Multilevel analysis of sleep arrangements and psychological well-being among older couples in Taiwan

Abstract

Background

This study investigated the simultaneous associations between sleep arrangements (e.g., separate rooms for sleep) and psychological well-being (PWB) among older Taiwanese couples when considering individual and couple characteristics.

Methods

The study sample comprised 860 heterosexual married older couples (1,720 individuals) residing in Northern Taiwan. PWB was operationalized as a latent variable by three indicators: happiness, life satisfaction, and fulfillment. Sleep measures were assessed at individual and couple levels; a multi-level structural equation modeling (SEM) was employed, with couples at the level 2 unit and individuals at the level 1 unit.

Results

The SEM results showed that older couples who slept in separate rooms experienced lower levels of PWB (β= − 0.12, p < .05) compared to those who slept together. In addition, a sensitivity analysis was also conducted, which controlled for previous relationship quality (e.g., conflict tactics), and the significant association between sleep in separate rooms and lower levels of PWB persisted (β= − 0.09, p < .01).

Conclusions

Sleep arrangement is a significant factor in a couple’s PWB. The present investigation also underscored the importance of considering sleep within the context of a couple’s relationship. Overall, the findings emphasize the need to consider the sleep arrangements of older couples when assessing their PWB.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-025-24856-w.

Public Significance

• Rather than living arrangements, sleep arrangements in older couples appear to be an important factor in psychological well-being (PWB).

• Older couples who slept in separate rooms experienced poorer PWB compared to those who slept together.

• Spousal conflict tactics in mid-life did not influence the association between sleep problems and the PWB of couples in later life.

Introduction

Sleep is one of the major public health issues, particularly in older people. A recent review indicated that sleep problems, such as insomnia or insufficient sleep, are associated with a higher risk of cancer, cardiovascular disease, and depression [1]. Studies have also shown that sleep problems are linked to cognition, health problems, and psychological well-being (PWB) among older adults [2, 3]. Although research has underscored the importance of the association between sleep, physical health, and PWB, most studies have considered sleep as an “individual” behavior. Yet, little research has argued that sleep in society should be viewed as a “social” activity [4] or an activity embedded in a social context, particularly at the couple level [2, 5]. As highlighted by Troxel^5(p578), “considering sleep in the couple context would seem a worthwhile area of scientific inquiry.” The World Health Organization defines PWB as the positive and subjective domain of well-being beyond the absence of mental health illness [6]. Therefore, when discussing couples’ PWB, it is important to consider sleep behavior at both the individual and couple levels. ^7,8

Prior studies produced mixed findings about the impact of couples’ sleep behavior on their mental health and PWB. The conventional sleep arrangement among married couples is sharing a bed, representing a symbolic marital status for couples; bed sharing is thus considered “a time of social interaction”[9, 10]. Most studies on the place of sleep in married couples were based on Western societies. They focused on health practices [11], where very few consider such arrangements that may be potentially associated with couples’ relationship satisfaction, and even fewer in non-Western societies. Compared to Western culture, Asian societies are relatively conservative, and sleep arrangements are related to intimate relationships, which are very private.

One line of research suggests that sleep can be an attachment behavior, as sleep often occurs when individuals lower their vigilance and awareness and are physically and emotionally secure [5]. In addition to the security and connection provided by sharing a bed, a spouse has also been identified as a social “zeitgeber” that helps establish a consistent and consolidated sleep-wake cycle [12]. Older couples who sleep together may thus enjoy better sleep quality. Some studies showed that couples report sleeping better when sharing a bed [13, 14]. By using sleep diaries, men who sleep alone reported lower sleep quality than those who sleep together [14]. However, other studies suggest that bed-sharing may harm sleep hygiene rather than improve the quality due to different wake-up and bedtime routines in couples [15–17] and movements to interrupt nocturnal sleep [16, 18].

Among a handful of studies on sleep at the couple level, the discussion has mostly revolved around how sleep behavior influences each other’s behavior and mental health [2, 17, 19, 20]. For instance, Chen [2] conducted a multi-level model to analyze a sample of 718 couples in the US and showed that different bedtimes influenced psychological distress. The other line of research focused on how the couple’s relationship influences their sleep hygiene or vice versa [12, 21–23]. By analyzing 29 heterosexual American couples, Hasler and Troxel [23] found that women reported more positive interaction during the day, leading to diary-based sleep efficiency for their partners.

Prior research has often treated the couple as a background to understand sleep and its possible association with PWB. However, three limitations can be improved upon. First, most studies [2, 12, 19, 23] assumed that couples sleep together, ignoring an interesting facet of a couple’s sleep arrangement. Troxel [5] mentioned that couples sharing a bed provide a sense of security and emotional attachment, which benefits sleep health and PWB. The couple’s sleep arrangement (i.e., same bed, different bed in the same room, different room) may thus influence their sleep and even PWB, as many studies have already shown a strong link between sleep and health among older adults [24–26]. One exception is a recent study that found that couples who slept in separate rooms experienced higher levels of loneliness [27]. Some may argue that, as mentioned above, if couples have discordant sleep-wake cycles[2], sleeping separately may positively affect both sleep and health. In addition, studies have also indicated that less optimal sleep characteristics (e.g., low sleep quality and duration) within couples flow from several different factors, such as discordant sleep time and snoring,^2,5,28, which was potentially solved by not sharing a bed. Second, in our literature search, no studies have used multiple indicators to represent PWB. Thirdly, the sample size is often limited and mostly from Western societies. While the social norm of sharing a bed with a partner for married couples seems to be socially prevalent, the reason for bed-sharing appeared to be subtly different for non-Western couples. To be specific, the traditional view of sex among older couples is for passing on the lineage, not for fun. Extending this, couples in East Asian, such as Taiwan, may be forced to sleep together due to limited household space. Hence, sleep arrangements are hypothesized to be associated with different levels of PWB in older couples.

The present study built upon the social-ecological theory [29, 30] which focuses on the complex interplay between the nested family levels that envelop older couples and the individuals in their immediate family environment in terms of their influence on PWB [29]. The theory emphasizes that individuals are embedded within a couple of family contexts, and to an extent, their sleep practices are hypothesized to be related to various aspects of spousal influences; these originate from their couple interaction and include conflict, communication, and attitudes such as egalitarian gender attitudes. Accordingly, this study aimed to significantly bridge the knowledge gap in the literature in three ways. First, we used 860 heterosexual couples of older couples from a non-Western society, Taiwan, to hypothesize that couples’ sleep arrangement is related to their PWB. Second, we employed multi-level structural equation models to properly hypothesize various types of couples’ sleep arrangement at the couple level that are related to different levels of PWB on the positive domain; and, PWB is constructed as a latent variable with three indicators: happiness, life satisfaction, and fulfillment. Lastly, the present study focuses on sleep, innovatively hypothesizing it as a couple-level social activity that is related to couples’ interaction and communication.

Methods

Sample

This study utilized the parental survey of the main subjects of the Taiwan Youth Project (TYP), which was conducted in 2019 by the Institute of Sociology, Academia Sinica in Taiwan. The study invited one of the parents of the focal young adults who were in the 2017 TYP survey, as well as their spouse, to participate in the survey. Each couple completed the survey independently. The survey comprised two sections: regular questions (e.g., quality of marriage and demographic characteristics) and private questions (e.g., sexual behavior). After obtaining informed consent, each couple member was interviewed by designated survey administrators for the first part, and the second part was completed by the participant alone. Of the 877 spouses who completed the survey, the final sample included 1,720 older adults (860 couples) who were still living together in the same household, and all were from the northern part of Taiwan. The present study was approved by the Research Ethics Committee of [blinded] (IRB Number: YM109021E).

Measures

Sleep arrangements (couple level; survey 2019) were categorized based on the self-report question: “Do you currently sleep with your spouse/partner in the same room and on the same bed?” Five response categories were presented (e.g., “yes” or “in the same room but on a different bed”). We further cross-classified the responses of participants and their spouses. If the couple provided the same response, their answer was recoded accordingly. To be specific, if the focal participant indicated that they slept together and their spouse endorsed the same category, the couple’s sleep arrangement was classified as “sleeping together.” In contrast, cases in which one partner did not respond or the partners’ responses were inconsistent were categorized as “unclear.” In the final analyses, this variable was captured by three dummy variables (i.e., “sleep in the same room but on a different bed;” “sleep in a separated room;” and “unclear”) with the first category (i.e., sleep on the same bed) as the reference group (N = 515; 59.88%).

Psychological well-being (PWB; individual level; survey 2019) was treated as a latent variable in subsequent analyses, which was measured by three observable variables: feeling happy, overall life satisfaction, and fulfillment [6]. Each item was based on individual self-report and with a five-point scale ranging from very unhappy/unsatisfied/unfulfilled (coded as 1) to very happy/satisfied/fulfilled (coded as 5), a higher score represents a higher level of PWB. While only the item “feeling happy” explicitly referred to a specific time frame, namely recent life, the other two items did not specify a temporal reference. Nevertheless, their general wording was interpreted as reflecting a global self-evaluation of recent life within the societal context. These three items captured the three aspects of subjective well-being, respectively: hedonic well-being (e.g., the feeling of happiness), evaluative well-being (e.g., life satisfaction), and eudemonic well-being (e.g., a feeling of soundness for life) [31–33].

The PWB measures appeared to be limited in number; however, each item was carefully selected to reflect a distinct dimension of PWB. This tripartite framework aligns with established conceptualizations in the literature, allowing for a parsimonious yet theoretically grounded assessment of well-being. Moreover, the items carry cultural significance. For instance, eudemonic well-being was assessed using the term xingfu (幸福), which in the Chinese context conveys a deeper sense of life fulfillment and personal meaning, consistent with indigenous conceptions of happiness [34]. In contrast, hedonic well-being was measured using the term kuaile (快樂), which primarily captures emotional and momentary experiences of positive affect, consistent with the hedonic well-being perspective. Thus, despite the brevity of the scale, our operationalization of psychological well-being is both theoretically informed and culturally resonant.

Finally, we included several control variables at the individual and couple levels. All these were measured in the 2019 survey. At the couple level, we included demographic and couple characteristics such as family location, the number of people living together, the number of children, age and education differences, years of marriage, discordant bedtime, and wake-up time. Two dummy variables captured family location (3 sites). The number of people living together and the number of children were continuous variables based on the main participant’s report. Three groups were created for age and education differences with no differences as the reference group. The years of marriage was the total number of years married at the interview from the main participant’s report. Discordant Bedtime and wake-up time were both dummy variables, with “non-discordant” as the reference group.

At the individual level, we included demographic variables, individual and cultural characteristics, and couple interactions. For the demographic variables, we included gender, retirement status, age, and education level. For gender and retirement status both were dummy variables, with female and non-retired status as the reference group. Age (in years) and education level (in formal education years) were continuous variables. For individual and cultural characteristics, we included self-rated health, sleep quality, sleep latency, total time in bed, depressive symptoms, self-esteem, and gender attitude. Self-rated health was based on participants’ reports on their overall health with five response categories (e.g., health to very unhealthy), and a higher score indicated lower health. Sleep quality was assessed using a self-reported measure with four response categories ranging from “very good” to “very poor,” where higher scores indicated poorer sleep quality. Sleep latency was derived from a single self-reported item indicating the duration between lying down and falling asleep based on four categories (e.g., “less than 15 minutes”). Sleep duration was measured based on participants’ self-reporting of sleep time in hours. Depressive symptoms were assessed using the mean score of sixteen items derived from the SCL-90-R (Cronbach’s α = 0.84)[35]. Each item featured five response categories ranging from “no such symptom (0)” to “severe symptom (4).” Self-esteem was measured using the mean score of six items from Rosenberg’s original self-esteem scale (Cronbach’s α = 0.81)[36]. Responses were recorded on a four-point Likert scale, with higher scores indicating stronger agreement with the statements. For both measures, higher scores reflected a more negative status, indicating more significant levels of depressive symptoms and lower levels of self-esteem, respectively. Gender attitude, which was captured by six items (e.g., “When there is a life conflict, it is the wife who should give up her career”) that measured an individual’s attitude toward gender roles and equality (e.g., strongly agree to strongly disagree). Previous research has consistently demonstrated that gender role attitudes and the division of domestic labor significantly influence couples’ relationship satisfaction [37]– [38], and even the likelihood of relationship maintenance [39]– [40]. The mean score was used for subsequent analyses, with a high score indicating a more egalitarian gender attitude (α = 0.69). Finally, we included the total conflict score and communication for couple interaction. The former asked participants for their responses or reactions when arguing or in conflict with their spouses. There are five different reactions, from regular oral altercations to physical fights. Each of these acts was dichotomized, and endorsement of a particular act received a score. A summation of these five items was used. The latter was based on a single item that asked participants if they had intimate interactions with their spouses. One of the responses was “close talk.” Participants who endorsed this item received a score (see Table S1 for further description of the online supplement). As highlighted in previous studies, these two measures were employed to capture each individual’s perspective on their marriage, which represents an important nuance in understanding couples’ relationships [41]. – [42].

Analytic strategy

Analysis began with descriptive tabulations that characterized the distribution of individual and couple characteristics. Then, to address our research questions, we used multilevel modeling techniques to study the association between sleep arrangements, namely separate rooms for sleep, and psychological well-being (PWB) among older Taiwanese couples. Owing to individuals nested within couples, we thus conducted the multilevel analysis to study the simultaneous associations between sleep arrangements and PWB, adjusting for individual and couple backgrounds. Accordingly, we applied the random-intercept model to test whether significant variations in PWB occurred at both individual and couple levels.

The nested data structure, in which individuals were nested within couples, clearly suggests that individual observations are not independent, which violates the assumptions of conventional regression models and inflates standard errors. Furthermore, the study aimed to examine how couples’ sleep arrangements are associated with PWB. Consequently, a two-level random intercept model was employed for the analyses.

Furthermore, PWB was modeled as a latent variable; the analysis accounted for both the non-independence of observations and the measurement error inherent in the construct. Specifically, the three indicators of PWB were allowed to vary across couples (i.e., random intercepts), which were subsequently used to estimate the latent PWB at Level 2, representing the overall random intercept of the latent construct. Finally, the relationship between the couples’ sleep arrangement and this latent PWB was estimated.

A two-level random intercept model was employed for the present analyses. In these analyses, we took several steps. The analyses were conducted in two sequential steps. First, a null model was fitted to evaluate the measurement model’s fit and calculate the intra-class correlation (ICC). This step was critical in determining the extent of dependency among individuals within clusters (i.e., couples). Consistent with the guidelines proposed by others [43], an ICC value exceeding 5% indicated the necessity of a model capable of accounting for such dependency. Figure 1 provides a visual representation of this analytical approach. Second, we included individual- and couple-level variables to estimate the association between the hypothesized correlation between outcome and primary exposure. In this model, we could see if sleep arrangement is significantly related to individual PWB and all the included variables, particularly the couple’s interaction. Finally, we exploratorily conducted a random slope analysis to examine whether couples’ individual sleep characteristics (e.g., sleep quality) varied as a function of their sleep arrangement. Additionally, we performed sensitivity analyses to evaluate the robustness of our proposed model using alternative coding schemes for couples’ sleep arrangement (e.g., imputing missing values based on one spouse’s response). Descriptive statistics were computed using Stata 14, and multilevel analyses were conducted in Mplus 8.10 (see Table S2 in the online supplement for the Mplus code) [44]. The rate of missing data in the present study was low (ranging from 1% to 5%), and full information maximum likelihood (FIML) was employed to account for missingness [45].

Fig. 1.

Illustration of estimated model^**. ^**The filled circle denotes the random intercept for each item measuring PWB, which is subsequently utilized to represent the latent PWB at the couple level as a random effect

Results

Table 1 presents the descriptive statistics. PWB for the subjects was quite positive, with a mean of 3.83 on a 5-point scale. Among these older couples, approximately 16.74% reported sleeping in a separate room. Interestingly, the majority of the couples had discordant bedtimes (71.16%) and wake-up times (76.86%), yet most of them reported good sleep quality, with a mean of 1.2 on a 4-point scale. The mean age was 61.53, which is considered young-old by others [46]. – [47].

Table 1.

Descriptive statistics for all variables

	%	Mean (SD) [range]	Skewness (Kurtosis)
Level 1 variables
Outcome variable
Psychological well-being1		3.83(0.61) [1,5]	− 0.32 (0.43)
Demographic variables
Age		61.53 (4.26) [50,91]		1.28 (−0.36)
Education level (in years)		11.08 (3.30) [6,18]		0.09 (−0.68)
Retirement status	36.05
Individual variables
Low self-rate health		3.02 (0.96) [1,5]		− 0.03 (−0.49)
Low sleep quality		1.20 (0.68) [0,3]		0.37 (0.26)
Sleep latency		0.91 (0.91) [0,3]		0.75 (−0.29)
Sleep duration		6.45 (1.20) [2,11]		− 0.40 (0.45)
Depressive symptoms		1.25 (0.33) [1, 3.69]		2.33 (7.72)
Low self-esteem		3.67 (1.96) [1,3.67]		− 0.39 (9.38)
Cultural Influence
Traditional gender attitude		2.75 (0.35) [1.17, 4.00]		0.14 (1.17)
Couple interaction
Total conflict score		0.67 (0.69) [0,5]		0.86 (1.14)
Communication	27.03
Level 2 variables
Demographic variables
Taipei City	35.12
Taipei County	31.62
Yilan County	33.26
Do not live with sons and daughters	25.12
Live with unmarried sons/daughters	51.05
Live with married sons	12.79
Live with married daughters and other situation	11.04
Couple variables
Bedtime is different	71.16
Wake-up time is different	76.86
Sleep in the same room and same bed	59.88
Sleep in the same room but not the same bed	8.61
Sleep in a separate room	16.74
Not clear	14.77

Level 1 (individual): n = 1,720; Level 2 (couple): n = 860

¹Psychological well-being is a latent variable measured by happiness, fulfillment, and overall life satisfaction. Here is the mean of these three items

Before presenting the results of the multilevel analysis, we first conducted a multilevel confirmatory factor analysis (MCFA) to evaluate the measurement model (see Table S3 in the online supplement). The model demonstrated an acceptable fit to the data based on commonly used indices, including the chi-square test and the Root Mean Square Error of Approximation (RMSEA), and all factor loadings were statistically significant at conventional levels. Following the MCFA, we assessed the fit of the unconditional (null) model, which also showed an acceptable fit (χ²(3) = 10.14, non-significance; CFI = 0.995; TLI = 0.991; RMSEA = 0.037). The Intra-class correlation (ICC) for each observable outcome used for the latent variable was 0.24 for “feeling happy,” 0.26 for “life satisfaction,” and 0.30 for “fulfillment.” All of these ICCs indicated the need to account for non-independence among the observations [48].

Model 1 incorporated sleep arrangement variables alongside all covariates at both the individual and couple levels. The findings reveal that compared to sleeping in the same bed, sleeping in a separate room had a significant lower level of PWB (β = − 0.12, p <.01). Similarly, sleeping in a separate bed was associated with a lower level of PWB (β = − 0.17, p <.05). At the individual level, several covariates were also significantly related; for example, older couples with lower sleep quality reported significantly lower levels of PWB (β = − 0.13, p <.05). Interestingly, none of the other sleep-related variables at the couple level reached statistical significance.

In Model 2, we incorporated older couples’ personal reflections on their interactions with their spouses. The results indicated that couples who reported higher levels of conflict with their spouses exhibited lower PWB (β = − 0.05, p <.05). Conversely, those who engaged in intimate communication (i.e., close talk) with their partners reported higher levels of PWB (β = 0.08, p <.05). Notably, the associations between sleep arrangement and PWB remained significant even after accounting for these critical individual-level covariates. Specifically, couples who slept in separate rooms continued to report lower levels of PWB (β = − 0.12, p <.05) compared to those who slept in the same bed. Moreover, we conducted additional analyses to examine the potential moderating effect at the couple level, specifically whether the association between differing bedtimes and wake-up times and PWB varied depending on sleep arrangements. The results, however, did not reach conventional levels of statistical significance.

Table 2.

Results from the multi-level model: couples’ sleep arrangement and psychological wellbeing [1–4]

	Model 1	Model 2
	β (SE)	β (SE)
Level 1 variables
Individual variables
Male	− 0.01(0.03)	− 0.02(0.03)
Age	0.00(0.00)	0.00(0.00)
Education level	− 0.01(0.00)	− 0.01(0.00)
Retired	− 0.01(0.03)	− 0.01(0.03)
Self-rate health	0.02(0.01)	0.02(0.01)
Low sleep quality	− 0.13(0.03)**	− 0.13(0.03)**
Sleep latency	− 0.01(0.02)	− 0.01(0.02)
Sleep duration	0.02(0.02)	0.02(0.02)
High depressive symptoms	− 0.47(0.05)**	− 0.45(0.05)**
Low self-esteem	− 0.38(0.04)**	− 0.37(0.04)**
Egalitarian gender attitude	0.02(0.04)	0.02(0.04)
Couple interaction
Total conflict score		− 0.05(0.02)*
Communication		0.08(0.03)*
Level 2 variables
Sleep in the same room but not the same beda	− 0.17(0.08)*	− 0.14(0.09)
Sleep in a separate room	− 0.12(0.04)**	− 0.12(0.03)**
Not clear	− 0.08(0.04)	− 0.07(0.04)
Discordant bedtime	− 0.01(0.03)	− 0.01(0.03)
Discordant wake-up time	− 0.01(0.03)	− 0.00(0.03)

¹ Psychological well-being is a latent variable measured by happiness, fulfillment, and life satisfaction

²All models were estimated with robust standard error

³All models were estimated with level-2 covariates mentioned in the context: family location, number of people in the household, number of children, education, and age differences between the couple

⁴ Several level-1 covariates were grand-mean centered: sleep quality, sleep latency, total time sleep time, self-rated health, depressive symptoms, self-esteem, gender attitude, total conflict score, and communication

^aReference group: Sleep in the same room and same bed

Level 1: n = 1,720

Level 2: n = 860

^* p <.05

^**p <.01

In addition to the primary analysis, we conducted several random slope analyses to explore potential cross-level interactions, suggesting the individual interactions within couple contexts. Prior research indicates that individuals with good sleep quality often report higher levels of PWB [49–51]. Building on this premise, we examined whether the associations between three individual-level sleep-related measures—sleep quality, sleep latency, and sleep duration—and PWB varied depending on sleep arrangements at the couple level.

The results revealed that the negative association between low sleep quality and PWB was exacerbated when individuals slept in separate rooms (β = − 0.16, p <.01). Conversely, the negative association between sleep latency (i.e., the time taken to fall asleep) and PWB became weaker when couples did not report their sleep arrangement (β = 0.09, p <.05) (see Table S4 of the online supplement). However, the association between sleep duration and PWB did not vary by different types of sleep arrangements.

While our study utilized cross-sectional data, we conducted additional analyses to support our primary findings by addressing a key counterargument: couples with unsatisfactory relationships may prefer to sleep apart, which, in turn, could negatively influence their PWB. This aligns with prior research indicating that older couples often sleep separately due to poor marital relationships[7, 20, 23], and that low marital quality typically has a negative impact on subjective well-being [52] and general well-being [51].

In these supplementary analyses, we incorporated data from the final wave of the first phase of TYP’s parental surveys, conducted approximately 10 years prior to the current survey. This earlier survey included responses from one older couple and captured both subjective and objective evaluations of marital quality. Specifically, one question asked, “Compared to other couples, would you say your relationship with your spouse is…?” with response options ranging from “a lot better (1)” to “a lot worse (5).” Additionally, conflict tactics were assessed through three dichotomous items (yes/no): “yelling at you with very negative words,” “throwing things or getting into physical fights,” and “making others nervous.” A summed score of these items was used to measure conflict intensity.

These variables were included as second-level covariates in the sensitivity analysis. The results were consistent with our primary findings (see Table A1 in Appendix 1). Specifically, older couples who slept in separate rooms or in the same room but in different beds reported lower levels of PWB compared to those who shared the same bed. As anticipated, couples who perceived their relationship quality as inferior to that of other couples exhibited significantly lower levels of PWB (β = − 0.13, p <.01).

Finally, as previously noted, our measure of sleep arrangement may be subject to arbitrariness. To assess the robustness of our findings, we conducted sensitivity analyses using alternative coding strategies. Specifically, in cases of discrepancy between the main participant’s and the spouse’s reports, we substituted either the main participant’s or the spouse’s response as the final classification. The results of these additional analyses are presented in Table S5 (online supplement). As shown, the direction and magnitude of the associations remained largely consistent. For instance, sleeping in a separate room was significantly associated with lower levels of PWB (β = −0.12, p <.01, Model 1; β = −0.11, p <.01, Model 2).

Discussion

Sleep is a fundamental need and a barometer of health and PWB [53–59]. Research has begun to address the importance of the mutual influence of couples’ sleep behavior on health[2, 17, 54, 60], and some argue that dyadic research provides a rich opportunity to observe the association between couples’ sleep and each other’s health and well-being [61]. We thus employed a multi-level model to examine how elder couples’ sleep arrangement was associated with individuals’ PWB and found that couples who sleep in separate rooms had worse PWB than couples who sleep together. This result remains significant even when we control for individual sleep quality and other important individual (e.g., age and self-esteem) and couple’s characteristics (e.g., different bedtime and wake-up time).

Our results may provide further insight into the two very different perspectives on couples’ sleep. On the one hand, previous studies have found that sleeping together can actually disturb each other’s sleep (e.g., resulting in more sleep movement). Hence, studies that measured sleep in an objective way (e.g., actigraphy) often support the idea that sleeping together may lead to less optimal sleep quality [16, 62, 63]. On the other hand, most individuals report sleeping better when they sleep with their spouses [64]. Our results indirectly support the second perspective, indicating that sleeping together seems to have a positive association with PWB. This is because our study found that sleeping in separate rooms was related to a lower level of PWB, as well as Chiao et al.‘s[27] results that sleeping separately was associated with older couples’ mental health, including high levels of social and emotional loneliness. Prior studies have often focused on the effects of sleep behavior, such as duration, bedtime, or movement, on sleep-related outcomes like insomnia or sleep quality[2], rather than exploring the impact of more structural sleep arrangements on mental health. As such, our finding, although novel, may be considered an important sleep issue for older couples.

Our results reveal that the better PWB of older couples who sleep together can be understood in two ways. First, sleep has been considered a social behavior, and research has found that Chinese older adults sleep better when staying with family members than alone [65, 66]. Given that sleep quality is highly related to health, it is expected that sleeping alone may be related to poor sleep quality and PWB. Secondly, in contemporary society, the social norm often expects couples to share a bed together because the bed is associated with intimacy [64, 67]. Hislop and Arber [68] argue that sleeping in a different location, temporarily or permanently, is a last resort to solve disturbances arising from a bed partner. Hence, sleeping in a separate location or even in different beds may run counter to this social norm, the “culture of togetherness“(p290) [65], which may be related to low PWB. Moreover, behavior that counters social expectations is often sanctioned by others or even oneself, which may negatively affect mental and psychological health [69]. Furthermore, this “difference” may be considered a social stress, and the negative effect of stress on health, in general, and psychological/mental health, in particular, has been found in many previous studies [70].

Biologically, sleep represents a vulnerable state for all organisms. Consequently, a safe and comfortable environment is beneficial to sleep quality. Some researchers have argued that attachment style may influence sleep problems, such as insomnia [71]. Extending this to the current situation, being physically present (i.e., sleeping together) should provide an attachment-like environment that increases the feeling of security and promotes sleep. As Troxel [5] mentioned, sleep is an attachment behavior. Furthermore, Williams (p91) [72] mentioned from a qualitative interview that sleeping together brought about “companionship, intimacy, mutual trust.” Hence, when Hislop [64] found that couples who slept apart temporarily while having restful sleep hesitated to make it a regular practice, it showed that emotional bonding between couples by sleeping together is far more important than actual good night’s sleep. This point was also supported by others who argued from an evolutionary perspective that sleeping together may reduce vigilance and arousal levels, which is beneficial to sleep quality and quantity [28]. As such, older couples can develop togetherness and companionship when sleeping together, which provides a basis for security, promoting good sleep[8], and better PWB.

While the above discussion presents possible reasons for why older couples may choose to sleep together, we further strengthened our results by considering their relational quality, although it was not ideally measured. Previous studies have shown that couples’ interaction, such as conflict or intimacy, can influence their sleep and mental health. Therefore, including such a measure in our analysis could further validate our results. Our sensitivity analysis showed very similar results, indicating that even when considering a possible and strong confounding variable, such as couple’s relationship quality, we still found that sleeping separately was negatively related to the PWB of older couples. While we do not claim that our measure accurately reflects our subjects’ current situation, it may still capture the essence of their interaction due to established interaction patterns for long-married couples (i.e., the non-significant difference between formal and current conflict tactics) (results not shown).

In addition to the aforementioned contribution, we also highlight another noteworthy finding from our multilevel confirmatory factor analysis. Specifically, we found that the factor loading of happiness was only moderate at the individual level but became substantially stronger at the couple level. We speculate that processes such as emotional dependence and emotion regulation within close relationships [73] may be at play. That is, individual emotional states—though transient and personal—may become consolidated or mutually amplified within intimate relationships. This interpretation aligns with the concept of emotional convergence among couples [74]. Given that our sample consisted of older couples, such contextual effects may be particularly pronounced, thereby strengthening the loading of happiness at the couple level. This result further supports our initial proposition that dyadic interactions are central to understanding psychological well-being.

Although this study provides insights into the relationship between older couples’ sleep and PWB, some limitations still need to be addressed. First, our results were associational based on cross-sectional data and the self-reported recall of sleep problems and PWB, raising the recall bias issue. Relatedly, several control variables in our study were measured using different time frames (e.g., “past one year” vs. “past one month”), which may introduce some measurement inconsistencies. However, this variation primarily reflects the design of the original scales, which applied time frames considered suitable for the specific behaviors or phenomena being assessed. Second, and relatedly, we did not consider the possible bidirectional relationship between PWB and sleep behavior. The negative association between sleeping in separate rooms and poor PWB is likely due to a reverse causality from those having poor PWB being more likely to sleep separately, and the potential benefits of separate rooms for a couple’s sleep. One study has shown that an individual’s mental health is related to sleep quality, which, in turn, exacerbates mental health [21]. However, our focus here is on a different facet of couples’ sleep behavior. Consequently, whether the bidirectional relationship between sleep arrangement and PWB remains a future investigation. Fourth, while the decision-making process regarding sleep arrangements among couples is related to actual sleep arrangements and PWB in general and life satisfaction in specific, such information was not available to the present investigation. Life satisfaction of couples may include the couple’s sex life. Our preliminary analysis included the measure of having had sex; however, it did not yield a significant association with PWB. Due to the model parsimony, we did not include it in our final model. Finally, we did not have access to their medical records; some important comorbidities, such as sleep apnea, cannot be included.

This study has several strengths. First, we employed a multilevel SEM model to accommodate the interdependence among individuals. This also responds to Troxel’s[5] and others [51, 75] who recognized that sleep is not an individual behavior but should be considered a couple’s behavior. In addition, the latent variable approach reduced the measurement error that may have been involved in the study. Second, we empirically examined the sleep arrangement among older couples and its effect on PWB. Finally, the number of couples in our sample, while moderate, was larger than in many previous studies[2, 12, 19], which corresponds to the appeal of Spiegelhalder et al.[76]. Given the global aging, understanding older couples’ well-being is important. Our study added to the current literature by showing that sleep arrangement matters. Older couples who slept on the same bed reported lower levels of negative PWB.

Appendix

Table 3.

Table A1 Results from the multilevel model: Couples’ sleep arrangement and psychological well-being with previous couple relationship^1,2

	Model 1	Model 2
	β (SE)	β (SE)
Level 1 variables
Individual characteristics
Self-rate health	-.03 (.01)	-.03 (.01)
Low sleep quality	.12 (.03)**	.13 (.03)**
Sleep latency	.01 (.02)	.01 (.02)
Total time in bed	-.02 (.02)	-.02 (.02)
High depressive symptoms	.45 (.05)**	.44 (.05)**
Low self-esteem	.37 (.04)**	.37 (.04)**
Egalitarian gender attitude	-.01 (.04)	.01 (.04)
Couple interaction
Total conflict score	.03 (.02)	.02 (.02)
Communication	-.07 (.03)*	-.07 (.03)*
Level 2 variables
Sleep in the same room but not the same beda	.15 (.05)**	.15 (.05)**
Sleep in a separate room	.09 (.03)**	.09 (.03)**
Not clear	.04 (.04)	.03 (.04)
Discordant bedtime	-.004 (.02)	-.01 (.03)
Discordant wake-up time	.01 (.03)	.01 (.03)
Couples’ previous interaction
Relationship comparison	.13 (.02)**	.13 (.02)**
Conflict tactics (Ref. No)
1 conflict tactics		.05 (.05)
2 or more conflict tactics		.04 (.07)

¹Psychological well-being is a latent variable measured by happiness, fulfillment, and life satisfaction

²All models were estimated with all individual demographic variables at level1 (i.e., gender, age, retirement status, and year of education) and level 2 (i.e., family location, age difference, and education difference, number of people in the household, number of children, and year of marriage)

^aReference group: Sleep in the same room and same bed

Level 1: n = 1,638

Level 2: n = 819

^*p <.05

^**p <.01

Authors’ contributions

WHL was responsible for development of the study hypotheses, data analysis, and drafting of the article. CC was responsible for conceptualizing and designing this study, article drafting, and critical revision and finalizing the article. She also supervised the structure of the manuscript.Both authors were involved in the writing of the paper, reviewed the manuscript, and approved the final submission.

Funding

This study was supported by the Ministry of Science and Technology (MOST) in Taiwan under grant 109-2410-H-010-006-MY2.

Data availability

The Taiwan Youth Project (TYP) data are publicly available and can be used for research with the approval of Academia Sinica in Taiwan (https://www2.ios.sinica.edu.tw/typ/). Our conclusions are based on the de-identified data and the link to access this information is approved by Academia Sinica in Taiwan. There were no preregistrations in the present study; we used Stata 14 to conduct descriptive analysis and Mplus 8.10 to analyze the multilevel modelling.

Declarations

Ethic approval and consent to participate

Taiwan Youth Project (TYP) data are publicly available and can be used for research with the approval of Academia Sinica in Taiwan (https://srda.sinica.edu.tw/datasearch_result.php). All TYP participants gave informed written consent at the start of their interviews. All procedures performed in this study followed the ethical standards of the Institutional Review Board, National Yang Ming Chiao Tung University, by Human Study Approval, IRB number: YM109021E.

Consent for publication

This manuscript has not been published or presented elsewhere in part or in entirety and is not under consideration by another journal.

Competing interests

The authors declare no competing interests.