The average people carry in their heads

Ask someone to guess the average, unprompted, and the figure they name tends to sit above what clinician-measured studies actually report. This is not one bias but several, pointing the same direction, and it is worth separating them because each one is a different kind of evidence problem rather than a single mistaken belief.

The sample of numbers people have actually seen

Nobody arrives at a guess about "average" from a random sample of measured men. They arrive at it from whatever numbers they have been exposed to - conversations, online claims, and self-reported figures that, as a body of evidence, run consistently higher than clinician-measured ones. If the numbers feeding someone's mental average are themselves inflated, the average they carry will be too, independent of anything about actual anatomy.

The sample of bodies people have actually seen

Pornographic and other selected media do not sample randomly from the population either; performers are cast, in part, on visible traits, which makes any visual impression drawn from that media a selected sample rather than a representative one. Separately, the ordinary experience of seeing other men is itself geometrically skewed: another man is typically seen from a distance and at an angle, while a man's view of himself is from above and foreshortened, a well known reason self-assessment in the mirror runs differently than a bone-pressed reading taken with a ruler pressed flat against the pubic bone. Neither of these is a measurement in any sense; both are visual impressions built from an unrepresentative vantage point, and impressions compound rather than cancel out.

Conversation itself is a selected sample

What gets discussed out loud skews the same way as what gets shown. A figure notably above average is more likely to come up in conversation than an unremarkable one, for the same reason a striking outcome gets talked about more than an ordinary one in almost any domain - the ordinary case rarely feels worth mentioning. Over enough conversations, the numbers a person has actually heard named skew toward the striking end even if no single person in any of those conversations was lying, which is a distinct mechanism from the self-report literature's finding that individuals round their own figure upward - this is about which figures get repeated, not about how any one figure gets generated.

The direction, not a number

Each of these mechanisms pushes the same way: toward a perceived average higher than what a clinician-measured study reports. That consistency across independent sources of bias is worth noting even without pinning down how much each one contributes - the pooled literature does not offer a breakdown by mechanism, and inventing a split between them would be exactly the kind of unsupported precision this site avoids. What can be said with more confidence is that the gap is structural rather than incidental: it would take a genuinely unbiased sample of numbers, bodies and conversation to produce an unbiased perception, and none of the three inputs described above is unbiased on its own. That is also why the gap persists even among people who have read that clinician-measured figures run lower - knowing the direction of the bias intellectually does not remove the lifetime of skewed numbers, images and conversations that built the intuition in the first place.

What this is not about

This is a claim about where the popular sense of "average" comes from, not about why self-reported figures specifically run high - that mechanism is covered on its own, and this post does not repeat it. It is also not a history of any particular round number in circulation, which is a separate and more specific question.

If what you want is your own number rather than a sense of where the average sits, the clinical measurement method removes every one of these biases at once, because it replaces an impression with a reading taken by a fixed method against a fixed landmark. A subjective visual rating inherits the same viewing-angle and selection problems rather than escaping them: Rate Cock, a numeric score built from a photo, a human judge's opinion, and an AI model reading an image are all still working from a selected, angle-dependent view, which is one more reason a rating and a measurement answer different questions.