The hot hand in the wild

Abstract

Streaks of success have always fascinated people and a lot of research has been conducted to identify whether the “hot hand” effect is real. While sports have provided an appropriate platform for studying this phenomenon, the majority of existing literature examines scenarios in a vacuum with results that might or might not be applicable in the wild. In this study, we build on the existing literature and develop an appropriate framework to quantify the extent to which success can come in streaks—beyond the stroke of chance—in a natural environment. Considering in-game basketball game situations, our analysis provides statistical evidence that individual players do indeed exhibit the hot hand in varying degrees, that is, individual players can consistently get in a streak of successful shots beyond random chance. However, as a whole, the average player exhibits shooting regression, that is, after consecutive makes he tends to perform below expectations. Even though our results are based on a sports setting, we believe that our study provides a path towards thinking of the hot hand beyond a laboratory-like, controlled environment. This is crucial if we want to use similar results to enhance our decision making and better understand short and long term outcomes of repeated decisions.

Quantifying implicit biases in refereeing using NBA referees as a testbed

“…However, given that I essentially perform multiple statistical tests—one for each player—I expect some of them to deem statistically significant results even by chance. Therefore, I perform a meta-test to calculate the probability that all of the data points that came out as statistically significant are false positives 25 , 26 . In particular, under the—realistic in this case—assumption that the tests are not correlated I can use the Binomial distribution for a meta-test.…”

Implicit Biases in Refereeing: Lessons from NBA Referees

“…However, given that we essentially perform multiple statistical tests -one for each player -we expect some of them to deem statistically significant results even by chance. Therefore, we perform a meta-test to calculate the probability that all of the data points that came out as statistically significant are false positives 8,9 . In particular, under the -realistic in our case -assumption that our tests are not correlated we can use the Binomial distribution for a meta-test.…”