He's on Fire? Strategic Decisions and Allocation Adjustments Under the Hot Hand Fallacy

Sammanfattning

This paper empirically explores 10 years of play-by-play data from the seasons 2011-2021 from National Basketball Association with a comprehensive dataset with 2.7 million shots to test for a hot hand, (i.e., predictability in future outcomes) using three separate models. I find proof for its existence in all short-term categories using panel data as well as when adjusting for shot difficulty using a linear regression prediction model. The measured values are significant where a hot hand equates to around a 0.70-0.77 standard deviation of player ability distribution. The findings are in contrast to most of the hot-hand research, which has typically reported either no hand or a very weak hand in sports when controlled shooting designs are used. This difference, I argue, is due to the lack of large datasets and the absence of the athletes’ natural environment, both of which are necessary for creating a realistic setting for evaluating a hot hand. In my regression discontinuity design, I found that hot players remain rational by not taking more difficult shots. There does not seem to exist any endogenous defensive response, i.e., the closest defender distance remains unchanged, and the coach begins to modify strategy first after longer streaks (a sequence of seven hits or longer).