Why it asks your age and sex, and what that buys.

A percentile is a statement about a comparison group. Without one it means nothing at all — and the same raw score genuinely lands in a different place for a 29-year-old man and a 50-year-old woman.

What the comparison group is

Your scores are read against John Johnson’s norms for the IPIP-NEO, banded by sex and by age. The 300-item version has two age bands per sex: under 21, and 21 and over. Raw sums become T-scores against that band’s mean and spread, and T-scores become percentiles through Johnson’s own cubic transform.

Several facets — Morality, Altruism, Self-efficacy, Dutifulness — pile up near the top of the scale, and a normal curve gets their tails badly wrong. Those tails are exactly where a report claims something is unusual, so the wrong transform there would manufacture distinctiveness.

What the sample actually is

A large internet sample, self-selected, administered in English, largely from the United States. Not census-representative of anywhere. It skews toward people who chose to take a long personality questionnaire on the internet, which is a real and unmeasured selection effect.

If you are not a US native English speaker

Trust the shape of your profile over the level. Structure replicates well in literate, English-fluent samples and degrades in others, and the degradation is driven by administration and respondent characteristics rather than by nationality as such. Your report shows both: absolute percentiles, and each facet relative to your own average. Where they disagree, the within-person column is the more trustworthy one.

That is why intake asks whether you are in the US and whether English is your first language. They control whether you are shown this caveat more prominently. They never touch a score. They are sent to the writer, which needs them to know when to say that a number is less trustworthy for you than the shape around it. Two yes-or-no questions rather than a country and a language, because that is the only distinction anything here draws.

The correlation matrix

Some findings come from how facets relate to one another in general rather than from any single score. Sympathy and Friendliness normally move together, so a profile with one high and the other low is saying something neither score says on its own.

That required a 30 × 30 correlation matrix, which exists in no package. It was computed for this engine from Johnson’s IPIP-NEO data repository over 306,835 administrations of the 300-item instrument on 2026-08-03.

Two details that would have silently corrupted it. That file has reverse-worded items already flipped at the time of administration, so flipping them again would have inverted half the instrument. And the correlations are computed on norm-referenced scores within each respondent’s own sex and age band — pooling raw scores would have invented a correlation between Neuroticism and Agreeableness purely out of the fact that women score higher on both.

Source: Johnson's IPIP-NEO data repository (osf.io/tbmh5), IPIP300.dat.

What this cannot tell you

Self-report is the only evidence here. You are the best available judge of your own internal states — anxiety, mood — and a worse judge of the traits that carry social evaluation. Nobody rated you but you.

And the effects are modest. They are real, and they are population-level and probabilistic. The validity figures you may have seen quoted were corrected downward when the corrections behind them were re-examined — Conscientiousness for job performance from .31 to .19, general mental ability from .51 to .31. None of it predicts an individual, including you.