How common is each type?

There is a straightforward answer to this question, and it is that nobody has one. Two large published samples of the same instrument disagree about which type is rarest, which is most common, and which side of two of the four dimensions holds the majority. Both are shown below with their samples attached, because a percentage without a sample is not a fact.

The two datasets

The figures most often repeated around the web trace back to a United States sample of about three thousand adults. A second, larger table drawn from twenty-three countries ranks the types differently at both ends. Neither table is a mistake — they are answers to different questions, and a single number cannot be both.

All sixteen types, rarest to most common

Both series share one horizontal scale, so bar lengths are comparable across rows. The mark in the last column flags the types where the two datasets differ by more than their own sampling error — that is, where the disagreement is about who was counted rather than luck.

Sorted by the average of the two datasets. The "differs beyond sampling error" note marks the 11 of 16 types where the two tables differ by more than their own confidence intervals allow — a difference that comes from who was counted rather than from luck.

The four groups: same population, different cuts

The totals do not land on exactly 100 because the published percentages are themselves rounded; the real sum is shown rather than smoothed away.

Even the simplest summary moves

Sides are rounded to whole percentages and the second side is derived from the first, so each pair sums to exactly 100.

Why a single percentage cannot be trusted

In the 1998 US sample, 1.5% of 3,009 people is about 45 individuals. Counting from that few, the 95% interval around the figure runs from roughly 1.1% to 1.9%. The same type sits at 2.3% in the larger table, whose own interval is about 2.1% to 2.5%. Those two ranges do not touch, and 11 of the 16 types show the same pattern: the disagreement is larger than the error bars allow, so it is a difference in the samples rather than in the arithmetic.

A larger sample would not fix this. Sample size controls random error and does nothing about bias. A very large self-selected sample — the kind a free online test collects — measures its own tilt precisely, and pooling 23 countries into one figure produces an average that no individual country matches.

What this site claims. Every figure on this page belongs to someone else, and none of them comes from a function test: both tables are from a four-letter preference instrument, while this site scores eight cognitive functions, which is a different measurement. We do not publish a distribution of our own, and the reason is not modesty — the test runs entirely in your browser, so there is no aggregate data to publish.

Sources: MBTI® Manual (1998), Form M normative sample, n = 3,009 US adults; and MBTI® Manual, 4th edition (2018), p. 165, n = 16,773 pooled across 23 countries. MBTI® and Myers-Briggs Type Indicator® are registered trademarks of The Myers-Briggs Company. This site is not affiliated with, endorsed by, or associated with The Myers-Briggs Company. The published figures are reproduced for attribution and comparison; the confidence intervals are our own calculation, assume simple random sampling, and are therefore narrower than the real uncertainty.

Keep reading

• Our methodology and findings
• How this compares to other approaches
• All sixteen type stacks
• The eight cognitive functions
• About this project
• Privacy and data handling
• Disclaimer and trademark notice