BEDDiT: A validated R package for inferring the bedtimes of Reddit users from the timestamps of their posts

Background: Social media usage is both a potential influencer of human sleep and a research tool for understanding human sleep more generally. For both reasons, there is a growing interest in understanding the sleep patterns of social media users. These studies require both social media usage data and sleep data. There is ample social media data that are publicly and immediately accessible, which could in principle make the field one of rapid discovery. However, the users’ sleep data generally must be collected by researchers directly from participants, which negates the benefits of large sample size and immediacy that social media data otherwise affords. Objective: To address this bottleneck, we developed a simple model for inferring the bedtimes of Reddit users from the timestamps of their posts. Methods: The model is split piecewise into wake-time and sleep-time hours: during wake-time hours, the model is constant; during sleep-time, there is a quadratic depletion in posting frequency. We train and test the model on 128 Reddit users who publicly report both their bedtime and time zone. Results: On aggregated data, the model has excellent fit; with R^2 > 0.99. In individual level data, the model has fair fit, Spearman rho 0.62, which is similar to the correlation between polysomnography-derived sleep parameters and self-reported sleep parameters. For additional validation, we apply the model to 26 Reddit users who responded to a private survey reporting both their bedtime and their time zone, which also produces a Spearman rho of 0.44 between inferred and self-reported bedtimes. We apply the model to estimate bedtimes for more than 50,000 Reddit users who publicly report their time zone and explore the characteristics of these bedtimes. Conclusions: Our model is freely available as an R package on GitHub and enables researchers to associate sleep patterns of social media users with the other characteristics users share.