Expected Goals Explained: What xG Measures and Misses
Expected goals explained simply is the sentence every analyst has to write at some point: a shot from the penalty spot is worth more than a shot from thirty yards, and expected goals puts a number on that difference. The model assigns each shot a probability of becoming a goal, then adds those probabilities up across a match or a season.
That is the whole idea, and almost every argument about the metric comes from forgetting how narrow it is. Expected goals describes shots; it says nothing about the player who never got into position to shoot.
Expected goals xG: what the model is counting
Expected goals xG models are trained on historical shots. Each shot is described by features such as distance, angle, the part of the body used, whether it followed a pass or a rebound, and the pressure on the shooter. The model then returns the share of similar shots that were scored.
The output is not a prediction about a specific shot, and it does not know who is shooting. A shot worth 0.12 is a chance that historically produced a goal twelve times in a hundred attempts, whoever took it.
Different providers use different feature sets and different training windows, which is why two public models can disagree by several goals over a season. The disagreement is usually small per shot and large once the shots are summed.
XG explained football: shot quality and shot volume
XG explained football is mostly a conversation about two things: how good the chances were and how many of them there were. A team generating 1.9 expected goals from twenty shots is creating something different from a team generating 1.9 from six, and both totals hide the distinction.
For scouting purposes the more useful split is between a player's shot volume and his average shot quality. High volume with low quality suggests a player who shoots from anywhere; low volume with high quality suggests a player who arrives in dangerous positions but rarely gets the ball.
- Volume: shots per 90 minutes, which is mostly a description of role.
- Quality: expected goals per shot, which is mostly a description of movement.
- Conversion: goals minus expected goals, which is mostly noise over one season.
Expected goals xG: penalties, deflections and distance
Expected goals xG carries a few known distortions. Penalties are worth around 0.76 each, which is a large number attached to a repeatable event, so most analysts strip penalties out before comparing forwards. Removing them is standard practice, and stating which version of the number you are using is not optional.
Deflections and rebounds land in a similar place. A shot that takes a heavy deflection is still recorded against the shooter, and a rebound from three yards counts as an excellent chance even when it came from a scuffed effort. Over 30 matches these effects mostly wash out, and over five they can define a narrative.
XG explained football: what the number cannot see
XG explained football properly has to admit three blind spots. The model cannot see the pass that was not played, the run that created space for somebody else, and the pressing work that produced the turnover five seconds before the shot.
It also cannot see the shooter's identity, which matters for scouting. Two players can post identical expected goals totals while one is failing to convert chances a stronger finisher would score and the other is scoring from nothing. The metric ranks the situations they found, not the finishing.
| Quantity | What it tells you | What it misses |
|---|---|---|
| Goals | Outcome | Repeatability across seasons |
| Expected goals | Quality of chances found | Who was shooting, and why |
| Expected goals per shot | Shot selection | Volume of the role |
| Goals minus expected goals | Finishing overperformance | Sample size and randomness |
| Expected goals per 90 | Chance involvement per minute | Minutes against strong or weak opponents |
Reading expected goals per 90 without fooling yourself
Per-90 numbers need minutes attached to them. A forward with 0.6 expected goals per 90 across 700 minutes and one with the same rate across 2,600 minutes look identical, and only one of them has shown a season-long pattern.
The second discipline is league context. Chance quality and defensive standard vary between divisions, so a rate carried over the border needs a translation factor before it means anything, and the model itself was usually trained on a specific league's data.
The third is opponent strength. Expected goals per 90 racked up against bottom-half sides tells a department more about the team than the player, and the split by opposition is where that becomes visible.
Where expected goals explained properly changes a shortlist
The metric earns its place in recruitment as a filter, not as a verdict. It is very good at surfacing forwards whose chance quality is high and conversion is currently poor, which is often the cheapest group of players on the market.
It is also useful in reverse. A forward with a strong goal tally and a weak expected goals number has probably been finishing an unsustainable share of low-value chances, and that is exactly the situation a passing profile and a second live viewing should interrogate.
At that point the model has done its job and the judgement belongs to a person. How a centre forward is then graded and written up is a separate exercise, and the scale used to rank the result is set out in our guide to the scouting report grading scale.
