Temporal Dynamics of Physical Activity and Sleep in University Students: A Cross-Lagged Panel Analysis Using Garmin-Derived Data

Sleep plays a central role in cognitive performance, emotional regulation, and mental and physical health, yet it is frequently disrupted during young adulthood (Hershner & Chervin, 2014; Pifer et al., 2024). Circadian rhythms naturally shift later in adolescence and young adulthood, often clashing with early academic schedules and contributing to chronic sleep deprivation (Pifer et al., 2024). Up to 70% of college students report insufficient sleep, and over half experience daytime sleepiness - factors linked to poorer academic performance, mood disturbances, and increased accident risk (Hershner & Chervin, 2014; Santana et al., 2023). Identifying modifiable predictors of sleep - such as physical activity (PA) - is therefore essential.

Traditionally, the relationship between PA and sleep has been conceptualized as bidirectional: adequate sleep is thought to facilitate higher daytime activity, while regular PA is believed to enhance sleep quality (Kline, 2014). However, empirical support for this relationship remains inconsistent and appears to depend on contextual factors such as exercise timing, intensity, and environmental cues (Korkutata et al., 2025). For instance, recent findings challenge the long-standing sleep hygiene recommendation to avoid exercise before bedtime (e.g., Breus, 2024). A meta-analysis by Stutz et al. (2019) found that evening exercise generally does not impair sleep and may even enhance sleep efficiency—particularly when moderate in intensity and not performed within an hour of bedtime. Similarly, Korkutata et al. (2025) reported that structured PA tends to increase slow-wave sleep and reduce nighttime awakenings, with vigorous activity close to bedtime disrupting sleep only for some individuals, not all.

This inconsistency in findings over time is likely attributable to several methodological limitations of earlier studies and a gradual shift in research paradigms. Earlier studies frequently relied on small sample sizes, cross-sectional designs, or experimental manipulations of PA. More recently, researchers have adopted intensive longitudinal approaches with larger samples collected in naturalistic, daily life settings to better capture the complexity of the PA-sleep relationship. However, many newer studies still face challenges, including reliance on subjective physical activity and sleep measures. Self-report data are vulnerable to recall errors and social desirability bias. This issue is especially pronounced in research involving younger populations. For instance, in a systematic review and meta-analysis by Memon et al. (2021), 93% of the 26 included studies in youth and young adults relied on self-reported sleep data, 90% on self-reported physical activity, and 99% were cross-sectional in nature, using a variety of questionnaire-based assessments. Unsurprisingly, the results were inconsistent: sixteen studies reported no relationship between PA and sleep, nine found positive associations, and four observed negative associations. These inconsistencies underscore the need for approaches that investigate the relationship between PA and sleep in adolescents and young adults using objective measures—such as actigraphy and smartwatch-embedded sensors—which are less prone to bias and can be collected over extended periods with minimal participant burden. However, resolving these challenges requires more than just sophisticated technology—it demands conceptual and methodological consistency. Even when objective tools are used, several challenges remaining include how to (1) accurately define physical activity, (2) reliably measure sleep, and (3) account for the bidirectional and autoregressive nature of these behaviors over time.

Currently, research in this study area employs a wide range of approaches to operationalize both PA and sleep. PA has frequently been quantified using step-based metrics such as total daily steps (Bisson & Lachman, 2023; Mead et al., 2019), minutes of moderate-to-vigorous physical activity (MVPA; Mead et al., 2019), or peak intensity (Bisson & Lachman, 2023). However, these metrics often fail to account for individual differences in physiological load or intensity of activity. For example, two individuals may accumulate the exact same number of steps in a 10-minute time frame, but the exertion involved can vary substantially depending on factors such as height, bodyweight, fitness level, and gait efficiency. To address this, more nuanced methods - such as estimating heart rate reserve - have been applied (Zapalac et al., 2024), which might lead to different conclusions than step count approaches.

Similarly, sleep has been assessed using a variety of indicators, such as total sleep time and wake after sleep onset (WASO; Bisson & Lachman, 2023; Mead et al., 2019), as well as sleep efficiency and estimated sleep stages derived from actigraphy-based parameters (Zapalac et al., 2024). However, these measures may not fully capture the multidimensional nature of sleep, as they often omit physiological indicators—such as heart rate (HR) during sleep—that are thought to reflect recovery status and autonomic balance during sleep (Nuuttila et al., 2024). Moreover, commonly used metrics like total sleep time are heavily constrained by fixed bed/wake times, social routines, and academic demands, limiting day-to-day variability and potentially attenuating the observable effects of PA. While device-based measures offer greater objectivity, their validity-particularly in detecting sleep stages and physiological recovery-remains limited (Chinoy et al., 2021).

Finally, differences in statistical modeling approaches also contribute to inconsistencies in the PA-sleep literature. These include mixed-effects models, with and without within-person centering, and in some cases, time-segmented analysis or compositional data techniques—each with varying degrees of adjustment for autocorrelation and repeated-measures interdependence. This variability in PA and sleep operationalization and methodological approaches is illustrated in three recent studies. First, Mead et al. (2019) used Fitbit Flex devices to monitor daily step count, active minutes (defined as periods of ≥10 consecutive minutes at ≥3 METs), total calories burned, total sleep time, and WASO over six consecutive days in a sample of 54 undergraduate students (mean age = 19.4 years, 70% female). Mixed-effects linear models were used to estimate both within- and between-person effects, with person-mean centering applied to distinguish intra- from inter-individual associations. Notably, they did not account for autocorrelation or include random slopes. The findings showed no significant between-subject associations. However, several within-subject effects emerged: nights with longer-than-usual sleep time or elevated WASO predicted lower PA the following day, including fewer steps and reduced energy expenditure. Conversely, days with more MVPA were associated with greater WASO that night. These findings suggest a short-term, potentially compensatory interplay between sleep and PA at the within-person level. Second, Bisson and Lachmann (2023) used Actiwatch-64 devices over six consecutive days in a diverse community sample of 427 adults (mean age = 54.2 years, range = 34–83, 61% female). Daily PA was quantified using total activity counts and peak intensity (i.e., the highest activity count during waking hours), while sleep was assessed using total sleep time (TST), WASO, and sleep latency. Multilevel modeling was used to assess both within-person fluctuations and between-person averages, including modelling random slopes. At the between-person level, greater average daily PA predicted shorter TST. Within individuals, higher-than-usual PA predicted shorter TST and lower WASO – but these effects were moderated by age. Specifically, the within-person association between greater PA and shorter sleep was significant only among younger adults, suggesting that age may shape how PA influences sleep on a day-to-day basis. Third, Zapalac et al. (2024) tracked sleep and PA over 35 days in 65 university students (mean age = 21.4 years, 65% female) using Fitbit Inspire HR devices. PA was quantified via heart rate reserve – based estimates of sedentary behavior, light activity, and MVPA. Sleep was assessed using device-based sleep staging algorithms combining movement and HR. Mixed-effects models with within-person centering and random slopes revealed that greater-than-usual sedentary time predicted longer same-night sleep within individuals, while individuals who were more sedentary overall slept less on average. Activity was also linked to changes in sleep architecture: both light and moderate-to-vigorous PA predicted more NREM sleep, less REM sleep, and longer REM latency, particularly when activity occurred in the evening. However, sleep staging relied on Fitbit’s proprietary algorithms, and HRR was based on the age-predicted maximum HR formula, which may reduce accuracy at the individual level.

These findings point to the importance of developing PA measures that are more precise in timing, grounded in physiology, and multidimensional in scope to better understand their link with sleep. The present study aims to provide an ecologically valid, high-resolution analysis of the day-to-day associations between PA and sleep in a sample of 97 university students. The sample was predominantly female (87%), with most participants aged 18-21 (64%), enrolled in bachelor’s programs (86%), and 38% identifying as international students. Building on the approach of Zapalac et al. (2024), we adopt metabolic equivalent of task (MET) - a metric integrating both heart rate and movement data – as our primary measure of activity intensity. This measure provides a physiologically meaningful estimate of energy expenditure beyond step counts alone. To investigate temporal dynamics and potential bidirectionality, we use random intercept cross-lagged panel models (RI-CLPM). Unlike traditional cross-lagged panel models, the RI-CLPM separates stable between-person differences from within-person fluctuations, enabling us to test whether deviations from an individual’s average activity level predict changes in their sleep that night, and vice versa. In addition, we will investigate how the timing of PA relates to sleep outcomes by aggregating MET-minutes into discrete time blocks anchored to each participant’s sleep onset, allowing for individualized, time-relative analyses. Sleep will be assessed using multiple metrics collected via the Garmin Vivosmart 4, including total sleep time, mean heart rate, and variability in minute-to-minute heart rate values during sleep. These indicators are leveraged as pragmatic, ecologically valid proxies of nocturnal autonomic activity. While less granular than traditional HRV metrics, these measures can still reflect meaningful fluctuations in physiological arousal during sleep in free-living contexts. By combining objective, time-sensitive measures with rigorous within-person statistical modeling, this study aims to clarify the nuanced and potentially bidirectional relationship between physical activity and sleep in young adults.