Do consistent and distinct sleep health phenotypes emerge among working adults over approximately ten years and, if so, what is the number and nature of those phenotypes?

What is the prevalence of “stayers” (i.e., those who remain in the same sleep phenotype over time) versus “movers” (i.e., those who transition to different sleep phenotypes over time) for each sleep health phenotype?

An increase in computer use frequency at work, time 1 to time 2 (7–13 years later) is associated with maintenance of suboptimal sleep health phenotypes over time or transitions from healthier to suboptimal phenotypes over time.

A second relevant element of job design is sedentary work, or extent of stationary time expending little physical effort at one’s job. Sedentary work has markedly increased in prevalence over the past several decades (Church et al., 2011), and this is often attributed to technological advancements that reduce physical labor needs (Peng et al., 2018). Scholars argue that because sedentary behavior associates with poor mental and physical health, it may interfere with healthy sleep (Yang et al., 2017). Frequent sedentary time has been associated with both too short (Lakerveld et al., 2016; Štefan et al., 2019) and too long sleep duration (Štefan et al., 2019), each of which indicate unhealthy sleep. Further, sedentary time relates to subjective difficulty falling asleep (Vancampfort et al., 2018) and physiological measures of sleep inefficiency (Madden et al., 2014). Meta-analysis also associates sedentary behavior with increased insomnia risk and sleep disturbance symptoms (Yang et al., 2017). Thus, a sedentary job may be a negative job design factor that, as it increases over time, relates to development of more suboptimal sleep health phenotypes, given its known links to individual sleep health problems.

An increase in sedentary activity frequency work, time 1 to time 2 (7–13 years later) is associated with maintenance of suboptimal sleep health phenotypes over time or transitions from healthier to suboptimal phenotypes over time.

A final relevant feature of temporal job design is work schedule type. Extensive previous research demonstrates that atypical work hours and work shifts extend sleep onset latency, increase sleep disturbances (i.e., wake after sleep onset or WASO; Chang & Peng, 2021) and daytime sleepiness (Booker et al., 2018), and reduce sleep quality and quantity (Booker et al., 2018). In fact, the prevalence of sleep issues among shift workers lead to the development of a shift work sleep disorder clinical diagnosis (American Sleep Disorders Association, 1990) involving severely disrupted sleep because of one’s work schedule. These detrimental effects can be explained by the misalignment of these work schedules with natural circadian rhythms and sleep/wake cycles (Costa, 1996). These negative sleep implications likely also extend to those who are not formally involved in shiftwork but engage in other overtime or flextime practices that functionally lead to nonstandard work schedules. Using information-communication technology to work after hours is meta-analytically connected to greater strain, including poorer sleep quality (Kühner et al., 2023). However, the specific consequences of these temporal job design factors for multidimensional sleep health are not known. We hypothesize that over-time changes to nonstandard work schedules will predict unfavorable transitions in multidimensional sleep health.

Maintenance of or transitions to nonstandard schedules from time 1 to time 2 (7–13 years later) is associated with maintenance of suboptimal sleep health phenotypes over time or transitions from healthier to suboptimal phenotypes over time.

To achieve the large sample size required for our person-centered analyses (Bray & Dziak, 2018) and access the over-time data required by our research questions, we used existing data from the Midlife in the United States study (MIDUS). MIDUS is an interdisciplinary, longitudinal data collection effort aimed at understanding health over the lifespan in a national sample. Please note that this study was not preregistered and data and study materials specific to the present study are not publicly available; however, multiple raw MIDUS datasets, including those used in the present analysis, are available online (https://www.midus.wisc.edu/data/index.php). Additional information on the MIDUS datasets and previous studies using the MIDUS datasets, including in total and by the current research team, can be found in Supplemental Table 1. We followed guidelines on the use of archival data in applied psychology research (Fisher & Chaffee, 2018) and ethical principles for repeated use of a dataset across multiple projects (Kirkman & Chen, 2011). Most notably, of the existing sleep research using MIDUS data, no other publications have used job design theory or examined longitudinal transitions in workers’ sleep health phenotypes, distinguishing our research (see https://www.midus.wisc.edu/findings/pubtopics.php?topic=Sleep).

Two waves of MIDUS data provide extensive work and sleep health information (MIDUS II and III) and were thus used in our analyses. Additionally, we used both core MIDUS data and Milwaukee MIDUS data, the latter of which aimed to add racial diversity to MIDUS datasets by targeting Black participants. MIDUS II (Core: 2004–2006; Milwaukee: 2005–2006) and III (Core: 2013–2014; Milwaukee: 2016–2017) are follow-ups to the original MIDUS I data collection (1995–1996). A national sample of adults 25–74 were recruited for MIDUS I via random digit dialing. The Milwaukee sample was selected by screening households in Milwaukee, Wisconsin in areas of the city that had higher concentrations of Black adults. Overall MIDUS data represent the U.S. adult population. We used data from the self-administered questionnaire (SAQ) portion of MIDUS, consisting of self-report surveys.

MIDUS II and III respondents who indicated that they were working 30 or more hours per week at both time points were included in our analytic sample to ensure that change in job design factors could be assessed (N=1,297). The analytic sample worked 45.61 hours per week on average, had an even sex distribution (53.97% male), and was primarily non-Latinx white (88.97%). Per Chou and Choi’s (2011) categories, approximately half (47%) of the sample worked in executive, managerial, and professional occupations, 35% worked in technical, clerical, service, and sales occupations, and 18% worked in crafts, labor, and military occupations. The sample reported slightly higher educational attainment, on average (M=7.11/12.00, indicating at least some college), than the national average. The majority of the sample was partnered (74.01% were married and/or cohabitating with a romantic partner) and had 2 children on average. Participants were 46.60 years old at MIDUS II (SD=7.47, Range=33 to 75) and 55.97 at MIDUS III (SD=7.36, Range=42 to 83).

Because participants needed to report job design and sleep experiences at both time points to be included in our analyses, there was no attrition between our time 1 analytic sample and time 2 analytic sample. However, to assess how our selection criteria might have impacted the characteristics of our analytic sample, we conducted comparison analyses. Compared to the full MIDUS core and Milwaukee combined sample (N = 3,676; M=55.03 at II and 63.36 at III), our working analytic sample (N = 1,297) was significantly younger (Cohen’s d=1.02 at II and .90 at III), likely due to our exclusion of retirees. There were also medium-sized significant differences in the number of children (d=.26) and self-rated physical health (d=.28), such that the analytic sample reported fewer children (M=2.12 compared to 2.54) and better physical health (M=2.67/5.00 compared to M=2.41/5.00). There were small significant differences in sex (d=.15) and race (d=.20), such that fewer women (46% compared to 53%) and racial minorities were in the analytic sample (11% compared to 19%). The analytic sample did not significantly differ from the full sample on marital status or education. Overall, the working subsample was younger, slightly healthier, slightly less diverse, and had fewer children than the full MIDUS core and Milwaukee sample.

MIDUS studies sometimes use adapted or abridged scales, but all measures are thoroughly vetted by subject matter experts before data collection and validated in subsequent studies after data collection. More information on MIDUS measurement can be found at their website https://www.midus.wisc.edu/data/index.php

Statisticians recommend using categorical indicators and latent class techniques (rather than continuous indicators and latent profile techniques) for mixture modeling when the indicator variables are skewed, as is the case with sleep variables (Wallace et al., 2018). Within the present data, most of the sleep health dimensions (i.e., sleep duration, regularity, sleep onset latency, and nap) exhibited non-normal distributions (i.e., skewness>|2| and/or kurtosis>|7|; West et al., 1995) at one or both time points. We therefore used empirical evidence to create cut-off scores differentiating optimal and suboptimal sleep for all sleep health variables to maintain consistency and interpretability across the dimensions (see Table 1 for a summary). These cut-offs align with previous values used in MIDUS research (Lee et al., 2022) but reorganize some sleep dimensions to provide a more detailed and accurate representation of sleep health (Buysse, 2014).

Based on holistic evaluation of model fit statistics, the three-class solution was chosen due to its low BIC and SSA-BIC; it also exhibited sufficient entropy (see Table 3). As shown in Table 4 and Figure 1, three similar classes were extracted at time 1 and time 2. The first class, good sleepers (~57% at time 1 and time 2) was characterized by relatively low proportions of unhealthy sleep across all characteristics. The second class, catch-up sleepers (~25% at both time points) was characterized by relatively healthy sleep except for high nap frequency and irregular sleep (i.e., long non-workday sleep compared to short workday sleep); as such, they were named catch-up because napping and non-workday sleep are considered ways to attempt to compensate or “catch up” on sleep (Leger et al., 2020). The third and final class, insomnia sleepers (~18% at both time points) was characterized by high insomnia symptoms, short duration, long sleep onset latency, high daytime tiredness, and somewhat irregular sleep, all of which map onto insomnia experiences. Membership to this class does not necessarily equate to clinical insomnia. In response to Research Question 1 (i.e., number and nature of sleep health phenotypes among working adults), three classes were identified: a highly optimal group, a highly suboptimal group, and a catch-up group.

In response to Research Question 2 (i.e., proportion of “stayers” vs. “movers” for each phenotype), Table 5 and Figure 2 shows the percentage of each phenotype at time 1 (good, catch-up, and insomnia sleepers) who transition to each phenotype at time 2. Participants were most likely to stay in the same class over time (81% for good sleepers, 58% for catch-up, and 90% for insomnia sleepers), though transitions also commonly occurred. Good sleepers and catch-up sleepers were more likely to transition to one another than to insomnia sleepers. Similarly, insomnia sleepers were less likely to transition to good (7%) or catch-up (3%).

Hypotheses 1–3 expected that an increase in computer use at work, sedentary work, or nontraditional work schedule, respectively would be associated with maintenance of suboptimal sleep health phenotypes or transitions from healthier to suboptimal sleep health phenotypes over time. Unexpectedly, an increase in computer use at work was unexpectedly associated with lower (rather than higher) likelihood of insomnia→insomnia transition (B=−1.26, SE=.53, p=.02) relative to good→good, contrary to H1. However, as expected, an increase in sitting at work was associated with higher likelihood of insomnia→insomnia (B=.38, SE=.15, p=.01) relative to good→good, supporting H2. Neither computer use nor sitting at work related to maintenance of or transitions into catch-up sleeper. Finally, largely as expected (H3), an increase in non-traditional work schedules from time 1 to time 2 was associated with transitions to catch-up sleep (good→catch-up: B=.66, SE=.29, p=.02; catch-up→catch-up: B=.481, SE=.23, p=.04; insomnia→catch-up; B=2.68, SE=1.16, p=.02) relative to consistently good sleep (good→good). Hypothesis 3 was partially supported in that nontraditional schedules related to maintenance of or transition to catch-up sleeper but not the insomnia sleeper. Full results are reported in Table 6.

Sleep is becoming a widely researched topic in the organizational sciences (see Crain et al., 2019), and, as a result, it is acknowledged as an essential ingredient to employee well-being and productivity (Barnes & Watson, 2019; Brossoit et al., 2019; Mullins et al., 2014; Schultz & Burton, 2018). Despite this growth, sleep is a relatively nascent topic in the organizational sciences, and our measurement has not kept pace with the sleep scholarship that recognizes the multidimensional (Buysse, 2014; Matricciani et al., 2018) and dynamic (Ohayon et al., 2017) nature of sleep health. We answer calls to assess sleep health as a phenotype (Chung et al., 2021; Matricciani et al., 2018) – or configuration of multiple dimensions of sleep health as they exist within a person – and the dynamic nature of these phenotypes over time. By examining these issues in working adults, we clarify how sleep health tends to manifest for this group that may be especially vulnerable to sleep difficulties (Swanson et al., 2011), describing common patterns of co-occurring sleep issues and, for the first time, how likely employees are to improve, maintain, or worsen in their holistic sleep health over a decade.

Findings from this study show that workers in the United States fit within three common sleep health phenotypes and that prevalence rates remain relatively consistent over time. At both time points, over half of working adults in our representative sample were good sleepers, exhibiting optimal sleep across all measured dimensions. One quarter were catch-up sleepers at each time point, exhibiting mostly optimal sleep (i.e., appropriate duration, low daytime tiredness, low insomnia symptoms, short sleep onset latency) but long weekend sleep and frequent daytime naps. At both time points, about 18% of our sample were insomnia sleepers, exhibiting suboptimal sleep consistent with insomnia symptoms (i.e., short duration, long sleep onset latency, high daytime tiredness, high insomnia symptoms). Working adults’ multidimensional sleep health seems to manifest in one of three ways over the course of a decade: good sleepers, catch-up sleepers, and insomnia sleepers. These sleep health phenotypes and proportions are largely consistent with those we previously identified in the same dataset cross-sectionally, across two independent subsamples of working adults (i.e., MIDUS II core and refresher samples; Smith et al., 2023). Although the MIDUS worker sample aged between the baseline (MIDUS II) and follow-up (MIDUS III) data collections, sleep health phenotypes can be consistently identified both cross-sectionally (Smith et al., 2023) and longitudinally. Other phenotypes identified in non-worker samples (e.g., nappers) were not found here, suggesting that work status may be an important determinant of sleep health phenotypes. Together with our new post hoc validation of these phenotypes using actigraphy and self-report sleep data (i.e., insomnia sleepers<catch-up sleepers<good sleepers in terms of sleep health), growing evidence supports these three sleep health phenotypes as common in working adults over time.

Related to our goal to understand the dynamics of multidimensional sleep health, organizational scientists have called for long-term study designs examining of the stability of various aspects of employee well-being to improve measurement and temporally sensitive theory (Mäkikangas et al., 2016). So far, research suggests that sleep health – albeit measured as independent dimensions rather than a multidimensional phenotype – may exist as a dynamic equilibrium wherein short-term fluctuation occurs (e.g. van Erde & Venus, 2018) but in the context of meso-term stability (Henderson et al., 2023). We add needed information about long- term stability and change. People are most likely to remain in the same sleep health phenotype over the long-term (i.e., approximately one decade), rather than transition to a new one. The prevalence of stability (60–90% of the sample) aligned with our expectations based on previous findings that 70–85% of people experience stable health and well-being in the long-term (Headey, 2006; Headey, 2010; Headey et al., 2014; Luhmann et al., 2018). Although dynamic equilibrium theory was developed to describe psychological well-being, our findings add to growing support for its application to other constructs related to worker health such as work- family conflict (Smith et al., 2022) and, here and in Henderson et al. (2023), sleep. Integrating this research with previous findings, sleep health can fluctuate from day to day and may even shift over months or years, but most people exhibit a stable constellation of co-occurring sleep dimensions that persist over a decade.

Beyond the tenets of dynamic equilibrium theory and Henderson et al.’s (2023) recent findings, however, the present study newly shows that the stability of sleep health differed based on a person’s baseline sleep health phenotype. Insomnia sleepers were the least likely to transition to a new phenotype over time; unfortunately, 90% of insomnia sleepers remained in the same highly suboptimal phenotype. Other research relatedly finds that people diagnosed with chronic insomnia do tend to experience persistent sleep problems one year later (Pillai et al., 2016), but our results indicate that co-occurring insomnia symptoms may be likely to persist over ten years. Indicating similar stability as insomnia sleepers, only 20% of good sleepers transitioned to a new phenotype, most of whom transitioned to catch-up sleeper (17%). Catch-up sleepers were the most likely to transition. Although catch-up sleepers often remained in the same phenotype (58%), they also commonly transitioned to good (36%) or to insomnia sleepers (6%). It is possible that catch-up sleep is primarily an adaptation to external circumstances (rather than a stable internal preference), making this phenotype likely to change when those circumstances change. The connection between nontraditional work schedules and catch-up sleep in the present study supports this potential explanation, as do previous findings that being younger or unpartnered relate to irregular sleep schedules (Lee et al., 2022). Applied more broadly, dynamic equilibrium perspectives should consider that the stability of health outcomes like sleep may not solely differ based on significant life events, as previously theorized, but also based on severity and nature of earlier health experiences (i.e., baseline sleep phenotypes here).

The present results also show that although sleep phenotypes are often stable, technology-related job design features seem to have implications for employees’ sleep health phenotype transitions when they do occur (i.e., for 10–40% of people depending on baseline phenotype). The connection between job design and sleep health phenotypes was robust even after accounting for sociodemographic factors and job demands. Lack of physical activity at work differentiated between good and insomnia sleepers. As expected, an increase in sedentary time at work related to remaining in or transitioning to the insomnia sleeper phenotype. This finding is consistent with past research showing the association between general (rather than job- specific) sedentary time and insomnia (Chen et al., 2014; Yang et al., 2017). In this study, we add rigorous evidence that sedentary time at work is associated with a maintenance of increase in a constellation of various, co-occurring insomnia symptoms over the course of a decade. Given the substantial increase in sedentary work (Church et al., 2011), its association with the maintenance or increase in a constellation of co-occurring insomnia symptoms over one decade is highly concerning. Counter to our hypotheses, an increase in computer use at work related to remaining in or transitioning to the good sleeper phenotype over insomnia sleeper. This finding could be attributed to the fact that MIDUS measurement of computer use does not specify timing of the use. Past work connecting screen time with poor sleep examines nighttime technology use (e.g., Vallance et al., 2015), rather than technology use during work hours. If computer use is limited to the daytime, use would not violate sleep hygiene recommendations to avoid active technology before bed (Gradisar et al., 2013). In fact, in some cases, workplace technology can help employees complete tasks in more efficient ways, to the benefit of their health (Bala & Venkatesh, 2016; Piszczek, 2017). Although definitive explanations could not be identified in the present data, perhaps the benefits of workplace computer use often outweigh the expected technostress associated, explaining the unexpected link to better sleep health found here.

People who remained in or who transitioned to nontraditional work schedules, specifically more frequent work during nighttime, were more likely to be consistent catch-up sleepers or transition to catch-up sleepers over time. While shift work has been linked to poor sleep quality in general (Chang & Peng, 2021), our findings add that any people working at night may attempt to deal with their misaligned sleep/wake cycle via a combination of daytime napping and weekend catch-up sleep. The catch-up sleep phenotype was also prominent among military, labor, and craft occupations in our post-hoc analyses. Based in past findings that many manual labor or skilled professionals (sometimes called “blue-collar”) have more physically demanding jobs (Azevedo et al., 2021) as well as erratic and/or on-demand work schedules, we speculate that the physical exhaustion and irregular schedules of people in these occupations may drive them to engage in “catch-up” strategies of napping and sleeping in on non-workdays despite seemingly getting healthy sleep during their normal sleep periods. Previous evidence in a general, national adult sample suggests that napping and long weekend sleep are insufficient in promoting healthy sleep (Leger et al., 2020), yet naps are empirically related to lower sleepiness during work for employees on non-standard schedules (Ruggiero & Redeker, 2014) and are clinically recommended for people diagnosed with shift work disorder (Wickwire et al., 2017). As such, not only is nontraditional work schedule a significant risk factor for the catch-up sleeper phenotype, but it is also possible that type of formally assigned work shift may moderate the health effects of being a catch-up sleeper. Of note, the catch-up sleeper phenotype exhibits sufficient sleep quantity and quality and, thus, would not have been identified by traditional one- or two-dimensional approaches typically applied in the organizational sciences. Examining multidimensional sleep health phenotype trajectories allowed the discovery of catch-up sleepers’ specific constellation of sleep experiences and identified the potential role of nontraditional work schedules in maintaining or inciting these experiences.

The above findings have important implications for the Balance Theory of Job Design. Balance Theory is under-utilized in occupational health psychology research, despite its unique ability to describe an essential question to the modern workplace: how technology-related changes to job design impact employee health (Carayon, 2009; Carayon & Smith, 2000; Kompier, 2006; Parker & Zhang, 2016; Smith et al., 1999). Our results show that Balance Theory can be extended to describe sleep health. Namely, we identify negative elements of job design (i.e., sedentary work, nontraditional schedules) that relate to suboptimal sleep health phenotypes. The counterintuitive association between work-related computer use and the good sleeper phenotype suggests that elements of a job’s design may be difficult to universally define as negative or positive within Balance Theory; the specific outcome measured (e.g., sleep health vs. other attitudes, behaviors, and health outcomes) may be necessary context when determining how to balance positive and negative job design factors. Moreover, Balance Theory should be integrated with theory and research specific to the relevant outcome to provide more detailed explanations. For instance, in our results, sedentary work and nontraditional schedules relate to different suboptimal sleep health phenotypes (i.e., insomnia sleeper and catch-up sleeper, respectively). Whereas Balance Theory would simply call both of these job design elements negative factors, integration with Buysse’s (2014) RuSATED sleep health framework and previous empirical work on job design better explain their differential links with specific sleep health phenotypes.

These results suggest that task-related, physical, and temporal job design are meaningfully linked to multidimensional employee sleep health over time and, as such, may be useful targets of intervention. Importantly, we urge organizations to consider the specific co-occurring sleep issues that are of high risk for their employees based on job design – and urge sleep intervention developers to consider the need to target particular job design challenges and sleep phenotypes. For instance, general sleep interventions attempting to improve either sleep quantity or quality may not be optimal for employees on nontraditional schedules, who instead may require help with their frequent napping and irregular sleep schedules. We wonder if the underwhelming effectiveness of many workplace sleep programs (Robbins et al., 2019) could partially be due to lack of specificity and targeting of workforce vulnerabilities and needs.

Organizations employing workers who operate on nontraditional schedules, specifically working during the night, should be aware that their workforces may be vulnerable to catch-up sleep. Some employees may work during non-traditional times by choice, such as in flexible work arrangements (Bolino et al., 2021), so self-imposed work at nighttime should also be considered a potential risk in addition to formal shiftwork arrangements. Both should be reduced when possible to decrease odds of catch-up sleep. When direct changes to work schedules are not possible, existing interventions developed to improve sleep for workers on nontraditional schedules could be applied, though their effectiveness for co-occurring suboptimal sleep dimensions (i.e., frequent napping, long weekend sleep) is not yet known. For example, light exposure (Olson et al., 2020) and improved scheduling adjusted for chronotype (Declercq et al., 2022) could serve as more immediately adjustable targets and should be tested in relation to the catch-up sleeper phenotype.

Similarly, existing interventions that aim to increase physical activity among sedentary workers (e.g., Carr et al., 2016) could be tested in the context of sleep to see if they are effective when targeting an interdependent pattern of sleep issues. Given that insomnia sleeper is the most suboptimal but also the most stable phenotype (i.e., least likely to change), future efforts should examine whether reducing sitting time at work may lead to transitioning from the insomnia sleeper to some other phenotype. Conversely, based on our findings, work-related computer use may not be a primary area for concern in terms of employee sleep health. Moreover, sleep health interventions can also supplement changes to job design, but our findings suggest that they require more detailed measurement of sleep health (i.e., beyond quantity and quality). Catch-up sleepers report sufficient sleep quantity and quality and, thus, may go undetected as suboptimal sleepers using traditional measurement, despite their irregular sleep and frequent napping. Further, insomnia sleepers report a constellation of sleep issues that may be interdependent and best tackled together. We extend calls to develop workplace interventions targeting multiple dimensions of employee sleep health (Chung et al., 2021) to also consider specific needs of common sleep health phenotypes.

All data for the main analyses were collected via self-report surveys, which may contribute to common method bias (CMB; Podsakoff et al., 2003). Objective job design data (e.g., from O*NET based on job title) could be incorporated in the future to deal with CMB. For sleep, although we include “objective” actigraphy data to validate our self-report profiles, actigraphy data was not used for profile identification. Subjective and objective measurement are considered related but not redundant and, therefore, complimentary when measuring sleep (Girschik et al., 2012; Zinkhan et al., 2014). For instance, in another study using MIDUS data, researchers found significant but small correlations (r=.10 to .28) between a composite of self-report sleep quality and actigraphy duration, midpoint, sleep onset latency, and wake after sleep onset (WASO) (Teas & Friedman, 2021). Moreover, a mismatch between subjectively and objectively measured sleep characteristics can indicate additional sleep issues (Bianchi et al., 2013). While incorporating objective measurement of sleep involves high cost, participant burden, and small sample size, joint assessment of subjective, objective, and match between subjective/objective sleep characteristics in determining sleep health phenotypes could potentially uncover new phenotypes or provide unknown information about existing phenotypes.

Several limitations should also be noted related to the archival sample and study design. First, although the data was collected at two time points rather than cross-sectionally, the design still does not test temporal precedence needed for strong causal inference (Shadish et al., 2002). We hope that our findings can ground future repeated-measures longitudinal studies that tease out directionality and timeline of change (Taris & Kompier, 2014). Additionally, in the MIDUS datasets, we could not definitively determine whether participants’ job design elements changed within the same job over the two time points or changed due to a job change. While we focused on the three technology-related job design elements (i.e., computer use, sedentary work, and nonstandard schedules), we were unable to determine whether changes in these elements were indeed driven by technology. The current design still allows for examination of general increases or decreases in each design element, but future studies could gather more information behind changes in job design elements. The MIDUS samples themselves also aim to be representative of U.S. adults – and are more representative than smaller-scale, more sparsely resourced research projects. Yet, as it applies to our study goals, the present analytic sample (i.e., a working adult subsample of MIUDS 2 and 3) is slightly less diverse than the general working adult population in the U.S in terms of race and ethnicity (BLS, 2021). We applaud MIDUS researchers for using targeted recruiting in their Milwaukee samples to increase diversity, and we encourage continued and expanded efforts in the future, among MIDUS and other ongoing data collections, to better describe more of the working population. For this reason, we also urge future researchers to replicate our findings in different and, especially, more diverse samples to test the generalizability of the sleep health phenotypes across working populations. These efforts should be grounded in strong sleep theory, such as the RuSATED framework (Buysse, 2014), when making decisions about sleep measurement and by best practice analytic guidelines for testing similarity of solutions across samples (Morin et al., 2016). Whereas consistent application of the same sleep dimensions and cut-off scores used here across different measurement techniques (i.e., self-report vs. actigraphy) and samples could enable tests of generalizability, inclusion of additional dimensions (e.g., timing, which was not included in the MIDUS dataset) or alternative measurement (e.g., evaluative measures of sleep quality, rather than sleep complaints, to measure satisfaction) could help expand understanding of the identified phenotypes and potentially identify new ones.

It is also notable that, at the time of this study, MIDUS only provided two waves of data assessing sleep health (MIDUS 2 and 3). Two waves of data are not sufficient to meet most standards of longitudinal design (Ployhart & MacKenzie, 2014). Thus, although the currently available data facilitates detailed understanding of how job design and sleep health associate over one decade, true causal attributions cannot be made. In particular, the two-wave design limited our ability to propose and test mediation, which requires at least three waves of data (Cole & Maxwell, 2003; O’Laughlin et al., 2018). Thus, we would also encourage future research in this area to rigorously test theoretically supported mediators to provide more explanatory detail about the mechanisms and causal chain of events connecting job design to employee sleep health (MacKinnon et al., 2007). The collection of additional MIDUS data would also allow extension of the timeline of sleep phenotype trajectories and provide more recent data, beyond 2017, when considering questions about current technology and sleep. Further, an extension could answer questions about how sleep phenotypes change across life stage, age, or, for example, when people transition from employment to retirement. Additionally, the first wave in our study (MIDUS 2) began data collection in 2004, and the second wave (MIDUS 3) ended data collection in 2017. Since then, not only has technology continued to advance but the COVID-19 pandemic has drastically changed where, when, and how work is done (e.g., Collins et al., 2020; McDowell et al., 2020). Purposeful additions to future MIDUS data collections could include variables that support the most up-to-date understanding of sleep health (i.e., also including timing) and, especially, of technology-related job design elements.

Finally, our model is limited in its ability to fully test Balance Theory. We test its first tenet by identifying two negative elements of job design as they relate to specific patterns of suboptimal sleep health. Now, additional efforts theoretically and practically needed to identify positive elements of job design that would counterbalance effects for each suboptimal sleep health phenotype (i.e., second tenet of the theory) and test mediators such as cognitive, psychological, or physical loads (i.e., mechanisms proposed by the theory; Carayon, 2009).