assessments. Eligibility criteria included couples who were married or cohabiting for at least 2 years with at least one school aged child at the time of recruitment.

For clarity, we refer to the two study waves utilized in this manuscript as T1 and T2. Given the focus of the study, and to ensure stability in family structure, men and women who were not continuously married or cohabiting across the study period were not included in the analytic sample. To reduce potential confounds, individuals who engaged in night shift work were excluded from the final sample.

The analytic sample included 135 cohabiting couples (90% married) at T1, where the average length of cohabitation was 11.54 years (SD = 7.33). Men and women’s mean ages were 39.9 years old (SD = 7.33) and 37.2 years old (SD = 5.93), respectively. Based on self-reports, a majority of the sample was White/European American (78% of men, 76% of women), 18% of men and 21% of women were Black/African American, and the rest (4% of men and 3% of women) reported other ethnicities. Couples were from diverse socioeconomic backgrounds with high representation of those living in or close to poverty (detail provided later). All couples in the sample self-identified as heterosexual. In the final analytic sample, 94% (N = 127) of couples who participated at T1 also participated at T2. No differences among primary study variables existed between couples lost to attrition and those who were retained across T1 and T2.

Procedures

At both waves, actigraphs were delivered to the families’ homes and both men and women were asked to wear them on their non-dominant wrist for seven consecutive nights. Men and women wore the actigraphs during the same week. To corroborate actigraphy data, each individual completed sleep diary logs nightly (Acebo & Carskadon, 2001). Upon completing actigraphy assessment, couples visited the research laboratory to complete questionnaires, typically on the day following the last night of actigraphy.

Socioeconomic status indices (SES).

Income-to-needs ratio (T1.)

Women reported annual family income according to the following categories: (a) $10,000 to $20,000; (b) $20,000 to $35,000; (c) $35,000 to $50,000; (d) $50,000 to $75,000; or (e) more than $75,000. Familial income and reported household size were used in deriving income-to-needs ratio, which was computed by dividing total family income by the federal poverty threshold for that family size in 2010, which was the year of data collection (U.S. Department of Commerce; www.commerce.gov). For example, the federal poverty line in 2010 was $22,050 for a family household of four individuals. For a family with a reported income between $20,000 and $35,000, their estimated income would be $27,500 (midpoint within the income bracket) and their income to-to-needs ratio would be 1.24, just above the poverty line. Approximately 27% of families were classified as poor and below the poverty line (ratio < 1); 29% were low income and lived near the poverty line (ratio between 1 and 2); 28% were classified as lower middle class (ratio between 2 and 3); 15% were middle class (ratio ≥3); and 1% were affluent and upper middle class (ratio ≥ 4) (Diemer, Mistry, Wadsworth, López, & Reimers, 2013).

Perceived Economic Well-Being (T1).

Men and women reported on their perceptions of economic well-being using the three well-established scales (Conger et al., 1992) of “can’t make ends meet,” “material needs,” and “financial cutbacks.” For “can’t make ends meet,” individuals rated how much they agreed with three statements concerning difficulty the family had in paying bills and other financial necessities over the past year (i.e., “Our income never seems to catch up with our expenses”; “Think back over the past year and tell us how much difficulty you had with paying your bills”). Per common practice, standardized scores were generated and averaged where higher scores represent less economic pressure. Individuals reported on “material needs” via 7 items that assess agreement with statements about their family’s economic situation using a 5-point scale (e.g., “My family has enough money to afford the kind of home we would like to have”; “We have enough to afford the kind of food we should have”; “My family has enough money to afford the kind of leisure and recreational activities we want to participate in”; “We have enough money to afford the kind of car we need”) where scores were reversed and higher scores indicate better economic well-being. Finally, participants reported on “financial cutbacks” by agreeing or disagreeing with 22 statements describing adjustments the family had to make over the last year due to financial need (e.g., “Reduced or eliminated medical insurance”; “Changed residence to save money”; “Fallen behind in paying bill”; “Postponed major household purchase”; “Reduced household utility use”). Higher scores indicate fewer financial cutbacks, thus a better economic situation.

Good internal consistency was observed for the three scales for men (αs =.80 - .93) and women (αs =.85 - .91). All three scales were moderately to highly correlated across individuals for men (rs = .60 - .65, p <.001) and women (rs = .43 - .67, p <.001), as well as correlated across men and women (rs = .63 - .75, p <. 001). The three subscales have demonstrated good reliability and predicative validity (Conger, 2000; Conger et al., 1992) and are typically aggregated to create a composite score (Conger & Conger, 2002). Thus, all subscales were standardized and summed to create a mean composite of perceived economic well-being for each partner, where higher scores indicate greater economic well-being.

Employment and Occupation (T1).

Men and women provided information about their employment status, job titles, and primary duties. Men were primarily employed (80%), and few identified as unemployed (8%) or non-workers (e.g., retired; 4%). A majority of women were employed (61%), while 14% were unemployed and 9% were non-workers. Across couples, 66% were dual earning and 34% had a single earner. For the purposes of data analysis, individuals were considered unemployed if they did not currently have a job, were a stay at-home caregiver, or were retired/disabled at the time of data collection. Employment status was missing for 16% of women and 8% of men.

Occupation type was coded using the 2010 Standard Occupation Classification (SOC) system and National Statistics Socio-Economic Classification (NS-SEC; Office for National Statistics, 2010), which categorized jobs into 7 categories based on job titles and duties. The SOC and NS-SEC are widely used measures in the literature examining SES across the world (Connelly, Gayle, & Lambert, 2016). Categories include managerial and professional occupations (e.g., doctor, office manager), supervisory and technical jobs such as skilled trade workers (e.g., plumber), factory and machine operators, as well as those with semi-routine (e.g., customer service), elementary and routine jobs (e.g., restaurant server, janitor). Occupation scores ranged from 1 to 7 where lower scores were indicative of more prestigious jobs. Men were primarily employed in lower managerial and professional occupations (34%), semi-routine and routine occupations (25%), and supervisory and technical work (12%). Women were primarily employed in lower managerial and professional occupations (38%), semi-routine occupations (6%), and intermediate occupations (e.g., administrative assistants; 18%). Participants who were unemployed or retired in the past year and provided information about their most recent occupations (n = 3; 2.2% men and women) were included in analysis.

Education (T1).

Men and women indicated their education level using the following categories: (a) < 7th grade (1% of women); (b) completion of 8th grade (0%); (c) 9th to 11th grade (4% men); (d) high school graduate (23% women; 40% men); (e) partial college or specialized training (33% women; 32% men); (f) bachelor’s degree (30% women; 15% men); or (g) graduate degree (13% women; 9% men). Education was examined as a continuous variable for men and women (1 = less than 7^th grade to 7 = graduate degree).

Sleep.

Objective sleep.

At T1 and T2, actigraphy was used to derive objective sleep parameters. Actigraphy is considered a reliable tool for measuring sleep in the home environment (Rupp & Balkin, 2011) especially when used for multiple consecutive nights (Acebo et al., 1999). Participants wore Octagonal Basic Motionlogger actigraphs, which measured motion in 1-min epochs using zero crossing mode. The Octagonal Motionlogger Interface with Actme software and Action W2 analysis software package (Ambulatory Monitoring, Ardsley, NY) were used. The established Cole-Kripke scoring algorithm (Cole, Kripke, Gruen, Mullaney, & Gillin, 1992) was utilized to derive the following sleep parameters as defined in the Ambulatory Monitoring manual: (a) sleep minutes - the number of minutes scored as sleep between sleep onset and wake time; (b) %sleep– the percentage of epochs scored as sleep between actigraphy-determined sleep onset and wake time; (c) long wake episodes (LWE) – the number of episodes (≥ 5 min each) scored as awake. Of note, the %leep variable is defined as “sleep efficiency” by the actigraph and its associated software (Ambulatory Monitoring). The definition of sleep efficiency varies across studies and software, however, it is frequently calculated using bedtime and not sleep onset (Ancoli-Israel et al., 2015), thus, we opted to refer to the variable examined as %sleep.

At T1, 78.7% of men and 85% of women had valid actigraphy data for at least five nights and their data were included in analyses; at T2, 82% of men and 82.5% of women had such data. At least 5 nights of actigraphy are recommended to achieve sufficient reliability and validity (Meltzer, Montgomery-Downs, Insana, & Walsh, 2012). Reasons for missing sleep data were primarily due to forgetting to wear the actigraph, and medication use for acute illnesses (e.g., flu) and exclusion of those nights from analyses; mechanical problems with the actigraphy happened rarely. Participants with missing actigraphy data were not excluded from analyses and per best practices, missing data were handled statistically using full information maximum likelihood (Enders & Bandalos, 2001).

Subjective sleep problems.

At T1 and T2, men and women reported on their own sleep problems using the 19-item Pittsburgh Sleep Quality Inventory (PSQI; Buysse, Reynolds, Monk, Berman, & Kupfer, 1989). The PSQI is a well-established measure that assesses the frequency of multiple dimensions of sleep including duration, latency, quality, and efficiency. Items are rated on a 4-point scale (0 = not in the past month to 3 = three or more times a week). A global sleep problems score is derived across items and scores > 5 indicate significant sleep problems (Buysse et al., 1989). The measure has established psychometric properties for assessing sleep problems with non-clinical samples (Buysse et al., 1989). At T1, 46% of men and 52.8% of women reported significant sleep problems (i.e., score > 5). At T2, 43.3% of men and 53.6% of women reported significant sleep problems. The PSQI had good internal consistency at T1 (men α = .80, women α = .85) and T2 (men α = .85, women α = .86).

Covariates (T1).

For rigorous assessment of the research questions, some potential covariates known to be associated with sleep were controlled in the models. These included age, race/ethnicity (0 = White/European American, 1 = Black/African American and a very small percentage [7%] who identified as non-African American minorities), medication use for chronic medical conditions (0 = no, 1 = yes), length of cohabitation, and season of sleep assessment (0 = period between fall daylight and spring daylight savings time; 1 =after spring daylight savings time); no participation occurred in the summer.

Plan of Analysis

Actor-partner interdependence models (APIM) were fit to examine longitudinal relations between the SES indices and sleep in couples. APIM (Kenny, Kashy, & Cook, 2006) allows for simultaneous examination of individual and partner effects of SES indices on individual and partner sleep parameters. Five SES indices were examined: income-to-needs ratio, perceived economic well-being, employment status, occupation, and educational attainment. Sleep duration was examined with actigraphy and sleep quality was examined with actigraphy (%sleep and long wake episodes [LWE]) and self-reports. To address the research questions, analyses involved a large number of SES and sleep variables, controlled for autoregressive effects (sleep at T1), covaried sleep across partners, and included assessment of actor and partner effects. Thus, in an effort to reduce potential multi-collinearity among SES variables and obtain more intelligible findings, models were fit separately for each SES indicator in conjunction with each sleep parameter. Family income and household size for the couple were only reported by one individual (primarily women) and path models were fit to examine relations between income-to-needs ratio and sleep Men and women’s SES and sleep variables were included simultaneously in the model.

Models were fit with AMOS 23 (Arbuckle, 2014), which uses full information maximum likelihood (FIML) estimation to handle missing data (Acock, 2005). For men, up to 10% of sleep data were missing, and up to 20%, 14%, and 8% of data were missing for subjective reports on economic well-being, education, employment and occupation, respectively. For women, up to 21% of data were missing across sleep parameters, 8% were missing for perceived economic well-being, 16% were missing employment information, and 30% were missing for educational status. Across couples, 12% were missing income information. Missing data was within permissible guidelines for the use of FIML (Enders & Bandalos, 2001).

SES indices and sleep parameters were inspected for outliers using histograms and skew statistics. Values that exceeded 4 SD were winsorized and recoded to the highest or lowest value (n = 6; Cousineau & Chartier, 2010), which included %sleep for men and women, as well as LWE for women. None of the other primary variables were skewed.

For each model, significantly correlated control variables and autoregressive effects of sleep at T1 were allowed to covary within and across partners. Similarly, indices of SES and residual variances for sleep outcomes were covaried within and across partners for all models. Controlling for the autoregressive effects of sleep allows for making conclusions about the role of SES as a predictor of change in sleep over time, and reduces bias in parameter estimates (Selig & Little, 2012). Autoregressive sleep and residual variances for sleep outcomes were covaried across partners to control for the effects of an individual’s sleep on that of his or her partner. Models were considered an acceptable fit if they satisfied at least two of the three following criteria: χ²/df < 3, CFI > 0.95, and non-significant RMSEA ≤ 0.06 (Hu & Bentler, 1999).

Results

Preliminary Analysis

Means, standard deviations, and correlations among primary study variables are presented in Table 1. Across SES indices, perceived economic well-being, educational attainment, and occupation type were significantly correlated within couples. Couple income-to-needs ratio is moderately correlated with all SES indices reported by men and women. For men and women, greater education was associated with their own greater perceptions of economic well-being and more managerial and professional occupations. Across all sleep parameters, sleep was significantly correlated within couples at T2, and only actigraphy-derived sleep variables were correlated within couples at T1.

Table 1.

Descriptive Statistics and Correlations among Sleep Parameters and Socioeconomic Indices for Cohabiting Men and Women

	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17
1. Income-to-Needs (T1)	___
2. Economic Well-being (T1) - W	.41***	___
3. Economic Well-being (T1) - M	.39***	.75***	___
4. Employment Status (T1) - W	.32***	.15	.17	___
5. Employment Status (T1)- M	.37***	.14	.14	.11	___
6. Occupation Type (T1) - W	−.40***	−.17	−.37**	−.01	−.05	___
7. Occupation Type (T1) - M	−.37***	−.44***	−.39***	.01	.01	.25*	___
8. Education (T1) - W	.49***	.26*	.27*	.18	.20	−.39***	−.27*	___
9. Education (T1) - M	.46***	.31**	.26**	.10	.18	−.19	−.37***	.43***	___
10. Sleep Minutes (T2) - W	.23*	.25**	.21	.03	.22*	−.16	.03	−.13	−.02	___
11. Sleep Minutes (T2)-M	.16	.03	.04	.00	.10	.10	−.14	−.09	.00	.40***	___
12. %Sleep (T2) -W	.31**	.39***	.25*	.02	.18	−.07	−.05	−.03	−.05	.65***	.15	___
13.% Sleep (T2) -M	.10	.15	.14	.06	.04	.00	−.23*	−.10	−.03	.17	.56***	.26**	___
14. Wake Episodes (T2) - W	−.24*	−.37***	−.24*	−.06	−.13	.02	.02	.05	.05	−.51***	−.06	−.94***	−.22*	___
15. Wake Episodes (T2)- M	−.09	−.14	−.14	.00	−.03	−.02	.18	.15	.05	−.15	−.38***	−.29**	−.92***	.26**	___
16. Subj. Sleep Problems (T2) - W	−.03	−.18	−.32***	−.06	−.02	−.02	.05	.05	−.12	−.09	−.10	−.17	−.11	.14	.06	___
17. Subj. Sleep Problems (T2) - M	−.07	−.22*	−.28**	.08	−.16	.10	.06	−.16	.08	−.20*	−.11	−.11	.01	.09	−.01	.25**	___
Mean	1.84	.14	.06	-	-	3.06	3.84	5.30	4.84	400.26	378.93	93.36	91.66	1.72	2.16	6.48	6.23
(SD)	1.02	.83	.83	-	-	1.63	2.09	1.02	1.02	69.48	71.93	6.82	8.08	1.62	1.95	3.27	4.09

Note. W = Women, M = Men; Employment (0 = Unemployed, 1 = Employed); %Sleep = percentage of epochs asleep from sleep onset to wake; Wake Episodes = Long Wake Episodes; Subj. Sleep Problems = Subjective sleep problems.

^*p < .05,

^**p < .01,

^***p < .001

Among covariates and sleep outcomes (not depicted in Tables), men’s age was associated with %sleep (r = −.24, p < .01) and LWE (r = .26, p < .01). Men’s (r = .30, p < .01) and women’s (r = .26, p < .01) use of medication was associated with greater subjective sleep quality problems. No significant associations were detected between sleep and season of assessment, length of cohabitation, race or mother’s age. Covariates were retained in all models due to broader theoretical considerations in the literature. ¹

Paired samples t-tests were conducted to examine differences between men and women on primary study variables (Table 2). In comparison to men, women had higher levels of perceived economic well-being, jobs with greater occupational prestige, as well as longer and greater %sleep. Men had higher levels of employment and greater long wake episodes (LWE) compared to women. Paired samples t-tests indicated that neither men’s nor women’s sleep differed between T1 and T2. Furthermore, correlations across T1 and T2 (not depicted in the Tables) indicated high stability for all of the objective and subjective sleep parameters for men (rs = .60 to .71, p < .001) and women (rs = .62 to .83, p < .001).

Table 2.

Differences in Socioeconomic Indices and Sleep Parameters among Cohabiting Men and Women

	Men		Women		t
	Mean	SD	Mean	SD
Income-to-Needs	1.84	1.02	1.84	1.02	-
Economic Well-Being	.06	.83	.14	.83	2.93**
Employment Status	80.5%	-	61.2%	-	−.3.78***
Occupation Type	3.84	2.09	3.06	2.84	−2.93**
Education	4.84	1.02	5.30	1.02	4.10***
Sleep Minutes	378.93	71.93	400.26	69.48	2.98**
%Sleep	91.66	8.08	93.36	6.82	2.48*
Wake Episodes	2.16	1.95	1.72	1.62	−2.68**
Subjective Sleep Problems	6.23	4.09	6.48	3.27	.18

Note: Employment Status (0 = Unemployed, 1 = Employed); %Sleep = percentage of epochs asleep from sleep onset to wake.

^*p < .05,

^**p < .01,

^***p < .001

Socioeconomic Status and Sleep in Couples

Income-to-Needs.

Longitudinal associations between income-to-needs ratio (income) and men and women’s sleep were assessed and model fit was acceptable for all models assessing these relations. For men, greater income was associated with increases in their sleep minutes (Figure 1; β = 11.78, p =.025; Model fit: χ²/df = 1.5, CFI = 0.92, RMSEA = 0.06, ns) and LWE (Figure 2; β = .37, p = .019; Model fit: χ²/df = 1.6, CFI = 0.91, RMSEA = 0.07, ns) over time. For women, greater income was associated with improvement in sleep quality indicated by fewer wake episodes (Figure 2; β = −.32, p =. 004). The models accounted for a large amount of variance in some sleep parameters. Specifically, R² = 52% for men’s sleep minutes, 54% for men’s LWE, and 59% for women’s LWE. Among men, income-to-needs accounted for 3% of the unique variance in sleep minutes and 1.1% in LWE, whereas for women, income-to-needs explained 5.7% and 7.3% of unique variance in minutes and LWE, respectively. No associations emerged between income and %sleep or subjective sleep problems for men or women.

Figure 1.

Solid bold lines represent significant actor/partner pathways, solid lines represent significant paths, and dashed lines indicate non-significant pathways. Unstandardized and standardized (in parentheses) coefficients are provided for pathways. Correlations are provided for covarying pathways. All models control for age, race, medication use, cohabitation length, season of assessment, and autoregressive effects for sleep (T1).

Note: * p < .05, ** p < .01, *** p < .001.

Figure 2.

Solid bold lines represent significant actor/partner pathways, solid lines represent significant paths, and dashed lines indicate non-significant pathways. Unstandardized and standardized (in parentheses) coefficients are provided for pathways. Correlations are provided for covarying pathways. All models control for age, race, medication use, cohabitation length, season of assessment, and autoregressive sleep effects (T1).

Note: * p < .05, ** p < .01, *** p < .001; LWE = long wake episodes.

For a more comprehensive presentation of the models, we depict associations between men’s and women’s sleep minutes (Figure 1) and long wake episodes (Figure 2). Also, please note the high stability in sleep parameters between T1 and T2 for both men and women (Figures 1 and 2).

Perceived Economic Well-Being.

Models examining longitudinal associations between perceived economic well-being and men and women’s sleep were assessed and model fit was acceptable. For women, greater perceptions of economic well-being were associated with increases in %sleep (Figure 3; β = 1.74, p =. 033; Model fit: χ²/df = 1.4, CFI = 0.94, RMSEA = 0.06, ns) and decreases in LWE (Figure 4; β = −.52, p =. 019; Model fit: χ²/df = 1.4, CFI = 0.94, RMSEA = 0.06, ns) over one year. The amount of variance accounted for by the models was 69% for %sleep and 57% for LWE. Specifically, perceived economic well-being accounted for 14.4% of the unique variance in %sleep and 12.7% in LWE. For men, no associations emerged between perceived economic well-being and sleep parameters. Furthermore, only actor effects emerged and perceived well-being of one person did not impact the sleep of the other member of the couple. Associations between men’s and women’s sleep at T1 and stability in the sleep variables over one year are shown in Figures 3 and 4.

Figure 3.

Solid bold lines represent significant actor/partner pathways, solid lines represent significant paths, and dashed lines indicate non-significant pathways. Unstandardized and standardized (in parentheses) coefficients are provided for pathways. Correlations are provided for covarying pathways. All models control for age, race, medication use, cohabitation length, season of assessment, and autoregressive sleep effects (T1).

Note: * p < .05, ** p < .01, *** p < .001. %Sleep = percentage of epochs asleep from sleep onset to wake.

Figure 4.

Solid bold lines represent significant actor/partner pathways, solid lines represent significant paths, and dashed lines indicate non-significant pathways. Unstandardized and standardized (in parentheses) coefficients are provided for pathways. Correlations are provided for covarying pathways. All models control for age, race, medication use, cohabitation length, season of assessment, and autoregressive sleep effects (T1).

Note: * p < .05, ** p < .01, *** p < .001; LWE = Long wake episodes.

Employment.

Models examining employment status and sleep yielded acceptable fit. No significant actor effects between employment and sleep emerged for men or women, however, partner effects on sleep were detected. Higher employment status of men was associated with greater sleep minutes (Figure 5; β = 32.12, p =. 011; Model fit: χ²/df = 1.5, CFI = 0.91, RMSEA = 0.06, ns) and increased percentage of the night spent asleep (Figure 6; β = 1.97, p =. 049; Model fit: χ²/df = 1.2, CFI = 0.96, RMSEA = 0.04, ns) for their partners. The amount of variance accounted for by the models was 52% for sleep minutes and 68% for %sleep. Employment status accounted for 5.1% of unique variance in women’s sleep minutes and 3.2% in women’s %sleep. Women’s employment status was associated with greater reported sleep problems for men (Figure 7; β = 1.60, p = 002; Model fit: χ²/df = 1.6, CFI = 0.90, RMSEA = 0.07, ns). This model accounted for 42% of the variance of men’s sleep problems and employment accounted for 3.3% of unique variance in this variable. Furthermore, relations between men’s and women’s sleep at T1 and autoregressive effects in the sleep measures are depicted in Figures 5 and 6.

Figure 5.

Solid bold lines represent significant actor/partner pathways, solid lines represent significant paths, and dashed lines indicate non-significant pathways. Unstandardized and standardized (in parentheses) coefficients are provided for pathways. Correlations are provided for covarying pathways. All models control for age, race, medication use, cohabitation length, season of assessment, and autoregressive sleep (T1).

Note: * p < .05, ** p < .01, *** p < .001.

Figure 6.

Solid bold lines represent significant actor/partner pathways, solid lines represent significant paths, and dashed lines indicate non-significant pathways. Unstandardized and standardized (in parentheses) coefficients are provided for pathways. Correlations are provided for covarying pathways. All models control for age, race, medication use, cohabitation length, season of assessment, and autoregressive sleep (T1).

Note: * p < .05, ** p < .01, *** p < .001. %Sleep = percentage of epochs asleep from sleep onset to wake.

Figure 7.

Solid bold lines represent significant actor/partner pathways, solid lines represent significant paths, and dashed lines indicate non-significant pathways. Unstandardized and standardized (in parentheses) coefficients are provided for pathways. Correlations are provided for covarying pathways. All models control for age, race, medication use, cohabitation length, season of assessment, and autoregressive sleep.

Note: * p < .05, ** p < .01, *** p < .001; Subj. Sleep Problems = Subjective sleep problems.

Occupation and Education.

None of the models were significant for occupation type or education. Please note that sleep parameters in models examining both education (βs =.53 – .84, p <.001) and occupation type (βs =.52 – .84, p <.001) were highly stable between T1 and T2 for men and women.

Discussion

The current study addressed relations between economic adversity and sleep by examining multiple indicators of SES and several objective and subjective sleep parameters in a community sample of cohabiting adults. Building on the literature that has primarily utilized cross-sectional designs, self-reported sleep, and examined sleep of individuals independent of their family context, the research questions were addressed with a two-wave longitudinal design across one year using an APIM framework that assessed couples’ sleep to simultaneously capture actor (the individual) and partner effects. Sleep duration was examined with actigraphy and multiple sleep quality parameters were derived using both actigraphy and self-reports.

In general, lower SES was associated with the worsening of men’s and women’s sleep over time. However, associations did not emerge across all SES indices or sleep parameters and there were some differential associations between men and women. Higher income-to-needs ratio was associated with increases in sleep duration, yet a decline in sleep quality (LWE) for men. For women, higher income-to-needs ratio was associated only with improvement in sleep quality (LWE). Similarly, for women, a higher level of perceived economic well-being was related to improvement in sleep quality over time indicated by both %sleep and LWE.

Previous investigations, mostly based on cross-sectional data, have found that low income is associated with poor sleep quality (Johnson et al., 2016) and short self-reported duration (Whinnery et al., 2014). Fewer studies have demonstrated that these indices are associated with objectively measured short sleep duration and poor sleep quality (El-Sheikh et al., 2015; Mezick, et al., 2008). Our findings contribute to this body of work by demonstrating longitudinal relations and utilizing actigraphic measures of sleep duration and quality. Because the autoregressive effects of sleep were controlled, the results add to the literature by demonstrating that some SES variables, namely income-to-needs, perceptions of economic adversity, and employment status predicted change in sleep over one year.

Furthermore, the few cross-sectional studies investigating links between perceived economic hardship and sleep corroborate our findings that greater hardship is associated with poor subjective sleep quality (El-Sheikh et al., 2015) and poor objectively derived sleep efficiency (Hall et al., 2009); note that findings from the present study were observed mainly for actigraphy-derived and not subjective sleep quality. Results build on this literature by demonstrating that economic hardship predicts worsening of sleep quality over one year. These findings underscore the importance of examining subjective economic well-being when assessing associations between SES and sleep.

Few studies have examined associations between couples’ employment and sleep. Those studies primarily focused on work-life responsibilities and their role in sleep disruptions (Venn, Arber, Meadows, & Hislop, 2008), as well as utilized samples where men were the primary earners (Maume & Ruppanner, 2017). Results of this study indicate that women’s employment status (i.e., returning to work) in single earner couples disrupt men’s sleep; however, in couples where both partners were employed, women reported greater sleep minutes compared to their spouses (Doumas, Margolin, & John, 2008).

Consistent with the literature on sleep in cohabiting relationships (Gunn et al., 2015; Meadows, Venn, Hislop, Stanley, & Arber, 2009), men’s and women’s sleep was associated both at T1 and T2 across the vast majority of the sleep parameters. However, few partner effects between SES and sleep were found. Men’s employment was associated with increased sleep duration and improvement in the percent of nightly sleep for women whereas women’s employment was associated with greater self-reported sleep problems for their partners. Though our study did not find individual (actor) employment effects, it may be that simply having an employed partner may reduce stress in this economically disadvantaged population, particularly for women who may be juggling multiple family and work responsibilities (Arber, Bote, & Meadows, 2009). In households with children, women often engage in “second-shift” child rearing and household management (Blair-Loy, Hochschild, Pugh, Williams, & Hartmann, 2015). Women’s employment outside of the home may therefore contribute to conflicts relating to their expected roles as mothers and wives, which may in turn lead to increases in men’s subjective sleep problems. Of course, this explanation is highly tentative and needs empirical assessment before conclusions can be drawn.

Additionally, it is possible that these gender differences may be specific to the geographical region of residency (semi-rural and relatively small towns in Alabama), the jobs that men and women held and their perceived prestige, as well as the broader socio-cultural milieu. For example, one study found that adults who have spent most of their time located in the Southeastern United States tend to endorse more traditional gender role attitudes compared to their non-Southern counterparts (Carter, Carter, & Corra, 2016), which in turn may contribute to such gender-related associations between employment and sleep. These interpretations are speculative and the results should be considered preliminary pending replication.

It is not evident why educational attainment was uncorrelated with sleep and did not have a significant effect on changes in adults’ sleep over time. Such associations have been reported where lower levels of educational attainment are associated with poor sleep (Grandner et al., 2013, 2016). Null effects may be due to the restricted variability in educational attainment within our sample. The majority of participants (76% of women and 56% of men) had partial college attendance or college degrees whereas 24% of women and 44% of men had a high school degree or less. Also, education and both income and perceived well-being were moderately correlated (.26 - .49) suggesting that education did not necessarily translate to proximal material resources or subjective economic well-being in the sample. Elucidating factors associated with education that may be influential across various contexts is warranted.

Additionally, it is unclear why employment status and occupation type were not associated with the individual’s own sleep. Some studies found relations where employment status and job security were associated with fewer sleep complaints (Grandner et al., 2010; Mai, Hill, Villa-Henninger, & Grandner, 2018). Others have posited that less control in one’s job (such as in elementary occupations) may be associated with greater stress that influences sleep quality (Burgard & Ailshire, 2009). Education and occupation are distal indicators of SES and representative of social prestige (Braveman et al., 2005) and may not influence sleep in couples in the same manner as more proximal variables such as lack of monetary resources or perceived financial stress. Also, employment and occupation type were moderately correlated with income and perceived economic well-being (−.36 - .32) for men and women. Status variables such as education and occupation have been found to predict sleep outcomes even after adjusting for financial and economic resources (Grandner et al., 2013; Jackson et al., 2013) and the lack of findings in this sample may suggest a need to better understand the role of SES indices within this particular cultural milieu.

Some longitudinal associations among SES indices and sleep parameters differed across men and women. Men’s sleep duration and quality were associated with the family’s income-to-needs ratio, an indicator of monetary resources. For women, their sleep was influenced by both, income-to-needs and more prominently, their perceptions of economic difficulties. Although no firm conclusions are justified due to the small number of gender-related effects, it is plausible that women may express greater concerns about socioeconomic security and financial resources (Malone, Stewart, Wilson, & Korsching, 2010) and report more negative assessments of their financial situation (Loibl & Hira, 2007) than men. At the same time, women in our sample reported higher perceived economic well-being than men raising questions about the comparability of our findings to those reported by others. It is notable that such studies do not generally consider how experiences of economic security may be different for men and women in cohabiting couples. Furthermore, beliefs about gender roles and division of labor (e.g., shopping for groceries, paying household bills) may contribute to differing perceptions of economic security and how different dimensions of SES influence sleep (Maume, Sebastian, & Bardo, 2010). It may be that women are the primary individuals in the household purchasing goods and services, which may make them more attuned to whether their current financial situation is consistent with the needs of the family.

Findings from this study need to be interpreted within the strengths and limitations of the sample and methodology. The use of a community sample of cohabiting heterosexual adults limits generalizability to clinical populations experiencing sleep problems and same-sex cohabiting couples. Although the sample is socioeconomically heterogeneous, few families were of higher SES based on income-to-needs ratio (ratio ≥ 4; upper middle class or higher), thus limiting generalizability to more affluent families. Additionally, income was only reported by the women in our sample and was ascertained using income brackets rather than directly asking about gross yearly income.

Couples were middle-aged adults with children and findings may not translate to other age groups or those without children. The presence of young and school-aged children in the household may serve as an additional stressor on socioeconomic resources (Umberson, Pudrovska, & Reczek, 2010) and the sleep of parents (Korous & El-Sheikh, 2017; Meltzer & Mindell, 2007; Meltzer & Montgomery-Downs, 2011), thus making it important to consider in future work. The present study did not have data on ages of all children residing in the household and could not control for potential confounding effects on men’s and women’s sleep.

Other studies have found differences among individuals who are employed, unemployed, retired, non-workers, and students and sleep (Grander et al., 2013). However, the present study was unable to delve into such associations due to limited variability in employment status. In addition, in the context of our analyses, SES was examined at T1 as a predictor of change in sleep between T1 and T2 and change in SES over time was not assessed. Furthermore, we did not directly compare the effects of the various SES variables to one another and thus results need to be interpreted in this context.

A critical avenue for research is the assessment of mechanisms of effects linking SES with sleep. Socioeconomic adversity is associated with many risk factors that could affect sleep including a suboptimal sleep environment. Poor sleep environments such as those with disruptive lights and sounds, uncomfortable sleeping surfaces, and lack of temperature control, serve as intervening variables in associations between disadvantaged and stressful environments (e.g., lower income) and poor sleep for both children and adults (Bagley, Kelly, Buckhalt & El-Sheikh, 2015; Chung, Wilson, Miller, Johnson, Lumeng, & Chervin, 2014; Spilsbury, Frame, Magtanong, & Rork, 2016).

Despite these limitations, study results have important implications. Sleep is predominantly treated as a solitary activity in clinical health settings, although a majority of adults share a bed with a partner. Therefore, it is important for clinicians to consider the sleep of both partners when treating sleep problems. Taken together, the findings indicate that income and perceptions of economic well-being were the most consequential predictors of sleep among couples with children in the Southeastern United States. Furthermore, results suggest that individual SES indicators should be examined carefully in future research as it appears that a “one size fits all” approach does not adequately capture SES effects on sleep.

Table 3.

Summary of Significant Effects

	Sleep Minutes	%Sleep	LongWake Episodes	Subjective Sleep Problems
Income-to-Needs	X
Economic Well-being		X	X
Education
Employment Status	X	X		X
Occupation Type