Estimating NBA Player Contribution with Regularized Logistic Regression

• Activity Number: 573
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Sports
• Author(s): Sameer Deshpande*+
• Companies: Wharton School
• Keywords: Logistic Regression ; Regularization ; Shrinkage ; Basketball analytics

Abstract

We propose a regularized logistic regression model to estimate NBA player contributions adjusted for all other players, teams, and other in-game variables. Traditional statistics, such as adjusted plus-minus, while useful for retrospectively comparing players, are often ill-suited to inform in-game lineup decisions. We instead consider the following regression problem: what does each player contribute to the odds of preserving a lead? By incorporating real-time box score information, we attempt to predict whether using a particular lineup at a specific point during a game helps or hurts a team's chances of winning. We perform two separate analyses: variable selection to identify the "best" and "worst" contributors and a fully Bayesian analysis to predict lineup efficacy.