What are the limitations of per-90 football statistics?

Per-90 statistics normalize output for minutes played, but they do not normalize the football environment around those minutes. Sample size, tactical role, team possession, game state, opposition, competition strength, set pieces, and provider definitions can all make two identical per-90 values mean different things.

The formula solves one problem: unequal playing time. It should not be asked to solve every other comparison problem.

1. Small samples produce unstable rates

A single event can transform a low-minute rate. That is mathematically valid and analytically fragile. The issue is not that per-90 arithmetic fails; it is that the observed minutes may not represent the player's normal role or future usage.

Always display minutes and appearances beside the rate. Tifolab uses sample context and keeps very thin profiles out of the public search corpus, but no universal minutes threshold turns a noisy rate into certainty.

2. Substitute minutes are not neutral

Substitutes often enter matches with a specific instruction, scoreline, opponent condition, or tactical shape. A winger introduced while chasing a goal may receive more attacking opportunities than the same player protecting a lead. A forward facing tired defenders may record a high dribble or shot rate that would not scale linearly to 90 minutes.

Per 90 describes the observed substitute output. It does not simulate a start.

3. Possession and opportunity differ

Standard per-90 rates do not adjust for possession. A defender on a team without the ball may face more duels, interceptions, clearances, and blocks. A midfielder on a possession-dominant side may make more passes because the team creates more passing opportunities.

The FIFA data analysis glossary also distinguishes 90-minute normalization from ball-in-play normalization. FIFA notes that the ball is typically in play for materially less than the full match, so two 90-minute performances can contain different opportunity windows.

4. Roles create different event expectations

Position labels are broad. Two central midfielders can be asked to perform opposite jobs: one protects transitions, while the other receives between the lines. Comparing their shot, pass, or defensive-action rates without role context rewards the player whose instructions generate more of the selected event.

This is why Tifolab pairs position cohorts with role presets. Even then, a preset is a lens rather than a tactical truth.

5. Match state changes behaviour

Teams behave differently when leading, drawing, or trailing. A leading team may defend deeper and concede possession. A trailing team may increase shot and crossing volume. Aggregated per-90 values hide when each event occurred.

Fixture history, scoreline, and opponent context help explain whether the rate reflects a stable trait or a recurring match situation.

6. Cross-league rates are not automatically equal

Ninety minutes remains 90 minutes in every competition, but the pace, tactical norms, team quality, schedule, and data coverage can differ. A common time denominator does not produce a league-strength adjustment.

Use cross-league per-90 values to screen for candidates, then test the profile against league context and video. The FIFA Talent Identification Guide treats data analytics as one part of a broader identification and selection system, not as a replacement for observation.

7. Event definitions can differ

Providers may classify key passes, assists, duels, pressures, or progressive actions differently. A comparison becomes unreliable if the underlying event definitions change between players or seasons.

Prefer a single provider and a consistent transformation pipeline inside one comparison. If sources must be mixed, disclose the boundary rather than treating the numbers as interchangeable.

A safer reading order

Read football statistics in this order:

1. Confirm the event definition and scope.
2. Check minutes, starts, substitute usage, and appearances.
3. Read the raw total and per-90 rate together.
4. Add position and role-aware percentile context.
5. Review league, possession, match state, and opposition.
6. Validate important conclusions with match observation.

Use Explore to screen profiles, then open Compare only after aligning the scope. For the calculation itself, see how per-90 statistics are calculated.