Separate Volume from Impact

Ask which question the number answers

“Who scored more?” and “Who scored more frequently while playing?” are different questions. A total is appropriate for the first. A rate may help with the second. Neither automatically answers who contributed more to the team's overall performance. A player can add value by creating space, progressing the ball or preventing a dangerous transition without appearing in a goals column.

Define the task before selecting the metric. If you are comparing forwards, specify whether you mean finishing, chance generation, pressing or availability. If you are comparing midfielders, decide whether the role is ball progression, defensive coverage or final-third creation. A single ranking often combines these tasks using unstated value judgments.

Make exposure visible

Consider two fictional players in the same competition and season. Player A scores six goals in 900 minutes; Player B scores eight in 1,800. Dividing by minutes and multiplying by 90 gives rates of 0.60 and 0.40 goals per 90 respectively. A has the higher observed scoring rate. B has the larger total and twice the recorded playing time.

The rate comparison does not prove A would maintain that output over B's minutes. Substitute appearances may occur against tired defenders or in unusual score states. Availability also has sporting value: a strong rate across a short spell does not supply the same season-long contribution as sustained performance.

Always show minutes beside per-90 rates. Any minimum-minutes threshold should be published and justified, not chosen after seeing which player it excludes. A threshold improves readability; it does not magically remove sampling uncertainty.

Separate opportunity from execution

Shot counts measure attempts, not their quality. Expected goals, usually abbreviated xG, assigns a scoring probability to an attempt using a model and its available features. Summing those probabilities describes the chances under that model. Providers may use different information, so their xG totals are not interchangeable.

A player's goals minus xG over a short period is not a clean estimate of permanent finishing skill. Chance models are imperfect, individual outcomes are variable, and penalties can dominate a small sample. If the question concerns open-play scoring, report penalties separately and say whether headers, set plays or rebounds are included.

The same principle applies to passing. A completion percentage mixes execution with difficulty. A conservative defender and an attacking midfielder may face very different decisions. To evaluate progression, add information about location, direction, pressure and the consequences of a failed attempt rather than rewarding completion alone.

Compare roles, not just positions

Two players listed as full-backs may perform different jobs. One holds width and crosses; another moves into midfield during possession. Tackles per match can reflect defensive involvement, but a player whose team rarely defends has fewer opportunities to tackle. A high defensive-action count may signal activity, exposure or both.

Possession-adjusted measures try to address this opportunity problem. They can be useful, but adjustment changes the question and depends on a formula. State whether an action rate is per match, per opponent possession or scaled to a reference possession level. Do not place different denominators in one ranking without a clear explanation.

Similar care is needed outside football. In NBA comparisons, per-possession scoring controls for pace better than raw points per game, but role and usage still matter. In MLB, a batting average and an on-base percentage describe different events. Normalization starts the comparison; sport-specific meaning completes it.

Use more than one view

A compact player profile should combine totals, rates, opportunity and a short role description. Add the competition, dates and provider definitions. If you include a percentile, name the comparison group: a high rank among all players may mean little for a specialist role. Changing the cohort can change the percentile without changing the player.

Video and event data should challenge each other. A striking clip is not a representative sample, while an aggregate can conceal repeated tactical details. Review sequences selected by a stated rule, not only successful actions. The aim is not to replace judgment with one metric, but to make judgment inspectable.

End with a claim proportional to the evidence. “Higher observed non-penalty scoring rate in this sample” is defensible when definitions match. “Clearly the better player” usually requires a broader case.