Myers-Briggs Type Indicator

What the Myers-Briggs Type Indicator is

The Myers-Briggs Type Indicator, usually shortened to MBTI, is a questionnaire that people complete about themselves. It sorts them on four pairs of opposites and gives each person a type made of four letters.

It is used in training and team building, and also for purposes beyond what it can support. What matters about it is what it measures, how it builds its result, and above all the question every decision based on it depends on: which conclusion can it support?

The four pairs

• Extraversion versus introversion: where energy comes from, being with people or being alone.
• Sensing versus intuition: whether attention goes to concrete detail or to the overall pattern and possibility.
• Thinking versus feeling: what settles a decision, logical consistency or its effect on people.
• Judging versus perceiving: whether closure and order are preferred, or keeping options open.

Two multiplied by itself four times gives sixteen, which is the number of possible types.

These pairs are presented independently of one another: nothing in the model says a leaning on one goes with a leaning on another. Anyone who reads a type of four letters as a single coherent description adds to the tool something it does not contain. It is four separate answers set side by side.

The first issue: a result in categories, built from data in amounts

This is the most important point here, and it concerns the tool’s structure rather than its results.

Responses to the statements are graded, and each axis yields an amount: how far the person leans towards one pole. Those amounts are then cut at a dividing point to give the letter. Someone one step above the cut gets one letter; someone one step below it gets the opposite letter.

Two consequences follow:

• Two people next to each other on the score are shown as different types, while two people far apart on the same side are shown as the same. The type hides the distance it was built on.
• Someone whose score is close to the cut may see their letter change with a small change in their circumstances or their day. That is not a fault in how they answered; it follows from cutting a continuous scale. It is the origin of the type changes people notice when they retake the questionnaire after a while.

What is required of a result’s stability is the subject of test reliability. What needs adding is that instability can sit in the letter while the score underneath is completely stable, so presenting results as letters lowers apparent reliability through arithmetic, not through weak measurement. We found no source, among those we reviewed, that estimates the share of people whose type changes on retesting, so we give no figure for it, and for the same reason we give none for how common each type is.

A worked example: how much a letter hides

Put numbers on it. Suppose one axis is scored from 0 to 100, with the cut at 50. The figures are assumed, to show the structure.

• Two employees scoring 49 and 51, 2 points apart, come out with different letters.
• Two employees scoring 51 and 95, 44 points apart, come out with the same letter.

In this example the letter separated two people 2 points apart and grouped two people 44 points apart. Anyone reading the two types as describing a difference in behaviour has read the exact reverse in both cases.

Now extend that to all four axes. An employee close to the cut on two axes out of four could plausibly be any of four types, two multiplied by itself twice, not one. Close on three axes, the possibilities reach eight, half of all sixteen types. In both cases what they are handed is a sheet showing one type, with nothing to show where their score sits relative to the cut.

That is why the most useful change in how this result is presented has nothing to do with the tool itself: show the score beside the letter. Someone who sees that their leaning on an axis is slight stops seeing themselves as a category, and what gets built on the result can fall away with it.

The second issue: which decision it can support

The principle behind test validity is decisive here: validity belongs to the inference, not to the sheet of paper. No test is valid in the absolute, and the question is always which decision it is valid for. So the claim that this indicator is valid, or invalid, cannot be answered as it stands.

The same principle asks for the inference to be written as a full sentence naming the score, the threshold, the person and the decision. Try it: “people of this type manage teams better than people of that type”. That sentence makes the reader ask where it came from, and what in the questionnaire corresponds to managing a team. We found no evidence, in the sources we reviewed, supporting that inference or others like it in selection or promotion decisions. That is not a finding that the tool is unfit for such decisions, nor that it is fit for them. It is a finding that the evidence any inference of this kind needs is absent from our sources, and that anyone who wants to use the tool in a decision has to bring that evidence before the decision, not after.

There is also a practical limit independent of the evidence: the tool is answered by the person it describes. In selection the candidate knows what is wanted and answers accordingly. That differs from a development setting, where nothing rides on the answer. The same tool produces different data depending on what rides on the answers; what changed is not the tool but the position of the person answering.

So what is stated firmly here comes to two things: the result is presented as a category and built from an amount, and the person answering has a stake in the result whenever anything rides on it.

The third issue: descriptions that fit both poles

The result comes with a description of the type, and its persuasiveness can come from the wording. A line such as “you enjoy people and still need time on your own” describes both poles at once, so every reader finds themselves in it.

This can be checked with a simple experiment: show a group the description of a type that is not theirs without telling them, then ask how well it fits. If acceptance comes close to their acceptance of their own type’s description, what convinced them was the wording, not the accuracy of the classification. Any organisation can run this on its own people before building anything on the tool.

Where it remains useful

None of this means it should be banned. It has a place, on one condition that is stated openly:

• A shared vocabulary for talking about ways of working. Someone saying they need to see the detail before they decide is useful information for the people who work with them, whether it came from this questionnaire or from a direct conversation.
• An opening for a team conversation about how decisions get made in it, who needs time before a meeting and who thinks during it.

The condition is to announce, before it is handed out, that the result will not enter any appraisal, promotion or assignment of a role. Once it does, the questionnaire becomes an exam answered to please, and the benefit it was used for is gone.

Uses that do harm

• Building teams on types. It is said that a team needs a balance of types, so roles are assigned by letter. The objection comes before any question of evidence: a type is a preference, not an ability, and assigning a role on that basis assigns it on something other than what the role is performed with. If the quality needed is known, the place to start is analysing the job and then gathering evidence for that quality, not four letters.
• Explaining disagreement by type. When a dispute between two people is put down to different types, the search for its real cause stops: an unclear standard, information that did not arrive, two interests in conflict. Explaining by type is comfortable because it commits nobody to anything, which is why it is the worst use of all.
• Keeping it in the employee file. Once the result stays on record, it gets consulted in later decisions it was never gathered for, even if nobody intends that. And since the result is unstable when shown as letters, the file keeps a description that may no longer describe its owner two years later.

What it is not

What places it among neighbouring instruments is what its result is made of, and its result is a reported preference.

• A cognitive ability test has right and wrong answers. The MBTI has neither, only a preference its owner reports, so comparing the two compares instruments of different kinds.
• A work sample test rests on performance that is observed; the MBTI rests on what people say about themselves.
• Integrity testing has another purpose entirely and belongs to a separate family.
• A preference is not a skill, so the result describes neither ability nor performance: someone who prefers solitude may run a meeting better than someone who prefers company.
• Nor is it a fixed classification of its owner, for the reason given under the first issue.

Related terms

• Test Validity
• Test Reliability
• Cognitive Ability Test
• Work Sample Test
• Situational Judgement Test