Goals Saved Above Expected: What the Goalkeeper Stat Measures

Goals saved above expected measures how many fewer—or more—goals a goalkeeper conceded than a model expected from the shots they faced. A positive number indicates better-than-expected shot stopping; a negative number indicates more goals conceded than expected. MLS describes it as the number of extra goals saved compared with an average goalkeeper.

The key is the type of expected goals used. For this job, a post-shot model is more useful than the ordinary xG total shown during a match.

How it is calculated

The basic calculation is:

Goals saved above expected = post-shot expected goals faced − goals conceded from those shots

Post-shot expected goals, often shortened to PSxG, estimates scoring probability using information about the shot after it has been struck. Opta calls its version expected goals on target (xGOT) and calls the resulting difference goals prevented. StatsBomb uses goals saved above average (GSAA). These labels describe a similar comparison, but their underlying models should not be treated as interchangeable. Opta’s methodology and StatsBomb’s explanation describe their respective approaches.

Here is a hypothetical four-shot match, with probabilities supplied by one post-shot model:

The goalkeeper conceded once from shots expected to produce 1.7 goals, so finished 0.7 goals above expected. That is an estimated difference, not a claim that someone literally saved seven-tenths of a goal.

Notice how saves receive different credit. Stopping the 0.60 shot adds much more than stopping the 0.10 shot. Conceding the difficult 0.80 shot costs only 0.20; conceding a 0.10 shot would cost 0.90.

Why ordinary xG is not enough

Pre-shot xG evaluates the chance before knowing how well the attacker executes the shot. Post-shot xG accounts for the resulting shot’s difficulty to save. Opta’s model includes where the ball reaches the goalmouth; StatsBomb’s published model explanation also describes trajectory and speed information. Their shot-stopping models focus on on-target attempts rather than treating every blocked or off-target effort as a save opportunity. Opta, StatsBomb

A real match illustrates the difference. Daniel Sturridge’s long-range equaliser for Liverpool against Chelsea in 2018 had 0.03 xG, but its top-corner placement raised it to 0.58 xGOT in Opta’s model. The original chance was unlikely to produce a goal; the actual shot was much harder to stop. Opta’s worked example shows why subtracting goals conceded from ordinary team xG can misrepresent a goalkeeper’s performance.

For the classification behind the shot sample, see our guide to what counts as a shot on target.

What to check before comparing goalkeepers

The denominator. The raw total has no denominator: it is an accumulated difference measured in goals. More shots faced mean more opportunities to gain or lose credit. A per-90 version divides that total by minutes played and multiplies by 90; a per-shot version instead divides by the number of included shots faced. These answer different questions about total contribution, contribution per playing time and performance per shot. Opta uses goals prevented per 90; StatsBomb’s GSAA% expresses actual save percentage minus expected save percentage. That difference is in percentage points: saving 80% when the model expects 75% gives +5 percentage points.

The model and included shots. Compare figures from the same provider and check whether penalties are included. Use the goals conceded from the same shot sample—not an unadjusted team goals-against total. StatsBomb’s published goalkeeper examples, for instance, explicitly use non-penalty goals. StatsBomb

The sample size. One difficult save or one soft goal can move a short-run total sharply. StatsBomb notes that shot-stopping rankings are noisy in small samples. A strong match is evidence of a strong performance, not enough to establish lasting superiority. StatsBomb

The rest of the job. This is a shot-stopping measure, not a complete goalkeeper rating. Claiming a cross before anyone shoots, sweeping behind the defence and distributing the ball require separate assessment. A goalkeeper can prevent danger without adding anything to this total. StatsBomb’s analysis framework separates shot stopping, claims and distribution.