Not a database.
A curated resource.

A minimum Cronbach's alpha or composite reliability of .80 is required across the primary validation sample. The widely cited .70 threshold, originating from Nunnally's early psychometric work, was a pragmatic heuristic rather than a theoretically derived standard — and one Nunnally later revised upward.

At .70, approximately 30% of observed score variance is attributable to measurement error rather than the construct of interest. For researchers building structural models or testing fine-grained hypotheses, that level of measurement imprecision matters. The .80 threshold reduces that error proportion meaningfully, producing more precise parameter estimates, stronger statistical power in hypothesis testing, and greater confidence that relationships observed between constructs reflect theoretical reality rather than shared measurement noise.

In practical terms, scales meeting the .80 standard are more likely to detect true effects, less likely to produce spurious ones, and more defensible in peer review. A small number of seminal legacy instruments with exceptional citation impact and established disciplinary use are assessed on a case-by-case basis.

Single-item scales are excluded. Without multiple indicators, it is impossible to separate true score variance from random measurement error, and reliability in the classical sense cannot be estimated at all — a scale of one item has no internal consistency to assess.

Single items are also acutely sensitive to question wording effects, momentary response biases, and context-order artefacts that multi-item scales partially cancel out through aggregation. Beyond reliability, single items preclude meaningful tests of dimensionality, convergent validity, or discriminant validity — the core evidence base that justifies treating a measure as capturing a distinct psychological construct.

Researchers selecting instruments from this collection are typically building structural models where measurement quality propagates directly into the validity of latent variable estimates. A single item introduced at that stage does not just underperform — it contaminates.

Scales that reproduce items verbatim from a prior instrument without substantive modification are generally excluded. The collection prioritises genuinely new instruments and bona fide adaptations — where the focal paper has made meaningful changes and established independent validation evidence.

However, the boundary is not purely syntactic. Where a prior instrument is deployed in a substantively novel context — a new population, domain, or theoretical framework — and the focal paper provides the first rigorous demonstration of predictive validity in that context, inclusion is considered on its merits. The relevant question is not whether the items are new, but whether the paper makes an original psychometric contribution.

Verbatim reproductions serving purely as manipulation checks, covariates, or convenience measures without independent validation evidence remain excluded.