Post-Shot xG and Shot-Stopping: definition and limits

Pre-shot expected goals values the chance before anyone strikes it. Post-shot expected goals values it after the ball has left the boot, which is the only way to judge the shot the goalkeeper actually faced.

Why the shot must be revalued

A pre-shot model sees a chance from twelve metres in the centre of the box and assigns a value near a quarter of a goal. It does not see whether the resulting shot went into the corner or straight at the goalkeeper. Judging a keeper on pre-shot values therefore charges him for chances he saved brilliantly and credits him for chances the attacker put wide.

Post-shot expected goals revalues the attempt after it is struck, using the trajectory, the placement within the goal frame and the speed of the ball. A shot aimed into the bottom corner keeps a high value; an identical attempt struck at the keeper's chest falls to a low one. The difference is precisely the save.

How the metric is computed

The model takes the pre-shot geometry as a starting point and adds the end point of the ball. Shots on target receive a value reflecting how far from the goalkeeper's starting position they finished; shots off target are reduced to zero, because no keeper saves a shot that misses. The resulting figure is the probability that an average goalkeeper concedes.

Summed across a match or a season, post-shot expected goals describes the quality of the shots that actually required a save. It is the correct denominator for shot-stopping, and the difference between it and the goals conceded is the standard measure of a keeper's contribution.

The tracking-data requirement

Post-shot values cannot be computed from coordinates alone. The provider must know where within the goal frame the ball finished and how fast it travelled, which requires either tracking hardware or detailed hand annotation of every shot. Providers without that pipeline publish pre-shot values only and label them accordingly.

This is the first place a reader should check when comparing keeper statistics from two pipelines. A post-shot figure from a tracking provider and a pre-shot figure from a video provider are not variations on the same number; they are different measurements, and ranking keepers across the two is a category error.

The penalty problem

Penalties present a specific difficulty. Their pre-shot value is enormous and nearly constant, and their post-shot value depends on placement in a way that gives the keeper a chance only if the penalty is poorly struck. Many models exclude penalties from post-shot keeper metrics entirely, which is the cleaner choice.

Where penalties are excluded, the exclusion must be stated, because a keeper who faced six penalties and saved two will look different under the two treatments. Reporting penalties as a separate line, as with every other metric on this site, avoids the ambiguity.

Reading it over a season

Post-shot expected goals is stable enough over a season to be informative, but keeper performance relative to it is not. Shot-stopping above expectation has low season-to-season repeatability, which means one strong season is weak evidence and three is stronger. Treating a single campaign as proof of a keeper's quality is the standard error.

The metric is nevertheless far better than save percentage for the same purpose, because it accounts for the difficulty of the shots faced. A keeper behind a dominant defence faces few shots, and the ones he faces are usually from distance; his save percentage will be high without him having done anything unusual.

Reporting keeper performance

The complete report is four numbers: shots faced, goals conceded, post-shot expected goals and the difference between the last two. The difference, expressed as goals prevented, is the headline; the other three are what make it interpretable. Without the shot count the reader cannot tell a busy keeper from an idle one.

Where the figure covers a partial season, the caption should say so, because a keeper who has faced a hundred shots has a meaningfully different sample from one who has faced four hundred. The same sample-size discipline that applies to strikers applies with more force to goalkeepers, whose opportunities are rarer.

Key reference points

• Post-shot expected goals revalues a shot after it is struck.
• It is the correct denominator for judging shot-stopping.
• It requires tracking or detailed annotation; pre-shot values are not a substitute.
• Many models exclude penalties; the exclusion must be stated.
• Shot-stopping above expectation repeats weakly between seasons.
• Report shots faced, goals conceded and post-shot expected goals together.

A goalkeeper can only be judged on the shots that actually arrived, which is why the post-shot model exists and why pre-shot figures flatter and punish him at random.