Applications and Implications of a Nested Dirichlet Model of Baseball Player Ability

• Activity Number: 325
• Type: Invited
• Date/Time: Tuesday, August 3, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Sports
• Author(s): Brad Null*+
• Companies: AOL Research
• Keywords: baseball ; Dirichlet ; forecasting ; hierarchical model ; Bayesian ; sports

Abstract

An understanding of the underlying abilities of the involved players is integral to many forecasting and optimization applications in baseball. Through the years, many approaches have been taken to assess these abilities, from simple aggregation of historical statistics to complex predictive methods. In this talk, we present a nested Dirichlet based model of player ability with several unique qualities. In particular, the model aims to accurately assess not just a player's estimated abilities, but the associated uncertainty with respect to those estimates, as well as the stochastic dynamics in these abilities that cause them to fluctuate over time. We will show how these and other features of the model lead to improved forecasting of player, play, game, and season results, as well as new insights that can be leveraged with respect to related optimization problems.