Using Recruiting Rankings and Team Level Measurements to Predict College Football Team Success

• Activity Number: 473 - For the Love of the Game: Applications of Statistics in Sports
• Type: Contributed
• Date/Time: Wednesday, July 31, 2019 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Sports
• Author(s): Ross Gosky* and Sydney Singleton
• Companies: Appalachian State University and Appalachian State University
• Keywords: model selection; cross-validation; multiple regression; regression tree

Abstract

We analyzed team performance for college football teams in major conferences during the years of 2007 through 2018, using each team’s end-of-season standardized Sagarin rating, given at www.sagarin.com, as the outcome measure of performance. We considered several potential predictor variables, including many from the team recruiting rankings at the website www.rivals.com, and other team attributes compiled from an annual college football prediction magazine. Forward, backward, and stepwise variable screening methods were used to select candidate models, as well as a regression tree as an additional candidate model. Cross-validation analysis was conducted to compare the prediction accuracy of the models, using each individual season separately as a sequence of test data sets. Our findings suggest that teams with higher recruiting rankings are predicted to perform better in a given season, even when other factors are taken into account, but also that other team-level factors are significant predictors of performance.