CHALLENGES
Located in Rehearsal Room 2 of The Egg
11:30am -1:15pm Monday through Friday
"Per 90" means a football statistic expressed as a rate per 90 minutes played: the player's total is divided by the minutes he has played and multiplied by 90. It lets analysts compare players with very different amounts of playing time on the same scale. Player statistics on RubiScore list minutes alongside totals, which is the starting point for every per-90 calculation.
This explainer covers how per-90 numbers are calculated, why they replaced per-match averages in most analysis, which situations make them misleading and how to read them with care. Player and match data are available at https://rubiscore.com for anyone who wants to work through the examples on real players.
The formula is simple: take the total, divide it by minutes played and multiply by 90.
A forward with 6 goals in 1,350 minutes scores 6 ÷ 1,350 × 90 = 0.40 goals per 90. A midfielder with 45 progressive passes in 2,025 minutes completes 45 ÷ 2,025 × 90 = 2.0 progressive passes per 90. The same method works for any counting statistic, from shots and tackles to expected goals (xG) and expected assists (xA).
The result is best read as "the rate at which this player produces this action for each full match's worth of playing time". It does not mean the player produced that number in a typical match, and it does not promise the same output over the next 90 minutes.
Per-match averages divide by appearances, and appearances are uneven units. A player who starts and plays 90 minutes, a player substituted at half-time and a player who comes on for the last ten minutes all record one appearance. Dividing by appearances therefore penalises substitutes and players who are regularly withdrawn, while flattering those who play every minute.
Per-90 rates remove that distortion by dividing by the actual time on the pitch. Two players with the same per-90 figure have produced the action at the same rate, regardless of how their minutes were distributed across matches.
Other sports use the same idea with their own time units. Basketball analysts often quote statistics per 36 minutes, and ice hockey uses rates per 60 minutes. Football settled on 90 because it matches the length of a regulation match.
Minutes sound unambiguous, but data providers make different choices at the edges:
These choices rarely change a conclusion, but they explain why the same player can show slightly different per-90 numbers on different sites.
Per-90 numbers are rates, and rates calculated from small samples swing wildly. A substitute who scores once in 45 minutes has a rate of 2.00 goals per 90, a figure no regular starter could sustain. One more blank appearance would cut it sharply.
For this reason, analysts usually apply a minimum-minutes threshold before ranking players. Public analysis often uses cut-offs such as 900 minutes, roughly ten full matches, though the right level depends on the statistic. Frequent actions such as passes stabilise quickly, while rare events such as goals or assists need far more minutes before a rate becomes informative. Expected goals and shots stabilise faster than goals, which is one reason analysts lean on them when samples are small.
Per-90 numbers assume that output scales evenly with time. In reality, minutes are not all equal. Substitutes often enter matches when opponents are tired, when the game has opened up or when their team is chasing a result and committing players forward. Each of those conditions can lift attacking output above what the same player would produce over a full match.
The opposite applies to starters who are routinely withdrawn early. Their minutes come mostly from the opening hour, when matches are often tighter. A per-90 comparison between an impact substitute and a regular starter therefore mixes two different contexts, and the substitute's rate usually overstates what he would produce if promoted to the starting line-up.
A rate per 90 minutes treats every minute as equal opportunity, but opportunity depends on the team. A defender in a side that has little of the ball faces far more opposition attacks, so he has more chances to make tackles and interceptions. His per-90 defensive numbers will usually exceed those of an equally good defender in a dominant side.
The common correction is possession adjustment, which scales defensive actions by how much of the ball the opponent had. Attacking numbers have the mirror problem: a forward in a high-possession side receives more service. Per-90 rates are a first normalisation, not the last one.
Some statistics translate cleanly into rates, and others need care:
Per 90 is not the only way to normalise. Some analysts express actions per touch, which measures how efficiently a player uses the ball he receives, or per team possession, which measures how much of his team's play runs through him. Each answers a different question. A playmaker in a dominant side may have modest per-touch numbers because he takes many safe touches, while a direct winger in a counter-attacking team may look efficient per touch but rarely see the ball.
Per 90 remains the default because minutes are recorded consistently and the unit is easy to understand. The alternatives are best used as a second view when a per-90 figure seems surprising, since they often reveal whether the surprise comes from the player or from the volume of ball his team gives him.
A short routine avoids most errors. First, check minutes before rates. On RubiScore, the minutes column beside a player's totals tells you immediately whether a per-90 figure rests on a meaningful sample. Second, ask how the minutes were earned: as a starter, as a substitute or as a mix. Third, consider team context, especially for defensive actions. Fourth, compare like with like, meaning players in similar roles, competitions and team styles. Finally, prefer stable underlying measures such as xG or shots over rare outcomes such as goals when samples are short.
Imagine two wingers in the same league halfway through a season. Winger A has 5 goals and 3 assists in 1,620 minutes, almost all as a starter. Winger B has 4 goals and 2 assists in 720 minutes, mostly from the bench.
Per 90, Winger A produces (5 + 3) ÷ 1,620 × 90 ≈ 0.44 goal involvements, while Winger B produces (4 + 2) ÷ 720 × 90 = 0.75. On the raw rate, Winger B looks far more productive.
Three checks temper that impression. Winger B's 720 minutes are below the threshold many analysts would use for goal involvements. Most of his minutes came against tiring opponents, often with his team chasing a result. And if his xG and xA per 90 sit close to Winger A's, much of the gap probably comes from finishing variance in a small sample. The fair conclusion is that Winger B deserves more minutes to test the rate, not that he is already the better player.
Per-90 statistics are the common language of modern player analysis because they solve the problem of unequal playing time. Used with a sample threshold, an eye on how the minutes were earned and an awareness of team context, they are one of the most useful tools for comparing players fairly.