MBTI vs Big Five: what each one rests on

The question needs splitting, otherwise it has no answer. An instrument can be perfectly reliable, in the sense that its items hang together, without the structure it proposes matching what the data show. The two are assessed separately, and the MBTI does not score the same on both.

On structure, the reference work dates from 1989. McCrae and Costa compared the MBTI with the five-factor model, using self-reports and peer ratings, in 267 men and 201 women aged 19 to 93. Their conclusion is unambiguous: nothing supports the view that the instrument measures truly dichotomous preferences or qualitatively distinct types. It measures four relatively independent dimensions, which overlap with four of the five major dimensions of personality.

A second piece of work defused the main argument in favour of types. In 2002, Bess and Harvey re-examined score distributions across roughly 12,000 participants in leadership development programmes. The two-humped distributions often cited as evidence that types exist turned out to be an artefact of the scoring software's default setting. With a finer setting, the distributions become strongly centre-weighted: most people sit in the middle, not at a pole.

A recent synthesis places the state of the literature. In 2025, in the Journal of Counseling & Development, Erford and colleagues reviewed 193 studies published between 1999 and 2024 on Form M of the instrument. Internal consistency is good there, from 0.845 to 0.921 across subscales and total scores, and convergence with six comparable instruments is described as robust. On the other hand, the authors report that studies of structural validity and test-retest reliability were absent from that twenty-five-year sampling.

A critical review published in 2005 by Pittenger, in a consulting psychology journal, had already gathered the measurement limitations that warrant, in its own words, considerable caution when drawing inferences from a four-letter type formula. Twenty years later, the 2025 finding about missing retest studies does not point towards a settled answer.

A type is a box: you are either in it or out of it. A dimension is a position on a scale: you are higher or lower, and the number can be compared. The consequence is easy to illustrate. Two people who answer almost identically, but on either side of the boundary, get two different letters and two different descriptions. Two people far apart from each other, but on the same side, get the same letter.

That would only be a curiosity if people clustered at the two extremes. Bess and Harvey's work shows the opposite: once the scoring artefact is corrected, scores concentrate around the middle. In other words, the boundary falls exactly where the most people are, and therefore where a small difference in answers is enough to change the label.

The difference has a very concrete practical consequence. With dimensions, you can retake the questionnaire two years later and look at whether a score moved, by how much, and in which direction. With a type, you have only two states: the same label, or another one. A real but moderate change becomes invisible, and a tiny wobble around the boundary looks like a change of personality.

Because it does several things well that dimensional models do badly, and it would be dishonest not to say so. A type is memorable, sayable out loud, shareable. It gives a team a common vocabulary in half a day. Above all, none of its poles is presented as the bad side, which makes the feedback far less uncomfortable than a score sitting below a group average.

• A result you remember, unlike five numbers forgotten on the way out of the room.
• No pole presented as a flaw, which makes group discussion possible.
• A shared vocabulary that allows talk about differences in working style without talking about the people.
• A short piece of feedback, readable without any prior training in measurement.

One mechanism should be added, without reading it as a criticism of anyone. Since scores concentrate around the middle, a description written for one pole in fact describes a great many people sitting near the centre. Recognising yourself in it is therefore the most likely outcome, and that recognition is experienced, legitimately, as evidence of accuracy. It is a property of the score distribution, not a lapse of judgement on the reader's part.

The answer depends entirely on what you want to do with it, and both uses are legitimate. To open a conversation, set a shared vocabulary in a team or simply enjoy comparing notes, a type-based format does the job well. To place yourself relative to a reference group, follow a change or combine your result with something else, you need continuous scores.

• Do you want a conversation or a measurement? The first is happy with a label, the second needs a number.
• Do you plan to retake the questionnaire later? Only continuous scores let you see a moderate shift.
• Will the result be used to decide something about a person? In that case neither format is enough, and the 2005 review explicitly warns against that kind of inference.

On stability, the best available reference is not about one instrument but about traits themselves. Roberts and DelVecchio's meta-analysis, published in 2000, pooled 3,217 correlations from 152 longitudinal studies. Rank-order consistency rises steadily: about 0.54 during the college years, 0.64 around age 30, then a plateau near 0.74 between 50 and 70, with the interval held constant at 6.7 years. Personality moves, slowly, and you need numbers to see it.

That work also keeps being revised. The inventory published by Soto and John in 2017 organises the five dimensions into fifteen facets, gaining detail without losing the brevity of the format. That is the sign of a living framework, not a fixed catalogue.

Brain Quiz publishes a self-observation questionnaire based on the five dimensions, reachable from the block at the end of this page. It does not reproduce the MBTI, which is a registered trademark, and offers no type-based equivalent: the reason is the one set out above, a fixed label cannot be deduced from a score sitting near the middle. No personality questionnaire, ours included, replaces professional advice.