Longitudinal Study of Toba/Qom Communities Reveals Delay and Shortening of Sleep in Real-Time

Introduction

The harnessing of electric power and then the invention of the electric bulb revolutionized human life in ways that are impossible to quantify. Electric light further allowed for the extension of wakefulness and chores into the evening hours and, eventually, the birth of the 24/7 society of work and entertainment in the current hyper-industrialized world. This process has significantly delayed sleep times, but whether it shortened sleep duration is still a matter of debate. A systematic review of studies reporting the duration of sleep between the 1960s and the 2000s failed to find a consistent shortening of sleep,¹ and a recent analysis of published sleep data suggests that in industrialized societies sleep may be longer than in non-industrialized ones.² In contrast, a poll by Gallup³ and a study on self-reported sleep suggest that in the last two decades sleep has shortened in the US.⁴ Similarly, a systematic review of available children’s data from 1905 to 2008 suggests a decrease in daily sleep of one hour in this period.⁵

In order to study the ways in which industrialization has affected human behavior, we need to be able to study populations in which this process is still ongoing. Pre-industrial societies offer unique opportunities to investigate the impact of the adoption of modern technologies and their related habits on sleep patterns, as they usually display differential access to electricity, high levels of exposure to the natural environment, weaker scheduling impositions, and rely more on environmental and ecological time references.^6–9

Our group is focused on the study of native Western Toba/Qom populations in the northern region of Argentina.¹⁰ These people, formerly hunter-gatherers, remain a very isolated and underserved group spread in the Formosa province, relying mostly on government assistance for subsistence. For the last 14 years, we have been studying sleep and circadian rhythms of both rural and semi-urban Toba/Qom communities (referred to as “urban”, for the sake of simplicity; see “Toba/Qom participants” in Methods), allowing us to contrast the behavior of communities in which the most relevant distinguishing characteristic was, until not so long ago, their access to electricity. Our previous work has shown that, indeed, urban Toba/Qom displayed later sleep onsets and shorter sleep duration than rural communities with no access to electricity.¹¹ This correlation between electric light and delayed sleep has also been reported by several other groups studying pre-industrial communities around the world.^12–15

Since we started working with Toba/Qom communities in 2012, electricity has been progressively introduced to most rural areas in this region of the country, including the most isolated settlements (Figure 1A). Since such efforts started around 2016, all the communities we have worked with are now connected to the electric grid, and at the time of writing this paper, every single family we have worked with has electricity in their homes. In recent years, television sets, smartphones, and internet connections, while in limited supply, are already available to many families.

We have longitudinally monitored sleep patterns through wrist-actigraphy in these Toba/Qom communities at different seasons and time points since 2012, generating a dataset of over 12,000 sleep records. This dataset represents a unique opportunity to assess the impact of the introduction of electricity on the sleep of these communities in real-time. The study we present here provides a close insight into the history of human sleep patterns in developing pre-industrial communities, highlighting the profound impact of modern technology on human behavior.

Results

Our sleep dataset from Toba/Qom communities includes over 12,000 sleep events recorded through wrist-actigraphy between 2012 and 2024 from a total of 156 participants (93 in the rural group, 62 in the urban group, and one who moved from one community to the other after 2018). Table S1 details the number of participants, recorded events, and age and sex composition of all data according to community, and the dates when it was collected. Starting around the year 2016, electricity was progressively introduced in the rural communities in the Vaca Perdida and neighboring areas, including the communities we studied (Figure 1A). Throughout this period of significant transformations for the rural Toba/Qom, we observed a progressive delay and shortening of sleep both at the individual and community level (Figure 1B and Table S2).

Sleep start, midpoint, and end times, sleep duration, and sleep regularity (measured by the Sleep Regularity Index) were analyzed through linear mixed-effects models including groups (urban or rural), date, photoperiod length, type of day (workday or free-day), and sex and age as fixed effects, while including random intercepts and random slopes across time for each participant. Interaction terms between group and date, and between sex and age were also included in the models. Due to differences in the actigraphic devices employed across time, we implemented a non-parametric multiple imputation strategy utilizing empirical error distributions based on a side-by-side comparison of Philips Actiwatch Spectrum Plus and Axivity AX3 devices. The results reported are therefore based on bootstrapping of one hundred independent simulations, effectively incorporating this variability as added imprecision in our study (see Methods for the complete details on the statistical analysis and for additional considerations on the older Actiwatch-L model).

Table 1 displays the full results of models for sleep onset, offset, and midpoint; Table 2 shows the results for sleep duration and sleep regularity. Coefficients are expressed along with their 95% CI. The models revealed photoperiod length as a strong predictor of sleep times, with 0.20 [0.17–0.23] h delays in sleep start times (t = 12.244, p < 0.001), 0.16 [0.13–0.19] h advances in sleep end times (t = 10.557, p < 0.001), and a reduction of 0.36 [0.33–0.40] h of sleep (t = 19.677, p < 0.001) per hour of increase in daylight. A longer photoperiod was also associated with slightly higher sleep regularity (1.0 [0.4–1.7] units/h of photoperiod extension, t = 3.097, p = 0.002). Weekends displayed slightly later sleep start times (0.18 [0.12–0.25] h, t = 5.623, p < 0.001) and later sleep offset times (0.21 [0.15–0.26] h, t = 7.055, p < 0.001), and thus did not affect sleep duration. Accordingly, the midpoint of sleep was minimally delayed in nights before free days (0.19 [0.15–0.24] h, t = 7.907, p < 0.001), indicating that social jet lag is minimal in these communities. Male participants started their sleep significantly later (0.59 [0.33–0.84] h, t = 4.438, p < 0.001), woke up later (0.46 [0.20–0.73] h, t = 3.446, p < 0.001), and had a lower sleep regularity (-2.85 [-4.88 to −0.82] SRI units, t = 2.762, p = 0.006), but slept about the same amount as female participants. Consistent with previous findings in these communities, the rural group showed overall earlier sleep start times than the urban group (0.32 [0.05, 0.6] h, t = 2.303, p = 0.021).

Figure 2 and Table S3 present the model-predicted changes in sleep variables for each group under a 12-h photoperiod length for a 22-year-old (the median of ages across all campaigns), while averaging for sex and type of day, for the 2014 to 2024 ten-year span. Models indicate similar delays in the times of start of sleep of 0.88 [0.49–1.27] h (t = 4.444, p < 0.001) for the rural group and of 0.90 [0.40–1.40] h (t = 3.528, p < 0.001) for the urban group. Evolution of sleep end times was asymmetrical between groups: while the rural group showed only a marginal, non-significant change in wake-up times over ten years (0.32 [-0.09 to 0.73] h, t = 1.540, p = 0.124), the urban participants delayed theirs substantially (0.87 [0.36–1.39] h, t = 3.324, p < 0.001). Consistently with these asymmetric shifts, changes in sleep duration were different between communities. While the rural group displayed the loss of 35 minutes of sleep (0.58 [0.30–0.86] h, t = 4.105, p < 0.001), this reduction was absent in the urban group. This asymmetry in the evolution of sleep times also implied that the urban group showed the larger shift in their midpoint of sleep, with a delay of 0.93 [0.45–1.41] h (t = 3.808, p < 0.001), as compared to the 0.61 [0.23–0.99] h delay of the rural group (t = 3.138, p = 0.002). Sleep regularity, measured by the Sleep Regularity Index, showed a significant decrease in both groups across time. Between 2014 and 2024, the model predicts a loss of 5.2 [1.6–8.8] units in the rural population (t = 2.876, p = 0.005) and of 10.2 [5.3–15.0] units in the urban group (t = 4.148, p < 0.001).

The 2012 to 2024 period involved important transformations in these Toba/Qom communities. First, starting around 2016, electricity was progressively introduced in rural communities. Secondly, while to our knowledge not a single participant from either community under study, nor anyone in their families, owned a smartphone by 2018, about half of the participants in our studies had their own smartphone, and about the same had access to a TV and/or a computer at their homes by October 2024. Moreover, even in very remote rural areas, where cell phone coverage is extremely limited, some participants can access the internet through free public Wi-Fi hotspots.

We set out to test whether either of these two events (i.e. introduction of electricity, and access to digital connectivity) could support the observed changes in sleep timing across this timespan, based on their potential for extending activity in the evening hours and thus delaying the start of sleep. To do this, we fitted a new linear model in which every sleep event was labeled as recorded either under conditions of access to electricity or not, and whether it was recorded during the “smartphones and internet era,” an umbrella term for all data from 2023 onwards (see “Test of candidate mechanisms” in Methods for further details on this analysis). This model revealed that neither the introduction of electricity into rural areas nor the “smartphone era” can explain by themselves the longitudinal changes in sleep onset observed in these communities, which were still consistent at an estimated delay of 0.12 [0.05–0.19] h per year (t = 3.321, p < 0.001; Table S4). Because both electricity access and the “smartphone era” indicators are, by construction, correlated with calendar date (see “Test of candidate mechanisms” in Methods), we cannot claim that this analysis fully rules out a contribution of either factor to the observed changes in sleep, particularly since both are independently plausible drivers of sleep timing, and a non-significant coefficient under any degree of shared variance with date does not, by itself, demonstrate the absence of an effect. Our diagnostics indicate, however, that this correlation is mild rather than severe, and that the residual date effect does not attenuate once electricity and smartphone use are included in the model (if anything, it remains of a similar, even slightly larger magnitude, and stays significant). We therefore interpret these results as evidence that a longitudinal trend in sleep persists beyond what can be attributed to these two milestones in the “modernization” of Toba/Qom lifestyle.

Discussion

Our analysis reveals the rapid evolution of sleep habits in two Toba/Qom communities undergoing early stages of modernization, marked by the sequential introduction of electricity, mobile devices, and internet access. Drawing on over 12,000 recorded sleep events spanning twelve years, our models document significant shifts in sleep timing and duration: in the span of a single decade, sleep onset was delayed by close to one hour, and total daily sleep was reduced by 35 minutes in rural communities. These findings echo patterns reported in historical records¹⁶ and self-reported sleep studies^4,5 tracking the transition from preindustrial to industrial (and more recently digital or postindustrial) environments. Yet, longitudinal investigations using objective sleep measures in communities where modernization, understood as a set of co-occurring environmental and social changes, unfolds incrementally have remained absent from the literature. Our results are consistent with the hypothesis that such transformations delay and shorten human sleep, and provide the longitudinal, objectively measured evidence of change over time that this hypothesis has so far lacked. Importantly, our findings imply there are other factors driving sleep changes beyond the availability of electricity and the use of electric light, and even smartphones, computers, or TV sets. Together, these results highlight the need to monitor how the introduction of modern amenities reshapes the lifestyle and biology of communities in transition.

In the time we studied these communities, we detected similar delays in the time of sleep start across groups, but only the urban community displayed a comparable delay in the offset of sleep. Rural communities, who live under even more limited conditions than their counterparts in the city of Ingeniero Juárez, may not be able to extend their sleep much later in the morning either due to being more exposed to natural elements (sunlight, temperature), to the need to start with sustenance activities (e.g. harvesting firewood, managing livestock, gathering), or to other necessities related to their isolated rural environments (e.g. distance to health centers or convenience shops, lack of transportation options). This contrast may explain the reduction in sleep duration in the rural group in the face of similar shifts in the timing of sleep onset. That is, the large shift in sleep start times appears to be the most relevant culprit driving changes in overall sleep behavior in these communities.

Internet access and devices such as smartphones, however limited in supply, allow Toba/Qom communities to connect to a world beyond their local environment, for example, through social media and video platforms. The adoption of these technologies in these communities has been very rapid. To the best of our knowledge, smartphones were virtually unavailable by 2018, yet by 2024, half of the Toba/Qom participants owned a smartphone, and about the same had access to a TV or a computer in their homes, reflecting a swift shift toward digital connectivity. A public internet antenna was installed as recently as 2022 in Isla García, providing free internet access to the community; one family in Vaca Perdida even had a satellite internet connection. These technological advances likely favor the delay of sleep and, whenever the time for waking up cannot be further delayed, the reduction of sleeping time by providing almost unlimited entertainment for the night hours. However, our models indicate that the introduction of smartphones and the internet cannot explain the observed delays in the start of sleep by itself. Moreover, this delay in sleep times could not be driven merely by late-night smartphone users. A recent study on these communities revealed that users of smartphones display only a 13-minute delay of sleep compared to non-users,¹⁷ making it impossible for just these users to explain the population-level changes reported here. Electronic devices are hypothesized to impact sleep via several simultaneous mechanisms: direct reduction of sleep time, effects of emitted light, promotion of arousal, and causing sleep disruptions.¹⁸ For the Toba/Qom population, an additional factor may contribute beyond these established pathways. Electronic devices such as smartphones, computers, and TVs, along with access to the internet, can open a door into the “modern” lifestyle, of which nightlife is a central element. This is particularly relevant in Argentina, where, for instance, the TV “prime time” extends from 21:00 to midnight and football matches can start as late as 22:00.¹⁹ This phenomenon merits further research incorporating Toba/Qom self-perceived experiences regarding access to electricity and electric light sources, as well as the roles of the internet, the TV, and other electronic devices.

Sleep loss is considered a modern world epidemic with significant consequences for public health, performance and safety, learning and memory, and overall economic impact.^20–23 Lack of sleep can affect health at several levels, including, but not limited to, cardiovascular, metabolic, immune, and mental health,^24,25 and is a strong predictor of all-cause mortality.²⁶ Although our models predict an average sleep duration still within the consensus recommended sleep duration range for these ages,²⁷ we note that these recommendations are aimed at people in modern developed societies. More research is needed to assess whether this reduction of sleep time in this population is associated with changes in physical health, psychological wellbeing, and overall quality of life.

The observed reduction in sleep regularity in both communities may be more concerning. Sleep regularity has gained attention recently as a marker of sleep health,²⁸ and has been proposed as a more precise predictor of the negative impacts of poor sleep than sleep duration itself.²⁹ Delayed sleep has already been associated with low sleep regularity in hyper-industrialized societies,³⁰ and in a study combining data from adolescents in the Toba/Qom and in rural communities of Southern Brazil, we already reported that sleep regularity was reduced in students exposed to electric light in the evening hours.¹⁴ Further investigation should determine whether this change is driven by the mere access to electric light and other electric appliances and devices at home, and/or if it is correlated with changes in evening habits that involve the whole community.

Numerous studies of preindustrial communities support the notion that electricity and other postindustrial environmental factors impact sleep timing.⁷ However, the hypothesis that industrialization has necessarily shortened sleep remains under debate. A recent meta-analysis, for instance, suggests that people in industrialized societies sleep longer than those in preindustrial settings, and attributes modern sleep-related health issues not to shorter sleep but to heightened circadian disruption.² However insightful, these observations need to be carefully weighed, considering the inherent diversity of pre-industrial and industrialized groups, without overlooking the unique environmental and sociocultural factors that can shape sleep. Our longitudinal study of the Toba/Qom communities provides a rare opportunity to assess the real-time impact of certain features of modernization, including the acquisition of electricity and related technologies. Our data reveals that in these Toba/Qom communities, this process correlates with a significant reduction in sleep duration and an increased circadian disruption —marked by a reduced sleep regularity and the further decoupling of sleep onset from sunset. We should also note that the transition from harsh preindustrial conditions to an electrically lit industrial environment can also have beneficial effects on sleep, for example, by improving comfort and sense of safety at bedtime.³¹ This underscores the intricate ways in which the environment and technology interact to shape human sleep patterns.³²

Some limitations in our study need to be addressed. First, it should be noted that the recordings in our dataset have been obtained using different devices and algorithms. Because of the long timespan of our study, this was an obstacle hard to overcome. Of note, the Actiwatch devices used from 2012 to 2018 do not provide raw accelerometric data, as the Axivity AX3 devices do, precluding us from using a common algorithm to extract sleep timing. These different methods are all based on longitudinal wrist-accelerometry and the sleep estimation in each of them was performed following well-tested protocols, based on similar criteria of activity/inactivity and aided by sleep logs to detect unreliable estimations.

Our side-by-side tests indicate that average sleep time estimations from AX3 and Actiwatch Spectrum devices are comparable within a margin of less than 20 minutes of difference between them (see Figure S2). However, because a discrepancy of this magnitude is biologically relevant, and because differences between actigraphic methods may even vary between individuals with different activity profiles,³³ we followed a rigorous procedure to integrate the available data by using multiple imputation of errors to account for the empirical disparities between devices. By presenting results pooled from one hundred simulated models for each sleep variable, this approach provides strong confidence that the detected longitudinal differences that survive the addition of this imprecision into our statistical analysis reflect a genuine temporal phenomenon rather than an artifact of changing instrumentation. Our correction method resamples from the empirical distribution of 206 paired differences obtained from non-Toba/Qom participants who wore both devices simultaneously, thus propagating the location, dispersion, and skew of this discrepancy into our uncertainty estimates. The assumption, nevertheless, is that such error can be comparable across groups. This is a limitation that cannot be overcome with the data available to us, and that demands further precautions in the interpretation of our results. Finally, because we could not perform a specific side-by-side validation of the older Actiwatch-L devices, we performed an additional sensitivity analysis over a reduced dataset excluding all data obtained with these devices, which revealed essentially the same results (Table S5).

Second, thanks to the improvement of fieldwork conditions, in terms of the available equipment and our relationship with the communities, our dataset is much denser in recent years, compared to the first four years of work. While this may pose a limitation in the model’s precision for those early years, the use of linear models, including random effects accounting for sex, age, and repeated measures, is robust enough to assess the dynamics of sleep across the studied period in these communities. Importantly, the lower number of participants in the rural community before the appearance of electricity may have contributed to the lack of a statistically significant effect of electricity on sleep onset in our analysis. Obviously, this result does not mean that the appearance of electricity was not critical for the sleep delay and shortening we observed; electricity not only allows for artificially lit evenings but also for internet hotspots, cell phones, TVs, and other sleep disruptors.

Finally, it is worth addressing the five-year interruption in data collection between 2018 and 2023. This covers the time of the COVID-19 pandemic, which of course impacted on the possibility of continuing fieldwork. The pandemic was associated with substantial changes in sleep in populations worldwide, including in Argentina, generally involving delayed and lengthened sleep together with reduced social jetlag.^34–37 However, the mechanisms proposed to drive pandemic-related sleep change (i.e., confinement indoors and consequent reduction of natural light exposure, the relaxation or pausing of fixed work and school schedules, the elimination of commuting, and the adoption of “working from home” routines) are largely tied to urban, densely populated, and wage-employed ways of life. These conditions are substantially attenuated, if not absent, in the Toba/Qom communities we study, where subsistence activities are largely outdoors, commuting is very infrequent, and formal employment is rare. Moreover, our results show the opposite trend in sleep duration to that reported during the pandemic, so the only dimension on which a pandemic legacy could plausibly act in the direction we observe is the delay in sleep onset. An additional sub analysis of sleep start times in the 2012 to 2018 data determined that the trend of sleep delay was already present at a comparable, albeit numerically larger, rate with overlapping 95% confidence intervals (Table S6), making it unlikely that the longitudinal trend we report is driven by an acute change during those years.

Our work suggests changes in sleep habits as one of the most plausible consequences and multifaceted impacts of technological innovation on human behavior and health. These impacts could differentially affect underserved communities like the Toba/Qom, who have finally attained the benefits of the access to electricity, smartphones, and the internet, but remain significantly behind in the access to many of the advantages of modern, developed societies in other dimensions such as infrastructure and health services. This disparity may exacerbate any potential negative consequences of the observed delayed and shortened sleep in these communities, which are commonly reported features of modern lifestyle. In this sense, the Toba/Qom case we present exemplifies what has been described as “cultural arrhythmia”, i.e., an imbalance between the pace of changes in human lives and their surrounding habitats.³⁸ Consequently, it is imperative that such transformative processes consider the unique features and conditions of each community, ensuring they enhance health and overall wellbeing while proactively addressing any potential challenges. Our results highlight the need to monitor sleep health in groups amid such rapid environmental transitions, the full implications of which remain to be established.

Resource availability

STAR Methods

Supplementary Material

Document S1. Figures S1–S2 and Tables S1-S6.