Abstract
Study Objectives
Diverse studies have reported longer sleep durations and later circadian timing during the initial COVID-19 lockdown period. Little is known about whether effects persisted after 2020. This analysis addresses three questions: (1) How did sleep timing and duration change from 2017 to 2023? (2) Did working from home explain trends? (3) Did effects differ by education, income, or race/ethnicity groups?
Methods
The American Time Use Survey is a nationally representative survey conducted by the US Bureau of Labor Statistics that collects 24-hour time diaries. These data are used to identify respondents who worked on the sampled day, their work location (home or not), and three sleep variables: wake-up time, bedtime, and 24-hour sleep total. Ordinary least squares regression is used to answer the study questions, comparing the COVID time period (May 2020 to December 2023) to PRECOVID (January 2017 to March 2020).
Results
Sleep duration was longer in the COVID time period compared to PRECOVID, by 0.23 hours (95% confidence interval = 0.17, 0.29), with earlier average bedtimes and later average waking times. There were no significant secular trends in sleep outcomes within the COVID time period, suggesting that these changes have continued through 2023. Controlling for working from home modestly attenuated, but did not eliminate, the COVID effects. Effects were generally similar across sociodemographic groups.
Conclusions
COVID-related changes in sleep for working adults in the United States, specifically later circadian timing and increased duration, seem to be sustained through 2023.
Statement of Significance
Evidence from different locations and diverse types of sleep data found that the initial months of lockdown following the beginning of the COVID-19 pandemic in March 2020 saw increased sleep durations across populations. Using a nationally representative sample of 24-hour time diaries in the United States from 2017 to 2023, this study confirms with national data that there were longer average sleep durations and later waking times after the beginning of the pandemic and that they have continued through 2023. These effects are not fully explained by working from home (WFH). While the pandemic-related increase in WFH varied greatly by education, income, and race/ethnicity groups, the COVID-19 effect on sleep characteristics was similar across these groups.
Keywords: COVID-19, sleep, circadian timing, time diaries, United States
Inadequate sleep has been implicated as a risk factor for many disease processes, including cardiovascular diseases, diabetes, and mental health conditions [1–5]. The prevalence of weekend catch-up sleep, that is, later wake times and longer sleep durations on non-workdays, implies that work, together with commuting time, is a common impediment to adequate or desired sleep [6, 7]. The COVID-19 pandemic altered work patterns dramatically, following lockdowns across the globe in March 2020. Using a labor market survey, economists found that working from home (WFH) increased from 14 per cent of workdays before the pandemic to 40 per cent in May 2020, and remained higher, close to 30 per cent, in June 2021 [8]. They speculated that the sustained high rate of WFH reflected an underlying trend that was simply accelerated by the pandemic.
Diverse studies have investigated the effects of the early pandemic, when lockdowns were widespread, on sleep duration, quality, and timing. However, the attention has largely been on the initial lockdown period and not longer-term impacts. A meta-analysis of 154 studies found an overall increase in sleep durations and a circadian shift to later bedtimes and waking times [9]. One study used a smartphone app (“sleep cycle”) to estimate sleep durations in several cities in North America, Europe, and Asia and found an average increase of 22 minutes when comparing sleep in April 2020 to April 2019 [10]. An online survey in India found average sleep onset to be 38 minutes later, while wake-up time averaged 51 minutes later during lockdown [11]. In the United States, an analysis of the Behavioral Risk Factor Surveillance System compared responses collected from March through December 2020 to earlier years and found a significant increase in sleep hours per day reported in response to a survey question asking for average sleep hours in a 24-hour period [12].
It seems intuitive that a key explanatory mechanism linking lockdowns to sleep duration and later timing must have been the sudden elimination of the need to wake up to get to work for many adults, either because the job was suspended during the lockdown or because it was moved to work at home, eliminating commuting time and directly observable arrival times. Early work start times have been shown to be associated with reduced sleep [13]. Nonetheless, one study found that the lockdown itself rather than WFH was associated with changed sleep patterns [14]. To the extent that WFH is itself a determinant of later or longer sleep, there may be sustained changes since the declaration of end of the pandemic in 2023, but those changes may differentially affect different population subgroups.
To understand at a national level whether sleep timing and duration changed with the pandemic and whether these changes have persisted for working adults in the United States through 2023, we use the 24-hour time diaries collected by the US Bureau of Labor Statistics. The American Time Use Survey (ATUS) is a large, nationally representative 24-hour time-diary survey that collects data monthly. Using these data from before and after the beginning of the pandemic in the United States in March 2020, we examine whether there is a population-level change in the average times on workdays when the night-time sleep interval begins and ends, and in the total minutes recorded as sleep over 24 hours. We investigate whether there are secular trends within the time periods before and after March 2020, and whether sleep characteristics returned to pre-pandemic population levels by 2023. We use the information about the location of work activity, whether at a workplace or at home, to examine the extent to which WFH explains changes in sleep patterns.
Further, we consider that the ability to WFH and the technology that facilitated it were resources that were intended not only to help prevent disease but also to improve quality of life, including reducing work-related constraints on sleep hours and timing. We ask whether being able to take advantage of these innovations differentially impacted different population subgroups such that socio-economic differences in sleep increased with the pandemic. This perspective is broadly consistent with “fundamental cause theory,” which posits that socioeconomic status and social factors are fundamental causes of disease because they lead to differential access to health-promoting resources [15]. Therefore, we investigate whether sleep patterns differed by levels of educational attainment, household income, or race/ethnicity, and further, whether these differences increased with the pandemic.
Materials and Methods
This analysis aims to answer the following questions: How did bedtimes, wake-up times, and 24-hour sleep change from 2017 to 2023? Does WFH explain these changes? And do these effects differ by demographic groups? We hypothesized that the pandemic would increase disparities in sleep between sociodemographic groups.
Data
ATUS data
ATUS is a nationally representative, cross-sectional survey conducted every year by the US Bureau of Labor Statistics [16]. Sampled individuals are asked to describe their activities over the previous 24-hour period from 04:00 am to 04:00 am. ATUS respondents are a randomly chosen subset of those who participated in the Current Population Survey (CPS). All ATUS interviews are conducted by phone using Computer-Assisted Telephone Interview technology. The interviewer asks respondents to walk through their activities starting from 04:00 am the previous day (diary day) to 04:00 am of the interview day and records responses verbatim, which are then coded using the ATUS activity lexicon, along with start and stop times for each activity, the location of the activity, and some other information. This analysis used ATUS data that were prepared by IPUMS, a center for data integration within the Institute for Social Research and Data Innovation at the University of Minnesota.ref ATUS data are de-identified and publicly available data, and this study was determined to be exempt by the University of Chicago Institutional Review Board.
The ATUS uses a three-tiered, six-digit activity code with 17 major activities, followed by two levels of additional detail. As an example, the major activity of “traveling” has a code of 180 000. One of the second-level codes is “travel related to work,” which is 180 500, and a third level of detail under this includes “travel related to job search and interviewing,” which is 180 504.
Sample
Our analytic sample includes 18 249 ATUS time diaries collected from January 2017 through December 2023 for those aged 25−65 who reported work activities on the day sampled.
Variables
Sleep
Sleep is a second-level activity under the major category “personal care” and includes three third-level distinctions: sleeping (includes activities such as napping and sleeping), sleeplessness (includes activities such as insomnia), and sleep not elsewhere classified. We define three sleep-related variables from the time diaries. It is appropriate to think of these variables as representing “time set aside for sleep” because there is no physiological measure of sleep, and there is likely individual variation in the completeness of reporting sleepless periods within the sleep activity, which would require knowing the clock times of those periods. Therefore, we consider all of activities under the sleep activity code to be sleep.
Our first sleep variable is “wake-up time.” This is determined by finding the first recorded sleep in the respondent’s diary day (which commences at 04:00 am) and taking the stop time of that activity. To account for interruptions (e.g. bathroom activities) that some respondents report within the sleep period, we check if another sleep activity begins within 30 minutes of the end of the previous sleep activity. If this is the case, then the stop time of the last sleep activity in the chain of sleep activities starting with the first sleep activity period is the respondent’s wake-up time.
The second sleep variable we define is “bedtime.” Bearing in mind that some respondents detailed short waking times during their sleep intervals while others did not (but likely at least some experienced them), we define bedtime as follows: first, find the start time of the respondent’s last recorded sleeping activity before the final 04:00 am and then check if another bout of sleep ended within 30 minutes before this start time. If so, record the start time of the previous bout of sleep and then repeat the search process, checking if another bout of sleep ended within 30 minutes of an earlier bout of sleep. This creates a chain of sleep periods that all occur within 30 minutes of each other. The start time of this entire chain of sleep is taken to be the “bedtime” of the respondent. “Bedtime” can be thought of as the beginning of the period that the respondent set aside for sleep at night but might not be the start of actual sleep. While this algorithm is specific to this analysis, it is consistently applied for all years and therefore trends identified should be meaningful, although bedtimes may not be comparable to other studies and data sources that determine the sleep interval using different methods.
The third sleep variable is “24-hour sleep.” This variable, constructed by ATUS, simply sums all sleep activities over the 24 hours from 04:00 am to 04:00 am, which would also include naps.
For respondents who held nightshift sleep schedules and have a main sleep period during the daytime (inferred by the respondent’s last recorded sleep ending before midnight), the wake-up time is the stop time of the respondent’s last sleep activity. The bedtime is the start time of the first sleep activity. This includes about 2 per cent of the sample.
Work-related variables
A respondent is considered to have worked the sampled day if there is an activity code for “work, main job” (050101) on their diary day.
WFH is determined by whether the first activity of “work, main job (050101)” had a location reported as “home (0101)” or not. “WFH” is a binary variable.
To further understand the role of commuting to work on sleep schedules, we also construct a variable “commute time.” This is determined by the respondent’s first recorded instance of “travel related to work (180501).” It is the duration of a one-way commute. This is a continuous variable that is the number of minutes of the commute. It is 0 if there is no commute time but there is “work, main job” indicated.
Temporal variables
The ATUS suspended data collection from March 18, 2020 to May 9, 2020 during the initial lockdown for COVID-19. We constructed four temporal variables. Two are binary variables: “PRECOVID” and “COVID.” PRECOVID is coded as “1” if the respondent’s interview day was during the period from January 1, 2017 to March 17, 2020, and otherwise “0”. The COVID variable, which includes both the pandemic and post-pandemic months, is coded “1” if the sampled day is between May 10, 2020 and December 31, 2023, and “0” otherwise.
Two additional temporal variables are needed to test whether there are significant secular trends within the “PRECOVID” or “COVID” time periods. The variable “PREMONTH” takes values from 0 to 38, indicating the order of the month, with 0 corresponding to January 2017 and 38 corresponding to March 2020. Similarly, the variable “POSTMONTH” takes values from 0 to 43, indicating the order of the month, with 0 corresponding to May 2020 and 43 corresponding to December 2023. PREMONTH is 0 for observations during the COVID period, and POSTMONTH is 0 for observations during the PRECOVID period.
Demographic variables
These analyses use standard sociodemographic variables collected in the CPS, all of which have frequently been found to be associated with sleep characteristics in diverse studies. Sex and age are included in all models, and age is centered at the mean. Race/ethnicity has been grouped into these categories: white non-Hispanic, black non-Hispanic, Asian, Hispanic, and other. Household income has been grouped into a six-level ordinal variable: “1” for income less than $35 000, “2” for $35 000-$49 999, “3” for $50 000-$74 999, “4” for $75 000-$99 999, “5” for $100 000-$150 000, and “6” for income greater than $150 000. Education is a five-level ordinal variable: “1” for no high school diploma, “2” for high school graduate or GED, “3” for some college/associates degree, “4” for bachelor’s degree, and “5” for a graduate degree.
Statistical analysis
To answer the first question of how COVID affected each of the sleep variables (wake-up time, bedtime, 24-hour sleep), we first show the unadjusted monthly averages, with a trend line derived from LOESS (locally estimated scatterplot smoothing) regression. Then, we employ ordinary least squares regression models for each outcome as a function of COVID status, with PRECOVID as the omitted referent category and include demographic variables age, sex, race/ethnicity, income, and education.
To examine whether there are secular trends in sleep variables within either the PRECOVID or COVID time periods, we employ no-intercept ordinary least squares regression models for each sleep outcome as a function of both the COVID and PRECOVID variables and include the two ordinal month variables PREMONTH and POSTMONTH, which are the variables of interest to determine secular trends. These models include the same demographic covariates.
To answer the second question of whether the COVID effect on sleep outcomes is explained by increased WFH or decreased commute time, we add those variables, separately, to the models that test the COVID effect and examine both the coefficients for WFH and commute time, as well as their confounding effects on the COVID time period variable.
To answer the third question, if COVID had a different effect on sleep for population subgroups formed by racial, education, or income categories, we add interaction terms between the COVID indicator variable and race/ethnicity, income, or education (in separate models) to the regression models for each of the three sleep variables.
All models include the observation weights provided by the Bureau of Labor Statistics. All regression analyses were done in R 4.3.2 using the “survey” package.
Model fit was assessed by inspecting plots of the Pearson residuals versus fitted values. The residual plots showed no leftover pattern and appeared randomly scattered about zero, indicating good fit for the data, independence of residuals, and no heteroskedasticity.
Results
This analysis included 18 249 respondents aged 25−65 who had a work activity on the diary day (Table 1). In 2020, about one-quarter were surveyed in the PRECOVID period. Age and sex remained similar in the two periods. Race/ethnicity distribution was slightly different in the two time periods with 1 or 2 per cent differences in the percentages in the race groups in the time periods. There were also small differences in the education and income distributions with about a 2 per cent increase in each case in the higher group in the later time period. The overall percentage WFH increased from 22 per cent in the PRECOVID time period to 38 per cent in the COVID time period. Commute time for those with a commute changed little in the two time periods (average 25 minutes PRECOVID versus average 24 minutes COVID), but was shorter in the COVID time period when the zero-minute commutes for those WFH are included in the average (19 minutes versus 15 minutes).
Table 1.
Distributions of Sociodemographic, Employment Location, and Work Commute Duration Characteristics for Adults Aged 25−65 on Workdays from 2017 to 2023 from Time Diaries Collected in the ATUS
| Characteristic | Overall n = 18 249* | PRECOVID‡ n = 9342* | COVID‡ n = 8907* | P-value† |
|---|---|---|---|---|
| Year | ||||
| 2017 | 3108 (17%) | 3108 (33%) | 0 (0%) | |
| 2018 | 2849 (16%) | 2849 (30%) | 0 (0%) | |
| 2019 | 2759 (15%) | 2759 (30%) | 0 (0%) | |
| 2020 | 2386 (13%) | 626 (6.7%) | 1760 (20%) | |
| 2021 | 2605 (14%) | 0 (0%) | 2605 (29%) | |
| 2022 | 2277 (12%) | 0 (0%) | 2277 (26%) | |
| 2023 | 2265 (12%) | 0 (0%) | 2265 (25%) | |
| Age | 45 (11) | 44 (11) | 45 (11) | .4 |
| Sex | .6 | |||
| Female | 8613 (47%) | 4429 (47%) | 4184 (47%) | |
| Male | 9636 (53%) | 4913 (53%) | 4723 (53%) | |
| Race | <.001 | |||
| Asian | 1154 (6.3%) | 537 (5.7%) | 617 (6.9%) | |
| Black | 2021 (11%) | 1097 (12%) | 924 (10%) | |
| Hispanic | 2715 (15%) | 1347 (14%) | 1368 (15%) | |
| Other | 313 (1.7%) | 149 (1.6%) | 164 (1.8%) | |
| White | 12 046 (66%) | 6212 (66%) | 5834 (65%) | |
| WFH | 5407 (30%) | 2059 (22%) | 3348 (38%) | <.001 |
| Commute time zeroes excluded, mean minutes (SD) | 24 (22) | 25 (23) | 24 (2) | .9 |
| Commute time including zeros, mean minutes (SD) | 17 (21) | 19 (23) | 15 (20) | <.001 |
| Education | <.001 | |||
| No diploma | 851 (4.7%) | 454 (4.9%) | 397 (4.5%) | |
| High school diploma | 3312 (18%) | 1753 (19%) | 1559 (18%) | |
| Some college/associates degree | 4481 (25%) | 2388 (26%) | 2093 (23%) | |
| Bachelor’s degree | 5466 (30%) | 2709 (29%) | 2757 (31%) | |
| Graduate degree | 4139 (23%) | 2038 (22%) | 2101 (24%) | |
| Income | <.001 | |||
| Less than $35 000 | 2795 (15%) | 1653 (18%) | 1142 (13%) | |
| $35 000−$49 999 | 2025 (11%) | 1152 (12%) | 873 (9.8%) | |
| $50 000−$74 999 | 3384 (19%) | 1759 (19%) | 1625 (18%) | |
| $75 000−$99 999 | 2737 (15%) | 1392 (15%) | 1345 (15%) | |
| $100 000−$150 000 | 3436 (19%) | 1668 (18%) | 1768 (20%) | |
| Greater than $150 000 | 3872 (21%) | 1718 (18%) | 2154 (24%) | |
* n (%); mean (SD).
†Pearson’s Chi-squared test; Wilcoxon rank sum test.
‡“PRECOVID” includes time diaries collected from January 2017 through March 17, 2020, when data collection was suspended. “COVID” includes time diaries collected from May 10, 2020 to December 31, 2023.
Figure 1 shows the raw monthly averages (with survey adjustment) and a smoothed regression line for each of the three sleep variables. There is substantial month-to-month variation; the number of observations per month averages about 220. Table 2 displays the adjusted regression models predicting wake-up time, bedtime, and 24-hour sleep as a function of the COVID time period and demographic variables. The beta-coefficients for COVID indicate significant effects on all three sleep outcomes. Wake-up time was significantly later, and Bedtime was significantly earlier in the COVID time period compared to the PRECOVID time period. There was also a significant increase of almost a quarter-hour for 24-Hour Sleep, compared to PRECOVID. Several of the sociodemographic variables were significantly associated with sleep outcomes. Income was significantly associated with all three sleep outcomes, with higher income associated with earlier wake-up times and earlier bedtimes, such that there was a modeled 30-minute earlier Wake-Up Time and 18-minute earlier Bedtime for the highest income group compared to the lowest income group. Income was also significantly associated with 24-hour sleep, with higher incomes associated with less sleep. The highest income group had about 12 minutes less sleep over the 24 hours, compared to the lowest income group. The education effect was only significant for bedtime, with greater education associated with later bedtime. Compared to the referent white race group, the Asian group had significantly later sleep timing, for both wake-up and bedtime, and also longer 24-hour sleep. The black race group also had later sleep timing compared to the white group, for both wake-up and bedtime, but no significant difference from the white group in 24-hour sleep. The Hispanic group had significantly earlier bedtime and longer 24-hour sleep compared to the white referent group.
Figure 1.
Monthly averages for wake-up time, bedtime, and 24-hour sleep for the PRECOVID time period (January 2017 to March 17, 2020) and the COVID time period (May 2020 through December 2023), with trend line from LOESS regression with 95% CIs. Data are derived from the 24-hour time diaries collected through the ATUS.
Table 2.
Regression Models of Wake-Up Time, Bedtime and 24-Hour Sleep on COVID Time Period (May 2020 Through December 2023) Compared to PRECOVID (January 2017 to March 17, 2020) Time Period, Adjusted for Gender, Age, Education, Income and Race/Ethnicity, from the ATUS 2017−2023
| Dependent variable | |||
|---|---|---|---|
| Wake-up time (am) | Bedtime (24-hour clock) | 24-hour sleep (hours) | |
| (1) | (2) | (4) | |
| COVID† | 0.12** | −0.08* | 0.23** |
| (0.03, 0.21) | (−0.1, −0.004) | (0.17, 0.29) | |
| Male | −0.07 | 0.0 | −0.19** |
| (−0.16, 0.01) | (−0.002, 0.15) | (−0.25, −0.13) | |
| Age (per year) | −0.025** | −0.012** | −0.011** |
| (−0.029, −0.021) | (−0.016, −0.009) | (−0.014, −0.009) | |
| Education (per level, 5 levels) | 0.02 | 0.09** | 0.009 |
| (−0.02, 0.07) | (0.05, 0.12) | (−0.022, 0.04) | |
| Income (per level, 6 levels) | −0.10** | −0.06** | −0.04** |
| (−0.13, −0.07) | (−0.09, −0.04) | (−0.06, −0.02) | |
| Race | |||
| White | (referent) | (referent) | (referent) |
| Asian | 0.38** | 0.22** | 0.30** |
| (0.22, 0.53) | (0.07, 0.37) | (0.18, 0.41) | |
| Black | 0.23* | 0.51** | −0.09 |
| (0.05, 0.41) | (0.35, 0.66) | (−0.22, 0.04) | |
| Hispanic | −0.05 | −0.16** | 0.11* |
| (−0.19, 0.09) | (−0.28, −0.04) | (0.01, 0.20) | |
| Other | −0.04 | 0.38* | −0.27* |
| (−0.37, 0.28) | (0.05, 0.70) | (−0.53, −0.01) | |
| Constant | 6.77** | 22.59** | 8.08** |
| (6.58, 6.95) | (22,42, 22.77) | (7.95, 8.21) | |
*p < .05.
**p < .01.
†COVID includes time diaries collected from May 10, 2020 through December 31, 2023. These are compared to time diaries collected from January 2017 through March 17, 2020, when data collection was suspended.
When the no-constant models were constructed to evaluate whether there were significant secular trends within either the PRECOVID or COVID time periods, none of the coefficients for the month trend variables were significant for any of the three outcomes (Table 3). Therefore, the secular trend variables are not included in the models assessing WFH effects and effect modification.
Table 3.
No-Intercept Regression Models of Wake-Up Time, Bedtime and 24-Hour Sleep. Four Time Variables are Included to Assess Secular Trend by Month within the PRECOVID (from January 2017 to March 17, 2020) and COVID (May 2020 through December 2023) Time Periods
| Dependent variable | |||
|---|---|---|---|
| Wake-up time (am) | Bedtime (24-hour Clock) | 24-hour sleep (hours) | |
| PRECOVID† secular trend per month | 0.001 | −0.002 | 0.003 |
| (−0.004, 0.007) | (−0.007, 0.003) | (−0.0004, 0.007) | |
| COVID secular trend per month | −0.003 | −0.0002 | 0.001 |
| (−0.007, 0.002) | (−0.005, 0.004) | (−0.003, 0.004) | |
| PRECOVID indicator | 6.74** | 22.62** | 8.02** |
| (6.53, 6.96) | (22.43, 22.82) | (7.88, 8.16) | |
| COVID indicator | 6.94** | 22.51** | 8.30** |
| (6.74, 7.14) | (22.32, 22.70) | (8.15, 8.44) | |
*p < .05.
**p < .01.
†PRECOVID includes time diaries collected from January 2017 through March 17, 2020, when data collection was suspended. COVID includes time diaries collected from May 10, 2020 through December 31, 2023.
All models are also adjusted for gender, age, education, income, and race/ethnicity as in Table 2, from the ATUS 2017−2023.
Figure 2 shows the trends in percentage WFH by month for both the PRECOVID and COVID time periods. With the beginning of the COVID time period, there is an abrupt increase in the proportion WFH and then a gradual trend downward during the COVID time period. The percentage WFH remains higher through 2023 than during the PRECOVID period. In the PRECOVID time period, there were both income and education gradients in WFH, and they were amplified in the COVID time period, with those with the highest income and most education having much higher percentages WFH in the COVID period. The race pattern changed with COVID. There was only slight evidence of differences by race/ethnicity in WFH during the PRECOVID period, but the Asian and white race categories had much higher proportions WFH in the COVID period.
Figure 2.
Percentages of working adults WFH from January 2017 through December 2023, overall, by education level, by household income level, and by race/ethnicity group. Data are derived from 24-hour time diaries collected through the ATUS.
When WFH was added to the regression models for the three sleep variables (Table 4), WFH was significantly associated with all three sleep outcomes, with those WFH having later sleep times and longer 24-hour sleep. Twenty-four-hour sleep averaged 23 minutes more for those who had WFH. However, the difference in the beta coefficients for the COVID effect when WFH was included in the model as a confounder (Table 4) versus when it was not (Table 2) did not indicate that WFH greatly attenuated the COVID effect, reducing it by about 20 per cent for both wake-up time and 24-hour sleep but increasing the COVID effect on bedtime by about 50 per cent. The other covariates (not shown) were similar when WFH was added to the model. Alternatively, when commute time in minutes was added to the regression models instead of WFH, it also had significant main effects, but the confounding effect as indicated by the change in the beta coefficients for the COVID effect was very similar to when WFH was included in the models.
Table 4.
Regression Models of Wake-Up Time, Bedtime and 24-Hour Sleep on COVID Time Period (from May 2020 to December 2023) Compared to PRECOVID Time Period (January 2017 to March 17, 2020), Adjusted for WFH or Commute Time in Minutes
| Dependent variable | |||
|---|---|---|---|
| Wake-up time (am) | Bedtime (24-hour clock) | 24-hour sleep (hours) | |
| Model 1 | |||
| COVID† | 0.10* | −0.12* | 0.18** |
| (0.01, 0.18) | (−0.20, −0.04) | (0.11, 0.24) | |
| WFH | 0.17** | 0.25** | 0.38** |
| (0.09, 0.26) | (0.17, 0.32) | (0.31, 0.44) | |
| Model 2 | |||
| COVID | 0.10* | −0.10* | 0.19** |
| (0.01, 0.19) | (−0.18, −0.02) | (0.13, 0.24) | |
| Commute (minutes) | −0.005** | −0.003** | −0.009** |
| (−0.007, −0.003) | (−0.005, −0.001) | (−0.011, −0.007) | |
*p < .05.
**p < .01.
†COVID includes time diaries collected from May 10, 2020 to December 31, 2023. These are compared to time diaries collected from January 2017 to March 17, 2020, when data collection was suspended.
All models are also adjusted for gender, age, education, income, and race/ethnicity, from the ATUS 2017−2023.
Table 5 presents the interaction terms between COVID and education, income, or race categories to assess whether the COVID effect on sleep outcomes differs between population subgroups defined by these demographic characteristics. None of the interaction terms between education level and COVID time period were statistically significant, indicating that education did not modify the COVID effect on sleep characteristics. Similarly, none of interaction terms between race/ethnicity groups and COVID time period was significant, indicating that the COVID effect did not differ by race/ethnicity. The interactions between income level and COVID for both Wake-Up Time (0.07; 95% confidence interval [CI] = 0.01, 0.11) and bedtime (0.06; 95% CI = 0.009, 0.11) were marginally significant (Table 5), indicating that the earlier sleep timing associated with a higher income level (Table 2) was reduced in the COVID time period compared to the PRECOVID time period. Specifically, the beta coefficient for Income level in the interaction model for wake-up time is −0.14 (95% CI = −0.18, −0.10), which is the modeled income effect in the PRECOVID time period. Combining that with the interaction term, the modeled income effect in the COVID time period is about half as large. Similarly, the beta coefficient for Income level in the interaction bedtime model is −0.09 (95% CI = −0.13, −0.06), indicating that is the income effect PRECOVID. Combining that with the interaction term indicates that the association between a higher income level and earlier bedtime is less than half of that in the COVID time period.
Table 5.
Interaction Terms from Regression Models of Wake-Up Time, Bedtime and 24-Hour Sleep on COVID Time Period (May 2020 to December 2023) Compared to PRECOVID Time Period (January 2017 to March 17, 2020), Including Interaction Terms between COVID and Education, COVID and Income, or COVID and Race/Ethnicity Categories
| Dependent variable | |||
|---|---|---|---|
| Wake-up time (am) | Bedtime (24-hour clock) | 24-hour sleep (hours) | |
| Interaction model 1 | |||
| Interaction between COVID† and education | 0.056 | 0.057 | 0.010 |
| (−0.02, 0.13) | (−0.01, 0.13) | (−0.042, 0.063) | |
| Interaction model 2 | |||
| Interaction between COVID and income | 0.07* | 0.06* | 0.03 |
| (0.01, 0.12) | (0.01, 0.11) | (−0.009, 0.06) | |
| Interaction model 3 | |||
| Interactions between COVID and race | |||
| Asian × COVID | 0.06 | 0.24 | −0.22 |
| (−0.25, 0.37) | (−0.06, 0.53) | (−0.46, 0.01) | |
| Black × COVID | −0.14 | −0.09 | 0.17 |
| (−0.49, 0.22) | (−0.40, 0.23) | (−0.09, 0.43) | |
| Hispanic × COVID | −0.06 | 0.06 | 0.08 |
| (−0.31, 0.20) | (−0.16, 0.28) | (−0.10, 0.26) | |
| Other × COVID | 0.45 | 0.19 | 0.28 |
| (−0.16, 1.06) | (−0.45, 0.82) | (−0.27, 0.84) | |
*p < .05.
**p < .01.
†COVID includes time diaries collected from May 10, 2020 to December 31, 2023. These are compared to time diaries collected from January 2017 to March 17, 2020, when data collection was suspended.
All models also include COVID, gender, age, education, income, and race/ethnicity, from the ATUS 2017−2023.
Discussion
We have used a large, nationally representative probability sample of 24-hour time diaries from the adult, working population in the United States to compare bedtimes, waking times and 24-hour total sleep time on workdays before and after the beginning of the COVID-19 pandemic. We found that sleep duration was longer in the time period that included the COVID-19 pandemic through the end of 2023, compared to a pre-COVID-19 time period from January 2017 through early March, 2020, with waking times later and bedtimes earlier. The increase in mean sleep duration was about a quarter-hour. There were no significant secular trends in sleep timing or duration within the three years prior to the pandemic, nor within the 3 years after the start of the pandemic, suggesting that these changes in sleep timing and duration have endured past the peak of the pandemic. While many studies investigated how the early months of the pandemic affected sleep [9–12], there has been little evidence about longer-term effects.
Key demographic variables were associated with sleep timing and duration, both before and following the start of the pandemic. Increasing age was associated with earlier sleep timing and shorter duration, which is consistent with prior research, as was the shorter duration for men [17, 18]. Our contrasting findings for income and education, with higher income being associated with earlier sleep timing and shorter duration, while more education is only associated with later bedtimes, differ from most prior studies that have investigated socioeconomic effects on sleep in the general population [19]. However, the income effect was muted in the COVID time period compared to earlier years. There were several significant associations between race/ethnicity groups and sleep timing and duration, adjusted for other demographic characteristics. Sleep timing was later for both the Asian and black race groups, compared to the white race category, while 24-Hour Sleep was much longer for the Asian category and somewhat longer for the Hispanic category, compared to the white category. Most prior research on race and ethnicity has focused on sleep duration for black and white groups, and generally found shorter duration for blacks [19–21]. There is more limited evidence about race/ethnicity and sleep timing, but there is some support for later sleep timing among black study participants [22–24].
The second question this analysis addressed was whether increases in WFH explained the timing and duration effects on sleep. WFH increased greatly with the beginning of the pandemic, and then trended downward through 2023. However, it remained higher at the end of 2023 than it had been pre-pandemic. There were much greater income, education, and race/ethnicity disparities in WFH after the beginning of the pandemic compared to before the pandemic. However, WFH only modestly confounded the pandemic effect on sleep. This was consistent with research during the initial lockdown period that found the lockdown, rather than WFH, was a determinant of changes in sleep, using household electricity consumption in Singapore [14]. Alternatively, measuring the work-from-home effect by including the minutes of commuting time had similar confounding effects on the time-period indicator coefficient.
Finally, we examined whether the pandemic changed the income, education, or race/ethnicity effects on sleep timing and duration. Generally, these effects were not changed. The exception was the income effect on sleep timing. Before the pandemic, higher income levels were associated with earlier sleep timing. This effect was muted in the pandemic time period. We had expected that the pandemic would increase disparities in sleep, especially sleep duration, because higher socioeconomic groups would have more latitude to WFH relative to lower socioeconomic groups and therefore more time to set aside for sleep, but we did not observe that. We did observe that the increase in WFH was strongly socially patterned by race/ethnicity, education, and income. However, WFH did not explain much of the pandemic effect on sleep timing and duration. Rather, most of the pandemic effect on sleep was independent of the WFH or commuting time effects. Nor has the pandemic effect of later wake-up times and bedtimes and longer 24-hour sleep durations returned to the pre-pandemic levels by the end of 2023. The pandemic seems to have had, for now, enduring effects on sleep among working adults.
There are both strengths and limitations to these data and thus our analyses. One strength is the size and generalizability of the sample itself. The ATUS is a national probability sample with monthly data collection, so it is possible to observe and model trends over time. There are sufficient numbers to assess the effects of standard socioeconomic factors. Twenty-four-hour time diaries were historically motivated by questions related to labor economics, and the data are well suited to identifying work characteristics, including days with work, places of work, and commuting times. Using such time diaries to examine sleep is less common, although a number of studies have done this [25–29]. The key advantage over survey questions that ask about sleep duration is that there is relatively little reason to suspect reporting biases that arise from respondents gravitating toward normative or socially desirable responses, since bedtime and waking time do not have widely held “normal” or “desirable” times. Sleep timing deduced from a 24-hour time diary is similar to the data that would be collected in a sleep log, which has been widely used in sleep research. What the time diary indicates as a sleep activity is not the same as a physiological sleep measure and the correlation between the two is unknown. However, there would need to be trends in the correlation itself for there to be a systematic bias in the comparisons made here over time or between groups, since the sleep variables are consistently identified throughout. Since we only have a single 24-hour time period for each respondent, we cannot assess the difference between workdays and non-workdays, nor can we be sure that changes in sleep after the beginning of the pandemic are not associated with possible but unknown changes in who was in the workforce over time. However, the sociodemographic characteristics of the respondents are very similar in the two time periods, with just the income increasing, which could be expected due to some inflation.
Conclusion
These data suggest that population-level changes in sleep timing and duration for working adults changed with the beginning of the COVID-19 pandemic, and these changes of longer time set aside for sleep and later waking times have persisted through 2023. Higher education and higher income groups, as well as the Asian and white race/ethnicity groups, greatly increased WFH with the pandemic, and WFH was strongly associated with longer sleep and later sleep timing. Nonetheless, these growing disparities in WFH and commuting time did not translate into increased disparitites in sleep timing or duration. More broadly, these data underscore that sleep patterns can change over time, in response to social, cultural, and economic context.
Supplementary Material
Acknowledgments
The authors thank Don Hedeker, PhD for statistical advice.
Contributor Information
Evan Mathura, The College, University of Chicago, Chicago, IL, United States.
Diane S Lauderdale, Department of Public Health Sciences, University of Chicago, Chicago, IL, United States.
Author contributions
Evan Mathura (Conceptualization [supporting], Data curation [lead], Formal analysis [equal], Funding acquisition [supporting], Investigation [supporting], Methodology [equal], Project administration [supporting], Resources [supporting], Software [lead], Supervision [supporting], Validation [lead], Visualization [lead], Writing—original draft [equal], Writing—review & editing [equal]), and Diane S. Lauderdale (Conceptualization [lead], Data curation [supporting], Formal analysis [equal], Funding acquisition [lead], Investigation [lead], Methodology [equal], Project administration [lead], Software [supporting], Supervision [lead], Validation [supporting], Visualization [supporting], Writing—original draft [equal], Writing—review & editing [equal])
Funding
This work was supported by a University of Chicago Biological Sciences Collegiate Division Summer Research Fellowship to Evan Mathura.
Disclosure statement
Financial disclosure: None.
Non-financial disclosure: None.
References
- 1. Jackson CL, Redline S, Emmons KM. Sleep as a potential fundamental contributor to disparities in cardiovascular health. Annu Rev Public Health. 2015;36(1):417–440. 10.1146/annurev-publhealth-031914-122838. PMID: 25785893; PMCID: PMC4736723. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Koo DL, Nam H, Thomas RJ, Yun CH. Sleep disturbances as a risk factor for stroke. J Stroke. 2018;20(1):12–32. 10.5853/jos.2017.02887 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Andreucci MSA, Campbell P, Dunn KM. Are sleep problems a risk factor for the onset of musculoskeletal pain in children and adolescents? A systematic review. Sleep. 2017;40(7). 10.1093/sleep/zsx093 [DOI] [PubMed] [Google Scholar]
- 4. Lee E, Cho HJ, Olmstead R, Levin MJ, Oxman MN, Irwin MR. Persistent sleep disturbance: a risk factor for recurrent depression in community-dwelling older adults. Sleep. 2013;36(11):1685–1691. 10.5665/sleep.3128 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Grandner MA, Seixas A, Shetty S, Shenoy S. Sleep duration and diabetes risk: population trends and potential mechanisms. Curr Diab Rep. 2016;16(11):106. 10.1007/s11892-016-0805-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Monk TH, Buysse DJ, Rose LR, Hall JA, Kupfer DJ. The sleep of healthy people: a diary study. Chronobiol Int. 2000;17(1):49–60. 10.1081/cbi-100101031 [DOI] [PubMed] [Google Scholar]
- 7. Di H, Guo Y, Daghlas I, et al. Evaluation of sleep habits and disturbances among US adults, 2017-2020. JAMA Netw Open. 2022;5(11):e2240788. 10.1001/jamanetworkopen.2022.40788 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Bick A, Blandin A, Mertens K. Work from home before and after the COVID-19 outbreak. Am Econ J Macroecon. 2023;15(4):1–39. 10.1257/mac.20210061 [DOI] [Google Scholar]
- 9. Ceolin C, Limongi F, Siviero P, et al. Changes in sleep duration and sleep timing in the general population from before to during the first COVID-19 lockdown: a systematic review and meta-analysis. Int J Environ Res Public Health. 2024;21(5):583. 10.3390/ijerph21050583 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Robbins R, Affouf M, Weaver M, et al. Estimated sleep duration before and during the COVID-19 pandemic in major metropolitan areas on different continents: observational study of smartphone app data. J Med Internet Res. 2021;23(2):e20546 https://www.jmir.org/2021/2/e20546 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Sinha M, Pande B, Sinha R. Impact of COVID-19 lockdown on sleep-wake schedule and associated lifestyle related behavior: a national survey. J Public Health Res. 2020;9(3):1826. 10.4081/jphr.2020.1826 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Hooper M, Reinhart M, Dusetzina SB, Walsh C, Griffith KN. Trends in U.S. self-reported health and self-care behaviors during the COVID-19 pandemic. PloS One. 2023;18(9):e0291667. 10.1371/journal.pone.0291667 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Kecklund G, Akerstedt T, Lowden A. Morning work: effects of early rising on sleep and alertness. Sleep. 1997;20(3):215–223. 10.1093/sleep/20.3.215 [DOI] [PubMed] [Google Scholar]
- 14. Raman G, Peng JC. Sleeping behaviors are determined by lockdown and not work-from-home arrangements. Sci Rep. 2024;14(1):20465. 10.1038/s41598-024-70461-z [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Link BG, Phelan J. Social conditions as fundamental causes of disease. J Health Soc Behav. 1995;35:80–94. [PubMed] [Google Scholar]
- 16. Sarah M. Flood, Liana C. Sayer, Daniel Backman, and Annie Chen. American Time Use Survey Data Extract Builder: Version 3.2 [Dataset]. College Park, MD: University of Maryland and Minneapolis, MN: IPUMS; 2023. 10.18128/D060.V3.2 [DOI] [Google Scholar]
- 17. Li J, Vitiello MV, Gooneratne NS. Sleep in normal aging. Sleep Med Clin. 2022;17(2):161–171. 10.1016/j.jsmc.2022.02.007 [DOI] [PubMed] [Google Scholar]
- 18. van de Langenberg SCN, Kocevska D, Luik AI. The multidimensionality of sleep in population-based samples: a narrative review. J Sleep Res. 2022;31(4):e13608. 10.1111/jsr.13608 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Grandner MA, Williams NJ, Knutson KL, Roberts D, Jean-Louis G. Sleep disparity, race/ethnicity, and socioeconomic position. Sleep Med. 2016;18:7–18. 10.1016/j.sleep.2015.01.020 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Ruiter ME, DeCoster J, Jacobs L, Lichstein KL. Normal sleep in African-Americans and Caucasian-Americans: a meta-analysis. Sleep Med. 2011;12(3):209–214. [DOI] [PubMed] [Google Scholar]
- 21. Lauderdale DS, Knutson KL, Yan LL, et al. Objectively measured sleep characteristics among early-middle-aged adults: the CARDIA study. Am J Epidemiol. 2006;164(1):5–16. [DOI] [PubMed] [Google Scholar]
- 22. Knutson KL, Reid KJ, Wong M, et al. Chronotype and sleep timing by race-gender: the CARDIA sleep study. J Biol Rhythms. 2025;40(2):171–180. 10.1177/07487304251315596 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Combs D, Hsu CH, Bailey O, et al. Differences in sleep timing and related effects between African Americans and non-Hispanic whites. J Clin Sleep Med. 2021;17(5):897–908. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Bailey O, Combs D, Sans-Fuentes M, et al. Delayed sleep time in African Americans and depression in a community-based population. J Clin Sleep Med. 2019;15(06):857–864. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Meyer JM. Sleep duration differences by education from middle to older adulthood: does employment stratification contribute to gendered leveling? J Health Soc Behav. 2024;65(2):182–199. 10.1177/00221465231199281 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Iao SI, Jansen E, Shedden K, et al. Associations between bedtime eating or drinking, sleep duration and wake after sleep onset: findings from the American Time Use Survey. Br J Nutr. 2021;127(12):1888–1897. 10.1017/S0007114521003597 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Yao CJ, Basner M. Healthy behaviors competing for time: associations of sleep and exercise in working Americans. Sleep Health. 2019;5(1):23–30. 10.1016/j.sleh.2018.10.001 [DOI] [PubMed] [Google Scholar]
- 28. Basner M, Dinges DF. Sleep duration in the United States 2003-2016: first signs of success in the fight against sleep deficiency? Sleep. 2018;41(4). 10.1093/sleep/zsy012 [DOI] [PubMed] [Google Scholar]
- 29. Antillón M, Lauderdale DS, Mullahy J. Sleep behavior and unemployment conditions. Econ Hum Biol. 2014;14:22–32. 10.1016/j.ehb.2014.03.003 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.