Abstract

Most public research into player performance in the National Hockey League makes use of data published by the NHL itself, which is limited in scope. In the context of evaluating player shooting, i.e. the ability of a player to affect the probability of a shot of theirs becoming a goal, the use of public NHL data may produce a bias, as it does not include data about the movement of the puck prior to a shot. This thesis makes use of a privately tracked dataset which does include information regarding pre-shot movement, and inspects whether the use of this more refined dataset has a material impact on the analysis of player shooting talent. The probability of a shot becoming a goal given external factors (such as location, pre-shot movement, and game state) are used to estimate the probability of a shot becoming a goal, independent of the shooter's ability. Then, the impact a shooter has on the goal probability is captured as a coefficient estimate in a logistic regression model. This analysis is run over both datasets, and the results show that there is evidence that the more granular dataset eliminates pre-shot movement bias in the evaluation of player shooting ability.