The aim of this study was to create a novel metric, Expected Pass Turnovers (xPT), that could evaluate possession retention from player-passing events in football. Event and positional data were analysed from all 380 matches in the 2020/21 English Premier League season, which encompassed 256,433 passes in the final dataset. A logistic mixed-effects model was implemented to attribute the probability of each pass getting turned over. The use of positional data enabled the identification of a) opposition players present in radii surrounding the ball carrier and b) availability of teammates with respect to the ball carrier. The addition of these positional features improved the accuracy (+6.1 AUC Score) of the model. xPT serves as a practitioner Key Performance Indicator, as analysts can identify players that lose possession more often or not than expected, given the situational context of each pass, from game to game. Future work may include modelling the turnover probability of dribble and carry actions, as this would lead to a more comprehensive understanding of turnover events in football.
Keywords: Expected pass turnovers; football metric; logistic regression.