Graduation Year

2018

Date of Submission

4-2018

Document Type

Campus Only Senior Thesis

Degree Name

Bachelor of Arts

Department

Economics

Reader 1

Darren Filson

Abstract

Researchers have tried to predict winning percentages for the National Hockey League (NHL) teams based on their performance in the previous seasons. However, these predictions have not been very accurate. This study hypothesizes that incorporating pair-wise game-level data with season-level data can be useful in improving the prediction of a team’s win percentage. Season-level data and pair-wise game-level data from the 2005-2006 season to the 2015-2016 season has been used to predict winning percentages for the pairs in each of the following seasons. Significant results were not found for any of the pair-wise game-level data variables except for two pair-wise variables. This helps establish the idea that including more granular information does not necessarily increase the predictive power of models. One of the pair-wise variables found to be significant (at the 10% level of significance) was when high goal differential was observed in the interaction term between high goal differential for a team in its home games against the other pair-wise team and the goal differential for a team in its home games against the other pair-wise team. This provides marginal support for the claim that extreme game-level outcomes from the previous season can help in predicting a team’s win percentage in the following season. Another pair-level variable found to be significant (at the 5% level of significance) was when high goal differential was observed and at least 4 games played was not observed in the interaction term between at least 4 games played against the other pair-wise team and high goal differential for a team in its home games against the other pair-wise team. This suggests that only in the games a team plays outside its own division, the extreme game-level data helps in predicting a team’s win percentage in the following season.

This thesis is restricted to the Claremont Colleges current faculty, students, and staff.

Share

COinS