Percentile Rankings

A rank turns a number into a verdict

Most football stats arrive without a sense of scale. Told that a midfielder completes 2.1 progressive passes per 90, almost nobody can say whether that is impressive, ordinary or poor, because the answer depends entirely on who else is being counted.

A percentile rank supplies the missing half. It says what share of comparable players that figure puts them ahead of. At the 84th percentile, the player is ahead of 84% of the group. At the 12th, they are ahead of 12% of it and behind the rest.

The point is that the raw number never changes and the verdict does. Those same 2.1 progressive passes per 90 can be the 84th percentile among centre backs and the 31st among central midfielders, because the two groups are asked to do different jobs. One reading says a defender who breaks lines unusually often. The other says a midfielder who does not.

The comparison group does most of the work

A percentile is only as meaningful as the pile it ranks against, so it is worth knowing exactly what goes into that pile.

Same position group. Forwards are ranked against forwards, full backs against full backs. Ranking a striker's tackle count against a defensive midfielder's would measure the team sheet, not the player.

Same league and season. Standards and styles differ enough between competitions and between years that mixing them blurs the comparison.

Cleared the minutes gate. Players below a minutes threshold are excluded from the group entirely. Without that floor, one excellent cameo would sit atop every leaderboard and push everybody else down a place.

That is why a percentile panel always states its cohort underneath: how many players are in it, what minimum they had to clear, and which season it covers. A percentile without its cohort attached is not really a statistic, it is a number with no denominator.

Why thin samples get values but no ranks

A per 90 rate built on few minutes is noisy on its own. A rank built on those same minutes is worse, because it compares one thin sample against everybody else's thin samples and compounds the noise across the whole group.

So the two are treated differently. Below the gate the per 90 values still appear, with the minutes played that produced them stated alongside so the figure can be weighed honestly. The rank is withheld until the sample supports it. Withholding it is not a judgement on the player, only on how much the evidence can currently carry.

Reading a percentile panel well

Read across, not down. One rank in isolation says little. A forward in the top decile for shot volume and the bottom quarter for conversion is telling a clear story about a player who gets into good positions often and finishes them poorly, which neither number says alone.

Separate volume from efficiency. Volume metrics, shots, touches in the box, progressive passes, mostly reflect role and instruction, and tend to hold from season to season. Efficiency metrics, conversion rate, duel success, swing far more, so a spike in one is weaker evidence about the future.

Check the role before judging a low rank. Percentiles measure output, not intent. A holding midfielder who never plays through balls may be doing exactly what has been asked. The rank is the beginning of the question, not the answer.

Watch the middle. Percentiles compress where players cluster. Around the median, a very small difference in the underlying number can move a player across many percentile places, so a gap between the 45th and the 55th is usually noise rather than a meaningful separation.

Common misreadings

"The 90th percentile means they scored 90." It means they are ahead of 90% of their comparison group. The underlying figure could be 0.4 or 40.

"A high percentile means a good player." It means a high output on one metric against one group. Players are not one metric, and the panel is designed to be read as a whole.

"Percentiles from two different sites should agree." Only if both use the same position groups, the same minutes threshold and the same underlying model. They rarely do, so treat ranks as comparable within a source rather than between sources.