Expected goals (xG), without the maths

In one linexG gives every shot a number between 0 and 1 that says how often shots like it have been scored in the past.

What xG measures

Expected goals, usually written xG, is a way of rating the quality of a chance. Every shot gets a value between 0 and 1. That value is the share of similar shots, taken in the past, that ended up in the net.

A shot worth 0.10 xG is the kind that goes in about one time in ten. A shot worth 0.50 goes in about half the time. A value of 1 would be a chance that is scored every single time.

The number rates the chance, not the finish. xG is worked out from what was known at the moment the ball was struck. Whether the shot then flew into the top corner or into the stand does not change it.

Where the numbers come from

Data companies build an xG model — a set of rules learned from past data. They collect a very large number of old shots, note which ones went in, and let a statistical method learn which features make a goal more or less likely. Opta, the data brand of Stats Perform, says its model learns from nearly one million shots and weighs more than 20 variables.

Typical inputs include:

• Distance and angle to the goal.
• Body part — foot or head, for example.
• Type of play — open play, fast break, corner, direct free-kick or throw-in.
• How the chance was made — the kind of pass or action before the shot.
• Where the goalkeeper and defenders were. Hudl StatsBomb records the position of every player in the camera frame at the moment of the shot, called a freeze frame. Opta’s model also considers defensive pressure and the goalkeeper’s position.

Penalties are usually given one fixed value, based on how often penalties are scored. Opta uses 0.79. Hudl StatsBomb uses 0.78. That small gap is a useful reminder: xG is an estimate from a model, not a fact about the match.

xG versus goals over a season

Add up the xG of every shot a player or team takes and you get their total xG. You can then compare it with the goals they actually scored.

In a single match the gap can be large. A team can create 2.5 xG and still lose 1–0. Good goalkeeping, bad luck and fine margins all play a part.

Over a season, the picture usually becomes clearer. A season’s worth of shots smooths out much of the match-to-match noise. A team that keeps creating more xG than it concedes is, on the whole, making better chances than its opponents — even during a run of poor results.

Scoring more goals than your xG is called overperforming it. Scoring fewer is underperforming. Research published by Hudl StatsBomb suggests that most players finish at close to the average rate. In its view, what separates the best forwards is mainly that they get into good shooting positions more often.

Useful variants

npxG — non-penalty xG

Penalties carry a large fixed value, and one designated player usually takes them. Removing them gives non-penalty xG (npxG). It is a fairer way to compare how many chances players or teams create from everything else.

xA — expected assists

Expected assists (xA) credits the passer instead of the shooter. Providers define it in different ways:

• Opta’s xA rates how likely a completed pass is to become an assist, using the type, end point and length of the pass. The pass earns xA whether or not a shot follows.
• Hudl StatsBomb’s “xG assisted” gives the passer the xG of the shot their pass set up. If no shot follows, there is nothing to credit.

Before comparing xA figures from two websites, check which definition each one uses.

Post-shot xG and goalkeepers

Ordinary xG is “pre-shot”: it ignores where the shot went. Post-shot xG (PSxG) adds what happened after the strike, above all where the ball was heading in the goal mouth. Hudl StatsBomb’s version also uses the shot’s path and speed. Opta calls its version expected goals on target (xGOT).

These models only rate shots on target. A shot that misses or is blocked scores zero, because it can no longer beat the goalkeeper.

That makes post-shot xG a tool for judging goalkeepers. Add up the post-shot xG of the shots a keeper faced, then subtract the goals they conceded. Own goals are left out, because the keeper faced no shot; Opta’s worked example leaves out penalties as well. A positive result suggests the keeper kept out more than an average keeper would have. Opta calls this figure goals prevented.

Limits and common misuse

• Small samples mislead. A striker’s xG over five matches says very little. Whole seasons say more than a few weeks.
• No shot, no xG. A dangerous attack that ends without a shot adds nothing to standard xG, however close it came.
• The score changes behaviour. A team defending a lead may sit deep and allow many low-value shots. Its xG against then partly reflects a choice, not only weakness.
• It assumes an average finisher. xG says how often a typical player scores from that position. It is not adjusted for the individual taking the shot.
• Providers disagree. Models use different data and different inputs, so the same shot can carry different values on different sites. Compare figures from one provider rather than mixing several.

Where to go next

• Football data: events, tracking and physical — where shot data comes from, and other common metrics such as per-90 figures and pressing measures.

Key terms

• xG — the probability, from 0 to 1, that a shot of that kind is scored.
• npxG — xG with penalties taken out.
• xA — expected assists; credit for the pass, defined differently by different providers.
• PSxG / xGOT — post-shot expected goals; rates shots on target using where they were heading.
• Goals prevented — post-shot xG faced by a goalkeeper minus goals conceded.