|
Abstract:
|
A key criterion used to determine the participants of the NCAA Women's Ice Hockey Championship tournament is a team's Rating Percentage Index. This rating system only takes into account a team's overall and its opponent's record, not game results against a specific opponent. Based on Mease's work, I propose a ranking system based on the results of individual games, ignoring game score. In women's hockey, approximately 10% of regular season games end in ties, and the proposed ranking system explicitly models the probability of two teams tying, given the relative strength of each team. A Bayesian approach is utilized, allowing for information on the probability of a tie from previous seasons to be included in the model. In addition, the prior distribution of a team's rating can be based on previous year's results, allowing for early season rankings to be indicative of the final rankings.
|
