It depends on the format. On 60,000 real answer sheets, a 20-question yes/no format gave the same person the same letter about 65% of the time per scale; this test’s 12 hand-picked five-point statements per letter reach about 81% (estimated). Because the four letters multiply, even that leaves fewer than half of people with all four letters unchanged.
What “accurate” means for an MBTI test
For a test, “accurate” can only mean that it agrees with itself, because there is no blood test for personality to check your letters against. Give the same person two equivalent sets of statements and see whether they land on the same side of each scale. If a test cannot do that, nothing built on top of it holds. Whether the scales measure something real is a second question, answered further down. MBTI is a registered trademark of The Myers-Briggs Company; this site is not affiliated with it.
How often an MBTI retake gives the same letters
We ran the check on 60,000 real answer sheets from Johnson’s public IPIP-NEO data, whose reference group holds more than 135,000 people. For each format we drew two non-overlapping sets of research statements, 40 times over, and counted how often the same person got the same letter from both sets.
| Format | All four letters the same | Same letter, per scale |
|---|---|---|
| 2 five-point statements per letter (8 questions) | 15% | 62% |
| 5 yes/no questions per letter (20 questions) | 18% | 65% |
| 12 yes/no questions per letter (48 questions) | 28% | 73% |
| 8 five-point statements per letter (32 questions) | 31% | 74% |
| 15 yes/no questions per letter (60 questions) | 32% | 75% |
| 12 five-point statements per letter, random research statements (48 questions) | 38% | 78% |
| 12 hand-picked five-point statements per letter, this test (48 for the four letters) | about 44% (estimated) | about 81% (estimated) |
Two things stand out. A short test and a yes/no format each cost agreement on their own, and the gap between the weakest and strongest format is wide: 15% against about 44% for getting the whole four-letter type back. If you want the length and format question on its own, it has its own post.
Four MBTI letters, four chances to flip
A per-letter figure sounds reassuring until you remember a type has four of them. Each letter is a separate chance to flip, and the chances stack. At 65% per letter, fewer than one person in five keeps all four. Even at about 81%, the best figure here, fewer than half do. No four-letter test can promise you the same type every time, and one that does is not telling you how it was checked.
Six letters, as on this site, multiply once more. Cut every scale hard down the middle and only 28% of people would get all six letters back. That number is the reason this test does not cut hard down the middle.
Why your MBTI type “changes”
Most of the time your personality did not change; the measurement did. Four things usually explain a different result.
You were close on a scale. If your answers put you near the middle of E/I, either letter can come out on a given day. A test that forces a side turns that coin toss into a label. Here, a close call gets a few more statements on that scale, and if it stays close the result says in between. With that rule, about 7 in 10 people would see none of their clear letters flip to the opposite side on a retake, and on any single scale only 4 to 7% see a clear letter flip.
The test was short, or yes/no. Two statements per letter cannot tell a mild lean from a strong one. Yes/no answers throw away the difference between “a bit like me” and “exactly me”.
You answered as a different self. Who you are at work, who you want to be and who you were last week give different letters. Answer for how you usually are.
The tests used different statements. Two sites can both label a scale N/S and ask about quite different things. Compare a result only with a retake of the same test.
An INFP who comes back as an INFJ has most likely moved on one close scale, not become a different person. A retake here tells you which letters moved, never that “your type changed”.
Is the MBTI test valid, or only reliable?
A test can be reliable, giving the same answer every time, and still not measure what it claims to. A bathroom scale that always reads two kilos light is perfectly consistent and always wrong. Getting the same answer twice is reliability, and the figures above measure only that. The other half is validity: whether the four scales describe something real about people.
The best-known check is McCrae and Costa’s 1989 study. They gave the official questionnaire and a Big Five questionnaire, the most studied model of personality, to 267 men and 201 women, and also collected ratings from people who knew them. The four MBTI scales lined up with four of the five Big Five traits: E/I with extraversion, S/N with openness, T/F with agreeableness and J/P with conscientiousness. The fifth trait, how easily you are thrown by stress, has no MBTI scale at all.
What the study did not support is the idea of types as separate boxes. It found no evidence that people split into two kinds on each scale; the scales measure four dimensions on which everyone varies by degree. In short, the letters are valid descriptions of real traits, and the sixteen boxes are a convenient way to talk about them.
This test is built on that reading. Its statements come from the same Big Five research, it adds a letter pair for the missing stress trait, and it reports a close call as in between rather than forcing a box. The method page shows where each scale comes from.
How accurate these figures are
The figures above are a simulation, not a promise, and three things flatter every row.
- Both sets of statements were answered in the same sitting. That makes these upper bounds; a real retake weeks later agrees less.
- Letters were split at the middle of each format’s own results. This is generous to yes/no formats.
- The short formats were built from validated research statements. Many quiz sites write their own questions instead, so the short rows are if anything too kind to them.
The last row of the table is a model estimate rather than a measured figure. The same model reproduced the random-12 row within half a point (78.0% against 78.3% measured), which is why we trust it enough to publish. The full account, including where the statements come from, is on the method page.
What to look for in an MBTI test
- Five answer options, not yes/no.
- About 12 statements per letter. Beyond that, gains are small.
- Statements from a published research pool such as the International Personality Item Pool, started by Lewis R. Goldberg, rather than written for the quiz.
- A result that admits close calls instead of forcing a side.
- A plain account of how the test was checked, with its limits stated.
- No account, email or subscription before you can see your result.
Ready to see how your letters hold up? Take the test, about 10 minutes, no sign-up.