Data · 2026-07-06
The share of people in your type is a range, not a number
Ask what percentage of people share your type and you will get a confident answer, usually with a decimal point. That answer is a point estimate from a sample somebody else recruited. Two of the largest published tables for the same instrument disagree about which type is rarest, and for eleven of the sixteen types the gap between them is wider than their own sampling error allows. The disagreement is not noise. It is a measurement of who was counted.
Every typology resource, including the ones that are careful about everything else, will give you a percentage for your type. It usually arrives with one decimal place. That precision implies a measurement, and the measurement is real — somebody counted people and divided. The difficulty is entirely in the denominator.
The table most people have seen comes from a United States sample of 3,009 adults, published in the 1998 manual for a four-letter preference instrument. It is a good sample for its purpose: recruited to be representative, large enough to detect a preference split, and still the reference point for most of what circulates today. It puts INFJ lowest at 1.5% and ISFJ highest at 13.8%.
A later edition of the same manual publishes a second table, drawn from 16,773 people pooled across 23 countries. It puts ENTJ lowest at 1.8% and ISTJ highest at 15.9%. Same instrument, same scoring, a different answer at both ends of the ranking. The four rarest types are the same four in both tables — INFJ, ENTJ, INTJ and ENFJ — but their order, and every number attached to them, moves.
It is tempting to read this as a sample-size problem and simply use the bigger table. It is not one. Sample size controls random error, and this disagreement is not random. For a type near 1.5% in a sample of 3,009, the 95% interval runs from roughly 1.1% to 1.9%, and the count behind it is about 45 people. The same type sits at 2.3% in the larger table, with an interval of about 2.1% to 2.5%. Those two ranges do not touch. Eleven of the sixteen types show the same pattern. When two samples disagree by more than their own uncertainty allows, the remaining explanation is that they contain different people.
They do contain different people, and predictably so. A representative sample recruits adults who were not thinking about typology at all; a volunteer sample recruits people who went looking. That difference is not small, and scale does not correct it. A very large self-selected sample does not approximate the public — it measures its own tilt with great precision. The global table also pools 23 countries into a single figure, which makes it an average across populations that plausibly differ from one another, and therefore a figure that matches none of them exactly.
None of this is a criticism of either table. Both are honest attempts to describe a population, and both are more transparent than the numbers that circulate with no sample attached at all. It is a criticism of the single number, quoted to one decimal place, with nothing said about where it came from. When you read that your type is 2% of the population, the useful question is not whether the figure is accurate but which population, recruited how, and in what year. Often the honest answer is that nobody knows, because the sample was never built to generalise that far.
It is also why this site does not publish a distribution of its own. We score eight cognitive functions rather than four preferences, which is a different measurement and would not reproduce these percentages even with a perfect sample. And because the test runs entirely in your browser, we hold no aggregate data at all — so a published share would have to be invented. What we can do instead is show the range, name where every figure came from, and let the uncertainty be part of the answer rather than something removed for the sake of a cleaner table.
Key points
- The two tables disagree at both ends: one puts INFJ last at 1.5%, the other puts ENTJ last at 1.8%.
- The rarest type in a 3,009-person sample is about 45 people — too few individuals to carry a one-decimal percentage.
- For 11 of the 16 types the two tables differ by more than their confidence intervals allow, so the difference lives in the samples rather than in the luck of the draw.
- Two of the four dichotomies change which side holds the majority between the two tables, which makes both "most people are F" and "most people are T" defensible.
- A bigger sample removes random error and does nothing about bias: a very large self-selected sample measures its own tilt very precisely.