Data
A centimetre near the middle
Because the middle of the distribution is crowded, a centimetre of measurement error near the mean moves a percentile far more than the same centimetre in the tails.
Guides on Data: How a size study is built, Reading a nomogram without fooling yourself, The Veale meta-analysis, read properly
A centimetre of measurement error does not cost the same number of percentile points everywhere on the curve. Near the mean it can swing your position by twenty points or more; out in the tails, the same centimetre barely moves the needle. The reason is where men are crowded on the curve, not anything about the ruler.
The shape that causes it
A normal distribution's cumulative curve - the one a nomogram draws - is steepest at the middle and flattens toward both ends. That steepness is the density of the population: near the mean, a huge share of men are packed into a narrow band of measurements, so a small step in centimetres crosses many of them at once. Out in the tails, the same step in centimetres crosses very few men, because very few men are out there to cross.
Roughly how much, near the mean
Using the Veale erect-length figures - a mean of 13.12 cm and an SD of 1.66 cm - a centimetre near the mean corresponds to a z-score step of roughly 0.6. Those two figures come from Veale et al. (2015), BJU International, where the erect-length row rests on 692 men measured by health professionals. Working from the z-score table, moving from the mean (the 50th percentile) up by about 0.6 SD lands around the 73rd percentile - a jump of over 20 percentile points from a single centimetre. Move the same centimetre starting two SDs out, near the 98th percentile, and the shift is a couple of points at most, because there is so little of the population left to cross.
Why the slope, not the distance, is what matters
It is tempting to think a centimetre is a fixed unit and should therefore mean a fixed amount everywhere, the way a centimetre of height means roughly the same thing wherever you measure it against a tape. On a percentile curve that intuition fails, because the curve is not measuring a distance in isolation - it is measuring how much population lies between two points, and that quantity depends on the local steepness of the curve, not on the centimetre gap alone. The same logic applies to any measurement summarised by a mean and an SD: the crowding is always worst near the centre of any roughly bell-shaped population, regardless of what is being measured or in what units.
What this means for your ± 0.3 cm
Careful self-measurement carries roughly ±0.3 cm of error on length, which sounds small next to a mean in the double digits. Near the mean that margin alone can move you several percentile points in either direction, and trained hands are not immune: in Habous et al. (2015), Journal of Sexual Medicine, how far seven andrology specialists' stretched readings fell short of erect length ranged from 16% to 27% depending on the observer. Given that, a figure quoted as a single percentile - "you are at the 65th" - is overstating the precision the underlying measurement can actually support. The honest version states a percentile range wide enough to cover the measurement uncertainty, not a single point read off a curve as though the input were exact.
Why this matters more here than it would elsewhere
Most men's figures sit somewhere in the crowded middle of the distribution, which is exactly where this effect bites hardest - a man near the tails, where headline claims tend to cluster, is the person least affected by a centimetre of error, and that part of the curve is thinner than intuition suggests anyway. The practical upshot: treat a percentile derived from a self-measurement as a band, not a point, and widen the band near the middle of the curve more than you would near either end.
A percentile, not a verdict
None of this arithmetic produces a judgement about anyone - it produces a range on a curve built from measured centimetres, and a range is the honest output. That is a different kind of statement from a score: Rate Cock rates a photograph rather than placing a measurement on this distribution, an AI model reading an image carries no percentile margin of this kind because it isn't measuring a length in the first place, a comparative photo score is a different scale with a different uncertainty, and a human judge's opinion has no centimetre error to propagate at all - each is worth reading for what it is rather than converted into a percentile it was never built to produce.