The Hot Hand Theory in NBA Basketball (Dissertation)

Abstract

For decades, the “Hot Hand” phenomenon has been labeled a statistical fallacy, even though it is a common belief in basketball. However, recent research suggests this conclusion was based on flawed analysis. This study addresses those limitations by analyzing the hot hand as a feature in a shot prediction task using logistic regression models.

The research also expands on previous literature by using Random Forest and XGBoost models, which achieved a moderately higher accuracy of 64% in live game scenarios compared to other shot prediction models. The study also uniquely investigated the hot hand effect during the annual NBA 3-point contest, an environment ideal for analyzing streaky shooting. Prediction models for the 3-point contest had lower accuracy scores (53-59%), likely due to a smaller sample size and less descriptive data.

By analyzing data from both the NBA 3-point contest (a semi-controlled environment) and live games (a naturalistic environment), the study provides a more nuanced understanding of the effect. We found significant evidence of a hot hand effect after a single made shot in both datasets: a +19.4% increase in the 3-point contest and a +4.2% increase in live games. The results for longer streaks were mixed, showing diminishing benefits.

If you are interested in reading the full dissertation, please contact the author at axelolafsson.work@gmail.com.