Beyond screentime: a 7-day mobile tracking study among college students to disentangle smartphone screentime and content effects on sleep

ABSTRACT

Health professionals and scholars alike consider evening smartphone use an important public health challenge. Existing research on the effects of smartphone usage on sleep has three limitations, namely reliance on self-report measures of smartphone use and/or sleep, limited attention to within-person effects, and a focus on general screentime rather than specific app usage. The current study addressed these limitations by conducting a seven-day study assessing smartphone use and sleep with a combination of subjective and objective measures among 75 students. The findings do not support the assumption that evening smartphone interferes with sleep. We did observe between-person relationships between specific indicators of smartphone usage and sleep, e.g. sleep quality was positively related to the use of meditation apps, and negatively related to the use of work-related apps. These findings indicate research should focus on what individuals do on their phones instead of how much time they spend.

KEYWORDS:

• Sleep
• smartphones
• data donation
• media effects
• well-being

Smartphones have become the dominant platform through which individuals consume media before bedtime (e.g. Chung, An, and Suh 2020; Rod et al. 2018; Van Kerkhof et al. 2017). Health professionals and scholars alike consider evening smartphone use an important public health challenge (Rod et al. 2018), because using a smartphone during the evening and especially right before bed is expected to negatively impact sleep. These health concerns are reflected in the advice that smartphones should not be used in the one to two hours before going to bed (e.g. LeBourgeois et al. 2017). Surprisingly, though, there is very little empirical evidence supporting this advice.

There is a burgeoning body of research on the negative effects of smartphone use on sleep. Recent reviews demonstrate that there is a small negative relationship (see, for example, Hale et al. 2019; Mac Cárthaigh, Griffin, and Perry 2020; Yang et al. 2020) or no relationship (e.g. Lund et al. 2021) between smartphone use and sleep. However, most of the studies included in these reviews have at least three important methodological limitations which confine our understanding of the link between smartphone use and sleep.

First, the complex nature of smartphone use is commonly ignored as researchers have focused on the effects of general time spent on smartphones (e.g. Hale et al. 2019; Orben and Przybylski 2020). This narrow focus on screentime is problematic as individuals differ highly in how many and which smartphone apps they engage with before going to bed, and differences in app usage patterns might affect sleep in unique ways (e.g. Huiberts, Opperhuizen, and Schlangen 2022). Therefore, scholars have called for moving beyond investigating screentime and suggest to examine the effects of specific smartphone behaviours, for example switching behaviour and content (e.g. Kaye et al. 2020).

Second, most studies predominantly used self-reported measures of sleep and/or smartphone use, thus neither or merely one of them was measured objectively (e.g. So et al. 2021). The sole reliance on self-reported measures is problematic for measures of media use (e.g. Naab, Karnowski, and Schlütz 2019; Parry et al. 2021) as well as sleep (Perez-Pozuelo et al. 2020). Therefore, Mac Cárthaigh, Griffin, and Perry (2020) identified improving the operationalisation of smartphone use as well as sleep as an important way to move the field forward. Specifically, Mac Cárthaigh, Griffin, and Perry (2020) argued that tracking of smartphone data has been underutilised, and that sleep requires researchers to include objective and subjective measures as both capture unique aspects of sleep.

Third, most existing studies employ cross-sectional designs or longitudinal designs with large time intervals (see also review by Brautsch et al. 2023). Although these studies provided us with preliminary evidence on the relationship between smartphone use and sleep, they offer only a partial view of this relationship. Specifically, cross-sectional studies only examine between-person processes (e.g. those using their smartphones more sleep less well) and neglect within-person dynamics in this relationship (e.g. do individuals sleep less after using their smartphone more on the previous evening?). Although longitudinal designs might inform us about the potential long-term effects, they rarely study the effects of smartphone use on sleep during the ensuing night at the level of the individual. Thus, at the moment we lack studies that make it possible to test whether evening smartphone use indeed has a short-term effect on sleep.

Due to these gaps in the literature, we currently know relatively little about whether and which types of smartphone use impact sleep in the night following smartphone use. The current study aims to remedy these existing gaps by (1) extending the operationalisation of smartphone behaviour through complementing time spent on smartphones with more specific smartphone behaviour data, namely number of unique apps used and exposure to specific app types, (2) adopting objective and subjective measures of sleep and tracking data for smartphone use, and (3) using a longitudinal design measuring smartphone use and sleep across seven consecutive days.

1.1 Evening smartphone use – screentime and sleep

There is a long tradition in media effects research studying the impact of electronic media use on sleep (e.g. Exelmans and Van den Bulck 2019). The unique affordances of smartphones – including their portability – led scholars to argue that smartphone-based media use might be especially impactful in the context of sleep. In a representative study among Dutch adults (N = 15,690) pre-bedtime smartphone use was especially characteristic of young adults; 52% of young adults indicated using their smartphone in the hour before bed, while this was reported by only 17% of the over 70-year-olds (Van Kerkhof et al. 2017). Several other studies also demonstrated that pre-bedtime smartphone use is a common practice among young adults. For example, in a study among 815 Danish young adults (Rod et al. 2018), 75% reported using their smartphone in the hour before bed. Moreover, research among Korean young adults demonstrated that smartphone usage was the most common form of leisure activity three hours before bed, especially among young adults who were high in so-called bedtime procrastination, i.e. going to bed later than intended (Chung, An, and Suh 2020). Studies to date have not consistently distinguished between evening smartphone use and pre-bedtime smartphone use. For this reason, in the current study we will explore to what extent effects are time contingent, distinguishing between evening (post-7pm) and pre-bedtime (two hours before sleep) smartphone use.

Typically, three mechanisms are put forward to explain the potential detrimental effects of evening and/or pre-bedtime smartphone use on sleep, namely time displacement, blue and short-waved light emitted from smartphones, and arousal (see Hale et al. 2019). It is important to note that the first two mechanisms – time displacement and screen light – imply that the effects of smartphone use on sleep are closely related to the device itself and hence are independent of the content that individuals are exposed to on these devices. If this is the case, smartphone screentime should be a sufficient indicator for assessing effects on sleep. This might partly explain why most existing studies focused solely on screentime. However, if arousal is considered an important mechanism, then we would expect that mainly arousing smartphone content would affect sleep. As studies mostly ignore content, i.e. collapsing time spent with arousing and non-arousing smartphone apps, they may underestimate the potential effect of smartphone use on sleep.

Overall, most research on smartphone usage and sleep shows that the time people spend with digital devices is negatively related to sleep, considering one or more sleep indicators. For example, a meta-analysis of 11 studies conducted among children and adolescents (Carter et al. 2016) demonstrated that there were consistent negative relationships between portable media use and sleep quantity, quality and daytime sleepiness. The effect sizes can be qualified as small or small/medium as the odds ratios ranged between 1.46 and 2.72. Similarly, two recent reviews, namely Brautsch et al. (2023) and Hale et al. (2019), concluded that the time spent with digital media was consistently related to various indicators of sleep quantity and sleep quality.

Despite these consistent and seemingly robust findings, it is important to recognise that many of these studies relied on cross-sectional designs, which solely provide insights in the differences between individuals. The studies are informative in that they show that individuals who use their smartphones more have lower sleep duration and quality (between-person effect), however, they do not provide any evidence for causal effects of smartphone use on subsequent sleep for each person (within-person effect). Disentangling between-person from within-person relationships is crucial as the negative relationship between smartphone use and sleep observed in earlier research obscures the causal direction at the level of the individual. For example, a negative relationship between sleep and media use could mean that individuals who have problems falling asleep might use digital devices longer (see also Christensen et al. 2016). In this case, the causal direction between smartphone use and sleep contrasts with the current health concerns.

Interestingly, studies that adopt more rigid methodological approaches to examine the effect of screentime on subsequent sleep show more mixed results or only very small effects. For example, So et al. (2021) studied the relationship between self-reported evening media use and subsequent objectively measured sleep over a five-day period among children. This study found no effects of media use on subsequent sleep during these five days. Similarly, Orben and Przybylski (2020) used data derived from retrospective self-reports of technology use and those from diary-based measures of sleep, and found only very small effects for retrospective measures on sleep and inconsistent and insignificant effects of the likely more accurate diary-based measures. These studies suggest that it is important to investigate the effects on subsequent sleep in more depth as cross-sectional studies relying on retrospective measures are not able to draw any directional conclusions.

In line with the evidence for a relationship between smartphone use and sleep from cross-sectional studies, we expect the following:

H1 – There is a small negative relationship between evening smartphone screentime and subsequent sleep, with evening smartphone use predicting more nighttime awakenings (H1a), shorter sleep duration (H1b), lower sleep quality (H1c), and longer sleep latency (H1d) both at the between-person level and the within-person level.

1.2 Beyond screentime: app count and app categories

Although most studies focused solely on the effects of smartphone screentime on sleep, an increasing number of scholars argue that screentime is an inadequate predictor for well-being in general (e.g. Bayer, Triệu, and Ellison 2020), and for sleep specifically (e.g. Orben and Przybylski 2020). Most importantly, what individuals do on their smartphones can vary substantially, and the effects of using a meditation app might be very different than those of watching an exciting TikTok video. This is particularly important as cognitive and physiological arousal are viewed as important underlying mechanisms in the relationship between digital media use and sleep (e.g. Exelmans and Van den Bulck 2017). Thus, engaging in arousing smartphone behaviours should be more detrimental than engaging in relaxing smartphone behaviours. To date, there is no definite theoretical framework that would explain which specific smartphone behaviours are detrimental for sleep.

We argue here that it might be fruitful to not only examine the various types of apps that someone uses, but also their usage patterns. This distinction mirrors the categorisation of internet use proposed by Blank and Groselj (2014) who stated that due to the breadth of possible uses of the internet, it was pertinent to find more meaningful indicators of internet use beyond time spent online. They introduced a three dimensional approach of internet use including (1) general time spent online, (2) the number of different online activities people engaged in, and (3) the type of activities that described people’s online behaviour. In the current study, we complement general time spent on smartphone apps, with two additional indicators of smartphone behaviour, specifically (1) app count (i.e. how many different apps are used) and (2) app categories (i.e. which types of apps are used).

App count. With regards to the concept of app count, research on the effects of general media use hints at the possibility that the number of types of media content people engage with during specific time slots could moderate the effects of screentime on sleep. With regards to the direction of the moderating effect of the sheer number of types of content people are exposed to, two contrasting hypotheses have been voiced.

It could be argued that a higher number of unique apps – indicative of smartphone users switching between apps – could be a sign of experienced boredom (e.g. Matic, Pielot, and Oliver 2015). Switching between apps might in this case indicate that the apps do not fulfill momentary needs (e.g. no new posts on Instagram, no interesting video to watch on YouTube). Consequently, individuals might decide to put their phone away and go to sleep. At least one study provides evidence for a negative relationship between media content diversity and sleep problems. For example, Orzech et al. (2016; data collected in 2011 among 261 first-year-students) demonstrated that when participants used more different types of media in the hour before bed this was related to earlier bedtime, more time in bed, and longer sleep duration. This study would suggest that a smartphone usage pattern characterised by a higher number of unique app categories might be beneficial for sleep.

However, engagement with a high number of app categories can also be viewed as an indicator of switching and/or multitasking. In this case, we may expect that this pattern of smartphone usage, i.e. engagement with higher number of unique app categories, would be detrimental for sleep. The literature on media multitasking has consistently demonstrated that switching among tasks results in increased arousal which is known to hinder cognitive and physiological downregulation making it more difficult to fall asleep (e.g. Yeykelis, Cummings, and Reeves 2014). This idea is supported by cross-sectional evidence of a negative link between media multitasking and sleep (e.g. van der Schuur et al. 2018).

Thus, due to current contradictory theory and empirical evidence, we formulate the following research question,

RQ1: How is app count (i.e. the number of unique apps used in the evening or pre-bedtime) related to sleep, i.e. number of awakenings, sleep duration, sleep quality and sleep latency at the between- and within-person level?

App categories. In addition to app count, it is also important to consider the possible unique impact of app categories. It can be argued that certain types of apps may disrupt sleep whereas other types of apps might promote sleep. No comprehensive theoretical framework for which types of apps could be more detrimental for sleep exists to date. However, research on traditional media and general screentime can be used to inform initial ideas about how different types of smartphone apps can be distinguished from each other in a meaningful way. In this study, we focus on app categories instead of a more granular content approach since the tracking of actual content is methodologically difficult.

Based on previous research, we used the theoretically most relevant and most frequently occurring app categories as a starting point. Thus, instead of investigating the effects of single apps, or a large amount of app categories (e.g. 27 from the Appstore), we focus here on seven categories that are theoretically relevant and for which we can expect sufficient variation in use among young adults (Schulz van Endert and Mohr 2020). The seven content categories (see also Table 1) are (1) video content, (2) social media (platforms), (3) social media (interpersonal mediated communication, including messaging and (video) calling), (4) games, (5) work and study, (6) music and audio, and (7) meditation apps. Although no studies have looked at the seven categories in the context of smartphone-based app usage’s impact on sleep, existing research provides us with some indication of expected relationships.

Table 1. App categories and descriptions.

Category	Description	Example*
Generic app categories
General Top 100	Apps from Top 100 most frequently used apps which could not be categorised.	n/a
Non Top 100	Apps outside Top 100 most frequently used apps.	n/a
Home Lock Screen	Unlocking of the phone.	n/a
Content specific app categories
1. Video Content	Apps to stream video content, including short and longform	YouTube
2. Social Media: Platforms (PF)	Apps to access image based social media platforms	Instagram
3. Social Media: Interpersonal communication (IC)	Apps for communication and messaging tailored to 1-to-1 or (small) group conversations	WhatsApp
4. Gaming	Apps to play (casual) smartphone games	Dots & Co
5. Work/Study	Productivity apps mainly used for non-entertainment purposes	Canvas (student app) Gmail
6. Music/Audio	Apps that provide audio content	Spotify
7. Relaxation	Apps used to relax, focus, and meditate	Calm

Note. The top panel lists generic app categories that do not match content specific categories which were theory driven.

With regards to the possible effects of exposure to video content, we build on traditional media effect research. Early studies on media use and sleep, focused primarily on the effects of bedroom TV on children’s sleep, and showed that bedroom TV had a negative effect on sleep (e.g. Nuutinen, Ray, and Roos 2013). As watching TV and videos on smartphones has become a common smartphone activity, we expected that these findings will generalise to smartphone-based video viewing. In line with this expectation, one cross-sectional study among adolescents showed that YouTube use was related to longer sleep latency (Varghese et al. 2021).

Following the emergence of social media, many scholars have studied their effect on sleep (e.g. Bhat et al. 2018; Levenson et al. 2017; Scott, Biello, and Woods 2019). Several studies support the idea that social media use, particularly before bedtime, is negatively related to sleep duration and sleep quality (e.g. Levenson et al. 2017). It has been argued that social media use is a particularly arousing pre-bedtime activity when it drives social interactions (Scott, Biello, and Woods 2019). Moreover, engaging in mediated-communication right before bed may not only trigger social interaction, e.g. incoming messages from others, but may also result in higher levels of smartphone-based social interaction vigilance, e.g. anticipating incoming messages after having fallen asleep. Both of which could impair sleep through a higher number of nighttime awakenings. For example, Rosen et al. (2016) showed that almost half of young adults check their phone at least once a night for messages. Based on this argument it might be especially interesting to distinguish between social media based on the centrality of social interaction. For this reason, the current study will attempt to make a first step in this direction by distinguishing platform-based social media (e.g. Instagram, Facebook) from interpersonal-communication social media (e.g. WhatsApp, Zoom).

Although television viewing and social media use seem to be most frequently mentioned in debates about media and sleep, research has also examined the relationship between sleep and the four other categories included in this study, i.e. gaming, work/study content, meditation and music/audio content. For example, Peracchia and Curcio (2018) conclude that gaming, particularly in the evening, is related to lower sleep quantity and sleep quality. In addition, smartphones do not only facilitate access to entertainment content but provide opportunities to engage in work and study-related activities outside school and work hours, even when already in bed. Several experience sampling (ESM) studies showed that using a smartphone for work-related tasks in the evening led to later bedtimes as well as lower sleep quality and interfered with psychological recovery (e.g. Lanaj, Johnson, and Barnes 2014; Sonnentag, Binnewies, and Mojza 2008).

For the last two categories (meditation and audio/music) it might be argued that these types of apps will not negatively impact sleep. Considering that exposure to blue light is a frequently mentioned argument for why smartphone use may disrupt sleep, it is important to recognise that meditation apps and apps that provide access to music and audio content make it possible to engage with the content without being exposed to screen light. Moreover, music is even seen as a sleep facilitator and engagement in meditation activities has at least a moderate positive impact on sleep (Kirk et al. 2022).

We currently lack studies that have specifically tested the impact of smartphone-based media activities on sleep. Moreover, there are no studies available that systematically assessed the unique impact of different types of app usage patterns based on objective measures of time spent on these apps while distinguishing between different categories. For this reason, we pose a second research question:

RQ2: To what extent does time spent with the pre-identified app categories in the evening relate to sleep, i.e. number of awakenings, sleep duration, sleep quality and sleep latency?

1.3 Current study

In sum, the current study will address the methodological gaps observed in previous studies to provide stronger insights into the relationship between smartphone use and sleep, specifically among college students. For this reason, the current study employs a 7-day longitudinal design to assess how objective measures of evening smartphone use, i.e. either from 7pm onwards or two hours before bedtime, are related to four sleep indicators, namely sleep duration (objective measure), sleep awakenings (objective measure), sleep quality (self-reported) and sleep latency (self-reported). In addition, the study will distinguish within- from between-person effects when assessing the relationship between smartphone use and sleep. Finally, in contrast to earlier research this study will not merely identify the effects of time spent using one’s smartphone, but also the unique effects of the number of unique apps accessed and engagement with specific app categories on sleep.

2. Method

2.1 Participants and procedure

After receiving ethical approval from the ethical review board of the Communication Science Department at the University of Amsterdam, we recruited 125 students (iPhone users) via the student recruitment pool for research. Students at this institute take approximately two classes per semester with each class meeting twice a week for two hour sessions. These classes are scheduled between 9am and 5pm between Monday and Friday. Most course work is independent study work which our student participants can work on at any time and schedule autonomously.

The study consisted of three stages. During stage 1, participants attended a one-hour introduction meeting in groups of five at the lab facilities. During this meeting, participants were instructed about the study, were given a Fitbit, and were made familiar with all technical aspects of the study. Participants also downloaded the MyPanel research app (mobilemarketresearch.com) on their phone which was used to administer daily short surveys. Finally, they completed an online survey to collect background information.

During stage 2, i.e. the next seven days, participants were asked to upload a screen video of their battery section every morning (see Measures below). A link to a how-to video was accessible via the app in case participants needed a reminder of the instructions on how to screen record and upload the videos. In addition, participants completed short surveys about sleep indicators in the morning. At stage 3, having completed the 7-day study, participants returned to the lab where they completed a final survey and returned the Fitbit. Participants could choose between 10€ or research credits for participation.

From the initial sample, we received smartphone data from 95 students. Of those, 93 also had accurate Fitbit data. For some participants, Fitbit data was missing due to malfunctioning of the Fitbit bracelet, forgetting to put it back on after removing it, or because the Fitbit ran out of battery. Of the 93 participants, 75 students (66% females; M_age= 21.02, SD_age= 3.08) reported evening smartphone use (see Measures below), and provided a total of 377 observations with an average of 5.02 nights per participant. Seventy-one students reported pre-bedtime smartphone use (see ‘evening smartphone use section below), and provided a total of 321 observations with an average of 4.52 nights per participant. As additional missing data occurred in some of our outcome variables; Table 4 denotes the number of observations.

2.2 Measures

Descriptive information for all core measures and relationships with the four sleep parameters are presented in Tables 2 and 3.

Table 2. Means and standard deviations.

	Evening: post-7pm		Pre-bedtime 2 hours before bedtime
	M	SD	M	SD
Sleep parameters
Number of Awakenings, count	1.27	1.30	1.24	1.29
Duration in minutes	472.68	108.37	477.46	109.63
Quality (low = 1, high = 5)	2.95	0.99	2.94	1.02
Latency in minutes	23.49	27.07	24.07	27.23
Time per app category in minutes
General Top 100	19.81	24.23	5.88	10.67
Non Top 100	7.01	19.85	1.26	3.35
Home Lock Screen	8.89	6.81	1.96	1.62
Video Content	11.68	24.30	6.01	14.55
Social Media: Platforms	23.32	27.05	6.22	9.67
Social media: Interpersonal communication	25.44	33.37	5.21	8.91
Gaming	1.32	7.60	0.94	5.02
Work/Study	4.96	7.60	0.98	2.41
Music/Audio	5.99	15.16	1.52	5.46
Relaxation	0.75	7.86	0.27	3.27

Table 3. Correlations of main constructs with sleep indicators.

	Sleep
	1	2	3	4
Sleep
1. Awakenings
2. Sleep duration	.15**
3. Sleep quality	−.03	0.31***
4. Sleep latency	.11*	−0.04	−0.21***	-
Apps
Generic app categories
1. General top 100	−.10	−0.08	−0.09	0.08
2. Non top 100	−.05	−0.08	−0.14**	0.16**
3. Home lock screen	−.05	−0.13*	−0.11*	0.05
Content app categories
1. Video content	−.07	0.09	0.09	0.13*
2. Social media: PF^a	−.06	−0.07	0.07	−0.03
3. Social media: IC^a	−.03	−0.10	−0.14**	0.04
4. Gaming	.00	0.04	0.11*	−0.06
5. Work/study	−.09	−0.04	−0.11*	0.05
6. Music/audio	.05	−0.09	0.02	−0.01
7. Meditation	−.06	0.06	0.17**	0.00

Note. Zero-order pearson correlations with pairwise-deletion.

^a PF = platforms, IC = interpersonal communication. *p < .05, **p < .01, ***p < .001. Analyses based on evening use data.

Evening smartphone use. Evening smartphone use was assessed by letting participants upload screen videos of their iOS battery section every morning. These videos covered the battery section information from participants’ evening smartphone use (from 7pm onwards). The battery section provides an account of the minutes each app is used per hour. These videos were automatically processed with a Python script (see Baumgartner et al. 2023). This script processed iPhone screen recordings of the battery section pages and provides a CSV file with the following information: which apps were used at each hour of the day, time in minutes that each app was open on screen per hour, and time in minutes an app was open in the background per hour. The script was validated by comparing the automatically obtained coding of 19 videos which included over 1000 app entries with coding by a manual coder, which resulted in an 86% accuracy rate. A detailed account of this procedure, including validation of the script, is described elsewhere (Baumgartner et al. 2023).

Based on these videos, two measures for evening smartphone use were gained. First, evening smartphone use covering all apps used from 7pm onwards until an individual’s sleep time (as assessed with the Fitbit). In line with previous studies (e.g. Orzech et al. 2016), and to make our findings more comparable to other studies, we also calculated smartphone use time in the two hours before a participant fell asleep as indicated by their Fitbit data. Since our data donation approach only provided app use data in hourly blocks, for data inclusion cutoff we considered the first 30 minutes within each hour as part of the hour preceding sleep onset. For instance, in the two-hour window analysis, for participants who fell asleep at 23:30, we consider smartphone use from 21:00 to 23:00, but for participants who fell asleep at 23:31, we consider smartphone use from 22:00 to 00:00.

Smartphone use: screentime, app count, and app categories. The total time participants spent on their phones in the evening or in the two hours before falling asleep was used as an indicator for App Time Overall. The number of unique apps that was used in a specific time period was the indicator for App Count. Finally, we manually categorised each of the 773 unique apps that participants used during the 7-day study in one of the seven content categories as described in Table 1. The sum of time spent with the apps in each of these categories was used as the indicator for App Time per Category.

Self-reported sleep quality and latency. A single item was used to measure sleep quality in the daily diary surveys (e.g. Lydon et al. 2016). Participants indicated how well they rated their last night’s sleep on a scale from 1 (poor) to 5 (excellent). In addition, participants indicated in minutes how long it took them to fall asleep the evening before (from the moment they closed their eyes) as an indicator of sleep onset latency.

Objective measures of sleep duration and sleep awakenings (Fitbit). Participants in this study wore a Fitbit flex for the whole seven days. Earlier research showed that Fitbits can provide reliable measurements of several sleep parameters (Thota 2020). Moreover, the Fitbit is comfortable to wear, has a long battery life and is water-resistant which means participants can wear the device for the full duration of a 7-day study without taking it off. We used number of nighttime awakenings and sleep duration in minutes as derived from the Fitbit raw data as objective measures of sleep. We compared the sleep duration data from the Fitbit with self-reports from the diaries (participants estimated the amount of time they were asleep) and found a strong overlap, r = .73, p < .001.

2.3 Analysis plan

We used a combination of Linear Mixed Models (Bates et al. 2015; Bolker et al. 2009; Yang et al. 2014) and a within- and between-person disaggregation approach to test H1 and to answer RQ1. For the linear mixed model, unless specified otherwise, we constructed mixed models with number of sleep awakenings measured with Fitbit (Awakenings), sleep duration measured with Fitbit (Duration), sleep quality as measured with self-reports (Quality), and sleep latency as measured with self-reports (Latency) as the outcome measures. Participants were entered as random intercepts, and either the number of unique apps (AppCount) or the time spent on apps (AppTime) as fixed effects. All models were run for the effects of evening smartphone use (i.e. post-7pm), as well as pre-bedtime smartphone use (i.e. two hours before sleep). Pre-bedtime smartphone use reflects the two hours before reported bedtime for each participant. All analyses were conducted in R 4.2.2. (R CoreTeam 2023; for the full R-script including all equations we refer the reader to the OSF project page [link to be added after unblinding]). The models were constructed using the lme4 package in R (Bates et al. 2015). The p-values were estimated using the lmerTest package (Kuznetsova, Brockhoff, and Bojesen 2018). All continuous outcomes were modelled using a Gaussian (normal) distribution and all count outcomes were modelled using a Poisson distribution.

To disentangle between- from within-person effects, predictors were centred in two ways, first using between-persons centering (cmc), where the mean value of each participant was subtracted by the mean value of that predictor and second, using within-persons centering (cwc), where each data point was subtracted by the mean value of each participant for that predictor (Curran and Bauer 2011; Wang and Maxwell 2015). This approach allowed us to investigate whether individuals who used their smartphones longer before falling asleep experienced more sleep problems than those who used their phones for a shorter amount of time (cmc). In addition, it allowed us to examine whether participants slept less in those nights that they had used their phones more than they usually did – reflecting short term interference of smartphone use at the level of the individual (cwc).

To answer RQ2, we constructed mixed models with similar sets of outcome measures and random effects as for H1, but for the fixed effects, we used the time spent on app categories, i.e. the seven content categories and the generic categories that were not content-specific, rather than overall screentime. The remaining apps, were collapsed into two additional categories, specifically a category containing apps that appeared in the top 100 most frequently used apps but could not be categorised under any of the main content categories (i.e. General Top 100 Apps), and one with apps that were not in the top 100 (i.e. Non Top 100 Apps). These two were added as additional predictors. Finally, unlocking of the screen was logged and included as a third non-category specific predictor (i.e. Home Lock Screen).

3. Results

3.1 The effects of evening and prebedtime app usage on sleep

For descriptive purposes, we present the zero-order correlations in Table 3. Next, for each model assessing the relationship between app usage and sleep, the person mean centred predictor (i.e. between-subjects effect, cwc) and the centred person mean predictor (i.e. within-subjects effect, cmc) is reported, respectively the model statistics presented in Table 4 are related to the objective sleep outcomes (Nighttime Awakenings and Sleep Duration) and those in Table 5 are related to te self-reported sleep outcomes (Sleep Quality and Sleep Latency). Significant relationships between app usage indicators and sleep related outcomes are described below.

Table 4. Number of nighttime awakenings and sleep duration predicted by app time and count.

	Night time awakenings^a				Sleep duration^b
	App time		App count		App time		App count
	7 pm to sleep	2 hours pre sleep	7 pm to sleep	2 hours pre sleep	7 pm to sleep	2 hours pre sleep	7 pm to sleep	2 hours pre sleep
Intercept	1.18 (0.01)	1.14 (0.01)	1.18 (0.01)	1.18 (0.01)	473.16 (6.18)***	477.82 (6.53)***	473.13 (6.10)***	477.65 (6.50)***
App usage cwc	1.002 (0.00)**	1.002 (0.00)**	1.00 (0.00)	0.99 (0.00)	−0.10 (0.09)	0.15 (0.33)	−1.35 (0.45)**	1.90 (1.66)
App usage cmc	1.001 (0.00)	1.001 (0.00)	0.99 (0.00)	1.00 (0.00)	−0.07 (0.12)	0.44 (0.46)	−0.51 (0.37)	−0.35 (2.43)
Model fit
AIC	1122.6	1122.6	953.6	1128.4	4620.71	3929.48	4594.40	3922.64
BIC	1138.4	1138.4	968.7	1138.7	4640.38	3948.33	4614.06	3941.50
Log Likelihood	−557.3	−557.3	−472.8	−560.2	−2305.35	−1959.74	−2292.20	−1956.32
No of obs	377	377	321	377	377	321	377	321
No group ID	75	75	71	75	75	71	75	71
Var: ID (SD)	0.16 (0.40)	0.16 (0.40)	0.17 (0.40)	0.17 (0.41)	568.4 (23.84)	375.9 (19.39)	554.2 (23.54)	355.4 (18.85)
Var: Res σ^2	0.61	0.61	0.63	0.61	11220.1	11720.3	10941.9	11729.0
Marginal R^2/conditional R^2	0.026/0.230	0.026/0.230	0.017/0.225	0.006/0.217	0.004/0.052	0.004/0.035	0.029/0.076	0.004/0.033
ICC	0.21	0.21	0.21	0.21	0.05	0.03	0.05	0.03

Note. Unstandardised regression coefficients, standard errors in parentheses.

^a Generalised linear mixed model fit by maximum likelihood, Laplace Approximation.

^b Generalised linear mixed model fit by maximum likelihood. cwc = person-mean centred predictor, cmc = centred person mean. *p < .05, **p < .01, ***p < .001.

Table 5. Sleep quality and sleep latency predicted by app time and count.

	Sleep quality				Sleep latency
	App time		App count		App time		App count
	7 pm to sleep	2 hours pre sleep	7 pm to sleep	7 pm to sleep	7 pm to sleep	2 hours pre sleep	7 pm to sleep	7 pm to sleep
Intercept	2.95 (0.69)***	2.94 (0.08)***	2.95 (0.07)***	2.94 (0.08)***	22.48 (2.11)***	22.93 (2.31)***	22.55 (2.08)***	22.94 (2.31)***
App usage cwc	−0.0004 (0.0007)	−0.0014 (0.0028)	−0.005 (0.004)	−0.0001 (0.014)	0.01 (0.02)	0.07 (0.07)	−0.03 (0.09)	−0.20 (0.33)
App usage cmc	0.0003 (0.001)	0.0059 (0.0052)	−0.007 (0.004)	0.005 (0.027)	0.03 (0.04)	−0.08 (0.15)	0.21 (0.12)	−0.54 (0.83)
Model fit
AIC	1060.27	922.05	1050.23	917.01	3468.04	2934.12	3440.86	2927.96
BIC	1079.84	940.83	1069.80	935.79	3476.42	2952.89	3460.43	2946.73
Log likelihood	−525.13	−456.02	−520.12	−453.50	−1719.23	−1462.06	−1715.43	−1458.98
No of obs	370	316	370	316	370	317	370	316
No group ID	75	71	75	71	75	71	75	71
Var: ID (SD)	0.17 (0.42)	0.20 (0.45)	0.16 (0.40)	0.20 (0.45)	219.4 (14.81)	251.9 (15.87)	211.4 (14.54)	251.4 (15.86)
Var: Res σ^2	0.81	0.84	0.81	0.85	501.75	472.82	501.50	473.83
Marginal R^2/conditional R^2	0.001/0.177	0.008/0.198	0.019/0.179	0.000/0.193	0.005/0.308	0.004/0.350	0.018/0.309	0.004/0.349
ICC	0.18	0.19	0.16	0.19	0.30	0.35	0.30	0.35

Note. Generalised linear mixed model fit by maximum likelihood. Unstandardised regression coefficients, standard errors in parentheses. Model 1 = grand-mean centred predictor, Model 2 = person-mean centred predictor and centred person mean. *p < .05, **p < .01, ***p < .001.

Number of nighttime awakenings. The number of nighttime awakenings was negatively related to evening smartphone use, cwc = 1.002 (.00), p < .01 and cmc = 1.001 (.00), ns. Individuals who used more apps tended to experience fewer nighttime awakenings (cwc); this effect was not observed at the within person level (cmc). The three other app usage indicators, i.e. pre-bedtime AppTime, evening AppCount and pre-bedtime AppCount were not significantly related to nighttime awakenings. For all model results see Table 4.

Sleep duration. Evening and pre-bedtime AppTime were not significantly related to sleep duration (see Table 4). The number of unique apps used in the evening (evening AppCount) was negatively related to sleep duration on the between-person level, cwc = −1.35 (0.45), p < .01, but not at the within-person level, cmc = −0.52, ns. Individuals who used more unique apps in the evening reported shorter sleep. No effects were found for pre-bedtime App Count on sleep duration, not at the between – nor at the within-person level.

Sleep quality. Evening and pre-bedtime AppTime and App Count were not significantly related to sleep quality (see Table 5).

Sleep latency. Evening and pre-bedtime AppTime and App Count were not related to sleep latency, neither at the between- nor at the within-person level (see Table 5).

3.2 The effects of different app categories on sleep indicators

To further investigate whether specific app categories predict the four sleep indicators, we conducted additional linear mixed models that predicted the four sleep indicators as a function of time spent per app category in the evening as well as pre-bedtime. An overview of all estimates can be found in Table 6.

Table 6. Time spent per app type in the evening and pre-bedtime predicting sleep.

	Awakenings^a		Duration^b		Quality		Latency^b
	Evening	Pre-bed	Evening	Pre-bed	Evening	Pre-bed	Evening	Pre-bed
Intercept	1.38 (0.41)	1.18 (0.41)	494.81 (10.88)	473.36 (11.4)	3.06 (0.10)	2.97 (0.11)	20.85 (3.11)	20.16 (3.12)
Top 100	0.99* (0.00)	0.99 (0.01)	−0.19 (0.24)	−0.49 (0.59)	−0.00 (0.00)	0.00 (0.01)	0.02 (0.06)	−0.05 (0.14)
Non-top 100	1.00 (0.00)	0.96 (0.02)	−0.18 (0.35)	−3.13 (1.81)	−0.00 (0.00)	−0.02 (0.02)	0.20* (0.09)	0.29 (0.40)
Home/lock	1.00 (0.01)	1.11*** (0.03)	−1.40 (0.97)	3.19 (3.96)	−0.01 (0.01)	−0.01 (0.04)	0.03 (0.23)	0.21 (0.85)
Video	1.00 (0.00)	1.00 (0.00)	0.48* (0.24)	0.69 (0.43)	0.00 (0.00)	−0.00 (0.00)	0.07 (0.06)	−0.02 (0.09)
Social media: PF	1.00 (0.00)	0.99 (0.01)	−0.25 (0.22)	0.49 (0.64)	0.00 (0.00)	0.01* (0.01)	−0.05 (0.05)	−0.08 (0.14)
Social media: IC	1.00 (0.00)	1.00 (0.01)	−0.16 (0.20)	−0.27 (0.71)	−0.00 (0.00)	−0.01 (0.01)	0.03 (0.05)	−0.26 (0.15)
Gaming	1.00 (0.01)	0.99 (0.01)	0.19 (0.76)	2.47 (1.24)	0.01 (0.01)	0.02 (0.01)	−0.04 (0.19)	0.04 (0.32)
Work/study	0.99 (0.01)	0.97 0.03	0.54 (0.89)	−0.57 (2.53)	0.00 (0.01)	−0.07** (0.02)	−0.17 (0.21)	3.22*** (0.51)
Music/audio	1.00 (0.00)	0.99 (0.01)	−0.57 (0.37)	−2.07 (1.13)	0.00 (0.00)	−0.01 (0.01)	0.05 (0.09)	0.72** (0.23)
Relaxation	0.98 (0.01)	0.99 (0.03)	0.92 (0.73)	1.88 (1.90)	0.02** (0.01)	0.04* (0.02)	0.03 (0.17)	0.13 (0.41)
Model fit
AIC	1130.9	953.6	4604.4	3907.3	1122.6	963.6	3473.7	2904.9
BIC	1178.1	998.8	4655.5	3956.3	1173.4	1012.4	3524.5	2953.7
Log likelihood	−553.43	−464.8	−2289.2	−1940.6	−548.3	−468.8	−1723.8	−1439.4
No of obs	377	321	377	321	370	316	370	316
No group ID	75	71	75	71	75	71	75	71
Var: residual σ^2	0.61	0.63	11149.06	11625.09	0.81	0.80	507.31	400.42
Marginal/conditional R^2	.067/.265	0.083/0.278	0.045/0.076	0.053/0.064	0.069/0.181	0.087/0.246	0.027/0.305	0.107/0.470
ICC	0.21	0.21	0.03	0.01	0.12	0.17	0.29	0.41

^a Generalised linear mixed model fit by maximum likelihood (Laplace approximation; Family: poisson, log); coefficients are odds ratios.

^b Linear mixed model fit by REML^,.t-tests use Satterthwaite's method, lmerModLmerTest. Uncl = unclassified; *p < 0.05, **p < 0.01, ***p < 0.001.

Evening app usage. Regarding the effects of specific exposure to specific app categories in the evening, we observed two significant relationships for content specific categories. Watching videos was associated with longer sleep duration, and the use of meditation apps was associated with better sleep quality. All other content specific predictors were nonsignificant. In addition, time spent with uncategorised Top 100 apps had a small negative relationship with number of awakenings (OR = 0.99), and a small positive relationship with sleep latency.

Prebedtime app usage. The use of meditation apps pre-bedtime was positively related to sleep quality, t = 2.34, p = .020. Interestingly, also the use of social media (platforms) in the two hours before falling asleep was positively related to sleep quality, t = 2.54, p = .012. Using work and study-related apps before falling asleep was associated with worse sleep quality t = −3.19, p = .002. Music and audio, and work/study app uses were associated with longer sleep latency, t = 3.12, p = .002 and t = 6.34, p < .001, respectively. Moreover, gaming was associated with longer sleep duration, t = 1.99, p = .048. The other predictors were nonsignificant, all t’s < 1.72, all p’s > .085. Table 6 provides a summary of these findings.

4. Discussion

That evening smartphone use interferes with sleep seems to be an almost universally accepted belief. However, due to the cross-sectional nature of most existing studies and their reliance on retrospective measures of smartphone use, robust evidence for the negative effect of smartphone use on sleep was still lacking (see for example Brautsch et al. 2023). We thus followed the call of researchers to investigate these effects with more advanced methodological approaches which go beyond self-reported screentime, recognise the complexity of smartphone behaviour, and specifically assess the within-person dynamics of smartphone use and sleep over time (e.g. Orben and Przybylski 2020). The findings of our study provide a more nuanced picture of the relationship between smartphone use and sleep. Our results are a cautionary tale for making sweeping statements that smartphone use in general will result in sleep problems.

First and foremost, the findings from this study do not seem to support the common assumption that evening smartphone use, specifically during the two hours pre-bedtime, has a negative effect on subsequent sleep (e.g. Exelmans & van den Bulck, 2016). Pre-bedtime app usage was not related to any of the four indicators of sleep – thus no support was found for H1 when looking at time spent on apps right before bed. In addition, only small effects were observed for app usage in the evening (post-7pm), and this was limited to number of nighttime awakenings. In this case, people who used their smartphones for longer after 7pm reported more nighttime awakenings. With regard to smartphone usage patterns (i.e. number of unique apps used; RQ1), we found no effects when considering the pre-bedtime time interval and only a small effect for usage post-7pm. Specifically, people who accessed more unique apps post-7pm had a shorter sleep duration. The latter effect could be seen as preliminary evidence for the time displacement hypothesis (e.g. Hale et al. 2019). It should be noted that all effects reflected between-subjects differences and we observed no evidence for effects of smartphone use on sleep experienced in the subsequent night.

The differences in effects observed for overall time spent on apps (H1) and the number of unique apps someone has engaged with (RQ1) highlight that relying solely on screentime might be problematic (Orben and Przybylski 2020). Although the magnitude of all observed effects were small, this was especially true for the overall screentime measure. The small effects are not surprising as smartphones are increasingly conceptualised as an ‘everything hub’ with the device being used for a wide range of activities from entertainment and socialising to working (e.g. Mason et al. 2022, 2). If the smartphone is used for ‘everything’ it seems evident that a general measure of screentime might obfuscate effects of specific smartphone activities which could even have opposite effects. This was also the underlying rationale for examining the effects of specific smartphone app categories on sleep.

Are smartphone effects on sleep app-content contingent? Although the effects observed for different categories were small, we did observe several app-specific effects on sleep (RQ2). In short, the results suggest that what someone does on their phone – as reflected in the type of app that smartphone users engage with – may be more telling than screentime itself.

First, engagement with apps from the meditation category, both in the evening as well as pre-bedtime, was related to better perceived sleep quality. This is in line with expectations that calming evening activities lead to better sleep quality (Kirk et al. 2022). This also supports previous empirical cross-sectional evidence for the positive effects of using a meditation app on sleep (Huberty et al. 2021a). Moreover, controlled trials reported positive long-term effects of meditation apps on sleep quality (Huberty et al. 2021b). The current study adds to these findings by showing that college students who choose to use meditation apps – as demonstrated by their smartphone log data – report better sleep quality.

Second, using so-called productivity apps (work/study category) during the two hours pre-bedtime, but not post-7pm, had a small negative relationship with sleep quality and latency, thus needing more time to fall asleep. This finding aligns with Lanaj, Johnson, and Barnes (2014) who reported negative effects of self-reported work-related smartphone use on sleep. It is expected that work-related tasks lead to cognitive rumination and activation (Vahle-Hinz et al. 2014). The present study supports these previous findings with objective measures of work-related smartphone use. It is interesting to note that using productivity apps in the evening does not seem to be detrimental per se as long as this does not occur immediately before going to bed. This might indicate that the negative effects of work/study-related engagement in the evening might be countered by engaging in recovery experiences before going to bed (Ebert et al. 2015).

Interestingly, the use of social media apps, both platform and messaging apps, during the evening or the two hours before bed, was not related to any of the sleep indicators. This finding is contrary to previous assumptions and cross-sectional findings indicating that social media use is detrimental for sleep (e.g. Scott, Biello, and Woods 2019). However, the few studies that employed ESM or longitudinal designs reveal similar findings to those observed in the current study. For example, Hamilton et al. (2020) found that adolescents who used more social media during the day went to bed later but did not sleep less. Similarly, Das-Friebel et al. (2020) did not find any effects of bedtime social media use on sleep during the subsequent night among students. Finally, a laboratory study also found no effects on arousal or sleep when participants used social media 30 minutes before bedtime (Combertaldi et al. 2021).

The discrepancy in findings between cross-sectional and more advanced studies (ESM, experiments) needs further scrutinisation. There are at least four explanations for this discrepancy across these two types of studies. First, cross-sectional studies are unable to assess short-term, causal effects. If these studies report relations, this could indicate that individuals who tend to suffer from sleep problems are more likely to use social media in the evening rather than vice versa. Second, the discrepancy between the different types of studies could indicate that the correlations found in cross-sectional studies were spurious and influenced by a third factor (e.g. traits such as anxiety that might be simultaneously related to social media use and sleep). Third, cross-sectional studies suffer from recall-bias and inaccuracies. In the context of sleep, people will most likely be able to give a subjective estimate of time spent sleeping rather than an objective estimation of their sleep duration and quality (e.g. Kaur et al. 2021). Finally, it is possible that as most cross-sectional studies rely on larger samples, they are able to detect even small effects that are of practical insignificance (see for example, the conclusion drawn by Orben and Przybylski 2020).

An alternative explanation for the absence of effects might be related to how social media is measured; both in the current study as well as earlier research. Most measurements of social media use, including the one adopted in this study, are still a proxy of people’s unique experiences on social media platforms and while exchanging social media messages. Consequently, these measures are unable to capture how individuals perceive their social media use nor which content they have encountered. For example, while social media use might trigger FOMO in one person, another individual might exchange reassuring messages via social media with a loved one before falling asleep. These diverging effects of social media have been referred to as the duality of social media effects (Kaur et al. 2021). To accurately assess the effect of smartphone-based social media use on sleep, this duality needs to be reflected in the measures that are used. Although this study has taken a first step in examining specific content effects by distinguishing between categories, future studies will need to even more closely examine the content smartphone users are exposed to, e.g. the messages they receive or the posts they view.

For two of the measures of engagement with apps belonging to specific categories, the study yielded results that were in contrast to earlier observations in the literature, i.e. video apps and audio/music apps. With regards to the first category, this study showed that watching videos on the smartphone was positively related to sleep duration (but no other sleep indicator). This could be an indication of the use of more immersive smartphone content on nights that students can sleep in the following day, e.g. during weekends. As the effects was only observed for evening use and not for pre-bedtime use, we caution against overinterpreting these effects. It is noteworthy that in general most effects for video content were not significant. This is surprising as earlier research (e.g. Nuutinen, Ray, and Roos 2013) would suggest that exposure to video content to be disruptive of sleep. The impact of exposure to video content might be lessened because of the screen size of smartphones, as arousal is shown to be screen size contingent (Szita and Rooney 2024). Finally, in line with social media app effects, video app effects are likely to depend on the content people watch. These alternative explanations will need to be addressed in future research.

In addition, with regard to time spent on music/audio apps, earlier research found music listening to have a positive effect on sleep (e.g. De Witte et al. 2020; Kirk et al. 2022). In contrast, we found that more time spent on music apps during the two hours before sleeping actually resulted in longer sleep latency, thus impairing sleep. Although many studies on music indeed showed positive effects on sleep (e.g. Tang et al., 2022), more recently researchers have also shown that some types of music can be disruptive of sleep. For example, Scullin, Gao, and Fillmore (2021) found that listening to music can result in experiencing so-called earworms. Earworms are ‘songs [that] can become “stuck” in one’s mind’ (Scullin et al., p. 985) and consequently intrude when someone wants to quiet their mind when trying to fall asleep. In three separate studies among young adults, Scullin et al. found evidence for the disruptive nature of earworms during the night – using self-report data as well as objective polysomnography data to assess sleep quality.

Overall, the current study displayed relatively few effects and observed effects were small in magnitude. This was most evident for sleep duration, which we might in part attribute to the fact that the study was conducted among a student sample. Students have rather flexible daily schedules and might be able to sleep in if they had watched a video longer than expected the following night, or even catch up on sleep through napping during the day. At the same time, we acknowledge that the sample, although reasonable and in line with earlier ESM studies, might still be somewhat low in power to observe very small effects. Being the first study to assess the effects of engagement with specific app categories, it is important that replications are conducted with diverse and larger samples to assess the robustness of the findings.

Considerations for future research. The present study has contributed to the field by using innovative methods and testing an initial framework to go beyond overall screentime measures by looking at the number of apps engaged with and the types of apps. Nevertheless, to draw more definite conclusions at least three limitations need to be addressed.

First, although measuring sleep has improved in recent years, and objective measurements have become available to scholars outside the specialised sleep research community (i.e. commercially available sleep sensors), the assessment of sleep is frequently debated. While the Fitbit has shown to be an adequate method to capture sleep overall, especially among healthy adults (Kang et al. 2017), actigraphs are still the golden standard in sleep research for most behavioural scientists. This specialist measurement tool can more accurately capture a variety of sleep indicators, such as sleep duration, sleep latency and sleep stages. More recent validation studies highlight the importance of using research-grade sleep sensors in future research (e.g. Kainec et al. 2024). At the same time the experience of sleep problems cannot be reduced to quantitative indicators. Sleep quality for example is understood as a complex phenomenon, which requires the inclusion of both objective as well as subjective indicators (Cudney et al. 2021). These two indicators often do not fully overlap, and many times subjective sleep quality is more informative when predicting well-being outcomes of sleep problems, and has been recognised as a valid indicator for sleep dysfunction (Mollayeva et al. 2016). It should be noted that this has been demonstrated in studies that used the complete scale to assess sleep quality, namely the Pittsburgh Sleep Quality Index (PSQI), rather than a one item measure as was the case in the current study. For this reason, future research might benefit from administering the full PSQI every morning to assess subjective sleep and to use actigraphs rather than commercially available devices to assess sleep objectively.

Second, the somewhat small sample size employed in the current study prevented the additional inclusion of covariates or moderators. The non-inclusion of covariates is expected to have only a small effect on our findings as we assume that important covariates included in earlier research will be relatively stable per individual over the seven days of the study. However, the inclusion of moderators could reveal smartphone effects that are now obscured, for example individuals might experience negative effects on some days but not all. Specifically, we might expect effects of smartphone use on sleep to differ on workdays versus free days. This distinction might be important as earlier research showed that sleep can differ between workdays and free days (e.g. Lenneis et al. 2021). Although the distinction between workdays and free days is sometimes equated with weekdays versus weekend days, this distinction does not hold for the current sample of university students. Students do often work next to their studies, but this is not restricted to specific days of the week. Moreover, in the current sample, participants’ week structure highly differed; they have a maximum of eight contact hours across weekdays for which they are at the university. For the remainder of the weekdays, some work, but others might study or engage in leisure activities. During the weekend they might work, engage in leisure activities, or study. As a result, time spent sleeping or catching up on sleep, or needing to wake especially early will differ highly in the sample.

While making a distinction between free days and work days would be insightful, in the current study this was not possible due to the relatively small sample, nor did participants report on their work schedule. Thus, to accurately address the moderating effect of free days versus non-free days, a more fine-grained measure of work is needed (e.g. daily assessment of amount of time spent on academic or profession related work) and a larger sample size. By employing a 7-day design we hoped to capture enough variance in the type of days, but recommend future research to assess the effect of these different types of days.

In short, to address the limitations above and to better understand the effects of social media on sleep we need further studies that employ at least 7-day ESM designs using larger sample sizes. Moreover, the samples should also be more diverse, reflecting a wider range of developmental stages and educational settings. For example, as the current study was conducted among a student sample, the unique characteristics of student life and their developmental stage could have impacted our results.

Third, although we were the first to consider the type of apps, the categorisation of the apps was theoretically driven as no consensus exists with regards to app categorisation. The apps were assigned to each category by the first and second author of the study, and for those apps for which the correct assignment was unclear, an ‘uncategorized’ label was used. This categorisation is publicly available (see OSF project page [link to be added after unblinding]) and has high face validity, but future research would benefit from a more rigid categorisation approach. Moreover, our measure of smartphone use can be improved upon as it assessed only hourly app-usage and not more specific usage (e.g. exact minutes before falling asleep; active vs passive use, etc.). The timing of app usage was limited by the way iOS logs app usages, namely giving times in minutes per hour timeslot. This means that it was impossible to assess app usage in the time up to the exact moment that someone went to bed. In addition, we were not able to distinguish between active versus passive app usage which also might have varying effects on sleep. Future studies will need to find data donations or tracking apps that provide timestamped app usage to make it possible to link it even more directly to the time participants go to bed and to assess actual in-app behaviours during that time.

Conclusion. In sum, although the findings indicate less consistent effects than expected, this study clearly shows that media effects research needs to identify characteristics and elements of media content that drive effects, rather than relying on broad categories such as screen-time (Bayer, Triệu, and Ellison 2020). Overall, the findings show that evening smartphone use in general might be less problematic than previously expected. Understanding the effects of smartphone use on sleep in more detail is important as it will (1) shape more specific advice about evening smartphone use that goes beyond not using one’s smartphone in the two hours before going to bed, and possibly (2) identify smartphone use that might be beneficial for sleep. To gain the necessary insights data donations or alternative direct measures of smartphone behaviours are crucial.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability

The data that support the findings of this study are openly available at https://osf.io/scvta/?view_only=3ae25ae4dd024248a4f9efe31ec80aba.

Additional information

Funding

This research was supported by a grant from the DigiComLab of the Amsterdam School of Communication Research.