Systematic and random error

Taking more readings fixes one kind of error and does nothing at all for the other. Knowing which kind you have decides whether repeating the measurement is worth your time.

Two kinds of error

Random error varies from reading to reading in no particular direction: a small wobble in ruler angle, a slightly different grip each time, a tape that sat a fraction differently around the shaft. Take several readings and the wobbles tend to cancel, because they are as likely to push the number up as down.

Systematic error pushes the same way every time, for a reason built into the method or the instrument. Take a hundred readings with a systematic error present and the average is still wrong by roughly the same amount, because every one of the hundred carries the same bias.

Examples from a real measurement

The fat pad above the pubic bone is the clearest systematic error in this subject. If pressure at the base is consistently too light, every reading comes in short by close to the same amount, because the same tissue is compressing the same way each time - repeating the reading does not press any harder. A stretched cloth tape is the girth equivalent: if the tape has lost its factory tension, it reads slightly long on every wrap, not just some of them.

Reading wobble is the random counterpart. Small differences in exactly where your eye lines up with the ruler's scale, or a slight shift in how the tape sits on one wrap versus the next, push the number in either direction largely by chance. A single stray reading caused by a moment of inattention also falls in this category, since nothing about it is built to recur the same way on the next attempt. Where measurement error comes from covers the fuller list of sources; this is about which category each one falls into and what that implies for fixing it.

Why repeats only fix one of them

Averaging - or taking a median of several readings - is a genuine fix for random error, because the errors above and below the true value cancel each other out as the count grows. It is not a fix for systematic error, because there is nothing to cancel: every reading is biased the same direction, so the average of ten biased readings is just as biased as one. This is why a tight, repeatable set of readings is not by itself proof of accuracy - see precision versus accuracy for the fuller version of that distinction. A systematic error needs to be found and corrected in the method itself: press harder, check the tape for stretch, confirm the landmark.

What this means for your own numbers

Most of the error in a careful self-measurement is systematic rather than random, and it mostly runs in one direction: under-pressing at the bone, and reading the underside instead of the dorsal line, both tend to move a figure the same way every time rather than scattering it. That is worth knowing before you reach for more readings as the fix, because more readings of a biased method just gives you a more confident wrong number - an honest uncertainty statement describes the random part of your error well and says nothing at all about a systematic one lurking underneath it. The fix for systematic error is technique review, not repetition: check the landmark, check the pressure, check the instrument for stretch, the way Rate Cock is a check on a different kind of number entirely, not a check on this one.

Where this idea does not reach

A photograph carries no repeatable readings at all, so the random-versus-systematic frame does not apply to what an image model outputs - it is inferring a figure from visual cues, not taking one, and there is no set of repeats to average. A rating from Penis Rater is not trying to converge on a true centimetre figure either, so the same distinction does not transfer there. And a person's impression, the kind a judge on Rate Penis gives, carries its own biases, but they are the biases of a person forming an opinion, not the systematic error of a ruler pressed too lightly against bone - a different problem, with a different fix again.