Study design, setting, and participant recruitment

The BSETS is an observational study in an ecologically valid setting that monitors participants for 7 consecutive days and nights, with preliminary assessments taking place online as well as in the laboratory where devices are handed over. All in-person assessments take place at a single center in the NET Building of Semmelweis University, in Budapest, Hungary.

The study is designed to be a cross-sectional, observational, quantitative study of volunteers, with an approximate total sample size of N = 250 participants, who are recruited via the method of branched diffusive convenience sampling. The selection criteria are: healthy volunteers over the age of 18 who are native speakers of the Hungarian language to ensure the comprehension of research protocols and requirements and capable of providing informed consent. Recruitment from clinics or clinical settings was deliberately avoided due to concerns about the generalizability of the sampling pool, as the primary purpose of the BSETS dataset is the investigation of sleep-DAE-trait relationships as opposed to sleep disorders or the effect upon these relationships of medical conditions that are implicated in disordered sleep. We collect data about daily medication use, permitting the post-hoc exclusion of participants seriously affected enough to be medicated. Although BSETS is an ongoing research study, we present summary statistics describing the features of our participant pool (e.g. sex ratios, average age, average level of educational attainment, test results) in the results section; further summary statistics will be made available as the study progresses.

Participant recruitment was initialized through the use of branched diffusive convenience sampling, for which candidates for research participation were solicited through in-person interactions, via email, and through solicitations publicized in specific online forums. No financial incentives or rewards were offered in return for research participation; compensatory incentives were restricted to the release of processed sleep data in the form of an automated weekly sleep report to participants following the completion of the study, as well as student credits in the case of participants or the students recruiting them already enlisted in pre-listed courses at Semmelweis University. Participants were encouraged informally to invite their family members, friends, and romantic partners to participate in the study following their own successful completion in our research. However, multiple members of the same household never participated in the study at the same time.

Data collection started in October 2021, and is expected to conclude in May 2023.

Ethics

The Institutional Review Board (IRB) of Semmelweis University as well as the Hungarian Medical Council (under 7040-7/2021/ EÜIG "Vonások és napi események hatása az alvási EEG-re" (The effect of traits and daily activities and experiences on the sleep EEG)) approved BSETS as compliant with the latest revision of the Declaration of Helsinki [40]. All participants gave written informed consent on a form reviewed and approved by the IRB.

During data collection, authors had access to information which could identify participants. This was necessary because personal contact via email was required to deliver and collect questionnaires and login data for the mobile EEG application. However, data is de-identified by using an internal code system after data collection. We only publish de-identified data from BSETS.

Measures, questionnaires, and outcomes studied

BSETS utilizes four discrete measurement domains for collecting data on sleep, traits, and experiences from participants prior to and during the 7 days and nights of the study period.

1. Psychological Trait Measures (or Traits); gathered via pre-study psychological testing
2. Objective Sleep Measures; gathered via portable EEG devices
3. Subjective Sleep Measures; gathered via participant self-report
4. Daily Activities and Experiences (DAEs or Experiences); gathered via participant self-reports

Detailed descriptions of the measures, tests, and tools utilized are provided below.

Psychological trait measures. In BSETS, the use of extensive pre-trial psychological test batteries is a means by which to establish a holistic snapshot of participant psychophysiological profiles. The role of these is both to serve as potential correlates of sleep themselves, but also to better contextualize findings related to day-to-day variation as moderators. When evaluating the suitability and significance of pre-trial psychological tests for inclusion in BSETS, the following criteria were applied:

1. The test must measure a trait with an established relationship with components or features of sleep, and a history of use in sleep research.
2. The test must be validated by previous studies.
3. The test must be intelligible to non-specialist general audiences and must be completable within reasonable time without researcher intervention or assistance.

Upon applying these criteria, we selected 10 psychophysiological test batteries and a further custom cognitive test that were selected for inclusion. These tests, the context of their use, their measurement target, their relevance for BSETS, and their associativity with sleep are described in Table 1 below.

Download:

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

Table 1. A summary of trait test batteries used in BSETS.

https://doi.org/10.1371/journal.pone.0288909.t001

Daily Activities and Experiences (DAEs): Evening diaries.

The evening diaries describe the DAEs of participants prior to sleep and at the end of a finished day. They are required to be filled out each night before sleep. Similar to the morning diaries, the structure of the evening diaries can be broken down into two sections.

Day Descriptions: here participants provide basic information about their day, and evaluate it in terms of emotion and affect (time required: ≤ 3 minutes) (Table 2).

Daily Experiences and Events: here participants respond to a series of common events and situations, and describe whether or not they experienced these in the preceding day (time required: ≤ 7 minutes) (Table 3).

Download:

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

Table 2. A summary of the evening diary sections used in BSETS for the overall description of the previous day.

https://doi.org/10.1371/journal.pone.0288909.t002

Download:

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

Table 3. A summary of the evening diary sections used in BSETS for the description of specific experiences during the previous day.

https://doi.org/10.1371/journal.pone.0288909.t003

7 identical copies of evening diaries are provided to each participant. Because university students make up the majority of the participant population, BSETS researchers anticipated that poor compliance in terms of diary completion timing might be higher due to the social schedules and nightlife habits common among student populations. The evening diaries therefore begin with a quality control question, asking participants if they filled out the questionnaire in the evening as required, or in the morning from memory.

After this quality control question, participants provide a free-form description of the day.

After this, the DAEs section of the evening diaries begins, starting with a prompt listed below (Table 3).

Objective sleep measures: EEG recordings, the Dreem 2 EEG headband, and monitoring.

In BSETS, EEG constitutes the primary method for measuring sleep quality and sleep features over a full night of sleep. In order to obtain sleep EEG recordings from research participants under naturalistic conditions with minimal invasiveness, we made a decision to utilize a novel remote EEG device, the Dreem 2 EEG headband (Fig 1), that can be used at home with a minimal level of researcher instruction.

Download:

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

Fig 1.

Ilustration of the Dreem 2 EEG Headband, with the layout of the device (left) and a person illustrating the way it is worn (right).

https://doi.org/10.1371/journal.pone.0288909.g001

The Dreem 2 EEG Headband device, as described in the corporate whitepaper [62] measures EEG signals with 3 electrodes in the frontal band (prefrontal position) and two more at the back of the head in the occipital position, measuring channels F7-O1, F8-O2, Fp1-F8, F8-F7, and Fp1-F7. An additional EEG electrode is used as a bias. Signals can be acquired through the medium of user hair, and a mechanical system distributes the pressure of electrodes to minimize discomfort. The sampling rate of EEG recordings is set to 250 Hz. By default, recordings are band-pass filtered to 0.4–18 Hz. A pulse oximeter measures the blood oxygen saturation and heat rate via reflectance oximetry. The device is controlled via the Dreem Coach mobile application that is available for both Android and iOS smartphones.

Participants are carefully instructed in the correct use of the Dreem 2 EEG Headband devices prior to the conclusion of the handover process; this includes instruction in the correct method of wearing, activating, deactivating, storing, and charging the devices, as well as in the method of connecting them to Wi-Fi and to the mobile application (Dreem Coach) used to operate the Dreem 2 EEG Headband. Participants are instructed to use the Dreem Coach mobile application to turn the devices on prior to sleep, or to restart the device if turning it off and then on again manually during the night. During data acquisition, a research assistant is available 24/7 to participants to respond to questions and solve problems that emerge during work with the headbands.

EEG recordings obtained from wearable remote EEG devices (Dreem 2 EEG headband) are uploaded to a cloud-storage database after the conclusion of each night of sleep and are inspected manually by researchers each morning. If the quality of recordings is judged as suboptimal, participants are contacted again with a reminder about the proper way to use the Dreem 2 EEG headband.

EEG recordings are scored automatically using an algorithm which has demonstrated high consensus with manual scorings using standard criteria of the American Academy of Sleep Medicine [63]. This algorithm also calculates sleep macrostructure metrics, including total sleep time, sleep onset latency, wake after sleep onset, sleep efficiency and the duration and percentage of wakefulness, N1, N2, N3, total NREM and REM in the recordings.

Artifactual epochs in EEG recordings are detected using a complimentary algorithm devised by Dreem Inc., which has been trained to reproduce the visual scorings of experts [62]. Each 2-second epoch, on each channel, is given a quality rating which corresponds to the likelihood of that epoch not being artifactual. Using these quality scores enables the exclusion of artifactual data from analyses on a channel wise basis. Several definitions could be used for identifying which epochs are to be excluded; of these, we have explored the effect of several on the tradeoff between data quality and the quantity of available data (see Results). In short, the determination was made to exclude data epochs if they did not reach a certainty of either 50 or 75% of not being artifactual, and all data from channels was also discarded if channel quality (the proportion of epochs from that channel scored as non-artifactual with at least 50% certainty) did not reach 20% or 50%. We excluded vigilance states from analyses if there were fewer than 20 epochs from that vigilance state fulfilling quality criteria from a channel. This aspect of the quality control process is necessary because even with high signal quality the number of data epochs fit for analyses can be scarce if a vigilance state (for example, REM) was rarely detected, resulting in noisy estimates of EEG metrics.

The availability of hypnograms and artifact scorings enables the use of standard time-series analysis methods, including the calculation of power spectral density (PSD), between-channel coherence and the detection of EEG oscillations. Of which, we elected to implement PSD analysis, and as a proof of concept. We attempted to replicate the well-known negative relationship between age and power in the delta and sigma frequency ranges, representing the loss of slow waves and sleep spindles in older individuals. PSD calculations were performed using EEGLAB’s periodogram() function with 2-second nonoverlapping epochs and Hamming windows.

Subjective sleep measures: Morning diaries and subjective self-reported assessments of sleep.

The morning diaries are required to be filled out each morning after waking up and can be completed in 5 to 10 minutes. The morning diaries can be subdivided into two sections (Table 4).

Sleep Diary Section: here participants provide subjective self-reports on the context, setting, and quality of sleep (time required: ≤ 5 minutes).

Dream Diary Section: here participants who affirm the recollection of dream experiences occurring in the preceding night describe their dreams and their contents (time required ≤ 7 minutes).

Download:

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

Table 4. BSETS morning diary, sleep diary section.

https://doi.org/10.1371/journal.pone.0288909.t004

7 identical copies of these diaries are provided to each participant, each containing an instructive preamble to remind participants when and how to complete the diaries (the main requirement being that they are filled out on all seven mornings of all seven days of the study, in the morning, not long after waking up). Participants record the time they begin to fill out the diaries, before answering a series of questions, which are as follows (Table 4).

After the Sleep Diary Section is completed, participants are asked to report their dream experiences using a recently developed dream affect inventory [65]. First, they are asked to respond to the following question (#6) with yes or no:

“Please read the following statement, and answer with a ‘true’ or ‘false’ as best applies to yourself:

‘I remember what kind of dream I had, what I was thinking about, and what feelings I felt before I woke up.’

In this question, by dream, we mean any experience, thought, or memory which crossed your mind when you were sleeping. This question applies to dreams in the morning, but also dreams you remembered when waking up during the night.”

Participants who respond “yes” to this question are prompted to continue with the dream component of the morning diaries (Table 5). Those who do not are requested to skip these questions.

Download:

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

Table 5. BSETS morning diary, dream diary section.

https://doi.org/10.1371/journal.pone.0288909.t005

This final question in the dream diary section, in which participants write as little or as much as they wish about their dreams; keywords, sentences, or complete paragraphs, enables semantic analysis and natural language processing tools to be applied to participants’ descriptions of dream content. In doing so, we can gauge the emotional valence of dreams, and conformity of dream content with participant’s descriptions of sleep quality, of daily experiences, or with EEG-evaluated sleep features.

Descriptive statistics

Data collection for BSETS is still ongoing. Here, we present preliminary results about data availability, sample demographics and quantitative EEG metrics.

So far, 208 volunteers took part in BSETS, of which 108 were females and 93 males. The gender of 7 participants remained unknown, as they did not respond to the trait questionnaires sent to them. The mean age in the sample was 28.83 years (SD = 12.49 years, range: 18–76 years). The age distribution of the sample was highly skewed, however, with 59.8% of the sample being younger than 25 years old, 11.3% being 26–35 years old, 6.7% 36–45 years old, 15.46% 46–55 years old and 3.1% older than 55 years old. The age distribution of the sample reflects the fact that most participants were friends or parents of medical students. 36.87% of the sample reported having advanced education (BA, MA or equivalent). 48.99% reported having a high school diploma, while 4.05% reported having vocational high school degrees. 5.56% of the sample reported only having completed eight grades of schooling, while 4.55% reported having a postgraduate degree. In a separate question, 56.6% reported being university students, likely representing students who will eventually obtain a university degree. Detailed descriptive statistics are provided in Table 6.

Download:

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

Table 6. Descriptive statistics of key variables in BSETS.

https://doi.org/10.1371/journal.pone.0288909.t006

In total, we recorded 1439 nights from participants in BSETS. We have registered 1383 diary reports (96.1% compliance rate) and 1399 EEG recordings (97.2% completion rate) from these nights. Some participants are still expected to return diaries in the future, which would further increase the diary compliance rate.

200 participants provided at least one night of EEG data. Some EEG recordings were lost due to technical problems. These usually happened because the recordings didn’t start or weren’t started by participants, or data was lost during the process of cloud uploading due to device errors. These nights were treated as missing data.

The mean number of nights per participant exceeds 7, which is the normal duration of the research protocol. This is because chose to continue the study beyond the original seven days. In the dataset, we included diaries for these additional nights, increasing the number of nights recorded from these participants. Furthermore, because of multiple missing recordings, typically due to the failure of the EEG device participants were wearing, five participants were requested to take part in the study for a second week. Our day-to-day analyses assume that all data comes from subsequent days and nights. This assumption is violated if participants do not repeat the study on subsequent weeks. This was the case for four of these participants, from whom partial data from the first week was discarded and only the second week was used. The fifth participant, however, took part in the study on two consecutive weeks: therefore, we included data from all 14 nights (some with missing EEG data, but all with valid diaries) from this participant.

In sum, while BSETS is not designed to be a population-representative sample, we found that participants are reasonably demographically diverse. Although most of our participants are young, and well-educated, participants with other demographic characteristics are also represented. We found high compliance with our protocol, with only 4% of participants dropping out of the study completely and over 95% of the nights from active participants providing diary and EEG data. The relatively young age of participants may have been a factor supporting high compliance, and in elderly or pathological population samples, which are more commonly studied in the sleep research field, may have different habits or engagement styles with respect to smartphone-activated technology or worn devices, and potential compliance issues arising from this should be considered in future research.

Quantitative EEG measures

Dreem 2 devices record EEG from five channels with various desirable and undesirable characteristics due to the topography of their references. Channels with more distant fronto-occipital references more closely approximate standard monopolar laboratory PSG, but are likely more contaminated by noise due to the less robust occipital electrodes of Dreem 2. Fronto-frontal channels are less likely to accurately measure topographically widespread activity such as sleep spindles [66, 67], and especially also slow waves [68], but they are also less likely to be contaminated by noise (Fig 2). Various definitions of the automatically detected artifacts are possible (see Methods), which may eliminate some of the problems resulting from the presence of noise. We inspected the mean PSDs from multiple channels and artifact detection methods in order to examine which method of quantitative EEG registration has the most favorable characteristics.

Download:

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

Fig 2. The distribution of channel quality (proportion of epochs with <50% artifact probability).

https://doi.org/10.1371/journal.pone.0288909.g002

First, we analyzed data availability (loss of channels/nights due to artifacts) and data quality (visual inspection of spectra and the ability to reproduce their known correlations with age) using four artifact definition methods. The results are summarized in Table 7.

Download:

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

Table 7. The effect of various artifact definitions on data availability and the reproducibility of the age-PSD correlation.

https://doi.org/10.1371/journal.pone.0288909.t007

We found that data availability is excellent on fronto-frontal channels, most prominently the F7-F8 channel, wherein over 80% of available nights provided a PSD estimate even with the most stringent artifact definition. However, data availability was significantly reduced on fronto-occipital channels (F7-O1, F8-O2), where more stringent methods excluded over half of nights as too noisy to provide a PSD estimate. Most of the channel quality (proportion of non-artifactual epochs) variance (56.7%-79.7%) was accounted for by participant identity, suggesting that some participants may have been more conscientious in their use of the Dreem 2 EEG devices. Conversely, because multiple recordings were available for most participants, the measure of average PSD could be calculated for a much higher proportion of participants than from individual nights.

On Fig 3, we illustrate findings from one definition which discards all epochs with a >25% artifact probability and all channels with an overall quality of <20%. We highlight PSDs by night and by participant on a channel with high data availability, F8-F7. Overall, we found that both night- and individual-level average PSD estimates resemble those obtained using standard PSG, with typical delta and sigma peaks. As expected given the frontal topography of key electrodes, sigma peak frequency was low at ∼12 Hz, possibly indicating slow rather than fast spindles. A minority of nights (and, due to the high interindividual similarity of channel quality, participants) still may be affected by artifacts, which most frequently contaminated activities above the sigma range. As we anticipated, fronto-frontal channels had lower voltage than fronto-occipital channels, but even these (with the exception of Fp1-F7) retained characteristic spindle peaks. The negative correlation between PSD, and NREM delta and sigma activity, was present on all channels, but most prominent on fronto-occipital channels, especially in the delta range. Similar visualizations are provided in the Supplementary data for other artifact definitions.

Download:

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

Fig 3. An illustration of qEEG data in BSETS.

Panel A: grand average PSDs calculated across all available nights, displayed by channel. Note the higher voltage of fronto-occipital channels and the relative absence of spindle peaks from the Fp1-F7 channel with a close bipolar reference. Panel B: the correlation between individual average PSD (in 0.5 Hz bins) and age. Panel C: average PSD estimates of all nights from F8-F7 (thin grey lines), overlain with the grand average (thick black line). Panel D: individual average PSD estimates (averages across all nights from the same participant) from F8-F7. All data is shown after the exclusion of channels with <20% data quality and epochs with >25% artifact probability.

https://doi.org/10.1371/journal.pone.0288909.g003

Quantitative EEG analyses in BSETS are limited by the frontal topography of available EEG channels. We quantitively assessed the scope of this limitation by comparing PSD estimates from frontal and other channels in an external dataset with a full PSG cap (S1 Text). We found that between-participant correlations of PSD estimates between frontal and other channels are reasonably high (typically r = 0.7–0.8). A local minimum is observed in the fast spindle frequency range, but even here a minimal correlation of >0.5 is observed. This suggests that the frontal PSD estimates in BSETS are reasonable proxies for activities which occur without a characteristic frontal topography.

In sum, our preliminary analyses found that the quality of EEG data in BSETS is sufficient for qEEG analysis. We tested the feasibility of this approach by calculating a simple and well-known EEG metric, PSD, and investigate if findings in the literature can be replicated using BSETS EEG data. While artifacts contaminated all channels, especially fronto-occipitally ones, this problem could be mitigated using a complimentary automatic artifact detection algorithm, which performed reasonably well regardless of settings. The majority of data from fronto-frontal and a large minority from fronto-occipital channels could be recovered even on a night-by-night basis. The resulting PSD values exhibited typical delta and sigma peaks and the expected correlations with age. Encouragingly, these findings were present even for the fronto-frontal F8-F7 channel which has high data availability. While our goal was to demonstrate how a fully automatic pipeline can process BSETS EEG data, we note that an additional visual inspection to eliminate nights with abnormal EEG spectra could further improve data quality.

An additional benefit of our use of mobile EEG devices [69] over laboratory PSG is the ease of protocol adherence and ecological validity, as participants can record EEG in their homes. While laboratory recordings are often affected by first night effects [70], recent research has shown that there is relatively little difference between EEG recordings undertaken in the laboratory or at home [71].

Examples of research opportunities

BSETS can be used to investigate a large number of hypotheses related to the relationship between sleep, normal and pathological human traits, and daily experiences. Here we enumerate a number of currently unanswered research questions which can be readily clarified using BSETS data.

1. It is known from some studies that at least moderate associations exist between self-reported circadian timing and objective (usually actigraphic) measures of actual sleep and wake timing [72–74]. However, to our knowledge no study compared to self-reported and EEG-measured patterns of the timing of daily activity.
2. Self-rated sleep quality has low but significant correlations with PSG-rated sleep structure [25]. Data from this question, however, comes from between-person studies. It is unknown if and to what extent within-person variation in subjective sleep quality is related to within-subject variation in objective sleep quality ‐ in other words, if objectively better sleep on a night leads to higher subjective ratings by the same person.
3. There is widespread belief that sexual intercourse improves sleep, with some empirical comparison in a recent diary-based study with subjective sleep reports [23]. It is unknown if sexual intercourse also improves objectively measured sleep. Similar replications are warranted for all previous findings relying exclusively on self-reports ‐ in other words, it is important if a daily experience affects objectively measured sleep or only the subjective perception of it.
4. There are numerous algorithms developed to detect morphological features, such as sleep spindles, in laboratory EEG data [75, 76]. While these algorithms may not be immediately applicable to mobile EEG data, after the development and validation of new algorithms morphological features can be detected and used as dependent variables in BSETS EEG data.

At the Editor’s suggestion, we include a proof of principle analysis investigating the effects of exercise on sleep. A recent daily observational study [77], similar in design to BSETS, found increased REM latency, reduced REM and slightly increased N3 after mild, but not after vigorous exercise in the elderly. A recent meta-analysis [16], following previous similar studies [18, 78, 79] highlights that current recommendations advise against exercise in sleep, and quantitively synthesizes 23 investigating the effect of evening exercise on sleep. It revealed borderline significant increases in REM latency and SWS, and a highly significant decrease in N1. However, the largest study in the meta-analysis had only 24 participants, and the total number of across all studies in the meta-analysis was approximately equal to the intended final size of BSETS (N = 266). We attempted to replicate these findings using a within-participant design with multilevel regression models, using EEG-derived sleep structure as the dependent variable, and the exercise and a random intercept per participant as independent variables, implemented in the lmer() function in the lmerTest R package, which is similar to lme4 but provides p-values from multilevel models using Satterthwaite’s approximation. Due to violations of the homoskedasticity and normality of residuals assumptions (see also Table 6) we log-transformed WASO before model fitting and for sleep onset latency and REM latency we specified a gamma distribution with a log link function, implemented with the glmer() function. We controlled all models for day of the week (specified as weekend or weekday). We found that exercise slightly but significantly (p = 0.006) increases sleep onset latency, but not other sleep macrostructure variables (Table 8).

Download:

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

Table 8. The effects of participant-reported exercise on subsequent sleep macrostructure.

https://doi.org/10.1371/journal.pone.0288909.t008

Our findings do not directly replicate the results of the previous meta-analysis or a recent daily observational study [77], but suggests that exercise may complicate the onset of sleep. The limitations of our analysis include that we did not control for the time of exercise relative to sleep, and thus could not distinguish between exercise any time during the day and specifically in the evening.

BSETS is designed to be an accessible database which can be used and re-used by both academic and independent researchers for a plethora of purposes, the most important among them being to further explore relationships between what we do, how we sleep, and whether that might change depending on who or what we are, psychologically. Our hope is that this dataset will empower scientists to address longstanding lacunae in sleep literature, and to enhance our understanding of what influences sleep, what sleep influences, and how. BSETS is projected to finalize data collection and cleaning by Summer 2023. Once data collection and preprocessing is completed by late 2023, the BSETS research is open for requests to release data or collaborate with others interested in our resources.

Participants in BSETS are mainly young and healthy. Nevertheless, we envision BSETS findings to have significant clinical applications for the elderly and for those with sleep disorders, for example, Insomnia disorder. This is because the substantial statistical power of BSETS can be leveraged to identify DAEs which are associated with changes in subsequent sleep quality and if such DAEs are found, they can be used to design evidence-based interventions in at-risk samples. For example, if certain dietary patterns are associated with improved sleep in BSETS, at-risk individuals can be encouraged to follow this dietary pattern to improve their sleep as well. Importantly, because such an intervention is based on an explicit hypothesis in contrast to the exploratory BSETS, a smaller sample can enough to confirm efficacy. In these smaller samples, the use of standard laboratory PSG may be feasible if, for example, an elderly population has technical issues with the use of smartphone-based mobile EEG technology in their homes.

Limitations

We designed BSETS to address several common issues affecting research into bidirectional relationships between daily experiences and sleep, most importantly issues concerning statistical power and the objective measurement of sleep. Nevertheless, BSETS has its own limitations which we discuss here.

During recruitment to BSETS, we prioritized sample size and high adherence to demographic representativity. As a result, BSETS participants are mainly healthy young participants, and its main goal is to study the effects of common daily experiences on normal sleep. BSETS findings may not automatically generalize to an elderly population or to those with sleep disorders. However, this limitation can be overcome in BSETS-based studies by using between-individual variation as a moderator. In other words, we can investigate if a certain relationship between DAEs and sleep is different as a function of demographic (e.g. age and sex) or health-related (e.g. trait depression, overall sleep quality) characteristics. Furthermore, BSETS data can be used to identify DAEs with effects on normal sleep, which can then be used in targeted interventions in more focused studies of the elderly or those with sleep disorders.

Our mobile EEG devices enable an accurate automatic extraction of hypnograms [63], and, as our own analyses presented here show, provide adequate recording quality to be recordings for at least basic quantitative EEG analyses. However, these devices only contain frontal and occipital electrodes, resulting in a limited topographical mapping of EEG activities. Most significantly, the lack of central channels affects our ability to detect fast spindles, although quantative analyses (S1 Text) suggest that frontal PSD estimates may be reasonable proxies for activity with other topographies. Furthermore, signal quality may not be sufficient for certain quantitative analyses, and there is overall little literature about quantitative EEG analyses with mobile dry EEG devices.

An important conceptual limitation comes from the fact that by observing the same individuals over multiple days in BSETS, it is only daily experiences with non-zero within-individual variance that is informative. Simply put, if participants experience very similar experiences and very similar sleep on all days, we can learn little about the relationship between these two variables in a daily observational study like BSETS, which effectively uses participants as their own controls. However, previous studies [20, 22, 24] reported moderate intraclass correlation coefficients, that is, indicators of within-individual similarity, of daily experiences and sleep variables, suggesting that in practice, we will see sufficient within-individual variance in BSETS to enable meaningful analyses.

As we aimed at a broad sampling of traits and daily experiences in our questionnaire-based measurements in BSETS, in some cases we were limited by practical constraints about the amount of information we could obtain using questionnaires about specific traits/experiences. For example, most daily experiences are only reported as either having or not having occurred, with no additional information such as for example, the specific time of day an event occurred, which may affect its ability to affect sleep. Our participants did not recruit caffeine intake, which we excluded due to the existence of a high-quality literature (see e.g. [15, 80]) on the effects of caffeine on sleep.

Finally, as a general limitation, while BSETS contains information about a wide variety of traits, experiences and sleep variables, there is a practically infinite number of possible relationships of interest between these constructs. Therefore, there will always be important research hypotheses about the relationship between daily experiences, traits and sleep which cannot be addressed with BSETS data.