Cognitive science

Which Personality Systems Have Real Evidence?

By PsyCove Editorial · 11 min read · Updated 2026-10-27 · Checked against the sources listed below

Dozens of systems claim to sort people, and the useful question is not which one feels most accurate. It is which ones survive the same four tests. This is a comparison rather than another list of MBTI criticisms: those are set out in the guide to the model's limits, and the head-to-head with the Big Five is in a separate guide. Here every major system is scored on the same criteria.

Key takeaways

Evidence is four separate questions. Does the structure replicate, do the scores hold still, do they agree with other methods, and do they predict anything outside the questionnaire.
The Big Five wins on the first three and does respectably on the fourth. Its effects on mortality, divorce and occupational attainment were indistinguishable from those of socioeconomic status and cognitive ability (Roberts et al., 2007).
The MBTI's problem is not reliability. Its scales show strong internal consistency and test–retest estimates (Capraro & Capraro, 2002). The problem is that the categories are cut from continuous scores and then treated as kinds (Bess & Harvey, 2002).
Feeling accurate is not one of the four tests. Identical feedback handed to a whole class is rated as personally accurate by most of it (Forer, 1949), which is why the systems without evidence are not obviously losing.

What Counts as Evidence for a Personality System?

Four tests, and a system can pass some of them while failing others. Most public arguments about personality frameworks go wrong by collapsing them into one.

The four tests

Structure. Does the proposed set of dimensions reappear when someone else collects data with a different instrument, in a different sample? A framework that only exists in its own questionnaire has not been tested.
Stability. Do scores hold still enough to describe a person? Not perfectly — enough to be more informative than a coin.
Convergence. Do self-reports, observer ratings and behavioral tests agree? A system that only works when the person is asked about themselves is measuring a story.
Prediction. Does the score say anything about outcomes outside the questionnaire — performance, health, relationships, longevity?

The order matters. A framework that fails structure cannot be rescued by prediction, because you would not know what the prediction was about. This ordering would be wrong if a framework failing structure turned out to predict something the Big Five misses; no such framework has been shown to.

How Do the Major Systems Compare?

The comparison table carries the argument, and everything after it is the reasoning behind the rows.

Six systems on the same four tests

System	What it claims	Structure and stability	Prediction
Big Five	Five continuous dimensions covering most trait vocabulary	Strong. Recovered across instruments and across self- and observer-ratings (McCrae & Costa, 1987), and in most cultures tested	Moderate and broad. Effects on mortality, divorce and occupational attainment match those of socioeconomic status and cognitive ability (Roberts et al., 2007)
HEXACO	Six dimensions, adding honesty–humility to the Big Five	Good. Derived from lexical studies in multiple languages rather than one (Ashton & Lee, 2007)	Moderate. The honesty–humility dimension adds prediction the Big Five misses
MBTI	Four dichotomies, sixteen types	Mixed. Scales are reliable (Capraro & Capraro, 2002) but the categories are cut points on continuous scores (Bess & Harvey, 2002)	Weak. No consistent prediction of managerial behavior or performance in the management literature (Gardner & Martinko, 1996)
Dark Triad	Three aversive traits: narcissism, Machiavellianism, psychopathy	Moderate. Three correlated but separable dimensions with distinct correlates (Furnham et al., 2013)	Moderate for short-term and antisocial outcomes; weaker over longer intervals
Enneagram	Nine types with wings and levels	Not established in the peer-reviewed literature	Not established
Astrology	Twelve signs from birth date	Refuted. Astrologers matching charts to personality profiles performed at chance (Carlson, 1985)	None

Where Does the Big Five Actually Hold Up?

The Big Five holds up best on the least glamorous question: whether the same five dimensions appear when you change the method.

McCrae and Costa compared self-reports, peer ratings and questionnaire scores built from different item pools. The five factors emerged across all of them, which is the kind of agreement a framework needs before anything else can be said (McCrae & Costa, 1987). That is convergence, not just internal consistency, and it is the reason the model became the default comparison point.

The prediction record is more modest than its advocates sometimes imply, and more substantial than its critics assume. In a review restricted to prospective longitudinal studies that controlled for background factors, personality traits predicted mortality, divorce and occupational attainment about as well as socioeconomic status and cognitive ability did (Roberts et al., 2007).

“About as well as socioeconomic status” is the honest headline. It is not a claim that a questionnaire beats your circumstances. It is a claim that both carry information, and that the trait measures were previously dismissed on a hunch about small effects.

What Does the MBTI Actually Fail?

Not reliability, and repeating that claim is the fastest way to lose an argument you could have won.

A meta-analytic review of the indicator's score reliability across studies found strong internal consistency and test–retest estimates, with variation between studies (Capraro & Capraro, 2002). The instrument is consistent. The failure is elsewhere.

Where the failure sits

The cut points. Continuous scores are split into halves, so a few points decides a letter, and the resulting categories look like kinds when they are divisions (Bess & Harvey, 2002).
The information loss. Distance from the midpoint disappears. A person just past the line and a person far beyond it receive the same letter.
The prediction. Two decades of management research using the indicator produced no consistent evidence that type predicts managerial behavior or performance (Gardner & Martinko, 1996).
The claim inflation. What the instrument measures is a preference for one side of a dichotomy. What it is commonly used to claim is a description of how someone thinks (Boyle, 1995; Pittenger, 2005).

What About HEXACO and the Dark Triad?

Both are real research programs with real evidence, and both are narrower than their popular versions.

HEXACO adds an honesty–humility dimension that the Big Five handles badly, and it came out of lexical studies in several languages rather than one, which is a methodological advantage rather than a marketing one (Ashton & Lee, 2007). Its practical value is concentrated in that sixth dimension: it predicts exploitative behavior that the Big Five largely misses.

The Dark Triad is three traits with overlapping but separable nomological networks — narcissism, Machiavellianism and psychopathy — and the review literature is careful to keep them apart even though short measures blur them (Furnham et al., 2013). It predicts short-term and antisocial outcomes reasonably well and longer-term ones less well, which is what you would expect from traits defined around impulse and exploitation.

Neither system is a rival to the Big Five in coverage. They are additions at specific edges, and treating them as complete accounts of personality is a category error in the same direction as treating a type code as one.

Why Do Systems Without Evidence Still Feel Accurate?

Because the feeling of accuracy is produced by the description, not by the system that generated it.

Generalities travel. Identical feedback given to a whole class was rated as accurate by most of it, and the text had been assembled from horoscopes (Forer, 1949).
Agreement is easy to manufacture. A description written to be recognizable will be recognized. A description written to be falsifiable will sometimes be rejected, which is why the second kind is rarer.
The system is never asked to lose. When a prediction fails, the reading is adjusted. When it succeeds, the system is credited. That is a test that cannot fail, and a test that cannot fail produces no evidence.

This is how a system with no research base can feel more accurate than one with a large literature. The literature constrains what the descriptions are allowed to say.

What About the Systems With No Research Base at All?

Enneagram typologies and astrology are the two most common examples, and they are in different positions from each other.

For astrology the question has been settled experimentally rather than by argument. In a double-blind study, astrologers matched birth charts to people with whom they were already familiar and did no better than chance, while doing well on the control task of matching charts to the correct questionnaire responses (Carlson, 1985).

For the Enneagram the situation is an absence rather than a refutation. There is no body of validation work comparable to the Big Five or HEXACO literature, and the honest statement is that the system has not been tested at the standard the others are held to.

That absence is not proof of uselessness. It is proof that nobody has done the work, and a framework that has not been tested should not be presented in the same voice as one that has.

What If Two Systems Disagree About You?

Two systems disagreeing about you is usually not a contradiction, because they are answering different questions at different levels. A trait score answers a population question: where you sit relative to other people on one dimension. A type code answers a sorting question: which side of a cut point you happened to fall on. Neither reading is wrong when the two are not measuring the same thing.

Three kinds of disagreement are worth telling apart before you decide which reading to keep.

Level. One system describes a population and the other describes a person. The population claim is the one with evidence behind it, and the personal claim is the one you can act on. Treating the second as if it inherited the first's evidence is the common error.
Construct. A dimension is continuous and a code is categorical, so a person can be high on a dimension and the wrong letter at the same time without either system being inaccurate.
Input. Self-reports, observer ratings and questionnaire scores agree only partly, which is why the convergence test above is not a formality. A disagreement between systems that used different inputs may be a disagreement between inputs rather than between systems.

When two readings conflict, prefer the one that stated something you could check and got it right. A reading that only ever confirms itself has not been tested, however accurate it feels.

Why Do Companies Still Pay for Systems With Weak Evidence?

The persistence of weak instruments in corporate training is not a mystery about evidence. It is a market. Someone sells the certification, someone delivers the workshop, and the buyer is rarely the person who bears the cost of a wrong label.

Three incentives keep that market running, and none of them is about what predicts anything.

The purchase is legible. A four-letter framework is easy to explain to a budget holder, easy to print and easy to remember. "A five-factor model with acceptable convergent validity" is none of those things.
The failure is invisible. A workshop that makes a team feel understood produces a satisfaction score the same afternoon. A prediction that was never made cannot be missed.
The language is the product. What organizations buy is a shared vocabulary for talking about colleagues. The Big Five measures better and offers no such vocabulary, which is a real advantage for the framework and a real disadvantage for the sale.

This is also why the evidence question does not settle the commercial one. If you are choosing a framework to buy, ask what the vendor is accountable for. If the answer is "the experience", you are buying a vocabulary, and that is a different purchase from a measurement.

How Should You Use This Table?

Use it to decide what kind of claim each system is allowed to make, not to pick a favorite.

Match the claim to the row. A Big Five score can support a statement about a population and a cautious one about a person. A type code can support a vocabulary and a conversation.
Do not upgrade a system by borrowing another's evidence. “The Big Five predicts job performance, and my type is basically Big Five” is two claims, and the second one needs its own test.
Treat reliability as necessary and not sufficient. A consistent instrument that discards the information you need is a consistent instrument that discards the information you need.
Ask what would count as the system being wrong. If nothing would, the system is not making claims and cannot be compared on these four tests at all.

We keep this test to four positions and one reference type, and we test every additional claim against what the score can support. Our own result page is in the same table as everyone else's rather than outside it. A framework that will not state its limits is telling you something about itself.

Frequently asked questions

Is the Big Five the most scientifically supported personality model?

It has the strongest record on structure and convergence, having been recovered across instruments, raters and cultures, and it holds up reasonably on prediction (McCrae & Costa, 1987; Roberts et al., 2007). HEXACO is a serious competitor at one edge of the model rather than a refutation of it.

Is the MBTI scientifically valid?

Its scales are reliable, which is not the usual criticism (Capraro & Capraro, 2002). The problems are the forced cut points, the information lost at them, and the absence of consistent prediction in applied research (Bess & Harvey, 2002; Gardner & Martinko, 1996).

Is the Enneagram pseudoscience?

It is untested rather than refuted. There is no validation literature at the standard the other systems in this table are held to, which is a different and weaker charge than having been tested and failed.

Has astrology been tested properly?

Yes. Twenty-eight astrologers each matched a natal chart to one of three personality profiles, one of which was the client's. They were right about a third of the time, the rate chance predicts, against their own prediction of half (Carlson, 1985).

If the MBTI is weak, why does it still feel accurate?

Because the feeling comes from the description rather than the system. Identical generalities are endorsed as personally accurate by most recipients (Forer, 1949), and a type description written to be recognizable will be recognized.

Which Personality Systems Have Real Evidence?

By PsyCove Editorial · 11 min read · Updated 2026-10-27 · Checked against the sources listed below

Key takeaways

Evidence is four separate questions. Does the structure replicate, do the scores hold still, do they agree with other methods, and do they predict anything outside the questionnaire.
The Big Five wins on the first three and does respectably on the fourth. Its effects on mortality, divorce and occupational attainment were indistinguishable from those of socioeconomic status and cognitive ability (Roberts et al., 2007).
The MBTI's problem is not reliability. Its scales show strong internal consistency and test–retest estimates (Capraro & Capraro, 2002). The problem is that the categories are cut from continuous scores and then treated as kinds (Bess & Harvey, 2002).
Feeling accurate is not one of the four tests. Identical feedback handed to a whole class is rated as personally accurate by most of it (Forer, 1949), which is why the systems without evidence are not obviously losing.

What Counts as Evidence for a Personality System?

Four tests, and a system can pass some of them while failing others. Most public arguments about personality frameworks go wrong by collapsing them into one.

The four tests

Structure. Does the proposed set of dimensions reappear when someone else collects data with a different instrument, in a different sample? A framework that only exists in its own questionnaire has not been tested.
Stability. Do scores hold still enough to describe a person? Not perfectly — enough to be more informative than a coin.
Convergence. Do self-reports, observer ratings and behavioral tests agree? A system that only works when the person is asked about themselves is measuring a story.
Prediction. Does the score say anything about outcomes outside the questionnaire — performance, health, relationships, longevity?

How Do the Major Systems Compare?

The comparison table carries the argument, and everything after it is the reasoning behind the rows.

Six systems on the same four tests

System	What it claims	Structure and stability	Prediction
Big Five	Five continuous dimensions covering most trait vocabulary	Strong. Recovered across instruments and across self- and observer-ratings (McCrae & Costa, 1987), and in most cultures tested	Moderate and broad. Effects on mortality, divorce and occupational attainment match those of socioeconomic status and cognitive ability (Roberts et al., 2007)
HEXACO	Six dimensions, adding honesty–humility to the Big Five	Good. Derived from lexical studies in multiple languages rather than one (Ashton & Lee, 2007)	Moderate. The honesty–humility dimension adds prediction the Big Five misses
MBTI	Four dichotomies, sixteen types	Mixed. Scales are reliable (Capraro & Capraro, 2002) but the categories are cut points on continuous scores (Bess & Harvey, 2002)	Weak. No consistent prediction of managerial behavior or performance in the management literature (Gardner & Martinko, 1996)
Dark Triad	Three aversive traits: narcissism, Machiavellianism, psychopathy	Moderate. Three correlated but separable dimensions with distinct correlates (Furnham et al., 2013)	Moderate for short-term and antisocial outcomes; weaker over longer intervals
Enneagram	Nine types with wings and levels	Not established in the peer-reviewed literature	Not established
Astrology	Twelve signs from birth date	Refuted. Astrologers matching charts to personality profiles performed at chance (Carlson, 1985)	None

Where Does the Big Five Actually Hold Up?

The Big Five holds up best on the least glamorous question: whether the same five dimensions appear when you change the method.

What Does the MBTI Actually Fail?

Not reliability, and repeating that claim is the fastest way to lose an argument you could have won.

Where the failure sits

The cut points. Continuous scores are split into halves, so a few points decides a letter, and the resulting categories look like kinds when they are divisions (Bess & Harvey, 2002).
The information loss. Distance from the midpoint disappears. A person just past the line and a person far beyond it receive the same letter.
The prediction. Two decades of management research using the indicator produced no consistent evidence that type predicts managerial behavior or performance (Gardner & Martinko, 1996).
The claim inflation. What the instrument measures is a preference for one side of a dichotomy. What it is commonly used to claim is a description of how someone thinks (Boyle, 1995; Pittenger, 2005).

What About HEXACO and the Dark Triad?

Both are real research programs with real evidence, and both are narrower than their popular versions.

Why Do Systems Without Evidence Still Feel Accurate?

Because the feeling of accuracy is produced by the description, not by the system that generated it.

Generalities travel. Identical feedback given to a whole class was rated as accurate by most of it, and the text had been assembled from horoscopes (Forer, 1949).
Agreement is easy to manufacture. A description written to be recognizable will be recognized. A description written to be falsifiable will sometimes be rejected, which is why the second kind is rarer.
The system is never asked to lose. When a prediction fails, the reading is adjusted. When it succeeds, the system is credited. That is a test that cannot fail, and a test that cannot fail produces no evidence.

This is how a system with no research base can feel more accurate than one with a large literature. The literature constrains what the descriptions are allowed to say.

What About the Systems With No Research Base at All?

Enneagram typologies and astrology are the two most common examples, and they are in different positions from each other.

That absence is not proof of uselessness. It is proof that nobody has done the work, and a framework that has not been tested should not be presented in the same voice as one that has.

What If Two Systems Disagree About You?

Three kinds of disagreement are worth telling apart before you decide which reading to keep.

Level. One system describes a population and the other describes a person. The population claim is the one with evidence behind it, and the personal claim is the one you can act on. Treating the second as if it inherited the first's evidence is the common error.
Construct. A dimension is continuous and a code is categorical, so a person can be high on a dimension and the wrong letter at the same time without either system being inaccurate.
Input. Self-reports, observer ratings and questionnaire scores agree only partly, which is why the convergence test above is not a formality. A disagreement between systems that used different inputs may be a disagreement between inputs rather than between systems.

When two readings conflict, prefer the one that stated something you could check and got it right. A reading that only ever confirms itself has not been tested, however accurate it feels.

Why Do Companies Still Pay for Systems With Weak Evidence?

Three incentives keep that market running, and none of them is about what predicts anything.

The purchase is legible. A four-letter framework is easy to explain to a budget holder, easy to print and easy to remember. "A five-factor model with acceptable convergent validity" is none of those things.
The failure is invisible. A workshop that makes a team feel understood produces a satisfaction score the same afternoon. A prediction that was never made cannot be missed.
The language is the product. What organizations buy is a shared vocabulary for talking about colleagues. The Big Five measures better and offers no such vocabulary, which is a real advantage for the framework and a real disadvantage for the sale.

How Should You Use This Table?

Use it to decide what kind of claim each system is allowed to make, not to pick a favorite.

Match the claim to the row. A Big Five score can support a statement about a population and a cautious one about a person. A type code can support a vocabulary and a conversation.
Do not upgrade a system by borrowing another's evidence. “The Big Five predicts job performance, and my type is basically Big Five” is two claims, and the second one needs its own test.
Treat reliability as necessary and not sufficient. A consistent instrument that discards the information you need is a consistent instrument that discards the information you need.
Ask what would count as the system being wrong. If nothing would, the system is not making claims and cannot be compared on these four tests at all.

Which Personality Systems Have Real Evidence?

What Counts as Evidence for a Personality System?

How Do the Major Systems Compare?

Where Does the Big Five Actually Hold Up?

What Does the MBTI Actually Fail?

What About HEXACO and the Dark Triad?

Why Do Systems Without Evidence Still Feel Accurate?

What About the Systems With No Research Base at All?

What If Two Systems Disagree About You?

Why Do Companies Still Pay for Systems With Weak Evidence?

How Should You Use This Table?

Frequently asked questions

Is the Big Five the most scientifically supported personality model?

Is the MBTI scientifically valid?

Is the Enneagram pseudoscience?

Has astrology been tested properly?

If the MBTI is weak, why does it still feel accurate?

Further Reading (primary sources & authoritative references)

Related reading

Which Personality Systems Have Real Evidence?

What Counts as Evidence for a Personality System?

How Do the Major Systems Compare?

Where Does the Big Five Actually Hold Up?

What Does the MBTI Actually Fail?

What About HEXACO and the Dark Triad?

Why Do Systems Without Evidence Still Feel Accurate?

What About the Systems With No Research Base at All?

What If Two Systems Disagree About You?

Why Do Companies Still Pay for Systems With Weak Evidence?

How Should You Use This Table?

Frequently asked questions

Is the Big Five the most scientifically supported personality model?

Is the MBTI scientifically valid?

Is the Enneagram pseudoscience?

Has astrology been tested properly?

If the MBTI is weak, why does it still feel accurate?

Further Reading (primary sources & authoritative references)

Related reading