A regression-based predictive model of student attendance at UVA men's basketball games

Abstract

A regression-based predictive model was developed to allow better prediction of attendance for the student general admission seats at University of Virginia men's home basketball games. The goal was to improve upon the existing prediction method that yielded prediction errors sometimes exceeding one thousand students. Based on the existing attendance prediction method and a literature review, twenty candidate factors were identified for potential use in an improved prediction model. Using a best subsets methodology and data from the previous four basketball seasons, a six predictor model was developed with an adjusted R2 value of 0.816. The predictors were based on whether UVA and/or the opponent were ranked, opponent popularity, and whether classes were in session. The resulting model was validated with data from home games from the 2002-2003 season. Its average prediction error of 263 students (standard deviation of 269 students) was a significant improvement over the existing prediction method