A live psychometric panel, open to research
594,590 completed assessments across 217 countries and territories in 28 languages, still growing every hour. Field your instrument on it, take a de-identified extract under agreement, or build a bid around it — without spending a grant on recruitment.
- 594,590
- Completed results
- 195
- Instruments
- 28
- Languages
- 217
- Countries & territories
Snapshot 2026-09-22, covering 2026-02-22 to 2026-09-22. 137 of the 195 published instruments are built on peer-reviewed frameworks; the rest are labelled as entertainment on their own pages and are not mixed into anything scientific. Reliability is published per scale, including the scales that score badly.
Four ways this usually works
Field an instrument
Put your scale in front of real respondents without running recruitment. We host it inside the normal assessment flow, respondents opt in, and you get the de-identified responses cut to your design — with the language and country spread of the live stream rather than an undergraduate subject pool.
Fits — Validation studies, translation and measurement-invariance work, norming a new scale
Human baselines for model evaluation
A model score is meaningless without a human distribution behind it. We hold one across cognitive and non-cognitive instruments with published internal-consistency reliability, in many languages, at a scale where per-item and per-country comparisons are actually powered.
Fits — AI evaluation, human-versus-model benchmarking, cross-lingual capability work
Joint grant applications
Funders ask for an industry partner with real infrastructure and real users. We bring the panel, the instruments, the engineering and the deployment surface; the university brings the research lead and the scientific design. We have no interest in being named on a bid we are not doing the work for.
Fits — Innovate UK, KTP, Horizon-style collaborative bids, doctoral funding
Secondary analysis
The aggregates are already public. Under agreement we go further: de-identified extracts, item-level statistics, longitudinal slices, and the metadata needed to reproduce a published figure — with a data-sharing agreement and a named purpose, not an open bucket.
Fits — Cross-cultural psychometrics, IRT and item analysis, replication work
What this sample is, and is not
It is a convenience sample. Respondents chose to take a test on a public website: self-selected, unproctored, unpaid, not quota-balanced on age, sex, education or country. It supports measurement work, cross-cultural comparison and item analysis. It does not support population prevalence estimates, and we will say so in writing if a design depends on one.
Geography is session-level. Country comes from the IP address of an anonymous session that completed at least one test — not a declared nationality. VPNs, travel and shared connections are all in there, and one person can appear as several sessions.
The counters do not all reconcile. Stored results and instrumentation events are separate tables counting different things, and we publish which is which rather than picking whichever is larger. The corpus passport names the source and the pull date for every figure.
Data governance
How extracts are handled
- De-identified before preparation — no names, emails or free text
- A data-sharing agreement with a named purpose and named PI
- Fields limited to the study design, not a bulk export
- A stated retention period, and deletion at the end of it
- Row-level security on every table; no aggregate exposes a row
The published aggregates are a different thing entirely: they are CC BY 4.0 and need no agreement at all — take them, cite us. Licence and attribution line. An extract is what you arrange with us when the aggregates are not enough.
What we ask for
- Co-authorship where we did the work — not honorary authorship
- Attribution when our data appears in a published figure
- The right to cite the findings in our own methodology
- Advance sight of a release, not approval over conclusions
We do not ask for editorial control, including over findings that are unflattering to our instruments. A partner who cannot publish a negative result is not a research partner.
Common questions
What data can a university get from JobCannon?
De-identified response data cut to an agreed study design, item-level statistics, and the metadata needed to reproduce published figures — all under a data-sharing agreement with a named purpose and a named principal investigator. The public aggregates (scale, distributions, reliability, geography, completion times) need no agreement at all and are already on this site.
Can we field our own questionnaire to JobCannon respondents?
Yes. We host the instrument inside the normal assessment flow with explicit opt-in, and return the de-identified responses. The live stream spans 217 countries and territories and, between them, 28 interface languages, which is the part a university subject pool cannot reproduce; the trade-off is that it is a self-selected sample rather than a quota-balanced one.
Is JobCannon data suitable for peer-reviewed research?
It is a large, multilingual convenience sample from self-administered, unproctored assessments — appropriate for measurement work, cross-cultural comparison and item analysis, and inappropriate for population prevalence estimates. Internal-consistency reliability is published per scale for 23 instruments (recomputed 2026-08-06), so the measurement properties can be checked before a study is designed rather than after it is reviewed.
What does JobCannon want in return?
Scientific rigour we cannot supply ourselves, and citation. Concretely: co-authorship where we did the work, the instrument or the findings being usable in our own published methodology, and attribution when our data appears in yours. We do not ask for editorial control over conclusions, including conclusions that are unflattering to our instruments.
How is respondent privacy handled?
Nothing identifying leaves the platform. Extracts are de-identified before they are prepared, free text and contact details are excluded, and geography is session-level and IP-derived rather than a declared nationality. Every published aggregate on this site is computed over counts, and no page exposes a single respondent row.
How long does it take to set up a research collaboration?
A scoping conversation, then a data-sharing agreement naming the purpose, the fields and the retention period. Fielding an existing instrument is engineering we already do weekly; a new instrument with its own consent flow is longer. Start with an email describing the design — that is enough to say yes, no, or "not in that form" quickly.
Tell us the design
One email with what you want to measure, in which languages, and at roughly what N is enough for us to say yes, no, or “not in that form” — usually within a few days. If the answer is no, you will get the reason rather than a follow-up sequence.
Institutional licensing and classroom deployment are a different conversation — see the institutional pages. How the instruments are built and validated: methodology.