Abstract
In the NHL, there have been numerous models created, both public and private, that seek to predict future performance for each team. These models have used a variety of available data and metrics to best describe what factors of the game at the
highest level predict future goals and wins. For this project, I seek to create a new predictive model that can be used for both Men’s and Women’s DI NCAA Hockey. The challenge is the lack of publicly available data at the NCAA level compared to what’s available in the NHL. We used data that can be found on College Hockey News such as shot attempts (Corsi), goals, shot percentages, and save percentages in different game states to calculate our model. We use ridge regression to calculate each team’s offensive and defensive rating by calculating each team’s
efficiency, or many goals they would be expected to score/give upper 100 shot attempts, which is adjusted for strength of schedule. We then use this, along with teams’ Corsi rating as well as other factors to calculate an expected winning percentage for each team. These can be used to predict the win probabilities for each future game. There are no public predictive models to compare this to, but my model performs very strongly over the course of the season in predicting the outcome of games. Analysts, fans, journalists can use this model to get a better idea on the actual strength of a team, compared to just looking at past wins and losses.
Sports and Society Statement
This project helps to show how we can make better decisions and understand the game more using the data available to us. While my project has a very specific application, it paints a broader picture that using data to evaluate sports helps us to understand the why and how of why certain things can happen in what we watch.