Goals Prevented — explained: definition and limits

Goals prevented subtracts goals conceded from post-shot expected goals faced. It is the single best public measure of shot-stopping, and it is routinely over-read because of how noisy the underlying quantity is.

The computation

Take the post-shot expected goals the keeper faced across the season and subtract the goals he conceded. A positive figure means he conceded fewer than an average keeper would have been expected to concede from those shots; a negative figure means he conceded more. The unit is goals, which makes the number unusually easy to interpret.

The simplicity is the metric's main attraction. Unlike a rate expressed in abstract units, a figure of plus eight says that eight goals did not happen that an average keeper would have allowed, which is a sentence anyone can use. The clarity of the unit is also what encourages people to over-trust it.

Why the variance is large

The quantity is a sum of small differences across a few hundred events, each with a large random component. The result is that the season-to-season correlation of goals prevented is weak: a keeper at plus ten in one campaign is as likely to be near zero in the next as to repeat, and keepers who lead the metric one season frequently regress sharply.

This does not mean the metric is useless. It means that a single season of it carries little information about a keeper's ability, while it does carry a great deal of information about that particular season. Distinguishing those two readings is the whole discipline of using it.

The confounds

Several factors inflate or deflate the figure without saying anything about shot-stopping. A defence that concedes many shots from distance produces a low post-shot value per shot and a keeper who looks efficient. A defence that concedes few but high-quality chances produces the opposite. Neither pattern is a property of the goalkeeper.

Score effects matter too. A keeper in a side that is frequently behind faces a different mix of shots than one who is frequently ahead, because the opponent's approach changes with the state of the match. Season totals blend both states and produce a number that describes a team's season as much as a keeper's hands.

Using it fairly

The fair use of goals prevented is comparative across many seasons or across many keepers in the same season, never as a verdict on one man from one campaign. Even then, the differences that matter are larger than the noise, and a gap of three or four goals between two keepers in a single season is close to meaningless.

It is also worth reporting beside the keeper's shot volume and the average post-shot value per shot he faced. Those two numbers describe the workload and its difficulty, and without them the headline figure is a difference without context, which is the most common way a good metric gets misused.

The penalty exclusion

Because penalties carry such a high and nearly constant value, including them in the computation adds a large variance term from a small number of events. A keeper who faced five penalties and saved none will lose nearly four goals from his figure through an event class that is largely independent of him.

Most careful treatments exclude penalties and report them separately, which is consistent with the treatment elsewhere on this site. Where a published figure includes them, the reader should expect a noisier number and should not compare it with a figure that excludes them.

What it does and does not show

Goals prevented measures one thing: how many goals the keeper conceded relative to the shots he faced. It says nothing about claiming crosses, sweeping behind the defence, distributing the ball or organising a back line, all of which are large parts of modern goalkeeping. Those have their own metrics on this site.

It follows that a low figure for goals prevented is not evidence that a keeper is poor overall. It is evidence about shot-stopping in a particular window, and the other columns of his profile are where the rest of the assessment lives.

Key reference points

• Goals prevented is post-shot expected goals faced minus goals conceded.
• Its unit is goals, which makes it easy to interpret and easy to over-trust.
• Season-to-season repeatability is weak; one campaign is not evidence of ability.
• Defensive shot profiles and score effects confound the figure substantially.
• Most careful treatments exclude penalties and report them separately.
• It measures shot-stopping only, not the keeper's other responsibilities.

Goals prevented is a good measure of a season and a poor measure of a goalkeeper, and keeping those two claims apart is what makes the metric useful.