An extended regularized adjusted plus-minus analysis for lineup management in basketball using play-by-play data

Abstract

In this work we analyse basketball play-by-play data in order to evaluate the efficiency of different five-man lineups employed by teams. Starting from the adjusted plus-minus framework, we present a model-based strategy for the analysis of the result of partial match outcomes, extending the current literature in two main directions. The first extension replaces the classical response variable (scored points) with a comprehensive score that combines a set of box score statistics. This allows various aspects of the game to be separated. The second extension focuses on entire lineups rather than individual players, using a suitable extended model specification. The model fitting procedure is Bayesian and provides the necessary regularization. An advantage of this approach is the use of posterior distributions to rank players and lineups, providing an effective tool for team managers. For the empirical analysis, we use the 2018/2019 regular season of the Turkish Airlines Euroleague Championship, with play-by-play and box scores for 240 matches, which are made available by the league website. The results of the model fitting can be used for several investigations as, for instance, the comparative analysis of the effects of single players and the estimation of total and synergic effects of lineups monitoring. Moreover, the behaviour of players and lineups during the season, updating the estimation results after each gameday, can represent a rather useful tool.