Reported sleep duration reveals segmentation of the adult life-course into three phases

Sleep duration varies widely within and between individuals and is a key determinant of health and cognition, influenced by genetic and environmental factors. Age explains a large share of variance in sleep duration, with adults sleeping less than children and younger adults often sleeping less than older adults. Prior research suggests national differences in reported sleep, with some Asian countries reporting shorter sleep. However, few large-scale epidemiological studies have examined general population sleep patterns across many nations; many past studies focused on specific age groups, clinical populations, specific sleep problems, or aggregated heterogeneous smaller studies, and most were conducted in high-income countries. Because sleep-related environmental factors (e.g., work time, artificial lighting) vary by development level, results may not generalize to low- and middle-income countries. This study aims to characterize global patterns of self-reported sleep duration across the adult lifespan, identify life-course phases in sleep duration, assess robustness across demographics and cultures, and examine associations between reported sleep duration and cognitive performance (spatial navigation).

The paper situates its work against prior reports of cross-national variation in sleep duration, including findings that some Asian countries report shorter sleep. Earlier epidemiological studies across nations were limited by modest samples, focus on particular age groups or clinical populations, or heterogeneity across combined datasets. Meta-analyses and reviews often centered on high-income countries, limiting generalizability to the global population. Environmental and socioeconomic factors known to influence sleep (e.g., work hours, artificial lighting) vary with development level, underscoring the need for large, standardized, cross-country datasets. Prior studies have also reported non-linear age associations in sleep duration and links between sleep and cognition, including inverted U-shaped relations with cognitive performance in older adults, but these findings were based on smaller, often WEIRD samples.

Data were collected via Sea Hero Quest (SHQ), a mobile video game designed to assess navigation ability, available in 17 languages for iOS and Android. Participants consented within the app and optionally provided demographics (age, gender, country; later, average sleep duration, education level, commute duration, and home environment). A total of 3,881,449 individuals played at least one level; 60.8% provided basic demographics, and 27.6% provided more detailed demographics including sleep duration. Inclusion and exclusion: participants older than 70 were excluded due to documented selection bias in older app users; participants reporting average sleep duration below 5 h (0.9%) or above 10 h (0.6%) were excluded; countries with fewer than 500 players were excluded to ensure robust within-country estimates. Final analytic sample: 730,187 participants from 63 countries. Demographics included 381,153 men and 349,034 women; mean age 38.71 years (SD 14.53); 526,170 with tertiary education, 204,017 secondary or less; 222,097 city and 508,090 non-city upbringing; commute durations: <30 min (291,822), 30–60 min (254,362), >1 h (183,764). Spatial navigation subset: N=418,152 who completed at least the first four wayfinding levels (levels 6, 7, 8, 11) and two training levels; only first attempts per level were analyzed. Navigation metrics: from 500 ms sampled trajectories, trajectory lengths (sum of Euclidean distances) were computed. Wayfinding performance (WF) was defined as the first principal component across trajectory lengths of the four wayfinding levels, each normalized by the sum of the two tutorial levels’ lengths to control for gaming skill (60.14% variance explained). WF was inverted and offset so higher values indicate better performance, then z-scored. Training performance (TP) was derived similarly from the two tutorial levels and z-scored. Change-point estimation: the relationship between reported sleep duration and age was segmented using a parametric global change-point detection (Matlab findchangepts with linear statistics), minimizing total residual error of piecewise linear least-squares fits with a penalty term for additional change points. Threshold tuning yielded two stable change points at 33 and 53 years across robustness checks. Statistical modeling: a linear mixed model (LMM) predicted reported sleep duration from age (linear), age squared, gender, education, home environment, commute duration, and interactions with age, with random intercepts by country: sleep ~ age*(gender + education + home_environment + commute) + age^2 + (1|country). For cognitive analyses, two ANOVAs predicted TP and WF with sleep duration (linear), sleep duration squared (to capture inverted U), and interactions with age groups (19–33, 34–53, 54–70), controlling for gender, education, home environment, and commute duration. Country-level analyses: computed country conditional modes (random intercept deviations from population-level fixed effects) from the LMM, correlated with raw national averages. Assessed clustering by assigning countries to 7 supra-national clusters based on historical/economic proximity and to cultural clusters from prior literature; significance tested via ANOVA including age and gender and by permutation shuffling country labels (100 iterations). Geographic associations were tested via regressions of country mean sleep duration (and conditional modes) on GDP per capita and absolute latitude. Ethics: UCL Ethics Research Committee approval (CPB/2013/015). Data and Matlab code are available at the provided OSF links.

• Global average self-reported sleep duration: 7.01 h (SD 1.07) across 730,187 adults in 63 countries. Women reported 7.5 minutes more sleep than men on average (Hedges’ g = 0.12, 95% CI [0.11, 0.12]).
• Life-course segmentation into three adult phases: sharp decrease from 19 to 33 years (e.g., ~7.4 h women, 7.3 h men at 19), slower decrease/plateau from 34 to 53, then increase from 54 to 70, reaching values similar to those at 25–30. Change points at 33 and 53 years for both sexes.
• Distributional shifts: frequency of short sleepers (5 h) increases with age; long sleepers (9–10 h) show a U-shaped association with age; sex differences larger among long sleepers.
• LMM results (effect sizes): age strongly associated with sleep (Hedges’ g = 0.49, 95% CI [0.47, 0.51]); age squared also strong (g = 0.50, 95% CI [0.48, 0.52]); gender modest (g = 0.12, 95% CI [0.12, 0.13]). Education overall weak (g = 0.059, 95% CI [0.054, 0.064]) but stronger among participants under 22 (g = 0.23, 95% CI [0.22, 0.24]), with tertiary education linked to shorter reported sleep. Commute >1 h associated with shorter sleep vs 30–60 min (g = 0.10, 95% CI [0.10, 0.11]); those commuting <30 min reported slightly less sleep than 30–60 min (g = 0.06, 95% CI [0.055, 0.066]). Home environment had a very weak association (g = 0.03, 95% CI [0.02, 0.03]).
• Cognitive association: wayfinding performance (spatial navigation) shows an inverted U-shaped relationship with reported sleep duration, strongest in ages 54–70, with optimal performance at 7 h in all age groups. Effects for WF (linear B ≈ 0.39, 95% CI [0.33, 0.46]; quadratic B ≈ -0.028, 95% CI [-0.033, -0.024]) exceeded those for training performance (motor; linear B ≈ 0.14, 95% CI [0.07, 0.22]; quadratic B ≈ -0.012, 95% CI [-0.017, -0.010]). Interactions of sleep duration terms with age group significant for WF but not TP.
• Cross-country variation: nearly 1-hour difference between some countries (e.g., Japan mean 6.63 h vs Albania 7.54 h; Hedges’ g = 0.79, 95% CI [0.68, 0.89]). Country conditional modes closely tracked raw averages (r = 0.97, p < 0.001). Reported sleep clustered into supra-national regions; ANOVA showed significant effects of global clusters (F(6,730181) = 936.97, p < 0.001), gender, and age; permutation tests confirmed clustering non-random.
• Geography and economics: positive correlation between national average sleep and absolute latitude (r = 0.52, p < 0.001); in regressions, both GDP per capita and latitude significantly predicted sleep duration using raw means and country conditional modes.
• WEIRD vs non-WEIRD: both groups exhibited similar U-shaped age patterns with comparable change points (WEIRD 33 and 55; non-WEIRD 31 and 52). No significant difference in mean sleep duration after controlling for GDP per capita and latitude (WEIRD status t(59) = 1.19, p = 0.24).

The findings delineate three robust adult life-course phases of self-reported sleep duration with change points at 33 and 53 years, consistent across sexes and across WEIRD and non-WEIRD populations. The decrease into mid-life aligns with increased childcare and work demands, while the increase after 53 likely reflects reduced caregiving and work pressures. The strong age and quadratic age effects, alongside modest influences of gender, education (especially in young adults), and commute, highlight the multifactorial determinants of sleep duration. Critically, reported sleep duration related to spatial navigation performance in later adulthood, showing an inverted U with an optimal at 7 hours; this effect was specific to navigation (not observed for motor training performance), suggesting domain-specific cognitive sensitivity to sleep duration in older adults. Cross-national clustering and the association with latitude and GDP indicate that cultural norms and environmental/geographic factors contribute to national sleep patterns. Together, these results extend prior smaller-scale, WEIRD-centric studies by providing a large, globally representative analysis linking self-reported sleep, age-related phases, culture/geography, and cognition.

This large-scale global study identifies two inflection points in adult sleep duration, segmenting the life-course into early adulthood (19–33), mid-adulthood (34–53), and late adulthood (54+), with a re-increase in reported sleep after mid-life. Spatial navigation performance shows an inverted U-shaped association with reported sleep duration, with 7 hours as a near-universal optimum across adult age groups and the strongest association in older adults. National sleep patterns are geographically clustered and related to latitude and economic factors, while the age-related pattern is remarkably stable across populations. Future research should incorporate objective sleep measures (e.g., polysomnography or wearables), assess sleep quality, include additional individual-level and contextual variables (e.g., ethnicity, socioeconomic status, health conditions, day/time of assessment, light exposure), and extend analyses to older ages beyond 70 where selection effects in app-based samples must be addressed. Leveraging large-scale digital platforms appears promising for globally generalizable sleep-cognition research.

• Self-reported sleep duration may diverge from objective sleep; concordance can vary by age, sex, and sleep disorders, potentially biasing associations.
• Important variables were not collected, including sleep quality, day of week, time of day of testing, and contemporaneous light levels, which can affect both sleep and cognitive performance.
• Potential unmeasured confounders (e.g., ethnicity, socioeconomic status, neurological/psychiatric conditions, inflammatory status) may influence sleep duration and cognition.
• Participants older than 70 were excluded due to selection bias in app users, limiting generalizability to the oldest adults.
• Smartphone/game-based sampling may underrepresent individuals with limited access to technology or lower digital literacy, potentially introducing demographic biases despite large sample size.
• Extreme reported sleepers (<5 h or >10 h) were excluded to reduce data entry errors and leverage consistent groups, which may remove true extremes and slightly narrow variability.