What makes a personality test accurate?

"Most accurate personality test" is a common thing to search and a hard thing to answer, because accuracy is not one property. A test can be excellent at one kind of accuracy and useless at another.

Here are the five checks that matter, in the order you can actually run them.

1. Does it give you the same answer twice?

The first question is consistency. If you take a test today and again in three weeks, having changed nothing important about your life, does it say roughly the same thing?

A test that moves you into a different category on a slightly different mood was not measuring a stable trait. This is the check that most category-based tests quietly fail, because a small shift near a boundary flips the whole label.

What to look for: any mention of test–retest reliability. Its absence is not proof of a problem, but no serious instrument omits it forever.

2. Do the questions hang together?

If several items are supposed to measure one thing, people's answers to them should correlate. If they do not, the scale is measuring several things at once and averaging them into a number that means nothing in particular.

The usual figure is Cronbach's alpha, reported per scale. Roughly: above .80 is solid for this kind of instrument, .70 is workable, below that is a scale still under construction.

What to look for: a published alpha for each scale, not one number for the whole test.

3. Does it agree with established measures?

A new instrument claiming to measure extraversion should correlate strongly with existing, well-validated extraversion measures. If it does not, one of the two is not measuring what it says.

What to look for: any statement about how the test relates to established public-domain instruments.

4. Does it tell you what it cannot do?

This is the fastest signal and it requires no statistics.

A test that measures personality traits cannot tell you what work you should do, it cannot tell you who to marry, and it cannot tell you anything about your health. Any instrument that offers all of those from one questionnaire is over-claiming, and over-claiming is the most reliable indicator that the numbers underneath are not being taken seriously either.

What to look for: a limitations page. If there is not one, that is the finding.

5. Are the scores continuous, or forced into boxes?

Personality traits are distributed continuously. People do not cluster into types; the middle of every dimension is the most crowded part.

When a test converts a continuous score into a category, it throws away the information that mattered and manufactures a hard boundary where the data has a gentle slope. Two people either side of a line are described as opposites while being nearly identical.

What to look for: does it show you a number and where you sit, or only a label?

How we do against our own checklist

It would be poor form to publish this and not apply it.

Continuous scores: yes. We report a 0–100 position on each dimension and treat the quadrant as a home region, never a type.

Stated limits: yes, on the methodology page, and we make no claims about work suitability and none about health.

Internal consistency: our target is .80 per scale, and the figures will be published on the methodology page as our own data accumulates. They are not published yet, because we do not have enough responses to report anything honest.

Test–retest: not yet either, for the same reason. It requires the same people returning months apart.

Convergent validity: items were written against public-domain constructs, and the formal comparison is on the same list.

That is three of five in place and two outstanding, stated rather than hidden. If another test publishes better numbers than ours, use theirs — the point of the checklist is that you should not have to take anybody's word for it, including ours.

You can see the instrument and its scoring on the methodology page, or take the free Personality Map and judge the output yourself.