Performance Evaluation of Women's Volleyball Players Using Bayesian Data Analysis

Abstract

Understanding player contribution is an important component of lineup construction, advance scouting, and performance evaluation in volleyball. However, traditional methods utilize oversimplified percentages that fail to acknowledge latent variables and situational nuance. These shortcomings are addressed in this work via a holistic framework mirroring the structure of a story, with each sub-component representing an angle by which player contribution can be investigated. The emphasis here is on modelling player contribution via player presence, and a Bayesian logistic regularized regression with a horseshoe prior and custom model structure is employed. This approach incorporates player roles, lineup matchups, and additional context-specific information into its estimates. The resultant model outputs are tested against substantive knowledge and posterior predictive checks, with direct application to volleyball coaches, players, and the community at large. Extensions to this analysis consider other factors of player contribution such as player action, event sequences, and player intent, with applications arising from advancements in computer vision, sports science technology, and data acquisition also discussed.