Evaluating NBA end-of-game decision-making

Abstract

This paper introduces a probabilistic method to evaluate the tactical decisions players and coaches make at the end of NBA games. For the purposes of this research, these decisions include whether to shoot a two-point or three-point field goal for the offensive team and whether to intentionally foul for the defensive team. With a win probability model built using logistic regression and player statistics, the optimal decision for both teams in a given possession is found. The End-of-game Tactics Metric (ETM) is the difference between the win probability of the optimal decision and the win probability of the actual decision. This research extends beyond current applications of win probability models to evaluate the actual on-court decision as opposed to evaluating the result of a possession. To evaluate the usefulness of ETM, the winning percentage of teams in games decided by a margin of five points or fewer can be compared with the mean ETM difference between a team and its opponent. The correlation coefficient of the relationship is -0.64. When combined with other variables that affect winning percentage in close games, a linear regression on those explanatory variables has an adjusted R2 value of 0.79. This analysis shows that the ETM difference has a significant effect on winning close games, despite having little reliance on player performance.