Gender differences in BaYaka forager sleep-wake patterns in forest and village contexts

Sleep studies in small-scale subsistence societies have broadened our understanding of cross-cultural sleep patterns, revealing the flexibility of human sleep. We examined sleep biology among BaYaka foragers from the Republic of Congo who move between environmentally similar but socio-ecologically distinct locations to access seasonal resources. We analyzed the sleep–wake patterns of 51 individuals as they resided in a village location (n = 39) and a forest camp (n = 23) (362 nights total). Overall, BaYaka exhibited high sleep fragmentation (50.5) and short total sleep time (5.94 h), suggestive of segmented sleep patterns. Sleep duration did not differ between locations, although poorer sleep quality was exhibited in the village. Linear mixed effect models demonstrated that women’s sleep differed significantly from men’s in the forest, with longer total sleep time (β ± SE =  − 0.22 ± 0.09, confidence interval (CI) = [− 0.4, − 0.03]), and higher sleep quality (efficiency; β ± SE =  − 0.24 ± 0.09, CI = [− 0.42, − 0.05]). These findings may be due to gender-specific social and economic activities. Circadian rhythms were consistent between locations, with women exhibiting stronger circadian stability. We highlight the importance of considering intra-cultural variation in sleep–wake patterns when taking sleep research into the field.

Gender-based differences in sleep ecology. Despite the biological necessity of sleep, individuals balance their sleep time with social and subsistence demands 22 . Humans incur sleep debt for increased opportunities of socialization, information gathering, caregiving, and work activities 17,23,24 . These activities are often rooted in social role expectations between men and women, leading to gender-differentiated sleep outcomes. For example, Maume and colleagues 25 , using data from the 2006 European Social Survey, found that having a child under 6 in the home significantly increased the odds of a woman sleeping restlessly, whereas it was not linked to men's sleep. This association may be different in cultural contexts in which both shared family sleep environments and some nighttime care by fathers is common.
Co-sleeping both with children and other adults is a valued cultural practice in many societies and is also often a necessity for households in which indoor sleeping space is limited, as is common among mobile foragers 6,26 . Among the Hadza of Tanzania, actigraphic analysis revealed that co-sleeping with a breastfeeding infant was not linked to altered sleep duration in mothers, but a higher number of co-sleepers was correlated with shorter sleep duration and quality 27 . However, breastfeeding was associated with earlier wake times 27 , which is consistent with findings from the United States 28,29 . Relatively little is known about fathers' nighttime caregiving 30 , but shared family sleep has been linked to lower testosterone levels in cosleeping fathers in the Gambia and the Philippines 31,32 . Additionally, and relevant to the present paper, BaYaka fathers are regularly involved in handson care of infants and young children 20,33 .
Beyond co-sleeping, gendered work and divisions of labor between men and women may also affect sleep, but this is little explored outside of industrial and post-industrial contexts. In all forager societies, men and women practice some specialization in different subsistence activities, with men often engaging in more hunting and women typically gathering more intensely, though considerable variation in these roles has been reported 34 . However, in Hadza adults, Samson et al. 5 found no overall difference in sleep patterns between men and women. Similarly, Yetish et al. 13 reported no difference in sleep duration in the Tsimane, though men exhibited more variability in sleep onset 17 . This may be linked to collaboration in subsistence activities. For example, in the BaYaka community surveyed here, Sarma et al. 35 found important complementarities in men and women's work-BaYaka men and women will work side-by-side in gardens, with men clearing plots and women planting. Unlike other societies where heightened gender distinctions in types and intensities of subsistence roles drive significant

Results
Descriptive sleep characteristics and Non-Parametric Circadian Rhythm Analysis (NPCRA) for this sample are presented in Table 1 (Table 2). Sleep values were similar between locations, apart from a higher probability of increased sleep efficiency in the forest (village = 65.1%, forest = 68.6%, Bayes Factor (BF) = 6.69) as well as decreased sleep fragmentation (village = 52.52, forest = 47.07, BF = 5.15). NPCRA analysis revealed that the BaYaka exhibited relatively consistent circadian rhythms between locations, with an overall interdaily stability of 0.55, intradaily variability of 0.11, and a relative amplitude of 0.86. While NPCRA analysis yielded weak evidence for a difference in intradaily variability between the forest and village locations (BF = 2.55), no other support for a difference in circadian rhythms was found.
To test gender-based sleep differences, we examined men and women's sleep between the village and forest locations. While there were no differences for men's sleep variables between the village and forest (all Bayes factors favored the null hypothesis), women's sleep substantially differed between the two sites. A strong association for differences in sleep duration and quality between locations was found; women in the forest had a greater probability of longer total sleep time compared to women in the village (village = 5.91 h, forest = 6.53 h; BF = 10.27), as well as longer TTST (village = 6.60 h, forest = 7.43 h; BF = 18.92). In addition, there was extremely strong evidence for higher sleep efficiency in the forest (village = 65.25%, forest = 71.43%; BF = 161.29), and strong evidence for lower sleep fragmentation (village = 53.28, forest = 45.58; BF = 8.72). This suggests more consolidated sleep patterns for women when they reside in the forest location as compared to the village.
When comparing gender-based sleep differences within each location (see Table 3), no differences were found between men and women in the village location, with the exception of weak evidence for longer total sleep time Table 1. Descriptive sleep quotas (n = 51, men = 23) and NPCRA analysis (n = 31, men = 17) of the BaYaka from actigraphic analysis. Overall age range is 17-72 years. Data are reported as mean (standard deviation). Nap period duration was calculated by subtracting TST from TTST. L5 is the activity during the 5 h of least activity during the diel cycle. M10 is the activity during the most active 10 h during the diel cycle. In other words, in the village, men go to bed and wake around the same time as the women. In the forest, however, men's sleep timing (both sleep onset and sleep end) is highly variable compared to women's sleep timing. NPCRA analysis between men and women (see Table 4) showed strong evidence that women had an increased probability of lower intradaily variability (women = 0.08, men = 0.14; BF = 30.93) as well as higher activity levels (Most active consecutive 10-h values, or M10) compared to men (women = 38,673, men = 26,539; BF = 19.76). However, the onset of men's M10 hours was significantly earlier compared to women (women = 07:42,  www.nature.com/scientificreports/ men = 07:14, p = 0.01). There were also slight differences in relative amplitude (BF = 3.02) and interdaily stability (BF = 1.66) between men and women. Linear mixed effects models also revealed significant associations between sleep and gender contingent upon location. With references set as women for gender and village for location for all models, an interaction between gender * location was negatively associated with TST (β ± SE = − 0. 22  We used functional linear modelling to demonstrate 24-h circadian activity patterns between men and women overall (see Fig. 3). Women exhibited significantly higher activity levels compared to men throughout most of the day. Figure 3 suggests that both men and women displayed a first peak of high-level activity in the midmorning; however, men's activity declined mid-morning and increased again just after noon (though to a lesser extent than their first peak), whereas women's activity levels plateaued in the mid-morning and declined slightly before increasing again to an even greater activity level in the afternoon. Women also exhibited lower activity levels around ~ 3:00 a.m. and a sharper incline of activity in the morning after waking (~ 6:00 a.m.) compared to men. Functional linear modelling of the forest and village locations (with no differentiation between men and women) was also conducted (Fig. 4). Results illustrated a significant difference between the locations in the evening hours, with more activity around midnight hours noted in the forest location as compared to the village. The forest location also displayed higher activity levels from ~ 9:00 a.m. until noon. Both locations displayed a similar pattern of a decrease in activity from mid-to late morning (although to a greater extent in the forest), followed by a short rebound of activity in the afternoon before the steady evening decline. While peak activity in the village occurred in the afternoon, the greatest activity in the forest occurred in the morning. Table 4. Descriptive sleep quotas (n = 51, men = 23) and NPCRA analysis (n = 31, men = 17) comparing women and men in the BaYaka study sample. Data are reported as mean (standard deviation). Bayes factor is reported to assess differences in sleep quotas between genders. Time value difference (sleep onset, sleep end, and M10 onset) is assessed using a t-test.

Discussion
This study examined sleep and circadian rhythms in a BaYaka community that resides in and moves between two socio-ecologically distinct locations. Our findings indicate that BaYaka sleep and circadian rhythmicity are highly entrained to the physical environment and remain consistent even with changes in location. We found moderate support for the hypothesis that sleep is of poorer quality in the village compared to the forest location.   www.nature.com/scientificreports/ Additionally, our hypothesis that sleep would be relatively similar between men and women was only partially supported. While sleep variables were similar in the village location, we found significant gender-based sleep differences in the forest location, pointing to the potential influence of gender-specific social and labor activities on sleep patterns within the same population. Our prediction that BaYaka would exhibit stronger circadian rhythmicity in the forest location was not supported-NPCRA analysis revealed consistent circadian rhythms irrespective of location, with no differences in relative amplitude (rhythm strength) and interdaily stability (rhythm consistency), and only a slight difference in intradaily variability (rhythm fragmentation). While sunlight and temperature are relatively similar in both locations, the differences in built environment between the forest and village were predicted to exert a strong enough effect to alter circadian rhythms between locations. However, differences in location-specific domicile construction did not appear to alter circadian rhythms, likely because this community continues to spend much of their waking time outside. Our findings for stable circadian rhythms in this community aligns with the circadian entrainment hypothesis proposed by Samson, Manus and colleagues 3 , where high levels of exposure to the physical environment (in particular, sunlight and outdoor temperature) during waking hours leads to greater www.nature.com/scientificreports/ levels of circadian stability and uniformity as compared to post-industrial populations that largely use artificial lighting and temperature control. There was moderate support for our prediction that sleep would be adversely affected by being in the village location. BaYaka exhibited lower sleep efficiency in the village (65.1%) compared to the forest (68.6%), with higher sleep fragmentation in the village. Based on ethnographic observation, we believe this may be due to the higher population density, greater noise, and more social interaction within the village. Interestingly, sleep duration did not substantially differ between the two locations, nor did average sleep time (both onset and end). BaYaka sleep end corresponded roughly to dawn; early wake times have been described in populations exposed to high levels of natural light throughout the day, which synchronizes their circadian clock with solar time 2 . In particular, a morning dose of bright sunlight is considered the most potent circadian entrainment factor for sleep timing 1 . Sleep onset occurred approximately 3 h after sunset, corresponding with use of fires in the evening hours to increase work and social opportunities-echoing findings of populations in similar subsistence settings by Yetish and colleagues 13 among San peoples, and Beale and colleagues 16 working in a rural community in Mozambique.
Associations between sleep quality and environmental conditions were found at both locations. Nocturnal temperature and humidity exerted a significant negative influence on sleep efficiency, with higher sleep fragmentation. As BaYaka domiciles in this community do not have doors, thermal stress could be reducing sleep quality-a similar finding to that of Samson, Crittenden and colleagues in their study with Hadza 5 . Total sleep time did not appear to be altered by any of the environmental factors examined; however, periods of increased rainfall influenced both TTST and sleep fragmentation, suggesting an increase in napping during rainy days, as www.nature.com/scientificreports/ people remain in-doors and tend to engage in fewer activities. Forest camps are in walking distance of the village, and thus it is unlikely that temperature, rainfall and humidity significantly differ between the two locations; however, future research that monitors specific environmental conditions within each location may provide more detail on the influence of environmental conditions on sleep behavior in this community. We found mixed results for our hypothesis that BaYaka would exhibit minimal gender-based differences in sleep quotas or circadian rhythms. No strong differences in sleep variables between men and women were found in the village location, but in the forest camp, women exhibited longer total sleep time and higher sleep efficiency than men. Overall, women also exhibited less circadian fragmentation than men, and slightly stronger and more stable circadian rhythms. It is possible that, despite labor coordination between BaYaka men and women 20,35 , gender-differentiated work and social activities, particularly in the forest location, may drive gender differences in sleep expression and in circadian rhythms. Functional linear modelling analysis revealed that women had significantly higher activity levels than men, with a sharp rise in activity almost immediately after waking, which remained higher than men for the remainder of the day. Men's activity levels, comparatively, were lower than women's, with a slower activity increase in the morning. This slower slope in men's morning activity is potentially due to the higher activity period for men throughout the night; men participate in nighttime socialization, hunting and fishing, which may be skewing the finding of an earlier Most Active 10 h (M10) in men (07:15) compared to women (07:42).
Conversely, women exhibited longer, higher quality sleep throughout the night in the forest location. This is surprising, given that a total of 5 women in the forest were breastfeeding and cosleeping with infants under two years old, which has been linked to more frequent nighttime arousals in mother-infant dyads in the global North 22,26,37 . The proportion of breastfeeding women (42% of women in the village, and 38% of women in the forest) are roughly equivalent, and thus likely do not explain the locational gender-based differences in sleep duration and quality. Sarma and colleagues 35 reported that BaYaka women in the village engaged for longer periods in moderate to vigorous subsistence activities compared with men. While BaYaka women work consistently over longer periods of the day, men participate in inconsistent spurts of high energy activity, such as hunting large game or climbing trees 35 . In the forest camps, where energetics of the BaYaka have not yet been reported, it may be that gender-based differentiation of forest hunting and foraging tasks drives the gender-differentiated sleep patterns observed in this study. While our study did not find differences between men and women's sleep in the village (apart from a slight difference in total sleep time), women's work activities in the forest may substantially differ from that of the village, creating stronger labor demands and thereby affecting sleep behavior. While women maintain gardens and continue to forage when living in the village, the forest portion of our study was conducted during caterpillar and bail-fishing seasons, leading to a potential increase in the intensity and duration of women's work. Specifically, women travel throughout the day between caterpillar trees, and, during bail-fishing, perform the majority of the digging as they change the course of streambeds to isolate fish for capture.
Activity patterns related to social space appear to have affected sleep timing variability. BaYaka camps are small and close-knit, particularly in the forest location. This social density may increase the likelihood that the nighttime activities of some individuals disrupt or fragment the sleep of others, or individuals may choose to join an activity if it is of interest to them. BaYaka women had more consistent wake and bedtimes in this study; this aligns with findings from studies in post-industrial societies 38 , and can be attributed to morning household responsibilities and childcare. Conversely, men displayed a high variance in both sleep onset and end, with an especially high standard deviation in sleep end in the forest camp. This is in contrast to the Tsimane horticulturalists of Bolivia, where sleep end was relatively stable from day-to-day 17 . Thus, the social influence on sleep timing may also influence circadian rhythm expression-the high variability in sleep timing in men may contribute to their higher circadian fragmentation and lower stability of circadian rhythms compared to women. BaYaka men opportunistically hunt and fish at night providing there is sufficient natural or artificial illumination-potentially reflecting the negative association between sleep efficiency and moon phase that we have documented elsewhere 39 . Our finding of high levels of sleep variability in men but not in women aligns with the hypothesis posited by Maume et al. 22 , whereby women in post-industrial societies often trade-off socialization opportunities for increased sleep time, while leisure and social time is often more incorporated into men's evening activities. Ethnographic observations confirm that BaYaka men preferentially choose to socialize at night in preparation for hunting or fishing, particularly in the forest location. In addition, men's socialization in the village location often involves alcohol ingestion; while this has demonstrated effects on sleep quality 40,41 , this was not demonstrated in our study, as no difference in sleep variables between men and women in the village location was found.
Overall, the BaYaka community in this study exhibited similar sleep patterns to Hadza 5 , Tsimane 13 , and San 13 communities, adding to and complementing the limited knowledge of sleep patterns in subsistence foraging societies. Nighttime sleep duration in all four populations averaged less than 7 h, notably less than the 8 h of monophasic sleep typically recommended by clinicians in Euro-American post-industrial populations 42 . Like Hadza 5 , low sleep duration among BaYaka occurs despite a high amount of time in bed during the night (8.95 h on average).
The high level of sleep timing variability in this BaYaka community, combined with high levels of sleep fragmentation, and longer 24-h total sleep time compared to nighttime sleep duration is suggestive of segmented sleep patterns. While a single, consolidated sleep bout is considered the healthy standard in post-industrial nations 42 , segmented sleep patterns were common in both European and equatorial populations prior to the widespread use of electricity in the twentieth century 43 . Plasticity in sleep and circadian expression is an essential feature of adaptation to different ecologies, and has been found to be more pronounced in urbanized populations exposed to electric light 15 . Our study suggests that subsistence and social factors may also drive high sleep flexibility, even when daylength is relatively stable and access to electrical light is minimal. The finding of sleep/ wake variability, particularly between locations, in association with consistent circadian rhythms may have implications for sleep and health disparities. In post-industrial society, negative health consequences reported www.nature.com/scientificreports/ with short sleep duration and high sleep variability are often exacerbated by circadian misalignment, increasing inflammation and diabetes risk 24,44 . To date, no study has examined the influence of sleep and circadian variables on health outcomes in foraging societies. Future research into associations between sleep timing variability, circadian rhythms, and health outcomes in small-scale subsistence societies could explore this possibility. For example, it is possible that circadian rhythms that are more closely aligned with the physical environment may be somewhat protective against the negative health consequences of poor sleep. In addition, future research that incorporates co-sleeping (particularly the number of co-sleepers), as well as variables such as nursing and pregnancy could elucidate further gender-based differences in the sleep behaviors of BaYaka adults. Interestingly, our study had two noted differences in comparison to the studies of the San and the Hadza. Our finding of gender-based sleep differences occurring within different residential locations has not been previously described in forager societies and highlights the importance of an understanding of sleep behaviors in varying contexts within a single community. In addition, circadian rhythms in this BaYaka community were stable and consistent, even more so than the circadian rhythms studied in the Hadza 5 . The intradaily variability of circadian rhythms in the BaYaka (0.11) is the lowest circadian fragmentation reported in small-scale, subsistence populations thus far 3,12 , and highlights the wide physiological variability in circadian rhythmicity, and the flexibility of these rhythms in response to social and ecological factors.

Conclusion
We examined sleep and circadian rhythms among BaYaka foragers inhabiting small forest camps and a relatively more populous village settlement-settings that are in relatively close geographical proximity and thus share environmental conditions, but socio-ecologically differ. Our results demonstrated gender differences in sleep behavior that we argue may be due to differences in subsistence activities and social opportunities between the two locations. Our findings add to our understanding of sleep in populations who are minimally exposed to artificial light and controlled, uniform temperatures. Like other subsistence societies, BaYaka participants showed stable circadian rhythms, short sleep durations, and relatively low sleep efficiency. Cross-cultural sleep research grounded in socioecological frameworks is needed to illuminate not only the plasticity of human sleep and its adaptive functions, but to provide further context for understanding both sleep needs and sleep disorders across the globe 4,45 . Our investigation of sleep among BaYaka adds to this knowledge by revealing how certain sleep profiles may be expressed in the same population under different conditions. In turn, understanding the relationship between socio-ecology and sleep patterns adds to the cross-cultural exploration of sleep, and provides insight into human variation of sleep and circadian biology.

Methods
Description of study site. BaYaka who participated in the present study are mobile forest foragers who spend half the year in a multi-ethnic village setting, and the other half living in small forest camps 46 . Forest camps have a population density of 7-43 individuals with the mean area per person averaging 11.5 m 2,47 . These camps are typically oriented around a core group of siblings, their spouses, children, and elders. In contrast, in the village these dispersed camps coalesce into a larger community of ~ 200 people situated in an ethnicallysegregated neighborhood adjacent to those of Bondongo fisher-farmers who occupy the village year round 46 . In both settings, BaYaka primarily rely on forest resources such as wild yams (e.g. Dioscoera sps.), edible leaves, mushrooms, fish, and wild game. Many BaYaka also keep swidden garden plots in the forest near the village where they grow plantains and manioc 46 . In the village they also trade forest products or labor for garden produce to the Bondongo for market goods (e.g. salt, cigarettes, Maggi bouillon cubes), alcohol or, occasionally, cash. Within the domain of subsistence, BaYaka maintain a gender-based division of labor 20,34,35 , although gender roles are conceived of as complementary 48,49 . Men hunt and climb trees to collect palm nuts and palm wine, whereas women perform bail fishing, and do a majority of gathering and food preparation tasks. In general, these gendered subsistence roles are collaborative and can be flexible 20,35 . Men and women will cooperate in barrier fishing and garden work, for instance. Additionally, BaYaka men provide hands-on infant and childcare 20 .
Data collection. Data collection occurred in two field seasons. In the village, data was collected in July through September 2017. Data from the forest camp was collected in July through September 2018. Overall, 39 participants took part in the village portion of the study (men = 20), and 23 individuals took part in the forest portion of the study (men = 10). A subset of individuals (n = 11; men = 7) participated in both the forest and village data collection. Throughout the study period, 11 women had a baby under 2 years who breastfed and co-slept with the mother (village = 8, forest = 5; 2 women participated in both field seasons). As BaYaka do not record their age in years, age data was estimated following methods by Diekmann et al. 50 . Ages in this study ranged from 17 to 72 years (mean = 36, standard deviation = 13.75). Participants wore actigraphy watches for 2 weeks each season. A total of 362 nights of sleep data was analyzed in this study (village = 226 nights, forest = 136 nights, range = 3-8 days per participant).
Average daily temperature, humidity, and rainfall data, as well as sunrise and sunset times, were retrieved from WorldWeatherOnline.com (2019) for each sleep night, from a weather station in Impfondo, Republic of Congo. We averaged the data for temperature and humidity four times each night (at 21:00, 0:00, 03:00, and 06:00) to obtain mean nightly averages for each day of the data collection period. Data on nightly lunar phase was obtained from the Astronomical Applications Department of the United States (http:// aa. usno. navy. mil/ data).
Actigraphy. Motionwatch  www.nature.com/scientificreports/ making it burdensome and difficult to implement in non-laboratory settings 9 . Actigraphy devices are better suited for foraging communities as they are non-invasive and unobtrusive for participants 9 . These devices have been validated against polysomnography for investigating sleep in field environments and allow for larger field sampling, as well 9,51 . All watches were set to record activity data in 1-min epochs, a common duration used to record human actigraphic sleep 9 . These devices have an algorithm that detects and differentiates periods of wakefulness and inactivity. The software then translates these into sleep and wake periods based on a threshold level 9 .
Statistical analyses. The CamNtech Motionware analysis program was used to score sleep data. The data was then compiled in Microsoft Excel and analyzed in R 52 . Mean averages of BaYaka sleep variables (men and women, both overall and within each location) were compared using Bayes levels of significance (R package BayesFactor), controlling for repeated nights of the same subject. Though the current standard of statistical analysis is frequentist testing (which reports rejection of the null using p values), a benefit of Bayesian statistical analysis is that Bayes quantifies both the alternative hypothesis (difference between two samples) and the null hypothesis (no difference between two samples) 53 . Bayes testing provides the probability of support for either hypothesis, rather than using a single point estimate to reject the null 53 . A higher, positive Bayes factor provides stronger support for the alternative hypothesis, while a higher negative factor provides stronger support for the null hypothesis. Interpretation of Bayes factor for the alternative hypothesis is adapted from Raftery 54 , using the terms "weak" (Bayes Factor = 1 to 3), "modest" (BF = 3 to 10), "strong" (BF = 10 to 30), and "very strong" (BF > 30) 53 . Non-Parametric Circadian Rhythm Analysis (NPCRA) was used to assess circadian rhythms between the forest and village, as well as between men and women. In addition, NPCRA analysis calculates the most and least active periods of individuals studied. As seven consecutive nights of sleep data are required for NPCRA analysis, 31 individuals (n = 16 village, n = 15 forest; n = 16 men, n = 12 women) are represented. Using NPCRA analysis, we measured location and gender differences in interdaily stability, intradaily variability, M10 values, L5 values, and relative amplitude using Bayes factors. Interdaily stability is the similarity of activity patterns from day-to-day, which range from 0 to 1, where 0 indicates no rhythm and 1 is total stability of rhythm. Intradaily variability is the degree of fragmentation in activity-rest periods which ranges from 0 to 2, where higher values indicate higher fragmentation. Most healthy adults will fall < 1. M10 values give the average activity level for the 10 most active hours in a 24-averaged period. M10 indicates how regular and active are the wake periods. L5 values are the least 5 active hours in averaged 24-h periods. L5 indicates how regular the inactive sleep periods are. Relative amplitude is calculated by dividing amplitude by the sum of L5 and M10 and ranges from 0 to 1, with 1 indicating a higher amplitude circadian rhythm. Additionally, the variability of sleep timing (sleep onset/ sleep end) was compared between men and women in each location using an F-test of variance.
Four linear mixed effects models were constructed using the lme4 package in R to test ecological predictors on sleep. Outcome variables for each of these models were (1) total sleep time (TST) defined as the amount of time scored as the main sleep bout in hours, (2) 24-h total sleep time (TTST) defined as the total sleep time across a 24 h period in hours, including sleep and daytime naps, (3) sleep efficiency, defined as the time asleep expressed as a percentage of time in bed, (4) sleep fragmentation, an index of the degree of fragmentation within the sleep period, and measured as the sum of mobile bouts (expressed as a % of inferred sleep time) and immobile bouts ≤ 1 min (expressed as a % of the total number of immobile bouts). Both sleep efficiency and sleep fragmentation are common quantitative measurements to assess sleep quality. Reference variables in all models were set as women for gender and village for location. An interaction effect between gender and location was included to test gender-based sleep differences between the village and forest locations. Environmental variables that are known to influence sleep (temperature, rainfall, humidity, and moon phase) were controlled for in the models; these predictor variables were scaled (mean-centered) for comparability with each other. Additionally, we included a random effect for subject to control for repeated night measures. Thus, the following model was used: Using the MultiModel Inference (MuMIn) package in R 55 , we averaged linear models with Akaike Information Criterions (AIC) < 10 and reported the importance of each factor within a 95% confidence envelope. To make Bayesian statistical inferences, we then used a shrinkage method to obtain the conditional model averages; shrinkage uses pooled information from more certain estimates of the regression model in order to improve less certain estimates. Pooling, in this instance based off a normal distribution, means that a category provides information that can be used to improve the estimates for all other categories 56 . For our results, we report the standardized coefficients and 95% confidence intervals for each factor.
Lastly, functional linear modelling (FLM) was employed to compare minute-by-minute individual sleep-wake activity patterns across two consecutive days. Two functional linear models were created: one comparing activity patterns in the forest and village locations, and one comparing activity patterns between men and women. FLM is a method developed specifically for actigraphy data that allows time-series data analysis over a 24-h period 57 . FLM uses raw actigraphy counts for each individual to avoid a masking of differences between groups that can sometimes occur with summary statistics 57,58 . FLM analysis was conducted using a nonparametric permutation test method in R using the package actigraphy 59 . A point-wise test (with 500 permutations) was used for the FLM models with a significance level of 0.05.

Data availability
The datasets generated during the current study are available from the corresponding authors on reasonable request.