Predicting NHL Player Types and Performance Outcomes

Sean Prentis; Greg Ackerman; Quinn MacLean

doi:https://doi.org/10.21985/n2-x6xs-dr46

Work

Predicting NHL Player Types and Performance Outcomes

Public Deposited

Download PDF

Download All Files (.zip)

The 2005 NHL Lockout instituted a hard salary cap, which is the most limiting salary cap of the major US sports leagues. The NHL as a result doesn’t have the luxury of overpaying or dumping a contract as other professional sports leagues have, which places a huge constraint and emphasis on player evaluations and relative performance. In fact, NHL teams have become so cautious in spending on free agents that Adam Gretz found that 21% of cap space amongst Stanley cup winners was used on free-agent acquisitions. This places an emphasis on proper evaluation of talent through scouting for internal talent but also places a need for a proper player evaluation application for when player performances are going to be at a premium to the organization. Our research questions for the project will aim at supplying organizations with a clear look at the type of playing styles that exist in the NHL, their attributes, and the indicators that lead to performance outbreaks or bounce-back seasons. Our specific research questions are: What are the different player performance clusters that exist in the NHL? What attributes lead to “breakout” seasons in the NHL? Can we use our predictive model to predict YoY increases in core metrics such as Goals Above Replacements (GAR), Standing Points Above Replacement (SPAR), and Wins Above Replacement (WAR), to identify value in existing players? For the purposes of the report, stats by the name of GAR_Delta, SPAR_Delta and WAR_Delta will be used. What are the reasons that we can infer from previous players expected for large increases in GAR_Delta, SPAR_Delta and WAR_Delta that didn’t pan out? What are the salary implications for predicted GAR_Delta, SPAR_Delta and WAR_Delta? Are there current limitations in available data that need to be addressed? What are some proposed advancements in data collection that would make the models meaningful? Is there streakiness that exists within players? Do lineup pairings impact the performance of a player?

Last modified