A0245
Title: G-RAPM: Revisiting players contributions in regularized adjusted plus-minus models for basketball analytics
Authors: Luca Grassetti - University of Udine (Italy) [presenting]
Abstract: The estimation of the players' value based on on-field performances is a very relevant topic in the sports management framework. In fact, coaches and team managers can view ratings based on the players' evaluation manifold. Many approaches are adopted to determine the players' ratings. Among others, Regularized Adjusted Plus-Minus model can be considered one of the most interesting solutions. The model-based approach allows computing the efficiency of players by accounting for the opposing lineup effect and including some exogenous variables to mitigate the presence of confounding. The existing proposals suggest treating the players' effects as real-valued parameters defining contributions to the team's score. This solution is efficient and allows for direct comparison among players, but the interpretation of the players' performance measures is not always straightforward from the sports management's point of view. Following the well-known stochastic production frontier econometric approach, the proposed model specification considers players' contributions to the total lineup's score as positive constraint parameters. An additional unconstrained effect is introduced to adjust the total players' effects for the actual lineup contribution, which can identify both positive and negative synergies among individuals. The proposed model is estimated feasibly within the Bayesian framework, allowing straightforward further generalisations.
