Save Percentage and Its Traps: definition and limits

Save percentage is the oldest goalkeeper statistic and still the most quoted. It is also the one most likely to produce the wrong conclusion, because it counts the shots a keeper faced rather than their difficulty.

The definition and its appeal

Save percentage divides saves by shots on target. It is a single ratio, it is available for every match ever played with a record of shots on target, and it needs no model, no tracking and no estimation. Its availability is the reason it has outlived almost every other statistic from the same era.

The appeal is real, and for a rough answer about a keeper's shot-stopping in a single match it is not misleading. The problems begin when it is used to compare keepers across a season or across teams, which is exactly the use it most often receives.

The shot-quality problem

A keeper behind a strong defence faces few shots, and the ones he faces tend to come from distance, because a good defence concedes possession in front of itself and rarely allows the ball into the box. His save percentage will be high for reasons that have nothing to do with him, and the same effect works in reverse behind a weak defence.

Post-shot expected goals exists specifically to correct this, by asking how hard the shots were rather than how many were stopped. Where both are available, the post-shot figure is the one to quote, and the save percentage survives only as a rough descriptive statistic.

The definitional drift

Providers disagree about what counts as a shot on target. A blocked shot, a shot that deflects off a defender, a cross that drifts goalwards and a clearance that heads towards its own goal are each treated differently by different providers. Those choices move the denominator and therefore move the percentage.

There is also the question of goals that are not attributable to the keeper at all, such as a deflected effort or a penalty. Most save percentages exclude penalties and some exclude free-kicks struck directly, and a figure quoted without that information is a figure whose denominator is unknown.

The small-sample effect

A keeper facing two hundred shots in a season has a save percentage with a confidence band wide enough that a few goals either way moves the figure by several percentage points. Differences of two or three points between two keepers in one season are well inside the noise, and they are nevertheless the differences that appear in published comparisons.

The problem compounds over part-seasons. A keeper with six appearances can post an extreme percentage, and the extreme figure will be quoted because it is remarkable. The remarkableness is a property of the sample size rather than of the keeper.

What save percentage is still good for

As a descriptive statistic inside one team and one system, save percentage is harmless. A keeper whose save percentage has collapsed over six weeks is conceding shots that he previously stopped, or facing better chances, and either way the fact is worth investigating. As a measure of trend within a stable context, the number works.

The failure is comparative. Across teams, across leagues and across eras the denominator changes so much that the ratio stops being a measurement of the keeper. The rule is therefore simple: use it within one system, never across systems, and prefer the post-shot model wherever it exists.

Reporting it

Where save percentage must be reported, it should be accompanied by the shots faced, the share of them taken from inside the box, and a statement of whether penalties are included. Those three pieces of context are enough for a reader to judge how much of the figure is the keeper and how much is the defence in front of him.

The stronger recommendation is to report post-shot expected goals and goals prevented instead, with save percentage as a legacy companion figure for continuity with older records. Where a keeper's career spans the era before tracking data, save percentage is the only comparable statistic that exists, and its limits should be stated rather than hidden.

Key reference points

• Save percentage divides saves by shots on target and needs no model.
• It measures the defence's shot profile as much as the keeper's ability.
• Providers differ on what counts as a shot on target, shifting the denominator.
• A season of two hundred shots has a wide confidence band; small gaps are noise.
• It is a useful trend statistic inside one system and a poor comparative one.
• Report shots faced, inside-box share and penalty treatment beside it.

Save percentage is a description of a defence wearing a goalkeeper's gloves, which is why it should be read inside one team and never across a league table.