Racial Disparities in Sleep: Potential Mediation by Discrimination and Psychological Distress

Abstract

Previous research has demonstrated that experiences of discrimination contribute to racial disparities in sleep, and that psychological distress mediates these relationships. However, previous research has not included race as part of the mediation pathway and has had limited dimensions of sleep health and psychological mediators. In the current study, we examine serial mediation pathways by which race and sleep health are mediated through discrimination and subsequently through psychological distress (i.e., depressive symptoms, chronic stress, and loneliness). Data were from the 2010 wave of the Health Retirement Study (HRS). The analytic sample (n= 7,749) included Black and White participants who were included in the Enhanced face-to-face interview in 2010 and who completed the psychosocial questionnaire. Race was reported as either Black or White. Sleep health was assessed with a 4-item questionnaire. Depressive symptoms were assessed with the shortened CES-D, chronic stress via the ongoing chronic stressor scale, and loneliness via the UCLA loneliness scale. Covariates were included in all serial mediation models. Relative to White participants, Black participants reported increased experiences of discrimination, which was associated with increased psychological distress, and poorer sleep health. Findings demonstrate the significant adverse impact that discrimination has on both psychological well-being and sleep health.

Introduction

Research has consistently demonstrated that racial and ethnic minorities are more likely to have poorer sleep health [1], which may also confer greater risk for chronic health conditions, such as cardiovascular disease [2]. More recent work has sought to identify the determinants of racial disparities in sleep, including one’s physical and social environment [3]. In addition, experiences of discrimination contribute to these disparities [4]. Individuals hold multiple identities (e.g., race, ethnicity, gender, etc.), all of which are tied to and defined by larger structures of privilege and oppression. The simultaneous existence of these larger structures affect individuals’ lived experiences [5]. As demonstrated in Meyer’s minority stress model, individuals with minority or marginalized identities have greater exposure to repeated experiences of discrimination, which take a toll on physical and psychological health [6-8]. This model has been applied to understanding disparities in sleep [1], such that increased experiences of identity-based discrimination are associated with poor sleep health [4,9]. While the impact of discrimination on sleep has been consistently demonstrated [4], the mediating pathways linking discrimination and sleep remain unclear.

A few studies have examined whether psychological distress mediates associations between discrimination and sleep. A study of Latinx adults tested whether a psychological distress (a combination of depression, stress, mental health) mediated associations between racial and ethnic discrimination and self-reported sleep disturbances, and found that experiences of discrimination were significantly related to increased psychological distress and poorer sleep. [10]. Similarly, Majeno and colleagues [11] tested whether stress and loneliness mediated associations between and discrimination based on race/ethnicity, or discrimination based on other factors, and sleep. In this sample of Latinx, European American, and Asian adolescents, loneliness and stress mediated associations between discrimination and sleep. Moreover, it was demonstrated that multiple facets of sleep may be differentially associated with experiences of discrimination. Using Health Retirement Study (HRS) data, Vaghela & Sutin [12] found that depression and anxiety mediated associations between self-reported sleep quality and non-restfulness. Taken together, these studies suggest that psychological distress mediates associations between discrimination and sleep. However, there are several limitations to the prior work. First, current research focuses on general sleep quality, using aggregated scores from self-report measures. However, as evidenced by findings from Majeno and colleagues [11], associations among discrimination, psychological distress, and sleep may vary among different facets of sleep. Sleep health is multidimensional [13], and various dimensions should be considered. Second, research to date has not tested whether discrimination and psychological distress mediate associations between race and sleep. The aforementioned studies either used race or ethnicity as a control variable or do not include race as a part of the mediation pathway. Incorporating race as part of this mediation pathway is critical because discrimination is not a uniformly experienced stressor. Racial and ethnic minorities are disproportionally targeted and often face more frequent experiences of discrimination across multiple sectors in society [14]. Veghela & Sutin [12] did stratify exploratory analyses by race, but this does not allow for a complete examination of explanatory factors contributing to racial disparities in sleep.

The current study will address these limitations by examining multiple facets of sleep and proposing a serial mediation model in which race is used as a factor in the model. We will also add to the literature by independently testing depression, loneliness, and chronic stress as types of psychological distress that may impact sleep in a diverse sample of adults. It was hypothesized that Black participants, relative to White participants, will report increased experiences of discrimination, which will be related to increased depression, stress, and loneliness, which will in turn be associated with poorer sleep health. Hypotheses and analytic plan were pre-registered at osf.io/3beh4.

Methods

The data for the current study were from the HRS. The HRS is a nationally representative longitudinal aging study in approximately 20,000 adults, and their spouses, over the age of 50 in the United States (US), focused on understanding how various psychosocial and biological factors inform well-being. The HRS study was granted approval from the Institutional Review Board (IRB) at the University of Michigan. The HRS study utilizes multi-stage probability sampling, and oversamples African American and Hispanic households to help obtain a nationally representative sample. See Sonnega and colleagues [15] for a more detailed description of the HRS study. The data for the current study are from several HRS datasets from the year 2010, including the 2018 Cross-Wave Tracker (this data set tracks data cross waves, including 2010), Health Status (C), Cognition (D), Leave Behind (LB), and the RAND HRS Longitudinal File (2016, V2). The data in year 2010 were chosen for the current study because all necessary variables were simultaneously collected at this time.

Analytic Sample

The initial sample included 43,403 participants. Only participants who were selected to be included in the Enhanced face-to-face interview (EFTF) for the year 2010 were retained in the analytic sample (n= 17,621). In addition, participants were included in our sample if they were alive, or presumed to be alive, available during the 2010 wave, completed data as identified by the HRS study team, and had identifiable racial identity data. In each EFTF wave, a random 50% of the core study participants were provided with additional questions, including the LB questionnaire, which contained key variables of interest including discrimination, loneliness, and chronic stress [16]. A total of 25,782 participants were excluded because they were not included in the 2010 EFTF, were removed from the sample prior to the initiation of EFTF, or did not have data to indicate their EFTF status. Participants with a status of alive or presumed alive were retained (n=12,399), while all others were removed (n= 5,222). One individual, present in the 2010 sample, decided to be formally dropped, lowering the analytic sample by 1 (n= 12,398). Individuals with a starting interview of the year 2010 or earlier were retained (n=12,163), and individuals with a starting interview year after 2010 were removed (n= 235). Individuals with a 2010 core interview were retained (n= 11,214), while others were removed (n=949). Individuals who indicated completing the LB questionnaire were retained (n=8,309), while those who indicated they did not complete the LB questionnaire or had no data for this were removed (n=2,905). Finally, individuals who indicated their race as Black/African American or White/Caucasian were retained (n=7,749), while those with no data or who indicated “other” were removed (n= 560). Therefore, the final analytic sample for the current study is 7,749 participants.

Measures

Discrimination.

Everyday discrimination scale [17] assessed the extent to which individuals experienced a range of discriminatory behaviors in their daily life. For instance, one sample item includes, “People act as if they are afraid of you”. Participants responded to all 6 items on a 1 (Almost every day) to 6 (Never) scale. The scale assessed the frequency of specific experiences, and not attributions about why the experiences occurred (e.g., on the basis of race/ethnicity, gender, etc.) All items were computed into a single mean score, in which higher values indicate greater experiences of everyday discrimination. The scale was considered reliable (alpha = .81).

Loneliness.

Perceptions of loneliness were assessed on an 11-item measure [18]. Participants indicated on a 1 (Often) to 3 (Hardly ever or never) scale the frequency with which they felt lonely or had reliable social support networks (e.g., “You lack companionship?”). The scale was considered reliable (alpha = .89), and all items were averaged into a mean score. The data were scored such that a higher score indicates greater perceptions of loneliness.

Ongoing Chronic Stressors.

The ongoing chronic stressor scale [19] was used to determine the subjective burden of chronic stress across eight areas in individuals lives. This measure included 8 items, and participants were instructed to indicate whether or not any of the eight areas were current/ongoing problems that have lasted at least 12 months on a scale of 1 (No, didn’t happen) to 4 (Yes, very upsetting). One sample item includes, “Ongoing physical or emotional problems (in a spouse or child)”. All 8 items were computed into a sum score, in which higher values indicate greater chronic stress burden (alpha= .68).

Depressive Symptoms.

Depressive symptoms were measured with a shortened version of the Center for Epidemiologic Studies Depression Scale (CES-D) [20,21]. Participants responded either “Yes” (1) or “No” (0) to all items. A sum score was computed across all items, and a higher total value indicates greater depressive symptoms. Responses outside “Yes” and “No” (e.g., don’t know) were coded as missing and were not included in the sum score total (alpha= .80).

Self-Reported Sleep Health.

Individual's perceptions of self-reported sleep was assessed individually over 4 items: Falling asleep (how often do you have trouble falling asleep?), wake during the night (how often do you have trouble with waking up during the night?), wake early (how often do you have trouble with waking up too early and not being able to fall sleep again?), and rested (how often do you really feel rested when you wake up in the morning?). Participants responded to each of these facets of sleep on a 1 (Most of the time) to 3 (rarely/never) scale. All items were coded such that a higher score indicates healthier sleep. These 4 items were all summed into a composite score of sleep health ranging from 4-12. Each sleep item was also tested separately to examine differential associations with experiences of discrimination and psychological distress.

Covariates

For all mediation models, we controlled for the same set of covariates. All covariates, with the exception of net worth data and health problems, were from the HRS 2018 tracker data file. This file is considered to be the most up-to date and the most definitive source for demographic data (HRS, 2016). These covariates included 2010 status for the following: age, gender (male/female), marital status (married/separated/widowed/never married; re-coded as married/not married), education (GED/high school diploma/two-year college degree/4-year college degree/master degree/professional degree), and self-reported health problems (Would you say your health is excellent, very good, fair, or poor?) (dataset C). Further, participant’s household net worth data was computed by the HRS study team as the sum of all wealth components variables less all debts variables. The individual variables included in the computed household net worth variable are based on responses in the participant interviews. Missing data were imputed by the HRS study team. Household net worth was from the RAND HRS longitudinal file. Net worth was included as the primary socioeconomic (SES) variable because research has demonstrated that for samples of older adults, net worth is a reliable indicator of SES [22].

Data Analysis

The PROCESS Macro requires complete data and is developed to implement listwise deletion in all statistical analyses. Scales for participants were considered missing if over 50% of the items were missing and is based on HRS documentation. For scales with no explicit instructions on how to handle missing data, the same criteria were used. Participants excluded from analyses were more likely to be Black, highest degree earned GED or high-school diploma, less likely to be married, more self-reported health problems, lower net worth, and older (all p’s <.001) Discrimination (skew= 1.89, kurtosis= 4.727) and net wealth (skew= 9.96, kurtosis= 177.97) were not normally distributed and were log10 and square root transformed, respectively. Transformed variables were used in all analyses.

Mediation analyses

The PROCESS Macro by Hayes for SPSS (V.27) was used to conduct multiple mediation models [23]. The Macro is based on ordinary least squares (OLS) regression. We examined the test of the indirect effect for indirect effect 1 (race and sleep mediated by discrimination), indirect effect 2 (race and sleep mediated by psychological distress), and the overall hypothesized serial mediation model (race and sleep mediated by discrimination and psychological distress), all using bias-corrected bootstrapped confidence intervals (5,000) (CI). If the confidence interval for a given indirect effect does not contain the value of zero, the mediation analysis is statistically significant. In the results, we report the indirect effects, the unstandardized coefficients for the individual pathways, and the direct effect [23,24].

Results

Descriptives

Participants (see Table 1) were on average 67 (SD= 11.4) years old, and majority White (82.9%) and female (58.4%). On average, participants indicated an experience of discrimination score of 1.5 (SD= .7), and an aggregated sleep score of 9.3 (SD= 2.1). Unadjusted t-tests demonstrated that Black participants were significantly more likely to report experiences of discrimination, greater psychological distress, and poorer sleep health (all p’s < .001).

Table 1.

Demographic and study variable characteristics

		Black	White	p value
n	7,749	1324 (17.1%)	6425 (82.9%)
Gender				< .001
Male, N(%)	3222 (41.6%)	448	2774
Female, N(%)	4526 (58.4%)	876	3651
Age, mean(SD)	67.06 (11.4)	63.37 (10.7)	67.81 (11.3)	.001
Marital status, N(%)				< .001
Not currently married	2957 (38.2%)	786 (59.4%)	2171 (33.8%)
Married	4787 (61.8%)	537 (40.6%)	4250 (66.1%)
Education, N(%)				<.001
GED	403 (5.2%)	84 (6.3%)	319 (5%)
High school diploma	3855 (49.7%)	621 (46.9%)	3234 (50.3%)
Two-year college degree	426 (5.5%)	82 9 (6.2%)	344 (5.4%)
Four-year college degree	1097 (14.2%)	103 (7.8%)	994 (15.5%)
Master’s degree	611 (7.9%)	61 (4.6%)	550 (8.5%)
Professional degree	151 (1.9%)	10 (0.8%)	141 (2.2%)
Net worth, mean(SD)	461,405.4 (1,004,328.7)	90)121,429 (281,4	531,466 (1,082,366)
Health problems, mean(SD)	2.78 (1.1)	3.14 (1.04)	2.71 (1.07)	.006
Experiences of discrimination, mean(SD)	1.55 (.7)	1.72 (.8)	1.51 (.7)	<.001
Depression, mean(SD)	1.81 (2.2)	2.28 (2.3)	1.71 (2.1)	<.001
Loneliness, mean(SD)	1.5 (.4)	1.59 (.4)	1.51 (.4)	<.001
Chronic stress, mean(SD)	12.2 (3.8)	13.06 (4.4)	11.97 (3.6)	<.001
Sleep Health Score, mean(SD)	9.3 (2.1)	9.2 (2.1)	9.4 (2.1)	<.001
Waking at night, mean(SD)	2.1 (.8)	2.3 (.8)	2.1 (.8)	<.001
Waking early, mean(SD)	2.4 (.7)	2.3 (.7)	2.4 (.7)	<.001
Falling asleep, mean(SD)	2.4 (.7)	2.3 (.8)	2.4 (.7)	<.001
Feeling rested, mean(SD)	2.4 (.7)	2.3 (.7)	2.5 (.7)	<.001

Depressive Symptoms

There was a statistically significant indirect effect (1) of race on sleep through experiences of discrimination (b= −.01; CI: −.03, −.004). The indirect effect (2) of race on sleep through depressive symptoms was not statistically significant (b= .04; CI: −.01, .01). There was a significant indirect effect of race on sleep (serial mediation) through experiences of discrimination and depressive symptoms (b= −.02; CI: −.03, −.005). Within the serial mediation indirect path, Black participants reported significantly higher experiences of discrimination (b= .02, p< .01), which was associated with increased depressive symptoms (b= 2.17, p< .001), which in turn was associated with poorer sleep (b= −.36, p< .001). After adjusting for the full model, the direct effect of race on sleep was still significant, in that Black participants reported significantly better sleep (b= .25, p< .001). Results are displayed in Figure 1.

Figure 1.

The pathway coefficients of the full serial mediation model including depressive symptoms. *p< .05, **p< .01, ***p< .001

Results for each sleep outcome were mostly consistent. The indirect effects of race through discrimination was significant for falling asleep (b= −.004; CI: −.008, −.001), waking early (b= −.005; CI: −.01, −.002), and feeling rested (b= −.005; CI: −.009, −.001). However, the indirect of race on waking at night through discrimination was not significant (b= .0001; CI: −.002, .003). The indirect effects of race on sleep through depressive symptoms were not significant for falling asleep (b= .01; CI: −.003, .03), waking at night (b= .008; CI: −.002, .02), waking early (b= .01; CI: −.003, .02), or feeling rested (b= .01; CI: −.004, .03). The indirect effects of race through discrimination and depressive symptoms were significant for falling asleep (b= −.004; CI: −.007, −.001), waking at night (b= −.003; CI: −.005, −.001), waking early (b= −.003; CI: −.006, −.001), and feeling rested (b= −.005; CI: −.008, −.002).

Loneliness

There was a statistically significant indirect effect of race on sleep through experiences of discrimination (b= −.02; CI: −.03, −.005). The indirect effect of race on sleep through loneliness was statistically significant (b= .03; CI: .009, .05), such that Black participants reported significantly lower levels of loneliness (b= −.04, p< .05) and healthier sleep. There was a significant indirect effect of race on sleep through experiences of discrimination and loneliness (b= −.01; CI: −.02, −.003). Within the serial mediation model, Black participants reported significantly more experiences of discrimination (b= .02, p< .01), which was associated with increased loneliness (b= .84, p< .001), which was related to poorer sleep (b= −.64, p< .001). After adjusting for the full model, the direct effect of race on sleep was still significant (b= .25, p< .001). Results are displayed in Figure 2.

Figure 2.

The pathway coefficients of the full serial mediation model including loneliness. *p< .05, **p< .01, ***p< .001

Results for each sleep outcome were mostly consistent. The indirect effects of race through discrimination were significant for falling asleep (b= −.005; CI: −.01, −.001), waking early (b= −.006; CI: −.011, −.002) and feeling rested (b= −.006; CI: −.011, −.002). However, the indirect effect of race on waking at night through discrimination was not significant (b= −.001; CI: −.004, .001). The indirect effects of race through loneliness were significant for falling asleep (b= .007; CI: .002, .01), waking at night (b= .004; CI: .001, .009), waking early (b= .007; CI: .002, .012), and feeling rested (b= .009; CI: .002, .02). The indirect effects of race through experiences of discrimination and loneliness were significant for falling asleep (b= −.002; CI: −.004, −.001), waking at night (b= −.002, CI: −.003, −.004), waking early (b= −.003; CI: −.005, −.001), and feeling rested (b= −.003; CI: −.006, −.001).

Chronic stress

There was a statistically significant indirect of race on sleep through experiences of discrimination (b= −0.02; CI: −.03, −.005). The indirect effect of race on sleep through chronic stress was not statistically significant (b= −.002; CI: −.02, .02). There was a significant indirect effect of race on sleep through experiences of discrimination and chronic stress (b= −.009; CI: −.02, −.003). Black participants experienced significantly more experiences of discrimination (b= .02, p< .001), which was associated with increased chronic stress (b= 6.08, p< .001), which was related to poorer sleep (b= −.08, p< .001). After adjusting for the full model, the direct effect of race on sleep was still significant (b= .27, p< .001). Results are displayed in Figure 3.

Figure 3.

The pathway coefficients of the full serial mediation model including chronic stress. *p< .05, **p< .01, ***p< .001

Results for each sleep outcome were mostly consistent. The indirect effects of race through discrimination were significant for falling asleep (b= −.005; CI: −.009, −.001), waking early (b= −.006; CI: −.01, −.002), and feeling rested (b= −.006; CI: −.01, −.002). However, the indirect effect of race on waking through the night was not significant (b= −.001; CI: −.003, .002). The indirect effects of race through chronic stress were not significant for falling asleep (b= −.001; CI: −.007, .006), waking at night (b= −.003; CI: −.006, .005), waking early (b= −.003; CI: −.005, .004), or feeling rested (b= −.004; CI: −.004, .006). The indirect effects of race through experiences of discrimination and chronic stress were significant for falling asleep (b= −.002; CI: −.004, −.001), waking at night (b= −.002; CI: −.004, −.0006), waking early (b= −.002; CI: −.003, −.0005), and feeling rested (b= −.003; CI: −.005, −.007).

Discussion

Research has consistently demonstrated racial disparities in sleep, and experiences of discrimination contribute to these disparities. The current study examined whether experiences of discrimination and psychological distress contribute to racial disparities in sleep. As expected, racial disparities in sleep were mediated through both experiences of discrimination and psychological distress. Specifically, Black participants reported more experiences of discrimination, which was associated with greater psychological distress, which was related to poorer sleep health.

These findings are consistent with previous studies demonstrating that psychological distress mediates associations between discrimination and sleep [10-12]. In addition, the current study adds to previous work by adding race to the mediation model, demonstrating that racial disparities in sleep can be explained by disproportionally greater experiences of discrimination, and consequently, greater psychological distress. Comparisons of adjusted vs. unadjusted means in self-reported sleep offer further insight into the role of discrimination in sleep disparities, in that after controlling for discrimination and psychological distress, Black participants had better sleep health. This emphasizes how disruptive discrimination is to psychological well-being and sleep, and Black individuals are more likely to experience discrimination. This is consistent with previous research demonstrating that racial discrimination mediates racial disparities in sleep [9,25]. Moreover, discrimination may be uniquely detrimental to sleep, even after accounting for psychological distress. For example, previous work has found that greater experiences of discrimination are related to poorer sleep, even after adjusting for perceived stress [26,27].

Findings were mostly consistent across depression, loneliness, and chronic stress, with one exception. In models not including discrimination in the pathway, depression and chronic stress were not significant mediators of race and sleep. However, the full mediation pathway was significant, suggesting that experiences of discrimination are a key factor explaining racial disparities in sleep. However, this indirect effect was significant for loneliness, such that Black participants reported lower loneliness, which was related to better sleep. When accounting for experiences of discrimination in the full model, however, Black participants had significantly higher loneliness and poorer sleep. This suggests that experiences of discrimination are disruptive to both psychological well-being, particularly loneliness, and sleep. It was also tested whether all mediation pathways would be consistent across multiple dimensions of sleep. In general, findings were consistent across sleep outcomes, with just one exception. Indirect effects of race on waking at night through discrimination were not significant, suggesting that the associations between discrimination and self-reported sleep continuity may be less robust. Given the multidimensionality of sleep health [13], future research could continue to examine how discrimination impacts sleep duration, continuity, and timing, and whether psychological distress mediates these associations.

These results should be interpreted in the context of its limitations. First, analyses were cross-sectional and temporal precedence cannot be determined among our study variables. While this is a limitation, cross-sectional mediation analyses are commonly used, including previously cited work in this area of research [10,11], and can provide a framework for future longitudinal research. Future analyses could utilize prospective or microlongitudinal data to establish temporal precedence. For example, Davenport and colleagues [28] conducted a daily diary study over a 4-week period, and found that sleep quality was worse during weeks when individuals reported more than usual experienced racial microaggressions. Ecological momentary assessment methodologies [29] may be particularly useful in determining temporality by measuring psychological distress and sleep directly following experiences of discrimination. Second, sleep health was self-reported, which does not always reflect objective estimates of sleep [30]. This may be particularly important in studies of racial disparities in sleep, as Black participants may be less likely to self-report sleep disturbances [31]. Future research could use actigraphy [32] to obtain objective estimates of sleep, and longitudinal tracking could also identify how experiences of discrimination impact sleep patterns over time. Lastly, our study focused solely on Black and White participants, and future research could examine whether experiences of discrimination and psychological distress mediate sleep disparities among other racial and ethnic minorities. Moreover, our measure of discrimination was not a measure of racial discrimination. Relatedly, Vaghela and Sutton [12] demonstrated that different types of discrimination may hold different associations with sleep, and this should also be delineated in future research. Lastly, the analytic sample consisted of mostly White (83%) participants, which reflected the parent study sample, and future studies could recruit a more balanced sample of Black and White participants.

The current findings add to our current understanding of factors contributing to racial disparities in sleep. Depression, loneliness, and chronic stress were all significant mediators of experiences of discrimination and sleep, and Black participants experienced significantly more discrimination. In addition, findings demonstrate the significant impact that discrimination has on both psychological well-being and sleep.