What Makes a Good Day?: An Interpretable Machine Learning Approach to the American Time Use Survey

What differentiates a happy day from a typical one? Answering this question can help to identify the building blocks of a happy life, but little is known about what makes some days better than others. To identify the ingredients of a good day, we used interpretable machine learning techniques to assess how time spent on over 100 activities predicted whether Americans rated their day as typical or “better than typical.” Using data from the 2013 (N = 9286) and 2021 (N = 6196) waves of the American Time Use Survey, we identified clear tipping points that distinguish better days from typical ones. Socializing was one of the most important activities for having a good day, but beyond 2 hours, additional socializing had little impact. Time spent with friends, in contrast, had an almost boundlessly positive effect. The impact of working was also highly time contingent. We found that working for up to six hours had no impact on whether people rated their day as better than usual. When individuals worked for more than six hours though, the effects rapidly turned negative. Taken together, our results suggest that good days emerge from the constantly shifting value of time; each minute in a day represents a choice between competing opportunities, with clear tipping points when benefits diminish and costs mount.