Examination of efficient roster design in the National Hockey League (NHL)
MetadataShow full item record
The study estimates the values of NHL roster positions. The analysis was conducted in two phases. First, cluster analysis was used to evaluate and rank players for their overall performance across positions. Second, regression analysis based on aggregated player classifications across team-games estimated the value of roster position and measured diminishing returns to talent across positions. Players were evaluated based on their regular season performance. The clustering of all skaters was administered separately for each position and each year. Standardized regular season-long variables were applied in the analysis. The variables used to cluster all positions were: points per time on ice, goals per time on ice, assists per time on ice, plus/minus per time on ice, shots differential per time on ice, blocks per time on ice, hits per time on ice and penalties per time on ice. Forwards were distributed amongst four lines and defensemen were allocated to three pairings. The linear regression analysis used play-by-play data from the 2010-17 NHL regular seasons. Results indicated that an increase in the quality of centers increased the win probability of a team the most. Teams make player acquisitions decisions based on the talent available and their current composition of players. A team’s hockey operations department can use the findings to evaluate their roster composition and identify positions with the greatest marginal benefit from player acquisitions.