The ‘Hot Hand’ An Investigation into Streakiness in Shooting

Abstract

The concept of the ‘hot hand’ is highly debated in the fields of statistics, data science, and psychology. The overwhelming consensus for the past four decades has generally revolved around the concept that ‘hot handedness’ is nothing more than a misinterpretation of randomness in small sample sizes. More recent research indicates that these conclusions, specific to the shooting performance in basketball, were founded on flawed analyses, and that a ‘hot hand’ may indeed exist. However, researchers have yet to successfully identified the ‘hot hand’ in modern NBA game data, and primarily focus on shooting data obtained from controlled experiments. In this work, we investigate streakiness in the context of player shooting data. Specifically, we develop a novel approach to the definition of streakiness that accounts for player specific effects. We then use two multivariate frameworks in the form of logistic regression and random forests, in order to measure the significance of streakiness in the context of predicting the outcome of a field goal attempt. Unlike previous research, we also examine this definition of streakiness with both pooled models and individual player-specific models. The results, similar to extant literature conflicted. Although the random forest models do not identify streakiness as a significant predictor, the logistic regressions provide several interesting insights. In particular, we find that while being ‘hot’ is significant at the .05 level in the pooled model, it is only inconsistently significant in the player-specific models. This indicates that individuals are likely to experience a ‘hot hand’ effect differently, if such an effect exists.