Why Overperforming xG Is Completely Normal — explained

A striker scoring more than his expected goals is not a statistical scandal. Variance, shot selection and genuine finishing skill all push a total away from expectation, and the interesting question is which of the three is doing the work.

The arithmetic of a small sample

Suppose a player takes eighty shots in a season with an average value of 0.15. His expected goals total is twelve. Because each shot is an independent low-probability event, the actual number of goals will land anywhere in a band around twelve that is roughly three or four goals wide, purely from chance. Scoring sixteen is entirely unremarkable.

This is the sense in which overperformance is normal rather than suspicious. The metric reports the centre of a distribution, and a single season is one draw from that distribution. Reading a draw as a deviation from truth inverts the statistics: the draw is the observation, and the expectation is the summary of many observations.

Selection: taking better shots

A second real mechanism is that players and teams influence which chances they take. A forward who declines low-value long-range efforts and waits for better positions will accumulate fewer shots and a lower expected-goals total than a volume shooter, while converting at a higher rate per attempt. His goals can outrun his expected goals without any finishing skill being involved at all.

That pattern is a shot-selection effect, and it is repeatable. Detecting it means looking at the number of attempts as well as their total value, because a low shot count with a high average value per shot describes a selective player and a high count with a low average describes a volume one.

Finishing skill, and how to see it

The third mechanism is the contested one. If a small number of players genuinely beat the average conversion rate of comparable positions across their careers, then finishing is a skill rather than noise. The evidence for this is real but modest: the repeatability of a player's conversion residual is weak-to-moderate, and it takes a long career to separate from luck.

The practical consequence is that overperformance should be treated as a hypothesis and not a verdict. One strong season creates a suspicion of skill; three consistent seasons begin to constitute evidence. A model that flagged every seasonal overperformance as luck would be as crude as one that treated every one as proof of genius.

Underperformance is the mirror

Everything above applies with the sign reversed. A player scoring six against an expectation of twelve is not necessarily wasteful; he may be unlucky, may be taking his chances from worse positions than his teammates, or may be a genuinely poor finisher. The three explanations are the same three, and the diagnostic order is the same as well.

The asymmetry people notice is psychological rather than statistical. Fans rationalise underperformance as bad luck and overperformance as class, and both readings are premature at the scale of a single season. Holding the same standard for both directions is the whole discipline.

Sample size is the deciding variable

The shorter the window, the more the residual is dominated by chance. Over ten matches the variance band is so wide that almost any result is consistent with the model. Over a career of several hundred attempts the band narrows enough that a persistent residual becomes something other than noise, and that is where the genuine skill debate lives.

A useful discipline is to ask how many attempts sit behind any claim about finishing. A residual built on forty shots carries almost no information; one built on four hundred carries some; one built on a thousand is a serious datum. The number of shots is the denominator of the whole discussion.

Why the residual can persist for a whole team

Teams can overperform their expected goals for longer than individuals, for a structural reason: a side's attacking model may systematically create chances the average model undervalues, such as repeated cut-backs or second-phase set plays. The model treats those as ordinary shots within their geometry, and the side converts them at a higher rate because its players are rehearsed in exactly that pattern.

That is the strongest argument that overperformance can be a signal rather than noise. It also has a practical test: if a club's overperformance survives a change of manager, personnel and season, the pattern is likely structural. If it evaporates when one player departs, it was that player, and the residual belongs to him rather than to the system.

Key reference points

• A single season is one draw from a distribution a few goals wide either side of the mean.
• Shot selection can raise a scoring rate without any finishing skill being involved.
• Finishing skill is real but weakly repeatable, and needs a long career to detect.
• Underperformance has the same three explanations as overperformance.
• Always ask how many attempts sit behind a claim about finishing.
• Team-level overperformance persists longer and is more likely to be structural.

The residual between goals and expected goals is a question, not an answer, and the honest response to it is to check the sample before choosing an explanation.