Known Quantity

Methods · November 2026 · Known Quantity for Good

How the free personality profile works, and what it cannot do

Six dimensions, public-domain questions, scores with honest ranges, and everything worked out in your browser.

Take the free profile

Our free personality profile asks you to rate 48 or 96 short statements about yourself. It turns your answers into six scores, each with a likely range, right in your browser. Nothing is sent to us. It gives no types and no letter codes, and it is not for hiring or for judging anyone. This page explains how it works, where every part comes from, and what it cannot do.

Words worth knowing: 25 plain-English explanations

The words this page leans on, each explained twice: first in plain English, then in one line for the expert.

Trait
A habit of thinking, feeling or acting that shows up across many situations.
For the expert A relatively stable individual difference in a pattern of behaviour, cognition or affect, estimated here by self-report.
Dimension
A line from low to high that everyone sits somewhere on.
For the expert A continuous latent variable; scores are treated as interval-level on a 1 to 5 mean metric.
Facet
A narrower piece of a dimension.
For the expert A lower-order trait; here four IPIP-HEXACO facets per domain, 4 items each in the full form.
Type
A box that sorts people into groups.
For the expert A categorical latent class; taxometric and mixture-model evidence does not support discrete personality types (McCrae and Costa 1989; Gerlach et al. 2018, disputed by Freudenstein et al. 2019).
Self-report
You describe yourself.
For the expert Data subject to self-presentation, limited self-insight and response styles; self-other agreement for broad traits is about .29 to .41 (Connelly and Ones 2010).
Reliability
How much a score would stay the same if you could take a fresh, equal test.
For the expert The proportion of observed-score variance that is true-score variance; estimated here by Cronbach's alpha.
Cronbach's alpha
A number from 0 to 1 for how well a set of questions hangs together.
For the expert alpha = (k / (k - 1)) x (1 - sum of item variances / total variance); a lower bound on reliability under tau-equivalence.
Test-retest
How similar your scores are when you take it twice.
For the expert Correlation between administrations; HEXACO-100 domains about .86 to .92 over about 13 days (Henry et al. 2022).
Validity
Whether a test measures what it claims and predicts what it should.
For the expert Evidence on content, internal structure, convergent and discriminant relations and criterion relations (APA/AERA/NCME Standards framing; not opened).
Standard error of measurement
How far your score could be off just from the test being short.
For the expert SEM = SD x sqrt(1 - reliability).
Likely range
The stretch where your score probably sits.
For the expert Observed score +/- 1.645 SEM (90%), centred on the observed score, not a regressed true-score estimate.
Norm group
The people your score is compared with.
For the expert A reference sample; IPIP advises local norms and supplies none.
Percentile
The share of a norm group scoring below you.
For the expert Requires a defined reference sample; not reported here.
Reverse keying
Some questions count backwards.
For the expert For a 1 to 5 item keyed negative, keyed value = 6 - response.
Balanced keying
Half the questions count each way.
For the expert Equal positively and negatively keyed items per scale, which cancels much acquiescence at the scale level.
Acquiescence
Agreeing with things no matter what they say.
For the expert A content-independent response style; distorts structure, stronger in some populations (Laajaj et al. 2019).
Unit weights
Every question counts the same.
For the expert Equal item weights; robust against sample-specific regression weights (Bobko, Roth and Buster 2007).
Ipsative
Scores that only compare your traits with each other.
For the expert Forced-choice scores summing to a constant; distort between-person comparison unless modelled (e.g. Thurstonian IRT).
Barnum effect
Believing a vague, flattering description was written just for you.
For the expert Forer (1949): high accuracy ratings for an identical generic profile.
Correlation
How strongly two things move together, from -1 to 1.
For the expert Pearson r; r squared gives the share of variance in common (r = .2 is 4%).
Meta-analysis
A study that pools many studies.
For the expert Quantitative synthesis; results depend on inclusion criteria and corrections (Sackett et al. 2022 on range restriction).
Heritability
How much of the difference between people in a group goes with differences in genes.
For the expert A population statistic, not a share of one person's trait; about .40 on average, .47 in twin and .22 in family and adoption designs (Vukasovic and Bratko 2015).
Lexical hypothesis
The idea that important traits show up as words in every language.
For the expert Basis for the Big Five and HEXACO; how many factors replicate across languages is disputed (De Raad et al. 2010; Ashton and Lee).
Measurement invariance
Whether a test works the same way in different groups.
For the expert Configural, metric and scalar invariance; often not met across cultures (Laajaj et al. 2019; Dong and Dumas 2020, reported).
Public domain
Free for anyone to use for anything.
For the expert IPIP items: "copy, edit, translate, or use them for any purpose without asking permission and without paying a fee."

1. The short version

  • Six broad dimensions, each split into four facets.
  • 48 questions (short form) or 96 (full form), all from a free public question bank.
  • Scored in your browser. Nothing is sent anywhere.
  • Every score comes with a likely range, because short questionnaires are not exact.
  • No types. No four-letter codes. No percentiles. Not for hiring.

2. Why we built another one

The popular tests each get something right and something wrong. The Myers-Briggs (MBTI) uses kind, neutral language, but it sorts people into boxes that the evidence does not support documented. Big Five tests have the best research behind them, but they spread honesty and fairness across other scores reported. The HEXACO model adds an Honesty-Humility dimension, but its official questionnaire may not be used in a public online tool documented. "Dark triad" quizzes point at something real, selfish and manipulative habits, but they put scary labels on noisy scores documented. The Enneagram helps people reflect, but its nine types have not held up in studies documented.

We kept the good parts and fixed what the evidence says is broken: kind language, continuous scores with ranges, an honesty dimension measured with plain behaviour words, and no types.

3. Why six dimensions, and not five or four letters

Lines, not boxes

A large study of the MBTI found no support for "truly dichotomous preferences or qualitatively distinct types" (McCrae and Costa, 1989) documented. Studies report that most people score near the middle of each scale reported. If a scale has a retest reliability of .80, about 80% of people land on the same side of the middle twice, and only about 40% keep all four letters our reading. So we show where you sit on a line, with a range around it.

Five or six

We use the six-dimension HEXACO structure. The sixth dimension, Honesty-Humility, covers being sincere, fair, modest and not greedy. Five-factor tests have no home for it. Whether six or five factors is "right" is debated in the research on trait words across languages debated. We do not claim six is the true number. We claim it is a useful, well documented description that covers honest and dishonest habits openly.

Why no "dark triad" labels

The three "dark" traits (often called narcissism, Machiavellianism and psychopathy) overlap a lot, and a large review called their measures "too simple" (Muris and others, 2017) documented. Most of what they share sits close to low Honesty-Humility plus low Agreeableness. How close is debated: one team estimated the hidden-factor link at about -.95, others argue the two are not the same debated. Labels like these name a kind of person, invite shame, travel as jokes and get used as weapons in arguments. So we measure the shared core with ordinary statements, describe it in plain behaviour words with costs and strengths at both ends, never print a label, and show that card last.

4. Where the questions come from

Every personality statement is an IPIP-HEXACO item from the International Personality Item Pool. The item key is at ipip.ori.org/newHEXACO_PI_key.htm. IPIP says its items "are in the public domain, which means that one can copy, edit, translate, or use them for any purpose without asking permission and without paying a fee" documented. We copied the text exactly and did not edit it. Where a word is hard, we added a short help line under the question.

The pool has 240 items, ten for each of 24 facets. We picked four per facet (96), two per facet for the short form (48). In every dimension, half the statements count toward the trait and half count against it. We picked short, plain items and left out ones that depend on money, schooling, where you live, religion or politics. For example, we left out "Have read the great literary classics" and "Prefer to eat at expensive restaurants", which track schooling and money, and "Leave a mess in my room", which assumes a room of your own.

Some items we kept still have known problems, and we flagged them for a psychologist to review. "Love luxury" depends on income. "Am mainly interested in money" can track hardship, not greed. "Avoid difficult reading material" can track reading in a second language. "Tire out quickly" can reflect health or disability. "Quickly lose interest in the tasks I start" is also common with ADHD, and the results never treat it as a sign of anything.

The two check questions ("please choose Disagree for this one") are our own wording and are never scored.

5. How scoring works

Reverse keying. For a statement that counts against a trait, we flip your answer: 1 becomes 5, 2 becomes 4, and so on (6 minus your answer).

Averages. Each score is the plain average of your answers to that trait's statements, on the 1 to 5 scale. Every question counts the same. We do not use weights from a sample, because weights found in one group usually shift in the next, and we have no sample at all: no answers ever reach us.

Missing answers. We never fill in a skipped answer. A dimension is scored only if you answered at least 75% of its statements (12 of 16, or 6 of 8 on the short form) and at least 2 that count each way. A facet needs 3 of its 4. Otherwise the page says there are not enough answers and offers a way back to the skipped questions.

The likely range. We estimate how much a score could be off just because the test is short. First, the reliability for the number of statements you answered, k, from a planning reliability a for n statements: r = a / (n - (n - 1) a), then reliability = k r / (1 + (k - 1) r). Then the standard error of measurement: SEM = SD x the square root of (1 - reliability). The likely range is your score plus or minus 1.645 x SEM (a 90% range), kept between 1 and 5. Skipping questions makes the range wider, so you can see it.

Planning values, said plainly. We cannot measure reliability ourselves, because no answers leave your device. So we use planning values from published studies our reading: a reliability of .80 for a 16-statement dimension (so about .67 for the 8-statement short form), and each facet's published 10-item reliability stepped down to 4 items. The spread value, SD, is set to 0.65 for every score. That 0.65 is a placeholder estimate until real data exists: we did not find a published figure for these exact scales, and a reviewer has not yet checked it. With these values the full form's ranges are about plus or minus 0.48, and the short form's about plus or minus 0.62.

The range is centred on your own score. A textbook alternative pulls it toward the average person's score, but that needs a population average we do not have.

Comparing your own scores. We call two of your scores clearly different only if the gap is more than 1.645 x the square root of (SEM1 squared + SEM2 squared), about 0.68 on the full form and 0.88 on the short form. Otherwise we say they are about the same.

Answer checks. A few gentle checks look for answers that would make the results meaningless: many identical answers in a row, very fast answering (a median under 1.5 seconds), opposite statements given the same answer, a check question answered differently from what it asked (skipping one is fine, like any question, and is not counted for or against you), and agreeing (or disagreeing) with nearly everything. If one trips, we say so kindly and still show your results. Timing stays in memory and is never saved. These thresholds are our own and have not been tested our reading.

6. Why there are no percentiles

A percentile tells you what share of a comparison group scored below you. That needs a defined group of people who took the same questions. IPIP says of norms: "None are available, nor should there be", and advises local norms documented. The one large public dataset for these scales we found has no stated licence. So we will not invent one. A 3 on our page is the middle of the answer scale, not the average person. People tend to describe themselves kindly, so a 3.6 on a trait most people value might be ordinary.

7. What the research says, and how sure it is

All of these are findings about groups of people, not predictions about you.

  • Work. Across 92 reviews and more than 1.1 million people, conscientiousness went with better results on 98% of 175 work outcomes, with an average correlation of .20 (Wilmot and Ones, 2019) documented. A correlation of .2 means about 4% of the differences between people. Useful across thousands, nearly useless for one person our reading.
  • Health. Conscientiousness relates to living longer, with effects of about .07 to .17 in single studies, larger than the effect of income in the same review (Roberts and others, 2007) documented.
  • Other people's view. People who know you agree with your self-ratings at about .29 to .41 for Big Five traits (Connelly and Ones, 2010) documented.
  • Stability. Scores settle in young adulthood; people get calmer and steadier with age; facets are less stable than broad dimensions (Bleidorn and others, 2022) documented.
  • Genes. About 40% of the differences between people in a group go with differences in genes on average, but twin studies give .47 and family and adoption studies .22, and that gap is debated (Vukasovic and Bratko, 2015) documented debated.

8. What this cannot tell you

  • A diagnosis. It does not test for anxiety, depression, ADHD, or any personality disorder.
  • Ability. It says nothing about how smart, skilled or capable you are.
  • Job fit. Do not use this for hiring, promotion, or choosing people for anything. It was not built or tested for that, and using personality tests in hiring has legal rules in many places. In the US, the EEOC treats personality tests as selection procedures covered by civil rights law, and psychological tests that can identify a mental disorder count as medical exams before a job offer documented.
  • Other people. This is for describing yourself. Answering for someone else does not tell you about them.
  • The future. Traits change, and behaviour depends on the situation as much as on traits.

9. Your privacy

Your answers stay in this browser tab. We cannot see them. The page sends nothing after it loads. It never puts answers or scores in the web address, and it carries a security rule that tells your browser to block any connection the page tries to make and to run only this site's own scripts, so no analytics or other outside script can run on it. Like any website, our host, Cloudflare, still sees the request for the page itself (see our privacy policy). If you leave it partway, your answers are kept in this tab's own session storage, so a reload does not lose them. Closing the tab, finishing, or pressing Clear my answers deletes them.

If you save your results, the file is made on your device and downloaded straight to you. It holds the date, which form you took, your answers and your scores, so you can open it again later. It never holds timing. Delete the file like any other file. Our privacy policy covers the rest of the site.

10. Known limits

  • These questions were written in English, mostly tested with Western volunteers, and may work less well elsewhere. Studies in low and middle income countries found the usual structure replicates poorly (Laajaj and others, 2019) documented.
  • No one has measured the reliability of this exact set of 96 questions. Our ranges use planning values, and the spread value is a placeholder.
  • The answer-check thresholds are untested.
  • Time estimates (about 5 and about 15 minutes) are planning figures until we measure real completion times.
  • English only for now.
  • A psychologist has not yet reviewed the item choices, the descriptions or the planning values. Until then, treat every result as a rough sketch.

11. Sources

Tags: documented means we opened the source and read the figure; reported means it comes from a secondary source or was not opened; debated means researchers disagree; our reading means our own reasoning or arithmetic.

These tools are free and stay free. If one helped and you can spare it, the tip jar helps keep them that way. Never required.