Building a Championship Contender: A Statistical Model for NBA Lineup Combinations

• 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): Ryan J. Parker*+ and Brian Reich and Jarad Niemi
• Companies: North Carolina State University and North Carolina State University and University of California at Santa Barbara
• Keywords: basketball ; lineup combinations ; team performance

Abstract

Several methods have been proposed to evaluate the contribution of individual basketball players to their team's success. In this talk we study the performance of combinations of NBA basketball players. In basketball, like many other popular American sports, team performance is not be well-predicted by the sum of the performance of the team members. For example, it may be that a three-point shooter's value is increased when paired with a play-making point guard or a post player that draws double teams. A challenging part of putting together an effective roster of players is accounting for these synergetic relationships while extrapolating to unobserved combinations of players. Combining individual player statistics and results for all 10-player combinations for the 2009-2010 NBA season, we build a statistical model for the performance of a 10-player combination that accounts for potentially-complex interactions between types of players, and we show that this model outperforms other proposed predictive models that simply use the average ratings of the players. Further, we demonstrate how to use our approach for free agent signing and setting lineups for particular opponent matchups.