Digital sleep logs reveal potential impacts of modern temporal structure on class performance in different chronotypes.

Stability of sleep and circadian rhythms are important for healthy learning and memory. While experimental manipulations of lifestyle and learning outcomes present major obstacles, the ongoing increase in data sources allows retrospective data mining of people's sleep timing variation. Here I use digital sleep-log data generated by 1109 students in a biology lab course at the University of Washington to test the hypothesis that higher variance in time asleep and later sleep-onset times negatively correlate with class performance, used here as a real-world proxy for learning and memory. I find that sleep duration variance and mean sleep-onset times both significantly correlate with class performance. These correlations are powerful on weeknights but undetectable on Friday and Saturday nights ("free nights"). Finally, although these data come with no demographic information beyond sex, the constructed demographic groups of "larks" and "owls" within the sexes reveal a significant decrease in performance of owls relative to larks in male students, whereas the correlation of performance with sleep-onset time for all male students was only a near-significant trend. This provides a proof of concept that deeper demographic mining of digital logs in the future may identify subgroups for which certain sleep phenotypes have greater predictive value for performance outcomes. The data analyzed are consistent with known patterns, including sleep-timing delays from weeknights to free nights and sleep-timing delays in men relative to women. These findings support the hypothesis that modern schedule impositions on sleep and circadian timing have consequences for real-world learning and memory. This study also highlights the low-cost, large-scale benefits of personal, daily, digital records as an augmentation of sleep and circadian studies.