Ranking Teams in the NCAA Division I Men's Basketball Tournament

• Activity Number: 439
• Type: Invited
• Date/Time: Wednesday, August 3, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Sports
• Author(s): Mark Brown*+ and Joel Sokol
• Companies: The City University of New York and Georgia Institute of Technology
• Keywords: Empirical Bayes ; Markov chains ; Prediction ; Quantitative sports analysis ; Applied probability ; Applied statistics

Abstract

The LMRC method for predicting NCAA tornament results from regular season game outcomes is a two part process. First using a logistic regression model we focus on those pairs of teams who have played one another during the regular season. For these pairs we estimate the probability of one team beating another on a neutral court. These probabilities are utilized in a probability transition matrix of a Markov chain. All teams are then ranked by the stationary distribution of this chain. This provides methodology to rank teams who have not played one another during the regular season, and in some cases have no common opponents either. In recent work, (Brown, M. and Sokol, J. (2010), Journal of Quantitative Analysis in Sports, Volume6, Issue 3, Article 4), we replaced the logistic regression analysis by an empirical Bayes approach. This resulted in significant improvement both on past tournament data and in the 2010 tournament in which the method correctly predicted the winning team in more tournament games than any of the few dozen competing methods found online. In this talk the methodology will be reviewed and we will report on its performance in the 2011 tournament.