Estimating NBA athlete performance using multilevel models

Abstract:

In basketball, the athlete performance evaluation are generally based on variants of plus-minus and PER statistics Optimizing Athlete Performance calculated through multiple regression, ridge, or lasso models using likelihood-based or Bayesian approach. We developed a novel methodology based on principal components analysis and multilevel model to create new indexes such as Oliveira-Newell Score that can be used to evaluate player performance during a match, relevance score used to rank players in a season, and the consistence score used to evaluate the player contribution for their team based on random effects.