Title: The hot hand theory in hockey: A multilevel logistic regression analysis
Authors: Ivor Cribben - Alberta School of Business (Canada)
Likang Ding - Alberta School of Business (Canada)
Armann Ingolfsson - University of Alberta (Canada) [presenting]
Abstract: The Hot Hand theory states that an athlete will perform better in the present if he/she has performed well in the recent past. This theory has been investigated for basketball, baseball, and other sports. We test this theory for National Hockey League (NHL) playoff goaltenders by estimating how their performance on recent shots influences the probability of saving the next shot on goal. We use multilevel logistic regression models, in which we allow either some coefficients to vary among the season-goaltender combinations. In our regression model, the recent performance of a goaltender is measured by the number of saves within the most recent shots he faced during the same game. We also include other control variables such as shot type, shot origin, and game score in our model. Our data consists of 36,235 shot-on-goal observations for 90 goaltenders who played in the NHL playoffs between 2008 and 2016. We compare the results of multilevel models to simple logistic models as well as to multilevel models with different settings. Our preliminary findings are that a good recent save performance has a negative effect on the save probability for the next shot, which is consistent with the opposite of the Hot Hand theory.