Grading Goalkeepers by the Shots They Should Have Let In

Regular expected goals is calculated before the shot is taken, from where the ball is on the pitch, the angle to goal, and how the chance arrived. That makes it a good measure of the chance a player was given, but a poor measure of a goalkeeper's performance, because it has no idea where the shot actually went. A player can be gifted a huge chance worth 0.6 xG and scuff it tamely into the keeper's arms, or the same chance can be lashed into the top corner. Pre-shot xG treats both attempts identically. The goalkeeper's job in each is completely different.

Post-shot expected goals, PSxG, fixes that by recalculating the shot's danger after it's been struck, using where the ball actually crossed the goal line, or would have if uninterrupted, along with its speed. A shot placed in the corner, inches from the post, generates a high PSxG regardless of how far out it was taken from, because keepers save shots hit straight at them far more often than shots placed into the corners. A tame effort straight down the middle keeps a low PSxG even if the pre-shot chance quality was excellent, because any goalkeeper at this level saves that shot most of the time.

Comparing PSxG to goals actually conceded is the cleanest widely available way to rate shot-stopping. A keeper who faces shots totaling 30 PSxG across a season and concedes only 24 goals has prevented roughly six goals above what an average keeper would be expected to stop, given the placement and power of what was fired at him. A keeper conceding 35 goals from 28 PSxG is doing the opposite, letting in shots that, on placement alone, an average keeper saves more often than not.

This matters because raw save percentage and goals-against totals both get distorted by the defense in front of the keeper. A back line that concedes a lot of low-quality, straight-at-the-keeper shots inflates a keeper's save percentage without him doing anything special. A back line that keeps conceding one-on-ones and shots into the far corner will drag a good keeper's save percentage down through no fault of his own. PSxG minus goals conceded controls for shot placement in a way raw saves can't, which is why it's become the standard public reference point for keeper performance, even though it still isn't a complete picture.

What it doesn't capture is everything that happens before the shot: positioning that cuts down the angle before the ball is even struck, command of the box on crosses, distribution that starts attacks, and the sweeper-keeper work of clearing danger outside the box entirely, which prevents a shot from happening at all and therefore never shows up in a PSxG number. A goalkeeper who consistently forces shots wide through good angle play will look ordinary on PSxG minus goals conceded simply because he's facing fewer high-value chances in the first place, which is a different kind of good goalkeeping that this particular number was never built to reward.

It also needs a reasonable sample before it means much. A handful of matches can swing wildly on one or two exceptional saves or one or two soft goals, the same variance problem that affects every shot-based metric over a small window. Over a full season, though, PSxG minus goals conceded does a better job than almost any other public stat of separating keepers who are genuinely shot-stopping well from keepers who are simply facing easier shots.

It picks up exactly where expected goals leaves off, once the shot has actually been struck, and it pairs well with non-penalty xG when judging the attacker on the other end of the same shot.