PSxG and Goalkeeper Saves: How Modern Stats Measure Shot-Stopping

Post-shot expected goals (PSxG) is a measure of how likely a shot was to become a goal at the moment it left the boot, judged after the strike using where the ball was actually heading. It is the modern game's most direct attempt to isolate one thing a goalkeeper is paid to do: stop shots that should, on balance, go in.

Why save percentage was never enough

For decades, the headline number for a goalkeeper was save percentage — the share of shots on target that he kept out. It is simple, intuitive, and badly flawed. A keeper who faces twenty tame shots from distance will post a glittering save percentage; one who faces ten point-blank chances a game will look worse, even if he is the superior shot-stopper. Save percentage treats every shot as equally difficult, and no two shots are. Worse, it rewards a keeper for the defence in front of him: a back line that funnels opponents into long-range efforts will inflate his numbers without him making a single difficult stop.

The whole point of a shot-quality model is to fix that. If you want to judge how well a goalkeeper handles what he faces, you first have to measure how hard what he faces really is. That is the gap post-shot expected goals was built to close.

What post-shot expected goals actually measures

Ordinary expected goals (xG) scores a chance at the instant the shot is taken, from factors the striker controls or inherits: distance, angle, body part, the type of pass that created it. It answers the question, how good was this chance?

PSxG asks a different question: how hard was this shot to save? It is calculated only for shots on target, and it adds the one piece of information ordinary xG ignores — where the ball was going. A shot drilled into the top corner carries a far higher PSxG than the same shot rolled straight at the keeper, because placement is most of what makes a shot unsaveable. By building the ball's trajectory into the number, PSxG estimates the probability that an average goalkeeper would concede from that strike.

This is why the two numbers diverge in useful ways. When a shot's PSxG is much higher than its pre-shot xG, the striker has found a corner the chance did not deserve. When it is lower, the effort was straight at the goalkeeper. Read across a season, that gap describes finishing quality on one side of the ball and shot-stopping difficulty on the other.

Goals prevented: the shot-stopping scoreline

The most useful thing PSxG unlocks is a single, honest scoreline for a goalkeeper's shot-stopping. Add up the PSxG of every shot on target a keeper faces, then subtract the goals he actually conceded. The result is usually called goals prevented, or goals saved above expected:

• A positive figure means he conceded fewer than an average keeper would have from those shots — he is saving the unsaveable.
• A figure around zero means he is stopping roughly what the model expects.
• A negative figure means shots are beating him that an average keeper would have kept out.

This reframes shot-stopping as a running total rather than a percentage, and it travels well across teams. A goalkeeper behind a leaky defence and one behind a watertight back line can finally be compared on the same terms, because the metric judges each only against the difficulty of the shots he personally faced. Platforms such as RubiScore track goals prevented alongside the raw save numbers, so a keeper's season can be read as how much he added to his team rather than how busy he happened to be.

Reading PSxG in a single match

Over one game, PSxG is a storyteller rather than a judge. A goalkeeper who faces 2.4 PSxG worth of shots and concedes once has had a strong night relative to what came at him; one who faces 0.6 and lets in two has had a poor one, or been undone by deflections the model cannot foresee. The single-match number is far too small to rate a career, but it answers the question a scoreline hides: was a clean sheet earned against real pressure or a quiet evening of half-chances, and did a heavy defeat sit with the keeper or with the defence that kept feeding shots to him?

It also reframes the spectacular goal. When a strike carries a very high PSxG — a thunderbolt into the top corner — the model is telling you it was always likely to beat any goalkeeper. Blaming the keeper for conceding it misreads the night, just as crediting him for saving a tame effort overstates it. Used this way, match-level PSxG turns highlight-reel arguments into something closer to evidence, and stops a single moment from defining how a goalkeeper's performance is remembered.

What the numbers leave out

A coach evaluating a goalkeeper on PSxG alone would be making the opposite mistake to the one save percentage invites. Shot-stopping is only part of the job, and it is the part these numbers cover. They say nothing about:

• Distribution. Modern goalkeepers start attacks. Pass completion, launch accuracy, and the ability to play through a press are separate skills with their own data.
• Command of the box. Claiming crosses, punching clear, and dominating set pieces prevent shots from ever happening, so they never appear in a shot-stopping model.
• Sweeping. A keeper who reads danger and clears outside his area kills chances pre-emptively. Defensive actions and average distance from goal capture this better than PSxG can.
• Positioning. The best positioners make hard saves look routine. Ironically, that can suppress their PSxG totals, because a shot taken at a well-positioned keeper is modelled as easier than the same shot into open space.

That last point is the subtle one. PSxG measures the outcome of a save attempt, not the positioning that set it up — so a goalkeeper can be elite precisely because he turns high-difficulty situations into low-PSxG shots before the ball is even struck.

A framework for reading a goalkeeper's season

Put together, a fair read of a goalkeeper follows a simple order:

• Start with goals prevented over a meaningful run of matches to gauge raw shot-stopping.
• Set it beside the volume and quality of shots faced, so the number has context.
• Layer in distribution data to judge what he offers in possession.
• Add cross-claiming and sweeping actions to capture the chances he prevents rather than saves.
• Only then weigh the eye-test moments — the command, the calm, the errors — against the numbers.

No single metric crowns a goalkeeper, but this sequence stops any one number from misleading you, which is exactly the trap save percentage set for so long.

The patience problem

One warning sits above all of this: goalkeeping data is noisy, and it settles slowly. A goalkeeper faces far fewer shots than an outfielder has touches, so goals prevented can swing wildly over a handful of games. A keeper riding a hot streak of fingertip saves will show a flattering total that regresses; one who has conceded two freak deflections will look worse than he is. Penalties, which distort the picture, are often stripped out and judged on their own.

The fix is patience. Shot-stopping numbers mean little over five matches and a great deal over fifty. Judging a goalkeeper on a small sample is how clubs talk themselves into and out of the wrong decisions, and the metrics only help if they are read across enough football to be real. Tracked match by match and totalled across a season, post-shot expected goals and goals prevented turn the hardest position on the pitch to evaluate into one that can finally be measured fairly — and the full set of goalkeeper statistics is published, season by season, at rubiscore.com.