Rating of sports teams via least squares and mean absolute deviation techniques : an empirical study of three NCAA sports

This is a decade-long study of rating sports teams using Least Squares and Mean Absolute Deviation methods. Three NCAA sports are studied: College Football, Men's Basketball, and Women's Basketball. Least Squares outperformed Mean Absolute Deviation in all three sports in terms of accuracy [number of correct games predicted]. The accuracy of the best Least Squares model is: 72.74% for Football, 74.61% for Men's Basketball, and 78.90% for Women's Basketball. This is the first study to consider Women's Basketball and first to use more than one year data set for the Mean Absolute Deviation model.

Four types of enhancements are applied to the basic models: Home Edge, Scaling of Differences, Start-of-Season Ratings, and Time Discounting. In three of these four, new ways of modeling these enhancements are introduced: Adjusted Average Home Edge (Home Edge), quasi-Bayesian (Start-of-Season Ratings), and the Time Discounting enhancements. The Adjusted Average Home Edge is the most effective Home Edge term. Start-of-Season ratings are used in all most accurate models. Five out of these six also have their start-of-season ratings regressed to the mean.