Abstract
In this meta-analysis, we aimed to evaluate the cross-sectional and longitudinal relationships between various forms of sleep problems and loneliness. A total of 84 articles (110 samples, N = 227,112) were identified for inclusion. Random effects models revealed a significant medium association between overall sleep problems and loneliness (r = .336, 95% confidence interval = [.315, .357]) as well as specific sleep complaints (i.e., insomnia, nightmares, poor sleep efficiency, and poor sleep quality) and loneliness (rs = .165–.354). The longitudinal relationships between overall sleep problems and subsequent loneliness, and vice versa, were also significant (rs = .249–.297). Although no consistent moderation patterns emerged, several significant moderators were identified for specific associations. Results support a robust association between more severe sleep problems and greater perceptions of loneliness; both also appear reciprocally associated longitudinally. Findings point to research directions that may enhance understanding of the interplay between sleep problems and loneliness—constructs with transdiagnostic relevance.
Sleep problems are relatively common across the life span. An estimated 25% of children experience some form of sleep problems in childhood (e.g., from short-term difficulties falling asleep to a diagnosis of a primary sleep disorder), and similar prevalence rates have been observed across countries (for review, see Owens, 2007). Likewise, representative data across four countries suggest that 25% of adolescents experience insomnia symptoms (Ohayon, Roberts, Zulley, Smirne, & Priest, 2000). Among adults, representative data from seven countries found that 35.2% of adults reported at least one sleep complaint (e.g., dissatisfaction with sleep duration, short sleep duration; Ohayon & Reynolds, 2009). In addition, a review of 50 epidemiological studies conducted in various countries worldwide revealed an estimated 33% prevalence rate of insomnia symptoms in the general population (Ohayon, 2002). Together, these data indicate that a significant proportion of individuals worldwide suffer from some form of sleep difficulties. Such data are concerning given that poor sleep has been linked with a range of physical and mental health problems (for review, see Ford & Kamerow, 1989; Irwin, 2015; Owens, 2007; Pigeon, Pinquart, & Conner, 2012; Sateia, Doghramji, Hauri, & Morin, 2000). Thus, efforts are needed both to identify factors that increase risk for the development of poor sleep and to better understand adverse outcomes to which sleep problems may contribute.
One factor that has been associated with increased risk for sleep problems is loneliness—also referred to as perceived social isolation. Loneliness, as defined by Peplau and Perlman (1982), represents a mismatch between an individual’s desired as opposed to actual social relations. This mismatch is then hypothesized to result in the painful perception and/or negative experience that one does not belong or is isolated from others (J. T. Cacioppo & Patrick, 2008). Loneliness and perceived isolation are also considered to be distinct from objective social isolation (for review, see J. T. Cacioppo & Cacioppo, 2014; S. Cacioppo, Grippo, London, Goossens, & Cacioppo, 2015). For instance, individuals may have few social contacts (objective isolation), but this degree of social connectedness may align with their personal preferences. Alternatively, individuals may be in frequent contact with family members or friends but perceive their social needs to be unmet (perceived isolation). Much like poor sleep, loneliness has been linked to a range of physical and mental health problems (for review, see J. T. Cacioppo & Cacioppo, 2014).
Relevant to this study, a growing body of research suggests that loneliness increases risk for various types of sleep problems. For instance, J. T. Cacioppo, Hawkley, Crawford, et al. (2002) found that young adults who reported elevated levels of loneliness on a self-report measure were significantly more likely than nonlonely young adults to demonstrate poor sleep efficiency and a greater time awake after sleep onset across in-laboratory and at-home assessments. Likewise, a 3-day study of 215 adults conducted by Hawkley, Preacher, and Cacioppo (2010) found that loneliness predicted greater daytime dysfunction independent of sleep duration. These results suggest that lonely individuals may be more likely to experience nonrestorative sleep even when sleeping the same amount of time as nonlonely individuals. Building on these studies, Kurina et al. (2011) found that greater self-reported loneliness significantly predicted greater sleep fragmentation among a sample of 95 adults living in a communal society, even after controlling for negative affect, age, sex, body mass index, and risk of sleep apnea. These studies are among those most commonly cited as support for the relationship between loneliness and subsequent sleep problems; yet, numerous other studies similarly have documented a significant relationship between these constructs (for review, see Hawkley & Capitanio, 2015).
At the same time, there is also preliminary support for the notion that various types of sleep problems increase risk for loneliness. For instance, Hom, Hames, et al. (2017) found in their study of 151 young adults that insomnia symptom severity at baseline significantly predicted levels of loneliness at 5-week follow-up, controlling for baseline loneliness severity. A study of 99 primary care patients conducted by Chu et al. (2017) also found a longitudinal relationship between self-reported sleep problems at baseline and thwarted belongingness (i.e., lack of meaningful social connection; Van Orden et al., 2010) at an average of 8 months of follow-up. In addition, the aforementioned study conducted by Hawkley et al. (2010) found that the relationship between daytime dysfunction and loneliness was bidirectional. This specific finding has been interpreted to suggest that sleep problems and perceived social isolation are reciprocally related (J. T. Cacioppo & Cacioppo, 2014). Even so, evidence for sleep problems as a risk factor for loneliness remains limited, especially compared with evidence for loneliness as a predictor of poor sleep. Thus, further research is needed to evaluate this longitudinal relationship.
It may be helpful to comment briefly on why poor sleep and loneliness might be related. Researchers have posited that given our reliance on other humans for survival, individuals who experience loneliness and perceived social isolation may experience hypervigilance to social threats (see predation; J. T. Cacioppo & Cacioppo, 2014; J. T. Cacioppo & Hawkley, 2009). This hypervigilance, in turn, may result in poor sleep quality and restless sleep because the brain remains vigilant even when an individual is asleep. Although human studies testing these hypotheses are limited, animal studies have revealed markers of disturbed sleep among mice that undergo objective social isolation (for review, see S. Cacioppo, Capitanio, & Cacioppo, 2014), which support these propositions. In terms of why poor sleep may contribute to loneliness, researchers have noted that disrupted sleep can lead to dysfunctional emotional reactivity and difficulties processing negative emotions (for review, see Baglioni, Spiegelhalder, Lombardo, & Riemann, 2010). These emotion-regulation difficulties may then contribute to interpersonal problems, including perceptions that one does not belong or difficulties engaging in social interactions. These problems then contribute to heightened loneliness. Researchers have also posited that insomnia itself may be intrinsically lonely—that is, the experience of being awake for many hours alone can lead to feelings of social isolation (Chu et al., 2016; Hom, Hames, et al., 2017; Littlewood, Kyle, Pratt, Peters, & Gooding, 2017). Research is needed, however, to rigorously test these theoretical explanations for the association between various types of sleep problems and loneliness.
Despite empirical and theoretical rationale for the association between sleep problems and loneliness, findings from studies that have assessed both sleep problems and loneliness have yet to be synthesized and meta-analyzed, to our knowledge. By doing so, we may be able to better evaluate the strength of the association between these constructs across diverse samples. Indeed, numerous studies have assessed both constructs even though their association was not the primary focus of the investigation. For instance, studies of suicide risk often include indices of sleep problems and perceptions of thwarted belongingness without directly seeking to examine the strength of their association (e.g., Oglesby, Capron, Raines, & Schmidt, 2015; Ringer et al., 2018; Silva et al., 2017). Evaluation of these studies alongside those focused on examining poor sleep and loneliness may expand our understanding of this association.
A meta-analysis of research in this domain will also reveal whether certain types of sleep complaints are uniquely associated with loneliness. Prior research suggests that insomnia in particular may demonstrate a robust relationship with mental health problems. For instance, insomnia symptoms, compared with nightmares, may not resolve spontaneously (Pigeon, Campbell, Possemato, & Ouimette, 2013) and remain even after posttraumatic stress disorder (PTSD) has been successfully treated (Zayfert & DeViva, 2004). Insomnia is also a strong predictor of depression, anxiety, and PTSD (Buysse et al., 2008; Pigeon & Gallegos, 2015; Taylor, Lichstein, Durrence, Reidel, & Bush, 2005). Together, these studies point to the especially pernicious effects of insomnia on mental well-being. The relationship between other sleep indices (e.g., sleep quality and sleep efficiency) and mental health problems is less well understood (e.g., for review, see Alvaro, Roberts, & Harris, 2013). Studies of sleep quality often take into account insomnia symptoms (e.g., difficulties initiating sleep), and nonrestorative sleep is a component of insomnia (Roth et al., 2010). Thus, it follows that sleep quality might be more strongly related to loneliness than other sleep complaints (e.g., nightmares) might be. However, given the extant literature, empirical support for these conjectures remains lacking. Loneliness, although strongly associated with mental health problems, is not necessarily a mental health problem itself. In this regard, a meta-analytic investigation will be especially useful.
Finally, a meta-analytic evaluation of the association between sleep problems and loneliness will facilitate the identification of moderators of this relationship. Such findings may reveal specific at-risk groups and other factors that strengthen this association. The following factors appear to be associated with greater prevalence and/or severity of sleep complaints: age (Ohayon, Carskadon, Guilleminault, & Vitiello, 2004), sex (Zhang & Wing, 2006), race (Grandner et al., 2013), and both physical and mental health problems (Stein, Belik, Jacobi, & Sareen, 2008). Likewise, each of these factors has demonstrated significant associations with loneliness (Hawkley et al., 2008; Holt-Lunstad, 2017; Meltzer et al., 2013; Stickley et al., 2013). It seems plausible, then, that the strength of the association between sleep and loneliness varies depending on these factors. In addition, research suggests that the severity and prevalence of sleep problems and loneliness have worsened in recent years (Holt-Lunstad, 2017; Keyes, Maslowsky, Hamilton, & Schulenberg, 2015). To inform intervention efforts, it will be helpful to investigate whether the sleep–loneliness association has also strengthened.
The Present Study
In sum, a growing body of research indicates that a link exists between sleep problems and loneliness—two constructs that independently and jointly have broad transdiagnostic significance. Efforts are needed, however, to synthesize data across studies that have captured these two constructs. The overarching goal of this investigation was to conduct a meta-analysis of the association between sleep problems—defined both broadly and specifically—and loneliness across the life span. Specifically, we aimed to investigate the cross-sectional association (a) between overall sleep problems (i.e., any sleep-related issue, e.g., disturbed indices of sleep health, general dissatisfaction with sleep, sleep disorder symptoms, and a diagnosis of a primary sleep disorder) and loneliness and (b) between specific types of sleep problems (i.e., insomnia, nightmares, poor sleep efficiency, and poor sleep quality) and loneliness. Specific types of sleep problems were evaluated to understand whether certain sleep complaints may be differentially associated with loneliness; insomnia, nightmares, poor sleep efficiency, and poor sleep quality were the four types of complaints most commonly represented in the literature, which facilitated meta-analytic evaluation. When possible, we also aimed to investigate the longitudinal relationship (a) between overall sleep problems and subsequent loneliness and (b) between loneliness and subsequent overall sleep problems. In addition, we sought to examine the influence of the following moderators on the association between sleep problems and loneliness: (a) year of publication, (b) sample type, (c) study aim, (d) demographic variables, and (e) geographic location of the study. These moderators were selected because of their common use in meta-analytic investigations (Borenstein, Hedges, Higgins, & Rothstein, 2009) and/or aforementioned associations with increased risk for sleep problems and/or loneliness.
Method
Search for studies
A comprehensive systematic search was first conducted on July 2, 2018, with a follow-up search conducted September 17, 2019, to identify potentially relevant studies for inclusion. The following electronic databases were searched: PubMed, MEDLINE, and PsycINFO. Search terms were the following: (insomnia OR sleep* OR nightmare*) AND (lonel* OR “social isolation” OR “sense of belonging” OR “belongingness”) AND English [la] NOT (animals[mesh] NOT humans[mesh]). These terms facilitated the identification of English-language studies probing various types of sleep problems and loneliness, excluding animal and nonhuman studies. We selected the three aforementioned sleep-related terms to balance precision and recall. Although “insomnia” and “nightmare*” are relatively specific, “sleep” is a broad term that is inclusive of various sleep-related constructs (e.g., sleep efficiency, poor sleep, sleep problems, sleep complaints, sleep-disordered breathing). We used test-sample identification (i.e., ensuring that known relevant studies were identified by the search) and queried the broad term “sleep” because of the discovery that using only specific sleep terms (e.g., “sleep complaints”) resulted in the omission of key studies. We also conducted reference list searches to locate articles not yielded by our initial search. This approach is consistent with recommendations for balancing rigor and feasibility in meta-analyses (Valentine, Hedges, & Cooper, 2009).
Inclusion and exclusion criteria
Articles were included according to the following criteria: (a) written in English; (b) study of humans (i.e., no animal studies); (c) inclusion of a measure of sleep problems, defined broadly (e.g., insomnia, nightmares, poor sleep efficiency, poor sleep quality, sleep-disordered breathing); (d) inclusion of a measure of loneliness (i.e., feelings of loneliness, perceived social isolation, lack of meaningful social connection, thwarted belongingness); and (e) reporting of at least one effect size for the association between sleep problems and loneliness. For articles in which an effect size was not reported, we contacted study authors to obtain this information; we made two attempts to contact each corresponding author. Articles were excluded according to the following criteria: (a) no quantitative data reported, (b) only case studies reported, (c) no distinct measure of sleep problems (e.g., conflation of poor sleep and depressed mood), and (d) no distinct measure of loneliness (e.g., conflation of loneliness and objective social isolation).
For this meta-analysis, we elected to focus on subjective sleep problems (e.g., difficulties falling asleep, poor sleep quality) and indices of poor sleep health (e.g., sleep efficiency). Other sleep indices—for instance, sleep duration, bedtime, and wake time—although informative, cannot be uniformly interpreted as being distressing across individuals. That is, individuals may sleep fewer hours compared with others yet not find their sleep duration to be problematic (J. T. Cacioppo & Hawkley, 2009; Hawkley & Cacioppo, 2007; Ohayon, 2002). Note that although sleep efficiency can be objectively assessed, we included this construct in this study because measures of sleep disturbance (e.g., Pittsburgh Sleep Quality Index; Buysse, Reynolds, Monk, Berman, & Kupfer, 1989) incorporate sleep efficiency into their conceptualization of perceived sleep problems. In addition, poor sleep efficiency is strongly associated with subjective sleep complaints, such as insomnia (Morin et al., 2011). For social isolation, we elected to focus on perceived social isolation (i.e., loneliness) rather than objective social isolation because perceived isolation is considered to be uniquely associated with adverse outcomes (J. T. Cacioppo, Hawkley, Berntson, et al., 2002; Hawkley et al., 2008; Hawkley et al., 2010). Furthermore, as noted, objective indices of social interaction—for example, number of close friends, size of one’s social network, time spent with others—do not serve as uniformly interpretable signals of distress.
Study selection
Using the above criteria, two independent coders conducted an initial screening of the titles and abstracts for all studies identified to determine appropriateness for the current meta-analysis. this screening identified 1,421 potentially relevant reports. After duplicates were removed (N = 645), a remaining 776 records were screened on the basis of their titles and abstracts. At this stage, an additional 374 articles were excluded on the basis of our search criteria. Agreement between coders during this initial screening was good (κ = .88). Then, a second round of screening was conducted with the full texts of the remaining 402 manuscripts; good agreement was observed between coders at this stage (κ = .89). Disagreements were resolved through discussion between the coders and consultation with a third independent coder. In instances in which multiple articles were published using the same data set, consistent with best practices (Valentine et al., 2009), we determined a priori that the article reporting findings from the largest sample size would be selected for inclusion to reduce redundancy; 10 redundant samples were excluded. We also contacted a total of 150 authors for effect-size information; 19 authors followed up with the requested data; the remaining authors either declined the data request or were unreachable. After excluding additional studies not eligible for inclusion on the basis of our exclusion criteria (k = 318), a total of 84 articles reporting on 110 distinct samples (N = 227,112) were selected for inclusion in the current meta-analysis. See Figure 1 for a detailed flow chart of the study selection process.
Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow chart for study selection.
Data extraction
Two independent coders then extracted the following variables from each study identified for inclusion: (a) first author; (b) publication year; (c) sample type; (d) study design (i.e., cross-sectional or longitudinal); (e) whether the study sought to evaluate the relationship between sleep problems, defined broadly, and loneliness; (f) length of study follow-up, if applicable; (g) geographic location of study (i.e., country in which data were collected); (h) sample size (N); (i) age (mean and standard deviation); (j) sex (percentage female); (k) race, as relevant (percentage White, percentage Black/African American, percentage Asian, percentage Native American/Alaska Native, percentage other, and percentage Hispanic/Latinx, if conceptualized as race rather than ethnicity); (l) ethnicity, as relevant (percentage Hispanic/Latinx); (m) measure of loneliness and its respective mean and standard deviation in the sample; (n) measure of sleep problems and its respective mean and standard deviation in the sample; and (o) bivariate correlations (i.e., Pearson’s r) or other index of effect size for the association between sleep problems and loneliness. For longitudinal studies with multiple time points, we extracted the effect sizes capturing both the shortest and longest periods of time. Interrater reliability for data extraction was excellent (κ = .92). Discrepancies in data extraction were resolved through discussion between coders or by a third independent coder if necessary.
Data-analytic strategy
Consistent with effect sizes reported by the majority of included articles, when possible, effect sizes were first converted to r values (.10 = small, .30 = medium, and .50 = large; Cohen, 1988).1,2 For all meta-analytic models, we used a Q test to examine whether substantial heterogeneity in effect sizes was observed across studies (Huedo-Medina, Sánchez-Meca, Marín-Martínez, & Botella, 2006). In addition, the I2 statistic was used to evaluate the degree of variability across studies that was due to heterogeneity as opposed to chance-related sampling error (.25 = small degree of heterogeneity, .50 = medium, and .75 = large; Higgins, Thompson, Deeks, & Altman, 2003). Random effects models were employed for all analyses because we anticipated that effect sizes would vary across studies given the diversity of samples included and methodologies leveraged. Forest plots were constructed to depict effect sizes with 95% confidence intervals (CI) for all samples in our overall cross-sectional model and two longitudinal models.
Specifically, to address study aims, we used random effects meta-analytic models to examine the cross-sectional relationship (a) between overall sleep problems and loneliness (k = 108) and (b) between specific sleep problems (i.e., insomnia, nightmares, poor sleep efficiency, and poor sleep quality) and loneliness (ks = 5–27). Given that not all studies used validated indices of sleep problems and loneliness, a series of meta-analytic models that included only effect sizes derived from validated indices of sleep problems and loneliness were conducted to investigate whether the use of unvalidated measures affected the pattern of cross-sectional findings (for details regarding which effect sizes were derived from validated vs. unvalidated measures, see Tables 1 and 2; also see Table S1 in the Supplemental Material available online). For all cross-sectional meta-analytic models, only one effect size from each sample was used; a preference for inclusion was given to the most general index of sleep complaints (i.e., poor sleep quality given preference over insomnia and poor sleep efficiency, insomnia given preference over nightmares). We used Fisher’s r-to-z transformations as exploratory analyses to compare the relative strength of effect sizes by sleep-complaint type (e.g., comparing the strength of the association between insomnia and loneliness with the strength of the association between nightmares and loneliness). We also leveraged meta-analytic models to investigate the longitudinal relationship (a) between overall sleep problems and subsequent loneliness (k = 6) and (b) between loneliness and subsequent overall sleep problems (k = 6) using effects sizes from the (a) shortest follow-up time points and (b) longest follow-up time points.
Sample Characteristics and Effect Sizes for Included Cross-Sectional Studies
Note: H = hypersomnia; I = insomnia; N = nightmares; r = Pearson’s r; S = sample; SDB = sleep-disordered breathing; SE = sleep efficiency; SF = sleep fragmentation; SPC = sleep-pattern changes; SQ = sleep quality; U.K. = United Kingdom; U.S. = United States.
This effect size was derived from measures for which the psychometric properties have yet to be established in peer-reviewed research. bThis effect size was included only in sleep-subtype analyses because the sample was redundant with another study. cThis study was a longitudinal study and its effect size appears in Table 2. dThis value is a Spearman’s ρ.
Cross-Sectional Effect Sizes for Included Longitudinal Studes
Note: If two effect sizes are given, the first is for the shortest follow-up time point, and the second is for the longest follow-up time point. L = loneliness; SP = sleep problem.
This effect size was derived from measures for which the psychometric properties have yet to be established in peer-reviewed research.
When sufficient data were available (i.e., at least six samples with metrics of interest for categorical moderators; Borenstein et al., 2009) and models yielded a significant effect, univariate metaregression analyses were used to examine whether variability in effect sizes was accounted for by (a) year of publication, (b) sample type (i.e., medical vs. nonmedical, psychiatric vs. nonpsychiatric, adult vs. child or adolescent 3 ), (c) whether the study specifically sought to examine the relationship between sleep problems and loneliness, (d) demographic variables (i.e., age, sex, race), and (e) geographic location of study (i.e., United States vs. another country). Moderators were examined as continuous variables whenever possible (e.g., age, percentage of females). When sufficient data were available, multivariate metaregression models were evaluated to compare the relative contributions of proposed moderators and to address concerns regarding multiple testing. We did not evaluate first author or research group as a moderator given that no researchers or research groups emerged as predominant generators of research in this domain.
Finally, two indices of publication bias were examined. First, given the inclusion of a large number of studies in our analyses, we used the Egger’s linear-regression-intercept test; significant t values for the Egger’s test signal potential publication bias (Egger, Smith, & Phillips, 1997). Second, we constructed funnel plots and employed Duval and Tweedie’s (2000a, 2000b) trim-and-fill method. This method iteratively removes the smallest studies from the positive side of the funnel plot, computing effects at each iteration until the resulting plot is symmetrical. Then, an overall effect estimate is calculated, correcting for potential bias. The evaluation of study aim and geographic location as potential moderators was, in part, an additional effort to identify possible publication bias. Given file-drawer effects—as well as research suggesting that U.S.-based behavioral studies in particular yield larger effect sizes than studies conducted in other countries (Fanelli & Ioannidis, 2013)—it seemed possible that both factors might moderate the sleep–loneliness relationship.
Analyses were conducted using Comprehensive Meta-Analysis (Version 3; Borenstein et al., 2013).
Results
Sample characteristics
Of the 110 unique samples included in this study, 99.1% reported cross-sectional findings, and 7.3% reported longitudinal findings (months of follow-up at the shortest time point from baseline: range = 1–84, M = 25.78, SD = 34.42; months of follow-up at the furthest time point from baseline: range = 1–84, M = 27.53, SD = 33.21). A plurality of samples was from articles published between 2016 and 2019 (40.0%); a plurality was also conducted in North America (42.7%), with all North American studies having been conducted in the United States. Sample sizes most commonly ranged in size from 100 to 199 (24.5%), and the majority (50.6%) of samples reported participant mean ages falling within the adult range (i.e., 25–59 years). On average, the percentage of female participants in each sample was 55.2% (of the 105 samples reporting sex), and the percentage of participants identifying as White was 73.0% (of the 50 samples reporting race; relatively few samples reported race, likely in part because of the prevalence of non-U.S. samples included). Across samples, 16.4% were composed of a medical population (e.g., spinal cord injury patients, adults with chronic fatigue syndrome) and 7.3% a psychiatric population (e.g., psychiatric inpatients, psychiatric outpatients). See Table 3 for details.
Summary of Characteristics of Included Samples (k = 110)
The categories are not mutually exclusive.
In terms of how sleep problems were assessed, 47.3% of samples (k = 52) used a nonspecific index of sleep problems, 25.5% (k = 28) an index of poor sleep quality, 22.7% (k = 25) insomnia, 4.5% (k = 5) nightmares, 5.5% (k = 6) poor sleep efficiency, 2.7% (k = 3) sleep-disordered breathing, 0.9% (k = 1) insomnia/hypersomnia, 0.9% (k = 1) sleep fragmentation, and 0.9% (k = 1) sleep-pattern changes. Regarding specific measures used, across the 110 samples, a total of 29 different indices of sleep problems were used. The most commonly employed indices of sleep problems were the Pittsburgh Sleep Quality Index (Buysse et al., 1989; k = 22), iterations of the Insomnia Severity Index (Bastien, Vallieres, & Morin, 2001; k = 18), and the Nottingham Health Profile (Hunt, McKenna, McEwen, Williams, & Papp, 1981; k = 11). Of the four specific sleep indices examined, the only index assessed via objective measures was sleep efficiency (actigraphy, k = 1; nightcap, k = 1); self-report measures were also used to capture sleep efficiency (researcher-developed questionnaire, k = 1; Verran/Snyder-Halpern Sleep Scale, k = 3). For loneliness, 22 different assessments of loneliness were used; iterations of the UCLA Loneliness Scale (Russell, Peplau, & Cutrona, 1980; Russell, 1996; k = 30), iterations of the Interpersonal Needs Questionnaire (Van Orden, Cukrowicz, Witte, & Joiner, 2012; k = 19), and the Nottingham Health Profile (k = 13) were the most commonly used. Detailed information regarding outcomes assessed and measures used are presented in Tables 1 and 2 and Table S1 in the Supplemental Material.
Cross-sectional relationship between sleep problems and loneliness
Any sleep problems and loneliness: all samples
A total of 108 samples investigated the relationship between any type of sleep problems and loneliness (N = 226,183). The Q test was significant (1,471.765), and the I2 value (92.730) suggested that a large degree of variability was due to heterogeneity as opposed to chance. The random effects model revealed a significant medium association between overall sleep problems and loneliness (r = .336, p < .001, 95% CI = [.315, .357]). Univariate metaregressions indicated that this association was significantly stronger among samples in articles published more recently (z = 2.64, p = .008, k = 108), nonmedical samples (z = −2.01, p = .044), adult compared with child or adolescent samples (z = 2.15, p = .032, k = 108), and U.S. samples (z = 3.01, p = .003, k = 108). None of the following emerged as significant moderators of this association: psychiatric status, study aim, mean age, sex, or race (ps > .05). When all potential moderators were evaluated in one multivariate metaregression model (k = 50), samples from studies published more recently (z = 2.92, p = .004), samples with a greater proportion of male participants (z = −2.55, p = .011), and U.S. samples (z = 2.55, p = .011) evinced strong effect sizes. Egger’s regression test did not suggest publication bias, b = −0.272, SE = 0.437, t(106) = 0.623, p = .535. Per Duval and Tweedie’s trim-and-fill procedure, five studies were trimmed (for funnel plot, see Fig. S1 in the Supplemental Material. Effect sizes and corresponding 95% CIs are presented in a forest plot in Figure 2.
Forest plot of all cross-sectional effect sizes (k = 110).
Any sleep problems and loneliness: validated measures only
Fifty-six samples explored the association between any type of sleep problems and loneliness with both constructs assessed via validated measures (N = 25,214). The Q test was significant (428.107), and the I2 value (87.153) indicated that a large degree of variability was due to heterogeneity rather than chance. The random effects model yielded a significant medium association between overall sleep problems and loneliness as assessed by validated measures (r = .346, p < .001, 95% CI = [.307, .385]). Univariate metaregressions revealed that this association was significantly moderated by year such that samples from articles published more recently evinced stronger effects (z = 2.72, p = .007, k = 56). This association was also significantly stronger among samples from studies conducted in the United States (z = 2.81, p = .005, k = 56). The association was not significantly moderated by sample type (i.e., medical or psychiatric status, age group), study aim, mean age, sex, or race (ps > .05). Multivariate metaregressions including all potential moderators (k = 32) revealed a stronger association between overall sleep problems and loneliness among samples in studies published more recently (z = 4.44, p < .001), samples with a greater proportion of male participants (z = −2.64, p = .008), and U.S. samples (z = 3.01, p = .003). Egger’s regression test did not indicate publication bias, b = 0.945, SE = 0.664, t(54) = 1.422, p = .161. Two studies were trimmed per trim-and-fill procedures; the funnel plot is presented in Figure S2 in the Supplemental Material.
Insomnia and loneliness: all samples
A total of 25 samples examined the association between insomnia and loneliness (N = 44,905). The Q test was significant (569.223), and the I2 value (95.784) indicated that a large degree of variability was due to heterogeneity as opposed to chance. The random effects model revealed a significant medium association between insomnia and loneliness (r = .354, p < .001, 95% CI = [.300, .409]). In univariate analyses, this association was significantly moderated by publication year, sex, and geographic location; samples from studies published more recently (z = 2.53, p = .012, k = 25), samples with a greater proportion of male participants (z = −2.54, p = .011, k = 25), and U.S. samples (z = 3.97, p < .001, k = 25) evinced stronger effects. None of the following were significant moderators of this association: sample type (i.e., medical or psychiatric status, age group), whether the study’s aim was to examine the association between sleep and loneliness, mean age, or race (ps > .05). In multivariate moderation analyses (k = 20), this association was stronger among samples from articles published more recently (z = 2.80, p = .005), U.S. samples (z = 2.83, p = .005), and studies with the aim of evaluating the sleep–loneliness association (z = 2.10, p = .036). Egger’s regression test signaled potential publication bias, b = 2.718, SE = 1.310, t(23) = 2.074, p = .049. Two studies were trimmed per Duval and Tweedie’s trim-and-fill procedure (for funnel plot, see Fig. S3 in the Supplemental Material).
Insomnia and loneliness: validated measures only
Fourteen samples investigated the association between insomnia and loneliness with both constructs assessed via validated measures (N = 5,384). The Q test was significant (146.977); the I2 value (91.155) revealed that a large amount of variability was attributable to heterogeneity rather than chance. The random effects model indicated a significant medium association between insomnia and loneliness, both assessed with validated measures (r = .412, p < .001, 95% CI = [.332, .486]). In univariate analyses, this association was significantly stronger among samples from articles published more recently (z = 2.14, p = .033, k = 14) and with a higher mean age (z = 2.46, p = .014, k = 14), greater proportion of male participants (z = −2.73, p = .006, k = 14), and greater proportion of participants identifying as White (z = 3.07, p = .002, k = 13). None of the following emerged as significant moderators of this association: psychiatric status, age group, study aim, and geographic location (ps > .05). There were insufficient data to evaluate medical status as a moderator (k < 6). Because of collinearity issues, a multivariate model comparing the relative contributions of each potential moderator could not be evaluated. Egger’s regression test did not suggest publication bias, b = 1.484, SE = 3.029, t(12) = 0.490, p = .633. No studies were trimmed per Duval and Tweedie’s trim-and-fill procedures (for funnel plot, see Fig. S4 in the Supplemental Material).
Nightmares and loneliness
Five samples were included in analyses evaluating the association between nightmares and loneliness (N = 2,173). All five of the samples that tested this association used validated measures; thus, there was no difference in findings between our analyses with (a) all studies examining this association and (b) studies using validated measures only. The Q test was not significant (5.755), and the I2 value (30.500) indicated that a small-to-medium degree of variability was attributable to heterogeneity as opposed to chance. A significant small-to-medium correlation was observed between nightmares and loneliness (r = .193, p < .001, 95% CI = [.140, .245]). Neither publication year, mean age, sex, nor race emerged as a significant moderator in univariate metaregression analyses (ps > .05). There were insufficient data to evaluate categorical moderators of this association (i.e., k < 6) and an insufficient number of samples to conduct multivariate metaregression analyses. Neither Egger’s regression test, b = 0.096, SE = 2.484, t(3) = 0.039, p = .972, nor Duval and Tweedie’s trim-and-fill procedures (no studies trimmed) suggested publication bias. Funnel plots are presented in Figures S5 and S6 in the Supplemental Material.
Poor sleep efficiency and loneliness: all samples
Six samples were included in analyses examining the association between poor sleep efficiency and loneliness (N = 600). The Q test was not significant (1.755), and the I2 value was < .001, which suggests that variability may have been due to chance. A significant small association was identified between poor sleep efficiency and loneliness (r = .165, p < .001, 95% CI = [.084, .246]). For univariate metaregression analyses, neither publication year, mean age, sex, race, nor age group was a significant moderator. There were insufficient data to evaluate medical or psychiatric status, study aim, or geographic location as moderators of this association (i.e., all studies fell under the same subcategory within each). Likewise, there was an insufficient number of samples to conduct multivariate metaregression analyses. Egger’s regression test did not indicate publication bias, b = −0.469, SE = 1.823, t(4) = 0.257, p = .810, and following Duval and Tweedie’s trim-and-fill procedures, we trimmed one study. For a relevant funnel plot, see Figure S7 in the Supplemental Material.
Poor sleep efficiency and loneliness: validated measures only
Five samples were included in analyses examining the association between poor sleep efficiency and loneliness assessed via validated measures (N = 462). The Q test was not significant (0.908), and the I2 value was < .001, which suggests that variability may have been due to chance. A significant small association was identified between poor sleep efficiency and loneliness (r = .144, p = .002, 95% CI = [.051, .237]). Neither publication year, mean age, sex, nor race emerged as significant moderators in univariate metaregression analyses. There were insufficient data to evaluate categorical moderators of this association (i.e., k < 6) and an insufficient number of samples to conduct multivariate metaregression analyses. Egger’s regression test did not indicate publication bias, b = 1.221, SE = 1.653, t(3) = 0.738, p = .514, and following Duval and Tweedie’s trim-and-fill procedures, we trimmed two studies. See Figure S8 for a relevant funnel plot.
Poor sleep quality and loneliness: all samples
A total of 27 study samples examined the association between poor sleep quality and loneliness (N = 18,254). The Q test was significant (82.768), and the I2 value (68.587) indicated that a large degree of variability was due to heterogeneity as opposed to chance. The random effects model revealed a significant small-to-medium association between poor sleep quality and loneliness (r = .285, p < .001, 95% CI = [.252, .318]). Univariate metaregressions indicated that this association was significantly moderated by geographic location such that this association was stronger among U.S. samples (z = 3.36, p = .001, k = 27). None of the following represented a significant moderator of this association: publication year, sample type (i.e., medical or psychiatric status, age group), study aim, mean age, sex, or race (ps > .05). Multivariate metaregression analyses (k = 14) did not reveal any significant moderators. Egger’s regression test did not indicate publication bias, b = 0.314, SE = 0.560, t(25) = 0.560, p = .580. Following the use of Duval and Tweedie’s trim-and-fill procedure, we trimmed five studies (for funnel plot, see Fig. S9 in the Supplemental Material).
Poor sleep quality and loneliness: validated measures only
Eighteen samples investigated the relationship between poor sleep quality and loneliness with validated measures to assess both constructs (N = 11,094). The Q test was significant (64.516), and the I2 value (73.650) revealed that a large amount of variability was attributable to heterogeneity rather than chance. The random effects model identified a significant small-to-medium association between poor sleep quality and loneliness as assessed by validated measures (r = .292, p < .001, 95% CI = [.248, .334]). In univariate metaregression analyses, this association was stronger among U.S. samples (z = 3.60, p < .001, k = 18). Neither publication year, medical status, age group, study aim, mean age, sex, nor race significantly moderated this association (ps > .05). There were insufficient data to evaluate psychiatric status as a moderator (i.e., all 18 samples were nonpsychiatric). There was also an insufficient number of studies to construct a multivariate metaregression model. Egger’s regression test, b = 1.058, SE = 0.709, t(16) = 1.492, p = .155, did not suggest publication bias; six studies were trimmed per Duval and Tweedie’s trim-and-fill procedures (for funnel plot, see Fig. S10 in the Supplemental Material).
Exploratory analyses comparing effect sizes by sleep complaint
The association between insomnia and loneliness (r = .354) was significantly stronger than that between (a) nightmares and loneliness (r = .193; z = 7.94, p < .001), (b) poor sleep efficiency and loneliness (r = .165; z = 4.94, p < .001), and (c) poor sleep quality and loneliness (r = .285; z = 8.76, p < .001). In addition, the association between poor sleep quality and loneliness (r = .285) was significantly stronger than that (a) between nightmares and loneliness (r = .193; z = 4.30, p < .001) and (b) between poor sleep efficiency and loneliness (r = .165; z = 3.04, p = .002). The same pattern of findings emerged when using validated sleep indices only. No other significant differences among effect sizes for specific sleep complaints were observed (i.e., ps > .05).
Longitudinal relationship between overall sleep problems and subsequent loneliness
Six samples evaluated the longitudinal relationship between sleep problems and subsequent loneliness.
Shortest time frames
Using effect sizes from the shortest available follow-up time point, we included 1,042 individuals in our meta-analytic model (for forest plot, see Fig. S11 in the Supplemental Material). The Q test was significant (12.084), and the I2 value (58.624) indicated that a medium-to-large degree of variability was due to heterogeneity rather than chance. The random effects model revealed a significant medium relationship between overall sleep problems and loneliness at follow-up (r = .297, p < .001, 95% CI = [.206, .382]. Univariate metaregression analyses did not reveal any significant moderators of this longitudinal relationship (ps > .05). There was an insufficient number of studies to construct a multivariate metaregression model. Neither Egger’s regression test, b = 2.791, SE = 4.246, t(4) = 0.657, p = .547, nor Duval and Tweedie’s trim-and-fill procedures (no studies trimmed) indicated publication bias (for funnel plot, see Fig. S12 in the Supplemental Material).
Longest time frames
A total of 941 individuals were included in our meta-analytic model evaluating effect sizes at the longest available follow-up time point (for forest plot, see Fig. S13 in the Supplemental Material). The Q test was significant (12.473), and the I2 value (59.915) suggested that a medium-to-large degree of variability was due to heterogeneity as opposed to chance. The random effects model revealed a significant medium relationship between overall sleep problems and loneliness at follow-up (r = .294, p < .001, 95% CI = [.196, .385]). Univariate metaregression analyses did not identify any significant moderators of this longitudinal relationship (ps > .05). There was an insufficient number of studies to conduct multivariate metaregression analyses. Egger’s regression test did not indicate publication bias, b = −1.717, SE = 4.711, t(4) = 0.364, p = .734), and neither did Duval and Tweedie’s trim-and-fill procedures (no studies were trimmed; for funnel plot, see Fig. S14 in the Supplemental Material).
Longitudinal relationship between loneliness and subsequent overall sleep problems
A total of six samples examined the longitudinal relationship between loneliness and subsequent sleep problems.
Shortest time frames
Using effect sizes from the shortest follow-up time point available, we included a total of 1,507 individuals in our model (for forest plot, see Fig. S15 in the Supplemental Material). The Q test was significant (13.021), and the I2 value (61.601) suggested that a medium-to-large degree of variability was due to heterogeneity as opposed to chance. The random effects model revealed a significant small-to-medium relationship between loneliness and overall sleep problems at follow-up (r = .267, p < .001, 95% CI = [.174, .356]). This relationship was significantly stronger among adults than children or adolescents (z = 3.09, p = .002, k = 6) and samples with a higher mean age (z = 2.89, p = .004, k = 6). None of the following were significant moderators: publication year, study aim, sex, geographic location, or length of the follow-up period (ps > .05). There were insufficient data to evaluate medical or psychiatric status (i.e., ks < 6) or race (k = 2) as moderators. There was also an insufficient number of studies to conduct multivariate metaregression analyses. Egger’s regression test did not indicate publication bias, b = 1.343, SE = 1.906, t(4) = 0.705, p = .520. Two studies were trimmed per Duval and Tweedie’s trim-and-fill procedures (for funnel plot, see Fig. S16 in the Supplemental Material).
Longest time frames
A total of 1,501 individuals were included in our meta-analytic model evaluating effect sizes at the longest follow-up time point available (for forest plot, see Fig. S17 in the Supplemental Material). The Q test was significant (13.680), and the I2 value (63.450) indicated that a medium-to-large amount of variability was due to heterogeneity rather than chance. The random effects model yielded a significant small-to-medium relationship between loneliness and overall sleep problems at follow-up (r = .249, p < .001, 95% CI = [.152, .341]). This relationship was significantly stronger among adults than children or adolescents (z = 2.52, p = .012, k = 6) and samples with a higher mean age (z = 2.76, p = .006, k = 6). None of the following significantly medium this relationship: publication year, study aim, sex, geographic location, or length of the follow-up period. There were insufficient data to evaluate medical or psychiatric status (i.e., ks < 6) or race (k = 2) as moderators. Neither Egger’s regression test, b = 0.703, SE = 2.026, t(4) = 0.347, p = .746, nor Duval and Tweedie’s trim-and-fill procedures (no studies trimmed) indicated publication bias (for funnel plot, see Fig. S18 in the Supplemental Material).
Detailed findings for all meta-analytic models are presented in Table 4.
Meta-Analytic Findings
Note: Values in square brackets are 95% confidence intervals. k = number of samples; ET = Egger’s regression test, two-tailed; T&F = Duval and Tweedie’s trim and fill test, random effects estimates reported.
p < .05.
Discussion
In this study, we meta-analyzed data from 84 peer-reviewed articles (k = 110, N = 227,112). We found a significant medium cross-sectional association between overall sleep problems and loneliness. We also found small-to-medium and medium associations between specific types of sleep complaints (i.e., insomnia, nightmares, poor sleep efficiency, and poor sleep quality) and loneliness. In addition, our longitudinal analyses indicated that overall sleep problems and loneliness may be bidirectionally associated. Although several significant moderators emerged for specific associations, no consistent moderators were identified across all models.
Findings from this meta-analysis support a robust association between sleep problems defined broadly and loneliness. Indeed, this association remained significant regardless of the specific sleep outcome evaluated (e.g., overall sleep problems or insomnia) and regardless of whether only studies using validated measures of sleep and loneliness were analyzed. The number and diversity of samples included in this investigation further underscore the robustness of our results. For instance, our cross-sectional analyses included 108 unique samples that differed substantially with regard to sociodemographic characteristics, geographic location, size, and type. Even so, no significant moderators were consistently observed across all meta-analytic models, which suggests that these effects were not being driven entirely by a specific sample type or demographic group. It is also worth noting that across samples, 29 distinct indices of sleep problems (e.g., disturbed indices of sleep health, general dissatisfaction with sleep, sleep disorder symptoms, a diagnosis of a primary sleep disorder) and 22 distinct indices of loneliness were used. Thus, it seems that our findings were not merely an artifact of a correlation existing between specific measures of sleep and loneliness.
Relatively few longitudinal studies were identified by our literature search—six evaluated the relationship between sleep problems and subsequent loneliness and six the relationship between loneliness and subsequent sleep problems. These longitudinal studies also differed with regard to the time frame evaluated. Thus, although our analyses point to the existence of bidirectional associations between overall sleep problems and loneliness, further longitudinal research is needed to understand how these factors relate to one another over time. It will also be important for such research to control for the baseline effects of the dependent variable; otherwise, as is well known, longitudinal associations may simply suggest that both constructs are stable and co-occur over time. Evaluation of the longitudinal relationship between specific sleep complaints and loneliness may also be informative.
Overall, our results provide strong support for a link between sleep problems defined both broadly and more specifically and loneliness. In this regard, our findings align with aforementioned theoretical and empirical work on the relationship between sleep and loneliness (J. T. Cacioppo & Cacioppo, 2014; J. T. Cacioppo & Hawkley, 2009; Hawkley & Capitanio, 2015; Hom, Hames, et al., 2017; Littlewood et al., 2017). Our findings may also explain other associations that have been observed in the broader literature on sleep problems, loneliness, and adverse mental health outcomes. For instance, our results support the conjecture that loneliness (see thwarted belongingness; Joiner, 2005; Van Orden et al., 2010) may be an explanatory factor that accounts for insomnia’s consistent link to suicide deaths (e.g., Bernert, Turvey, Conwell, & Joiner, 2014; Robins, 1981). Our findings may also explicate why disturbed sleep and depression appear to be reciprocally related (Manber & Chambers, 2009). Sleep problems may exacerbate loneliness, contributing to depressive symptoms; conversely, depression, which often affects social functioning, may exacerbate sleep problems via the pathway of loneliness. In this way, sleep problems and loneliness may scaffold one another—a concept with transdiagnostic relevance. On this point, although our study was not designed to delineate the mechanisms underpinning the sleep–loneliness association, our results highlight the value and potentially broad impact of research that aims to do so.
Given the multiple indices of publication bias included in our study, it did not appear that such bias notably influenced our patterns of results. Just one of our Egger’s regression tests (for the model evaluating the association between insomnia and loneliness) was statistically significant, and few studies were trimmed on the basis of Duval and Tweedie’s trim-and-fill procedures. At most, six studies were trimmed for our meta-analytic model that evaluated the association between poor sleep quality and loneliness among 18 separate samples using only validated measures; in some cases, no studies were trimmed. Furthermore, across models, adjusted estimates after studies were trimmed still indicated significant associations between sleep problems and loneliness, which suggests that even after accounting for potentially biased studies, our results remained the same. In addition, as discussed above, one of the goals of our meta-analysis was to synthesize findings from studies that did not necessarily endeavor to demonstrate that a relationship exists between sleep problems and loneliness (e.g., Ringer et al., 2018; Silva et al., 2017). By including these studies in our analyses, our confidence that publication bias minimally influenced our results was further enhanced. Across our meta-analytic models, samples from studies that sought to examine the sleep–loneliness relationship did not evince significantly stronger effect sizes than those that did not focus specifically on this relationship. The only exception was in the multivariate metaregression model evaluating moderators of the association between insomnia and loneliness. After controlling for other potential moderators, studies that aimed to evaluate this association evinced stronger effects than those that did not. This result aligns with the Egger’s regression test being significant for the insomnia–loneliness meta-analytic model and signals a need for that model’s results to be interpreted cautiously.
Although no clear pattern of significant moderators emerged across all meta-analytic models, it may still be informative to examine moderators that were identified for specific associations taking into account low power as a consideration. The most common moderator identified across models was geographic location. U.S. samples evinced stronger effects across five of the 10 models for which we were able to evaluate categorical moderators. Even more critically, in three out of four of our multivariate metaregression models, geographic location remained a significant moderator after accounting for the influence of other moderating factors. This finding emphasizes the strength of geographic location as a moderator. These findings are consistent with research suggesting that U.S.-based behavioral research studies, on the whole, yield larger effect sizes than those conducted in other countries (Fanelli & Ioannidis, 2013). Although a degree of publication bias may account for these results, there are also sociocultural factors that may underlie this association. For example, social-media usage (Jackson & Wang, 2013), individualism and family fragmentation (Lykes & Kemmelmeier, 2014), opioid use problems (Manchikanti et al., 2012), increasing suicide rates (Curtin, Warner, & Hedegaard, 2016), and vigilance associated with gun ownership (Karp, 2018) may all play a role. Research is needed to test these potential explanations.
Our moderation analyses also indicated that age may influence the relationship between sleep problems—broadly and specifically—and loneliness. Three of our models yielded stronger effects among samples with a greater mean age, and three models yielded stronger effects among adult samples compared with child or adolescent samples. These findings dovetail with research suggesting that loneliness exacerbates sleep problems as individuals age (Hawkley & Cacioppo, 2007) and that variability in sleep is normative among children and adolescents (Jenni & Carskadon, 2007). Yet we would be remiss not to note research suggesting that puberty is strongly associated with both sleep problems and depression (for review, see Colrain & Baker, 2011). By examining mean age and age group rather than developmental stage as moderators, our age-related findings are limited in interpretability. It is possible that the sleep–loneliness association was especially strong among a subset of younger individuals (e.g., those undergoing puberty); however, we did not have access to data that would allow us to examine developmental stage as a moderator. Moving forward, efforts are needed to evaluate the role of development in addition to chronological age. Considering other demographic moderators, samples with a greater proportion of male participants yielded stronger effects in two multivariate metaregression models and two univariate models. Interestingly, the literature indicates that sleep problems and feelings of loneliness both appear to be more common among female participants than male participants (e.g., Meltzer et al., 2013; Zhang & Wing, 2006). Our multivariate moderation findings suggest, however, that sleep problems defined broadly and loneliness may be particularly strongly linked among males such that exacerbation of one may be more likely to be associated with the exacerbation of the other. Efforts are needed to better understand why this may be the case. A reluctance to seek help for mental health problems among males (e.g., Milner & De Leo, 2010), in particular, may be one possible explanation requiring examination.
Finally, both models evaluating the association between overall sleep problems and loneliness as well as both models examining the insomnia–loneliness relationship found stronger effects among samples from studies published more recently. Note that publication year was also identified as a significant moderator in three of four multivariate regression models, which highlights the influence of publication year above and beyond the effects of other potential moderators. These findings indicate that the relationship between these constructs has potentially worsened over time—findings consistent with what appear to be recent population-wide trends in sleep (Keyes et al., 2015) and loneliness (Holt-Lunstad, 2017) worsening. Here, too, however, further research is needed.
Regarding implications for clinical practice, our study suggests that it may be advantageous to monitor loneliness among individuals with various types of sleep problems as well as various types of sleep problems among individuals experiencing loneliness. Although our investigation did not establish a causal link between these constructs, the robust cross-sectional association observed between both suggest, at a minimum, that they are likely to co-occur. Fortunately, in examining the scales administered to the samples included in this study, it appears that sleep problems and loneliness can be assessed using relatively brief scales (e.g., single-item measures). We also emphasize that there exist numerous evidence-based treatments designed to therapeutically affect sleep problems and loneliness, such as interventions based on cognitive behavioral therapy (e.g., for insomnia, Edinger & Means, 2005; for social anxiety disorder, Rapee & Heimberg, 1997) and interpersonal psychotherapy (Klerman & Weissman, 1994). Thus, after the identification of a sleep problem or loneliness, clinicians have the opportunity to offer a range of treatment options.
Considering the especially robust relationship observed between insomnia and loneliness compared with other types of sleep complaints and loneliness, insomnia may be a critical sleep complaint both to assess and to treat. We emphasize, however, that research is needed to understand why certain sleep complaints may evince stronger associations with loneliness than other complaints. Possible issues with publication bias for our insomnia analyses underscore the necessity for further research in this domain. Findings from analyses of sleep efficiency specifically should also be interpreted with caution because of the variability in assessment format (i.e., objective and subjective indices) and the relatively few studies examining sleep efficiency. We recommend that future studies evaluate distress associated with poor sleep efficiency and use multimodal assessments of sleep efficiency. Such research would enhance confidence in our understanding of the relationship between sleep efficiency and loneliness. Finally, in terms of the clinical relevance of our findings, we note that further efforts are needed to understand how the effect sizes identified in this study translate clinically considering that the effect-size cutoffs established by Cohen (1988) do not speak to how consequential certain effect sizes may ultimately be (Funder & Ozer, 2019).
Recommendations for future research
In evaluating the extant literature on sleep problems and loneliness, we also identified a number of key areas for future research in this domain. For one, although samples included in this meta-analysis varied with respect to sociodemographic characteristics, certain groups remained underrepresented. Relatively few studies were conducted with children and adolescents (12.9%), and a substantial portion of studies (23.0%) did not report the mean age of their sample, which limits our ability to evaluate age as a moderator. On average, of the studies reporting participant race, 73.0% of participants identified as White. Furthermore, of the samples included in our meta-analysis, only 31.0% reported the prevalence of a non-White racial group, and only 5.3% reported whether participants identified their ethnicity as Hispanic/Latinx (for details, see Table S1 in the Supplemental Material). Thus, we encourage increased evaluation of sleep problems and loneliness among individuals across the life span and of varying racial and ethnic backgrounds. Routine assessment of other characteristics that may affect sleep and loneliness (e.g., sexual orientation; Chen & Shiu, 2017; Mereish & Poteat, 2015) across studies will also facilitate meta-analytic investigation of their potential role in this association. In terms of geographic diversity, included samples were largely from studies that had been conducted in North America (42.7%), Europe (29.1%), and Asia (23.6%), and all North American studies were conducted in the United States exclusively. Consequently, further research on sleep and loneliness globally is needed.
Regarding study design, only 7.3% of samples reported longitudinal findings, and as alluded to above, few studies have evaluated how sleep and loneliness interact over both acute and multiyear time frames. Studies leveraging daily assessments (e.g., Kurina et al., 2011) as well as long-term epidemiological studies will be informative in this regard. On the topic of study design, although our pattern of findings remained the same regardless of whether we included only samples using validated measures of sleep and loneliness, we recommend that future investigations seek to exclusively use psychometrically sound measures of these constructs. In so doing, confidence in the interpretation of meta-analytic findings can be enhanced.
Evaluation of specific types of sleep complaints will also improve the understanding of whether each is differentially associated with loneliness. Only 56% of the samples that we identified used a specific index of sleep problems, and of these samples, the majority assessed insomnia or poor sleep quality. Thus, we did not have a sufficient number of samples to evaluate moderators for all models evaluating the association between specific sleep complaints and loneliness. Because different moderation patterns emerged depending on the specific sleep index examined, it is important that future studies more precisely assess the nature of sleep problems experienced by individuals. Such information might also provide insight into the mechanisms that explain the sleep–loneliness link. For instance, we found that the association between insomnia and loneliness (r = .354) was significantly stronger than the association between other types of sleep complaints and loneliness (rs = .165–.285). We are hesitant to draw conclusions about what might account for these differences, particularly because few studies assessed nightmares and sleep efficiency and because of the potential influence of publication bias in studies examining insomnia. However, this pattern of findings aligns with research suggesting that insomnia is especially robustly related to other mental health problems (Buysse et al., 2008; Pigeon & Gallegos, 2015; Taylor et al., 2005).
Furthermore, although overlap exists between insomnia and nightmares (Ohayon, Morselli, & Guilleminault, 1997), research indicates that insomnia may be more strongly related to psychiatric problems than nightmares (Pigeon et al., 2013; Zayfert & DeViva, 2004). Thus, it is unsurprising that in this study, insomnia was more strongly related to loneliness than nightmares were. As discussed, the relationship between other sleep indices (i.e., sleep quality and sleep efficiency) and mental health problems is less well understood (e.g., for review, see Alvaro et al., 2013). Yet it is worth noting that in this study, sleep quality evinced a stronger relationship with loneliness (r = .285) than did nightmares (r = .193) or poor sleep efficiency (r = .165). This finding may be driven by overlap between sleep quality and insomnia. That is, assessments of sleep quality often take into account insomnia symptoms (e.g., difficulties initiating sleep), and nonrestorative sleep is a unique component of insomnia (Roth et al., 2010). In addition, sleep efficiency is a more objective index of sleep disturbance and may not necessarily be perceived as markedly distressing, which potentially explains why it may be less likely to contribute to perceptions of loneliness than insomnia or sleep quality might be.
Ultimately, it is clear that additional research is needed to facilitate an understanding of what might underlie specific associations. Studies would benefit from the assessment of specific sleep problems that have been understudied by comparison (e.g., nightmares, sleep-disordered breathing, hypersomnia). Furthermore, because research suggests that mental health problems are differentially associated with onset, maintenance, and terminal insomnia (e.g., Taylor et al., 2005), an evaluation of specific types of insomnia would be useful. Only one study identified by our search examined the association between loneliness and various types of insomnia. Finally, routine assessment of other possible moderators not evaluated here (e.g., emotion regulation skills, overall physical health, anxiety and depression symptom severity) may shed light on the degree to which this association remains robust in the presence of other mental health and physical health problems.
Limitations of the current meta-analysis
This meta-analysis was also limited in several ways. First, with regard to search strategy, we included only findings from peer-reviewed research. Although this study selection approach facilitated the inclusion of studies that had been, to a degree, vetted for methodological rigor, the gray literature was not included in our analyses, which potentially biased findings. Second, only articles published in English were included in our search. In addition, as noted, although study locations spanned various countries, some regions (e.g., Africa, South America) were underrepresented. Thus, findings may not be generalizable to non-English speakers and regions not well represented in this meta-analysis. Third, we were underpowered to examine moderators for a subset of our meta-analytic models, especially those examining the association (a) between nightmares and loneliness and (b) between poor sleep efficiency and loneliness. Further work is needed to expand our understanding of these associations. Fourth, we were limited in the moderators that we were able to examine among our samples. Research is needed to meta-analytically investigate moderators not evaluated here. Fifth, although we made multiple efforts to contact authors to obtain effect-size information not reported within articles, we were able to reach just a minority (i.e., 19/150). Finally, although our publication bias tests did not identify significant issues with publication bias across models, it is still possible that such bias influenced our pattern of results.
Conclusions
This meta-analysis sought to synthesize existing data on the relationship between sleep problems—defined both broadly and specifically—and loneliness. Our cross-sectional analysis of 108 unique samples revealed a medium association between overall sleep problems and loneliness. We also found small-to-medium and medium associations between specific types of sleep complaints (i.e., insomnia, nightmares, sleep poor efficiency, and poor sleep quality) and loneliness. Although relatively few samples were available for longitudinal meta-analytic investigation, findings from these analyses suggested that overall sleep problems and loneliness are also bidirectionally associated. Taken together, our results point to the existence of a robust relationship between sleep problems and loneliness. Moving forward, research is needed to identify mechanisms that underlie this relationship and to delineate how sleep problems and loneliness influence one another over time.
Supplemental Material
Hom_Supplemental_Material – Supplemental material for A Meta-Analysis of the Relationship Between Sleep Problems and Loneliness
Supplemental material, Hom_Supplemental_Material for A Meta-Analysis of the Relationship Between Sleep Problems and Loneliness by Melanie A. Hom, Carol Chu, Megan L. Rogers and Thomas E. Joiner in Clinical Psychological Science
Footnotes
Acknowledgements
Opinions, interpretations, conclusions, and recommendations are those of the authors and are not necessarily endorsed by the Military Suicide Research Consortium or the Department of Defense.
Transparency
Action Editor: Kelly L. Klump
Editor: Scott O. Lilienfeld
Author Contributions
M. A. Hom, C. Chu, and M. L. Rogers developed the study concept and design. Data collection and extraction were performed by M. A. Hom, C. Chu, and M. L. Rogers. M. A. Hom and C. Chu performed the data analysis and interpretation under the supervision of T. E. Joiner. M. A. Hom drafted the manuscript, and C. Chu, M. L. Rogers, and T. E. Joiner provided critical revisions. All of the authors approved the final version of the manuscript for submission.
Notes
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
