A probabilistic and dynamic reformulation of expected goals (xG): Methodological advances and modelling perspectives

Expected Goals (xG) is a probabilistic metric that estimates the likelihood of a shot resulting in a goal and is based on factors such as shot location, angle, body part used, and match context. It has become highly relevant in football analysis and is widely used to evaluate team performance, assess individual players, and guide recruitment and tactical decisions. In this study, we review the limitations of xG and propose strategies to improve its accuracy and applicability. Despite its popularity, xG is often misinterpreted as a deterministic measure. An xG value of 1 does not guarantee a goal but represents the expected outcome on the basis of shot probabilities. Summing shot-level xG values can be misleading, as multiple low-probability shots are not equivalent to a single high-probability attempt. Treating each shot as an independent probabilistic event, for example, using Bernoulli trials and Poisson binomial distributions, captures football’s inherent uncertainty more accurately. Matches are dynamic, with teams adapting strategies throughout; thus, xG must be interpreted in context. Goals may occur outside traditional xG events, such as deflections or self-goals, and the low-scoring nature of football highlights that xG alone may not fully reflect performance. Player- and position-specific variations, along with situational factors such as match location, scoreline, and opponent strength, further affect outcomes. Moving from simple metrics to probabilistic modelling, future developments could integrate richer datasets, contextual variables, and real-time updates, transforming xG into a context-aware, actionable decision-support tool for coaches and analysts.