Methodology

My Path publishes self-report questionnaires — about personality, interests, work style and relationships, and about the people and places around you — plus an IQ screener and practice material for school subjects and exams. This page describes how those tests are scored and what the numbers on a result rest on. It also lists, plainly, the validation work that has not been done. If a sentence here and a test page disagree, tell us: one of them is wrong.

None of the tests is a clinical, diagnostic or hiring instrument. A result describes how you answered on one occasion, compared — where we have the data — with other people who took the same test here.

One row per test, generated from the same figures as the rest of this page. The sample is the number of completed sittings a figure rests on; a dash means there is none yet.

The items are our own. They are written for this platform on well-studied models — Holland's RIASEC interests, the Big Five traits, DISC styles, a sixteen-type preference model after Jung, and others — or adapted from public-domain item pools such as the IPIP. They are not the licensed commercial instruments those models are best known through, and published reliability or validity figures for those instruments do not transfer to our forms.

The evidence behind the models differs a great deal. The Big Five and RIASEC structures have a large research literature; type models and many workplace frameworks have much less. Each test page says which kind of model it uses.

Every psychometric test is scored by one engine that executes a declarative configuration per test: which items belong to which scale, which are reverse-keyed, and which are attention checks excluded from scoring. The same answers always give the same scores, and no AI model is involved in computing a score.

Two answer formats are used. Rating-scale (Likert) items are averaged per scale and shown as a 0–100 position on the answer scale. Forced-choice items — pick the statement more like you — produce shares of one total, so raising one scale lowers another by construction. Such scores describe the balance inside your own answers rather than where you stand on each scale against other people, and when no style leads by more than the form can tell apart from chance, the result says so instead of naming one anyway.

A percentile on My Path compares you with people who completed the same test on this platform — never with a representative population sample. We show one only where a test has a norm table with at least 60 completions; elsewhere the result is read against the answer scale itself, such as its midpoint, and the page says so.

Norm tables currently exist for the Big Five (239 completions), the Career RIASEC test (3,987), Multiple Intelligences (283), Learning Styles (138) and Emotional Intelligence (96), and for the IQ test, whose sample is shared with my-iq-path.com (96 completed full-form and 288 short-form sittings as of October 2026). These samples are self-selected people taking tests online, pooled across all six languages and not grouped by age, sex or country. Read a percentile as your position among this site's test-takers, not in the population.

Where a test has enough completions we measure internal consistency (Cronbach's α) per scale on our own sample rather than quoting a range from the literature.

Big Five. Conscientiousness 0.72, Extraversion 0.81, Emotional Reactivity 0.74 (n = 239). The Openness and Agreeableness items were rewritten after that sample was collected, so there is no α yet for the current items, and their percentiles still compare against the earlier item version.

Career RIASEC. 0.65 to 0.76 across the six interest scales (n = 3,987).

Multiple Intelligences. 0.55 to 0.78 across eight scales (n = 283). The page used to print a literature range of .70–.85; it now prints the measured figures.

Emotional Intelligence. 0.53 to 0.76 across five scales (n = 96), where a range of .85–.95 was printed before.

Internal consistency is undefined for forced-choice shares, so four of those forms report something they can measure about themselves: how often the same choice is made when the same pair of styles comes back, against what random answering would give. Conflict Resolution Style agrees with itself on 85.8% of repeated pairs against a chance level of 75% (n = 12), the Four-Preference Style Profile 70% against 60.5% (n = 87), DISC 73.5% against 68.7% (n = 54), and Love & Affection Styles 78.7% against 75% (n = 100). The figure describes the instrument, never one person's sitting.

For the other tests no reliability figure has been measured yet. Where a test page quotes a published figure for the underlying framework, it says that the figure is the framework's, not ours.

The IQ test has a 12-item short form and a 60-item full form, both free, covering six domains: fluid, verbal, numerical and spatial reasoning, working memory and processing speed. It is untimed unless you choose a timing mode, and the timing mode only affects how processing speed is scored.

Accuracy in each domain is converted to a percentile against completed sittings of the same form; until a domain has at least 35 of them, a fixed conversion table derived from published norms is used instead. The six domains are combined with weights following each domain's loading on general ability in the literature — fluid reasoning highest, processing speed lowest. On the full form, that weighted combination of your accuracies is ranked against all full-form sittings, so the overall figure is a true percentile rank. On the short form, the overall figure is a weighted average of the six domain percentiles, which is not itself a percentile rank — a known simplification we keep for a 12-item form.

The range shown with the result is a fixed placeholder — ±12 percentile points on the short form, ±7 on the full form — not a measured standard error. "Estimated IQ ≈ N" converts the overall percentile to the mean-100, SD-15 scale, clamped to 60–145. It is an estimate of your position among this site's test-takers, labelled as such and always shown with its range. It is not age-normed and it is not a clinical IQ.

No item has been calibrated with item response theory (IRT) or a Rasch model. The per-item difficulty our norms job computes is the share of people who answer correctly — a review statistic, not an IRT parameter.

Several tests include instructed items ("choose Agree for this one") that are excluded from scoring. At submission each sitting is checked for patterns that make a result less informative: the same answer throughout, a completely flat profile, implausibly fast answering, a missed instructed item, or — on forced-choice forms — always picking the same side. A flag does not delete a result. It qualifies the claim it undermines — a flat profile, for example, means the order of your scales is rounding — and the result page says what it affects.

Results are deterministic and free. AI-written reports — single-test narratives, cross-test reports and deeper analyses — are optional and metered. A large language model writes them from what the scoring produced: each test's named scales, bands, percentiles and norm sample, its separated or not-separated verdicts and any response-quality flags, in the same form the result page uses. The model receives these numbers; it does not compute them. A fixed output contract, which editing the prompt cannot override, sets each report's length and sections.

A language model can still write a fluent sentence the numbers do not support. Read an AI report as an interpretation of your result, not as part of it, and tell us when one overreaches.

Every test is available in English, German, Spanish, French, Portuguese and Ukrainian. The non-English item sets were produced by machine translation and then read item by item against the English by an AI model — not by native-speaker reviewers or professional translators — and every review record says so. That reading found and fixed real defects that parity checks cannot see: the French Social Skills items asked something other than the English on 35 of 36 items, and the Spanish, French and Portuguese Enneagram items assumed a male reader on 21 of 55.

No analysis of whether items work the same way across languages (differential item functioning or measurement invariance) has been run, and norms are pooled across languages. Treat comparisons between language groups as unverified.

So that nothing above reads as more than it is:

• No test has been calibrated with item response theory or a Rasch model.
• No test–retest study has been run on any test, so we cannot say how stable a result is from one sitting to the next.
• No test has been compared with an established published instrument (convergent validity), and none has been checked against outcomes such as job performance or wellbeing.
• No bias (DIF) or measurement-invariance analysis has been run across languages, ages or other groups, and no fairness audit has been carried out.
• No norm sample is representative of any population, and nothing is age-normed.
• No test has been validated for clinical, diagnostic, educational-placement or hiring use. Do not use a result for those purposes.

This list gets shorter only when the work has been done and its result is published on this page.

Self-report questionnaires measure how you describe yourself: your typical behavior, preferences and feelings as you see them on the day you answer. Mood, sleep, stress, the situation you had in mind and the wish to come across well all move answers. A result is a snapshot of one sitting, not a fixed property of you, and two results a few points apart usually do not differ meaningfully.

The "around me" tests ask you to describe a partner, a team, a manager, a family, a child or a community. They describe your perception of that situation, not the other people, and are not evidence about anyone's character.

We do not sell personal data and do not run third-party ads. Results stay private until you share one, one test at a time, through a link you can revoke. You can download your data, and delete your account with all its results, from your profile. The privacy policy has the details.

Claims we published and later changed, dated by the day the change reached the live site.

• October 18, 2026 — This page: removed the “Last reviewed” date. Nobody besides the author reviews this page; what is dated now is when its figures were last refreshed.
• October 18, 2026 — Multiple Intelligences test page: the reliability figures gave the highest measured alpha as 0.77; the norm table says 0.78. The page now prints the figures from the same generated file as this page.
• October 6, 2026 — This page: removed descriptions of IRT calibration, “representative subsamples” and a fairness audit. That work was never done.
• September 20, 2026 — Multiple Intelligences result notes: the reliability figure was presented as describing the current questions. It was measured on an earlier version of the test, and the notes now say so.
• September 19, 2026 — Result notes of four workplace and family check-ins described correlations from a pilot study. No such study was ever run; the claim was removed from every test.
• September 18, 2026 — Big Five: the Openness and Agreeableness items were rebuilt after their measured consistency proved weak. Those two scales show no reliability figure until enough people have answered the new items.
• September 17, 2026 — Multiple Intelligences: the result printed a published-literature alpha range (0.70–0.85) as if it described our questions. Measured on our own sample it was lower; the measured figures replaced it.
• September 17, 2026 — Grit: eleven of the thirteen questions followed published Grit-S / Grit-O wording too closely. They were rewritten in-house.