Analysis · evergreen
Expected goals: what the number actually tells you
In one line: Chance quality as a probability: where xG came from, what the model eats, why every site shows a different number, and the honest limits of football's most quoted stat.
Every broadcast graphic now carries it, every pundit leans on it, and most arguments about it start from a misunderstanding. Expected goals (xG) is not a prediction and not a rating of teams. It is one number per shot: the probability that an average shot from that situation becomes a goal, between 0 and 1. A chance rated 0.3 should score roughly three times in ten. Over a season, those probabilities tell you more about a team than the scoreboard does — which is exactly why it also gets misused every single week.
Where it came from
The modern model was first implemented in April 2012 by Sam Green, then an analyst at the sports data company Opta, adapting ideas that American sports analytics had been using for years. It stayed a niche tool until BBC's Match of the Day began building it into its coverage from the 2017-18 season, and it has been part of football's everyday language since. That history matters, because it explains the metric's shape: it was built to compare chance quality, not to settle bar arguments about who deserved a single result.
What the model actually eats
A shot's xG is estimated from the things that make shots miss: distance and angle to goal, the body part used, how the chance was built (a through ball versus a scramble), the position of defenders and keeper, and pattern of play. Two honest consequences follow. First, the number is a model's opinion, built from historical shots of the same type. Second, different providers run different models — Opta, StatsBomb and the rest weigh the ingredients differently, which is why the same goal gets 0.34 on one site and 0.41 on another. Both are doing their job; the differences are usually smaller than the arguments about them.
What it is good for
Over months, not minutes. Chance creation and chance suppression are among the most stable signals of how good a team actually is — results wobble around them. xG separates a striker in a finishing hot streak from one getting genuinely better chances, shows when a defence is conceding dangerous opportunities rather than surviving on luck, and gives recruitment departments a shared language for “does this player get into good positions?”
The honest limits
Small samples lie. A single match is noise; a team that “won the xG” has not won an argument, it has taken one shot-quality snapshot. Game state bends everything: a leading team stops shooting and a chasing one piles forward, so the numbers describe the situation as much as the quality. The goalkeeper is invisible to it — xG rates the chance, and the save is the other half of the story (a keeper who concedes above his chances' rating is facing, or failing, his own column). And a 0.95 chance that stays out is not a scandal: probability is not a promise, which is the sentence most xG arguments forget first.
The fair summary: xG measures chances, not football. Used over a season it is the most honest number in the game; used to pronounce on one Saturday it is astrology with decimal points. This desk uses it the first way — the same rule applied to every number we publish: know what it measures, then say so.
Sources
Next