The xG model debate
- A Substack essay argues there is no single 'perfect' xG model and that model choice depends on use case and context. - The piece frames the trade-offs around accuracy, bias and the intended application of a predictive model. - That methodological debate is playing out alongside fan-facing supercomputer forecasts, for example predictions naming Arsenal likely Champions ( ).
Expected goals (xG) measures the quality of a shot based on factors like distance, angle, assist type, and game situation, predicting the probability of it being a goal. xG helps analysts evaluate team and player performance beyond actual scores. (understat.com) Football analytics platforms like StatsBomb and Opta develop xG models using machine learning on thousands of historical shots. Different models exist because data sources and weighting vary between providers. (statsbomb.com) A recent Substack post argues there is no single "perfect" xG model—choices depend on use case like live predictions versus post-match analysis. Each model trades off accuracy, bias, and computational speed for specific applications. (thexgfootballclub.substack.com) xG models are "live" for real-time predictions or "post-match" for full-game analysis, with live models updating rapidly during play. The post highlights how model choice affects predictions, like title race forecasts. (thexgfootballclub.substack.com) Fan-facing supercomputer simulations, such as those predicting Arsenal as likely Champions League winner, rely on xG-derived probabilities. These forecasts run thousands of match simulations to project outcomes with 60-70% Arsenal chances in some models. (eldestapeweb.com) Opta maintains their xG model emphasizes shot quality and body position, while StatsBomb includes more situational factors like pressure from defenders. Users pick models based on whether they prioritize predictive power or explanatory insight. (fbref.com) The debate underscores how seemingly minor modeling decisions propagate through simulations, altering projected winners by up to 15% in tight races. No model perfectly captures chaos of football, so analysts cross-reference multiple sources. (thexgfootballclub.substack.com)
