What Expected Goals Measures and Misleads

Expected goals, usually shortened to xG, has moved from specialist analysis into mainstream football coverage. It appears in broadcasts, match reports and debates about players and teams. Yet the metric is often used as though it were a verdict on who deserved to win, which is not what it was designed to tell us.

This explainer sets out what xG is, how models are typically built, and where caution is needed.

The basic idea

An expected goals model estimates the probability that a given shot will result in a goal. It does this by looking at large numbers of historical shots with similar characteristics and measuring how often they were scored.

A shot from close range in the centre of the penalty area is scored far more often than one from distance or from a tight angle. The model captures that difference by assigning each shot a value between 0 and 1. Adding up the values of all a team's shots gives its total xG for a match.

Common inputs include:

Shot location: distance from goal and the angle to it.
Body part: headers are generally converted less often than shots with the foot.
Type of assist: a cross, a through ball or a cut-back can lead to different conversion rates.
Situation: open play, set piece, counter-attack or penalty.

More advanced models may include information about defensive pressure or the goalkeeper's position, depending on the data available. Penalties are usually given a fixed value reflecting the historical conversion rate, which in most models is a little over three in four.

What xG does well

The strength of xG lies in aggregation. Over a single match, scorelines are shaped by randomness: a deflection, a goalkeeping error, an unusually good finish. Over many matches, the quality of chances a team creates and concedes tends to be a more stable indicator of performance than goals alone.

It helps separate process from outcome.
It gives context to a team's results over a longer run of games.
It provides a common language for comparing the volume and quality of chances.

Expected goals measures the quality of chances, not the quality of a performance. It is most useful when aggregated over time and read alongside what happened on the pitch.

Where it misleads

The limits of xG are as important as its strengths:

Finishing skill: most models treat shots as if taken by an average player. An elite finisher may consistently outperform xG, although separating genuine skill from variance requires large samples.
Non-shots: dangerous situations that do not end in a shot, such as a cross that narrowly evades a forward, contribute nothing.
Game state: teams that are leading often defend deeper and concede more shots, which can distort comparisons.
Small samples: a single match's total can be dominated by one or two big chances, making it a poor summary of the game.
Model differences: providers use different inputs and methods, so figures for the same match can vary.

There is also a subtler issue. xG measures chances that ended in shots. A team whose players pass up low-quality attempts in favour of better ones may produce fewer shots but more valuable ones. Shot selection is itself a tactical choice, and xG reflects it only indirectly.

Reading xG sensibly

A sensible reading treats xG as a lens rather than a verdict. It is valuable for questions such as whether a team's results look sustainable, or whether a striker is consistently getting into good positions. It is far less valuable as a final judgement on a single match.

A few habits help:

Prefer season-long or multi-match figures to single-game totals.
Look at the distribution of chances, not just the sum.
Check which provider produced the numbers before comparing them.

Combined with other information, such as pressing metrics and an understanding of what possession statistics can and cannot tell us, xG becomes part of a broader picture. On its own it answers one specific question: given the shots taken, how many goals would an average finisher be expected to score from them?

That is a useful question. It is just not the only one.