Explainer · Plain-language

Predictive Validity: Definition, Meaning & Examples | CASRAI

Predictive validity is the extent to which a measure taken now accurately forecasts a relevant outcome in the future. It is a specialised form of criterion validity central to selection, screening, and risk-assessment tools.

CASRAI plain-language explainers — clear answers to recurring research-administration questions

The step most authors miss

Doing CRediT right? Don’t stop at the statement.

A CRediT statement credits you inside one paper. The recognition CRediT was built for happens when those roles are tied to you, persistently. Sign in with your ORCID — free — and claim your CRediT contributions on casrai.org, the home of the standard. They become a verified, portable part of your identity, not a line that disappears into one PDF.

Free: claim your contributions, then export a journal-ready CRediT statement, schema.org structured data, JATS XML, CSV or BibTeX — and preview your public profile. A membership publishes that profile publicly and verifies the journals you serve.

Forecasting a future outcome

Predictive validity asks whether a measure obtained at one point in time anticipates an outcome observed later. A university entrance exam is predictively valid to the degree that exam scores correlate with subsequent academic results; a job-selection test is predictively valid if it forecasts later performance ratings. Because the criterion lies in the future, predictive validity requires a longitudinal design — collect the measure, wait, then collect the outcome — which makes it more demanding to establish than concurrent evidence.

Why it matters for decisions

Predictive validity is the evidence base for any instrument used to make forward-looking decisions about individuals: admissions tests, recruitment assessments, credit scoring, and clinical or actuarial risk tools. Where such tools influence life chances, weak predictive validity is not merely a technical flaw but an ethical and legal concern, because decisions are then being made on a poor forecast. This is why selection instruments are expected to publish predictive validity coefficients.

Threats and attenuation

Reported predictive validity is often lower than the true relationship because of measurement artefacts. Restriction of range is a major one: if only high scorers are admitted or hired, the observable spread of both predictor and outcome shrinks, deflating the correlation. Unreliability in either the measure or the criterion attenuates it further, and the future criterion may itself be contaminated — for example, performance ratings biased by the very scores being validated. Sound studies correct for, or at least report, these factors.

Predictive validity within criterion validity

Predictive validity is best understood as one half of criterion validity, paired with concurrent validity. Both correlate a measure against an external criterion; the difference is timing — concurrent validity uses a present benchmark, predictive validity a future one. In a complete validation, predictive evidence is combined with content and construct evidence so that an instrument is justified both as a forecaster and as a sound representation of its underlying construct.

Key facts

At a glance

Definition: How well a present measure forecasts a future outcome
Type: A sub-form of criterion validity defined by a time gap
Design: Requires a longitudinal (predict-then-observe) study
Used for: Admissions, selection, screening, risk-assessment tools
Evidence: Correlation between early measure and later criterion
Main threat: Restriction of range deflates the coefficient

Common misconceptions

What people often get wrong

Often heard: Predictive validity is separate from criterion validity.

Actually: No — it is a sub-type of criterion validity, distinguished only by predicting a future rather than a concurrent criterion.

Often heard: A modest predictive correlation means the measure is useless.

Actually: No — even moderate coefficients can add real value over chance when applied at scale, though they should be reported honestly and not over-sold.

Often heard: A test with good content coverage will predict the future well.

Actually: No — covering the domain (content validity) does not guarantee forecasting power; predictive validity must be demonstrated with longitudinal data.

Going deeper