Four-Letter Type Validity: What the Evidence Actually Says

The criticism you have heard is mostly accurate, and it is usually delivered more harshly than the evidence supports. Three findings are solid: the result is less stable than the theory requires, the sixteen boxes do not appear in the data as separate groups, and the code does not predict how well someone does a job. Those are enough to rule out the uses people most want to put it to, including hiring, team assignment and career advice.

What the harsh version gets wrong is the leap from that to nonsense. The four scales correlate respectably with four of the five dimensions that mainstream personality research has converged on, which means the questionnaire is measuring something, even if the box it reports is a crude summary of it. A 2017 systematic review concluded that construct validity is reasonable and that three of the four subscales have acceptable retest reliability.

So the fair position is unglamorous and specific. Four letters are a decent way to start a conversation about how you work and a bad way to decide anything about anyone, including yourself.

Reliability is consistency. If a bathroom scale reads 70 kilograms, then 76, then 71 in the same minute, it is unreliable, and nothing else about it matters. For a questionnaire, the usual test is to give it to the same people twice, weeks apart, and see how closely the two sets of answers agree.

Validity is whether the thing measures what it claims to. A scale that reads the same number every time is perfectly reliable and completely invalid if that number is your height. For personality questionnaires this splits further: construct validity asks whether the scales relate sensibly to other measures of the same idea, and predictive validity asks whether the result forecasts anything in the world, like performance, satisfaction or how long someone stays in a role.

A measure can pass one and fail another, and the four-letter system does exactly that. It holds up moderately well on construct validity and poorly on the type structure and on prediction. Keeping the three questions separate is the whole trick to reading this literature without being pushed around by whoever is talking.

This is the finding that does the most damage, because a type theory claims that preferences are set by adulthood and stable through life. If that were true, retest agreement should be near perfect.

It is not. Pittenger's 2005 review reports that the instrument's own 1998 manual found 35 percent of people received a different four-letter code after a four-week interval, and that McCarley and Carskadon found in 1983 that 50 percent received a different classification on at least one of the four scales across a five-week interval. Howes and Carskadon, in Pittenger's summary, reported that among people whose original score was close to the midpoint, between a quarter and a third of labels moved on the second testing, depending on the scale.

The reason is structural rather than sloppy. Each axis is cut at a midpoint to produce a letter, and most people sit near that midpoint, so a small shift in mood or in how a few items are read moves them across the line. Randall, Isaacson and Ciro put the test-retest reliability of three subscales at .75 or higher and the thinking and feeling subscale at .61. Those are ordinary numbers for a trait questionnaire. They are poor numbers for an instrument that hands back a label and tells you it describes who you are.

A type theory makes a testable prediction about the shape of the data. If there are genuinely two kinds of people on an axis, the scores should pile up in two humps with a dip between them, the way height does when you plot adult men and women together on one chart.

That dip is not there. Pittenger summarizes evidence from several independent sources, including Stricker and Ross in the 1960s, Hicks in 1984 and McCrae and Costa in 1989, all finding a continuous distribution across each dimension, with Bess and Harvey replicating the single-hump finding in 2002 using modern scoring. Most people cluster near the center of each scale, which is the opposite of what the theory needs.

This matters more than it sounds. Pittenger, summarizing Harvey and Murry, puts the loss from converting a continuous score into a letter at between 26 and 32 percent of the information on each scale, and the broader statistical literature on splitting continuous variables at a cut point is blunt about the cost: MacCallum and colleagues show it reduces the association you can detect and can produce spurious results. When the scoring step is the part that creates the sixteen types, a finding about the scoring step is a finding about the types.

This is where the strongest claims are made in workplaces and where the evidence is thinnest. Gardner and Martinko's review of type research in management, which Pittenger summarizes, concluded that efforts to detect simple links between type preferences and managerial effectiveness had been disappointing, and that no definitive conclusion about those relationships could be drawn from the available studies.

The National Research Council volume edited by Druckman and Bjork, quoted in the same review, found that there was not sufficient well-designed research to justify using the four-letter system in career counseling programs. The publisher of the original instrument agrees on the narrower point: its own ethical guidance states that the assessment does not measure ability, competence or skill and is not designed for hiring.

There is a wider context worth holding alongside this. Meta-analytic work on personality and job performance sets the ceiling low. Barrick and Mount's 1991 meta-analysis found one trait, conscientiousness, with consistent relations to performance across occupations and criterion types, at a corrected correlation of about .22, which is roughly five percent of shared variance. The other four traits predicted in particular occupations rather than across them, and the paper describes the size of those estimates as small, below .10 in the cases it singles out. That is the best the better-evidenced instruments manage. Anyone promising that a personality questionnaire will sort out your hiring is overselling the whole field, not just this part of it.

A one-sided page would stop there, so here is the other half. The 2017 systematic review by Randall, Isaacson and Ciro, the first review of this question conducted to systematic-review standards, found the available construct validity studies agree that the instrument has reasonable construct validity, and pooled retest reliabilities of .75 or above on three subscales. The authors are careful about both results: they pooled only three retest studies and flag substantial heterogeneity between them, and they note that most studies used college-age samples and that careful thought is needed before applying the results elsewhere.

Pittenger himself, the most-cited critic, is more measured than his reputation suggests. He writes that the available evidence suggests the instrument does measure constructs related to personality, and that conclusions about the superiority of one instrument over another are premature. His argument is against the type formula and against corporate uses of it, not against the idea that the questions tap something.

And McCrae and Costa, whose 1989 paper is often cited as the demolition, actually concluded that the four scales measure aspects of four of the five major dimensions of normal personality. That is a reinterpretation rather than a dismissal: the scales work, the type story on top of them does not.

The Big Five, or five-factor model, is the framework academic personality research has settled on. It describes five broad dimensions: openness to experience, conscientiousness, extroversion, agreeableness and neuroticism, each reported as a position on a scale rather than as a label.

The four-letter scales map onto four of them. In McCrae and Costa's analysis, the energy axis lines up with extroversion, the information axis with openness, the decisions axis with agreeableness and the structure axis with conscientiousness. The missing one is neuroticism, the dimension covering emotional volatility, anxiety and how heavily setbacks land, which the original four-letter system never included.

What the Big Five has that the four letters do not is a large replicated literature linking the dimensions to outcomes, measurement that keeps the whole scale instead of cutting it, and instruments like the BFI-2 with published facet structure and norms. What it lacks is memorability: nobody puts their conscientiousness percentile in a dating profile. That difference in appeal, rather than any difference in evidence, is most of why four-letter codes are the ones people know.

Shared language is the real product. Two colleagues who keep irritating each other can get further in ten minutes with the words sensing and intuition than in an hour of calling each other pedantic and vague. The letters give a neutral name to a difference that would otherwise be read as a character flaw, and neutral names de-escalate.

Self-reflection is the second use. Reading a well-written description of a type and noticing which three sentences make you wince is a genuinely useful exercise, regardless of whether the box is real. The value is in the noticing, not in the label that prompted it.

Both of those survive every criticism on this page, because neither requires the instrument to be accurate. What does not survive is any use where the result decides something: who gets hired, who gets the project, which degree someone takes, whether two people should date. The moment a code starts closing options rather than opening a conversation, the evidence stops supporting it.

Four decisions follow from the evidence above. First, our 16 Personalities Test reports where you fall on each of the four axes, not only the code, so a near-even split shows up as a near-even split. Second, every result tells you that the axes are spectrums and that a neighboring type may fit you well if one of your letters was a mild lean.

Third, our test pages are labeled for what they are. The 16 Personalities Test carries a note that it is informed by Jungian typology and the Big Five, and it is not presented as a validated clinical instrument or as the official instrument published by the Myers and Briggs organizations. Where we have a better-evidenced test for the same question, such as the Big Five Personality Test, we say so and link it.

Fourth, we do not publish type-based career advice or compatibility verdicts as if they were findings. Where popular sources give a confident answer that the research does not support, our pages say which part is folklore. That costs us some traffic and it is the right trade.

If you want the four letters, take the free 16 Personalities Test and read your position on each axis before you read the code. An optional in-depth report is available for a longer write-up. If any axis comes back close to the middle, treat the corresponding letter as provisional and read the neighboring type as well.

If what you actually want is the better-measured version of the same question, take the Big Five Personality Test instead. It gives you five scores rather than a label, including the emotional stability dimension the four-letter system leaves out, and it is the framework most of the research quoted on this page was conducted in.