Percentiles

A percentile answers one specific question: what share of a reference sample measured below a given figure. The 50th percentile is the median - half the sample measured below it, half above. The 90th percentile means nine in ten men in that sample measured less than the figure in question, and one in ten measured more. Nothing else is packed into the term, which is exactly where two common misreadings creep in.

Misreading one: percentile as a percentage of the mean

A figure at the 90th percentile is sometimes read as "90 percent of the average", as though the number were a fraction of the mean. It is not. The 90th percentile is very likely well above the mean - by definition, it is a figure that only a tenth of the sample reached or exceeded - and it has no arithmetic relationship to "90 percent of" any other number. Percentile and percentage are different operations that happen to share a word.

Misreading two: percentile as a score

The other common confusion treats a percentile like a grade on a test: higher is simply "better", the way a 90 out of 100 beats a 50 out of 100 on an exam. A percentile is a position in a distribution, not a mark of quality - it says where a measurement sits relative to a specific measured population, and that population is a group of men measured by clinicians for a research study, not a graded standard anyone is being judged against.

What it takes to compute one

Getting from a raw measurement to a percentile needs a reference mean and standard deviation and a short piece of arithmetic - worked through in full with the z-score mapping - which this note deliberately leaves aside to keep the definition clean. It also depends entirely on which population the percentile is measured against, a detail that changes the answer even for an identical figure.

That population dependency is also the whole difference between this kind of number and a rating. A percentile is a fact about a distribution of measured lengths; a score from Rate Cock is a judgement about a photograph, and the two do not convert into each other. An AI system estimating size from an image, a comparative score built for ranking photos, and an individual human judge's opinion are three separate kinds of output, and none of them is a percentile in this sense - each is worth reading on its own terms rather than translated into one.