On the relationship between +/– ratings and event-level performance statistics

Abstract

This work considers the challenge of identifying and properly assessing the contribution of a single player towards the performance of a team. In particular, we study the use of advanced plus-minus ratings for individual football players, which involves evaluating a player based on the goals scored and conceded with the player appearing on the pitch, while compensating for the quality of the opponents and the teammates as well as other factors. To increase the understanding of plus-minus ratings, event-based data from matches are first used to explain the observed variance of ratings, and then to improve their ability to predict outcomes of football matches. It is found that event-level performance statistics can explain from 22% to 38% of the variance in plus-minus ratings, depending on player positions, while incorporating the event-level statistics only marginally improves the predictive power of plus-minus ratings.