Quantifying implicit biases in refereeing using NBA referees as a testbed

Abstract

AbstractImplicit biases occur automatically and unintentionally and are particularly present when we have to make split second decisions. One such situations appears in refereeing, where referees have to make an instantaneous decision on a potential violation. In this work I revisit and extend some of the existing work on implicit biases in refereeing. In particular, I focus on refereeing in the NBA and examine three different types of implicit bias; (i) home-vs-away bias, (ii) bias towards individual players or teams, and, (iii) racial bias. For this study, I use play-by-play data and data from the Last 2 min reports the league office releases for games that were within 5 points in the last 2 min since the 2015 season. The results indicate that the there is a bias towards the home team—particularly pronounced during the playoffs—but it has been reduced since the COVID-19 pandemic. Furthermore, there is robust statistical evidence that specific players benefit from referee decisions more than expected from pure chance. However, I find no evidence of negative bias towards individual players, or towards specific teams. Finally, my analysis on racial bias indicates the absence of any bias.