Study location and participants

Members from a Kaqchikel Maya community living in a small agricultural town in the highlands of western Guatemala were recruited to participate in this study. Houses in this town are built close to each other around a main square that includes the municipal building, a health post, a school and a pila (a communal water source and water basins for clothes washing). The town is surrounded by agricultural land, as small-scale agriculture was the main economic activity until recently. Most houses in the community are built with adobe or concrete blocks and have tin roofs, with windows covered by wood panels or curtains (Fig 1). In general, houses do not include technological means for temperature regulation; hence, this population tends to be more exposed to meteorological changes than those in urban contexts with access to air conditioning or central heating. This community is undergoing an economic and technological transition, but access to electricity is not constant and powered lighting or appliances such as televisions vary among households.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Houses in the Kaqchikel Maya community.

The tin roofs that cover most houses are visible, and both adobe and concrete building material can be seen.

https://doi.org/10.1371/journal.pone.0277416.g001

Many families are still involved in subsistence-level economic activities including farming and fishing. However, as the transition to a market economy advances and subsistence pattern shifts, both men and women are now increasingly working for comparatively wealthier national and international chalet owners that populate the Lake Atitlán region, now a popular tourist destination. As a result, social and economic stratification is advancing. Traditional agriculture has in many cases given way to weekend agriculture as a source of supplemental income. Most economic activities (agriculture, fishing, commerce, building and services) are manual and, like most social activities, diurnal. That said, men do spend time out with friends in the evenings and there are church activities and community committees that take place after work. On weekends, people go to the market, visit with family, and devote time to religious activities. Over the last two decades, men’s agricultural roles have declined as some have found jobs in construction and house maintenance for chalet owners. Women are also increasingly working outside the home as house cleaners, babysitters, hotel workers, and makers and sellers of handicrafts such as weaving, jewelry, and beadwork to visitors and re-sellers. In addition to following traditional divisions of labor by gender, traditional gender differences are pronounced in social activities as well, with men often devoting more time than women to nighttime socialization starting in late adolescence.

Data were collected from January to April 2017 for a total of 464 nights of data from 37 participants, 14 men (mean age = 30.71 years, range = 18–45 years, nights of data = 85) and 23 women (mean age = 45.76 years, range = 34.4–62 years, nights of data = 379). Nine men (mean age = 32.44 years, range = 23–45 years) and 20 women (mean age = 46.62 years, range = 34.4–62 years) had at least seven days of complete 24-hour data, and were thus included in Nonparametric Circadian Rhythm Analysis (NPCRA).

Informed consent

Informed consent was obtained from all participants prior to the start of data collection. Participants were informed about the study protocols and goals in their native Kaqchikel language by Kaqchikel-speaking field assistants. They were informed that participation in the study was voluntary, and that they could withdraw from the study at any time. Those who agreed to participate indicated their consent with an X, thumbprint, initial, or signature (personal preference). Study protocol followed Simon Fraser University Research Ethics Board, SFU-REB protocol 2016s0577 and the University of Toronto, Mississauga Research Ethics Board, UTM-REB protocol 35696.

Equipment

Sleep-wake data were obtained with the CamNtech Motionwatch 8, a wrist actigraphic device that provides measures of sleep and wake activity. Polysomnography, which measures brain activity to distinguish sleep and wake activity as well as sleep stage, is still considered the gold standard in sleep studies in terms of reliability and accuracy. However, polysomnography often necessitates laboratory-based study, with invasive equipment attached to the participant. Actigraphs are noninvasive, and do not interfere with daily activity. Using a built-in accelerometer, actigraphs quantify movement in one-minute epochs; these data are then algorithmically translated to a binary sleep-wake determination. Using the high-sensitivity setting to determine sleep, our wake threshold was defined as a period of 20 consecutive bouts that were categorized as wakefulness. To assist in actigraphy scoring accuracy, participants were asked to press an event marker button upon going to bed and getting up, which increased reliability in determining sleep onset and offset. Actigraphs are validated against polysomnography at 96.5% sensitivity and 86.3% accuracy [30]. Specific validation of the CamNtech Motionwatch 8 demonstrates its reliability in estimating sleep parameters including sleep duration and efficiency [31].

Statistical analysis and hypothesis testing

Raw actigraphy data were processed and compiled using CamNtech Motionware analysis software, and analyzed using R Core Team [32]. To test the circadian entrainment hypothesis, we used NPCRA, which is a method of assessing the day-night rhythm in activity level [33] and is used to generate individual values for relative amplitude (RA; a measure expressing the ratio of inactivity during sleep periods and activity during wake periods, expressed as a range of values between 0 and 1 with higher values indicating higher rhythm amplitude), interdaily stability (IS; the similarity of activity patterns on different days, expressed as a range of values between 0 and 1 with 0 indicating a complete lack of rhythm and 1 indicating complete stability of rhythm), and intradaily variability (IV; a measure of fragmentation of rest and activity periods, expressed as a value between 0 and 2 with higher values indicating more variability in rhythm) [34]. Calculation of RA, IS, and IV provide a quantitative description of the amplitude (i.e., strength), stability, and variability of the rest-activity rhythm [33]. These non-parametric variables infer key characteristics of circadian rhythm, including synchronization with a zeitgeber [17]. Additionally, these non-parametric rest-activity variables have been used previously to infer and describe circadian rhythms in nonindustrial populations [16,22], as well as in clinical assessment of circadian function [35]. It is recommended that a minimum of 7 days of complete 24-hour actigraphy data be used to compute NPCRA [34]; we therefore focused on the 29 Maya participants who satisfied this requirement.

Using unpaired, two-sample Student’s t-tests or Mann-Whitney U tests, we compared the Maya participants’ sleep and rest-activity rhythm averages with those from participants in Madagascar and the United States. Comparison between Malagasy participants and industrial populations have been previously reported by Samson and colleagues [16]. Malagasy participants are small-scale, rural agriculturalists who live in a community in northeastern Madagascar. This community has no electric infrastructure, and houses are exposed to environmental temperature fluctuation and nighttime noise. Therefore, Malagasy sleep is considered representative of a group living with minimal exposure to the effects of electricity and in-home temperature and noise buffering. We used comparative sleep data from a study conducted in July and August 2015 and 2016 from 21 Malagasy individuals (female mean age = 40.30 years; male mean age = 36.80 years; age range = 19–59 years), published by Samson and colleagues [16]. Written informed consent was obtained from participants following the protocol provided by the Duke University Institutional Review Board for human subjects research. All participants were informed of the study objectives and procedures. Upon obtaining written consent, they were instructed to wear the watch continuously for the duration of the study period and asked to use the event marker button to indicate when they went to bed and got up. Movement was measured in 60-second intervals and scored using CamNtech software [16].

For comparative data from individuals living with continuous, 24-hour access to electricity characteristic of industrial populations, we used data from the Midlife in the United States (MIDUS) cohort. The MIDUS project recruited participants from throughout the United States, who were interviewed by phone and self-administered questionnaire, answering questions related to demographic variables, health, employment, and psychological factors [36]. Actigraphy data are available in the MIDUS II and MIDUS Refresher datasets. MIDUS II was conducted in 2004 and included follow-up of the data collected in MIDUS I as well as cognitive, neurological, and comprehensive biomarker assessments, and sleep data from subsamples of respondents [37]. MIDUS Refresher data were collected between 2012 and 2016 and include comprehensive biological assessments of 3,577 respondents [38]. Actigraphy data were generated with the Mini Mitter Actiwatch-64 (Philips Healthcare, Amsterdam, Netherlands) activity monitor. Participants were instructed to wear the watches continuously for seven consecutive days, with movement collected in 30-second intervals. Data were scored and analyzed using Aciwatch software [39]. Protocol for data collection and written informed consent was approved by the Education and Social/Behavioral Sciences and the Health Sciences Institutional Review Boards at the University of Wisconsin-Madison.

We selected MIDUS participants with ages between 25 and 62 to match the Maya sample, and excluded participants who reported sleep disorders and those taking sleep medication more than once per week. After excluding based on these criteria, 289 individuals (mean age = 47.07 years; age range = 25–62) were included for sleep comparisons, with 276 (mean age = 47.08 years; age range = 25–62) of these having circadian rhythm data for rest-activity comparisons.

To date, relatively few studies have reported population-wide sleep characteristics that include both sleep and rest-activity rhythm quotas; therefore, we chose these three populations as example groups to better contextualize Maya sleep and rest-activity rhythms. Once these data become available, future work should perform formal comparative analysis across cultures (especially subsistence strategies). Although it is possible that the chosen populations are not representative of their general demographic and geographic characteristics, they still reveal the capacity of the population in question [40].

To test the gendered sleep hypothesis, we used the lme4 package in R to model data using linear mixed effects models, controlling for temperature, rainfall, and humidity. We included temperature in our models because of the reported associations between temperature and sleep in small-scale subsistence groups [8,22]. While rainfall and humidity are less well reported in the sleep literature, these variables have the potential to influence comfort during the sleep period and thus to affect sleep patterns, particularly in sleep spaces where there is minimal buffering from changing environmental conditions [23,34].

Temperature, humidity, and rainfall measures were obtained using Historical Forecast Weather (HFW) data downloaded from World Weather Online (https://www.worldweatheronline.com/hwd/hfw.aspx). HFW data are generated using raw data from world weather agencies and then applying factors such as terrain and altitude. HFW provides hourly weather forecasts, and distinguishes between daytime and nighttime readings. After excluding daytime measures, we calculated nightly averages for temperature and humidity and nightly cumulative rainfall using hourly HFW readings between 19:00 (7:00 pm) on the day of interest and ending the following day at 5:00 (5:00 am), which are the hours categorized as night by World Weather Online.

We built three separate linear mixed effects models to investigate the predictors of sleep duration, sleep efficiency, and central phase measure (e.g., Sleep duration ~ Age*Gender + Rainfall + Humidity + Temperature + 1|subject), with subject as a random effect for individual participant to control for repeated measures [34]. Male was set as the reference gender for all models. Sleep duration, or total sleep time, is the amount of time scored as sleep on actigraphic analysis. Statistical analysis and results are reported as total sleep time in hours. Sleep efficiency is the percentage of time spent asleep out of the total time in bed in the period scored as sleep on actigraphic analysis and is one of the quantitative measures that contributes to sleep quality [34]. Central phase measure indicates the midpoint of the sleep period as a minute value before or after midnight and is used to assess circadian timing of sleep [34]. Values can be positive or negative, with higher values indicating a later sleep midpoint and a more evening-preferring chronotype, and lower values indicating an earlier sleep midpoint and greater morning preference. Models were averaged using the MuMIn package in R, which pools information from more certain estimates in the model to improve less certain estimates [41]. Along with estimate, standard error, and confidence interval, we report the importance of each predictor in our averaged model. Importance is reported in a range of 0 to 1.

Sleep timing consistency was compared by measuring variance in consistency of sleep onset (i.e., falling asleep) and offset (i.e., waking up), using an F-test of variance between men and women. Men’s and women’s sleep averages were compared with the BayesFactor package in R, using default priors [42]. This Bayesian approach (e.g., ttestBF(sleep parameter ~gender + ID) allowed us to control for repeated measures by ID and to compare the probability of the alternative to the null hypothesis (no effect). Bayes factor results indicate the likelihood of the data occurring if the null hypothesis is true compared to the alternative hypothesis being true. Negative Bayes factor values that are further from zero provide stronger evidence for the null hypothesis, and positive values further from zero indicate stronger support for the alternative hypothesis. Interpretation for Bayes factor values is adapted from Jarosz and Wiley [43].

We report additional sleep variables that were generated with CamNtech Motionware analysis software. These include sleep latency (time in hours between lights out and sleep onset), wake after sleep onset (WASO; periods of awakening during the sleep period, expressed in hours), and sleep fragmentation (a quantitative index of restlessness in the sleep period, calculated as the sum of the percentage of sleep epochs spent moving and the percentage of immobile periods during the night).

The circadian entrainment hypothesis

Our results are consistent with the circadian entrainment hypothesis. Sleep was less efficient among Maya participants than among those in more industrialized settings. Maya participant interdaily stability, a quantitative measure of regularity in the rest-activity pattern, was greater compared to the industrial United States sample, and persisted after controlling for age and gender. Placed within the broader context of sleep in other nonindustrial populations, sleep in the Maya sample was longer and more efficient than sleep in Hadza hunter-gatherers (6.25 hours, 68.9% efficiency) [22] and Himba pastoralists (5.47 hours, 66.4% efficiency) [44]. However, Maya participant sleep duration was about the same as that reported in Tsimane forager-horticulturalists (6.63 hours) [45], and shorter and of lower efficiency than rural Haitians (7.0 hours, 88.7% efficiency) [46] and industrialized populations in Italy (7.60 hours, 94.2% efficiency) [47] and the United States (7.02 hours, 89.9% efficiency) [48].

The sleep-delaying effects of artificial lighting are evident in a number of comparative studies. For example, greater exposure to outdoor nighttime lighting in urban United States settings is associated with later sleep and wake time, shorter sleep duration, and increased daytime sleepiness than in rural ones [11]. The use of light-emitting E-readers at night has been shown to result in longer time to fall asleep and reduced daytime alertness, fueling the concern that round the clock access to indoor lighting and use of light-emitting technologies are delaying bedtimes, leading to insufficient sleep duration [10]. In a comparative study between two small-scale subsistence groups in Argentina, exposure to natural light only was linked to longer average sleep duration than when there was exposure to electric infrastructure. This decrease in nightly sleep duration in those using electricity was associated with delayed bedtimes [12]. Similarly, individuals with access to electricity in rubber plantations in the Amazon rainforest experienced later bedtime and shorter sleep duration on workdays than those without access [13]. Participants who used artificial lighting exhibited delayed melatonin onset, demonstrating the potential physiological effect of nighttime exposure to light. In contrast, Hadza and Himba populations experience comparatively shorter, lower quality sleep despite lack of electric infrastructure. Similarly, our results of moderately long sleep duration and efficiency from the Maya community suggest that a presence of technology is not always associated with shorter, poorer quality sleep. Therefore, our findings add to a growing understanding that the relationship between industrialization and sleep is not linear, and is rather shaped by a complex interaction of environmental and social factors.

In our linear mixed effects models of sleep duration, efficiency, and central phase within the Maya sample, we found that rainfall was significantly associated with shorter sleep duration. It is possible that the tin roofs that are present on many of the houses in the sample community lead to noise from rainfall that delays or interrupts sleep. Our data were collected during January, a relatively dry part of the year. A comparison with data collected during the rainy season and inclusion of noise data in our models would help to clarify the association between rainfall and sleep duration. Neither temperature nor humidity had a significant association with sleep variables in our sample. While temperature in particular has been previously associated with sleep duration and timing, it is possible that the houses of the Maya participants—which are often made of adobe, a good temperature and humidity buffer—are more protected from the elements than typical housing in the small-scale subsistence groups discussed (e.g., Hadza), and may offer sufficient environmental protection to protect sleep from substantial disturbance due to temperature extremes or high humidity.

Our results add to previous findings pointing to the fortifying effect of environmental exposure on circadian rhythms. Using rest-activity patterns to infer characteristics of circadian rhythm, we found that relative amplitude (rest-activity rhythm strength) of 0.92, interdaily stability (rest-activity rhythm stability) of 0.58, and intradaily variability (rest-activity rhythm fragmentation) of 0.53 indicate that, in general, Maya participants’ rest-activity pattern robusticity was strong. This rest-activity rhythm robusticity was apparent when the Maya sample’s results were compared to an industrial population from the United States. Maya participants had more stable rhythms (interdaily stability) and less fragmented rhythms (intradaily variability) than their United States counterparts, although we did not find any difference in the expression of rest-activity amplitude.

Interestingly, Mayan participants did not appear to differ from their Malagasy counterparts in rest-activity rhythm stability, had marginally less rest-activity fragmentation, and showed no significant difference in the expression of relative rest-activity amplitude. In the Maya participants, exposure to nighttime lighting was minimal during the study period, illustrated in the representative actograms in Fig 5. These findings suggest that having more environmentally-buffered sleep sites, with reduced exposure to natural light and temperature fluctuations, combined with individual exposure to the natural environment, daily outdoor activities, and/or social regulation of sleep-wake timing may also influence circadian rhythm expression. If natural sunlight and limited use of artificial lighting are important for synchronization between environmental cues and physiology [9], it follows that urbanization and technological development could explain the increase in reports of sleep disturbance in metropolitan populations [49,50]. This increase may be due to circadian disruption. Overall, our results emphasize that sleep patterns’ complexity cannot be fully explained by population industrialization, and provide further support for the beneficial effects of some technological buffering from environmental exposure.

The gendered sleep hypothesis

Our test of the gendered sleep hypothesis yielded mixed results. Consistent with our predictions, we found that women’s sleep was longer than men’s, and that men had significantly later chronotype (later bedtime and wake times). Furthermore, men showed significantly more variable sleep onset and offset, suggesting that work and social activity patterns differ between men and women. These results are consistent with our personal observations of the daily work and social schedules in this community. Women are responsible for making breakfast for the household, which involves early morning trips to the mill to obtain corn for tortillas. Men tend to socialize more at night, which could plausibly drive later bedtimes. Alternatively, we considered that ecological drivers (i.e., timing of sunset and sunrise that entrain circadian rhythms) and meteorological variables (i.e., temperature, humidity, and rainfall) that are uniform across the population may be a more important driver of sleep patterns, in which case we predicted no sleep differences by gender. While we observed significant differences between men and women in sleep duration and timing, no significant gender differences were found in rest-activity rhythm variables. Therefore, it appears that socio-ecological influences shared by both men and women in this population are important and that they may be contributing to circadian rhythm regulation.

Our findings in tests of the gendered sleep hypothesis are consistent with other findings in similar cultural contexts. Recent results from nonindustrial settings point to the importance of cultural factors in shaping gender differences in sleep. Prall and colleagues [44], for example, report extremely short sleep in Himba pastoralists of Namibia, who have been recorded to sleep just 5.47 hours per night on average [44], in stark contrast to sleep duration averages in industrialized populations, which tend to fall within the 7–8 hour range (e.g., [47,48]). Short sleep durations are especially pronounced in Himba men, with women presenting longer sleep duration (5.92 hours) and higher efficiency (70.2%) on average than men (average duration = 4.76 hours, average efficiency = 60.3%). The authors posit that high levels of young male social activity combined with daily labor demands of pastoralism are likely responsible for the dramatically low average sleep duration exhibited by men. A wide age range (15–78 years) of Himba individuals were included in this sleep analysis, with older age predicting longer sleep duration and higher sleep efficiency. This finding supports the idea that short sleep in this population and gender differences in duration and efficiency may be the result of high levels of evening socialization, especially in men and younger individuals [44]. Overall, it appears that socio-ecological contexts may evoke gendered effects in sleep-wake regulation.

Bolivian Tsimane forager-horticulturalists exhibit highly variable sleep onset timing that is associated with social and work activities which differ by occupation and between men and women, such as hunting and fishing; these activities were found to contribute to a decrease in total sleep time of 1.3 and 2.32 hours, respectively [45]. Tsimane women’s sleep duration is on average longer than men’s, a pattern which is attributed in part to differential labor demands [45]. Similarly, it appears that a combination of labor and social demands is contributing to men’s decreased sleep duration in the Maya community as well, where men also appear to be involved in more nighttime social activity than women.

The interaction between gender and age suggests that age predicts longer sleep duration in women, but shorter sleep duration in men (Fig 2 and S3 Table). It is surprising that age was negatively associated with sleep duration in men, as we expected younger men to spend more time participating in late-night social activities that contribute to shorter sleep duration and more variation in bedtime and wake time. It is possible that the younger men in our sample, some of whom are in their late teens and early 20s, are more constrained by household rules and therefore limited in how late they are allowed to stay out, while older men many be more likely to have the freedom to stay out later at night. Furthermore, our results add to the consistent pattern of men showing a more evening-preferring chronotype than women, perhaps due to a possible benefit to social networking [26,28], as well as cultural norms and prescriptive social rules that limit women’s nighttime socializing. Our findings help to clarify gender differences in sleep patterns and speak to the importance of considering cultural factors, such as gender expectations, labor roles, and social activities.

For men in particular, supplemental agriculture work on weekends may in part explain Maya sleep durations that were shorter than those in industrial populations, where longer sleep on weekends and free days could contribute to higher population averages. Other weekend morning activities that are common in this Maya community, such as visiting the market or attending church, could influence sleep duration as well. Specific activity patterns should be studied in future work for a more complete understanding of how Maya economic and social activities may be contributing to the relatively shorter sleep durations observed in our analysis.

Limitations

There are limitations to our study which warrant attention. First, we lack quantitative data on the number of households with access to electricity, home electric devices, and information on the frequency of power outages. While access to electricity and ownership of home electric devices such as televisions and refrigerators has increased over the past two decades, there is still the possibility that individual and household variation in the use of electricity could play a role in sleep and rest-activity patterns. Future research should employ further quantitative measurement of these factors.

A second limitation involves the comparability of our comparative samples. The data for our comparisons between Maya, Malagasy, and MIDUS samples were generated using CamNtech MotionWatch 8 actigraphs, collected in 60-second intervals (Maya and Malagasy), and Mini Mitter Actiwatch-64 activity monitors, collected in 30-second intervals (MIDUS). Sleep data scoring protocols corresponded to their respective accelerometers, and data cleaning and analysis was done by different individuals, which introduces possible rater inconsistencies. However, we addressed these limitations to the best of our ability by using comparative sleep variables that were generated with high consistency in scoring protocols. Differences in 60-second intervals vs 30-second intervals have been previously found to be insignificant in the generated actigraphy results [51]. A number of other socio-ecological factors could contribute to differences in sleep and rest-activity patterns that are difficult to account for in our comparative analyses between these samples as well. For example, differences in season and latitude in which data were collected, which can affect sleep patterns through temperature and sunlight exposure, could be partly responsible for some of the differences between populations that we report here.

Finally, a potential limitation concerns the chronotype results. Sleep mid-point on free days, when bedtime and waketimes are not constrained by work responsibilities, would more accurately serve as a proxy for individual’s sleep patterns and chronotype. However, we do not have data available on participants’ work schedules, and are therefore limited in our conclusions regarding the relative contribution of physiology and cultural factors in influencing chronotype patterns reported here.

Conclusions

In sum, sleep data from this Maya community that is currently undergoing an economic transition which affects zeitgeber exposure helps in furthering our understanding of the environmental, economic, and cultural drivers of human sleep. Our results are consistent with those of other recent studies suggesting that, despite the sleep disruptive effects of electricity on sleep, industrialization does not inherently drive significant reductions in sleep duration and efficiency compared to nonindustrial, or “natural” human sleep (e.g., [8,16,52]). In the cultural West, there is evidence suggesting harmful effects of artificial lighting on sleep patterns. Yet, it appears that sleep in the Mayan sample—which was longer and more efficient than sleep in many small-scale populations with less industrialization—benefits from the safety and comfort of sleeping sites that are relatively buffered from the environment. Thus, these findings suggest that despite arguments implicating urbanization in giving rise to a sleep loss epidemic [14,15], the climate control and safety of urban settings may contribute to longer, better quality sleep while paradoxically also be leading to circadian rhythm weakening and desynchronization through decreased exposure to these zeitgebers [16]. For individuals living in industrialized environments, our results suggest that intentional exposure to natural sunlight, minimizing use of electrical devices around bedtime, and maintaining consistent bedtime and wake schedules can improve sleep health and general wellbeing.