Do personality tests work if you're autistic or have ADHD?

You finished the questionnaire and the result described someone withdrawn, anxious and disorganised. Some of it landed. Some of it felt like a description of your worst week rather than your life. And underneath both reactions sat a more specific question: does this thing even work on me, or am I being measured against a ruler built for somebody else?

That is a better question than most people asking it realise, and the research has a genuinely surprising answer — not the reassurance you might want, and not the dismissal you might fear.

A note on scope first, because it matters. The usable evidence here covers autism and ADHD. There is essentially nothing on Big Five measurement in dyspraxia, dyslexia or Tourette’s, and nothing on intersectional groups like late-diagnosed autistic women. Where this post says “neurodivergent”, read it as “autistic or ADHD”, because that is all anyone has studied.

The short answer

Autistic and ADHD people do score differently as groups, and the differences are large. In a meta-analysis of Big Five traits in autism, group contrasts ran from g = −0.88 to g = −1.42. For ADHD, one meta-analysis of 2,023 adults found neuroticism d = 1.11 and conscientiousness d = −0.89.

But nobody has checked whether the questionnaire means the same thing in both groups. There is no published test of measurement invariance for any Big Five inventory between autistic and non-autistic respondents, or between ADHD and non-ADHD respondents. Not a weak result — an absence.

And the differences, however large, are far too small to identify anyone. Telling two groups apart at even 80% sensitivity and 80% specificity requires d = 1.66. Every effect above is smaller. When researchers actually tried to classify people from their Big Five scores, they got 70% right — barely better than a coin weighted by base rates.

So the test is not diagnosing you. It also has not been shown to be measuring you on the same scale as everyone else.

What do the score differences actually look like?

Consistent, large, and easy to misread in one specific way.

The autism meta-analysis pooled 14 studies and found that every Big Five trait correlated negatively with autism characteristics. Here is the trap: the fifth trait was coded as emotional stability, not neuroticism. “Lower on all five” therefore means lower emotional stability — which is higher neuroticism, not lower. Plenty of secondhand summaries get this backwards.

The individual study with the clearest numbers found conscientiousness d = −0.90, agreeableness −0.85, neuroticism +0.84, extraversion −0.81 and openness −0.55.

For ADHD, the pattern is similar in shape: neuroticism up substantially, conscientiousness down, extraversion and agreeableness modestly down, openness essentially unchanged (d = 0.15, not significant). A separate meta-analysis adds a useful detail — conscientiousness tracks inattention specifically, while agreeableness tracks hyperactivity and impulsivity.

Two cautions on the ADHD numbers. Heterogeneity across studies was severe (I² = 85–97%), meaning the studies disagree a lot. And ADHD screening scales share item content with conscientiousness scales — “I leave my belongings around” sits comfortably on both — so an unknown share of that d = −0.89 is the two questionnaires asking the same question, not a substantive difference.

Is the test measuring me, or just re-describing being autistic?

Partly the second, and there is now direct evidence for it.

A 2025 study mapped autism-related traits onto the five-factor space and found that socially oriented autistic traits behave like facets of extraversion — they sit inside the domain rather than beside it. Behaviourally oriented traits, like repetitive behaviour, sat outside the model entirely.

That is worth sitting with. If the social features of autism are located in the extraversion domain, then a low extraversion score for an autistic person is not an independent finding about their personality. It is, at least in part, the instrument restating the thing you already knew, in different vocabulary.

This does not make the score meaningless. It does mean “you scored low on extraversion” carries less new information for you than it would for someone else — and that reading it as a separate fact about your sociability would be double-counting.

Can I even answer these questions accurately about myself?

More accurately than the stereotype suggests — and this is the finding most likely to be misreported.

The concern is real enough on its face. Around half of autistic people meet the clinical threshold for alexithymia, difficulty identifying and describing one’s own feelings, against roughly 5% of non-autistic people — a risk ratio of 6.5. And on interoception, autistic participants were more confident while less accurate at a heartbeat-counting task.

But when researchers tested self-insight about personality specifically, the picture changed. Comparing self-reports against parent reports, agreement in the autistic sample ran at a median r = .34, against .38 for the comparison group — close. And the small advantage the non-autistic group did show was driven entirely by normative agreement: their profiles were simply more average, which makes any self-report look more accurate. On distinctive agreement — the actual measure of knowing your own particular self — the groups did not differ.

So: difficulty naming emotions in the moment is common. Difficulty describing your own stable characteristics is not the same thing, and on the evidence available it does not appear to be impaired. Be wary of anyone who collapses the two.

Two honest limits: the alexithymia figure comes from a scale that itself turned out to need repair for use with autistic people, so some of that effect is measurement artefact. And the self-insight finding rests on one 2014 study of 37 adults. It is promising, not settled.

What about masking?

It correlates with exactly the traits you would expect. Whether it distorts your score is untested.

Camouflaging is a measurable construct with a validated scale, developed from autistic adults’ own accounts. And the one study linking it to Big Five scores found camouflaging related negatively to extraversion, agreeableness and conscientiousness, and positively to neuroticism.

Here is where to be careful. That study is correlational. It cannot distinguish “masking makes your extraversion score misleading” from “people lower in extraversion mask more.” Those are different claims with different implications, and no study has tested whether camouflaging causally biases Big Five self-report. Anyone telling you your masked score is definitely wrong is going beyond the evidence.

Why does “nobody has tested invariance” matter?

Because the profession’s own rulebook says it should have been tested, and because when people do test it on other instruments, the results split roughly down the middle.

The Standards for Educational and Psychological Testing — published jointly by the AERA, APA and NCME, and open access since 2021 — includes Standard 3.6: where credible evidence indicates scores may differ in meaning for relevant subgroups, developers and users are responsible for examining the validity of score interpretations for those subgroups. Large, replicated group differences are precisely the credible evidence that is supposed to trigger that examination.

It has not happened for the Big Five. And you cannot assume the answer, because the adjacent literature genuinely splits:

• One study tested two questionnaires in the same autistic sample. One failed invariance at the first hurdle — every item on one subscale functioned differently. The other held completely. Invariance is a property of a specific instrument, not of questionnaires in general.
• A parent-report measure tested in autistic adolescents had 18 of 25 items functioning differentially, with the problems concentrated in reverse-coded items — and Big Five inventories lean heavily on reverse-coded items.
• But a 2025 evaluation of an autism-trait measure across 1,322 autistic and 1,279 non-autistic adults did demonstrate invariance, across both autism status and gender.

That last one is the constructive point: this check is achievable. It has simply never been run on the instruments that millions of people take.

The best worked example of what happens when someone does the work comes from Williams and Gotham (2021) — an autistic researcher examining the standard alexithymia scale. Eight of nine items showed statistically significant differential functioning; one was practically significant, a question about “physical sensations that even doctors don’t understand.” Dropping it produced a shortened score that can legitimately be compared across groups. The response to a biased instrument was to repair it, not to abandon it or to tell people their scores mean something different.

Does this matter if a test is used to hire me?

Yes, and this is the setting where the gap stops being academic.

A 2026 study ran three common selection tools across 99 non-autistic and 107 autistic respondents. All three produced differences unfavourable to autistic candidates. On the five-factor inventory: emotional stability d = −0.98, conscientiousness −0.69, agreeableness −0.59. On a situational judgment test, d = −0.84 — and autism remained a significant negative predictor of the score even after controlling for general mental ability, all five personality traits, gender and age.

Note the sample is small and the study measures score differences, not adverse impact in a real hiring process — actual impact depends on cut scores and how many people are hired. But the direction is consistent, and it is the reason the missing invariance work matters beyond curiosity.

So should I take one?

Yes, with a clear idea of what it can and cannot do.

• It is not a diagnosis, and that is arithmetic rather than a disclaimer. Distinguishing groups at 80% sensitivity and specificity needs d = 1.66. The largest effect anywhere above is 1.42, most are under 1.0, and an actual attempt to classify people from Big Five scores managed 70%. At d = 1.0, 48% of autistic people still fall within one standard deviation of the non-autistic average.
• Read low extraversion carefully. For an autistic reader it may be partly restating autistic social difference rather than adding a separate fact.
• Read conscientiousness carefully if you have ADHD. Some of that score is the questionnaires overlapping.
• Treat the norm group as part of the result. A percentile says how you answered relative to a comparison sample. It is not a statement about who you are — and the same Standards (5.8) require that sample to be described well enough for you to judge whether it applies to you.
• Use the facets, not the headline. A single extraversion number collapses warmth, gregariousness and assertiveness — which can move independently, and often do.

And if what you actually want to know is whether you’re autistic or have ADHD, a personality questionnaire cannot tell you. That is a clinical question for a clinician, and the numbers above are the reason why — not caution, but measurement.