What Is xG? Expected Goals, Explained in Depth

What xG measures — and what it deliberately ignores

xG (expected goals) assigns each shot a goal probability grounded in historical data: a tight-angle strike from outside the box is worth around 0.03 xG, while a tap-in from the six-yard line can exceed 0.9 xG. Summing every shot gives each team a match total that says who created the better chances.

The power of xG lies in what it ignores: finishing luck, goalkeeping heroics and shots that rattle the post. The scoreboard tells the story of a moment; xG tells the story of the production behind it. That is why xG is a core input for process-driven match analysis rather than result-driven storytelling.

How xG is calculated: from shot features to probabilities

An xG model is trained on hundreds of thousands of historical shots. For every new shot it typically evaluates:

• Distance and angle: how far from goal and how wide the visible target is — the two strongest predictors.
• Body part: headers from the same spot convert less often than footed shots.
• Play situation: open play, set piece, fast break or penalty (about 0.76 xG).
• Assist and pressure context: through ball, cross or turnover, plus how much defensive pressure the shooter faced.

The model then asks how often historical shots with those exact features were scored, and outputs a value between 0 and 1. xG is not an opinion — it is a summary of historical frequencies.

xG vs the actual score: why they disagree

Football is a low-scoring sport, and single-match variance is enormous. A team generating 2.1 xG can finish goalless without anything being statistically "wrong" — rare events simply misbehave in small samples.

Three forces drive the gap: finishing skill (some players convert above the model's average), goalkeeper performance (an over-performing keeper suppresses the scoreline), and plain small-sample luck. Over longer windows — usually 8-10 matches — goals scored converge toward cumulative xG. A one-match gap between xG and the score is not an error; it is the nature of the game, and it never means the next result is somehow settled in advance.

Why xG matters more than the scoreboard over time

A league table is a backward-looking record of outcomes; xG carries forward-looking signal. Analytics research has repeatedly found that a team's xG history predicts its future goal output more reliably than its actual goal history does.

Practical readings: a side that keeps producing high xG but scores little is usually a good team hitting bad finishing variance; a side scoring plenty from low xG is usually riding an unsustainable hot streak. When judging form, the last 5-10 matches of xG and xGA (expected goals against) mislead far less than a bare win-loss strip. For the underlying terminology, see the glossary.

How 11Stat uses xG inside its 12-engine analysis

At 11Stat, xG is not a verdict — it is an input. Team attack and defence strengths are weighted by xG-based production profiles instead of raw goal counts, and those profiles feed Poisson and Dixon-Coles score-distribution engines as well as form and momentum factors. Twelve independent engines then meet in a consensus layer whose probabilities are calibrated against historical performance.

The 0-10 confidence score and the most likely scores you see on an analysis card are derived from these xG-informed distributions. The output is always a probability: when the model leans toward an outcome, that outcome is more likely on the data — never certain.

A worked example: reading a full match in xG

Take a hypothetical match. The home side attempts 14 shots, mostly from distance, totalling 1.1 xG. The away side manages only 6 shots, but three are close-range chances, totalling 1.6 xG. The match ends 1-0 to the home team.

The scoreboard says the hosts won; the xG reading tells a different story — the visitors created the better chances, and repeating that performance would earn more goals over time. The right conclusion is neither "the score lied" nor "xG is infallible": read them together. The last-5-match xG sparklines on 11Stat cards exist precisely to surface these divergences — as context, not as a promise about any future result.